Pricing Behaviour and Non-Price Characteristics in the Airline Industry (Advances in Airline Economics) 1780524684, 9781780524689

Citation preview

PRICING BEHAVIOR AND NON-PRICE CHARACTERISTICS IN THE AIRLINE INDUSTRY

ADVANCES IN AIRLINE ECONOMICS Series Editor: James Peoples Recent Volumes: Volume 1:

Competition Policy and Anti-Trust, Darin Lee

Volume 2:

The Economics of Airline Institutions, Operations and Marketing, Darin Lee

ADVANCES IN AIRLINE ECONOMICS VOLUME 3

PRICING BEHAVIOR AND NON-PRICE CHARACTERISTICS IN THE AIRLINE INDUSTRY EDITED BY

JAMES PEOPLES University of Wisconsin-Milwaukee, WI, USA

United Kingdom – North America – Japan India – Malaysia – China

LIST OF CONTRIBUTORS Volodymyr Bilotkach

Senior Lecturer in Strategic Management and International Business, Newcastle Business School, Northumbria University, Newcastle upon Tyne, UK

Jeffrey P. Cohen

Associate Professor of Economics, University of Hartford, CT, USA

Cletus C. Coughlin

Senior Vice President and Policy Adviser, Office of the President, Federal Reserve Bank of St. Louis, MO, USA

Yan Du

PhD of Economics, Oregon State University, OR, USA

Alexander Eisenkopf

Professor, Zeppelin University, Friedrichshafen, Germany

David Gillen

Professor, Sauder School of Business, University of British Columbia, Vancouver, Canada

Tim Hazledine

Professor, University of Auckland, Auckland, New Zealand

Kevin E. Henrickson

Associate Professor of Economics, Gonzaga University, WA, USA

Manuel A. Hernandez

Postdoctoral Fellow, International Food Policy Research Institute, Washington DC, USA

Christian Hofer

Assistant Professor of Supply Chain Management, University of Arkansas, AR, USA

Marc Ivaldi

Toulouse School of Economics, School of Advanced Studies in the Social Sciences, Toulouse, France vii

viii

LIST OF CONTRIBUTORS

Andreas Knorr

Professor of Economics, German University of Administrative Sciences Speyer, Speyer, Germany

Andreas Lueg-Arndt

Professor of Economics, Cologne Business School. European University of Applied Sciences, Cologne, Germany

Dan Mahoney

Department of Economics, University of Oregon, OR, USA

B. Starr McMullen

Professor, Oregon State University, OR, USA

Marija Pejcinovska

University of California, Irvine, CA, USA

James Peoples

Professor of Economics, University of Wisconsin-Milwaukee, Wisconsin, USA

Nicholas G. Rupp

Associate Professor of Economics, East Carolina University, NC, USA

Ian Savage

Department of Economics and the Transportation Center, Northwestern University, IL, USA

John Scott

Gonzaga University, WA, USA

Anirban Sengupta

Director, Analysis Research and Planning Corporation, Washington DC, USA

Senay Sokullu

Assistant Professor, Department of Economics, University of Bristol, Bristol, UK

Tuba Toru

Toulouse School of Economics, School of Advanced Studies in the Social Sciences, Toulouse, France

Steven N. Wiggins

Professor, Texas A & M University, TX, USA

Wesley W. Wilson

Professor of Economics, University of Oregon, OR, USA

Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2012 Copyright r 2012 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. The opinions expressed in these chapters are not necessarily those of the Editor or the publisher. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-1-78052-468-9 ISSN: 2212-1609 (Series)

Emerald Group Publishing Limited, Howard House, Environmental Management System has been certified by ISOQAR to ISO 14001:2004 standards Awarded in recognition of Emerald’s production department’s adherence to quality systems and processes when preparing scholarly journals for print

CHAPTER 1 PRICING BEHAVIOR AND NON-PRICE CHARACTERISTICS OF THE AIRLINE INDUSTRY: INTRODUCTION AND OVERVIEW James Peoples Air transportation service is an important part of globalization for businesses and leisure travelers in part because it provides relatively affordable and fast transport across long distances. Society benefits from easy access to an efficient air transportation system that enhances face-to-face business transaction among international companies and promotes tourism worldwide. Travelers have apparently taken advantage of the benefits arising from the use of air transportation, as global passenger trends indicate impressive industry growth. Information from the International Air Transport Association reports leisure and business travel by air grew 7 percent annually from 1980 to 1990. The same IATA data show the largest share of air passengers are from flights originating in the United States, Canada, and Western Europe. High consumer demand for air transportation service, though, is hardly limited to traditional developed countries as air travel originating from emerging economies in Asia is growing rapidly at an annual rate near 9 percent (IATA). Given the economic and social benefits associated with global access to an affordable air transportation system, it is important to understand the factors contributing to low airfares as well as examine whether the practice of charging low fares is common on all routes.

Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 1–9 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003003

1

2

JAMES PEOPLES

Regulatory reform beginning in the latter part of the last century played a key role stimulating worldwide competition in passenger transportation services.1 For instance, removal of regulatory rate setting and easing of entry restrictions had an immediate impact on the air transportation business environment in the United States. The number of certificate of services tripled in the six years following airline deregulation (Card, 1998), and average passenger airfares fell by 33 percent (Winston, 1998). In this more highly competitive environment, carriers had greater incentive to incorporate efficiency-enhancing operating techniques and cost-saving technology. Arguably the most prominent cost-saving technique employed by legacy carriers was the development of hub-and-spoke networks. The introduction of this technique contributed to increased load-factors by allowing carriers to transport passengers from originating cities to a major airport, which is the airline’s hub. From the hub travelers from different originating cities are grouped together to take connecting flights to a common destination. Savings arise because flights carrying loads near passenger-capacity levels reduce per passenger unit costs. Continued investment in fuel-efficient aircrafts further contributes to carrier cost-savings.2 Despite impressive passenger welfare gains associated with a more competitive and efficient airline industry, past research indicates that the introduction of the hub-and-spoke system as well as carrier dominance of a few computer reservation systems contributed to noncompetitive pricing for routes served primarily by the hub carrier (Borenstein, 1989). Chapters in this volume contribute to the debate on fares charged by airline carriers by examining pricing behavior such as nonlinear and competitive pricing. Analyses performed by these studies include examination of pricing decisions associated with strategic alliances, competition from low-cost carrier (LCC), and fare setting when facing financial distress. Research contributions on pricing behavior by these studies present a nuanced analysis of fare determination that includes new ways of examining whether the prices charged by carriers are welfare enhancing. Examination of airline carrier pricing, though, only tells part of the story of firm behavior, as analysis of carriers’ ability to provide quality services and the social externalities associated with providing air transportation services is also significant in understanding the economics of this industry. This volume contributes to the analysis of non-price characteristics of air transportation business operations by examining frequency of service, passenger safety, availability of aircraft types for different transportation services and the geographic distribution of air transportation related noise by aircraft departures and arrivals at airports.

Introduction and Overview

3

COMPETITION, ONLINE TICKET AGENCIES, CARRIER FINANCIAL CONDITIONS, AND AIRLINE PRICING Research analyzing competition and individual carriers’ financial condition’s influence on pricing behavior in the airline industry are presented in the first part of the book. The initial chapter by Manuel Hernandez, Anirban Sengupta, and Steven Wiggins directly tests whether competition creates a challenge for legacy carriers practicing nonlinear pricing such that passengers are charged different prices for the same flight without cost justification. The authors use transactions level data, which has the advantage of allowing an empirical examination of the effect of Southwest and other LCCs on both the level and structure of fares of legacy carriers. The analysis is performed by defining a menu of airline fare types according to restrictive ticket characteristics to then evaluate the competitive effects of both Southwest and other LCCs, including adjacent and potential competition from Southwest, on the relative pricing behavior of major carriers. Findings suggest that competition from Southwest has an important effect on both the level and fare structure of legacy carriers. In particular, direct and potential competition from Southwest both lower the fare per mile and compress the fare structure by decreasing the premia of the highest fares, including first-class tickets, over the lowest fares. Adjacent competition from Southwest and direct competition from other LCCs only seem to significantly affect the fare level. The following chapter by David Gillen and Tim Hazledine further contributes to the analysis of nonlinear pricing by focusing on the influence of LCCs and the adoption of Internet-based ticket sales systems on legacy carriers’ ability to charge prices that differ by the date of purchase. The authors use price data for over 1,700 flights on 39 Canadian and transborder US routes observed in May 2006 to explore fare determination of legacy carriers in this more competitive business environment. Their findings suggest that there continues to be extensive price discrimination based on date of purchase of ticket and other factors. However, average prices are nevertheless still significantly determined by the number and size of airlines supplying a route. The authors also find that established ‘‘legacy’’ carriers can still charge a substantial price premium over LCCs, and the internet fare systems may have made it easier for legacy carriers to coordinate the typically substantial increases in their fares over the last two weeks before the departure flight date.

4

JAMES PEOPLES

While Gillen and Hazledine’s analysis reveals the significance of considering how competition influences the pricing power of legacy carriers when tickets are sold close to the fight departure, the need to fill remaining seats might strengthen the negotiation advantage of passengers who wait and shop for the best fares as departure time nears. In the succeeding chapter Volodymyr Bilotkach and Nicholas Rupp explore the possibility of passengers benefiting from purchasing tickets at a late date by constructing and analyzing price-offer curves (the dynamics of offered prices as the departure date approaches) for 105 specific round-trip itineraries on 50 busy US routes. Data used for this analysis were collected from the three leading online travel agents’ websites. The benefit associated with using these data is that it permits the authors to exploit across-route variation in the level of competition and presence of LCCs, in particular Southwest Airlines (a carrier that does not sell its tickets through online travel agents). Findings suggest that fares and yields are consistently higher along the entire price-offer curve on less competitive markets, and on routes without LCC presence. Price changes are smoother on competitive routes than on markets with one or two competitors. Price drops are observed across a spectrum of the markets, and at any day prior to departure. In particular, at least one price drop was observed within 10 days before the flight for about half of all the round trips they tracked. In about one-third of round-trip itineraries, they observed price drops in the last week prior to departure. At the same time, the shape of the average price-offer curve is as expected – flat up to about three weeks before the flight, and rising rapidly afterwards. The findings do not suggest systematic differences across online travel agents; however, differences in price quotes are not infrequent. Simple cost-benefit analysis also shows that when booking a ticket closer to the planned departure date a traveler should comparison-shop. Chapters 1–4 reveal the increasing importance of online travel agencies to post information on fares. In Chapter 5 Volodymyr Bilotkach and Marija Pejcinovska explore the ability of these agencies to use their market power to post fare quotes that may provide a competitive advantage to carriers. These authors use evidence from the travel services distribution industry to examine the impact of intermediaries that are supposedly pure distributors on the final price quotes observed by the customers. They analyze a sample of fare quotes, collected from the three leading online travel agents on 50 large US airline markets. Looking at the lowest fares available reveals interesting results. For example, the fewer the agents who offer the lowest fare, the higher the fare is. The effect on fares is more pronounced in those airline markets that are less competitive. Finally, they find individual agents

Introduction and Overview

5

appear to discriminate for or against individual airlines when deciding whether or not to report airlines’ lowest offered fare quotes. Pricing behavior by carriers is shaped not only by product market characteristics such as the number of competitors and the ability of travel agents to indulge in price discrimination, but also by the financial conditions that individual carriers face. In Chapter 6 Christian Hofer considers the effect of changing financial conditions on the price behavior of financially distressed carriers. He notes that past research has presented evidence that distressed carriers’ fares are lower, all else equal. Yet, little is known about how such price effects vary throughout the turnaround process. Based on the contention that the effect of distress on airfares is nonlinear, this research aims to provide more detailed insights into the nature and magnitude of distressed carriers’ pricing actions during the downturn and recovery phases of the turnaround process. Several hypotheses are developed and tested using a large panel data set from the US domestic airline industry. This chapter’s empirical analyses provide ample support for the contention that a distressed firm’s strategic options and, thus, its pricing behavior change as the firm proceeds through the downturn and recovery phases. A potential strategy for addressing financial woes expressed in the previous chapter is to form alliances through code-sharing. This type of alliance depicts an agreement between carriers where a carrier operating a specific flight allows another carrier to market and sell seats on that same flight. Gains derived from such collaboration can arise from the enhancement of a carrier’s presence on the new route served by the operating carrier, from an increase in fight options for the marketing carrier without the cost of additional investment, and from the potential to dominate the code-sharing route. In their contribution to this volume, Starr McMullen and Yan Du identify factors that determine which routes will stay in a code-share agreement and which will be dropped. Their findings also show airport dominance and yields (fares) are the most important factors affecting alliance firms’ decisions to code-share on an individual route. For example, alliance carriers take advantage of their hub dominance to build code-shared routes with their partners. Their findings indicate that alliances may be weak on routes with few competitors, as high route-level concentration was found to discourage carriers from staying in a code-share arrangement after initial entry. In the concluding chapter of this volume’s section on price behavior, Kevin Henrickson and John Scott examine the increasingly common practice of charging fees in addition to airfares. These authors note that the dramatic increase in oil prices over the past several years has put tremendous pressure on the airline industry’s ability to be profitable. As such, many airlines have

6

JAMES PEOPLES

sought out new sources of revenue, and have turned to fees for services that were previously provided free of charge, the most well known of which is a charge on checked baggage. Their empirical analysis incorporates a spatial autoregressive model to estimate the impact of the introduction of baggage fees on airfares. Their results indicate that firms charging baggage fees lower their ticket prices to appear more competitive, and that a traveler with one checked bag will pay $0.76 more for every $1 increase in baggage fees. These fees also have allowed Southwest Airlines, which advertises ‘‘Bags Fly Free,’’ to differentiate itself from other carriers and charge higher airfares on routes in which they compete with airlines charging baggage fees. Specifically, a passenger can expect to pay $0.73 more on Southwest for every $1 increase in baggage fees on the same route.

NON-PRICING CHARACTERISTICS AND EXTERNALITIES Chapters examining non-price characteristics are initially presented in this section, followed by analysis of air transportation externalities. The first chapter by Andreas Knorr, Andreas Lueg-Arndt, and Alexander Eisenkopf examines the significance of airport networks as determinants of carriers’ choice of aircraft. These authors motivate their study by identifying the different business approaches of the two major passenger aircraft manufactures, Boeing and Airbus, regarding their assessment of the future development of airline (and alliance) networks. They note that Boeing, having long predicted major growth in intercontinental point-to-point operations – based on the so-called fragmentation (‘‘dehubbing’’) hypothesis –, has consistently opted for the development of midsized wide-body aircraft, while Airbus continues to forecast a substantial increase in hub-to-hub traffic, which it argues will be best met with its A380 super jumbo. Knorr, LuegArndt, and Eisenkopf empirically test these competing hypotheses using data covering the 1982–2007 observation period. Their findings produced no empirical evidence for the dehubbing hypothesis. They report that while some fragmentation does indeed take place, especially on the North Atlantic market, it did not occur in the form of point-to-point traffic that entirely bypasses the hub airports. Rather, airlines have bypassed overseas hubs by deploying smaller aircraft to connect secondary markets abroad to their primary hubs. As a result the number of spokes has increased for the biggest hub airports.

Introduction and Overview

7

Further analysis on the role of airports as a platform for providing air transportation is examined by Marc Ivaldi, Senay Sokullu, and Tuba Toru in Chapter 10. They focus on the critical influence of airports on airline services and passenger demand by providing a methodology to analyze airports under two-sided market setting. Making this market observation on airports represents a significant contribution to the literature on airline economics in part because airports are platforms that airlines and passengers use to interact and make business transactions. After developing a structural model, the authors use data on US airports to provide empirical evidence for two-sidedness. Starting with a monopoly platform, they derive the demand equation of passengers and pricing equation of airlines that are then estimated simultaneously. To verify the structure of the market, they examine the significance of network effect parameters. Their findings suggest airports are indeed two-sided platforms since the coefficients of flight frequencies and airport characteristics are significant in passenger demand. In addition, their findings indicate that airports can cross-subsidize the two sides with respect to their elasticities. In the succeeding chapter, Dan Mahoney and Wesley Wilson address the question asking what is the appropriate size of an airport to provide service that satisfies consumer demand and what are the determinants of airport size and growth? Their chapter uses concepts from classical consumer theory to construct an empirical model of airport size determination. This model includes traditional determinants of consumer demand such as local income level and population size, as well as airline-specific factors such as location of alternative airports and the attributes of these airports and the observed airport. Findings are consistent with consumer theory suggesting that airport size is positively associated with high local income and high local population levels. What is new are findings suggesting hub airports and airports servicing low-cost traffic promote airport growth. Findings also indicate that the ‘‘presence’’ of an alternative airport depends on the distance to the nearest airport and, as distances to the rival increases, passengers at the airport increase, underscoring the effects of alternative airports in attracting passengers. While findings from the earlier chapters in this section indicate the significance of airports as facilities that contribute to the convenience of travelers, the frequency of flight arrivals and departures can create externalities that negatively influence individuals who live in close proximity of the airport. Past research examining social externalities associated with airport activity suggests that often Hispanic residents are subjected to higher noise levels than residents in other neighborhoods near an airport (Sobotta, Campbell, & Owens, 2007). In Chapter 12 Jeffrey Cohen and Cletus Coughlin argue that

8

JAMES PEOPLES

analysis of spatial heterogeneity would provide a more accurate description of the distribution of noise among different demographic groups living near an airport. These authors explore the issue of spatial heterogeneity in the context of the determinants of airport noise using houses sold near the Atlanta airport in 2003. They estimate ordered probit locally weighted regressions (OPLWR) and produce results substantively different than those using a standard ordered probit model. Findings reveal notable differences in both the signs and magnitudes of the parameter estimates for different individual observations (e.g., houses) using OPLWR. For instance, findings using the OPLWR approach indicate greater spatial heterogeneity among Hispanic households. The majority of these households still reside in high-noise locations, but a significantly larger percentage lives in low-noise locations when using the OPLWR approach compared to the standard ordered probit model. In the final chapter of this volume, Ian Savage further explores externalities associated with airline services. He observes that safety is arguably the most important ‘‘quality’’ attribute of commercial aviation, yet it rarely figures into overt interfirm rivalry. This chapter lays out the underlying economic models of safety provision and the demand for safety by passengers, and concludes that profit-maximizing firms should seek to diversify their safety offering. However, crucial failures in market processes remove the desirability to diversify and lead to minimal product differentiation among mainstream airlines. That said, the chapter concludes by pointing to two examples of competitive markets where there are actual or imagined differences in safety between the rival airlines.

CONCLUDING REMARKS Information presented in this volume contributes to our gaining a better understanding of the effect of market forces on the choice of fares and service quality available in the airline industry. These studies show that while the opportunity for passengers to receive low fares exist and is even relatively common, the practice of price discrimination still persists in this industry. The ability for carriers to form horizontal and vertical alliances can create market advantages that promote the adoption of noncompetitive pricing strategies. Nonetheless, findings in this volume do not suggest the need for a major overhaul of the regulatory apparatus to address pockets of relatively low consumer welfare. Rather, select attention to fares set on noncompetitive hub-spokes and on fares posted where there is identifiable price discrimination among online travel agencies are identified as areas that might benefit from of oversight attention.

Introduction and Overview

9

Analysis of non-price characteristics in this industry reveals an airline industry that is responsive to the demands of their customers. Carriers’ choice of aircrafts and airport systems as well as growth of airports in strategic locations presents passengers with a convenient and efficient transportation service. Findings on externalities associated with the use of air services reveal where oversight agencies should investigate to locate potential market failures. Observations presented in these chapters, though, do not indicate an industry characterized with unbridled external cost heavily borne by a few groups. Rather, new evidence presented in this volume indicates that noise distribution is not as inequitable as previously thought. In addition, evidence presented in this volume indicates that cost cutting that can potentially lead to unsafe conditions for passengers is a risky business endeavor that carriers would be wise to avoid given the ability of passengers to identify airlines with poor safety records.

NOTES 1. Legislation introducing airline deregulation occurred in the late 1970s in the United States and in the early 1990s for EU countries. 2. Research indicates relatively small gains in fuel efficiency due to carriers’ increased reliance on regional jets. The fuel consumption performance of these types of aircrafts does not compare favorably to larger long-haul carriers (Babikian, Lukacku, & Waitz, 2002; Air Transport Department, Cranfield University, 2010).

REFERENCES Air Transport Department, Cranfield University. (2010). Fuel and air transport. Report for the European Commission. Babikian, R., Lukacku, S., & Waitz, I. (2002). The historical fuel efficiency characteristics of regional aircraft from technological operational and cost perspectives. Journal of Air Transport Management, 80(6), 389–400. Borenstein, S. (1989). Hubs and high fares: Dominance and market power in the US airline industry. Rand Journal of Economics, 20, 344–363. Card, D. (1998). Deregulation and labor earnings. In: J. Peoples (Ed.), Regulatory reform and labor markets (pp. 183–230). Boston, MA: Springer Publishers. Sobotta, R. R., Campbell, H. E., & Owens, B. J. (2007). Aviation noise and environmental justice. Journal of Regional Science, 47, 125–154. Winston, C. (1998). US industry adjustment to economic deregulation. Journal of Economic Perspectives, 12(3), 89–110.

CHAPTER 2 EXAMINING THE EFFECT OF LOW-COST CARRIERS ON NONLINEAR PRICING STRATEGIES OF LEGACY AIRLINES Manuel A. Hernandez, Anirban Sengupta and Steven N. Wiggins INTRODUCTION Firms with linear pricing offer their customers the same price for each unit of a good or service. Anything else is nonlinear pricing. Nonlinear pricing in imperfect markets indicates a fundamental asymmetry in information between firms and consumers. Consumers are commonly expected to exhibit quality- or quantity-preference differences and have different reservation values for different product attributes. The firms, however, cannot observe consumers’ preferences. When complete information regarding preferences is not observable, nonlinear pricing strategies with firms offering a menu or schedule of prices allow consumers to sort themselves according to their own preferences, resulting in market segmentation. The airline industry provides an ideal setting to examine this particular form of price discrimination.1 By offering a schedule of fares accompanied with varying ticket characteristics, travelers are required to self-select into different ‘‘bins’’ or fare types based on their individual preferences and Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 11–53 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003004

11

12

MANUEL A. HERNANDEZ ET AL.

requirements, leading to market segmentation. The menu of fare types, in turn, rarely differs across routes with different numbers and types of competitors. The testing of nonlinear pricing strategies in the airline industry, however, has been significantly limited due to the lack of adequate data, particularly data on individual ticket restrictions.2 The study by Hernandez and Wiggins (2010) is the first study that uses a detailed, ticket-level data set to analyze the impact of market competition on nonlinear pricing strategies in the US airline industry. They separate tickets into broad fare groups defined by restrictive ticket characteristics and examine the impact of increased competition on the relative pricing of the menu of fare types. They find that the ratio of high-to-low-quality fares increases as competition rises, but that the ratio of medium-to-low-quality fares decreases with competition. Similarly, they find that competition from Southwest substantially affects both the level and structure of fares on a route. Similar to Hernandez and Wiggins (2010), we take advantage of a detailed, ticket-level data set to group tickets into different fare types based on restrictive ticket characteristics and focus on the impact of low-cost carriers (LCCs) on the nonlinear pricing strategies of legacy carriers. The legacy carriers include American, United, Delta, Continental, US Airways, and Northwest. In a market setting where firms offer a menu of prices, the presence of low-cost firms presumably affects prices on the lower end of the fare schedule of the other competitors. However, there is no clear prediction regarding the effect of low-cost firms over the entire fare schedule. Our identification strategy relies on the fact that legacy carriers generally offer the same menu of fare types across routes with and without competition from LCCs. Considering the importance of Southwest in the industry, we account separately for the direct presence of Southwest and other LCCs on a route. We also examine the relative pricing behavior of legacy carriers in face of adjacent competition from Southwest (i.e., competition from secondary airports on the same route) and potential competition from Southwest (i.e., situations where Southwest operates at both endpoint airports but does not fly the route). Similarly, we allow for a differentiated impact of Southwest and other LCCs across the major carriers. The existing literature on nonlinear pricing in the airline industry lacks a comprehensive study of nonlinear pricing strategies by legacy carriers in the presence of Southwest and other LCCs. The present study aims to fill this gap in the literature. Our study is in line with Alderighi, Cento, Nijkamp, and Rietveld (2004), who investigate the response of full-service carriers to the entry of LCCs in the European market, but we have richer data to

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

13

perform the analysis over a broader set of fare types. We also control for an exhaustive set of ticket, flight, and market characteristics. The estimation results indicate that competition from Southwest has an important effect on both the level and fare structure of the major carriers. In particular, we find that both direct and potential competition from Southwest lower fares, as documented in previous studies, but also compress the entire fare structure of legacy carriers.3 We observe a decrease in the premia of the highest fares, including first-class tickets, relative to the lowest fares with actual and potential presence of Southwest on a route. Adjacent competition from Southwest and direct competition from other LCCs also decrease average fares, but do not seem to compress the fare structure. The results, however, cannot be generalized across all the legacy carriers as suggested by a differentiated analysis by carrier. The remainder of the chapter is organized as follows. The second section provides a literature review on nonlinear pricing, particularly in the airline industry, as well as some discussion on the potential impact of LCCs on airline fares. The third section describes the data used in the analysis. The fourth section discusses the empirical methodology and presents the estimation results, while the last section concludes.

LITERATURE REVIEW Nonlinear pricing strategies by competing firms is a standard practice in any industry where competing firms offer a menu of prices to consumers who self-select based on their preferences, resulting in market segmentation. There exists an abundance of empirical literature testing the theoretical predictions of nonlinear pricing strategies in gasoline, beverages, advertising, auto rentals, and airlines, among others. The discussion below highlights some of the most important works on nonlinear pricing in the different industries, with emphasis on the airline industry, as well as on the potential impact of LCCs on prices. Shepard (1991) and Borenstein and Shepard (1996) find evidence of second-degree price discrimination in the retail market for gasoline. Studies by Cohen (2004) looking at pricing of paper towel packages of different sizes, Clerides (2002) studying intertemporal pricing in the book market, and McManus (2007) investigating the optimality of nonlinear pricing of coffee drinks also provide empirical evidence of firms in oligopoly markets using distinct product qualities or quantities to segment customers.

14

MANUEL A. HERNANDEZ ET AL.

Nonlinear pricing is also a standard practice in the advertising industry; evidence of its application has been documented in Yellow Pages advertising. Busse and Rysman (2005) examine the effect of competition on second-degree price discrimination in display advertising in Yellow Pages directories. In particular, they evaluate the effect of competition on prices at high versus low ends of the quality distribution (large versus small advertisements) and find evidence that competition increases the curvature of the price schedule. Large advertisements experience the largest fall in prices with increased competition. These findings are consistent with the predictions of Rochet and Stole’s (2002) model on nonlinear pricing and imperfect competition, but inconsistent with Stole’s (1995) model in which high-valuation customers are also associated with high brand loyalty. Auto rental firms offer a menu of prices that differ across car quality (e.g., midsized versus economy cars) and consumer types (business versus leisure customers). Using a database that includes the number of auto rental operators at every commercial airport in the United States along with rental prices for different consumer segments and car types, Khan, Singh, and Zhu (2009) find that one additional competitor in the market reduces average prices by approximately 4.5 percent. This effect, however, is not uniform across segments or quality levels. While prices for both business (weekday) and leisure (weekend) segments fall with competition, price for the leisure segment falls even faster, resulting in increased dispersion with increased competition. The converse is observed for the weekend market where the prices for large-size cars fall faster, resulting in decreased price dispersion. These findings suggest that the impact of competition on the shape of the nonlinear price schedule depends on the underlying distribution of customer preferences. The existing literature on nonlinear pricing strategies in the airline industry is also fairly exhaustive, particularly in the US market. Borenstein (1989), using DB1B data from the Bureau of Transportation Statistics (BTS), finds that increased competition leads prices to drop faster at the lower end of the fare distribution. The results are, however, contrary to the findings of Busse and Rysman (2005), who attribute this contradiction to the brand-loyalty programs in the airline industry.4 With more competition, airlines compete more fiercely for the low-type consumers who are less brand-loyal, such that low fares decrease at a faster rate than the high-end fares. Stavins (2001) approximates price discrimination with marginal implicit prices of ticket restrictions using a large cross-section of tickets on various routes. She finds that price discrimination increases with market competition. Verlinda and Lane (2004) examine the relationship between nonlinear pricing and consumer search costs in the airline industry during a period of

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

15

increasing internet use. They find that the spread between unrestricted and restricted fares increases as internet usage increases. They also show that the ratio of unrestricted to restricted ticket fares increases with more competition, suggesting that increased internet penetration induces more competitive behavior. More recently, Hernandez and Wiggins (2010) take advantage of a detailed transaction data set on airline tickets and find that market concentration differentially impacts various types of fares. In particular, they find a nonnegligible decrease in the quality premium of high-to-low-type fares as we move from competitive to monopoly markets, while the ratio of medium-to-low-type fares increases with market concentration. Turning to the impact of LCCs on airline fares, Morrison (2001) provides some of the first insights of the effects of Southwest by accounting for actual, adjacent, and potential competition from this carrier. He estimates that Southwest effects generate about $12.9 billion in savings, of which $9.5 billion were attributed to actual, adjacent, and potential effects that Southwest had on its competitors’ fares, while the remaining $3.4 billion were passed on to the passengers directly. Wang (2005) also shows that Southwest competition decreases the average price in the market it operates and induces the other carriers in those markets to lower their fares. Lee and Luengo-Prado (2005) find a strong negative impact of LCCs on prices when examining hub premiums from major carriers. Sengupta and Wiggins (2006) document that Southwest posits a downward pressure on average prices on the route. Goolsbee and Syverson (2008) show, in turn, that Southwest’s potential entry on a route causes the incumbent to reduce its fares significantly. These fare cuts were only on threatened routes and not in nonSouthwest competing airports. Hernandez and Wiggins (2010) also find that (direct and potential) competition from Southwest substantially affects both the level and fare structure on a route when examining the impact of market concentration on different fare types. Brueckner, Lee, and Singer (2010) recently extended the literature on the fare impacts of LCCs by incorporating an adjacent-airport approach to offer a comprehensive picture of the competitive effects of both legacy carriers and LCCs. Their analysis measures the impact of airport-pair competition and adjacent competition for both types of carriers, and also captures the impact of potential competition from LCCs. The analysis is separated in nonstop and connecting markets. The authors find evidence of weak effects on fares when legacy carriers compete contrary to the significant effect on prices due to competition among LCCs, irrespective of market – airport-pair, at adjacent airports, or potential competition.

16

MANUEL A. HERNANDEZ ET AL.

The present study contributes to the above literature by directly examining the impact of LCCs on the fare structure of various fare types of legacy carriers. In a market setting where firms offer a menu of prices, the presence of low-cost firms presumably affects prices on the lower end of the price schedule of the other competitors. However, studies regarding the effect of these lowcost firms over the entire price distribution of legacy carriers are absent in the existing literature, and the present study intends to fill this gap. Our analysis is related to Alderighi et al. (2004), who investigate the response of full-service carriers to the entry of LCCs in the European market and find that competition with LCCs reduces the fares of full-service carriers in all service classes (business and leisure segments) proportionally. We differentiate our work from that of Alderighi et al. (2004) by exploiting transactional level data that allows us to work with various fare types and examine the impact of Southwest and other LCCs on the relative pricing of a broader fare menu. As noted earlier, we follow Hernandez and Wiggins (2010) and group fares according to certain ticket characteristics and restrictions. We specifically focus on the competitive effects of Southwest and other LCCs, including adjacent and potential competition from Southwest, on the nonlinear pricing behavior of legacy carriers. We further allow for a differentiated impact across the major carriers.

DATA The main source of data for this study is a census of airline ticket transactions from a major computer reservation system (CRS). The database consists of tickets purchased between June and December 2004 for travel in the fourth quarter of the same year. It includes tickets purchased directly from airlines, including their websites, and through travel agents and several online travel sites. Overall, the data represent around 30 percent of all domestic ticket transactions in the United States during the corresponding quarter. For each ticket sold or itinerary, we have information on the fare paid, origin and destination airports, segments involved in the itinerary, carrier and flight number, cabin and booking class, online/offline purchase, and dates of purchase, departure, and return. Due to confidentiality reasons, the major CRS vendor did not provide information on ticket restrictions, including advance purchase requirements, refundability, and travel and stay restrictions. As a result, we merged the transaction data set to historical data from a travel agent’s CRS containing a large subset of ticket fares offered for travel in the last quarter of 2004.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

17

The matching procedure merges an itinerary from the transaction data set to a fare from the travel agent’s data set based on route, carrier, and prices.5 The matching process ensured that fares matched within 2 percent and that the itinerary matched advance purchase requirements and travel and stay restrictions.6 Consistent with the previous literature, a route is defined as an airport-pair, regardless of direction. We restrict the analysis to nonstop itineraries and exclude itineraries except one-way and roundtrip itineraries. Tickets that involve travel on different airlines (interline tickets) are also excluded. Prices are measured as roundtrip fares, and in the case of one-way tickets the fare is doubled. To avoid holiday peaks, transactions involving travel on Thanksgiving, Christmas, and New Year’s are dropped.7 As in previous studies, we also eliminate tickets with very low prices, which presumably are frequent-flyer tickets.8 As noted above, we restrict our analysis to tickets for flights operated by American, United, Delta, Continental, US Airways, and Northwest. Besides Southwest, each of these six legacy carriers transported at least 5 percent of all domestic travelers during the fourth quarter of 2004. We also limit the analysis to matched itineraries such that there are at least 1,000 observations per route and 100 observations per carrier-route. The final analysis includes 755,842 tickets on 215 routes or 339 carrier-routes. The list of routes is provided in Table 1. Following Hernandez and Wiggins (2010), we group the matched itineraries into four broad quality groups. These four categories characterize the main differences in terms of ticket restrictions and explain the main differences among ticket types. Group F fares correspond to first class tickets. Group 1 fares include refundable business, full coach, and coach tickets. Group 2 fares represent nonrefundable tickets without travel or stay restrictions, while Group 3 fares include nonrefundable tickets with travel and/or stay restrictions. Group F through Group 3 ranks the overall quality of tickets in descending order, with Group F representing the highest quality ticket and Group 3 representing the lowest. The top panel of Fig. 1 shows that average fares decline as we move across groups. In particular, the average fare per mile among tickets in Group F through Group 3 is 97, 65, 27, and 17 cents. This pattern confirms the positive correlation between the grouping of tickets and prices, which is recurrent across all legacy carriers, as shown at the bottom of Fig. 1.9 The transaction data is supplemented with several market-level variables consistent with the airline literature (see Borenstein, 1989; Borenstein & Rose, 1994; Brueckner, Dyer, & Spiller, 1992; Gerardi & Shapiro, 2009; Lee &

18

Table 1.

Routes by Southwest and Other LCCs’ Presence. BWI-STL PHL-PBI

DTW-PHX PHL-PHX

LAX-LAS SEA-SLC

LAX-SMF SJC-SAN

LAX-SJC SJC-SNA

Routes with other LCCs’ presence ATL-BOS ATL-DEN ATL-MEM ATL-RDU BWI-ATL DCA-ATL DEN-MCO DEN-PHL DEN-STL DFW-BWI EWR-ATL EWR-MDW JFK-SEA LAS-ATL LGA-MCO MCI-DEN MIA-ATL MSP-LAS ORD-FLL ORD-IAD SFO-ATL SNA-DEN

ATL-DFW ATL-TPA DEN-BOS DEN-SEA DFW-DEN FLL-BOS LAS-DEN MCO-IAD MSP-DEN ORD-TPA

ATL-FLL BOS-BWI DEN-IAD DEN-SAN DFW-LAS FLL-JFK LAX-ATL MCO-JFK MSP-MDW PHL-ATL

ATL-IAD BOS-IAD DEN-IAH DEN-SFO DFW-MSP FLL-LGA LAX-HNL MCO-MSP MSP-PHX PHL-BOS

ATL-LGA BOS-MCO DEN-LAX DEN-SJC DTW-LAS IND-DFW LAX-MSP MCO-ORD MSP-SFO PHX-DEN

ATL-MCO BOS-TPA DEN-LGA DEN-SLC DTW-LAX JFK-LAS LGA-DTW MDW-DFW MSY-ATL SAN-JFK

BDL-DCA BOS-LGA CLE-LGA DEN-ORD DFW-MCI DFW-SNA EWR-LAX EWR-TPA

BOS-DCA BOS-MIA CLE-ORD DFW-DCA DFW-MCO DFW-TPA EWR-MIA HPN-ORD

BOS-DFW BOS-ORD CLT-LGA DFW-DTW DFW-PHL EWR-DCA EWR-MCO IAD-DFW

Routes with both Southwest and other LCCs’ presence FLL-PHL PHL-LAX PHL-MCO

PHL-TPA

Other routes ATL-CVG BOS-DTW BOS-PIT CLT-ORD DFW-EWR DFW-PHX EWR-DTW EWR-ORD

ATL-ORD BOS-LAX CLE-EWR DEN-EWR DFW-LGA DFW-SFO EWR-LAS EWR-SFO

ATL-DTW BOS-EWR BOS-SFO CVG-LGA DFW-FLL DFW-SAN EWR-FLL EWR-PBI

ATL-IAH BOS-IAH BWI-ORD CVG-ORD DFW-LAX DFW-SEA EWR-IAH EWR-PHX

MANUEL A. HERNANDEZ ET AL.

Routes with Southwest presence BDL-MCO BWI-LAX OAK-LAX PHL-LAS SLC-LAX STL-LAX

IAD-LAX IAH-MSY LAX-MIA LGA-MIA LGA-TPA MSP-ORD ORD-IAH ORD-RDU PDX-SFO SJC-AUS

IAD-SFO IAH-TPA LAX-MCO LGA-MSP MCI-ORD OAK-DEN ORD-LAS ORD-SAN PHL-SFO SJC-DFW

IAH-DFW JFK-MIA LAX-ORD LGA-ORD MCO-DCA ORD-BDL ORD-MSY ORD-SEA PHL-PIT STL-DFW

IAH-LAS JFK-LAX LAX-TPA LGA-PBI MIA-ORD ORD-CMH ORD-OMA ORD-SFO SAN-SFO

IAH-LAX JFK-SFO LAX-SEA LGA-PIT MKE-MSP ORD-DCA ORD-PHL ORD-SNA SEA-SFO

IAH-LGA LAS-SFO LAX-SFO LGA-RDU MSP-DTW ORD-DFW ORD-PHX ORD-STL SNA-SFO

Note: Southwest and other LCCs are considered present on a route if they have 5 percent or more of the market share on the route. Airport codes: ATL, Atlanta; AUS, Austin; BDL, Hartford; BOS, Boston; BWI, Baltimore; CLE, Cleveland; CLT, Charlotte; CMH, Colombus; CVG, Cincinnati; DCA, Washington-National; DEN, Denver; DFW, Dallas-Ft. Worth; DTW, Detroit; EWR, Newark; FLL, Fort Lauderdale; HNL, Honolulu; HPN, NY-White Plains; IAD, Washington-Dulles; IAH, Houston; IND, Indianapolis; JFK, NY-Kennedy; LAS, Las Vegas; LAX, Los Angeles; LGA, NY-La Guardia; MCI, Kansas City; MCO, Orlando; MDW, Chicago-Midway; MEM, Memphis; MIA, Miami; MKE, Milwaukee; MSP, Minneapolis-St. Paul; MSY, New Orleans; OAK, Oakland; OMA, Omaha; ORD, Chicago-O Hare; PBI, West Palm Beach; PDX, Portland; PHL, Philadelphia; PHX-Phoenix; PIT, Pittsburgh; RDU, Raleigh-Durham; SAN, San Diego; SEA, Seattle; SFO, San Francisco; SJC, San Jose; SLC, Salt Lake City; SMF, Sacramento; SNA, Santa Ana; STL, St. Louis; TPA, Tampa.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

IAD-LAS IAH-MCO LAX-FLL LGA-DCA LGA-STL MSP-EWR ORD-DTW ORD-PIT PBI-BOS SNA-SLC

19

20

MANUEL A. HERNANDEZ ET AL. By ticket type 160 140

cents per mile

120 97

100 80

65

60 40

27

17

20 0 Group F

Group 1

Group 2

Group 3

By ticket type and carrier 160 140

128

cents per mile

120 80

80 60 40 20

56

77 65

66

72

52

47 25 17

98

93

91

100

30 19

24 14

25 16

32 20

32 23

0 American

United Group F

Delta Group 1

Continental USAirways Northwest Group 2

Group 3

Fig. 1. Average Fare per Mile by Ticket Type and Carrier. Group F: First-Class Tickets; Group 1: Refundable Business, Full Coach, and Coach Tickets; Group 2: Nonrefundable Tickets without Travel or Stay Restrictions; Group 3: Nonrefundable Tickets with Travel and/or Stay Restrictions.

Luengo-Prado, 2005; Stavins, 2001). In addition to controlling for the presence of Southwest and other LCCs on a route, we also control for hub presence, carriers’ market share, Herfindahl-Hirschman Index (HHI) as a measure of competitiveness on a route, slot-controlled airports, distance, flight frequency, per capita income, temperature difference, and tourism index.10 To identify the presence of Southwest and other LCCs on a route, we use the T-100 Domestic Segment Database from BTS. This data set contains domestic nonstop segment data reported by all U.S. carriers and is

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

21

consistent with our matched transaction data based on direct flights. Market share and the HHI are also computed using this data. Considering the importance of Southwest in the industry, its effect on the average price on a route, and also the effect on competitors’ pricing, and based on the results documented by Morrison (2001), Lee and LuengoPrado (2005), Sengupta and Wiggins (2006), Goolsbee and Syverson (2008), Brueckner et al. (2010), and Hernandez and Wiggins (2010), we separately account for the presence of Southwest and other LCCs on a route. The other LCCs include AirTran, Allegiant, ATA, Frontier, Independence, JetBlue, Primaris, Spirit, and Sun Country. We acknowledge Southwest’s presence on a route only if it has a market share of 5 percent or more on the route. The 5 percent threshold is to account for highly circuitous routes and potential data errors. Similarly, we identify other LCCs’ presence on a route if they collectively have 5 percent or more of the market share on the route. We also account for adjacent and potential competition from Southwest, assuming that this carrier (as well as other LCCs) also operates from secondary airports. We do not account for adjacent and potential competition from other LCC since Hernandez and Wiggins (2010) show, using a broader sample of routes (and carriers), that they do not have a major impact on both the level and structure of fares. As shown in Fig. 2 (on the left panel) and following Morrison (2001), given route or airport pair A-B, we define adjacent routes as C-B, A-D, and C-D, where airport C is located in the same city as airport A while airport D is located in the same city as airport B. In the case of potential competition, we follow the definition of Goolsbee and Syverson (2008), who define a route with a threat of entry as one where Southwest does not operate on the route but operates at the two endpoint airports. In Fig. 2 (on the right panel), there is potential competition from Southwest on route A-B since the carrier does not operate the route but operates in both airport A and airport B, for example, in routes A-E, A-D, B-F, and B-E. As in the case of direct competition, we use a 5 percent threshold to identify Southwest’s presence on an adjacent route as well as Southwest’s airport presence at each endpoint. Table 2 presents the summary statistics of all the variables used in the analysis. The average fare paid for a roundtrip in one of the six legacy carriers is 483 dollars or 33.4 cents per mile. First class (Group F) tickets represent 6 percent of the sample, Group 1 tickets 18 percent, Group 2 tickets 29 percent, and Group 3 tickets the remaining 47 percent. Around 61 percent of the tickets are bought less than two weeks prior to departure (25 percent in the last three days), 73 percent of the tickets are for roundtrip travel, 65 percent of the itineraries are during peak times (defined as Monday through Friday between 7–10 am and 3–7 pm), and only 10 percent of the tickets are bought on

22

MANUEL A. HERNANDEZ ET AL. Direct and adjacent competition on A-B

A

Potential competition on A-B

A

C

d E

a

e

b c F

B

a b, c, d

Fig. 2.

B

D

= direct competition = adjacent competition

e

D

= potential competition

Identifying Direct, Adjacent, and Potential Competition on Route A-B.

the internet. In terms of the distribution of tickets by carrier, 32 percent of the tickets in the sample correspond to American, 23 percent to United, 17 percent to Delta, 13 percent to Continental, 9 percent to US Airways, and 6 percent to Northwest.11 About 83 percent of the itineraries involve travel to/ from a hub of the operating carrier. The average market share of the operating carrier is almost 60 percent and the average concentration ratio or HHI on a route is 0.57. Finally, for 5 percent of the itineraries (9 percent of the routes) there is direct competition from Southwest, while for 32 percent of the itineraries (35 percent of the routes) there is direct competition from other LCCs. In the case of adjacent and potential competition from Southwest, for 12 percent of the itineraries (11 percent of the routes) there is adjacent competition from Southwest, and for 2 percent of the itineraries (3 percent of the routes) there is potential competition from this carrier.

Preliminary Analysis In general, the data suggests that Southwest’s presence on a route has, on average, an important effect on both the level and structure of fares charged

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

23

Table 2. Summary Statistics of Variables Used in the Analysis.

Fare (dollars) Fare per mile (cents)

Mean

St. Dev.

Min.

Max.

483 33.4

490 33.7

62 3.4

4,806 305.9

Dummies for ticket type Group F Group 1 Group 2 Group 3

0.06 0.18 0.29 0.47

0.24 0.39 0.45 0.50

0.00 0.00 0.00 0.00

1.00 1.00 1.00 1.00

Southwest and other LCCs’ presence Southwest on route Southwest on adjacent route Southwest potential entry LCC on route

0.05 0.12 0.02 0.32

0.22 0.32 0.14 0.47

0.00 0.00 0.00 0.00

1.00 1.00 1.00 1.00

Ticket and flight controls Adv0_3 Adv4_6 Adv7_13 Adv14_21 Adv22_over One-way Online purchase Mean deviation of load factor Peak time

0.25 0.14 0.22 0.16 0.23 0.27 0.10 0.01 0.65

0.43 0.35 0.41 0.37 0.42 0.44 0.30 0.19 0.48

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.97 0.00

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.23 1.00

0.83 0.59 0.57 0.22 1,016 2,862 38,797 9.15 0.01

0.37 0.25 0.20 0.42 664 1,413 3,474 6.68 0.02

0.00 0.00 0.19 0.00 185 341 31,811 0.10 0.00

1.00 1.00 1.00 1.00 2,704 6,576 48,150 26.70 0.13

0.32 0.23 0.17 0.12 0.09 0.06

0.47 0.42 0.38 0.33 0.29 0.24

0.00 0.00 0.00 0.00 0.00 0.00

1.00 1.00 1.00 1.00 1.00 1.00 755,842

Carrier and market controls Hub for carrier Carrier market share HHI Slot-controlled airport Distance Total flights on route Per capita income (dollars) Temperature difference Tourism index Main carriers American United Delta Continental US Airways Northwest # Observations

24

MANUEL A. HERNANDEZ ET AL.

by legacy carriers. As shown in Table 3, the average fare per mile decreases from 34 cents in routes where Southwest does not operate to 22 cents in routes where Southwest operates. The decrease in fares is uniformly observed across all ticket groups: Group F fares decrease from 96 to 62 cents, Group 1 fares from 60 to 26 cents, Group 2 fares from 29 to 19 cents, and Group 3 fares from 20 to 15 cents. The larger decrease of higher end fares relative to Group 3 fares (the lowest fares) results also in a compression of the entire fare distribution. The lower section of Table 3 shows that the ratio of Group F to Group 3 fares decreases from 5.6 to 4.7 as we move from routes where Southwest is not present to routes where there is direct competition from Southwest. The ratio of Group 1 to Group 3 fares, in turn, decreases from 3.6 to 1.9, while the ratio of Group 2 to Group 3 fares decreases from 1.5 to 1.2.

Table 3.

Absolute and Relative Fares per Mile by Ticket Type and Southwest and Other LCCs’ Presence. Southwest (SW) on Route

Southwest (SW) on Adjacent Route

Southwest (SW) Potential Entry

SW

SW

No SW

SW

No SW

LCC

No LCC

120.5 40.4 33.9 19.6 30.4

92.7 60.5 27.6 19.5 33.9

39.2 18.2 17.4 8.2 11.5

96.0 56.5 28.5 19.8 33.9

91.9 58.1 23.6 17.0 29.3

100.0 55.3 31.8 20.7 35.4

No SW

A. Average fare per mile Group F 61.5 Group 1 26.4 Group 2 18.8 Group 3 14.7 Total 22.1 Mean test Pr(|T|W|t|) 0.00

(in cents) 96.0 60.0 28.5 19.8 34.1

0.00

B. Relative fare per mile (relative to Group 3) Group F 4.7 5.6 6.8 Group 1 1.9 3.6 2.7 Group 2 1.2 1.5 1.8

0.00 5.5 3.6 1.5

5.5 2.5 2.1

Other LCCs on Route

0.00 5.5 3.3 1.5

5.9 3.6 1.4

5.7 3.3 1.6

Note: Group F: first-class tickets; Group 1: refundable business, full coach, and coach tickets; Group 2: nonrefundable tickets without travel or stay restrictions; Group 3: nonrefundable tickets with travel and/or stay restrictions. Southwest and other LCCs are considered present on a route if they have 5 percent or more of the market share on the route. A similar 5 percent threshold is used to identify adjacent and potential competition from Southwest on a route. Absolute and relative fares are a weighted average of fares by flying distance for routes with and without Southwest and other LCCs’ presence.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

25

Table 3 also suggests that potential competition from Southwest has a significant effect over all fare types, but not all higher end fares are forced down toward Group 3 fares. In particular, on routes without a threat of entry from Southwest, the average fare per mile priced by major carriers is roughly 40 cents, while on routes with a threat of entry the fare per mile is less than 12 cents. By fare type, Group F through Group 3 fares decrease from 96 to 39 cents, 57 to 18 cents, 29 to 17 cents, and 20 to 8 cents, respectively. In terms of relative prices, however, we only observe a decrease in the ratio of Group 1 to Group 3 fares (from 3.3 to 2.5). Adjacent competition from Southwest and direct competition from other LCCs affect the average level of fares. On routes with adjacent competition from Southwest, the average fare per mile is 30 cents as compared to 34 cents on routes without adjacent competition. Similarly, on routes where legacy carriers face direct competition from other LCCs, the average fare per mile is 29 cents, about 6 cents lower than on routes where they face no competition from these carriers. A closer look at the data confirms that Southwest’s presence on a route affects the fare structure of legacy carriers. Fig. 3, for example, presents Delta’s average relative prices, by fare type and day of purchase (prior to departure), for two of their largest routes in the sample, with similar distances and levels of concentration. Salt Lake City (SLC)–Seattle (SEA) and Atlanta (ATL)–New York-La Guardia (LGA) represent short/mediumdistance routes and both exhibit medium levels of concentration, but there is only direct competition from Southwest on SLC-SEA route. It is clear that the compression of the fare structure in routes with direct competition from Southwest is independent of the time of purchase, at least for some airlines.

EMPIRICAL ESTIMATION We now turn to a more formal examination of the effect of Southwest and other LCCs on both the level and fare structure of legacy carriers. We are particularly interested in measuring the competitive effects of both Southwest and other LCCs, including adjacent and potential competition from Southwest, on the nonlinear pricing behavior of the major carriers. The econometric analysis requires a careful accounting of the impact of these LCCs on the relative pricing of the different fare types while controlling for several cost and market-specific factors.

26

MANUEL A. HERNANDEZ ET AL.

The Empirical Model We implement a regression analysis at the ticket level to separate the effect of Southwest and other LCCs on fares from cost and other market factors that could also explain the pricing behavior of legacy carriers. We use similar controls at the ticket, flight, and market level as are used in Hernandez and Wiggins (2010). The idea is to account more accurately for

SLC-SEA (SW presence) 10

ratio

8 6 4 2 0 0

3

6

9

12

15

18

21

18

21

Days prior to departure Group F / Group 3 Group 1 / Group 3 Group 2 / Group 3

ATL-LGA (No SW presence) 10

ratio

8 6 4 2 0

0

3

6

9

12

15

Days prior to departure Group F / Group 3

Group 1 / Group 3 Group 2 / Group 3

Fig. 3. Delta Relative Fares by Ticket Type, Day of Purchase, and Southwest Presence. Group F: First-Class Tickets; Group 1: Refundable Business, Full Coach, and Coach Tickets; Group 2: Nonrefundable Tickets without Travel or Stay Restrictions; Group 3: Nonrefundable Tickets with Travel and/or Stay Restrictions. Relative Fares are the Ratio of Average Fares per Mile for a Given Day prior to Departure. Southwest is Considered Present on a Route if It has 5 Percent or More of the Market Share on the Route.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

27

possible differences in costs, particularly shadow costs, across fares as well as for differences in market characteristics, such as relative demands for different fare types, across routes. In particular, we model fares as a function of group dummies for fare type, Southwest and other LCCs’ presence on a route, and a vector of controls. The group dummies capture the quality premia of the different ticket groups over the lowest priced Group 3 tickets. We estimate two fare models. In the first model, we assume that LCCs may affect the fare level but not the fare structure or quality premia captured through the group dummies. In the second model, we allow the quality premia to be affected by the presence of Southwest and other LCCs on a route, so the group dummies are interacted with the variables indicating the presence of LCCs on the route. To distinguish between both specifications, hereafter we refer to the first model as the no-interaction model and to the second model as the interaction model. The log-linear fare per mile equation of the no-interaction model is defined as ln pijkt ¼ b0 þ

2 X

bf 1 qf i þ bc2 SWcj þ b3 LCCj þ X ijkt l þ w1j þ f1k þ eijkt (1)

f ¼F

where pijkt is the fare per mile of ticket i charged by legacy carrier j on route k at time t; qfi is a dummy variable for Group f fare, f ¼ F,y,2; SWcj , c ¼ fd; a; pg, are dummy variables representing direct (d), adjacent (a), and potential (p) competition from Southwest on a route; LCCj is a dummy variable indicating the presence of other LCCs a route; and Xijkt is a vector of ticket, flight, and market controls. The error term is assumed to have a carrier effect w1j , a random route effect j1k common to all carriers on a route, and a white noise error eijkt . The parameters of interest are bc2 and b3 , which capture the average effect of Southwest and other LCCs on fares. The log-linear fare per mile equation of the interaction model is defined as ln pijkt ¼ d0 þ þ

2 P

f ¼F

2 X

df 1 qf i þ dc2 SWcj þ

2 P

f ¼F

dcf 3 ðqf i  SWcj Þ þ d4 LCCj

df 5 ðqf i  LCCj Þ þ X ijkt g þ w2j þ f2k þ uijkt

(2)

f ¼F

The parameters of interest in this second model are dc2 , dcf 3 , d4 , and df 5 , which approximate the impact of Southwest and other LCCs on both the base fares and the entire fare structure. More specifically, the sign of dcf 3 and

28

MANUEL A. HERNANDEZ ET AL.

df 5 indicates if the premia of the different group fares, over Group 3 fares, vary with the presence of Southwest and other LCCs on a route. Any variation of these premia will suggest that legacy carriers modify their nonlinear pricing strategy when they face competition from LCCs. As indicated above, the vector of controls is intended to account for costand market-specific effects other than the presence of LCCs that may affect airlines’ pricing behavior. A detailed discussion of these controls is found in Hernandez and Wiggins (2010). In addition to ticket-specific characteristics like one-way tickets, we control for scarcity factors since prices may vary with load factor (Dana, 1998, 1999a, 1999b; Gale & Holmes, 1993). The scarcity variables include date of purchase as compared to departure, the mean deviation of the load factor at purchase, and whether the itinerary involves departure and/or return during peak periods. We also distinguish online purchases from offline purchases given that tickets purchased online are generally cheaper. The market controls, in turn, are widely used market variables for hub presence, carrier market share, market concentration (HHI), slot-controlled airports, distance, flight frequency, average per capita income at endpoint cities, the absolute temperature difference between the origin and destination, and a tourism index defined as the ratio of accommodation to personal income at the destination city. Airlines, for example, seem to charge higher prices for service to and from their hub airports (Borenstein, 1989; Lee & Luengo-Prado, 2005), while slotcontrolled airports are supposed to raise the costs of serving a market. The level of concentration in a market, despite the presence of LCCs on the route, may also affect the pricing behavior of firms. The absolute temperature difference and the tourism index should account for potential tourist effects (fraction of leisure travelers on a route) and, consequently, for differences in the relative demand for different fare types (Borenstein, 1989; Borenstein & Rose, 1994; Brueckner et al., 1992; Stavins, 2001). Finally, it is worth noting that some of the explanatory variables in Eqs. (1) and (2) are likely to be endogenous, particularly the carrier market share and route HHI. Market share is potentially endogenous because it is a function of price and is likely correlated with the error term. If market share is endogenous, HHI is also endogenous. We use the same instruments as Borenstein (1989) and Borenstein and Rose (1994) to instrument these variables. The carrier market share is instrumented with the carrier’s enplanement share at the two endpoint airports, while the HHI is instrumented with the square of the fitted value of market share (from its first-stage regression) plus the rescaled sum of the square of all other carriers’ share on the route.12

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

29

We also recognize the potential endogeneity of the dummy variables indicating the presence of Southwest (and other LCCs) on a route due to the potential endogeneity of their entry: the entry of Southwest on a route could be correlated with unobservable route characteristics that also determine airline fares. Unfortunately, we are unaware of appropriate instruments for these variables.13

Estimation Results Base Estimations Table 4 presents the results of estimating Eqs. (1), the no-interaction model, and (2), the interaction model, by both ordinary (OLS) and two-stage least squares (2SLS). The 2SLS approach addresses the potential endogeneity of the carrier market share, route HHI, and flight frequency. Carrier effects are treated as fixed and route effects as random in order to account for routespecific effects (like LCCs’ presence) in the estimations. The standard errors reported are robust, clustered on the route. Note that Kleibergen and Paap’s (2006) LM under-identification test and Wald weak-identification test indicate that the instruments used are not weakly correlated with the potentially endogenous variables (carrier’s market share, route HHI, and flight frequency).14 As shown at the bottom of Table 4, the LM under-identification test rejects at the 1 percent level of significance the null hypothesis that the excluded instruments are not correlated with the suspected endogenous variables. The Wald F weakidentification test shows that the correlation between the instruments and the instrumented variables is not weak. The estimated coefficients of the control variables are very similar across the two models and generally have the expected signs. For clarity of exposition, we focus on the 2SLS results of the no-interaction model. The estimation results indicate that online purchases, time of purchase, and oneway tickets have an important effect on the prices charged by legacy carriers. Tickets bought on the internet are, on average, 25 percent less than tickets bought offline; tickets purchased closer to departure time (0–6 days prior to departure) are between 16 and 21 percent more expensive than tickets bought well in advance (more than 21 days in advance), while one-way tickets are 14 percent more expensive than half the price of roundtrip fares. The flight scarcity variables are statistically significant, but their economic magnitudes are small. A one standard deviation increase in the deviation of the load factor from its average at the time of ticket purchase (0.19) only

30

Log of Fare per Mile Regressions.

Table 4.

No-Interaction Model OLS

2SLS

Coeff.

Std. Err.

1.582 0.779 0.299 0.387 0.315 0.260 0.134

0.035 0.070 0.027 0.055 0.058 0.068 0.035

0.189 0.158 0.131 0.068 0.170

0.025 0.024 0.021 0.014 0.019

Coeff.

Std. Err.

OLS Coeff.

Std. Err.

Dependent variable: log of fare per mile 1.674 0.037 1.556 0.052 0.792 0.079 1.019 0.062 0.285 0.025 0.318 0.030 0.541 0.087 0.122 0.044 0.381 0.084 0.191 0.040 0.610 0.123 0.190 0.084 0.141 0.045 0.113 0.038 0.232 0.125 0.654 0.088 0.218 0.044 0.234 0.068 0.734 0.076 0.152 0.061 0.267 0.118 0.486 0.121 0.019 0.075 0.085 0.050 0.072 0.131 0.036 0.052 0.212 0.027 0.199 0.024 0.165 0.025 0.174 0.024 0.130 0.021 0.124 0.019 0.068 0.014 0.069 0.013 0.144 0.023 0.155 0.017

2SLS Coeff.

Std. Err.

1.610 1.031 0.309 0.300 0.197 0.522 0.133 0.401 0.581 0.194 0.152 0.863 0.070 0.211 0.480 0.064 0.190 0.021 0.027 0.226 0.186 0.124 0.068 0.128

0.047 0.064 0.031 0.091 0.061 0.111 0.041 0.152 0.078 0.050 0.088 0.092 0.074 0.123 0.172 0.060 0.066 0.122 0.056 0.025 0.024 0.018 0.013 0.022

MANUEL A. HERNANDEZ ET AL.

Group F Group 1 Group 2 Southwest on route Southwest on adjacent route Southwest potential entry LCC on route Group FSouthwest on route Group 1Southwest on route Group 2Southwest on route Group FSouthwest adjacent route Group 1Southwest adjacent route Group 2Southwest adjacent route Group FSouthwest potential entry Group 1Southwest potential entry Group 2Southwest potential entry Group FLCC on route Group 1LCC on route Group 2LCC on route Adv0_3 Adv4_6 Adv7_13 Adv14_21 One-way

Interaction Model

0.240 0.136 0.021

0.010 0.016 0.006

Hub for carrier Market share HHI Slot-controlled airport Log distance Log total flights on route Log per capita income Log temperature difference Tourism index American United Delta Continental US Airways Constant

0.194 0.069 0.124 0.055 0.706 0.010 0.086 0.004 1.203 0.104 0.052 0.201 0.062 0.229 8.283

0.051 0.080 0.094 0.030 0.023 0.030 0.167 0.020 0.409 0.069 0.072 0.065 0.087 0.082 1.805

Under-identification test Kleibergen-Paap rk LM stat. Chi-sq(1) P-val Weak identification test Kleibergen-Paap rk Wald F stat. # Observations R-squared

0.251 0.012 0.232 0.010 0.129 0.017 0.134 0.015 0.016 0.009 0.019 0.006 Dependent variable: log of fare per mile 0.397 0.115 0.217 0.047 0.071 0.122 0.051 0.078 0.841 0.356 0.064 0.089 0.150 0.052 0.036 0.028 0.904 0.066 0.700 0.024 0.681 0.205 0.019 0.028 0.080 0.249 0.152 0.166 0.017 0.020 0.013 0.021 1.477 0.631 1.391 0.363 0.100 0.087 0.095 0.063 0.161 0.072 0.040 0.068 0.017 0.109 0.194 0.060 0.034 0.104 0.094 0.081 0.030 0.101 0.296 0.072 13.450 3.498 9.037 1.733 10.21 (0.001)

755,842 0.790

7.02 755,842 0.720

0.243 0.128 0.014

0.011 0.016 0.008

0.405 0.091 0.808 0.127 0.889 0.660 0.028 0.028 1.671 0.090 0.143 0.004 0.077 0.118 14.327

0.108 0.124 0.321 0.047 0.062 0.195 0.233 0.018 0.562 0.082 0.065 0.099 0.096 0.094 3.331 10.12 (0.001)

755,842 0.808

6.40 755,842 0.746

31

Note: Fare per mile ¼ roundtrip fare (in cents)/(2  nonstop origin to destination mileage). White robust standard errors reported, clustered on route. Market share and HHI instrumented using the same instruments as Borenstein (1989) and Borenstein and Rose (1994). Log of total flights instrumented with the log of population. The under- and weak identification tests for the instruments are the LM and Wald versions of the Kleibergen and Paap (2006) rk statistic and are heteroskedastic-robust.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

Online purchase Mean deviation of load factor Peak time

32

MANUEL A. HERNANDEZ ET AL.

increases fares per mile by 2.4 percent, while tickets that involve travel during peak periods are less than 2 percent more expensive than those during off-peak periods. Most of the effects of the market controls are also consistent with the literature on airline pricing. For example, fares per mile for travel to and/or from hub airports of the operating carrier are almost 40 percent higher than fares for service not involving a carrier’s hub. Route distance and flight frequency decrease the average fare per mile, while slotcontrolled airports and a higher per capita income at the endpoint cities increase fares. Regarding the tourist effect variables, which approximate the proportion of tourist travelers on a route, only the tourism index has a statistically significant negative effect on prices, although economically small. The results also indicate, however, that the average fare per mile decreases with market concentration.15 The no-interaction model also confirms the asserted price/quality differences of the different ticket groups found by Hernandez and Wiggins (2010) using a broader sample of routes (and carriers). The quality premium over Group 3 fares declines progressively as we move from Group F through Group 2 fares. The corresponding relative premia are 167, 79, and 29 percent. Turning to the variables of interest, we find that Southwest has a much larger negative effect on the prices charged by legacy carriers than the other LCCs. On routes with direct competition from Southwest, the fare per mile is 54 percent lower than on routes where Southwest does not operate, while on routes with direct competition from other LCCs, fares are 14 percent lower. Potential and adjacent competition from Southwest also significantly reduces prices. In particular, on routes where legacy carriers face a threat of entry from Southwest, the fare per mile is 61 percent lower than on routes without a threat of entry. On routes with adjacent competition from Southwest, the fare per mile is 38 percent lower than on routes without adjacent competition. These results confirm the importance of Southwest in the industry, specifically its effect on competitors’ pricing, and are qualitatively similar to the results found in previous studies, which analyze the impact of Southwest on airline fares.16 Relative to the average effect of other LCCs on the fare per mile, the (negative) effect of direct and potential competition from Southwest on fares is roughly four times larger, while the effect of adjacent competition from Southwest is roughly three times larger. Moving to the effect of LCCs’ competition on the quality premium of the different group of fares (over Group 3 fares), the results of the interaction model reveal that Southwest’s direct competition significantly affects the

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

33

structure of fares of legacy carriers (see Table 4). Group 3 fares fall by 30 percent with Southwest’s presence on the route, and the relative premia of all other fare groups also fall substantially. The impact of Southwest on the fare structure is more easily seen in Table 5, which presents the estimated quality premia of the different groups of fares, over Group 3 fares, for routes with and without Southwest (and other LCC) presence. The estimated relative premium of Group F fares decreases by 41 percentage points (from 169 to 128 percent) with direct competition from Southwest, the relative premium of Group 1 fares decreases by 58 percentage points (from 91 to 33 percent), and that of Group 2 fares decreases by 20 percentage points (from 31 to 11 percent). It is interesting to observe that Southwest’s presence on a route seems to affect, on average, the entire fare distribution of the major carriers and not just the prices of the lowest fares. The results also indicate that potential competition from Southwest affects both the base fares (they decrease by 52 percent) and most of the fare structure. Table 5 shows that the premium of Group F over Group 3 fares decreases by 21 percentage points (from 167 to 146 percent) with potential competition from Southwest, while the relative premium of Group 1 fares decrease by 48 percentage points (from 89 to 41 percent). The relative premium of Group 2 fares does not seem to be affected by a threat of entry from Southwest.

Table 5.

Group F Group 1 Group 2

Predicted Quality Premia of Various Fares to Group 3 by Southwest and Other LCCs’ Presence. Other LCCs on Route (%)

Southwest (SW) on Route (%)

Southwest (SW) on Adjacent Route (%)

Southwest (SW) Potential Entry (%)

SW

No SW

SW

No SW

SW

No SW

LCC

No LCC

128 33 11

169 91 31

180 12 36

165 98 29

146 41 23

167 89 30

179 87 28

160 89 31

Note: Group F: first-class tickets; Group 1: refundable business, full coach, and coach tickets; Group 2: nonrefundable tickets without travel or stay restrictions; Group 3: nonrefundable tickets with travel and/or stay restrictions. Southwest and other LCCs are considered present on a route if they have 5 percent or more of the market share on the route. A similar 5 percent threshold is used to identify adjacent and potential competition from Southwest on a route.

34

MANUEL A. HERNANDEZ ET AL.

As in the preliminary analysis, adjacent competition from Southwest and direct competition from other LCCs do not appear to compress the fare distribution. Adjacent competition from Southwest decreases base fares by almost 20 percent and significantly decreases the premium of Group 1 over Group 3 fares by 86 percentage points (from 98 to 12 percent); however, it also slightly increases the relative premium of Group F fares by 15 percentage points (from 165 to 180 percent). Direct competition from other LCCs decreases base fares by 13 percent, but the relative premia of the different fare types do not generally change with the presence of other LCCs on the route; the premium of Group F over Group 3 fares actually increase by 19 percentage points (from 160 to 179 percent) with the presence of other LCCs. In sum, two interesting patterns emerge from the analysis above. First, after controlling for cost- and market-specific factors, Southwest seems to have an important effect on both the level and fare structure of legacy carriers, particularly when these major carriers face actual and potential competition from Southwest. These results closely match the results in Hernandez and Wiggins (2010), which use a broader sample or routes and carriers. Second, adjacent competition from Southwest and direct competition from other LCCs basically only affect the level of fares but have a limited impact on the fare structure. Later we examine if these results are common across all major carriers. Alternative Estimations We now perform alternative estimations to examine the robustness of our results. In particular, we test if our main results hold if we modify the 5 percent threshold to identify Southwest and other LCC presence on a route. Similarly, we consider an alternative data source that includes both direct and connecting service to identify LCCs’ presence on a route and examine if the results are sensitive to the data source used. Alternative Thresholds to Identify Market Presence. We modify the 5 percent threshold to acknowledge Southwest and other LCCs’ presence on a route. We consider both a less restrictive threshold (1 percent) and a more restrictive one (10 percent). These thresholds are similar to the ones used by Lee and Luengo-Prado (2005) when accounting for the presence of LCCs on a route. The estimation results are reported in Table A.1 (for the 1 percent threshold) and Table A.2 (for the 10 percent threshold). It is clear that the results remained unchanged when using these alternative thresholds. The estimated coefficients of the control variables are very similar to those in

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

35

Table 4 when using a 5 percent threshold, and the impact of Southwest and other LCCs on the fares of legacy carriers is basically unchanged. Using a 1 percent threshold, direct competition from Southwest reduces fares by 54 percent, potential competition by 55 percent, adjacent competition by 40 percent, and direct competition from other LCCs by 15 percent. In the case of the 10 percent threshold, these figures are 55, 65, 37, and 15 percent. Regarding the impact on the fare structure, Southwest’s direct presence on a route reduces both base fares and the premia of the different ticket groups, over Group 3 tickets, when using either a 1 or a 10 percent threshold. This can be more easily seen in the upper and middle panels of Table 6, which present the predicted relative premia of the different groups of fares for routes with and without Southwest (and other LCC) presence using these alternative thresholds. For the 1 percent threshold, the decrease in the premia of Group F

Table 6. Predicted Quality Premia of Various Fares to Group 3 by Southwest and Other LCCs’ Presence (Alternative Estimations). Southwest (SW) on Route (%)

Southwest (SW) on Adjacent Route (%)

SW

SW

No SW

LCC

No LCC

A. Using a 1 percent threshold to identify market presence Group F 127 169 180 165 150 Group 1 34 92 11 99 45 Group 2 14 31 35 30 18

167 90 31

179 87 26

160 89 33

B. Using a 10 percent threshold to identify market presence Group F 133 168 179 165 150 Group 1 32 89 10 97 27 Group 2 10 30 35 28 28

167 88 29

181 84 28

161 88 30

C. Using DB1B dataset Group F 128 170 Group 1 30 94 Group 2 10 31

167 89 30

172 90 25

164 89 32

166 100 28

SW

Other LCCs on Route (%)

No SW

175 14 39

No SW

Southwest (SW) Potential Entry (%)

161 112 42

Note: Group F: first-class tickets; Group 1: refundable business, full coach, and coach tickets; Group 2: nonrefundable tickets without travel or stay restrictions; Group 3: nonrefundable tickets with travel and/or stay restrictions. In Panels A and B, a 1 and 10 percent threshold is used to identify Southwest and other LCCs’ presence on a route. In Panel C, Southwest and other LCCs’ presence on a route, and market structure measures are derived from DB1B dataset.

36

MANUEL A. HERNANDEZ ET AL.

through Group 2 fares, over Group 3 fares, are of 42, 58, and 17 percentage points, respectively. For the 10 percent threshold, the corresponding changes are of 35, 57, and 20 percentage points. Potential competition from Southwest also decreases the base fares of legacy carriers and compresses most of the fare structure by forcing down the high-end fares toward the base fares. More specifically, with a threat of entry from Southwest, the changes in the relative premia of Group F through Group 2 fares are of 17, 47, and 12 percentage points in the case of the 1 percent threshold and of 17, 61, and 2 percentage points in the case of the 10 percent threshold. As in our base results, adjacent competition from Southwest and direct competition from other LCCs affect the base fares but do not compress the fare structure. Alternative Data Source to Identify Market Presence. The data source that we used to identify the presence of LCCs on a route (as well as to derive the market structure measures) is the T-100 Domestic Segment Database, which only accounts for nonstop service; however, it is largely compatible with our working sample based on direct flights. We now use an alternative data source, the widely used DB1B from BTS, which is a 10 percent random sample of all domestic U.S. tickets. We account for both direct and connecting service (up to one stop as in previous studies) to identify Southwest and other LCCs’ presence and to derive the other market structure measures on a route. The estimation results using DB1B to identify market presence on a route are presented in Table A.3. The results show that the effect of the control variables and the variables of interest are qualitatively similar to our base results. Direct competition from Southwest reduces fares by 50 percent, potential competition by 22 percent, and adjacent competition by 34 percent. Direct competition from other LCCs, in turn, reduces fares by 10 percent. Turning to the impact on the fare structure, Southwest’s presence on a route affects both base fares and the relative premia of the different fare groups. As shown in the bottom panel of Table 6, the estimated premia of Group F through Group 2 fares, over Group 3 fares, decrease by 42, 64, and 22 percentage points, respectively. Potential competition from Southwest also decreases base fares but does not compress the fare structure in this case. The same occurs with adjacent competition from Southwest and direct competition from other LCCs. In general, the alternative estimations performed support our main findings in that Southwest competition, including (in most cases) potential competition, has an important effect on both the fare level and structure. Adjacent competition from Southwest and direct competition from other LCCs mainly affect the fare level.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

37

Distinguishing Effects by Carrier We now turn to an alternative specification that accounts for potential differences by carrier in the direct impact of Southwest and other LCCs on the fare level and structure.17 Our base results capture the overall effect of Southwest and other LCCs on the pricing behavior of legacy carriers, but these effects are not necessarily recurrent across all of these carriers. We interact the dummy variables for Southwest’s and other LCC’s presence on a route and the interactions of these dummy variables with the group dummies, with the carrier dummies (Northwest being the base group). Table 7 reports the estimation results of this alternative specification. The results of the no-interaction model show that direct competition from Southwest decreases the fares of all carriers. American, for example, reduces fares by 49 percent, United by 34 percent, Delta by 27 percent, US Airways by 19 percent, and Northwest by 74 percent.18 The results also indicate that all legacy carriers except Delta reduce their fares when they face direct competition from other LCCs. The fare reductions range between 10 percent (for United) and 49 percent (for Northwest). Note that potential and adjacent competition from Southwest also reduces fares, although we are not able to distinguish effects by carrier for these two variables. In terms of the fare structure, the results for the interaction model in Table 7 reveal that major carriers behave differently when they face direct competition from Southwest and other LCCs. Direct competition from Southwest reduces the base fares of all carriers, but only Delta and American also force Group F fares toward the base fares, as shown in Table 8. In the case of Delta, the premium of Group F over Group 3 fares decreases by 57 percent, while in the case of American, the relative premium of Group F fares decreases by 21 percent. However, all carriers seem to decrease the relative premium of Group 1 fares when facing direct competition from Southwest. The corresponding reductions are of 55, 27, 91, 15, and 48 percentage points for American, United, Delta, US Airways, and Northwest. Similarly, all carriers except United decrease the ratio of Group 2 over Group 3 fares with Southwest’s presence on the route. It follows that the main differences in nonlinear pricing behavior across carriers is regarding the pricing of first-class (Group F) tickets relative to the lowest fares (Group 3 tickets) when Southwest operates on the route. As before, potential and adjacent competition from Southwest reduces, on average, the base fares but only potential competition additionally compresses most of the fare structure. The estimation results also show that major carriers behave differently when they face direct competition from other LCCs. All carriers except US Airways reduce their base fares with the presence of other LCCs on the route; however,

Log of Fare per Mile Regressions (Accounting for Potential Differences in the Impact of Southwest and Other LCCs’ Presence on the Route by Carrier). No-Interaction Model OLS Coeff.

Dependent variable: 1.584 0.035 0.774 0.068 0.300 0.027 0.646 0.099 0.312 0.058 0.226 0.063 0.407 0.056 0.157 0.115 0.309 0.110 0.377 0.147 0.459 0.113 0.252 0.061 0.417 0.068 0.340 0.055 0.203 0.074 0.010 0.093

Interaction Model

2SLS Coeff.

Std. Err.

log of fare per mile 1.664 0.036 0.784 0.077 0.276 0.025 0.743 0.116 0.392 0.088 0.585 0.140 0.486 0.062 0.051 0.119 0.279 0.120 0.191 0.159 0.385 0.122 0.255 0.063 0.387 0.056 0.593 0.121 0.241 0.055 0.255 0.139

OLS

2SLS

Coeff.

Std. Err.

Coeff.

Std. Err.

1.569 0.990 0.315 0.554 0.191 0.162 0.347 0.425 0.286 0.534 0.408 0.279 0.357 0.174 0.191 0.372 0.175 0.443 0.229 0.256 0.745 0.181 0.275

0.051 0.066 0.028 0.071 0.039 0.078 0.058 0.082 0.073 0.114 0.065 0.069 0.077 0.055 0.081 0.072 0.105 0.071 0.031 0.070 0.078 0.057 0.111

1.622 1.012 0.308 0.633 0.207 0.493 0.425 0.213 0.190 0.380 0.323 0.304 0.317 0.422 0.202 0.631 0.112 0.479 0.219 0.160 0.889 0.096 0.229

0.047 0.069 0.028 0.083 0.062 0.127 0.056 0.133 0.107 0.117 0.080 0.063 0.069 0.094 0.066 0.122 0.105 0.075 0.031 0.094 0.098 0.071 0.100

MANUEL A. HERNANDEZ ET AL.

Group F Group 1 Group 2 Southwest on route Southwest on adjacent route Southwest potential entry LCC on route Southwest on routeAmerican Southwest on routeUnited Southwest on routeDelta Southwest on routeUS Airways LCC on routeAmerican LCC on routeUnited LCC on routeDelta LCC on routeContinental LCC on routeUS Airways Group FSouthwest on route Group 1Southwest on route Group 2Southwest on route Group FSouthwest adjacent route Group 1Southwest adjacent route Group 2Southwest adjacent route Group FSouthwest potential entry

Std. Err.

38

Table 7.

Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group Group

FSouthwest on routeAmerican 1Southwest on routeAmerican 2Southwest on routeAmerican FSouthwest on routeUnited 1Southwest on routeUnited 2Southwest on routeUnited FSouthwest on routeDelta 1Southwest on routeDelta 2Southwest on routeDelta FSouthwest on routeUS Airways 1Southwest on routeUS Airways 2Southwest on routeUS Airways FLCC on routeAmerican 1LCC on routeAmerican 2LCC on routeAmerican FLCC on routeUnited 1LCC on routeUnited 2LCC on routeUnited FLCC on routeDelta 1LCC on routeDelta 2LCC on routeDelta FLCC on routeContinental 1LCC on routeContinental 2LCC on routeContinental FLCC on routeUS Airways

0.454 0.037 0.124 0.179 0.220

0.117 0.070 0.079 0.103 0.053

0.439 0.048 0.140 0.171 0.213

0.146 0.056 0.080 0.097 0.051

0.365 0.301 0.154 0.143 0.087 0.136 0.609 0.443 0.168 0.158 0.308 0.061 0.246 0.479 0.008 0.438 0.462 0.043 0.282 0.489 0.385 0.684 0.213 0.248 0.596

0.154 0.037 0.047 0.103 0.154 0.055 0.152 0.061 0.075 0.144 0.064 0.038 0.169 0.108 0.056 0.083 0.179 0.050 0.074 0.097 0.050 0.077 0.097 0.052 0.133

0.324 0.071 0.288 0.008 0.213 0.229 0.677 0.429 0.061 0.121 0.333 0.106 0.343 0.350 0.030 0.543 0.517 0.044 0.383 0.481 0.337 0.819 0.129 0.386 0.654

0.190 0.113 0.081 0.140 0.190 0.086 0.147 0.056 0.096 0.140 0.073 0.047 0.188 0.130 0.061 0.102 0.178 0.045 0.094 0.104 0.050 0.064 0.099 0.119 0.134

Dependent variable: log of fare per mile

39

1Southwest potential entry 2Southwest potential entry FLCC on route 1LCC on route 2LCC on route

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

Group Group Group Group Group

40

Table 7. (Continued ) No-Interaction Model OLS Coeff.

Std. Err.

Interaction Model

2SLS Coeff.

Std. Err.

Group 1LCC on routeUS Airways Group 2LCC on routeUS Airways Dependent variable: 0.193 0.025 0.162 0.024 0.134 0.020 0.069 0.013 0.170 0.018 0.235 0.010 0.133 0.016 0.021 0.006 0.200 0.051 0.029 0.087 0.162 0.090 0.052 0.029 0.703 0.024 0.001 0.030 0.044 0.159 0.005 0.021 1.230 0.419 0.259 0.077 0.164 0.074 0.414 0.070 0.216 0.084

log of fare per mile 0.219 0.026 0.172 0.024 0.132 0.020 0.070 0.014 0.142 0.025 0.247 0.012 0.127 0.017 0.016 0.009 0.369 0.109 0.131 0.126 0.817 0.393 0.151 0.054 0.911 0.074 0.690 0.230 0.053 0.238 0.015 0.021 1.459 0.634 0.084 0.096 0.052 0.093 0.360 0.093 0.219 0.106

2SLS

Coeff.

Std. Err.

Coeff.

Std. Err.

0.527 0.060

0.103 0.048

0.570 0.006

0.112 0.055

0.214 0.187 0.129 0.069 0.149 0.220 0.131 0.020 0.226 0.033 0.077 0.032 0.701 0.015 0.048 0.019 1.223 0.259 0.186 0.405 0.262

0.021 0.021 0.018 0.012 0.016 0.009 0.015 0.005 0.047 0.086 0.089 0.026 0.024 0.030 0.154 0.021 0.357 0.059 0.056 0.050 0.069

0.245 0.202 0.130 0.071 0.120 0.231 0.129 0.015 0.384 0.129 0.800 0.125 0.894 0.660 0.004 0.032 1.479 0.102 0.082 0.362 0.276

0.022 0.021 0.017 0.013 0.023 0.011 0.016 0.008 0.105 0.131 0.361 0.048 0.069 0.218 0.230 0.019 0.550 0.079 0.073 0.073 0.088

MANUEL A. HERNANDEZ ET AL.

Adv0_3 Adv4_6 Adv7_13 Adv14_21 One-way Online purchase Mean deviation of load factor Peak time Hub for carrier Market share HHI Slot-controlled airport Log distance Log total flights on route Log per capita income Log temperature difference Tourism index American United Delta Continental

OLS

0.354 7.909

0.086 1.723

0.249 14.103

0.113 3.911

0.431 8.086

0.060 1.601

10.60 (0.001)

755,842 0.794

7.25 755,842 0.724

0.342 14.269

0.091 3.778 10.44 (0.001)

755,842 0.821

7.03 755,842 0.760

Note: Fare per mile ¼ roundtrip fare (in cents)/(2  nonstop origin to destination mileage). White robust standard errors reported, clustered on route. Market share and HHI instrumented using the same instruments as Borenstein (1989) and Borenstein and Rose (1994). Log of total flights instrumented with the log of population. The under- and weak identification tests for the instruments are the LM and Wald versions of the Kleibergen and Paap (2006) rk statistic and are heteroskedastic-robust.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

US Airways Constant Under-identification test Kleibergen-Paap rk LM stat. Chi-sq. (1) P-val Weak identification test Kleibergen-Paap rk Wald F stat. # Observations R-squared

41

42

Table 8.

Predicted Quality Premia of Various Fares to Group 3 by Carrier and Southwest and Other LCCs’ Presence on the Route. American SW

No SW

A. Southwest (SW) on route (%) Group F 127 148 Group 1 41 96 Group 2 27 24

United

US Airways

Northwest

No SW

SW

No SW

SW

No SW

SW

No SW

SW

No SW

189 75 27

177 101 26

115 9 20

172 100 36

– – –

132 80 38

137 51 14

138 66 25

170 36 3

159 84 25

United LCC

No LCC

B. Other Low Cost Carriers (LCC) on route (%) Group F 114 162 205 164 Group 1 105 87 123 88 Group 2 5 29 15 32

Delta

Continental

US Airways

Northwest

LCC

No LCC

LCC

No LCC

LCC

No LCC

LCC

No LCC

185 116 43

161 85 31

68 57 48

164 87 31

84 15 11

164 89 31

150 70 9

164 87 31

Note: Group F: first-class tickets; Group 1: refundable business, full coach, and coach tickets; Group 2: nonrefundable tickets without travel or stay restrictions; Group 3: nonrefundable tickets with travel and/or stay restrictions. Southwest and other LCCs are considered present on a route if they have 5 percent or more of the market share on the route.

MANUEL A. HERNANDEZ ET AL.

No LCC

Continental

SW

American LCC

Delta

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

43

US Airways and Northwest (to a minor extent) are the only carriers that force down the premia of all group fares, relative to Group 3 fares, with direct competition from these other LCCs. The premia of Group F through Group 2 fares, over Group 3 fares, decrease by 79, 74, and 21 percentage points in the case of US Airways, and by 14, 17, and 21 percentage points in the case of Northwest. Recall that in our base estimations, direct competition from other LCCs does not compress the average fare structure of legacy carriers. Continental also appears to decrease the relative premium of Group F and Group 1 fares with the presence of other LCCs on the route, while American decreases the relative premium of Group F and Group 2 fares and United only decreases the relative premium of Group 2 fares.

CONCLUDING REMARKS This study adds further insights to the existing literature about the impact of LCCs on the pricing behavior of legacy carriers. We use a detailed transaction data set to build a menu of fare types according to ticket characteristics and restrictions and examine the effect of Southwest and other LCCs on the absolute and relative pricing of the different fare types. The estimation results indicate that competition from Southwest has an important effect on both the level and fare structure of the major carriers. In particular, it is interesting to observe that both direct and potential competition from Southwest lower fares and generally compress the entire fare structure by decreasing the premia of the highest fares, including firstclass tickets, over the lowest fares. Adjacent competition from Southwest and direct competition from other LCCs only appear to considerably affect the fare level. The results, however, cannot be generalized across all the legacy carriers as indicated by a differentiated analysis by carrier.

NOTES 1. In the literature, nonlinear pricing is synonymous to second-degree price discrimination. 2. Ticket restrictions act as fencing mechanisms. They induce travelers to reveal their individual preferences for different attributes by preselecting into a ‘‘bin,’’ defined by a set of ticket characteristics, and consequently to reveal their willingness to pay for those specific characteristics. Sengupta and Wiggins (2006) find that ticket characteristics along with carrier- and route-specific effects explain almost 80 percent of the variation in ticket prices.

44

MANUEL A. HERNANDEZ ET AL.

3. Hernandez and Wiggins (2010) find a similar result using a broader sample of routes and carriers. 4. See also Busse and Rysman (2002). 5. For a more detailed description on the matching procedure, please refer to Sengupta and Wiggins (2006), Puller, Sengupta, and Wiggins (2009), and Hernandez and Wiggins (2010). 6. Kernel densities of the fares of the matched transactions (our working sample) and the original data show that the matched tickets are generally representative of the original transactions from the major CRS. Results are available upon request. 7. We exclude travel on the Wednesday prior to Thanksgiving through the following Sunday, as well as all travel beginning on December 22 through the end of year. 8. In particular, roundtrip tickets of 20 dollars or less were dropped (10 dollars or less for one-way tickets). 9. Only in the case of Continental, Group F fares are, on average, below Group 1 fares. 10. For a full description of all these other variables, and the sources of information used to construct them, refer to Hernandez and Wiggins (2010). 11. It is worth noting that this distribution of tickets by carrier is very similar to the distribution derived from the number of passengers reported by each carrier in the T-100 nonstop segment database for the corresponding routes, although we have a slightly higher proportion of tickets from American in our sample. 12. See Borenstein’s (1989) and Borenstein and Rose’s (1994) discussion for the validity of these instruments. Similar to Borenstein and Rose (1994), we also account for the potential endogeneity of flight frequency (total number of flights on a route), and we instrument it with the average population at the two endpoint cities. 13. However, we still examined if the potential endogeneity of Southwest’s entry on a route could be affecting our results. We followed Hernandez and Wiggins’ (2010) approach, who argue that Southwest’s presence at each endpoint of a route is unlikely to be endogenous to demand on any specific route. The idea basically consists of replacing in the regression analysis the dummy variables for direct, potential, and adjacent competition from Southwest by a single dummy variable that takes the value of 1 if Southwest is present at both endpoint airports or operates on an adjacent route. The results obtained suggest that the potential endogeneity of Southwest’s entry on a route does not appear to be biasing our estimation results. Further details are available upon request. 14. The LM and Wald versions of the Kleibergen and Paap (2006) rk statistic are a generalization of the well-known Anderson’s LM test of canonical correlations and Cragg and Donald’s Wald test for weak identification for the case of non-i.i.d. errors. 15. Refer to Hernandez and Wiggins (2010) for a possible explanation for this result, which has also been found by Borenstein (1989) and Stavins (2001). 16. See, for example, Morrison (2001), Lee and Luengo-Prado (2005), Sengupta and Wiggins (2006), Goolsbee and Syverson (2008), Brueckner et al.(2010), and Hernandez and Wiggins (2010). 17. Unfortunately, due to the lack of sufficient observations by fare group, carrier, and route type, we are not able to evaluate differentiated effects across carriers for adjacent and potential competition from Southwest.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

45

18. We are not able to derive a differentiated effect for Continental since we do not observe in our working sample routes where Continental operates and faces direct competition from Southwest.

REFERENCES Alderighi, M., Cento, A., Nijkamp, P., & Rietveld, P. (2004). The entry of low cost airlines: Price competition in the European airline market. Tinbergen Institute Discussion Paper No. TI 04-074/3. Borenstein, S. (1989). Hubs and high fares: Dominance and market power in the U.S. airline industry. RAND Journal of Economics, 20, 344–365. Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the U.S. airline industry. Journal of Political Economy, 102, 653–683. Borenstein, S., & Shepard, A. (1996). Dynamic pricing in retail gasoline markets. RAND Journal of Economics, 27, 429–451. Brueckner, J. K., Dyer, N. J., & Spiller, P. T. (1992). Fare determination in airline hub-andspoke networks. RAND Journal of Economics, 23, 309–333. Brueckner, J. K., Lee, D., & Singer, E. (2000). Airline competition and domestic U.S. airfares: A comprehensive reappraisal. Mimeo. University of California, Irvine. Busse, M., & Rysman, M. (2002). Competition and price discrimination in yellow pages advertising. Mimeo. Boston University. Busse, M., & Rysman, M. (2005). Competition and price discrimination in yellow pages advertising. RAND Journal of Economics, 36, 378–390. Cohen, A. (2004). Identifying price discrimination when product menus are endogenous. Board of Governors of the Federal Reserve System Finance and Economics Discussion Series, 2004-10. Clerides, S. K. (2002). Book value: Intertemporal pricing and quality discrimination in the U.S. market for books. International Journal of Industrial Organization, 20, 1385–1408. Dana, J. D. (1998). Advance-purchase discounts and price discrimination in competitive markets. Journal of Political Economy, 106, 395–422. Dana, J. D. (1999a). Using yield management to shift demand when peak time is unknown. RAND Journal of Economics, 30, 456–474. Dana, J. D. (1999b). Equilibrium price dispersion under demand uncertainty: The roles of costly capacity and market structure. RAND Journal of Economics, 30, 632–660. Gale, I. L., & Holmes, T. J. (1993). Advance-purchase discounts and monopoly allocation of capacity. American Economic Review, 83, 135–146. Gerardi, K. S., & Shapiro, A. H. (2009). Does competition reduce price dispersion? New evidence from the airline industry. Journal of Political Economy, 117, 1–37. Goolsbee, A., & Syverson, C. (2008). How do incumbents respond to the threat of entry? Evidence from the major airlines. Quarterly Journal of Economics, 123, 1611–1633. Hernandez, M. A., & Wiggins, S. N. (2010). Nonlinear pricing strategies and market concentration in the airline industry. Mimeo. Texas A&M University. Khan, R., Singh, V., & Zhu, T. (2009). Price discrimination in the auto rental industry. Mimeo. Retrieved from http://www.ckgsb.edu.cn/mrf2009/papers/c27283ab-d942-4189-9bce3c30e55d83fc.pdf

46

MANUEL A. HERNANDEZ ET AL.

Kleibergen, F., & Paap, R. (2006). Generalized reduced rank tests using the singular value decomposition. Journal of Econometrics, 133, 97–126. Lee, D., & Luengo-Prado, J. (2005). The impact of passenger mix on reported ‘‘Hub Premiums’’ in the U.S. airline industry. Southern Economic Journal, 72, 372–394. McManus, B. (2007). Nonlinear pricing in an oligopoly market: The case of specialty coffee. RAND Journal of Economics, 38, 512–532. Morrison, S. A. (2001). Actual, adjacent, and potential competition: Estimating the full effect of Southwest airlines. Journal of Transport Economics and Policy, 35, 239–256. Puller, S. L., Sengupta, A., & Wiggins, S. N. (2009). Measuring evidence of peak-load pricing and consumer heterogeneities in airline pricing. Mimeo. Texas A&M University. Rochet, J. C., & Stole, L. A. (2002). Nonlinear pricing with random participation. Review of Economic Studies, 69, 277–311. Sengupta, A., & Wiggins, S. N. (2006). Airline pricing, price dispersion, and ticket characteristics on and off the internet. Mimeo. Texas A&M University. Shepard, A. (1991). Price discrimination and retail configuration. Journal of Political Economy, 99, 30–53. Stavins, J. (2001). Price discrimination in the airline market: The effect of market concentration. Review of Economics and Statistics, 83, 200–202. Stole, L. A. (1995). Nonlinear pricing and oligopoly. Journal of Economics and Management Strategy, 4, 529–562. Verlinda, A. J., & Lane, L. (2004). The effect of the internet on pricing in the airline industry. Mimeo. Retrieved from http://ssrn.com/abstract¼965788 Wang, C. (2005). The effect of a low cost carrier in the airline industry. MMSS Honors Seminar. Retrieved from http://mmss.wcas.northwestern.edu/thesis/articles/get/548/ Wang2005.pdf

Table A.1.

Log of Fare per Mile Regressions (Using a 1 percent threshold to identify Southwest and Other LCCs’ Presence). No-Interaction Model OLS Coeff.

1.582 0.779 0.301 0.389 0.333 0.272 0.153

2SLS Coeff.

Std. Err.

Dependent variable: log of fare per mile 0.036 1.676 0.037 0.069 0.794 0.078 0.027 0.287 0.024 0.055 0.545 0.086 0.059 0.399 0.087 0.059 0.548 0.101 0.038 0.150 0.044

OLS

2SLS

Coeff.

Std. Err.

Coeff.

Std. Err.

1.553 1.021 0.344 0.130 0.205 0.218 0.119 0.237 0.651 0.190 0.237 0.739 0.135 0.165 0.411 0.044 0.083 0.058

0.053 0.062 0.026 0.046 0.040 0.073 0.038 0.120 0.093 0.046 0.068 0.076 0.057 0.106 0.126 0.068 0.051 0.133

1.611 1.041 0.332 0.312 0.211 0.470 0.131 0.417 0.576 0.172 0.155 0.877 0.055 0.168 0.452 0.122 0.188 0.026

0.049 0.065 0.031 0.090 0.065 0.100 0.040 0.149 0.079 0.052 0.087 0.094 0.073 0.121 0.189 0.047 0.068 0.124

47

Group F Group 1 Group 2 Southwest on route Southwest on adjacent route Southwest potential entry Low cost carrier on route Group FSouthwest on route Group 1Southwest on route Group 2Southwest on route Group FSouthwest adjacent route Group 1Southwest adjacent route Group 2Southwest adjacent route Group FSouthwest potential entry Group 1Southwest potential entry Group 2Southwest potential entry Group FLCC on route Group 1LCC on route

Std. Err.

Interaction Model

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

APPENDIX

48

Table A.1. (Continued ) No-Interaction Model OLS

Interaction Model

2SLS

OLS

2SLS

Std. Err.

Coeff.

Std. Err.

Coeff.

Std. Err.

Coeff.

Std. Err.

Group 2LCC on route Adv0_3 Adv4_6 Adv7_13 Adv14_21 One-way Online purchase Mean deviation of load factor Peak time

0.190 0.159 0.131 0.068 0.171 0.236 0.136 0.020

0.025 0.024 0.021 0.013 0.019 0.010 0.016 0.006

0.214 0.166 0.130 0.068 0.144 0.246 0.128 0.014

0.027 0.025 0.021 0.014 0.023 0.012 0.018 0.009

0.090 0.201 0.175 0.122 0.068 0.155 0.229 0.134 0.017

0.057 0.024 0.024 0.018 0.012 0.018 0.010 0.015 0.006

0.069 0.228 0.187 0.122 0.068 0.128 0.238 0.127 0.011

0.061 0.025 0.024 0.018 0.013 0.022 0.011 0.016 0.008

Hub for carrier Market share HHI Slot-controlled airport Log distance Log total flights on route Log per capita income Log temperature difference Tourism index American United Delta Continental

0.190 0.083 0.112 0.050 0.698 0.009 0.182 0.007 1.120 0.134 0.022 0.232 0.099

0.049 0.080 0.094 0.031 0.022 0.030 0.163 0.019 0.437 0.059 0.062 0.055 0.078

0.405 0.064 0.871 0.148 0.896 0.684 0.005 0.020 1.246 0.042 0.098 0.047 0.098

0.114 0.124 0.357 0.052 0.066 0.205 0.255 0.019 0.618 0.073 0.054 0.103 0.094

0.211 0.063 0.052 0.031 0.692 0.019 0.256 0.018 1.317 0.124 0.012 0.219 0.130

0.045 0.077 0.089 0.029 0.023 0.028 0.155 0.019 0.380 0.054 0.060 0.052 0.074

0.414 0.087 0.856 0.126 0.883 0.673 0.111 0.033 1.462 0.038 0.085 0.051 0.140

0.108 0.126 0.325 0.049 0.063 0.197 0.236 0.017 0.554 0.072 0.050 0.097 0.088

MANUEL A. HERNANDEZ ET AL.

Coeff.

0.255 9.288

0.076 1.768

0.082 14.292

0.095 3.606

0.322 10.126

0.066 1.624

10.28 (0.001)

755,842 0.792

6.95 755,842 0.720

0.165 15.349

0.092 3.470 10.43 (0.001)

755,842 0.810

6.79 755,842 0.745

Note: Fare per mile ¼ roundtrip fare (in cents)/(2  nonstop origin to destination mileage). White robust standard errors reported, clustered on route. Market share and HHI instrumented using the same instruments as Borenstein (1989) and Borenstein and Rose (1994). Log of total flights instrumented with the log of population. The under- and weak identification tests for the instruments are the LM and Wald versions of the Kleibergen and Paap (2006) rk statistic and are heteroskedastic-robust.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

US Airways Constant Under-identification test Kleibergen-Paap rk LM stat. Chi-sq(1) P-val Weak identification test Kleibergen-Paap rk Wald F stat. # Observations R-squared

49

Log of Fare per Mile Regressions (Using a 10 percent threshold to identify Southwest and Other LCCs’ Presence). No-Interaction Model OLS Coeff.

1.580 0.774 0.297 0.382 0.300 0.220 0.120

0.189 0.158 0.132 0.070 0.170

Coeff.

OLS

Std. Err.

Dependent variable: log of fare per mile 0.035 1.675 0.038 0.070 0.783 0.079 0.027 0.281 0.025 0.058 0.546 0.101 0.057 0.371 0.085 0.069 0.651 0.146 0.036 0.148 0.047

0.025 0.024 0.020 0.013 0.018

0.214 0.166 0.131 0.071 0.143

0.027 0.025 0.021 0.014 0.023

2SLS

Coeff.

Std. Err.

Coeff.

Std. Err.

1.553 1.008 0.304 0.122 0.179 0.177 0.106 0.196 0.664 0.220 0.230 0.730 0.156 0.178 0.456 0.043 0.084 0.112 0.014 0.198 0.170 0.123 0.069 0.157

0.049 0.061 0.030 0.049 0.038 0.087 0.040 0.133 0.077 0.049 0.067 0.075 0.062 0.131 0.153 0.078 0.053 0.145 0.053 0.025 0.024 0.019 0.012 0.018

1.609 1.016 0.300 0.317 0.185 0.572 0.145 0.352 0.570 0.204 0.140 0.867 0.065 0.166 0.607 0.016 0.200 0.036 0.018 0.227 0.185 0.124 0.070 0.127

0.046 0.063 0.030 0.108 0.060 0.131 0.044 0.161 0.076 0.054 0.091 0.094 0.076 0.156 0.179 0.061 0.066 0.130 0.057 0.026 0.024 0.019 0.013 0.022

MANUEL A. HERNANDEZ ET AL.

Group F Group 1 Group 2 Southwest on route Southwest on adjacent route Southwest potential entry LCC on route Group FSouthwest on route Group 1Southwest on route Group 2Southwest on route Group FSouthwest adjacent route Group 1Southwest adjacent route Group 2Southwest adjacent route Group FSouthwest potential entry Group 1Southwest potential entry Group 2Southwest potential entry Group FLCC on route Group 1LCC on route Group 2LCC on route Adv0_3 Adv4_6 Adv7_13 Adv14_21 One-way

Interaction Model

2SLS

Std. Err.

50

Table A.2.

0.241 0.138 0.021

Hub for carrier Market share HHI Slot-controlled airport Log distance Log total flights on route Log per capita income Log temperature difference Tourism index American United Delta Continental US Airways Constant

0.187 0.074 0.125 0.068 0.718 0.010 0.030 0.004 1.533 0.090 0.063 0.197 0.041 0.262 7.749

0.010 0.016 0.006

0.012 0.018 0.009

Dependent variable: log of fare per mile 0.050 0.385 0.116 0.081 0.049 0.118 0.095 0.894 0.364 0.030 0.164 0.051 0.023 0.926 0.069 0.030 0.705 0.211 0.160 0.128 0.258 0.020 0.017 0.021 0.461 1.938 0.699 0.062 0.116 0.081 0.065 0.173 0.065 0.059 0.033 0.106 0.082 0.016 0.100 0.080 0.076 0.099 1.757 13.294 3.661

Under-identification test Kleibergen-Paap rk LM stat. Chi-sq(1) P-val Weak identification test Kleibergen-Paap rk Wald F stat. # Observations R-squared

0.253 0.132 0.015

0.234 0.136 0.018

0.010 0.015 0.006

0.245 0.131 0.013

0.011 0.016 0.009

0.208 0.062 0.061 0.047 0.712 0.020 0.097 0.014 1.628 0.083 0.050 0.192 0.074 0.325 8.535

0.047 0.080 0.090 0.028 0.025 0.028 0.156 0.021 0.420 0.057 0.061 0.055 0.076 0.068 1.639

0.396 0.076 0.870 0.142 0.915 0.694 0.025 0.029 2.054 0.108 0.155 0.023 0.061 0.157 14.235

0.111 0.121 0.330 0.047 0.065 0.203 0.239 0.019 0.623 0.078 0.059 0.097 0.092 0.092 3.492

10.25 (0.001)

755,842 0.788

6.83 755,842 0.714

10.19 (0.001)

755,842 0.806

6.91 755,842 0.737

51

Note: Fare per mile ¼ roundtrip fare (in cents)/(2  nonstop origin to destination mileage). White robust standard errors reported, clustered on route. Market share and HHI instrumented using the same instruments as Borenstein (1989) and Borenstein and Rose (1994). Log of total flights instrumented with the log of population. The under- and weak identification tests for the instruments are the LM and Wald versions of the Kleibergen and Paap (2006) rk statistic and are heteroskedastic-robust.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

Online purchase Mean deviation of load factor Peak time

Log of Fare per Mile Regressions (Southwest, Other LCCs’ Presence, and Market Structure Measures Derived from DB1B). No-Interaction Model OLS Coeff.

Coeff.

OLS Std. Err.

1.582 0.776 0.299 0.331 0.252 0.008 0.090

Dependent variable: log of fare per mile 0.034 1.674 0.036 0.071 0.792 0.078 0.026 0.284 0.024 0.051 0.502 0.081 0.056 0.336 0.079 0.103 0.222 0.113 0.034 0.100 0.041

0.189 0.156 0.132 0.072 0.167 0.243

0.024 0.024 0.020 0.014 0.018 0.011

0.212 0.164 0.131 0.072 0.141 0.256

0.026 0.025 0.020 0.014 0.023 0.012

2SLS

Coeff.

Std. Err.

Coeff.

Std. Err.

1.603 1.028 0.339 0.054 0.138 0.137 0.051 0.282 0.724 0.241 0.166 0.757 0.172 0.156 0.257 0.202 0.006 0.021 0.093 0.197 0.172 0.122 0.071 0.152 0.234

0.054 0.065 0.022 0.038 0.045 0.088 0.035 0.109 0.096 0.045 0.068 0.078 0.065 0.079 0.112 0.067 0.055 0.119 0.056 0.022 0.023 0.017 0.012 0.018 0.010

1.659 1.039 0.322 0.245 0.167 0.331 0.075 0.418 0.640 0.215 0.097 0.869 0.108 0.060 0.233 0.124 0.075 0.013 0.069 0.222 0.182 0.121 0.070 0.128 0.246

0.053 0.065 0.027 0.073 0.064 0.104 0.036 0.142 0.079 0.053 0.085 0.091 0.078 0.081 0.159 0.078 0.062 0.111 0.060 0.024 0.023 0.017 0.013 0.022 0.011

MANUEL A. HERNANDEZ ET AL.

Group F Group 1 Group 2 Southwest on route Southwest on adjacent route Southwest potential entry LCC on route Group FSouthwest on route Group 1Southwest on route Group 2Southwest on route Group FSouthwest adjacent route Group 1Southwest adjacent route Group 2Southwest adjacent route Group FSouthwest potential entry Group 1Southwest potential entry Group 2Southwest potential entry Group FLCC on route Group 1LCC on route Group 2LCC on route Adv0_3 Adv4_6 Adv7_13 Adv14_21 One-way Online purchase

Interaction Model 2SLS

Std. Err.

52

Table A.3.

Hub for carrier Market share HHI Slot-controlled airport Log distance Log total flights on route Log per capita income Log temperature difference Tourism index American United Delta Continental US Airways Constant

0.135 0.025 0.199 0.059 0.160 0.070 0.698 0.012 0.040 0.007 1.333 0.089 0.043 0.205 0.051 0.231 7.734

0.016 0.007

0.017 0.009

Dependent variable: log of fare per mile 0.049 0.402 0.113 0.086 0.080 0.133 0.100 0.802 0.334 0.031 0.163 0.052 0.023 0.895 0.062 0.031 0.682 0.198 0.163 0.077 0.226 0.020 0.021 0.020 0.415 1.618 0.643 0.070 0.116 0.093 0.075 0.155 0.076 0.069 0.013 0.111 0.083 0.022 0.099 0.087 0.029 0.109 1.784 13.406 3.179

Under-identification test Kleibergen-Paap rk LM stat. Chi-sq(1) P-val Weak identification test Kleibergen-Paap rk Wald F stat. # Observations R-squared

0.127 0.020

0.132 0.020

0.014 0.006

0.127 0.017

0.015 0.008

0.215 0.040 0.100 0.048 0.691 0.019 0.118 0.019 1.472 0.079 0.032 0.191 0.082 0.303 8.581

0.044 0.082 0.093 0.030 0.025 0.030 0.161 0.019 0.364 0.064 0.069 0.064 0.076 0.075 1.705

0.401 0.146 0.666 0.133 0.871 0.632 0.047 0.034 1.722 0.098 0.133 0.001 0.068 0.124 14.134

0.105 0.122 0.284 0.048 0.056 0.178 0.212 0.017 0.561 0.085 0.066 0.099 0.089 0.094 2.900

10.52 (0.001)

755,842 0.784

6.77 755,842 0.715

10.45 (0.001)

755,842 0.806

6.33 755,842 0.749

Note: Fare per mile ¼ roundtrip fare (in cents)/(2  nonstop origin to destination mileage). White robust standard errors reported, clustered on route. Market share and HHI instrumented using the same instruments as Borenstein (1989) and Borenstein and Rose (1994). Log of total flights instrumented with the log of population. The under- and weak identification tests for the instruments are the LM and Wald versions of the Kleibergen and Paap (2006) rk statistic and are heteroskedastic-robust.

Effect of LCCs on Nonlinear Pricing Strategies of Legacy Airlines

Mean deviation of load factor Peak time

53

CHAPTER 3 THE NEW PRICING IN NORTH AMERICAN AIR TRAVEL MARKETS: IMPLICATIONS FOR COMPETITION AND ANTITRUST David Gillen and Tim Hazledine INTRODUCTION The passenger air travel market has recently been impacted by two major innovations: first, in the 1990s, the rise of ‘‘low-cost carrier’’ (LCC) airlines offering cheap one-way point-to-point tickets, and then, in the new millennium, the emergence and enthusiastic adoption by consumers of online Internet booking systems. It has been suggested that the transparency of Internet booking would result in only the lowest fare offerings being sustainable in the market, and the simplicity and efficiency of LCCs would mean that it would be their fares that would be the lowest. For the traditional ‘‘network’’ or ‘‘legacy’’ carriers, these developments have implications beyond just possible increased competitive pressure on average fares. In every industry, there is in general a wide variation in customers’ willingness to pay, but the airline industry is perhaps unique in the extent to which it has been able take advantage of this by deploying price discrimination techniques that convert consumer to producer surplus – basically by means of ‘‘fences’’ placed around the lowest price tickets aimed Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 55–82 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003005

55

56

DAVID GILLEN AND TIM HAZLEDINE

at leisure travelers, which deter the higher-value business and other lastminute customers from purchasing these fares. The most famous and effective of these fences has been, since its introduction in the 1980s, the requirement that a discounted ticket be purchased, in advance, for a return journey involving a Saturday night stayover at the destination. This restriction cleverly sorted out leisure travelers – for whom the weekend or Saturday stay may be part of the reason for the trip – from the business customer who would be keen to get home for the weekend and might also have a more complicated itinerary than a simple return journey. Obviously, if an airline switches to selling one-way fares, it cannot impose any stay-over requirement, and so, if the legacy carriers have been forced by LCC competition to adopt offers of a one-way-trip pricing structure, they may have thereby lost some or all of their ability to fence off leisure from business travelers. One aviation economist has suggested that the implications of these innovations are far-reaching and radical: Perhaps the most important impact of the LCC business model on [legacy carriers] has been the introduction of low one-way fares. This has undermined the price discrimination ability of the [legacy carriers], and is the most important pricing development in the industry in [the] past 25 years. (Tretheway, 2004)

These propositions have already been carried through to antitrust regulation, in two recent cases involving the legacy carriers Qantas and Air New Zealand, who sought approval to in effect cartelize markets on which they held joint market shares in excess of 80%, on the grounds that the discipline imposed on legacy carriers by actual or potential competition from LCCs would be sufficient to prevent any significant increase in airfares. The role of LCCs and the issue of whether structural competition between legacy carriers is still relevant are the focus of the empirical work reported in this chapter. We made use of the new Internet booking sites to generate more than 20,000 observations of airfares offered in May 2006 on 39 North American routes: 22 within Canada and 17 transborder routes between Canada and the United States. The impact of the Internet and LCCs is immediately apparent: the old restricted return fare product has completely disappeared from the Canadian market – all 13 of the airlines in our database offer one-way Internet fares, which can be purchased either directly from the airline’s web site or from a travel agency that collects all the available fares and displays them so as to make it immediately obvious to the consumer where the best deal is coming from.

The New Pricing in North American Air Travel Markets

57

Thus, in this sense, the traditional legacy carrier business model has indeed been broken up and replaced with something much closer in its processes to the LCC model, at least in flights within or originating in Canada. But then the interesting finding from our research is that the market outcomes generated by the new systems are quite familiar. We find that the legacy carriers are still able to generate substantial intertemporal price discrimination, with prices offered for a specific flight often being two or three times higher in the week before the flight than the fares available eight weeks before (where our observations begin). We observe this pattern of fare increases even in markets served also by LCCs, whose prices throughout – and especially in the last week – are significantly lower than those of the legacy carriers. We find that structural competition still matters: the legacy carriers’ fares are lower, on average, if there are more of them operating on a route. And we find evidence that the very transparency of Internet-based fare systems may actually have made it easier for legacy carriers to coordinate their pricing, in particular to take advantage of the generally higher willingness to pay of business travelers making their purchase decision close to the day of the flight. This chapter continues with a review of the econometric literature on air travel pricing, which has focused on the US market. The third section introduces our new pricing database and illustrates pricing patterns on five routes. The fourth section sets out the pricing model and describes the other variables collected for this study. The fifth section reports the results of the econometric modeling. The sixth section concludes with some implications for policy and future research.

PREVIOUS STUDIES OF AIRLINE PRICING There is an interesting econometric literature on the relationship between airline (average) fares and differences or changes in market structures, mostly restricted to the US domestic market. The reasons for this restriction are, first, that the United States has, of course, a large and important domestic aviation industry to study, and second, that American researchers have available to them an unusually good database on prices charged on domestic flights. This is the well-known Department of Transport ‘‘Database 1a’’ (DB1a), which is a 10% quarterly sample of all tickets sold, by carrier and route. The US literature has had three main focuses of attention: on the link between number of competitors and prices; on the existence of

58

DAVID GILLEN AND TIM HAZLEDINE

‘‘hub premiums’’ earned by airlines dominating an airport that is used as a hub for connecting flights; and on the competitive impact of LCCs, especially the most successful of the US LCCs, Southwest Airlines. Many of these studies are surveyed by Tretheway and Kincaid (2005). Kim and Singal (1993) examined the impact of the 14 US airline mergers (national and/or regional carriers) that took place over the 1985–1988 period, comparing routes affected by mergers with a control group of unaffected routes. They found that over the period from the initiation of merger talks through merger completion, the merging firms increased fares on average by 9.44% relative to unaffected routes, and any competitors on the affected routes raised their prices by even more – 12.17%, on average. Peters (2006) is a sophisticated analysis of five of these mergers, also finding price increases. Borenstein (1990) focused on the 1986 Northwest Airlines/ Republic merger and found average fare increases of 6.7% and 22.5% on routes served premerger by both airlines, depending on whether there was or was not at least one other competitor also serving the route. Clougherty (2006) found that domestic airline merger activity of the late 1980s and early 1990s had an important international dimension, as domestic mergers improve an airline’s international competitive position. Hurdle, Johnson, Joskow, Werden, and Williams (1989) compared 850 nonstop city-pair routes in the United States and reported an average price differential between routes supplied by one carrier and routes supplied by two carriers of about 20%. Borenstein (1991) found average fare differences between monopoly and duopoly, and between duopoly and triopoly, of 8% in each case. Morrison and Winston (1990) report that the number of effective airlines in a particular market over the 1982–1988 period affected the fare charged with an elasticity of 0.12 – that is, a route served by two carriers has fares 12% lower than a route served by just one airline. Turning to studies focusing on the ‘‘hub premium,’’ Hofer, Dresner, and Windle (2004) examine whether LCCs affect network carriers’ ability to benefit from market concentration and power and whether LCCs earn hub premiums. Their conclusion is that airport market power and market concentration are positively correlated with average fares and that the presence of an LCC in a market leads to lower fares. Ciliberto and Williams (2010) have probed the mechanics of the hub premium, which they find can be directly linked to the ability of hub airlines to control competitive access to airport gates. Some studies have looked for a ‘‘Southwest effect’’ qualifying the hub premium. Morrison and Winston (2000) compared the average fare at 11 concentrated airports with the average fare across all airports in the United

The New Pricing in North American Air Travel Markets

59

States, using 1996 DB1a data. They found a hub premium estimate of 22%. They next controlled for Southwest by simply removing airports served by this airline and found fares at the concentrated airports were found to be 6% lower than the remaining airports; fares at airports served by Southwest were approximately 39% lower than fares at airports not served, averaging across all airports. In another study, Morrison and Winston (1995) found that average fares for travel to and from the 12 concentrated airports were actually 6% lower on average than those for trips to and from other domestic airports if those served by Southwest were excluded from the latter group. They noted that if airports served by Southwest were excluded from both the concentrated and the remaining unconcentrated hub airports, average fares between the two groups differed by only about 1% after adjusting for differences in passengers’ trip lengths and use of frequent-flyer award tickets. However, Goolsbee and Syverson (2008) find that an increase in the probability of Southwest entering a route1 has a large (20% þ ) and significant impact on prices when the existing level of concentration on the route is above average, but not when the route is of below-average concentration. So, there may be some disagreement about the extent to which the ‘‘Southwest effect’’ is qualified by the extent of existing competition, but there seems little doubt that, overall, Southwest and perhaps some other LCCs have had a substantial impact on US airfares well beyond the price of the tickets they sell for their own flights. The above papers have all studied the average fare or price charged. Two other articles have reported research into the dispersion of fares: in particular, the gap between highest and lowest fares observed, which in the airline context is generally attributed to price discrimination. Given the pervasiveness of price discrimination in this industry, it is of course interesting to investigate its determinants: in particular, the relationship, if any, between the dispersion of high and low fares and industry structure. A priori, we could expect that more elastic segments of market demand would be more prone to price-cutting in the presence of competition. In a model with a linear market demand curve, elasticity is of course higher (in absolute value) at the top of the curve than the bottom, and Hazledine (2006) has shown in an extension of the standard Cournot–Nash model to allow for price discrimination that because of this the dispersion between highest and lowest prices is largest in a monopoly market. However, it is possible and perhaps likely that air travel market demand curves are nonlinear, such that the leisure (low willingness to pay) segment is more elastic than the high-value business/last-minute segment. In this case,

60

DAVID GILLEN AND TIM HAZLEDINE

increasing the number of competitors may have more effect on pricing at the low end of the market. Borenstein and Rose (1994) call this ‘‘competitivetype’’ price discrimination and find evidence for it in the DB1a data, as does Stavins (2001) using Official Airline Guide data on prices offered under the advance purchase return ticket restrictions regime. Overall, it is difficult to neatly summarize the US results, even though nearly all studies make use of the same database (albeit for different time periods). It seems evident that market concentration does tend to have a link with average prices, especially concentration at hub airports, though Tretheway and Kincaid (2005) note that the size of the estimated hub effect has reduced as controls have been introduced for other factors affecting price, such as route distance and the leisure/business customer split. It is clear that the entry of LCCs – in particular, Southwest Airlines – has had a quite dramatic impact on US domestic airfares, though it is not settled whether this effect totally superseded the role of market structure, that is, competition between legacy carriers. And we note that all the studies surveyed above use data on airfares generated before legacy carriers had adjusted to the new realities of the LCC business model – in particular, oneway cheap fares – and to the Internet booking sites that have largely superseded the opaque computer reservation systems previously used by airlines to monitor and manage their price discrimination techniques.

PRICING BEHAVIOR IN CANADIAN AND TRANSBORDER ROUTES Tretheway and Kincaid (2005) conclude that In part, the lack of research on the effect of market concentration on average airline fares outside the United States is due to the fact that, in other countries, publicly available data on individual ticket purchases is difficult to obtain.

Indeed, no such data on ticket purchases are available for flights originating in Canada. However, the new one-way Internet-bookable fare system has been wholeheartedly adopted by Air Canada (AC), WestJet (WJ), and all other carriers serving the domestic and/or transborder Canadian market.2 This system is itself a rich potential source of price data – though not price paid, but rather price offered. The product is tightly defined – a particular one-way journey at a particular time and date – and the price it is offered at is observable not just to consumers, but also to any researcher willing to take the time and effort to make frequent observations.

The New Pricing in North American Air Travel Markets

61

We used the Expedia.ca web site as our source of price data. The web sites have to be observed in real time, pages printed out and then manually transcribed to spreadsheets, and therefore, there is a real constraint in terms of research assistants’ time limiting the scope of our database. We chose to observe flights departing in the last four of the five Wednesdays in May, 2006, and we took up to nine observations3 on prices for each flight, beginning eight Wednesdays before flight date and continuing weekly up to one week before flight date, and then finally the price the day before (i.e., Tuesday) flight date. We observed fares on 39 routes, with a total of 438 daily flights, all nonstop or one-stop, flown by 13 airlines.4 The sample of routes was chosen to give us a range of trip lengths and city size combinations. The sample includes most of the routes on which AC offers some nonstop service. The major exclusion is the ‘‘triangle’’ of routes linking Toronto, Ottawa, and Montreal, which were excluded because of the sheer volume of flights that would need to be recorded. There are, for example, 25 flights each way between Toronto and Ottawa flown daily by AC and WJ. Our price data actually have three advantages over the US data used in the studies surveyed in the previous section. First, they are at the level of individual flights, whereas DB1a is aggregated to the route level. Second, they are for specific flight dates, whereas DB1a is compiled on a quarterly basis. And, third, our observations are also dated by the time before flight that the fares were offered, so we will be able to observe the process of any intertemporal price discrimination. On the other hand, since we are just making, in essence, up to nine spot observations on fares offered, we do not know how many tickets were actually purchased at these or other prices, whereas DB1a, with its 10% sample of all the tickets sold, will yield good estimates of the true average ex post price paid. Also, we limit our observations to the lowest fare offered at any time. AC, in particular, offers a suite of economy fares, varying by the features attached to the fare, such as frequent flier points and flexibility with respect to changes. Sometimes one or two of the lower fare classes disappears from the market soon before the flight, but not always, and in general, we must expect that some tickets will be sold at fares above those observed in our database and that some of our observed fares will be in fare classes offering more features or fewer restrictions than the lowest fares. In this section, we illustrate the results of our data collection. The most striking feature of the data is simply that the lowest fare offered for a particular flight really does exhibit a strong tendency to increase the closer the observed fare offer is to the flight date. The time path of prices is often, but not always, monotonic – there are occasions where we observe a fall in

62

DAVID GILLEN AND TIM HAZLEDINE

the lowest price, though this is nearly always reversed in the weeks following. Here, we show the lowest fares available at each observation time – defined by the number of weeks before flight date – averaged across all the flights offered by each airline serving a route and across the four Wednesday flight dates. Fig. 1 shows the YVR-YEG (Vancouver–Edmonton) route, which has a somewhat higher than average proportion of business travelers. The route was served by AC and WJ. We can see both carriers start to increase prices five weeks out, marked on the figure as P5, and that the average price increases steadily, but with AC at a higher level throughout than WJ. The biggest jump is from P1 to P0 (the day before flight date). So, on this route, both carriers are using intertemporal price discrimination, but the legacy carrier more so than the LCC.5 In contrast, on Fig. 2 showing the Vancouver–Halifax (YVR–YHZ) route, we had three carriers – two LCCs (WJ and CanJet) and AC. This is a relatively long-haul route, at approximately 4,600 km. The carriers behave in strikingly different ways. WJ gradually increases prices by a small amount, CanJet kept prices relatively constant, and AC increased prices very little in the first six weeks but two weeks before the flight lifted prices significantly. So we have one LCC that does not price discriminate at all, one that does to a small degree, and a legacy carrier that discriminates to some degree over six weeks and significantly in the last two weeks. Compare this behavior with Fig. 3, which shows prices in one of the transborder markets. The route is Toronto–New York (YYZ–LGA), as served by four legacy carriers – AC, AA (American airlines), Delta, and UA

Fig. 1.

Representative Fare for Short-Haul Domestic Route.

The New Pricing in North American Air Travel Markets

63

Fig. 2.

Representative Fare for Short-Haul Domestic Route-Multiple Low-Cost Carriers.

Fig. 3.

Representative Fare for Long-Haul Transborder Route-Multiple Carriers.

(United Airlines) – and one LCC, CanJet. This can be regarded as a predominately business market. Prices are relatively constant until two weeks before the flight with the LCC charging the lowest price. At two weeks before the flight, every carrier except CanJet raises prices significantly, as then does CanJet one week out. The legacy carrier prices increase significantly and in an apparently coordinated way.

64

DAVID GILLEN AND TIM HAZLEDINE

In Fig. 4, we show the high-volume domestic route Toronto–Vancouver, which is about average in terms of the relative proportions of business and leisure passengers. There are four carriers: AC, WJ, and two scheduled charter operators that would be classified as LCCs. The legacy carrier sets prices 60%, or more, higher than the other carriers. All carriers increase their prices by a small amount over weeks eight through two before flight date, but only the legacy carrier increases prices significantly within two weeks of takeoff. Note that Harmony is the only carrier in our sample that appears to systematically offer last-minute cheap seats on those of its flights that are not sold-out. Our final illustration is Fig. 5, showing another transborder route (Vancouver–Los Angeles), which was served by five legacy carriers plus the US-based LCC, America West. Here, the pricing behavior is rather different. There is a wide range right up to the week before the flight, with American Airlines the highest at around $370 and America West and Alaska in the low 200s. But then, in the last week, four of the legacy carrier sharply increase prices, with three of them doing so in parallel, such that they almost catch up with the American Airlines fares. Alaska, the regional legacy carrier, however, persists in offering low prices, which even fall slightly the day before the flight. Figs. 1–5 illustrate pricing activity in markets with differing numbers of carriers as well as differences in route length, market type, domestic versus transborder, and mix of carriers. These data show quite marked differences in prices charged for flights on the same route and the same day – differences

Fig. 4.

Representative Fare for Long-Haul Domestic Route.

The New Pricing in North American Air Travel Markets

65

Average Fare by Carrier YVR-LAX May 2006 $600.00 $500.00

AC

$400.00

Alaska America West

$300.00

AA NW

$200.00

UA

$100.00 $P8 P7 P6 P5 P4 P3 P2 P1 P0

Fig. 5.

Representative Fare for Short-Medium-Haul Transborder Route.

both across carriers and over time on each carrier. The data also reveal some striking apparent coordination of price increases as the flight date approaches. The next sections explore these pricing patterns systematically.

MODEL SPECIFICATION AND DATA We will be specifying econometric models to explain variations in measures of the lowest fare offered on each flight/date in terms of demand, cost, and market structure factors. To eliminate price differences solely due to the length of the flight, all the fares as shown in the previous section are divided by the nonstop length of the journey in kilometers. Here, we introduce and define first the dependent (price) variables, then the explanatory regressors. Table 1 notes means, maxima, and minima of all variables.

Price Variables The dependent variables are as follows: Pwavk: a weighted average of all nine lowest fare observations (divided by ‘‘k’’ – the nonstop flight distance). Ideally, the weights would reflect the number of tickets sold at each fare, but we do not have such information. We do have

66

DAVID GILLEN AND TIM HAZLEDINE

Table 1.

Pwavk P8k P82k P0k P10k RCOST HHIFLIGHTS DOMINANT FRINGE SOLDOUT LEISURE STOPS RETURN WESTJET LCCXWJET

Descriptive Statistics for Data Set. Mean

Maximum

Minimum

0.257 0.171 0.186 0.379 0.341 0.116 0.431 0.080 0.039 0.278 61.6 0.272 0.130 0.219 0.056

0.908 0.672 0.835 1.514 1.239 0.292 1.00 1.00 1.00 1.00 90 1.00 1.00 1.00 1.00

0.064 0.060 0.060 0.055 0.070 0.052 0.191 0.00 0.00 0.00 40 0.00 0.00 0.00 0.00

a ‘‘guesstimate’’ of the proportions of leisure and business travelers using each route (see below), and we believe that most business tickets are purchased within the last two weeks before travel. So, we construct weights based on the assumptions that leisure travel tickets are purchased evenly throughout the eight weeks, whereas three quarters of business travel tickets are purchased at the fares we observe one week and one day before the flight date. We can report that using a simple unweighted average of the nine fare observations does not make much difference to the results. P8k: the first fare observed, eight weeks before flight date. This is usually but not always the lowest fare that a traveler could purchase over this period for a given flight. P82k: the (unweighted) average of the up to seven fare observations from week eight through week two before flight date. This is an indicator of the average fare that would have been paid by most leisure travelers. We noted in the previous section that fares tend to not increase sharply until within two weeks before travel, and this is reflected in the mean value of this variable compared to P8k. P0k: the day-before fare, paid by last-minute travelers, which is more than twice as high, on average, as the average price over the previous seven weeks. If the flight was sold out before this day, we use the most recent available fare quote. P10k: the average of the day-before and the week-before fares, being a plausible estimate of the fare paid by most business travelers.

The New Pricing in North American Air Travel Markets

67

Regressors RCOST. In most airline pricing studies, route distance is used as a proxy for costs. This is clearly a very important determinant of flying costs and is always highly successful econometrically, but it does suppress any variation in costs due to other factors such as aircraft type, wage costs, and overall operating efficiency. For this study, we built up our own cost variable based on the ‘‘block hour’’ operating costs of different aircraft types as reported by airlines to the US Dept of Transportation. We were able to ascertain the type of aircraft used on each flight, and so could estimate the cost per flight, to which we added an estimate of the marketing costs attributable to each flight. We found that the explanatory power was increased by aggregating per flight costs up to the route level, implying that this is the basis on which airlines set their prices. That is, if an airline has three flights on a route, two of which it operates using a modern lower cost aircraft type, and one with an older airplane, then it does not appear to systematically charge a higher price on the third flight. Thus, the variable RCOST is an airline’s cost of supplying a seat-kilometer on a particular route. The mean value of this variable is well below the mean value of the price variables. Part of this is simply because only a proportion (typically around 70–80%) of flown seats are filled with fare-paying passengers and part because the direct ‘‘trip costs’’ measured here make up, on average, only about 70% of total airline costs, including head office and other fixed costs (Swan & Adler, 2006). Note too that differences across airlines in overall operating efficiency – such as those due to different turnaround times and ground handling costs – are not allowed for, which in particular means that he cost advantage of LCCs is likely to be underestimated. Details of the construction of the RCOST variable are given in the appendix. HHIFLIGHTS. We proxy structural competition with the standard Hirschman–Herfindahl Index, being the sum of squared market shares. We have two possible market share measures: share of seats flown and share of flights, which can differ of course depending on the size of aircraft flown, as well as the number of seats allocated to code-shares and to the component sectors of one-stop itineraries. We found that the flights-based measure performed slightly better, suggesting that it is the number of flights an airline can offer on a route, rather than the total number of seats supplied, which is the better indicator of competitive presence, although the differences in results are not substantial. The HHI is bounded by zero (atomistic structure) and one (monopoly). Symmetric duopolies and triopolies have HHI values of 0.5 and 0.333, respectively, and the sample mean value of 0.43 is between these values.

68

DAVID GILLEN AND TIM HAZLEDINE

DOMINANT; FRINGE. The HHI appears to be a useful summary statistic of structural competition. However, there are other dimensions that may matter. In particular, two routes with the same HHI may have different numbers of competitors, depending on the size distribution. For example, a symmetric duopoly, as noted, has an HHI of 0.5, but so too will be the HHI of a triopoly in which one firm has two-thirds of the market and the other two one-sixth each. Competition authorities have often worried about the possibility of ‘‘dominance,’’ meaning a situation where the market share of the largest supplier is so high as to allow it to, in effect, dominate pricing in the industry (i.e., price like a monopolist), even in the presence of a ‘‘fringe’’ of much smaller suppliers, who may be unable to apply significant pressure, or may not even wish to, being happy to raise their own prices under the ‘‘umbrella’’ held over them by the industry leader. On the other hand, a firm may achieve dominance through superior cost performance and/or other genuine competitive advantages that will be manifested in a lower price, other things equal. And, if implicit collusion is a potential factor in pricing, an industry with more firms and these with differing capabilities and objectives might find it harder to coordinate or collude than would an industry with fewer, more similar suppliers. It is difficult to find a precise definition of what constitutes ‘‘dominance,’’ but our reading of the competition regimes of smaller economies such as Canada, Australia, and New Zealand is that an industry in which the market share of the largest firm equaled or exceeded 70% would be likely to arouse concerns about dominance if it came before the authorities. In our sample of 39 routes, there are five non-monopoly routes on which the largest carrier’s share of total seats supplied exceeds 70%.6 We construct two dummy variables: DOMINANT ¼ 1 if the carrier has market share greater than 70% (but less than 100%); FRINGE ¼ 1 for another carrier flying a route, which has a dominant carrier. SOLDOUT. More than one quarter of the 1,749 flights disappeared from the web site before flight date; presumed sold-out. Sometimes a flight can sell out for noneconomic reasons such as a booking by a large tour group. Links with price could go in either direction: an underpriced flight could thereby sell out or airlines may be able to predict the popularity of a flight (e.g., at peak travel periods) and set a higher price to take advantage of this. LEISURE. Constructed by one of the authors (Gillen) as an informed guess of the percentage division between leisure (i.e., tourists and ‘‘visiting friends and relatives’’) and business passengers traveling on each route, to allow for systematic lower willingness to pay for flights on which a larger proportion of passengers are not traveling on business.

The New Pricing in North American Air Travel Markets

69

WEEKi. We included a dummy for each of the first three Wednesdays in our May observation period. STOPS. More than one quarter of the flights have a stop-over, sometimes involving a change of plane. A stopover involves an additional aircraft movement and additional ground handling costs and so must be expected to add to costs. However, on the demand side, one-stop itineraries must in general be less attractive to travelers, if only because they last an hour or more, longer than the nonstop trip.7 And, we can expect this disadvantage will weigh more heavily on business travelers, for whom the opportunity cost of travel time will in general be higher than the time cost of leisure travelers. RETURN. This is a dummy variable for the four routes in our database that originate in the United States. WESTJET; LCCXWJET. We include dummy variables for WJ, which operates more than one in five flights in our sample, mostly within Canada, and for LCCs apart from WJ, of which there are four, with less than 6% of the flights, in total. Note that the leading North American LCC, Southwest Airlines, does not fly transborder or within Canada.

ECONOMETRIC RESULTS We have seen that there is substantial variation in the lowest prices offered for air travel flights on particular routes, both over time before the flight date and across the carriers serving the route. In this section, we estimate a model to explain these variations. In particular, we are interested in  the extent to which market structure factors are relevant in explaining price differentials across routes;  the influence of LCCs on competition and pricing; and  any differences in the determinants of the prices charged to leisure and business travelers. Table 2 notes regression models estimated for the five pricing variables defined in the previous section, with a common set of regressors. As noted, we observed prices for 1,749 ‘‘flights,’’ each being a journey between two airports (with or without a stop en route), leaving at a certain time on one of four Wednesdays in May 2006. In this section, we examine the results for each explanatory variable in turn. The regression model is estimated by EViews 5.1, using the EGLS procedure with cross-sectional random effects and the flight number used as the cross-sectional identifier.

70

Table 2. Explaining Differences in Lowest Fares Offered for Canadian and Transborder Flights. Regression: Dependent Variable (2) Log fare in week eight before flight (lnP8k)

(3) Log average fare weeks eight to two (lnP82k)

(4) Log fare day before flight (lnP0k)

(5) Log average fare week one and day before (lnP10k)

1.454 (14.90) 1.029 (21.06)

0.593 (7.27) 0.863 (21.12)

0.620 (7.36) 0.866 (20.54)

2.020 (15.74) 1.223 (19.04)

1.806 (14.58) 1.151 (18.56)

0.356 (4.48)

0.369 (5.55)

0.445 (6.47)

0.201 (1.92)

0.317 (3.14)

0.306 (6.69)

0.293 (7.69)

0.305 (7.72)

0.298 (4.96)

0.305 (5.25)

0.249 (4.07)

0.290 (5.67)

0.278 (5.26)

0.191 (2.39)

0.216 (2.79)

0.044 (3.65)

0.020 (2.11)

0.027 (3.00)

0.015 (0.72)

0.049 (2.68)

0.011 (8.44)

0.010 (8.80)

0.009 (7.65)

0.007 (3.78)

0.008 (4.76)

DAVID GILLEN AND TIM HAZLEDINE

Constant Log of route cost per km (lnRCOSTb) Herfindahl Index of flights (HHIFLIGHTS) Carrier has W 70% seat share on route (DOMINANT) Small carrier on dominated route (FRINGE) Flight sold out (SOLDOUT) Percent leisure passengers on route (LEISURE)

(1) Log weighted average fare (lnPwavk)

0.080 (8.26)

0.075 (0.98)

0.110 (15.15)

0.013 (0.74)

0.055 (3.61)

0.041 (4.21)

0.025 (3.25)

0.063 (8.60)

0.008 (0.44)

0.019 (1.29)

0.036 (3.80)

0.021 (2.71)

0.044 (6.03)

0.039 (2.21)

0.030 (1.98)

0.152 (5.33)

0.059 (2.49)

0.093 (3.77)

0.237 (6.27)

0.210 (5.77)

0.227 (5.67)

0.180 (5.36)

0.187 (5.39)

0.284 (5.39)

0.261 (5.13)

0.137 (3.69)

0.152 (4.92)

0.072 (2.25)

0.186 (3.80)

0.177 (3.76)

0.406 (8.01)

0.213 (5.02)

0.198 (4.53)

0.653 (9.79)

0.601 (9.33)

0.579 0.808

0.576 0.814

0.564 0.794

0.497 0.723

0.496 0.735

Notes: t-Statistics in parentheses. All regressions are panel EGLS (cross-section random effects) with the flight number as the cross sectional identifier. All regressions have 1,749 observations.

The New Pricing in North American Air Travel Markets

May 10, 2006 flight (Week 1) May 17, 2006 flight (Week 2) May 24, 2006 flight (Week 3) Number of stops on flight (STOPS) Flight originates in US (RETURN) Carrier is WJ (WESTJET) Carrier is low-cost, excl. WJ (LCCXWJ) Weighted R2 Unweighted R2

71

72

DAVID GILLEN AND TIM HAZLEDINE

Note first that our constructed cost variable RCOST is highly significant and, in particular, has a coefficient in regression 1 remarkably close to 1.0, which is the value we would expect if airlines set their price structures with the aim of achieving profit margins, which do not systematically differ because of factors such as route distance and aircraft type that account for differences in costs. That is, a 1% increase in route costs results in a 1% increase in prices charged, on average. We would not necessarily expect the one-to-one price-cost link to hold exactly for components of the price structure, and we see from regressions 2–5 that is does not hold exactly. The coefficient on unit costs is somewhat smaller than one for the lower part of the price distribution and greater than one for the higher end. This probably reflects differences in the variability of prices at the low and high price ends of the distribution. The performance of the market structure measure in a pricing equation is of considerable interest, especially in the context of recent claims that structure no longer matters because of the competitive threat imposed by LCCs and or by increases in customer price responsiveness resulting from Internet-based ticket marketing systems and the adoption of one-way discounted fares. We see that across these 39 Canadian and transborder routes, market concentration as measured by the HHI index of flights offered shows a linkage with airfare prices that is both statistically and economically significant. The coefficient of 0.356 from regression 1 implies that monopoly routes have fares around 18% higher than (symmetric) duopolies. Also of interest is the result that the concentration-price linkage seems to be relatively larger at the low-price end of the price distribution, even though this is where actual or potential competition from LCCs as well as willingness of (leisure) travelers to shop around are likely to be more important factors in the marketplace. Indeed, the impact of structural competition on day-before-flight fares (regression 4), which of course are mainly purchased by last-minute business travelers, is quite weak, consistent with the rather striking pictures of apparently parallel pricing we showed in the previous section. We included the dummy variables DOMINANT and FRINGE to allow for dimensions to structural competition that may be suppressed by the HHI summary statistic. The regression results are quite striking: the coefficients on these dummy variables are large, significant, negative, and similar.8 That is, given two routes with similar HHIs, the route with more firms (and thus, necessarily, more, smaller firms) will have lower prices charged by all firms, including the largest. The result also implies that adding even quite a small

The New Pricing in North American Air Travel Markets

73

amount of competition to a route that otherwise would be a monopoly would have a very substantial downward effect on prices. However, we caution that this inference comes from a sample with just five examples of supposed dominance. We were not sure a priori whether the observation that a flight sold out before flight day would be associated with airfares, and, if so, which way the causation would run. Would a mistake in pricing too low result in a flight selling out or do the airlines correctly predict that certain flight/dates are more likely to be unusually popular and set prices higher to take advantage of this? The results favor the second hypothesis, with regression 1 implying that prices are overall around 4–5% higher on flights that prompt the value 1 for the coefficient on SOLDOUT. This relationship however does not hold for either of the weeks at the extremes of our observations periods (regressions 2 and 4). The LEISURE variable9 turns out to have substantial predictive power for pricing. The coefficient in regression 1 implies that each percentage point difference in the proportion of leisure passengers on a route results in an approximately 1% difference in average lowest fares charged, and the coefficients in the other regressions tell us that this effect holds through all the weeks before flight date, right up to the day before. It may seem surprising that leisure-dominated routes have lower prices as far away as eight weeks before the flight date, given that we believe just about all customers on all routes purchasing this far in advance will be leisure travelers. But it may be that airlines maintain a higher overall fare structure on more business-oriented routes to deter leisure travelers from purchasing seats, which they expect to be able to sell more profitably to business travelers closer to flight date. Dummy variables for the first three of the four weeks in our observation period reveal a fairly steady ramping up of the price structure as the Canadian summer holiday traveling season builds up through the month of May. This is most pronounced, as would be expected, at the advance purchase end of the distribution (regression 3), rather than the last-minute market (regression 4). For the dummy STOPS indicating a nondirect itinerary, we were unsure a priori whether cost or demand (or neither or both) effects would show through in the pricing model. The results imply quite firmly that demand effects dominate. Overall, regression 1 shows that one-stop flights are offered at a 14% discount (exp[0.152] ¼ 0.86) to nonstops. And this discount is largest for day-before flight fares, where business travelers are prevalent. Given that costs will generally be higher for one-stop itineraries,

74

DAVID GILLEN AND TIM HAZLEDINE

we seem here to be observing another variant of price discrimination, with legacy carriers using the relatively greater unattractiveness to business travelers of unnecessarily longer journeys to separate them from lower value leisure customers. The coefficient on the dummy variable for the flights on the four routes in the sample that originate in the United States, and so can be seen as a RETURN flight from the Canadian perspective, seems to suggest that these flights are heavily discounted, at all times before flight date. However, we obtained the ‘‘fares’’ for these flights by subtracting the one way ex-Canada lowest fare from the lowest return fare offered.10 It seems likely, then, that despite the deregulation that has occurred in the North American industry, return fares are still being offered, in effect, at a discount to one-way fares for transborder itineraries. Plausibly, this represents another form of price discrimination based on the leisure/business traveler split: leisure travelers are more likely to be able to make the double commitment, in terms of flight time and date, required in the purchase of a return ticket. WJ’s fares are around 13% lower than other airlines, on average – a differential that widens, though not substantially, in the last two weeks before flight date. Given that AC’s domestic market share remains substantially higher than WJ’s (in this sample, 57% vs. 36% of withinCanada flights), the results imply that the incumbent legacy carrier is still able to charge a significant price premium over its ‘‘value-based’’ airline competitor. We see that the behavior illustrated in Figs. 1–5 is confirmed by the econometric results: the fringe LCCs have price distributions that are both lower and flatter than those of the legacy carriers, with prices that are around 20% lower for the first seven weeks, and which remain low right up to the week and day before travel, by which time the differential has widened to around 50% – that is, last-minute LCC fares are available for about one half the fares offered by the legacy carriers (exp[0.653] ¼ 0.52).

IMPLICATIONS AND CONCLUSIONS We have found that a data set built up from observations of the airfares offered on a travel agency web site yields interesting insights into pricing in the new world of LCC-inspired one-way tickets sold directly or near-directly to consumers, in a setting that allows easy comparison of the fares of competing airlines.

The New Pricing in North American Air Travel Markets

75

Perhaps, we could summarize the essence of our results as suggesting that the ‘‘new’’ airline pricing systems and transaction technologies actually continue to yield ‘‘old’’ market outcomes. Specifically, (i)

(ii)

(iii)

(iv)

(v)

Even having abandoned the Saturday night stay-over return ticket requirement that used to be deployed to good effect to discriminate between low-value leisure and high-value business travelers, the airlines are able to implement price discrimination schemes with results that would be spectacular in almost any other service industry. The main method of achieving price differentials based on willingness to pay is by raising the lowest fare as the flight date approaches. On average, we find, lowest available fares eight weeks out are less than one half the lowest fare offered the day before flight date, with most of the increase occurring during the last two weeks. We identify two other likely price discrimination practices. First, despite their higher cost, one-stop itineraries are systematically priced lower than nonstop flights, taking advantage of business travelers’ lower willingness to put up with the added inconvenience of longer journeys. And second, transborder return fares do appear to be discounted, perhaps also reflecting the lower ability of business travelers to commit to both legs of a journey in advance. Factors plausibly to be linked with shifts in market demand show through with significant effects on price. The airlines have some ability to predict which flights will sell out and price these higher. They steadily raised average fares through the month of May (2006), as the summer high travel season approached. On routes with a smaller proportion of leisure travelers, prices are set systematically higher. As for competition, we find that antitrust’s traditional structural measure, the Hirschman–Herfindahl Index, has predictive power: more concentrated routes do charge higher prices, other things equal. As well, we find that, controlling for the HHI, routes with one carrier offering more than 70% of seats facing competition from one or more much smaller airlines have lower prices, not higher as would be predicted by dominant-firm models. This suggests that a little competition can go a long way in terms of disciplining large firm pricing in this industry. We find that LCCs and their variants (such as WJ) have not come to dominate these air travel markets. They are present on many routes and do offer low fares, but these always coexist with significantly higher fares offered by legacy carriers for the same flights. In particular, LCC

76

DAVID GILLEN AND TIM HAZLEDINE

competition has not suppressed the importance of market structure for pricing at the leisure end of the market. On the contrary, structural competition for the leisure traveler appears to be more important than competition for the high-fare business traveler, where there are on many routes clear signs of parallel pricing between legacy carriers, even when three or four of these are serving the route. Our results have particular relevance to competition and competition policy in Canada as well as to international aviation policy. This is not just because the sample is made up of domestic and transborder Canadian flights, but, more fundamentally, because the nature of the Canadian market differs from the United States. The geography of the more northern country is such as to not encourage the use of the full hub-and-spoke system favored by legacy carriers in the United States. This means that most flights are nonstop, and all but one of the 39 routes in our sample (Moncton–Calgary) is served, nonstop, by a legacy carrier. Thus, LCCs do not get an advantage over legacy carriers from their point-to-point services. Also, and again largely due to geography, Canada has few secondary airports near major cities that can be served at lower cost by LCCs. For these and other reasons, it may be true that LCCs do not provide the same competitive discipline on legacy carriers in domestic and transborder routes that they may have exercised – Southwest Airlines, in particular – in the United States. Coupled with the apparent reality – again, geography-related – that the Canadian market seems to have difficulty in supporting more than two home-grown airlines (witness the recent departure of CanJet), there may be a pro-competition case for opening the market to entry by established carriers through extended ‘‘5th Freedom’’ Rights, which would allow a foreign carrier to fly between domestic Canadian points having started (or ending) its journey in its home market as well as a more aggressive pursuit of Open Skies agreements with major countries.11 Having more 5th Freedom agreements would enhance competition in domestic markets while more Open Skies agreements would provide an opportunity for the growth of the aviation industry and perhaps foster the entry of new carriers. In theory, price discrimination could mitigate the inefficiencies of market power because it decouples price from output restrictions. A perfectly discriminating monopolist, for example, would generate all of the possible surplus available under a demand curve, because it would be selling right up to the last customer willing to pay a dollar more than marginal cost. The fact that the monopolist would be itself appropriating all that surplus would be relevant to ‘‘old style’’ antitrust under which setting higher than

The New Pricing in North American Air Travel Markets

77

competitive prices to most customers is deemed undesirable in itself, but would not matter under modern net benefit criteria, which may be totally focused on how much surplus is generated, not who captures it – that is, allocative, not distributional efficiency. However, there may still be net benefit concerns about the implications of prices higher than average and marginal cost for production efficiency (X-inefficiency) and dynamic efficiency (innovation). Note too that the airlines seem to be able to effectively bifurcate the market into leisure and business/last minute travelers. Even just focusing on the leisure segment, our results imply quite substantial price differentials associated with differences in competitive conditions on routes, and these differentials will have the usual allocative efficiency implications. A number of questions remain open. Certainly one is the differences between markets and the extent to which we observe similar behavior under differing market structures. An example currently being explored is the apparent quite different extent of intertemporal price discrimination between Canadian markets and those in New Zealand and Tasman market. A second question is the extent to which differences in fares reflect vertical differentiation in the product, in which case the quoted fares should be adjusted for quality differences.

NOTES 1. The perceived and actual probability of Southwest eventually entering a route increases sharply when it begins operating from both end point airports on that route. 2. Round trip fares can still be purchased but the fact one-way fares are offered means airlines cannot enforce some fences they previously employed to separate customer groups. 3. Flights sometimes sell out before flight date and disappear from the web site. We actually observed flights daily in the last week, but do not use these data here, apart from the day-before fare, unless the flight sold-out, in which case we take the last observed fare offer. 4. The airlines are AC, Alaska, America West, American Airlines, CanJet, Continental, Delta, Harmony, Northwest, Skyservice, United, US Air, and WJ. Note that in September 2006, CanJet announced its exit from the scheduled passenger services business. 5. It should be noted that WJ is by now what could be called an ‘‘LCC þ .’’ It allows connections on its network, has a frequent flier program, and offers other features (in particular to attract business travelers) that are similar to the traditional legacy carrier model.

78

DAVID GILLEN AND TIM HAZLEDINE

6. The routes are Moncton–Calgary (WJ, AC, and CanJet), Ottawa–Vancouver (AC and WJ), Toronto–Calgary (WJ, AC, and CanJet), Toronto–Denver (AC and United), and Calgary–Phoenix (America West and Alaska). We use share of seats not flights because seats correlate with passengers and thus revenues that would be used by the competition agencies in measuring market share for dominance purposes. 7. The airlines offer itineraries on some routes that are many hours longer than the nonstop flight. We excluded any itineraries twice or more as time consuming than the nonstop trip. 8. The similarity in the DOMINANT and FRINGE coefficients is such that if the two variables are combined into one, the adjusted R2 of the regression model increases very slightly, in all five cases. However, since little else changes in the models, we show here the specification with the separate coefficients. 9. The variable has a maximum value of 90 (Toronto–Orlando) and a minimum of 40 (e.g., Toronto–Washington, DC). The mean value of 61.6% can be compared with the actual figure of 70% for the proportion of all Canada–US air travelers giving non-business reasons for their trip. Note that there are relatively few of the major US tourist destinations in our sample. 10. The Canadian-based web site Expedia.ca does not sell one-way tickets for ex-USA flights. 11. The sharp competition observed on trans-Tasman routes between Australia and New Zealand appears to be partly due to the aggressive use of 5th Freedom rights by Emirates.

ACKNOWLEDGMENTS We are indebted to Heibin Huang and Jixin Charlene Sun for excellent research assistance, and to the Auckland University Research Committee and the Canadian Studies program of the Government of Canada for financial support.

REFERENCES Aviation Daily (2008, various issues). Aviation Week. McGraw Hill. Borenstein, S. (1990). Airline mergers, airport dominance, and market power. The American Economic Review, 80, 400–418. Borenstein, S. (1991). The dominant firm advantage in multi-product industries: Evidence from the U.S. airlines. Quarterly Journal of Economics, 106, 1237–1266. Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the U.S. airline industry. Journal of Political Economy, 102, 653–683. Ciliberto, F., & Williams, J. W. (2010). Limited access to airport facilities and market power in the airline industry. Journal of Law and Economics, 53(3), 467–495.

The New Pricing in North American Air Travel Markets

79

Clougherty, J. A. (2006). The international drivers of domestic airline mergers in twenty nations: Integrating industrial organization and international business. Managerial and Decision Economics, 27, 75–93. Goolsbee, A., & Syverson, C. (2008). How do incumbents respond to the threat of entry? Evidence from the major airlines. Quarterly Journal of Economics, 123, 1611–1633. Hazledine, T. (2006). Price discrimination in Cournot-Nash oligopoly. Economics Letters, 93, 413–420. Hofer, C., Dresner, M., & Windle, R. (2004). Hub premiums in an era of low-cost carriers and financial distress, July 2, Address at the 2004 Air Transport Research Society World Conference in Istanbul, Turkey. Hurdle, G. J., Johnson, R. L., Joskow, A. S., Werden, G. J., & Williams, M. A. (1989). Concentration, potential entry, and performance in the airline industry. The Journal of Industrial Economics, 38, 119–139. Kim, E., & Singal, V. (1993). Mergers and market power: Evidence from the airline industry. The American Economic Review, 83, 549–569. Morrison, S., & Winston, C. (1990). Dynamics of airline pricing and competition. The American Economic Review, 80, 389–393. Morrison, S., & Winston, C. (1995). The evolution of the airline industry. Washington, D.C.: The Brookings Institution. Morrison, S., & Winston, C. (2000). The remaining role for government policy in the deregulated airline industry. In: S. Peltzman & C. Winston (Eds.), Deregulation of network industries: What’s next. Washington, D.C.: AEI-Brookings Joint Centre for Regulatory Studies. Peters, C. (2006). Evaluating the performance of merger simulation: Evidence from the U.S. airline industry. The Journal of Law and Economics, 49(October), 627–649. Stavins, J. (2001). Price discrimination in the airline market: The effect of market concentration. Review of Economics and Statistics, 83, 200–202. Swan, W., & Adler, N. (2006). Aircraft trip cost parameters: A function of stage length and seat capacity. Transportation Research E, 42, 105–115. Tretheway, M. E. (2004). Distortions of airline revenues: Why the network airline business model is broken. Journal of Air Transport Management, 10, 3–14. Tretheway, M., & Kincaid, I. (2005). The effect of market structure on airline prices: A review of empirical results. Journal of Air Law & Commerce, 70, 467–498.

80

DAVID GILLEN AND TIM HAZLEDINE

APPENDIX: CONSTRUCTING THE COSTS VARIABLE We tried three different ways of building up a costs variable. Two of these apply formulations developed in a recent article by Swan and Adler (2006). The third measure, which is used in the regressions noted in Table 1, is our own ‘‘block hour’’ measure based on Form 41 costs collected by the US DoT/FAA and reported in Aviation Daily (2008, various issues). Swan and Adler report two methods for constructing costs. First is what they term as a ‘‘planar-form cost function’’ based on engineering data, which specifies that aircraft trip costs are determined by seat numbers (S) and route distance (D). For single-aisle operations from 1, 000 to 5000 km, the relationship is as follows: C ¼ ðD þ 722Þ  ðS þ 104Þ  $0:019 And for longer haul twin-aisle operations: C ¼ ðD þ 2200Þ  ðS þ 211Þ  $0:0115 The latter function has more application to long-haul transocean flying, whereas our data covers markets that are essentially domestic flying; our longest route is 5,335 km and the average nonstop flight length is 1,683 km. These formulae are designed to incorporate only aircraft trip costs, which Swan and Adler state are around 50–60% of total costs, with general administrative costs around 30% and commissions and sales expenses the remainder. Note that the constants added in these expressions to both D and S ensure that there will be increasing returns to both these factors: doubling distance for a given number of seats, or doubling aircraft seat capacity for a given distance, will in both castes less than double the total costs of making the trip, so the costs per seat kilometer will fall. The other cost formula provided by Swan and Adler is a Cobb–Douglas cost function estimated as a log-linear regression on the same information as they used in constructing the planer cost functions. The two specifications estimated for narrow body and wide body operations, respectively, with cost per seat kilometer as the dependent variable, are as follows: c ¼ 2:44S 0:40 D0:25 and c ¼ 0:64S 0:345 D0:088

The New Pricing in North American Air Travel Markets

81

Note that in these equations, the impact of distance on costs is captured through a distance elasticity and varies between .09 and .25. This ‘‘cost taper’’ results from lower fuel burn at higher altitudes and spreading the costs of getting to altitude over more kilometers. Our third method builds up flight costs from available measures of ‘‘block hour’’ operating costs. The cost associated with operating aircraft on a route; referred to as flying operations cost (FOC) can be measured by the cost per block hour multiplied by the number of block hours required for the route: FOCs ¼ Bs  H s  f ðY s Þ where Bs is cost per block hour for the aircraft used, Hs are the block hours required for segment S, and f(Ys) is flight frequency that depends on number of passengers on segment, Ys. A route can have one or more segments and we examine a flight so frequency is 1. Obviously, the cost per block hour for a given aircraft and given segment will depend on the labor prices of the carrier as well as other input prices. However, the other input prices such as fuel, maintenance, and capital costs will not vary significantly across carriers for that segment. We constructed this cost measure using information submitted by carriers to the USDOT and assembled as ‘‘Form 41’’ data. These data are available by aircraft type by carrier by activity – for example, costs per block hour are available for United, Southwest, American, and USAir flying a Boeing 737–400 series. The data are also available on a cost per available seat mile (CASM) basis broken out for crew cost, fuel/oil, aircraft cost, insurance, taxes, and maintenance burden. The cost of operating an aircraft type by carriers using that aircraft were reported by carrier, and an average across carriers for an aircraft type is also reported. We had available all costs (2005) for all aircraft types used in our data set. The cost variable was measured as the block hour operating costs for the aircraft used times the block hours for a flight. These were divided by the seat km of the flight and expressed as ‘‘cost per seat km.’’ These represented flight costs. These costs can differ for the same airline flying the same route if it uses different aircraft type. Accordingly, we also calculated a route-based operating cost measure for each airline and route, which was a capacityweighted average of the individual flight costs. We will let the econometrics tell us which measure is the more appropriate. There are additional nonoperating costs attached to a flight. These include marketing and transaction costs, passenger and baggage groundhandling costs, and airport fees. We have little direct information on these

82

DAVID GILLEN AND TIM HAZLEDINE

costs, but we do know that they are incurred on a per passenger basis, so that excluding them from our costs measure would bias upwards the relative costs of longer flights. Therefore, we proceeded on a ‘‘top-down’’ basis, as follows. We calculated the total of operating costs for each airline, based on the block hour-based per flight costs we had already computed aggregated over all the flights observed, applied the Swan and Adler rule of thumb that these costs are around 20% of flight operating costs, divided by the total number of seats offered on all flights by the airline, to get marketing costs per seat, and then for each flight divided this number by flight distance and added to our flight operating cost measure to arrive at our total costs per seat kilometer measure. We thus have 12 possible measures of the cost per seat kilometer for a specific flight by a specific airline. There are the three different methods for constructing the operating cost variable, each of which can be measured at the flight or route level, and to which can be added or not the top-down estimate of marketing and ground costs. We found that neither of the Swan/Adler measures performed well econometrically. This may be because our routes are clustered near or below the bottom end of the flight distance range specified by Swan and Adler – 1,000–5,000 km. The nonlinearity built in to the Swan/Adler formulas could overestimate flight costs for these shorter flights. Also, we observe both single- and double-aisle aircraft flying transcontinental routes, but the shorter distance Swan/Adler formula is designed just for single-aisle aircraft. We therefore used the measure that we developed ourselves from the block hour data as the basis of our cost variable. Perhaps surprisingly, we found that adding our, admittedly quite crude estimate of marketing costs to the operating cost data improved the explanatory power of the variable, and we also found that prices were better explained by operating costs measured at the route, not the flight level.

CHAPTER 4 A GUIDE TO BOOKING AIRLINE TICKETS ONLINE Volodymyr Bilotkach and Nicholas G. Rupp INTRODUCTION This study traces the evolution of offered airfares on 50 busy routes on the US domestic market. Our approach differs from that in the literature in the following ways. First, we trace the lowest offered fares for specific roundtrip itineraries, acknowledging both that many trips involve return travel and that the round-trip airfare is often not equal to the sum of the two oneway fares. Many previous studies (e.g., Escobari, 2009; Escobari & Gan, 2007) either looked at fare quotes for specific one-way flights or examined the lowest round-trip quote available. Second, our sample of half of the top 100 domestic routes includes itineraries from markets with varying number of competitors as well as from markets with and without the presence of low-cost carriers (LCCs). Third, we have collected fare quotes simultaneously from three leading online travel agents. Thus, our research design allows us to see whether any systematic airfare differences exist across the different online distributors of travel services. We arrive at five stylized facts regarding the dynamics of airline price-offer curves. First, both fares and yields are consistently higher along the entire price-offer curve on less competitive markets and on the routes without LCC presence. Second, price changes are smoother on competitive routes than on markets with one or two competitors. Third, price drops are observed across Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 83–105 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003006

83

84

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

a spectrum of the markets and at any day before departure. In particular, at least one price drop was observed within 10 days before the flight for about half of all the round trips we tracked. In about one-third of round-trip itineraries, price drops within the last week before departure were observed. Fourth, the shape of the averaged price-offer curve is as expected – flat up to about three weeks before the flight departure and rising rapidly afterwards. Fifth and finally, we do not document systematic differences across online travel agents; however, differences in price quotes are not infrequent. From a consumer point of view, this chapter offers some insight into the optimal booking strategy for an air traveler. First, substantial price increases generally do not happen until three weeks before the flight date – this suggests that the optimal time to purchase a ticket for a price-conscious traveler is about three to four weeks before the departure date. Second, for airline passengers making last-minute purchases, waiting for future price reductions is a risky proposition, since the price trajectory clearly trends upward in the days immediately before departure. Hence, waiting for price changes has a negative expected value. Looking at cumulative price changes from fourteen to one day before departure in our sample, we find that the average price quote increases by 120% over the last two weeks before the flight departure date. At the same time, last-minute price drops are more common than conventionally believed. Finally, a simple cost-benefit analysis of checking fare quotes from different travel agents shows that shopping around appears justified when booking three or fewer weeks before departure. Overall, our message to the traveler is: if you book in advance, wait until about four weeks before the planned departure date and do not shop around. If you have to book closer to the departure date, do not wait, but do shop around. Studies analyzing samples of offered fares started re-emerging recently (the first such work can be traced to Stavins’ 2001 study of price dispersion in the US airline industry), and this literature is expected to grow at a fast pace in the near future. Among the topics addressed in this literature are price dispersion (Bilotkach, 2006; Escobari & Jindapon, 2011; Escobari & Gan, 2007; Gaggero & Piga, 2011; Giaume & Guillou, 2006), price dynamics (Button & Vega, 2007; Piga & Bachis, 2011), and fare comparison across booking engines (Chen, 2006; Clemons, Hann, & Hitt, 2002). Our contribution to this literature is methodological (we are tracing fare quotes for specific round-trip itineraries, which is not a common practice) and practical (we both analyze how price-offer curves differ depending on the underlying market structure and are able to provide some stylized facts to the traveling public). The rest of the study is organized as follows. The second section discusses the data collection process. The third section describes the data analysis

A Guide to Booking Airline Tickets Online

85

exercise. The fourth section concludes. A list of the markets included into our study appears in the appendix.

DATA COLLECTION Our data collection methodology assumes a week-long round-trip journey, which could be either a business trip or a visit to friends or relatives. We semirandomly selected 50 of the top 100 (based on nonstop passenger traffic) US domestic routes. Our route selection procedure was not exactly random, as we strove for inclusion of markets with varying degree of nonstop competition into the study. Thus, we first determined which markets were monopolies (as far as nonstop travel is concerned, at the airport-pair-market level) and duopolies, classifying the other markets as competitive. Then, we randomly selected routes out of each of these three subsets of markets. For each route, we randomly assigned whether we would track fares for morning (8 am), midday (noon), or evening (5 pm) departures. Airport-pair markets we have chosen reflect a variety of route competition options, from a single carrier to four carriers offering nonstop service. And perhaps more importantly, LCCs (Jetblue, Airtran, Frontier, Virgin America, and Southwest Airlines) are also present on some of the routes. In addition to being the leading low-cost airline in the United States,1 Southwest Airlines is notable for its ticket distribution strategy: among the major US carriers, Southwest is the only one that does not sell its tickets through the major online travel agents, relying primarily on its web site (see Bilotkach, 2010, for more detailed discussion of this aspect of Southwest Airlines’ strategy). The list of markets selected for our data collection exercise appears in the appendix. Fare quotes have been collected simultaneously through three leading online travel agents’ web sites. Online travel agents as a group sell about one quarter of all domestic airline tickets in the US market. The online travel agent segment of the ticket distribution market is dominated by the three major players: Travelocity (owned by Sabre, currently a privately held company), Expedia (founded within Microsoft in 1995, and an independent publicly traded company since 2005), and Orbitz (started through a partnership of several major airlines in 2001, currently a subsidiary of Travelport, owned by the Blackstone Group – a private equity company). According to the US Department of Transportation, in 2002, 28.5% of all bookings with online travel agents were on Travelocity; 28.7%, on Expedia, and 21.3% on Orbitz, for a total of 78.5% of all US travel reservations through an online travel agent.2 We chose to collect prices from the three different sources in light of studies suggesting one could observe substantial

86

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

differences in fare quotes across the major travel agents (Bilotkach & Pejcinovska, 2011; Bilotkach & Rupp, 2010). Instead of collecting the lowest available round-trip fare, as other similar studies have done, we focus on specific round-trip itineraries. This means we trace the lowest offered prices for a specific flight number beginning eight weeks before departure. Fares are pulled weekly up to two weeks before departure; then, fares are collected approximately every three days within two weeks of departure. If a particular flight number is no longer available in the airline’s schedule, then we select the same airline’s flight operating as close as possible to the original flight. Only a few such substitutions were made with these occurring typically well in advance of the departure date. The flight departure dates are Wednesday, June 17, 2009, and Friday, June 19, 2009, with the return flight a week later. For each of the markets included into our study, we have picked one specific round-trip itinerary for each airline operating nonstop service on the route. For example, on the Atlanta–Newark route, Delta, Air Tran, and Continental operate nonstop flights on this route; hence, we collected the lowest fare offered for a specific flight for three itineraries in this market. Given that we track fare quotes simultaneously for three online travel agents’ web sites, each day of data collection for this route results in nine fare quotes (three airlines  three travel agents). As 50 markets we selected varied in terms of the number of competitors, we ended up tracking 105 specific round-trip itineraries, representing all of the major carriers except Southwest Airlines. Details are in the appendix. Fare quotes were collected 56, 48, 42, 35, 28, 21, 14, 10, 7, 4, and 1 day before the scheduled flight date. We increased the frequency of data collection starting two weeks before departure, as recent evidence from the data set of actual ticket purchases (Puller, Sengupta, & Wiggins, 2009) suggests that most of the actual ticket purchases occur at the last minute. In the end, for each round-trip itinerary, we have obtained three so-called price offer curves or profile of fare quotes as the departure day nears. Analysis of these data is presented in the following section.

DATA ANALYSIS General This section addresses the following issues: (i)

Trends of fare quotes as the flight date approaches. Specifically, we would like to know the following:  Do fares typically rise as the departure date nears?

A Guide to Booking Airline Tickets Online

(ii)

(iii)

87

 Does route competition including the presence of LCCs affect either the level or dynamics of price changes before departure? Instances of price drops as the departure date approaches. In particular, we are interested in the following questions:  How frequent are price reductions?  Are price drops more common the closer to the departure date? Differences across online travel agents. The relevant questions include the following:  Are there systematic differences in fare quotes for travel agents?  Are fare differences across travel agents more prevalent the closer to the departure date? We expect to find more differences for two reasons. First, consumers buying last-minute tickets may not have time to shop around. Second, airline-agent contracts may reserve the best (lowest) last-minute fares with a particular preferred online travel agent, hence the discrepancy in offered fares.

Our goal is to inform the traveling public on the behavior of offered prices across time and hence help customers make more informed decisions on the optimal time to book airline tickets.

Trends in Price Quotes Fig. 1 presents the general trend of averaged offered fares as the departure date nears. It is clear from this figure that the average quotes are rather flat up to 28 days before the flight departure; afterwards, a clear upward trend is observed. One lesson for travelers is that the lowest fares typically are offered for passengers booking tickets more than a month in advance of departure. We observe little price differences between fares offered many weeks in advance, as compared to fares four weeks before departure. Substantially lower fares, however, are unlikely in the final four weeks before departure. In sum, for passengers considering whether to book a ticket more than four weeks in advance, it may be beneficial to wait before buying, as future price reductions are possible (we will discuss this issue in more detail later), while significant price increases are unlikely. Descriptive analysis of both fare quotes and corresponding yields (fare per mile flown) relative to the market structure produces rather expected results. Specifically, Fig. 2 provides a graphical representation of the

Fare ($)

88

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP $600 $550 $500 $450 $400 $350 $300 $250 $200 $150 $100

56

49

42

35

28

21

14

10

7

4

1

Days prior to departure

Fig. 1.

Average Fare Quotes.

0.40

Yield (fare/miles)

0.35 0.30 0.25 Diff

0.20

95% CI 0.15

95% CI

0.10 0.05 0.00 56

Fig. 2.

49

42

35

28

21

14

10

7

4

1

Difference in Average Yields: Monopoly and Non-Monopoly Routes before Departure.

difference in average yields based on whether the route is served by a monopolist: Difference ¼ Yieldt ðMonopolyÞ  Yieldt ðNon-monopolyÞ where t represents the date before departure. We conduct a similar calculation for the yield differences for routes with and without a LCC (and Southwest Airlines) presence: Difference ¼ Yieldt ðLCCÞ  Yieldt ðNon-LCCÞ

A Guide to Booking Airline Tickets Online

89

0.30

Yield (fare/miles)

0.25 0.20 Diff

0.15

95% CI 95% CI

0.10 0.05 0.00

Fig. 3.

56

49

42

35

28

21

14

10

7

4

1

Difference in Average Yields for Routes without and with LCC Presence.

We present both differences between the means and the corresponding 95% confidence intervals. The 95% confidence interval on Fig. 2 indicates significant differences for average yields on monopoly versus non-monopoly routes every day that sample fares are collected before departure. Fig. 3 shows that yields are also significantly higher in markets without the presence of a LCC carrier (i.e., Southwest, Frontier, Jetblue, Air Tran, and Virgin America). Fig. 4 presents a similar depiction of yield differences on routes served by Southwest Airlines compared to non-Southwest routes. In each of Figs. 2–4, we note that the corresponding differences in yields actually increase as the departure day nears. Comparing how these differences in yields change before departure reveals that the largest changes occur on monopoly routes versus non-monopoly markets, with the difference in yields rising from 5 cents per mile 21 days in advance to over 25 cents per mile the day before departure (a five-fold increase in the yield differential). In comparison, difference in yields for non-LCC versus LCC routes increases by a smaller magnitude as departure date nears, rising from 5 cents (21 days in advance) to about 20 cents (1 day before departure). To put this into perspective, fare quote for a 2,000-mile round-trip journey (such as a trip from Washington, DC, to Miami) the day before the flight would be about $400 higher if no LCCs were present on the route. An even smaller increase occurs when comparing non-Southwest routes to routes where Southwest is present; the yield difference rises from about 5 cents (21 days before departure) to about 17 cents (a day before departure). That is, not only are travelers on routes without a LCC presence asked to pay more per mile flown; the differential increases as the departure date nears.

90

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

We also detect rather curious facts about the dynamics of price changes. Looking at the percentage changes in fare quotes as the departure date nears, Fig. 5 shows that the presence of an LCC does not make the trend much smoother than its absence. In comparison, the corresponding graph from Fig. 6 looks much smoother for the competitive markets (markets with three or four airlines offering nonstop services). Figs. 5 and 6 also demonstrate that the largest percentage increases in offered fares are observed 10 and 4 days before the flight’s departure.

0.30

Yield (fare/miles)

0.25 0.20 Diff 95% CI 95% CI

0.15 0.10 0.05 0.00

Fig. 4.

56

49

42

35

28

21

14

10

7

4

1

Difference in Average Yields for Routes with and without Southwest Airlines’ Presence.

Percentage change

60 50 40 30

All

20

LCC NoLCC

10 0 -10

0

10

20

30

40

50

Days before departure

Fig. 5.

Dynamics of Price Changes for Routes with and without LCC.

A Guide to Booking Airline Tickets Online

91

Percentage change

70 60 50 40 Monopoly

30

Duopoly

20

Competitive

10 0 -10

0

Fig. 6.

10

20

30

40

50

Days before departure

Dynamics of Price Quotes Depending on Market Structure.

Price Drops Throughout this analysis, we will define price drops as more than a $1 reduction in fare quotes between the two dates of data collection. We did not consider cases where price quotes were unavailable on the next date of collection as price drops. The following stylized facts stand out. First, drops in offered fares as the departure date nears are not uncommon. In fact, in our sample, we have observed price drops both 49 days and 1 day before the scheduled departure of the flight (and for every single day of data collection in between). At the same time, price drops are clearly more likely to occur 28 or more days before the flight’s departure. This is visible from Fig. 7. Combined with our previous observation that prices do not start increasing until about 21 days before departure; we can say that it generally makes sense to wait until four weeks or so before the planned trip date before buying the ticket. Even if you have to make the booking two to three weeks in advance or sooner, waiting for the price drop can pay off. This decision, however, is like playing a lottery with a negative expected value, since the price is more likely to increase than to decrease. The second stylized fact is that we observed at least one price drop over the data collection process for most of the itineraries included into our analysis. Fig. 8 shows that no price drops have been observed for only about 8% of all the itineraries (more specifically, no price drops were observed for

92

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

Proportion of itineraries

0.90 0.80 0.70 0.60 0.50 decrease

0.40

no chg

0.30

increase

0.20 0.10 0.00 49

42

35

28

21

14

10

7

4

1

Days before departure

Fig. 7. Share of Itineraries with Decreasing, Increasing, and Unchanging Price Quotes Observed during the Sample Period (56 Days to 1 Day before Departure).

Share of Itineraries with Drops

0.35 0.3 0.25 0.2 Expedia

0.15

Orbitz

0.1

Travelocity

0.05 0

0

1

2

3

4

5

6

Number of Price Drops per Itinerary

Fig. 8.

Price Drops by Itineraries and Agents for Fares Collected Up To 56 Days in Advance.

six itineraries on Orbitz, eight on Expedia, and nine on Travelocity). The modal number of price drops per itinerary is two for all travel agents; the median is two for Expedia and Travelocity; three for Orbitz. This means that for about half of the itineraries in our sample, the price went down

A Guide to Booking Airline Tickets Online

93

Proportion of itineraries

0.25 0.20 0.15 < 0.3 mkt share 0.10

0.3-0.5 mkt share 0.5+ mkt share

0.05 0.00 49

42

35

28

21

14

10

7

4

1

Days before departure -probability of a weekly price reduction

Fig. 9.

Likelihood of Price Drops Depending on Carrier’s Market Share on the Route.

three or more times as the departure date approached. The maximum number of price drops observed for an itinerary was six (recorded for one itinerary on Expedia and Travelocity). Looking at the relationship between the airline’s market share and the likelihood of a price drop, we observe that more dominant carriers on the route tend to decrease their fare quotes less frequently. Differences in the frequency of price drops are especially visible for carriers with 50 þ percent market share 28, 14, and 7 days before the flight departure (Fig. 9). We noted above that we have tracked fare quotes for specific round-trip itineraries. We have also determined how many nonstop flights operate on the day of the flight on each of the markets under consideration, at both the airline and the market levels. Figs. 10 and 11 demonstrate that fewer price drops occur on ‘‘thinner’’ markets (those with fewer flights) closer to the departure date. Interestingly, one to four weeks before departure, airlines that operate more flights on the route are less likely to drop their fares compared to carriers operating fewer daily flights.

Differences across Agents Next, we turn to price differences across the three major online travel agents. We find that on average, agents offer nearly identically priced

Proportion of itineraries

94

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP 0.30 0.25 0.20 0.15

< 14 flights

0.10

14-20 flights 21+ flights

0.05 0.00 49

42

35

28

21

14

10

7

4

1

Days before departure -probability of a weekly price reduction

Proportion of itineraries

Fig. 10.

Total Daily Flights on Route and Price Reductions.

0.25 0.20 0.15 < 5 flights 0.10

5-8 flights 9+ flights

0.05 0.00 49

Fig. 11.

42

35 28 21 14 10 7 Days before departure -probability of a weekly price reduction

4

1

Carrier Daily Flights on Route and Price Reductions.

itineraries. Differences in fare quotes, however, were not uncommon, especially within three weeks of departure. Indeed, one day before departure, only 7 out of 10 itineraries are similarly priced by all three travel agents (meaning that the difference in fares quoted by any two of the three agents is within $5). Again, these observations are visualized on the figures below.

A Guide to Booking Airline Tickets Online

95

From Fig. 12, we can see that Expedia does come out as somewhat of a ‘‘low price leader’’ for fare quotes 14 and 7 days before the flight. In absolute terms, however, this ‘‘leadership’’ does not amount to any significant difference in prices. Fig. 13 shows the frequency of similar fares by all three travel agents across airlines. Two figures are reported for each day of data collection: the share of itineraries relative to all itineraries we tracked (105) and the same

$550 $500 Fare ($)

$450 $400 Travelocity

$350

Expedia

$300

Orbitz

$250 $200

56

49

42

35

28

21

14

10

7

4

1

Days prior to departure

Proportion of Itineraries

Fig. 12.

Average Fare Quotes by Different Agents.

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1day

4days 7days 10days 14days 21days 28days 35days 42days 49days 56 days Where all agent report a quote

Fig. 13.

Relative to all itineraries

Share of Itineraries Where All Agents Report Fare Quotes within $5 of Each Other.

96

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

relative to the number of itineraries for which all three travel agents reported a fare quote. Note that in some cases (more so closer to the departure date), some travel agents did not report any fare quotes for the specific flight combination we have chosen. Our guess is that this could be due to differences in travel-agent-airline contracts regarding presentation of overbooked flights. For the purpose of constructing Figure 13, we treated fare quotes within $5 of each other as identical. It is visible that agents report nearly identical fare quotes for the same itineraries about 80% of the time, or at least in 7 out of 10 cases where all agents report quotes for the itineraries we have been tracking. The largest share of itineraries with identical quotes occur 10, 28, and 35 days before the flight’s departure. Fare discrepancies are somewhat more prevalent closer to the flight departure date. As we noted above, a day before departure date, we observe either differences in fare quotes ($5 þ ) or some agents not offering fares for about 7 out of 10 itineraries.

Cost-Benefit Analysis of Searching for Additional Fare Quotes One lesson for the traveler from this analysis is that even though on average all travel agents report identical fare quotes, it still may make sense to shop around. Checking one more travel agent’s web site is relatively costless, and you might find a different (not necessarily a better) deal less than 20% of the time. Hence, the expected benefit still appears to outweigh the costs. For example, consider a case of booking a round trip from Atlanta to Boston with Delta Air Lines seven days before departure. In our sample, the fare quoted by Travelocity for that trip was $323, while both Orbitz and Expedia quoted $313, or $10 lower fare. A customer starting his/her search from Travelocity would save $10 if he/she checked other travel agents’ web sites. The marginal cost of performing this additional check is $2 (based on the assumption that consumers need three minutes per each additional fare check, and the consumer’s time is valued at $20 per hour). Of course, our hypothetical traveler would not gain anything if he/she started searching with Orbitz or Expedia to begin with. Yet, assuming the first site to check the fare quotes is chosen randomly, it still makes sense to perform additional searches, in our Atlanta–Boston trip example. One-third of the time, you gain $8 ($10 difference in fares net of $2 marginal cost of search), and with probability of two-thirds, you lose $2, so the expected value is positive. Additional search, however, does not yield positive expected value in our example for a traveler with a time valuation of $33.33/hour or higher.3

A Guide to Booking Airline Tickets Online

Table 1.

Potential Savings from Shopping Around.

Days before Flight

56 49 42 35 28 21 14 10 7 4 1

97

Savings from Search

Highest Value of Time for Searching, $/hour

Average

Standard deviation

Maximum

$9.22 $4.13 $6.61 $1.97 $0.55 $7.36 $9.72 $8.47 $24.50 $6.12 $5.62

$28.47 $31.32 $32.85 $11.31 $3.14 $34.97 $25.91 $42.56 $87.85 $34.09 $25.83

$182.00 $316.00 $316.00 $82.00 $30.00 $311.00 $150.00 $311.00 $667.00 $321.00 $195.00

$30.72 $13.78 $22.03 $6.57 $1.84 $24.54 $32.42 $28.23 $81.68 $20.42 $18.73

Notes: ‘‘Highest value of time for searching’’ refers to the highest value of time (per hour) for which marginal benefit of checking quotes across three travel agents exceeds marginal cost, given the average savings from the search. Assumes three minutes for checking an additional fare quote and supposes two out of three agents offer the lowest fare. See also note 3.

Averaged across 105 itineraries for fare quotes 7 days before the flight departure, maximum hypothetical savings from searching around is $24.50. This means that on average, a traveler with value of time of $81.68 or less should continue searching 7 days before the departure, assuming two out of three agents offer the lowest fare quote for a specific itinerary. To see how much a hypothetical traveler can save from searching around, Table 1 presents the average highest potential savings from checking all three travel agents’ sites (defined as the absolute difference between the highest and the lowest quote for a given itinerary on a given day). Savings are clearly the lowest for fare quotes collected 35 and 28 days before departure, and the highest for quotes obtained 7 days before the flight date. Overall, average savings from searching across travel agents appear modest; however, given low marginal cost of search, a traveler with value of time of $15/hour will be advised to shop around when booking a flight three weeks to one day before departure. Combining results reported in Table 1 with the previously outlined information on dynamics of fare quotes, we can make the following suggestions to the traveler. If you are booking in advance, wait until about four weeks before the planned flight date and take the first price quote you encounter (assuming you book with one of the three major online travel

98

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

agents included in this study). If you are booking three weeks or sooner before your planned departure date, do not wait (as prices will rise rather than fall); however, do shop around – the money you may save will likely be worth the effort.

The Dynamics of Price Changes The analysis of changes in price quotes and differences across travel agents is summarized in the simple regression analysis reported below. Using the magnitude of the change in the fare quote between the two consecutive days (PiPj) of data collection, we explore whether market structure, airline market share, or presence of LCCs have any effect on fare changes. Specifically, we estimate the following model: ðPi  Pj Þ ¼ a þ b Mkt:Share þ d Tot:Flights þ Z Mkt:Structure þ x LCC þ o Agents þ l Airlines þ  where Mkt.Share represents a carrier’s market share on the route; Tot.Flights are the total daily nonstop flights on the route across all carriers; Mkt.Structure is an indicator variable for routes served by a single carrier (monopoly) or two carriers (duopoly); LCC is another indicator variable taking the value of 1 on routes where a LCC is present; Agents is an indicator variable for the travel agent that offered the fare quote; and Airlines is an indicator variable for each of the 11 carriers whose fares we observe. We ran 10 regressions, corresponding to 11 days of data collection. Results reported in Table 2 note that carriers with larger market share are more likely to increase prices before departure. We observe significant price increases 49, 28, and 7 days before departure for the carriers with larger route market share. The overall size of the airline market also influences the dynamics of price changes before departure. We observe slightly lower prices (ranging from $1.38 to $3.51) on routes with more daily nonstop service offerings 49, 28, 14, and 4 days before departure. Market structure does appear to influence price changes with modest price cuts ($62 to $70) offered on monopoly routes well before departure (49 and 28 days in advance) followed by steep price increases ($123) 10 days before departure. A similar yet smaller magnitude price changes occur for duopoly routes with price reductions ($26) coming 35 days in advance followed by a nearly equal and offsetting price increases 14 days in advance

(1) (P49–P56) DP49 Carrier market share Total daily flights Monopoly Duopoly Route_LCC Travelocity Expedia Constant n R2

Examining the Dynamics of Airfare Pricing before Departure. (2)

(3)

(P42–P49) (P35–P42) DP42 DP35

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(P28–P35) DP28

(P21–P28) DP21

(P14–P21) DP14

(P10–P14) DP10

(P7–P10) DP7

(P4–P7) DP4

(P1–P4) DP1

69.57 22.89 1.74 62.31 4.66 46.10 3.81 141.61 81.62 8.38 (23.20) (25.15) (30.39) (23.38) (24.07) (31.86) (34.88) (48.45) (94.75) (67.60) 0.50 0.96 1.38 0.07 3.51 2.33 2.22 3.39 1.03 1.57 (0.50) (0.64) (0.79) (0.68) (0.56) (0.88) (0.95) (1.68) (1.86) (1.06) 18.16 22.85 45.19 2.50 1.14 122.86 55.41 8.17 4.69 37.65 (21.49) (21.60) (33.54) (21.26) (22.28) (30.63) (42.25) (68.75) (80.97) (55.09) 8.64 8.36 26.63 23.98 4.02 75.53 34.63 0.67 0.78 26.05 (9.56) (12.17) (14.51) (8.41) (9.06) (15.83) (17.20) (21.84) (36.45) (26.90) 8.26 4.75 22.81 10.03 37.17 5.36 39.75 125.98 19.91 14.46 (7.70) (7.66) (12.25) (7.71) (7.14) (12.63) (14.26) (27.03) (28.44) (22.83) 1.70 2.03 1.65 1.70 3.72 4.67 9.10 7.88 17.86 0.50 (6.67) (6.73) (8.26) (8.00) (6.56) (10.01) (13.04) (19.37) (23.64) (16.17) 2.25 1.24 1.48 0.08 3.34 3.44 1.07 0.25 18.36 1.64 (6.44) (6.84) (8.56) (8.10) (6.13) (10.11) (12.85) (18.22) (24.06) (16.98) 56.16 249.28 16.52 25.63 33.06 19.14 1.21 56.06 79.06 54.09 (17.24) (23.06) (20.03) (20.70) (15.05) (25.53) (37.02) (37.54) (68.25) (47.18) 311 303 302 311 308 311 313 305 288 271 0.2167 0.1518 0.1292 0.117 0.2001 0.2237 0.271 0.1412 0.2164 0.1513

99

Notes: Robust standard errors appear in parentheses. Carrier dummies for each of the 11 individual airlines are also included in estimations above, yet do not appear for the sake of brevity. Fifty routes were randomly selected from the top 100 busiest domestic nonstop routes based on passenger traffic. We observed fare quotes for the three major online travel sites on particular flights by carrier on the following days before departure: 56, 49, 42, 35, 28, 21, 14, 10, 7, 4, and 1 before departure. The sample size fluctuates since some travel agents do not provide a fare quote for that particular flight. This was more common the week before departure. Statistical significance at 10% level. Statistical significance at 5% level. Statistical significance at 1% level.

A Guide to Booking Airline Tickets Online

Table 2.

100

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

with the largest price increase ($76) coming immediately before departure (4 days). In sum, the largest monopoly and duopoly effects occur just before departure with monopoly fares escalating 10 days before departure and duopoly routes experiencing the largest price increases 4 days before departure. The presence of a LCC on the route has an influence on the pricing dynamics. Well in advance of departure (49, 28, and 14 days), LCC routes exhibit modest price increases ranging from $14 to $37 (see Table 2). This upward pattern, however, reverses immediately before departure as we find significantly lower prices ($126) just four days before departure, which more than offsets the higher observed prices in advance of departure. LCCs clearly influence last-minute fares, this is the most noticeable effect of LCCs on a route. In addition, the inclusion of carrier dummies reveals that JetBlue pricing consistently (in 6 of 10 regressions in Table 2) and significantly lowered its prices compared to the omitted carrier (American). Finally, we observe no systematic differences in offered prices at any time before departure among any of the three major online travel agents as no individual travel agent offers significantly higher (or lower) fares at any time before departure in our sample.

CONCLUSIONS This chapter examined the dynamics of offered airline prices for flights between 1 day and 56 days before departure. Tracking the same flight numbers and itineraries across the three major online travel agents for 50 of the top 100 busiest US domestic nonstop routes, we were able to make some inferences about the dynamics of airline price-offer curves. We found five stylized facts for airline pricing. First, both fares and yields are consistently higher along the entire price-offer curve on less competitive markets, and on routes without LCC presence. Second, price changes are smoother on competitive and LCC routes than on markets with one or two competitors. Third, price drops are observed across a spectrum of markets and at any day before departure. In particular, at least one price drop was observed within 10 days before the flight for about half of all round trips tracked. In about one-third of round-trip itineraries, price drops within the last week before departure were observed. Fourth, the shape of the averaged price-offer curve is as expected – flat up to about three weeks before the flight and rising rapidly afterwards. Fifth and finally, we do not document systematic

A Guide to Booking Airline Tickets Online

101

differences across online travel agents; however, differences in price quotes are not infrequent. These findings suggest that consumers looking for airline tickets well in advance from the departure date may benefit from waiting to make an airline purchase. As the departure date nears (especially within three weeks), fares are much more likely to rise than fall, so we encourage potential passengers not to wait for further price drops within the final 21 days before departure. Finally, we do encourage passengers to seek a second (or third) price quote when making last-minute ticket purchases since these fares exhibit the most discrepancies among online travel agents.

ACKNOWLEDGMENTS We thank Jan Brueckner and participants at the International Industrial Organization Conference for their useful comments. Nicholas Rupp gratefully acknowledges the support of East Carolina University’s Research and Creative Activities Grant.

NOTES 1. At the time that these data were collected this statement is correct; however, as of January 2011, American Airlines is no longer selling its tickets on either Expedia or Orbitz due to a contract dispute with these travel agents. 2. Travel agency market share data are from ‘‘Computer Reservations System Regulations: Final Rule,’’ 14 CFR Part 255. 3. For $Y of maximum savings from checking two additional web sites, assuming that the marginal cost of searching two other travel agents’ sites is $X, and supposing, as in our example, that a traveler picks the first site randomly, and that two agents offer the lowest fare for a given itinerary, searching provides a positive expected value if 1/3(YX)(2/3)XZ0. This gives Xr(y/3) or the corresponding value of time of 10Y/3 per hour (recall we assumed that searching two other agents’ web sites takes a total of six minutes). This cutoff value of time will be higher, if we suppose that only one agent offers the lowest fare quote.

REFERENCES Bilotkach, V. (2006). Understanding price dispersion in the airline industry: Capacity constraints and consumer heterogeneity. In D. Lee (Ed.), Advances in airline economics (Vol. 1, pp. 329–345). Cambridge, MA: Elsevier.

102

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

Bilotkach, V. (2010). Reputation, search cost, and airfares. Journal of Air Transport Management, 15, 251–257. Bilotkach, V., & Pejcinovska, M. (2011). Distribution of airline tickets: A tale of two market structures. In J. Peoples (Ed.), Advances in airline economics (Ch. 5, pp. 83–105). Bingley, UK: Emerald Group Publishing Limited. Bilotkach, V., & Rupp, N. (2010). Do consumers benefit from low-price guarantees? Evidence from the airline industry. University of California, Irvine Working Paper, Irvine, CA, USA. Button, K., & Vega, H. (2007). Uses of the ‘‘Temporal-Fares-Offered Curve’’ in air transportation. Journal of the Transportation Research Forum, 46(2), 83–99. Chen, J. (2006). Differences in average prices on the Internet: Evidence from the online market for air travel. Economic Inquiry, 44, 656–670. Clemons, E. C., Hann, I-H., & Hitt, L. M. (2002). Price dispersion and differentiation in online travel: An empirical investigation. Management Science, 48, 534–549. Escobari, D. & Jindapon, P. (2011). Price discrimination through Refund Contracts in airlines. Working Paper. University of Texas, Pan American. Escobari, D. (2009). Systematic peak-load pricing, congestion premia and demand diverting: Empirical evidence. Economics Letters, 103(1), 59–61. Escobari, D., & Gan, L. (2007). Price dispersion under costly capacity and demand uncertainty. NBER Working Paper 13075. Gaggero, A., & Piga, C. (2011). Airline market power and intertemporal price dispersion. Journal of Industrial Economics. Forthcoming. Giaume, S., & Guillou, S. (2006). Concentration, market share inequality and prices: An examination of European airline markets. In D. Lee (Ed.), Advances in airline economics (Vol. 1, pp. 273–296). Cambridge, MA Elsevier. Piga, C., & Bachis, E. (2011). Low cost airlines and online price dispersion. International Journal of Industrial Organization, 29(6), 655–667. Puller, S., Sengupta, A., & Wiggins, S. (2009). Testing theories of price dispersion and scarcity pricing in the airline industry. NBER Working Paper 15555.

A Guide to Booking Airline Tickets Online

103

APPENDIX: LIST OF MARKETS Origin

Destination

Monopoly (nonstop routes) Atlanta Savannah, GA Atlanta Salt Lake City Charlotte, NC Tampa Cincinnati Atlanta Denver Washington, DC (IAD) Miami Dallas (DFW) New Orleans Houston (IAH) New York (LGA) Charlotte, NC New York (LGA) Dallas (DFW) San Diego Dallas (DFW) Duopoly (nonstop routes) Atlanta Boston Atlanta Baltimore Atlanta Orlando Atlanta Pittsburgh Atlanta Ft. Myers, FL Chicago (ORD) Dallas (DFW) Chicago (ORD)

Los Angeles

Chicago (ORD)

San Francisco

Chicago (ORD)

St. Louis

Detroit

New York (LGA)

New Orleans New York (LGA)

Atlanta Chicago (ORD)

Orlando

Detroit

Phoenix

Dallas (DFW)

San Francisco

Atlanta

Airline(s)

Flight Date

Delta Delta US Airways Delta United

17 17 17 19 17

June June June June June

American Continental US Airways American American

19 19 19 19 19

June June June June June

Delta, AirTran Delta, AirTran Delta, AirTran Delta, AirTran Delta, AirTran American, United American, United American, United American, United American, Northwest Delta, AirTran American, United AirTran, Northwest American, US Airways Delta, AirTran

17 17 17 17 17 19

June June June June June June

19 June 17 June 17 June 17 June 19 June 19 June 17 June 19 June 19 June

104

VOLODYMYR BILOTKACH AND NICHOLAS G. RUPP

APPENDIX (Continued) Origin San Francisco Tampa

Destination Washington, DC (IAD) Atlanta

Competitive (nonstop routes) Atlanta Newark Chicago (ORD)

Boston

Chicago (ORD)

Seattle

Dallas (DFW)

Denver

Denver

Las Vegas

Denver

Kansas City

Denver

San Diego

Denver

San Francisco

Denver

Los Angeles

Denver

Minneapolis

Denver

Salt Lake City

Detroit

Chicago (ORD)

Houston (IAH)

Denver

Las Vegas

Seattle

Airline(s)

Flight Date

United, Virgin America Delta, AirTran

19 June

Delta, AirTran, Continental American, United, Jetblue American, United, Alaska American, United, Frontier United, Frontier, Southwest United, Frontier, Southwest United, Frontier, Southwest United, Frontier, Southwest United, Frontier, American, Southwest United, Frontier, Northwest, Southwest United, Frontier, Delta, Southwest American, Northwest, United Continental, United, Frontier Alaska, US Airways, Southwest

17 June

19 June

19 June 17 June 19 June

17 June 17 June 17 June 17 June 17 June

17 June

17 June

17 June

19 June

17 June

A Guide to Booking Airline Tickets Online

105

APPENDIX (Continued) Origin

Destination

Los Angeles

Washington, DC (IAD)

Newark

Chicago (ORD)

Philadelphia

Chicago (ORD)

Phoenix

Los Angeles

Phoenix

Chicago (ORD)

San Francisco

San Diego

Seattle

Phoenix

Seattle

Denver

Seattle

Los Angeles

Airline(s) American, United, Virgin America American, Continental, United American, United, US Airways United, US Airways, Southwest American, United, US Airways United, Virgin America, Southwest Alaska, US Airways, Southwest Alaska, Frontier, United, Southwest Alaska, United, Virgin America

Flight Date 19 June

17 June

19 June

19 June

19 June

19 June

19 June

19 June

19 June

Notes: Southwest Airlines does not offer fares through any of the three major online travel agents, hence we do not have Southwest fare data. We do, however, observe which routes Southwest flies, hence they are included in the above route competition characterization. Airport codes used are as follows:

 ORD: Chicago O’Hare  IAD: Washington Dulles  IAH: Houston George Bush Intercontinental  DFW: Dallas-Ft. Worth International  LGA: New York LaGuardia

CHAPTER 5 DISTRIBUTION OF AIRLINE TICKETS: A TALE OF TWO MARKET STRUCTURES Volodymyr Bilotkach and Marija Pejcinovska INTRODUCTION The vertical relationships literature has considered situations where both producers and retailers have a degree of market power. On one hand, retailers may have certain freedom in deciding what price to charge to the final consumers. On the other hand, large retailers may also pressurize producers (Wal-Mart is a classic example; see also Comanor and Rey (2000) for a formal treatment of this topic). At the same time, producers may pressurize retailers via resale price maintenance. The interplay of producers’ and retailers’ bargaining power, in addition to the structure (both horizontal and vertical) of product and distribution markets, eventually determine the sticker price faced by an unsuspecting consumer. In the end, the final sticker price consumers will observe can be affected by the structure of the distribution market as much as (if not more than) by the competition on the market for the good or service itself. The price you will pay for your next car will be determined not only by how many car manufacturers are out there, but also by how many dealerships there are in your area (other factors, such as your ability to bargain, will also play a role).

Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 107–138 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003007

107

108

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

On some important markets, the role of retailers appears to be limited to mere distribution of the product or service, at prices determined by the producers. The producers use such retailers to ensure that as many customers as possible are aware of the product or service; the retailers make money out of commission paid by producers, or by charging the final customers a fee. One such market is that for distribution of travel-related services (airline tickets, hotel rooms, rental cars, etc.). Producers on this market retail their services both independently and via travel agents, linked to the computer reservation systems. Retailers do not specifically mark up the producer’s price (they can and do charge consumers some booking fees – see next section for more details). It therefore appears that the role of travel agents on this market is ‘‘technical,’’ so the prices consumer observes should be determined only by the competition between the producers. In this chapter we challenge this contention and show that in fact the structure of the distribution market matters. Analyzing airlines’ and distributors’ incentives given the structure of the market, we can suggest the following. First, individual distributors will have a strong incentive to obtain exclusive rights for distribution of the airlines’ discounted tickets. Second, the competing airlines can potentially exploit their repeated interaction and presence of the less informed customers to sustain an equilibrium whereby each carrier offers its discounted fares via different distributor(s) than its competitor(s); the fares will then be higher than they would be were all competitors to offer their best deals via all the available channels due to presence of the ‘‘underinformed’’ consumers. Taken to data, these conclusions will imply that: (a) certain distributors may discriminate for or against certain airlines; and (b) the more travel agents offer the airlines’ discounted fares, the better a deal a customer may be able to get. For the empirical study, we compare last-minute fare quotes obtained (nearly simultaneously) via the three leading online travel agents (Expedia, Travelocity, and Orbitz) for a sample of 50 US airport-pair markets. We collected the data in October–November 2006, twice a week for four weeks. The data analysis revealed some discrepancies in offered airfares across the travel agents (even where the agents were technically linked to the same computer reservation system). We also found that when an additional travel agent offered the lowest available fare, this fare was about 3 percent ($11.80 on average) lower, holding other things constant. The downward effect of the ticket distribution market structure was more pronounced for less competitive airport-pair markets. Further analysis revealed that individual travel agents appear to discriminate either for or against individual airlines when choosing whether or not to report the lowest fare quotes offered by the carriers, consistent with what analysis of the incentives on the market predicted. This conclusion is

Distribution of Airline Tickets: A Tale of Two Market Structures

109

also supported by the limited information we have been able to gather from the industry players. Our results have the following implications. Broadly, we show that the structure of the distribution market should not be taken for granted even where its role appears to be that of pure distribution, and where retailers simply sell the product at prices set by the producer. Also, our study adds to the evidence that spread of the internet leads to more product differentiation (as suggested by Clay, Krishnan, & Wolff (2001), Baye, Morgan, & Scholten (2004), and Bilotkach (2010) for the airline industry) rather than convergence to the law of one price due to elimination of the search cost (evidence to this effect can be found in Brown & Goolsbee (2002)). Studies analyzing samples of offered fares started reemerging recently (the first such work can be traced to Stavins’ (2001) study of price dispersion in the US airline industry), and this literature can be expected to grow at a fast pace in the near future. Among the topics addressed in this literature are price dispersion (Bilotkach, 2006; Escobari, 2005; Escobari & Gan, 2006; Giaume & Guillou, 2006); price dynamics (Button & Vega, 2007; Piga & Bachis, 2006); and fare comparison across booking engines (Chen, 2006; Clemons, Hann, & Hitt, 2002). The latter two papers are the most similar to our study, and exhibit different results. Chen showed little disparity in fares quoted by the major online travel agents (Travelocity and Orbitz) and by the airlines themselves on the New York– Los Angeles air travel market. Her results do suggest structure of the distribution market is important. In particular, fares quoted by a single source in her sample are higher than those quoted by multiple sources. This is the sort of pattern we observe in our sample, for 50 airport-pair markets. Including more markets allows us to also talk about the role of the airline market structure, in addition to that of the distribution market. Clemons et al., however, observed substantial differences in fare quotes across the five unidentified online travel agents. The rest of the chapter is organized as follows. The second section describes institutional details of the ticket distribution market. The third section analyzes airlines’ and distributors’ incentives in light of the present market structure. The fourth section describes the data, and the following two sections analyze it. The last section discusses our results and concludes. List of the airport-pair markets used for our analysis is in the appendix.

TICKET DISTRIBUTION MARKET The scheme of distribution of travel services (using the example of airline tickets) is presented in Fig. 1. An airline can sell its tickets either directly

110

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Airline

Computer Reservation Systems (Sabre, Worldspan, Amadeus, Galileo)

Brick-andmortar travel agents

Direct distribution – call center or web site

On-line travel agents

Customer

Fig. 1.

Airline Ticket Distribution Scheme.

(using its call-center or web site) or via the travel agents, by posting its fares into one or several computer reservation systems (CRS), which those travel agents access to make the bookings. Arrangements whereby a travel agent accesses a single CRS are most common; CRS do not normally charge the agents for such access. Prior to 2001, the total price the end customer paid for the ticket did not depend on the source via which the ticket was distributed: the airline paid both commission to the travel agent selling the ticket and a booking fee to the CRS involved. Following the events of September 11, 2001, the airlines (looking for ways to control costs) announced they will stop paying commissions to brick-and-mortar agents. The agents had to start charging their customers booking fees (which mostly fell in the range of $5–25 per ticket1). However, according to our sources,2 the airlines still pay commission to the large online travel agents. The main difference between the online and brick-and-mortar travel agents is supposed to be that with the online agent the end customer can observe the search results directly, whereas with the brick-and-mortar agency it is the agent who looks at the screen and communicates available options to the customer. Brick-and-mortar agents are claimed to offer more personalized service; yet, online agencies are moving into that territory as well.

Distribution of Airline Tickets: A Tale of Two Market Structures

111

Obviously, travel agents are in business of more than mere distribution of airline tickets: they provide a range of travel-related services (hotel and rental car reservations; selling vacation packages, etc.) and are able to provide their customers some complementary (and usually free) expert information on travel services providers and/or trip destinations. At the same time the airlines also are moving in the direction of offering the full range of services one can expect from a regular travel agent.3 For the travel agent to see the airline’s fare quote, it must be posted with the CRS this agent connects to. The CRSs were originally developed and owned by the airlines, but have later taken a life of their own as independent companies. On this side of the market we currently have four major players: Sabre (with about 45 percent market share in the US market and over 30 percent in the global market), Worldspan (over 25 percent US market share and 15 percent worldwide), Galileo and Amadeus (share of these two systems on the US market keeps declining while they remain solid players on the worldwide arena, with combined market share of over 50 percent). The airlines are currently free to choose which systems to participate in and at what level to do so.4 Most carriers do participate in multiple systems actively. An exception is Southwest Airlines, which only participates in Sabre at a low level, so that a potential customer will not be able to book this airline’s flights via an online travel agent linked to Sabre.5 Another such carrier at the time of our data collection was JetBlue Airways. This airline, however, started actively participating in Sabre in late 2006 and began posting its fares with Worldspan later in 2007. As for the online travel agents covered by this study, Travelocity is linked to Sabre, whereas Expedia and Orbitz6 are both linked to Worldspan CRS. Internet has altered the ticket distribution business dramatically. Emergence of online travel agents was the major innovation in the industry. Airlines also saw a huge potential in selling their tickets via their own web sites. As a result, in a matter of about three years (from 1999 to 2002) the airline ticket distribution business went from the one dominated by the brick-and-mortar agents who sold almost three quarters of all tickets in 1999 (the airlines sold the rest directly, predominantly offline) to the one with online travel agents’ market share of 15 percent, and 10 percent of all tickets sold by the airlines online.7 According to the most recent available estimate by Citigroup Investment Research, quoted by Forbes, in 2005 online travel agents captured over 25 percent market share in the airline ticket distribution industry.8 This share could have only grown since.

112

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

The online travel agent segment of the ticket distribution market is in turn dominated by the three major players that are included in our study: Travelocity (owned by Sabre CRS), Expedia (founded within Microsoft in 1995, and an independent publicly traded company since 2005), and Orbitz (started through a partnership of several major airlines in 2001, currently a subsidiary of Travelport, owned by the Blackstone Group – a private equity company). According to the US Department of Transportation,9 in 2002, 28.5 percent of all bookings with online travel agents were on Travelocity, 28.7 percent on Expedia, and 21.3 percent on Orbitz, for a total of 78.5 percent of all online travel reservations. Assuming the combined market share of these three biggest players remained similar as of current time, we can say that approximately one in five trips are booked via these three big online travel agents. Thus, our study encompasses the online travel agent industry quite well, and the segment we are looking at is a nontrivial part of the US air travel distribution market.

AIRLINES’ AND DISTRIBUTORS’ INCENTIVES Contractual Relationships Between Airlines and Agents According to our sources in the airline industry, airlines typically enter into 3–5 year long contracts with individual online travel agents. These contracts specify commission the airline pays to the travel agents, as well as the conditions for travel agent’s access to the airline’s inventory. While the details of the contracts are confidential, the crucial parameters that affect the structure of commissions paid by the airlines are known in the industry as content and participation. Content refers to the inventory that the airline makes available to the agent. We believe that the airlines would prefer to offer low level of content to the agents, meaning that carriers would seek to keep the lowest fares to themselves rather than selling those tickets via intermediaries. Retaining the lowest fares for the airline suggests that the structure of commissions usually paid to the travel agents is likely regressive (i.e., a fixed booking fee per ticket is an example of a regressive commission). The agents clearly prefer to obtain full access to the airlines’ inventory – this is known as full content. Participation refers to whether the agents are able to access the airline’s inventory in real time. When participation is asynchronous, the fare quoted on the agent’s first screen may differ – in either direction – from the actual fare at the point of booking the flight.10 Clearly, agents prefer real-time

Distribution of Airline Tickets: A Tale of Two Market Structures

113

participation, as it increases their reliability. No data exists, that we are aware of, which reveals the reliability of an agent’s quotes. Our industry contacts estimate that the agents’ ‘‘failure rate’’ (i.e., the likelihood that a different fare appears between the first query and the time of booking) is about 5 percent. The above-mentioned description suggests that the airlines’ preferred contract with the travel agents entails less than full content, since this enables airlines to sell the lowest fares themselves while still maintaining their visibility via the online agents. In comparison, travel agents would prefer to have full content with real-time participation. An important implication from what we have learned about airline–agent relationships is that they need not be standardized – an airline can have different contracts with different agents, and the agents can also treat the airlines differently.

Airlines’ and Agents’ Incentives Let us see what the structure of the distribution market implies for incentives of both airlines and distributors. The analysis that follows will make use of the following simple stylized facts about the markets in question. First, an airline sells both discounted and full-fare tickets, and a discounted ticket is more likely to sell due to the law of demand. Second, when searching for a ticket, consumers vary in how informed they are about the available options. As an illustrative example, consider the case of a single airline and two distributors, with the airline selling both discounted and full-fare tickets at exogenously determined prices. Discounted tickets sell with certainty, but full-fare tickets sell with probability lower than one (since some consumers’ willingness to pay is lower than the price of the full-fare ticket). Further suppose the number of tickets is determined outside of our analysis. Assume a distributor receives a fixed fee regardless of the price of the ticket it sells. Next, suppose consumers are heterogeneous in that some of them observe price quotes only from a single distributor, whereas others are able to observe quotes offered by the airline via both distributors. At the same time, suppose that each distributor is visited by a sufficient number of customers, so that the airline could sell all its discounted tickets via a single distribution channel. Given that ticket prices and the number of tickets offered for sale are determined outside of our exercise, the airline is interested in simply selling as many tickets as possible. Each of the ticket distributors will be interested in making sure it sells as many of the airline tickets as possible. Next, since

114

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

the discounted tickets sell with certainty, the airline will not really care via which channel they are sold.11 At the same time, each of the travel agents will have a clear incentive to ensure the airline sells as few of the discounted tickets via its competitor as possible. As for the full-fare tickets, since those sell with lower probability, the airline will want to retail those via both distribution channels, since overlooking one of the channels will lead to some of the full-fare tickets going unsold. The above thought experiment is a very stylized one; at the same time, it shows an important feature of the way incentives are likely to be structured in the travel distribution industry. Specifically, each of the travel distributors will want to retail as many of the tickets that sell with higher probability as possible (i.e., travel agents will want full content). The airline, on the other hand, may be indifferent between retailing its discounted tickets via only one or both distributors. The situation with the full-fare tickets is however different: here the airline will want to ensure that those tickets are sold via as many channels as possible. Thus, if the above-described incentives are at work in this industry, we should expect the following from our data:  There will be differences in the lowest fare quotes across distributors (stemming from possible exclusive dealing between the airlines and the agents).  Travel agents may be observed ‘‘preferring’’ some airlines over the others, consistently reporting an airline’s lower fare quotes while other agents will not do it. When we introduce competing airlines, focusing on the discounted fares, we can take our thought experiment further, yielding more interesting results. The trick here is that if you are the only airline offering discounted fare tickets through a certain distributor, you can charge higher price to take advantage of the less informed consumers. This is a typical result that emerges in the pure price competition models where some of the consumers do not observe prices charged by both competitors. Such equilibrium can be sustained via repeated interaction among players both across time and markets (e.g., Evans & Kessides, 1994, show importance of the multimarket contact in pricing by the US airlines). The above fact has a very interesting implication for the data analysis. First of all, due to various across-airline and across-distributor heterogeneities, equilibria where airlines only offer their discounted fares via some of the available distribution channels are likely to both be asymmetric in terms of prices, and imply higher price as compared to the cases where all airlines

Distribution of Airline Tickets: A Tale of Two Market Structures

115

distribute their fares via all the distributors. Such a behavior will likely be manifested via the lowest observable fare quotes. Specifically, where competing airlines will not be offering their discounted fares via all available distribution channels, we may observe the lowest fare quote offered by fewer distributors. Moreover, this lowest fare quote that will be offered by fewer distributors may be higher than the lowest fare quote (other things equal, of course) resulting from competing airlines distributing their quotes via all available outlets. We can therefore anticipate that more distributors offering the lowest fare will be associated with lower price quotes in the data. The exercise we have described here is obviously simplistic, naı¨ ve, and not formal at all. Yet, through our thought experiments we are able to outline the basic incentive structure that can be expected from the interaction between airlines and distributors of their services, and to formulate some concrete expectations as to how those incentives will likely be reflected in the data.

DATA COLLECTION AND DESCRIPTION Collection Process For this study, we collected fare quotes via the three leading online travel agents (Travelocity, Expedia, and Orbitz) for randomly selected 50 out of 100 top US airport-pair markets, as measured by the number of passengers traveling nonstop.12 This approach necessarily meant we have included many markets originating at a hub of a major carrier (in particular, Atlanta, Denver, and Dallas-Fort Worth airports). The data collection was centered around the notion of a traveler embarking on a short trip, and making his travel arrangements shortly before the departure. This allows working under an assumption that the traveler’s uncertainty about whether or not he is going to fly has been realized. Thus, whether the ticket is refundable or not is beside our hypothetical customer’s concern, allowing him to simply search for the best deals available.13 Our traveler was assumed to have strong preference for a given airport pair, while willing to accept one-stop flights in either or both directions. We considered the case of directional airport-pair markets, so that, for instance, JFK–Los Angeles market was different from Los Angeles–JFK one. The list of all airport pairs included in our analysis is presented in the appendix. We were collecting the fare quotes from October 20 to November 17, 2006. Fare quotes were collected on Tuesdays (for departure on Thursday and return on Saturday of the same week) and Fridays (for departure on the

116

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

nearest Sunday and return on the following Tuesday). We ended up collecting the data on nine different dates. For each airport-pair market on each day of data collection, we attempted achieving near simultaneity of obtaining the fare quotes by launching three browser windows in parallel. It should be clearly noted here that we were searching for the lowest fare quotes for a given combination of the departure and return days, without giving any regard to the departure time of the flights. We should note, however, that most of the time the lowest available fare quotes are applicable to a number of roundtrip journey combinations, involving flights at different times of the day. The data collection process was not, unfortunately, devoid of problems. Due to the computer troubles on our side, we were unable to complete the data collection process on October 20, November 3, and November 7, covering only half of intended markets on the latter date and about 85 out of 100 on the other two. As a result, we obtained 823 date-airport-pair markets observations out of 900. Further, Orbitz web site was down for maintenance at the time of data collection on October 27, meaning we did not observe any fare quotes by this online travel agent for that date. On each of the other dates, we were able to obtain offers from each of the three online travel agents.

Descriptive Statistics Normally, each request for fare quotes to an online travel agent results in multiple offers from various airlines, from which a customer selects the most preferred option using (un)certain criteria related to characteristics of an itinerary. Price and convenience (departure time, total duration of the trip, and time between flights, if any) are the most obvious such characteristics. Additionally, some customers may be willing to pay a premium for flying with their ‘‘preferred’’ airline. In this study we largely ignore the nonprice characteristics associated with various offers and focus on price alone. The main reason for doing so (except for trying to avoid complicating the analysis) is that differentiation with respect to some nonprice characteristics of a trip is horizontal (e.g., passengers will have different preferred departure/arrival times); and even where we have vertical product differentiation (e.g., other things equal, everyone will prefer to travel nonstop rather than make a stop en route14), customers will still differ in terms of their willingness to pay for this higher quality. Thus, let us look at the lowest fare quotes and see if we can observe any difference across the three major online travel agents. Before we proceed, it

Distribution of Airline Tickets: A Tale of Two Market Structures

117

pays to clarify that for all the analysis that follows we rounded the fare quotes to the nearest dollar using conventional rounding rules. This way, $300.49 and $299.51 quotes will be considered equal (as they both round to $300), whereas $300.49 and $300.51 quotes will be viewed as different (the first one being rounded to $300, while the second one to $301). We will shortly revisit the issue of treatment of ‘‘small’’ differences between the observed quotes. Having said this, for an average search the lowest fare quote was offered by 1.4 airlines (with standard deviation of 0.86); the median number of carriers offering the lowest quote is one, with the maximum of five airlines pricing their best offers at the same level. The median number of travel agents offering the lowest fare is also one. Table 1 presents some basic facts about our sample. From there we can see that Orbitz was least likely to offer the lowest fare quote, and in quite a

Table 1.

Comparison of Fare Quotes across Travel Agents. Travelocity

Average lowest fare quote for agent Cases offering the lowest fare Share of cases offering the lowest fare Difference between lowest offered by agent and lowest across the three agents Cases offering the second lowest fare Share of cases offering second lowest fare Cases offering neither lowest nor second lowest fare Share of cases offering neither lowest nor second lowest fare

Expedia

Orbitz

$407.15 (141.69) 586 71.20% $10.99 (30.34) 282 34.26% 169

$409.55 (138.74) 605 73.51% $16.16 (45.48) 317 38.52% 132

$413.38 (147.59) 71 9.82% $14.62 (31.69) 398 55.04% 271

20.53%

16.04%

37.48%

Notes: 1. Numbers in parentheses are standard deviations. 2. The $1–2 differences between the second lowest and the lowest available fare quotes are treated as nonnegligible. 3. Difference between the lowest fare offered by the agent and the lowest across the three agents is calculated only for those cases where the agent does not offer the lowest fare quote. 4. Number of cases where Orbitz offers neither the lowest nor the second lowest fare omits the date where the agent’s web site was down for maintenance. 5. All percentages for Orbitz are relative to the number of cases this agent was online (i.e., omitting the date when www.orbitz.com was closed for maintenance).

118

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

number of cases (even if we correct for the fact that fare quotes from this online travel agent were unavailable on one of the dates) did not even offer the second lowest fare. Expedia and Travelocity offer the lowest fare quote with almost equal frequency – their average lowest fare quotes are similar too. At the same time, the average lowest fare quote for Orbitz is ‘‘too close’’ to that for the other two online travel agents, given how infrequently this agent actually offers the lowest fare. A look at Fig. 2 explains this puzzle. Specifically, a substantial share (53.7 percent, to be more precise) of differences between the lowest and the second lowest fare quotes is within the range of two dollars. In fact, out of 379 cases where Orbitz offered the second lowest fare quote, it was only $1–2 higher than the lowest one found through a different travel agent in 341 cases. Thus, if we consider this 1–2 dollar difference (which, given the rounding we used, can mean actual differences from $0.02 to $2.98) negligible, Orbitz can be said to offer the lowest fare quote in 396 cases,15 or 54.7 percent of all cases where a quote from this agent could be obtained. Still, the average difference between the second lowest and the lowest fares across all the data points is $13.26, or around 3.3 percent of the average lowest fare quote (which is $393.53 with standard deviation of $134.45), and in 14.5 percent of cases this difference is above $20 or approximately 5 percent of the lowest fare quote. Before we move to a more detailed data analysis, let us see if the raw data can point us to any kind of correlation between the number of agents offering the lowest fare and the level of this fare. The relevant numbers are presented in Table 2 below. As one can see, the rough answer to the question of relationship between the number of travel agents offering the lowest fare and the level of this fare depends on whether we treat differences between the second lowest and the lowest fares in the amount of $2 or less as negligible. If we do not, then it appears all three travel agents offer the same lowest fare in very few cases. It also appears that the more travel agents offer the lowest fare, the lower the quoted price is. However, if we consider $1–2 differences between the second lowest and the lowest fares as negligible, the picture changes, even though the lowest fare still appears higher when only one agent offers it. Observe also that treatment of $1–2 differences between the lowest and the second lowest fares as negligible increases the median number of travel agents offering the lowest quote from one to two. Numbers presented in Table 2 are of course raw. Even though they do suggest there is some association between the observed competition between the travel agents and the lowest fare quote a potential customer can obtain, we need to control for a number of variables to make a more definite

0.3

0.25

Share

0.2

0.15

0.1

0.05

0 1

2

3

4

5

6

7

8

9

10

10 to 20 to 30 to 40 to 50 to 60 to 70 to 80 to 90 to 100 20 30 40 50 60 70 80 90 100 to 150 Difference

150 to 200

over 200

Distribution of Airline Tickets: A Tale of Two Market Structures

0.35

Fig. 2. Histogram of Differences Between the Second Lowest and the Lowest Fare Quotes. 119

120

Table 2.

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Lowest Price Quotes versus Number of Agents Offering Same.

Number of Agents Offering the Lowest Fare

Treatment of $1–2 Differences Between Second Lowest and the Lowest Fares As nonnegligible

One Two Three

As negligible

Number of cases

Average price (st. dev.)

Number of cases

Average price (st. dev.)

413 381 29

$427.35 (134.48) $377.76 (110.73) $141.21 (130.81)

298 264 261

$464.98 (122.50) $352.00 (120.05) $357.53 (84.50)

statement to this effect. In addition, we have not touched on the issue of competition on the airline markets in question. These are precisely the issues we leave for the next section of the chapter.

DATA ANALYSIS: THE ROLE OF TWO MARKET STRUCTURES The aim of our analysis is to understand the relationship between the structure of the ticket distribution markets (leading to the above-discussed incentives of airlines and seemingly ‘‘technical’’ distributors that the online travel agents supposedly are) and the fare quotes that the customer observes. We will continue using the lowest available price quote two days before departure (recall this is how far in advance we collected the data) as an indicator of the extent of competition for traveling public’s dollars. Table 2 showed there is some raw association between this indicator and the number of travel agents offering this lowest fare. Of course, to be more certain these numbers mean anything, our analysis needs to control for a number of factors raw averages are unable to capture. We conducted a fairly simple regression analysis using the natural logarithm of the lowest observed fare as the dependent variable. As a measure of intensity of competition between the three online travel agents, we use the number of agents offering the lowest fare quote observed. As before, we construct this measure treating the $1–2 differences between the lowest and the second lowest fares as nonnegligible and negligible. We use the airport-pair market Herfindhal index for nonstop flights (calculated using T-100 data) as the (imperfect) measure of competition between the

Distribution of Airline Tickets: A Tale of Two Market Structures

121

airlines on a route. We did not believe it was necessary to instrument this variable, as for the time period we consider the market structure on the airline side can be treated as fixed. In some of the specifications we report below, we also interact this Herfindhal index with the number of travel agents offering the lowest fare. For our regressions, we dropped the two markets (Oakland-San Diego and Oakland-Burbank), where JetBlue and/or Southwest were the only nonstop competitors. We believed one-stop fare quotes we observed on those markets were rather meaningless (such trips would involve significant time increase, and we could not observe how those quotes compared to the lowest nonstop fares available). This reduced the number of observations at our disposal from 823 to 795. The controls we used in all regressions are listed below. Variables that we report in Table 3 that follows are:  Natural logarithm of the lowest nonstop travel time between airports – as a proxy for distance.  Geometric average of endpoints’ income per capita, at the metropolitan statistical area level.  Temperature difference between the trip’s origin and destination for November of 2006 – we use this variable as a measure of the route’s attractiveness for vacation travelers.  Indicator variable for fare quotes for travel to Florida and Las Vegas (conventional vacation destinations).  Geometric average of endpoints’ population, at the metropolitan statistical area level.  Number of carriers observed offering the lowest fare.  Indicator for whether the lowest fare is for the nonstop trip in both directions. This variable is also interacted with the shortest nonstop flight time in all regressions.  Indicator variable for routes within Hawaii – there are two such markets in our list.  Indicator variables for markets where carriers whose fare quotes we observe compete (at the city-pair market level) with JetBlue Airways and Southwest Airlines. The following variables are also used in all regressions, but the corresponding coefficients are not reported to save space:  Airline-specific indicator variables taking the value of one if an airline offers the lowest fare quote. We use seven such variables for major airlines

122

Table 3.

Estimation with Number of Travel Agents Offering the Lowest Fare. Treatment of $1–2 Differences Between Second Lowest and the Lowest Fares As nonnegligible

Constant Log of flight time Average per capita income Average population Temperature difference Las Vegas/Florida destinations Hawaiian routes JetBlue Southwest Number of carriers offering the lowest

As negligible

(1)

(2)

(3)

(4)

(5)

(6)

5.032 (0.169) 0.248 (0.029) 5.8E-06 (2.3E-06) 7.3E-09 (3.6E-09) 0.0002 (0.0006) 0.018 (0.040)

5.048 (0.171) 0.248 (0.029) 5.7E-06 (2.2E-06) 7.3E-09 (3.6E-09) 0.0002 (0.0006) 0.020 (0.040)

4.913 (0.178) 0.248 (0.029) 6.3E-06 (2.3E-06) 7.1E-09 (3.6E-09) 0.0002 (0.0006) 0.025 (0.040)

4.991 (0.164) 0.251 (0.029) 5.6E-06 (2.3E-06) 7.1E-09 (3.7E-09) 0.0002 (0.0006) 0.023 (0.040)

5.011 (0.166) 0.251 (0.029) 5.5E-06 (2.4E-06) 7.2E-09 (3.7E-09) 0.0002 (0.0006) 0.025 (0.040)

4.922 (0.175) 0.252 (0.029) 5.7E-06 (2.4E-06) 7.2E-09 (3.7E-09) 0.0002 (0.0006) 0.027 (0.040)

0.905 (0.056) 0.148 (0.029) 0.081 (0.028) 0.022 (0.019)

0.905 (0.056) 0.149 (0.029) 0.082 (0.028) 0.022 (0.019)

0.887 (0.057) 0.146 (0.029) 0.078 (0.027) 0.021 (0.019)

0.960 (0.055) 0.152 (0.029) 0.086 (0.028) 0.021 (0.019)

0.960 (0.054) 0.153 (0.029) 0.087 (0.028) 0.021 (0.019)

0.951 (0.055) 0.153 (0.029) 0.085 (0.028) 0.018 (0.019)

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Independent Variables

Nonstop flight  flight time Number of travel agents offering the lowest fare Airport-pair market Herfindhal index Herfindhal  number of travel agents offering the lowest fare Adjusted R-squared

0.314 (0.049) 0.001 (0.0002) 0.062 (0.020) — —

0.314 (0.049) 0.001 (0.0002) 0.062 (0.020) 0.031 (0.082) —

0.768

0.768

0.321 (0.049) 0.001 (0.0002) 0.035 (0.039)

0.173 (0.063)

—

0.313 (0.049) 0.001 (0.0002) 0.030 (0.012) 0.039 (0.082) —

0.770

0.767

0.767

0.235 (0.131)

0.313 (0.049) 0.001 (0.0002) 0.030 (0.012) —

0.315 (0.049) 0.001 (0.0002) 0.014 (0.029) 0.113 (0.133) 0.079 (0.050) 0.767

Significant at 10% level; significant at 5% level.

Notes: 1. Dependent variable is natural logarithm of the lowest fare quote. 2. Number of observations is 795. 3. Numbers in parentheses are standard errors. 4. Treatment of $1–2 differences as negligible affected measure of the number of travel agents offering the lowest fare. 5. Results corrected for heteroskedasticity using White robust variance–covariance matrix. 6. Controls for date of collection, airline-specific dummies, airline–travel agent interactions, and airport-specific indicator variables have been included into all regressions, but are not reported.

Distribution of Airline Tickets: A Tale of Two Market Structures

Nonstop flight

123

124

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

(American, Delta, United, Continental, Alaska, Northwest, US Airways), plus a separate variable for smaller carriers whose fare quotes we observe (Air Tran, Frontier, Midwest, Spirit). For the purposes of this analysis, fares offered by America West were included in US Airways category.16  Dummies for different (eight out of nine, naturally) dates of data collection.  Airline-travel agent interaction variables.  Thirty-one airport-specific dummies to control for relevant heterogeneities. Regression results are presented in Table 3. We report six specifications, three for each treatment of small differences between the lowest and the second-lowest fare quote. As can be seen from Table 3, coefficients on all the control variables are stable across specification and do not depend on whether the small differences between the lowest and the second lowest fare are treated as negligible. The coefficients on logarithm of flight time, JetBlue and Southwest variables have expected signs. The negative and significant effects of average per capita income and population on fare are somewhat puzzling at first; yet, Bilotkach (2010) obtained similar result when studying determinants of differences in last-minute fares quoted by Southwest Airlines and Orbitz. When the fare quote is for nonstop flight, it is about 31 percent lower than that for the one-stop trip; however, this effect is distance (proxied by scheduled flight time for a nonstop trip) dependent. More specifically, reading our regression results we can see that one-stop and nonstop lowest fare quotes are equal for about a three-hour flight; for a fivehour coast-to-coast flight, the nonstop fare quote will be about six percent higher than same for a one-stop flight. This finding is quite intuitive: for shorter flights, a stop en route entails significant increase in travel time. Therefore, an airline choosing to offer one-stop fares on shorter-haul markets where a nonstop competitor is present may be inclined to set high prices, both because of higher cost relative to a nonstop competitor and in anticipation of a loyal customer within few miles from earning an award ticket. For longer-haul flights, however, an airline offering a one-stop service becomes an effective competitor to carriers flying nonstop, and prices its tickets lower to attract less time-sensitive travelers. As for variables we do not report in Table 3, the following facts are worth mentioning. First, fares collected on Tuesdays were typically higher than those collected on Fridays. Second, few airline coefficients turned out significant: Delta Air Lines offered consistently higher lowest fares, whereas

Distribution of Airline Tickets: A Tale of Two Market Structures

125

quotes by ‘‘other’’ airlines were marginally lower. Airport effects are insignificant to marginally significant. Only Atlanta, Denver, Las Vegas, and San Francisco international airports show marginally significant coefficients. These airports are hubs for major airlines and are dominated (Las Vegas to a lesser degree than the other three) by a single airline. We also experimented with including dominant-airline-hub-airport dummies into our regressions. However, the results did not change in any meaningful way, and adjusted R-squared fell marginally. Travel agent–airline interactions are also typically insignificant; however, they are jointly significant, as suggested by conventional F-tests. Unlike control variables, the variables of interest do change depending on how small differences between the lowest and the second lowest fares are treated. When $1–2 differences between the lowest and the second lowest fares are treated as nonnegligible, first and second specifications suggest that additional agent offering the lowest fare quote will bring the same down by over 6 percent ($24.40 on average). At the same time, we observe strong dependence of this effect on the level of nonstop competition on a given airport-pair market, as well as the expected positive sign on the Herfindhal index itself. Treatment of small differences between the lowest and the second lowest fares as negligible (resulting in counting more travel agents as effectively offering the lowest fare) yields decreases the magnitude of the effect of observed competition between travel agents to about 3 percent ($11.80 on average). As before, we observe that the effect of apparent competition between travel agents is dependent on the level of competition between the airlines. More specifically, specification 6 in Table 3 suggests that for the monopoly airport-pair market an additional travel agent offering the lowest fare brings down this fare quote by almost 8 percent. This is $31.48 given the observed average for the entire sample and over $36 given the observed average lowest fare quote for the airport-pair markets in our sample with only one carrier providing nonstop service.17 For a symmetric duopoly, the effect of adding another travel agent is reduction in fare by about 4 percent (again, $15.74 on average for the whole sample, and $14.89 on average for the markets with Herfindhal index for nonstop services between 0.4 and 0.6 in our study). The effect for the most competitive nonstop market we have (AtlantaDenver) is only 2.2 percent (or $9.65, given the average lowest fare quote we observe on this route is $438.53). For deeper understanding of the association between additional travel agent offering the lowest fare quote and the level of the same, the following table presents results of regressions which, instead of the number of travel

126

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

agents offering the lowest fare quote, used dummy variables for the cases where the lowest fare quote was offered by two online travel agents, as well as by two or three agents. This way, we will obtain the ‘‘marginal effects’’ of the second and the third online travel agent on the lowest fare quote. Note that in all specifications reported in Table 4 we used exactly the same control variables as in regressions we report in Table 3. Moreover, all coefficients on those control variables are very similar in terms of both magnitude and statistical significance across all regressions reported in the two tables. Therefore, reporting coefficients on control variables in Table 4 is redundant. Results for the cases where small differences between the lowest and the second lowest fares are treated as nonnegligible do not add much to our understanding of the impact of competition between airlines and travel agents. On the other hand, if we treat the $1–2 differences between the best and the next best deals as negligible, it becomes apparent that it is the third online travel agent that appears to bring in the ‘‘competition’’; the effect is again more pronounced for less competitive airline markets. It is also interesting to note that the so-called ‘‘marginal effect’’ of the second online travel agent is to bring fares up, not down, and more so the less competitive the airline market is. We have up to now considered the three main online travel agents as rather indistinguishable entities. That is, we effectively believed that the agents were offering deals that were similar in nonprice dimensions.18 This might actually not be the case. An author’s observations about the online ticket distribution business, for example, has led him to conclude that Orbitz tends to offer more option for complex international itineraries; however, a number of those options entailing long layover times (sometimes as long as 20 hours) are not offered by Orbitz’s main competitors. It may thus happen that, as an example, the lowest fare offered by one agent involves scheduling that most customers will view as hugely inconvenient – therefore, such an offer may be rejected by most passengers in favor of a more expensive but a more convenient option. We did mention above that differentiation by the nonprice characteristics we did not capture (these may include operating carrier, type of aircraft, departure and/or arrival time) is horizontal, and therefore any ‘‘ranking’’ of fare quotes by these characteristics we may attempt will necessarily be subjective; not to mention the possibility of the same fare quote being offered for flights departing at different times, operated by different carriers using different aircraft types, etc. It is true that we can rank one-stop trips more objectively by the total duration of the trip (with a caveat that the

Independent Variables

Treatment of $1–2 Differences Between Second Lowest and the Lowest Fares As nonnegligible

Two travel agents offer the lowest fare quote Two or three travel agents offer the lowest fare quote Airport-pair market Herfindhal index Herfindhal  two travel agents Herfindhal  two or three travel agents Adjusted R-squared

As negligible

(1)

(2)

(3)

(4)

(5)

0.034 (0.053) 0.099 (0.057)

0.033 (0.653) 0.098 (0.057)

0.006 (0.101) 0.045 (0.102)

0.052 (0.020) 0.071 (0.026)

0.052 (0.020) 0.071 (0.026)

0.041 (0.054) 0.055 (0.063)

–

0.036 (0.083) –

–

–

0.768

0.768

0.075 (0.091) 0.019 (0.151) 0.233 (0.146) 0.769

–

–

0.033 (0.081) –

0.052 (0.110) 0.170 (0.103) 0.242 (0.107) 0.768

– –

–

0.767

0.767

(6)

Significant at 10% level; significant at 5% level.

Notes: 1. Dependent variable is natural logarithm of the lowest fare quote. 2. Number of observations is 795. 3. Numbers in parentheses are standard errors.

Distribution of Airline Tickets: A Tale of Two Market Structures

Effects of Additional Travel Agents on the Lowest Fare Quote.

Table 4.

4. Treatment of $1–2 differences as negligible affected measures of the number of travel agents offering the lowest fare. 5. Results corrected for heteroskedasticity using White robust variance–covariance matrix. 6. The same control variables have been included as those included into specifications reported in Table 3. Coefficients are similar in both magnitude and statistical significance.

127

Estimation Including Travel Agent Dummy Variables.

Independent Variables

Treatment of $1–2 Differences Between Second Lowest and the Lowest Fares As nonnegligible (1)

Expedia Adjusted R-squared

0.088 (0.066) 0.188 (0.074) – –

(2)

As negligible (3)

0.087 (0.066) 0.187 (0.074)

0.056 (0.106) 0.041 (0.109)

0.031 (0.081) –

0.074 (0.090) 0.023 (0.151) 0.230 (0.146) 0.071 (0.049) 0.068 (0.030) 0.771

–

–

0.075 (0.048) 0.071 (0.030) 0.769

0.074 (0.048) 0.071 (0.030) 0.769

(4)

(5)

0.119

0.119

(0.029) 0.212 (0.051)

(0.029) 0.212 (0.052)

– –

0.032 (0.082) –

–

–

0.091 (0.029) 0.068 (0.029) 0.770

0.091 (0.029) 0.068 (0.029) 0.770

(6) 0.027 (0.060) 0.078 (0.078) 0.054 (0.109) 0.167 (0.101) 0.237 (0.106) 0.089 (0.025) 0.068 (0.024) 0.771

Significant at 10% level; significant at 5% level.

Notes: 1. Dependent variable is natural logarithm of the lowest fare quote. 2. Number of observations is 795. 3. Numbers in parentheses are standard errors. 4. Treatment of $1–2 differences as negligible affected measures of the number of travel agents offering the lowest fare, as well as the travel agent dummy variables. 5. Results corrected for heteroskedasticity using White robust variance–covariance matrix. 6. The same control variables have been included as those included into specifications reported in Table 3. Coefficients are similar in both magnitude and statistical significance.

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Two travel agents offer the lowest fare quote Two or three travel agents offer the lowest fare quote Airport-pair market Herfindhal index Herfindhal  two travel agents Herfindhal  two or three travel agents Orbitz

128

Table 5.

Distribution of Airline Tickets: A Tale of Two Market Structures

129

shortest layover time may not be uniformly the most preferred one due to higher possibility of missing the next flight if something goes wrong with the first one); but here we too may observe (and in fact have observed) the same fares for trips of different duration. What we can do to capture possible ‘‘agent-specific’’ heterogeneity is to simply supplement regressions reported in Table 4 with travel agent dummy variables. The corresponding estimation results are presented in Table 5. Note that the only difference between all specifications reported in Table 5 as opposed to those described in Table 4 is that the former includes Orbitz and Expedia dummy variables; all other control variables are the same, and their coefficients are very similar in terms of magnitude, sign, and significance to same reported in Table 3. Comparing results reported in Tables 4 and 5 we see the following. First, in those specifications where the airport-pair market Herfindhal index was not interacted with measures of observed competition between travel agents, the magnitude of the effects of additional travel agent offering the lowest fare quote nearly doubled when we tried controlling for possible travel agent heterogeneity. However, results did not change for specifications where the airport-pair market Herfindhal was interacted with variables measuring competition between travel agents (specifications 3 and 6 in the table). Therefore, if there is a bias due to possible travel agent specific heterogeneity, correcting for it only reinforces our results.

DATA ANALYSIS: AIRLINE–AGENT RELATIONSHIPS The supposition that individual distributors might discriminate either for or against the individual airlines, as implied by our theoretical analysis, can be looked into directly. Table 6 reports two indicators of interaction between individual airlines and each of the three main online travel agents. First, we show simple correlation coefficients between airline dummies (taking value of one if the carrier offers the lowest fare quote), and the travel agent indicator variables, treating small differences between the lowest and the second lowest fare quotes as negligible. We can see from that table that the picture is not exactly random. For example, when American Airlines offers the lowest fare quote, it is more likely to appear on Expedia or Orbitz than on Travelocity; Travelocity, on the other hand, is more likely to report other carriers’ lowest fare quotes as compared to other agents. We also computed correlations between the number of carriers offering the lowest fare and individual travel agent indicator variables. Here we can see that fare quotes

130

Table 6.

How Travel Agents Present Airlines’ Lowest Fare Quotes.

American Airlines Delta Air Lines United Airlines US Airways/America West Alaska Airlines Continental Airlines Northwest Airlines Other carriers (Frontier, Midwest, AirTran) Number of Carriers

Expedia

Orbitz

Correlation

Percent shown

Correlation

Percent shown

Correlation

Percent shown

0.076 0.029 0.082 0.064 0.158 0.217 0.064 0.232

62.6 74.6 78.1 74.2 41.3 23.5 85.7 89.1

0.157 0.061 0.063 0.056 0.065 0.002 0.038 0.146

86.2 81.7 78.8 69.8 84.7 73.5 64.3 61.7

0.152 0.073 0.134 0.107 0.138 0.117 0.075 0.019

75.5 72.4 74.9 51.1 27.3 90.8 81.6 60.1

0.113

–

0.045

–

0.174

–

Notes: 1. In all calculations, small differences between the lowest and the second lowest fare quotes were treated as negligible. 2. Correlations are simple correlation coefficients between the airline dummy variable (taking the value of one if the airline is one of those offering the lowest fare quote) and the travel agent indicator (equal to one if travel agent offers the lowest available fare quote). 3. ‘‘Percent shown’’ is the percent of cases, where a given travel agent featured the lowest fare quote, when it was offered by a given airline. 4. Measures for Orbitz were computed taking into account the day when the agent’s quotes were unavailable due to site maintenance.

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Travelocity

Distribution of Airline Tickets: A Tale of Two Market Structures

131

presented by Orbitz are associated with higher ‘‘diversity’’ – this agent is more likely to give same lowest fare quotes by multiple carriers. The second measure we report is so-called ‘‘percent shown,’’ or the share of lowest fare quotes we observe offered by an airline actually offered by a given travel agent. We can see a substantial variation in this measure as well. On one extreme, of all the cases Continental Airlines offered the lowest fare, Travelocity bothered to show it to prospective customers in less than a quarter of instances. On the other end of the spectrum, in 9 out of 10 cases the lowest fare was quoted by Continental Airlines, we could find it on Orbitz. We can say that Expedia appears to be the most ‘‘objective’’ agent in a sense that it does not seem to openly discriminate either for or against a particular carrier in any systemic way; however, Orbitz appears to discriminate against Alaska and to a lesser extent US Airways/America West (or maybe those airlines do not cooperate with this agent closely), while clearly favoring Continental Airlines by showing its lowest fares to potential customers. Finally, Travelocity appears to have better relationships with Northwest and the group we termed ‘‘Other carriers,’’ while not being eager to show Alaska’s and Continental’s fares where these carriers offer the cheapest deals. To be more confident that numbers reported in Table 6 are not a realization of some random process, we conducted formal pair-wise tests of equality of the above-reported correlation coefficients, by airline, for three possible pairs of travel agents (Travelocity–Expedia, Travelocity–Orbitz, and Expedia–Orbitz). To perform the test, we applied the following Fisher transformation to each correlation coefficient r, as follows:   1 1þr (1) Z ¼ ln 2 1r Then, the usual z-test can be applied to the test statistic: Z1  Z2 z ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1=ðN 1  3Þ þ 1=ðN 2  3Þ

(2)

The resulting values of the above z-statistic are presented in Table 7. Table 7 shows that the three major online travel agents do treat most of the airlines differently. Specifics of the airline–agent (or maybe airline–CRS and CRS–agent) interaction behind the numbers we report can be subject to various speculations; however, what we report above is in line with what we suspected when analyzing airlines’ and distributors’ incentives.

132

Table 7.

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

Testing for Equality of Correlation Coefficients from Table 6. Travelocity vs. Expedia Travelocity vs. Orbitz Expedia vs. Orbitz

American Airlines Delta Air Lines United Airlines US Airways/America West Alaska Airlines Continental Airlines Northwest Airlines Other carriers (Frontier, Midwest, AirTran) Number of Carriers

4.95 0.64 0.39 2.40

4.86 0.88 1.06 3.41

0.85 0.25 1.44 1.01

4.48 4.44 2.02 7.62

0.42 6.74 0.23 4.32

4.06 2.30 2.26 3.30

1.36

1.24

2.60

Notes: Reported are values of test statistic (2) for pair-wise comparison of correlation coefficients reported in Table 6. Significant at 5% level; significant at 10% level.

Here is what our findings can mean for an average traveler. Suppose you are searching for a ticket in a market where US Airways and American Airlines are the two major nonstop competitors (e.g., Dallas–Philadelphia route). One day, American Airlines offers a lower fare as compared to US Airways. If you search on Travelocity, you will be less likely to observe that lowest fare than if you use any other online travel agent. This may not mean American Airlines’ fare quote on Travelocity will be higher than same offered by US Airways – all we are saying here is that American Airlines’ fare quote observed on Travelocity will be higher than same displayed by Expedia or Orbitz. When US Airways offers the lowest fare on same route, however, the situation will be reversed, and a potential customer looking on Orbitz will observe higher US Airways’ fare quote than the one displayed by Travelocity, again based on our results. Additionally, travel agents may have different policies as far as offering tickets on overbooked flights is concerned (provided an agent itself knows the flight is overbooked). In this case, when one agent chooses not to offer a ticket on such a flight, whereas the other one chooses to show this option to the potential customer, we will get two agents – even linked to the same CRS – offering different quotes. For such cases to influence our results, however, we would also need most, if not all, flights between the two cities overbooked on the dates of data collection at the time of data collection, so that such instances would show up as the lowest fare quotes. To be more

Distribution of Airline Tickets: A Tale of Two Market Structures

133

confident, we have computed correlation between the number of travel agents offering the lowest fare and ‘‘price-per-minute’’ or the ratio of the lowest fare quote to duration of a nonstop flight. Such correlation is indeed insignificant (negative 0.065 when small differences between the best and the second-best fare quotes are treated as nonnegligible, and positive 0.043 when such differences are treated as negligible).

DISCUSSION AND CONCLUSIONS In this study we examined a sample of airfare quotes with the aim of understanding the interplay of two market structures: those of the airline and the ticket distribution markets. We compared the lowest fare quotes by the three leading online travel agents, which together appear to sell approximately 20 percent of all airline tickets on the US market. Moreover, while Expedia, Travelocity, and Orbitz dominate the online travel distribution business, none of the three can be singled out as a clear leader. Analysis of incentives implied by the structure of the ticket distribution market suggested the following. First, individual distributors will have strong incentive to obtain exclusive rights for distribution of the airlines’ discounted tickets. Second, the competing airlines can potentially exploit their repeated interaction and presence of the less informed customers to sustain an equilibrium whereby each carrier offers its discounted fares via different distributor(s) than its competitor(s); the fares will then be higher than they would be were all competitors to offer their best deals via all the available channels due to presence of the ‘‘underinformed’’ consumers. Taken to data, these conclusions will imply that: (a) certain distributors may discriminate for or against certain airlines; and (b) the more travel agents offer the airlines’ discounted fares, the better a deal a customer may be able to get. The data analysis suggested that, controlling for various factors, when more travel agents offer the lowest fare quote, the level of same is lower. Initial results suggest that an additional travel agent offering the lowest fare quote will bring the same down by $12–24 on average (depending on how small differences between the lowest and the second lowest fare quotes are interpreted); this effect is more pronounced for the airline markets that are less competitive (ranging from about $9 decrease in fare quote per additional agent for the most competitive market in the sample – in terms of nonstop competition – to over $36 for the market with a single carrier performing nonstop flights). Further analysis revealed, however, that it is the third travel agent that brings the fare quote down, whereas going from one agent offering

134

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

the lowest fare quote to two we observe an increase in the price level. Again, the downward pressure of the third travel agent is more pronounced for the airline markets, which themselves are less competitive. That is, we have found the evidence to support the second of the above conjectures. Analysis of the market players’ incentives and our (admittedly limited) knowledge of the market structure allowed us to suspect the existence of special relationships between individual airlines and travel agents. We found support for this supposition in our data. Our analysis methodology can be criticized on the basis of selection of markets, agents, booking scenario, looking only at the lowest fare quote, and rounding off observed offered prices. In defense, we can say the following. The markets we selected are rather diverse in terms of the airline competition and length of haul. One might wish to see more diversity in terms of market size, but then many studies of the airline industry look at the ‘‘top’’ markets. It is true that many of the markets in our sample involve one or even two major hub airports. Yet, such is the structure of major US carriers’ networks that their nonstop flights are predominantly to/from their hubs. Additionally, we control for airport-specific and airline-specific heterogeneity in our analysis. It is true that we only looked at three of the available multitude of online travel agents. However, the agents covered by our study are the most important players, covering 80 percent of the nondirect online distribution of airline tickets in the USA; other agents have small market share compared to Orbitz, Expedia, or Travelocity. In addition, none of the big three agents is itself the dominant player – their market shares are rather symmetric. Next, we selected the booking scenario we use to avoid running into a major problem associated with choosing to examine fare quotes collected further away from the intended departure date. Specifically, a customer looking at offered prices in advance may be rightly concerned about the possibility of obtaining a refund in case he/she has to cancel the trip after it has been booked (and the airlines, understanding this, might price their offers strategically). Looking at last-minute fare quotes, on the other hand, we can work under an assumption of no (or at least very little) uncertainty as far as canceling the trip is concerned. Each query to each of the travel agents yielded multiple options with different price quotes. We believed looking only at the lowest fares offered is justified for two reasons. First, our theoretical analysis allowed us to formulate certain hypotheses as these relate to the lower end of the price distribution. Second, we assumed our hypothetical customer was interested in getting the best deal available – which is reinforced by assuming he was booking travel after it has become certain there will be no need to cancel, so potential refund is not of our traveler’s concern.

Distribution of Airline Tickets: A Tale of Two Market Structures

135

Rounding off was used for the sake of comparing the fare levels across airlines and travel agents. We realize that rounding as we use it can mean that what we measure as $2 difference between the lowest and the second-lowest fare quote can in fact be between $1.01 and $2.99 difference, but we believe in this case consistent methodology with potential for creating case-specific measures is more justified than consistent measures created using case-specific methodology. On the other hand, by rounding off we would treat the $200.34 fare by one carrier, $200.45 by the second one, and $199.60 by the third one as indistinguishable, which appears a more appropriate treatment as opposed to claiming $199.60 as the lowest fare quote, while $200.45 the third lowest. Next, we have not, as indicated above, used nonprice characteristics of fare quotes in our analysis. Also, we chose not to do so since consumers’ preferences over those characteristics are quite likely to be heterogeneous (indicating horizontal rather than vertical differentiation). Besides, we have often observed the same fare quotes for itineraries with very different nonprice characteristics. Additionally, we have controlled for the vertically differentiated trip characteristic by including the indicator variable for the nonstop itinerary. As a final note, this study examined a market where agents from which a consumer purchases the product appear to be acting as mere distributors and not as price-setters, in addition to competing with the producers directly. We find that competition between such agents exists (both theoretically and empirically); it is an important determinant of the price the potential consumer will face, and interplays with competition between producers of the service in question.

NOTES 1. Some agents (e.g., www.priceline.com) waive booking fees on some itineraries; others may charge different fees depending on the itinerary’s complexity. 2. We have been able to interview several airline and travel agency industry professionals. 3. When booking a ticket on an airline’s web site, one now has a possibility of adding a hotel and a rental car to his/her reservation, without leaving the site. Some airlines also offer vacation packages through their (or related) web sites. 4. Initial CRS regulation, adopted in 1984, stipulated that an airline owning or marketing a CRS must participate in competing systems. This rule was scrapped in 2004. 5. Bilotkach (2010) demonstrates that this strategy allows Southwest Airlines to take advantage of the travelers making their arrangements shortly before the departure. 6. Orbitz is also owned by Travelport, an owner of Galileo, the other major CRS. We have however found no indication of direct linkage between the travel agent and

136

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

this CRS, possibly related to the low market share of Galileo on the US market, where Orbitz operates. 7. The numbers are from ‘‘Computer Reservations System Regulations: Final Rule,’’ 14 CFR Part 255. 8. Online Travel Gets Personal, Forbes.com, posted 02/17/2006. 9. See ‘‘Computer Reservations System Regulations: Final Rule,’’ 14 CFR Part 255. 10. When participation is not real time, the agent first shows you the fare it encountered the last time it queried the airlines’ inventory. If you want to follow through on that fare, the agent queries the inventory again. 11. This will actually be true (with some qualifications regarding the shares of customers informed about only one of the two distributors) even if discounted tickets sell with higher probability than full-fare tickets – details are available from the corresponding author upon request. 12. We used T-100 dataset for 2005 to determine what those markets were. 13. One may suggest our hypothetical traveler might prefer some flexibility regarding the time of the return flight. However, once the fundamental travel uncertainty has been realized, the problem of purchasing a refundable versus a nonrefundable ticket becomes similar to the choice between a lottery and a certain outcome (nonrefundable discounted tickets carry the possibility of making a change for a fee). 14. We do use a dummy for nonstop flights to control for this. 15. This number obtains by summing 71 and 341, and subtracting 16 cases where this agent offers both the lowest and the second lowest fare quotes. 16. The two airlines were at the time in process of finalizing their merger approved over a year ago. However, at the time of data collection we still observed fare quotes by America West Airlines. 17. Disregarding Southwest and JetBlue as possible competitors, as these airlines’ quotes are not observable with the travel agents. 18. The only measure of nonprice characteristics of the fare quote we did include was the nonstop flight indicator variable.

REFERENCES Baye, M. R., Morgan, J., & Scholten, P. (2004). Price dispersion in the small and in the large: Evidence from an internet price comparison site. Journal of Industrial Economics, 52, 463–496. Bilotkach, V. (2006). Understanding price dispersion in the airline industry: Capacity constraints and consumer heterogeneity. In Lee Darin (Ed.), Advances in airline economics (Vol. 1, pp. 329–345). Elsevier. Bilotkach, V. (2010). Reputation, search cost, and airfares. Journal of Air Transport Management, 15, 251–257. Brown, J. R., & Goolsbee, A. (2002). Does the internet make markets more competitive? Evidence from the life insurance industry. Journal of Political Economy, 110, 481–507. Button, K., & Vega, H. (2007). The uses of the ‘‘Temporal-Fares-Offered Curve’’ in air transportation. Journal of the Transportation Research Forum, 46, 83–99.

Distribution of Airline Tickets: A Tale of Two Market Structures

137

Chen, J. (2006). Differences in average prices on the internet: Evidence from the online market for air travel. Economic Inquiry, 44, 656–670. Clay, K., Krishnan, R., & Wolff, E. (2001). Prices and price dispersion on the Web: Evidence from the online book industry. Journal of Industrial Economics, 49, 521–539. Clemons, E. C., Hann, I-H., & Hitt, L. M. (2002). Price dispersion and differentiation in online travel: An empirical investigation. Management Science, 48, 534–549. Comanor, W. S., & Rey, P. (2000). Vertical restraints and the market power of large distributors. Review of Industrial Organization, 17, 135–153. Escobari, D. (2005). Are airlines price discriminating? Tourist versus business travelers. Working Paper. Texas Pan American University. Escobari, D., & Gan, L. (2006). Price dispersion under costly capacity and demand uncertainty. Working Paper. Texas A&M University. Evans, W. N., & Kessides, I. N. (1994). Living by the golden rule: Multimarket contact in the U.S. airline industry. The Quarterly Journal of Economics, 109, 341–366. Giaume, S., & Guillou, S. (2006). Concentration, market share inequality and prices: An examination of European airline markets. In L. Darin (Ed.), Advances in airline economics (Vol. 1, pp. 273–296). NY: Elsevier. Piga, C., & Bachis, E. (2006). Pricing strategies by European low cost airlines: Or, when is it the best time to book online? Loughborough University Department of Economics Working Paper WP 2006–14. Stavins, J. (2001). Price discrimination in the airline market: The effect of market concentration. Review of Economics and Statistics, 83, 200–202.

138

VOLODYMYR BILOTKACH AND MARIJA PEJCINOVSKA

APPENDIX: LIST OF AIRPORT-PAIR MARKETS Anchorage–Seattle Atlanta–Boston Atlanta–Washington (National) Atlanta–Denver Atlanta–Dallas-Fort Worth Atlanta–Fort Lauderdale Atlanta–Jacksonville Atlanta–Las Vegas Atlanta–Los Angeles Atlanta–Orlando Atlanta–Miami Atlanta–Chicago (O’Hare) Atlanta–West Palm Beach Atlanta–Philadelphia Atlanta–San Francisco Atlanta–Salt Lake City Austin–Dallas-Fort Worth Burbank–Oakland Baltimore–Atlanta Dallas (Love Field)– Houston (Hobby) Denver–Dallas-Fort Worth Denver–Los Angeles Denver–Chicago (O’Hare) Denver–San Francisco Dallas-Fort Worth–Las Vegas

Dallas Fort Worth–Los Angeles Dallas-Fort Worth–New York (LaGuardia) Dallas-Fort Worth–Orlando Dallas-Fort Worth–Miami Dallas-Fort Worth–Chicago (O’Hare) Dallas-Fort Worth–San Diego Dallas-Fort Worth–San Antonio Dallas-Fort Worth–Seattle Detroit–Minneapolis-St. Paul Fort Lauderdale–New York (JFK) Honolulu–Lihue Honolulu–Kahului Houston (Intercontinental)–Los Angeles New York (JFK)–Los Angeles New York (JFK)–Orlando New York (JFK)–San Juan, Puerto Rico Las Vegas–Phoenix Los Angeles–Seattle Los Angeles–San Francisco New York (LaGuardia)–Chicago (O’Hare) Miami–San Juan, Puerto Rico Oakland–San Diego Chicago (O’Hare)–San Francisco Tampa–Atlanta Seattle–Minneapolis-St. Paul

Note: Markets selected for the study are 50 airport-pair markets randomly chosen from among the top 100 US airport-pair markets by nonstop traffic in 2006. Source: T100 Segment dataset, US Department of Transportation.

CHAPTER 6 ON AIRLINE PRICING BEHAVIOR DURING FINANCIAL TURNAROUNDS Christian Hofer INTRODUCTION Few industries may be better suited to study the effects of financial distress on managerial decision making than the airline industry. Economic recessions, natural catastrophes, and terrorist attacks are just some of the factors that frequently take a particularly heavy toll on the airline industry. Thus, coping with and overcoming financial distress is a critical aspect of airline management. Many studies have examined the effect of financial distress and bankruptcy on airfares (e.g., Borenstein & Rose, 1995; Busse, 2002; Kennedy, 2000), service levels (Ribbink, Hofer, & Dresner, 2009), and market entry and exit decisions (Benoit, 1984; Liu, 2009). Hofer, Dresner, and Windle (2005), for example, found that distressed carriers generally have lower ticket prices, all else equal. Moreover, Hofer et al. (2005) presented evidence that airfares decrease in the time periods just before, during, and following bankruptcy filings. Hofer, Dresner, and Windle (2009) further explored the effect of financial distress on airfares and concluded that firm and route market characteristics moderate this relationship. This research aims to further our understanding of airlines’ pricing behavior in times of financial distress. Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 139–155 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003008

139

140

CHRISTIAN HOFER

Distressed firms’ turnaround processes typically consist of several stages, including, most notably, the downturn phase and the recovery phase (Schendel, Patton, & Riggs, 1976). Multiple studies have concluded that a firm’s strategic options vary throughout this process and that different turnaround strategies may be indicated at different stages of the turnaround process (Arogyaswamy, Barker, & Yasai-Ardekani, 1995; Hofer, 1980; Robbins & Pearce, 1992; Whitaker, 1999). Such strategies include, for example, cost cutting strategies, revenue increasing strategies, or asset reduction strategies (Hofer, 1980). Hence, it is conceivable that a distressed firm’s pricing behavior varies throughout the turnaround process. In this research, distressed carriers’ pricing behaviors during the turnaround process are further explored. It is suggested that the distress–price relationship is nonlinear and that a distressed firm’s pricing actions vary with the firm’s position in the downturn and recovery cycle. The empirical analysis of a large sample of carrier-route level observations from the U.S. domestic airline industry provides ample support for the hypotheses set forth in this chapter. Hence, this research provides new insights into the anatomy of turnaround processes in the airline industry, with a focus on the role of prices as a mechanism to respond to and recover from financial hardship. As such, this research is of interest to airline industry professionals, policy makers, and academics alike: With price being one of the key competitive parameters in the airline industry, it is paramount for airline managers to understand how and when financial distress may be expected to impact a distressed competitor’s airfares. Similarly, this study enables policy makers to better understand the implications of financial distress on (price) competition in the U.S. airline industry. Finally, from an academic perspective this research contributes to our knowledge of the role of pricing actions in distressed carriers’ turnaround efforts. The remainder of this chapter is structured as follows: The relevant literature is reviewed and hypotheses are developed in the second section. Data, variables, and methodological issues are discussed in the third section, followed by the presentation of the empirical results in the fourth section. The results are further discussed and concluding remarks are offered in the fifth section.

LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT Economists have long contended that a firm’s financial condition should affect its output and pricing decisions (Brander & Lewis, 1986). In this vein,

On Airline Pricing Behavior During Financial Turnarounds

141

Hendel (1996) concludes that distressed firms may increase their likelihood of survival and return to profitability in the long term by lowering prices or selling off assets. Borenstein and Rose (1995) were among the first to study the effect of financial distress on ticket prices in the airline industry. Reviewing distressed carriers’ airfares, they observed some price reductions before bankruptcy filings but found no evidence of further price effects during or after such filings. These results are broadly consistent with Kennedy (2000), who identified patterns of revenue decreases and recoveries prior to and following bankruptcy filings, respectively. Busse (2002), on the other hand, found that bankrupt and highly leveraged carriers were more likely to engage in price wars. Similarly, Hofer et al. (2005) found that fares fall before, during, and after carriers file for Chapter 11 protection. They suggest that both demandside and supply-side rationales may explain the observed erosion in fares: As the public learns of a carrier’s financial difficulties, passenger demand may decrease given the uncertainty about the carrier’s future. Prices may then decrease in response to the decline in passenger demand. From a supply perspective, bankrupt airlines can often shed costs by negotiating more favorable terms with lenders. Such cost reductions may then enable these carriers to offer lower fares (see also Barla & Koo, 1999). Hofer et al. (2009) adopted a broader perspective and investigated the effects of financial distress, rather than of bankruptcy per se, on airfares. Specifically, these authors found that firm and route market characteristics – operating costs, market shares, firm size, and market concentration – moderate the distress– price relationship. In summary, most studies present at least some evidence that financial distress affects an airline’s ticket prices. In accordance with the results of these studies, the baseline hypothesis is proposed as follows: H1a. The greater the degree of financial distress, the lower the distressed firm’s airfares. At the same time, the aforementioned studies acknowledge that the magnitude of this relationship varies with certain firm and market characteristics. The purpose of this research, in turn, is to explore how the nature of distress–price effect changes throughout a distressed firm’s turnaround process. Multiple studies have explored the anatomy of corporate turnarounds and have investigated the appropriateness of various turnaround strategies as a function of a distressed firm’s position in the turnaround process (Arogyaswamy et al., 1995; Fredenberger & Bonnici, 1994; Hofer, 1980; Robbins & Pearce, 1992; Schendel et al., 1976; Whitaker, 1999). The central tenet of these studies is that

142

CHRISTIAN HOFER

a firm’s strategic options change with its financial situation and as it proceeds through downturn and recovery phases. Hence, this study applies the strategic options framework to explore distressed airlines’ pricing behaviors throughout the turnaround process. Hofer (1980) suggested that the choice of operating turnaround should be a function of the severity of the firm’s financial difficulties. This contention is mirrored in the work of Robbins & Pearce (1992) and Arogyaswamy et al. (1995). All three studies concluded that cost and price cutting strategies may be viable in situations of moderate distress, while asset reduction (downsizing) strategies may be required in situations of severe distress. Thus, the relationship between distress and airfares should be nonlinear. Price reductions may be a viable instrument in a firm’s turnaround process when there is a reasonable expectation that such reductions will spur greater passenger demand and positively affect cash flows (Eisdorfer, 2007). As financial distress persists or even worsens, however, further price cuts may not be economically viable in the long run. As such, it is suggested that the marginal effect of financial distress on airfares decreases as the degree of financial distress increases. H1b. The effect of financial distress on airfares is curvilinear. That is, the incremental effect of financial distress on airfares decreases as the firm’s level of distress increases. As indicated above, prior research has typically differentiated between the downturn and recovery phases of the turnaround process. As illustrated in Fig. 1, the downturn phase is characterized by financial decline, while the recovery phase is characterized by financial improvement. H2a/b and H3a/b explore differences in distressed carriers’ pricing behaviors during downturn and recovery phases, respectively. Thus, a central tenet of this research is that a distressed firm’s pricing behavior is not necessarily a function of the degree financial distress only. Rather, a firm’s position in the turnaround Fin. distress

downturn

recovery Time

Fig. 1.

The Downturn and Recovery Cycles.

On Airline Pricing Behavior During Financial Turnarounds

143

process defines its strategic options which, in turn, will be reflected in the firm’s pricing behavior. In the context of this research, the downturn phase is defined as the time period starting with a carrier’s entry into financial distress, followed by a period of financial deterioration and ending with the financial turnaround (see Fig. 1). As Eisdorfer (2007) pointed out, maintaining and improving cash flows is an all important objective of financially distressed firms’ turnaround efforts. Yet, there is ample evidence in the literature suggesting that firms operating under financial distress are more likely to engage in aggressive price competition (Bowman, 1982; Busse, 2002; Maksimovic & Zechner, 1991). To the extent that unit revenue losses are not offset by increases in passenger demand, aggressive pricing behavior may, thus, be considered risky. Much of the literature on corporate turnarounds stresses the importance of unabsorbed slack, that is, freely available ‘‘excess, uncommitted resources in organizations’’ (Singh, 1986, p. 567) as a determinant of the risk distressed firms are willing to take as they implement turnaround strategies (Barker & Duhaime, 1997; Francis & Desai, 2005; Singh, 1986). As a firm experiences a period of financial downturn, its slack resources are depleted over the course of the downturn, thereby increasingly limiting its ability to engage in aggressive price competition with potentially adverse cash flow effects. Hence, it is expected that a distressed carrier’s fares will decrease as it enters and proceeds through the downturn phase. However, the longer the carrier remains in the downturn phase, the smaller the incremental price reductions. These expectations are summarized in Hypotheses 2a and 2b and in Fig. 2 below: H2a. The longer the downturn period, the lower the fares. H2b. The relationship between the length of the downturn period and fares is convex. Fare

Length of downturn

Fig. 2.

The Relationship Between Downturn Length and Airfares.

144

CHRISTIAN HOFER Fare

Length of recovery

Fig. 3. The Relationship Between Recovery Length and Airfares.

A successful turnaround, in turn, can only be achieved if profit margins return to levels that are sustainable in the long run. Hence, it is expected that airfares rise as financially distressed carriers achieve the turnaround and recover. These fare increases are expected to accelerate as the distressed firm continues to recover. This expectation is, in part, motivated by the observation of Hofer et al. (2005) who noted that a carrier’s financial distress may deter passengers from buying tickets given the uncertainty surrounding the carrier’s future operations. Ultimately, this is expected to depress ticket prices. Hence, the signal of improving financial health should contribute to increasing passenger demand and, thus, ticket prices. H3a,3b and Fig. 3 summarize these contentions: H3a. The longer the recovery period, the higher the fares. H3b. The relationship between the length of the recovery period and fares is convex.

DATA, VARIABLES, AND METHODOLOGY The hypotheses are tested using data from the U.S. domestic airline industry. This selection is suitable given the amount and quality of the data available (Borenstein, 1989; Hofer et al., 2005; Morrison, 2001). Airlines continue to be required to provide detailed information on every tenth (domestic) ticket sold. In addition, airlines must report detailed financial and operational data. The data, which are publicly available, were purchased from Database Products Inc, a service provider that screens

On Airline Pricing Behavior During Financial Turnarounds

145

the data for accuracy and creates custom data sets. Specifically, route-carrier level data for the top 2,000 U.S. domestic airport-to-airport route markets (by passenger volumes) were obtained for all quarters in the 2003–2006 time period. The large scope of this sample selection ensures that both major trunk route markets and smaller niche route markets are considered in the analysis. The 2003–2006 period was deemed an appropriate selection since most carriers experienced at least some degree of financial distress during this time. Data prior to 2003 were not included due to the extraordinary impact of the September 11 attacks on the (U.S.) airline industry, and more recent data were not yet available at the time of data collection. The resulting data set contains 213,982 complete observations, where each observation pertains to a particular carrier in a particular O&D route market in a particular quarter.

Variables For each observation, the carrier’s average fare1 (Fare) and the number of passengers transported on the particular route in a given quarter (AirlinePax) are recorded. Fare is the primary dependent variable of interest in this study. Prior research has shown that various route, airport, and carrier characteristics affect both prices and demand in air travel markets (e.g., Borenstein, 1989; Hofer, Windle, & Dresner, 2008). Each of these groups of variables are discussed in turn. Route-specific characteristics include the distance between the origin and destination airports (Miles),2 the tourist orientation of a route3 (Tourist), and the degree of concentration of the route market (RouteHHI). The RouteHHI variable is calculated as the sum of the squared market shares of all carriers operating in the route market. A carrier’s share in a route market (RouteShare) is assessed as well. Similarly, the degree of airport market concentration4 (MaxAirportHHI) and the carrier’s market power in the airport market5 (MaxAirportShare) is measured as well. Carrier-specific operating characteristics include the number of coupons6 (Coupons) and the carrier’s load factor7 (LoadFactor). A carrier’s operating expenses per available seat mile (OpEx) are used to assess systematic differences in carriers’ cost structures and their strategic orientation.8 Carrier financial characteristics and their effect on carriers’ pricing behavior are of particular interest in this research. In line with prior research, Z’’ score values are used to assess a carrier’s financial condition (Hofer et al., 2005; Hofer et al., 2009). Developed by Altman (2002), this

146

CHRISTIAN HOFER

composite measure assesses a firm’s current and past performance as well as its relative liquidity and debt. Prior research has shown that the Z score and its variants are suitable predictors of firm failure and valid measures of a firm’s financial condition (Calandro, 2007; Gritta, 2004). The Distress variable is defined as (1)Z’’ score, such that greater positive values of the Distress variable indicate greater levels of financial distress. Squared terms of the Distress variable (Distress^2) are also included in the estimation models to assess nonlinearity of the distress–price relationship. The Distress variable is also used to split the sample into subsamples of healthy and distressed carriers. Specifically, the DistressDummy variable is a binary variable that takes on the value of 1 if the carrier has a positive Distress score and the value of 0 if the carrier has a negative Distress score.9 Since this study focuses on the analysis of distressed firm’s pricing behaviors, only observations of carriers that are considered distressed are included in the subsequent analyses.10 In line with the hypotheses presented in this paper, a firm’s position in the turnaround process is identified and its pricing behavior is assessed as a function of downturn and recovery characteristics. Specifically, DownturnLength is the number of consecutive quarters a distressed carrier’s financial condition has been declining at the time of the observation. Likewise, RecoveryLength is the number of consecutive quarters a distressed carrier’s financial condition has been improving.

Descriptive Statistics Descriptive statistics for subsample of financially distressed carriers, as identified by the DistressDummy variable, are shown in Table 1. Bivariate correlations for the subsample of distressed carriers are shown in Table 2. As the data in Table 1 show, there is significant variability in all variables. Due to their potential skewness and to facilitate the interpretation of coefficient estimates, the Fare, AirlinePax, Miles, RouteHHI, and MaxAirportHHI variables enter the regression models in logged form.

Empirical Methodology The baseline model is shown in Eq. (1) below. The primary objective of this study is to investigate the effect of financial distress on carriers’ airfares. Due to the endogeneity of the Fare and AirlinePax variables, the latter is

On Airline Pricing Behavior During Financial Turnarounds

Table 1. Variable Fare ($, one-way) AirlinePax (10% sample) Miles Tourist Coupons RouteHHI MaxAirportHHI RouteShare (%) MaxAirportShare (%) Distress DistressDummy OpEx ($/ASM) LoadFactor (%) DownturnLength RecoveryLength

147

Descriptive Statistics (n ¼ 151,880). Mean

Std. dev.

155.27 503.52 1171.89 0.27 1.87 4673.69 2983.77 12.89 13.94 2.06 1 0.13 77.05 2.19 0.54

75.27 1401.62 652.9 0.44 0.42 2125.3 1319.55 20.94 15.57 6.77 0 0.02 6.23 2.73 0.9

estimated in a first-stage model, as shown in Eq. (2). Population and Income are exogenous instrumental variables used in the AirlinePax model. Following the approach of Hofer et al. (2005), Population is defined as the product of the metro area populations of the endpoints of a route market. Similarly, Income is defined as the weighted average income level in the origin and destination metro areas. lnðFarecrt Þ ¼ b0 þ b1 lnðAirlinePaxcrt Þ þ b2 lnðMilesr Þ þ b3 ðln Milesr Þ2 þ b4 Couponscr þ b5 Touristr þ b6 Distressct þ b7 LoadFactorct þ b8 OpExct þ b9 lnðRouteHHI rt Þ þ b10 lnðMaxAirportHHI rt Þ þ b11 RouteSharecrt þ b12 MaxAirportSharecrt þ Sfirm fixed effectsc þ Stime fixed effectst þ vcrt

ð1Þ

lnðAirlinePaxcrt Þ ¼ a0 þ a1 lnðFarecrt Þ þ a2 lnðMilesr Þ þ a3 ðln Milesr Þ2 þ a4 Touristr þ a5 Distressct þ a6 Populationrt þ a7 Incomert þ Sfirm fixed effectsc þ Stime fixed effectst þ ucrt

ð2Þ

Subscripts c, r, and t denote carrier, route and time specific characteristics, respectively. Firm and time fixed effects are added to control for inter-firm

148

Table 2. Bivariate Correlations (n ¼ 151,880).

Fare AirlinePax Miles Tourist Coupons RouteHHI MaxAirportHHI RouteShare MaxAirportShare Distress OpEx LoadFactor DownturnLength RecoveryLength

2

3

4

–0.112 0.250 0.165 0.154 0.084 0.061 0.043 0.059 0.024 0.021 0.069 0.089 0.071

0.156 0.044 0.536 0.169 0.070 0.676 0.564 0.068 0.030 0.051 0.057 0.021

0.109 0.302 0.540 0.093 0.062 0.096 0.031 0.063 0.110 0.021 0.003

0.035 0.055 0.202 0.023 0.010 0.001 0.003 0.035 0.002 0.002

5

6

7

8

9

10

11

12

13

0.165 0.066 0.412 0.594 0.040 0.028 0.525 0.087 0.127 0.757 0.029 0.010 0.003 0.037 0.058 0.001 0.033 0.050 0.016 0.011 0.051 0.124 0.043 0.060 0.134 0.128 0.168 0.127 0.033 0.101 0.028 0.067 0.004 0.011 0.000 0.006 0.026 0.002 0.004 0.022 0.018 0.084 0.047 0.071 0.479

Note: Correlation coefficients significant at the 5% level are printed in bold.

CHRISTIAN HOFER

1 2 3 4 5 6 7 8 9 10 11 12 13 14

1

On Airline Pricing Behavior During Financial Turnarounds

149

heterogeneity and systematic differences in airfares and passenger demand over time. The system of equations composed of Eqs. (1) and (2) is estimated via a three-stage least squares (3SLS) procedure. Given the endogeneity of the Fare and AirlinePax variables an ordinary least squares (OLS) method would result in biased estimates. As an alternative to the 3SLS procedure, the two-stage least squares (2SLS) method was also used, and the estimation results are largely consistent.11 The 3SLS procedure is preferred since it is asymptotically more efficient than the 2SLS procedure (Kennedy, 2003).

EMPIRICAL ANALYSIS AND RESULTS The baseline estimation results for Eqs. (1) and (2) are shown in Table 3, panels A and B, respectively. Both models are highly statistically significant as evidenced by the chi-squared statistics. Looking at the Fare equation, the results suggest that greater passenger volumes are associated with lower fares, all else equal. Consistent with prior research, the relationship between distance and fares is found to be nonlinear. Moreover, fares increase as routings become more circuitous (Coupons), but are lower in tourist markets. Market concentration at both the route and airport market levels is found to contribute to higher airfares. Similarly, route and airport market power is associated with higher airfares. Higher operating costs and higher load factors are both found to result in higher fares. The variable of particular interest, financial distress, carries a statistically significant negative coefficient. This finding is consistent with the baseline hypothesis (H1a) and prior research (e.g., Hofer et al., 2009). In the AirlinePax model, the coefficient of the endogenous Fare variable is negative as expected, suggesting that higher fares result in lower passenger demand, all else equal. Moreover, passenger demand is found to increase with distance in a nonlinear fashion and is higher in tourist markets. Both Income and Population variables carry the expected positive coefficient estimates. Finally, it is noteworthy that greater financial distress is associated with lower passenger demand, all else equal, as contended by Hofer et al. (2005). Table 4 summarizes additional regression results. Specifically, the estimates shown in Panel A test H1b by adding a squared term of the Distress variable. Panel B, in turn, focuses on the analysis of carriers that operate in the downturn phase of the turnaround cycle and estimates the effect of downturn length on airfares. Similarly, Panel B shows the results

150

CHRISTIAN HOFER

Table 3.

3SLS Baseline Regression Results (n ¼ 151,880).

Panel A lnFare Intercept lnAirlinePax lnMiles (lnMiles)2 Tourist Coupons lnRouteHHI lnMaxAirportHHI RouteShare MaxAirportShare Distress OpEx LoadFactor w2

Panel B Coef. 5.227 (0.094) 0.073 (0.003) 0.495 (0.029) 0.052 (0.002) 0.150 (0.002) 0.128 (0.005) 0.032 (0.003) 0.017 (0.002) 0.003 (0.000) 0.004 (0.000) 0.002 (0.000) 0.564 (0.053) 0.003 (0.000) 30,631

LnAirlinePax Intercept LnFare LnMiles (lnMiles)2 Tourist Distress Income Population

w2

Coef. 0.695 (0.808) 0.268 (0.060) 0.098 (0.156) 0.029 (0.012) 0.219 (0.015) 0.024 (0.001) 0.105 (0.049) 0.014 (0.000)

14.310

Standard errors are shown in parentheses below the respective coefficient estimates. Coefficient estimates marked with  and  are statistically significant at the 1% level and 5% level, respectively.

for observations of carriers that are in the recovery phase and estimates the effect of recovery length on ticket prices. In the interest of brevity, the subsequent discussions focus on the respective variables of interest only. In Panel A, the squared term of the Distress variable carries a positive and statistically significant – albeit small – coefficient estimate, providing evidence in support of H1b. That is, the relationship between financial distress and airfares is convex.

On Airline Pricing Behavior During Financial Turnarounds

Table 4.

lnFare Intercept lnAirlinePax lnMiles (lnMiles)2 Tourist Coupons lnRouteHHI lnMaxAirportHHI RouteShare MaxAirportShare Distress Distress2

3SLS Regression Results With DownturnLength and RecoveryLength. Panel A Coef.

Panel B Coef.

Panel C Coef.

5.203 (0.094) 0.072 (0.003) 0.483 (0.029) 0.051 (0.002) 0.150 (0.002) 0.124 (0.005) 0.030 (0.003) 0.016 (0.002) 0.003 (0.000) 0.004 (0.000) 0.006 (0.000) 0.0001 (0.000)

4.759 (0.153) 0.071 (0.004) 0.062 (0.048) 0.013 (0.003) 0.157 (0.003) 0.121 (0.008) 0.061 (0.004) 0.008 (0.003) 0.008 (0.000) 0.005 (0.000)

5.613 (0.290) 0.053 (0.009) 0.998 (0.085) 0.080 (0.006) 0.192 (0.006) 0.379 (0.017) 0.119 (0.008) 0.033 (0.006) 0.006 (0.000) 0.001 (0.000)

0.014 (0.002) 0.001 (0.000)

DownturnLength DownturnLength2 RecoveryLength RecoveryLength2 OpEx LoadFactor w2 No. of observations

151

0.641 (0.054) 0.003 (0.000) 30,098 151,880

0.653 (0.094) 0.003 (0.000) 24,922 65,985

0.236 (0.044) 0.044 (0.008) 2.854 (0.172) 0.004 (0.000) 7,351 21,428

Standard errors are shown in parentheses below the respective coefficient estimates. Coefficient estimates marked with  and  are statistically significant at the 1% level and 5% level, respectively.

152

CHRISTIAN HOFER

The estimation results shown in Panel B are based on the analysis of carriers in the downturn phase of the turnaround cycle.12 In reality, the turnaround process may not be as smooth as shown in Fig. 1. Rather, a carrier’s financial condition may deteriorate in one quarter, improve in the next, and then worsen again in the subsequent quarter. In the long run, however, patterns of downturn and recovery should be discernible. Hence, only observations of carriers that have faced worsening financial conditions for at least two consecutive quarters are included in this analysis. The results clearly indicate that the longer a carrier faces deteriorating financial conditions, the more the carrier’s fares tend to decrease, ceteris paribus (H2a). The squared term of the DownturnLength variable, however, is positive and statistically significant, indicating that the magnitude of the fare decreases diminishes with the length of the downturn period (H2b). Panel C focuses on carriers that have achieved the turnaround and are in the recovery phase. Symmetrically to the DownturnLength analysis above, only those observations of carriers that have seen their financial condition improve for at least two consecutive quarters are considered in this analysis. The expectation expressed in H3a is that prices will increase as the carrier’s recovery progresses. H3b further suggests that the relationship between the length of the recovery period and airfares is convex. The coefficient estimate of the RecoveryLength variable is negative and statistically significant, while the coefficient of the squared term is positive and statistically significant. This finding suggests that distressed carriers may not immediately change their pricing policy upon achieving the financial turnaround. Rather, the evidence indicates that fares continue to erode, presumably due to increasing passenger demand.13 However, fares ultimately rise as recovery progresses. Thus, the regression results provide at least some support for H3a and H3b.

CONCLUDING REMARKS The preceding analyses provide compelling empirical evidence that the relationship between a carrier’s financial distress and its pricing behavior may be more complex than reflected in prior research. The analyses presented here reveal that this relationship is curvilinear and that a distressed firm’s pricing behavior changes as a function of a firm’s position in the turnaround process. Specifically, prices generally decrease during the downturn phase. However, the incremental price reductions from one time period to the next become smaller the longer the carrier’s financial condition deteriorates.

On Airline Pricing Behavior During Financial Turnarounds

153

Once the turnaround is accomplished, prices continue to decline for some time before they increase as the carrier’s recovery progresses. The key finding of this study, thus, is that it may not be the degree of financial distress per se that best explains a distressed firm’s pricing behavior. Rather, the position of a firm in the turnaround process significantly impacts the firm’s pricing strategy. This result adds to our understanding of the distress–airfare relationship, in particular, and of airlines’ pricing behavior, in general. It is noteworthy that this study focuses on the analysis of distressed firms only. Symmetrically, one might be expect that a healthy carrier’s pricing strategy also is a function of its financial situation such as, for example, the availability of slack resources. Do carriers that enjoy great financial health and success seize the opportunity to lower prices and gain market shares, or do they maintain or even increase prices in the interest of greater short term profitability? The investigation of these questions is suggested for future research. The findings presented in this chapter should be of interest to both policy makers and airline managers. Both policy makers and industry professionals have long been concerned about the potentially dysfunctional effects of airline financial distress and bankruptcy protection, in particular, on the health and long-term viability of the airline industry as a whole. While the results of this study confirm that distress leads to lower airfares, ceteris paribus, it is also evident that these price reductions are only temporary and are reversed as soon as the carrier achieves the financial turnaround, even though the carrier continues to be financially distressed. This finding may help alleviate concerns about the above described sick industry problem. From a managerial perspective, knowledge of a competitor’s pricing strategy is paramount in an industry that is as competitive as the U.S. airline industry. This research sheds light on the role of prices in distressed carriers’ turnaround efforts and, as such, may help airlines anticipate a distressed competitor’s pricing actions.

NOTES 1. Fares are reported as average one-way fares (in $) based on round-trip purchases. 2. A squared term of the Miles variable is included in the estimation model to capture the nonlinearity of the distance–fare relationship. 3. All route markets originating or ending in Nevada or Florida are defined as tourist markets (Dresner, Lin, & Windle, 1996).

154

CHRISTIAN HOFER

4. Airport market concentration is measured as the sum of the squared enplanement shares of all carriers serving an airport. In line with Hofer et al. (2005), the larger of the concentration measures for the endpoints of the route market (MaxAirportHHI) is entered in the empirical analysis. 5. Carrier and time specific airport market shares are calculated based on passenger enplanement data. Similar to the logic discussed above, the maximum of a carrier’s enplanement shares at the endpoints of a route market (MaxAirportShare) is retained for the empirical analysis. 6. The number of flight coupons per one-way trip is an indicator of service quality. A single-coupon itinerary is a nonstop flight, while a two-coupon itinerary indicates that passengers make one en-route connection, etc. 7. Load factors are calculated as the ratio of revenue-passenger miles and available seat miles. 8. Prior research differentiates between ‘‘high-cost’’ legacy carriers and low-cost carriers (e.g., Hofer et al., 2009). 9. While the definition of this cutoff is somewhat arbitrary, this selection facilitates the interpretation of the estimation results. Also, it is noted that this cutoff is restrictive and, thus, conservative than the cutoff originally proposed by Altman (1968). 10. The entire data sample comprises 213,982 observations. Of these, 151,880 observations (71%) pertain to carriers that are considered financially distressed. 11. These results are not reported here due to space constraints. 12. It is noted that the Distress variable is not included in the Downturn and Recovery models due to concerns of multicollinearity and to facilitate interpretation of the regression results. 13. While not shown here, the effect of RecoveryLength on passenger demand is found to be positive as expected. As demand increases, the carrier may experience economies of scale which, eventually, result in lower fares, ceteris paribus.

REFERENCES Altman, E. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance, 23, 589–609. Altman, E. (2002). Bankruptcy, credit risk, and high yield junk bonds. Malden, MA: Blackwell Publishers. Arogyaswamy, K., Barker, V., & Yasai-Ardekani, M. (1995). Firm turnarounds: An integrative two-stage model. Journal of Management Studies, 32, 493–525. Barker, V., & Duhaime, I. (1997). Strategic change in the turnaround process: Theory and empirical evidence. Strategic Management Journal, 18, 13–38. Barla, P., & Koo, B. (1999). Bankruptcy protection and pricing strategies in the US airline industry. Transportation Research Part E, 35, 101–120. Benoit, J. (1984). Financially constrained entry in a game with incomplete information. The Rand Journal of Economics, 15, 490–499. Borenstein, S. (1989). Hubs and high fares: Dominance and market power in the U.S. The Rand Journal of Economics, 20, 344–365. Borenstein, S., & Rose, N. (1995). Bankruptcy and pricing behavior in U.S. airline markets. The American Economic Review, 85, 397–402.

On Airline Pricing Behavior During Financial Turnarounds

155

Bowman, E. (1982). Risk seeking by troubled firms. Sloan Management Review, 23, 33–42. Brander, J., & Lewis, T. (1986). Oligopoly and financial structure: The limited liability effect. The American Economic Review, 76, 956–970. Busse, M. (2002). Firm financial condition and airline price wars. The Rand Journal of Economics, 33, 298–318. Calandro, J. (2007). Considering the utility of Altman’s Z-score as a strategic assessment and performance management tool. Strategy & Leadership, 35, 37–43. Dresner, M. E., Lin, J. C., & Windle, R. J. (1996). The impact of low-cost carriers on airport and route competition. Journal of Transport Economics and Policy, 28, 309–328. Eisdorfer, A. (2007). The importance of cash-flow news for financially distressed firms. Financial Management, 36, 33–48. Francis, J., & Desai, A. (2005). Situational and organizational determinants of turnaround. Management Decision, 43, 1203–1224. Fredenberger, W., & Bonnici, J. (1994). Turnaround phases: Extending the life cycle theory. American Business Review, 12, 59–65. Gritta, R. (2004). Air carrier financial condition: A review of models for measuring the financial fitness of the U.S. airline industry. Proceedings of the 2004 Conference of the Air Transport Research Society. Hendel, I. (1996). Competition under financial distress. The Journal of Industrial Economics, 44, 309–324. Hofer, C. (1980). Turnaround strategies. Journal of Business Strategy, 1, 19–31. Hofer, C., Dresner, M., & Windle, R. (2005). Financial distress and US airline fares. Journal of Transport Economics and Policy, 39, 323–340. Hofer, C., Dresner, M., & Windle, R. (2009). The impact of firm financial distress on prices: A contingency approach. Transportation Research Part E, 45, 238–249. Hofer, C., Windle, R., & Dresner, M. (2008). Price premiums and low cost carrier competition. Transportation Research Part E, 44, 864–882. Kennedy, R. (2000). The effect of bankruptcy filings on rivals’ operating performance: Evidence from 51 large bankruptcies. International Journal of the Economics of Business, 7, 5–25. Kennedy, P. (2003). A guide to econometrics. Cambridge, MA: MIT Press. Liu, C. (2009). Entry behaviour and financial distress: An empirical analysis of the US domestic airline industry. Journal of Transport Economics and Policy, 43, 237–256. Maksimovic, V., & Zechner, J. (1991). Debt, agency costs, and industry equilibrium. The Journal of Finance, 46, 1619–1643. Morrison, S. (2001). Actual, adjacent, and potential competition: Estimating the full effect of Southwest Airlines. Journal of Transport Economics and Policy, 35, 239–256. Ribbink, D., Hofer, C., & Dresner, M. (2009). The impact of financial distress on customer service in the U.S. airline industry. Journal of the Transportation Research Forum, 48, 89–104. Robbins, K., & Pearce, J. (1992). Turnaround: Retrenchment and recovery. Strategic Management Journal, 13, 287–309. Schendel, D., Patton, G., & Riggs, J. (1976). Corporate turnaround strategies – A study of profit decline and recovery. Journal of General Management, 3, 3–11. Singh, J. (1986). Performance, slack, and risk taking in organizational decision making. Academy of Management Journal, 29, 562–585. Whitaker, R. (1999). The early stages of financial distress. Journal of Economics and Finance, 23, 123–133.

CHAPTER 8 BAGGAGE FEES AND CHANGES IN AIRLINE TICKET PRICES Kevin E. Henrickson and John Scott INTRODUCTION The past several years have seen dramatic increases in oil prices, which have adversely impacted airlines, with the average price of jet fuel increasing from $1.34 per gallon between 1995 and 2005 to $2.81 per gallon between 2006 and 2009. As a partial response to these increases in costs, many airlines have introduced fees for services that were previously provided to their customers free of charge. One such charge is a fee on checked baggage, which most airlines introduced in 2008. These charges have been successful in increasing airline revenues, so successful that many airlines have increased their fees multiple times over the past two years. Baggage fees have also enabled airlines to avoid dramatic increases in their airfares, which may result in significantly fewer customers, as these additional fees generate revenues, but since they are not collected when passengers book their tickets, the cost of air travel on these airlines appears lower than it actually is. The most notable exception to this pattern of charging baggage fees is Southwest Airlines, which has launched a ‘‘Bags Fly Free’’ advertising campaign in an attempt to differentiate their product from that of fee charging airlines. In this chapter, we use a spatial autoregressive model to analyze what impact the increase in fuel costs, and the introduction of baggage fees have had on ticket prices. Our results suggest that increases in jet Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 177–192 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003010

177

178

KEVIN E. HENRICKSON AND JOHN SCOTT

fuel prices are passed along to travelers in the form of higher ticket prices but that baggage fees actually reduce ticket prices, as airlines may substitute baggage fee revenue for ticket revenue to become more competitive on their airfare. We also find that Southwest Airlines has increased their ticket prices on routes in which they compete with fee charging firms, leveraging their ‘‘Bags Fly Free’’ product differentiation to increase their revenues. This interaction between the pricing of low-cost carriers, such as Southwest Airlines, and large legacy carriers is not uncommon in the literature. Many of these studies focus on the impact of airport hubs and route market power on pricing differentials, with seminal articles by Borenstein (1989), Evans and Kessides (1993), and Borenstein and Rose (1994), and a more recent example by Gerardi and Shapiro (2009). Other studies focus on the impact of low-cost carriers entering markets, and the competitive response of incumbent airlines (e.g., Dresner, Lin, & Windle, 1996; Goolsbee & Syverson, 2008; Whinston & Collins, 1992; Windle & Dresner, 1999). One common theme through the results of these studies is that airlines with route and/or airport power tend to charge higher ticket prices, but that entry of low-cost carriers can increase competition and temper this pricing power. A second theme in the airline pricing literature that is relevant to this present chapter is the airport substitution literature (e.g., Daraban & Fournier, 2008; Dresner et al., 1996; Fournier, Hartmann, & Zuehlke, 2007; Morrison, 2001). This literature clearly demonstrates that airline competition spills over from airports to nearby airports. That is, an increase in airfares from Los Angeles International Airport (LAX) to Chicago O’Hare International Airport is likely to impact fares from John Wayne Airport to Chicago Midway Airport as John Wayne Airport is approximately 41 miles from LAX and Chicago Midway is only 27 miles from Chicago O’Hare, meaning that travelers can view these airports as substitutes for one another. Empirically, these studies also demonstrate how spatial econometrics can be used to directly model this spatial substitution of airports. We add to these literatures by looking at year-over-year quarterly changes in airfares in the top 150 domestic routes from 2007 to 2009. This period of time allows us to both examine the introduction of, and subsequent increase in, baggage fees, and the impact of changes in jet fuel prices on airfares. Using a spatial autoregressive model to account for airport substitutability, our findings indicate that a one dollar increase in jet fuel costs increases airfares by $1.60, which may seem like a small impact, but equates to an average of $320,000 in extra traveler expenditures every year for each airline route. Our results also show that a one dollar increase in baggage fees

Baggage Fees and Changes in Airline Ticket Prices

179

reduces airline ticket prices on the fee charging airline by $0.24 and increases Southwest’s ticket prices on routes in which they compete with baggage fee charging airlines by $0.73. Thus, for a passenger checking one bag, they can expect to pay $0.76 more for each one dollar increase in baggage fees on airlines charging bag fees, and $0.73 more for each one dollar increase in baggage fees charged by competing airlines on Southwest Airlines. These results point to baggage fee charging airlines lowering their ticket prices to appear more competitive on airfares and making up for this revenue later when travelers check-in and pay their baggage fees. Likewise, the results show how Southwest has used its ‘‘Bags Fly Free’’ advertising campaign to differentiate their product and have used this differentiation to increase their airfares, in effect getting consumers to pay for their bags through higher ticket prices, as there is very little difference between the change in costs on baggage fee charging firms and on Southwest. The remainder of this chapter is divided into four sections. Section two presents background information regarding the impact of fuel prices on airlines and outlines the introduction and evolution of baggage fees. Section three then outlines our empirical model and discusses the data used in this analysis. Section four presents the results of this study, while Section five offers concluding comments.

BACKGROUND ON JET FUEL PRICES AND BAGGAGE FEES In 2009 American Airlines, Delta Air Lines, Southwest Airlines, United Air Lines, and US Airways combined to consume nearly 7.8 billion gallons of jet fuel, consumption that represents approximately 25–30% of their total operating costs for the year.1 Industry wide, Mazraati (2010) finds that the aviation sector accounts for approximately 5.8% of total worldwide oil consumption. In addition, Mazraati (2010) shows that jet fuel demand from this industry is relatively inelastic to changes in prices. Combining these observations with the changes in jet fuel prices shown in Fig. 1, a global recession and the terrorist attacks on 9/11, the past decade has been a challenging time for airlines (e.g., Drakos, 2004; Guzhva & Pagiavlas, 2004; Ito & Lee, 2005).2 Focusing specifically on the impact of oil prices on airlines, jet fuel prices have increased from an average of $1.34 per gallon between 1995 and 2005 to $2.81 per gallon between 2006 and 2009, a 109% increase. This increase in

KEVIN E. HENRICKSON AND JOHN SCOTT

3 2 1 0

Average Jet Fuel Prices (in dollars)

4

180

1980

1990

2000

2010

Year

Fig. 1.

Average Jet Fuel Prices by Year.

jet fuel prices has had a dramatic effect on airline costs and profits as shown in Table 1. Specifically, jet fuel prices peaked in 2008 at an average of $3.27 per gallon, and Table 1 shows that industry costs increased 46.2% while profits decreased 203% in 2008 relative to their 2007 levels. However, it should be noted that while all airlines were impacted by these increasing jet fuel prices, the large legacy carriers suffered more than the smaller low-cost carriers. For example, Delta Air Lines had their costs increase by 49.6% and profits fall by 122.8%, while Southwest Airlines had their costs increase by 39.4% and their profits fall by only 43.2%. One potential response to these increases in costs is for airlines to pass these costs along to their consumers through higher airfares. However, with price sensitive consumers and competition from both other carriers and modes of transportation, there are limits to how much airlines can increase their fares. As a result, airlines began introducing fees for services that were previously provided free of charge starting in 2008. Perhaps the most publicized of these fees are fees on checked baggage. As shown in Table 2, most airlines allowed each passenger at least one checked bag free of charge before 2008, but have since both introduced, and subsequently raised, fees on baggage. The most notable exception to this strategy is Southwest Airlines, which has had successful advertising campaigns regarding their ‘‘Bags Fly Free’’ policy, enabling them to differentiate their product from

Baggage Fees and Changes in Airline Ticket Prices

Table 1. Company

AirTran Alaska American Continental Delta Frontier Southwest United US Airways Total

181

The Impact of Fuel Prices on Costs and Operating Profits. Fuel Costs (Millions of $)

Operating Profit/Loss (Millions of $)

2007

2008

2009

2007

2008

2009

765 779 5,810 3,167 7,626 438 2,530 4,910 1,872 27,897

1,130 1,052 7,960 4,685 11,441 557 3,526 6,910 3,518 40,779

629 531 4,857 2,623 6,681 264 2,900 3,377 1,785 23,647

144 123 702 621 2,129 10 790 952 543 5,994

72 25 2,054 378 485 87 449 1,746 1,774 6,172

177 208 1,163 210 125 67 262 248 121 415

their competitors. Table 2 also offers some insight into the profitability of these fees, as many of the airlines have increased their baggage charges multiple times within a relative short period of time, with many airlines charging $25 for a first checked bag by the end of 2010. By charging fees on services, rather than increasing airfares, airlines are attempting to both increase their profitability and minimize the number of passengers lost due to the higher costs of traveling on the airline. Focusing on the latter of these goals, service fees may be more effective in retaining consumers, as opposed to fare increases, because of the timing of when consumers purchase their tickets versus when they pay these additional fees. Specifically, travelers most often purchase their airline tickets before the day of departure, at which point they compare the airfare charged by various carriers. However, any baggage fees are not included in the calculation of these airfares, preventing consumers from directly comparing the true cost of each ticket unless they determine each airline’s baggage fee and incorporate this into the airfare themselves, something that increases the search costs for airline tickets. The actual baggage fees are then only collected when each passengers checks in for their flight, within 24 hours of departure.3 Therefore, by separating the time at which consumers pay for their ticket and when they pay their baggage fees, it enables airlines to lower the impact of the increase in the cost of flying through baggage fees versus increasing all fares by an equivalent amount. Finally, Table 3 illustrates the increased revenues generated by these baggage fees by quarter.4 As shown in this table, airlines earned approximately $2.2 billion in baggage fees in 2009, a 131.2% increase from the amount of revenue earned from baggage fees in 2008. Focusing on

182

KEVIN E. HENRICKSON AND JOHN SCOTT

Table 2. Airline

Baggage Fees by Airline.

Effective Date (Tickets Purchased)

Baggage Fee

AirTran

May 15, 2008, to November 11, 2008 After November 12, 2008

1st bag free, $10 for a 2nd bag $15 for 1st bag, $25 for a 2nd bag

Alaska

May 21, 2008, to May 1, 2009 After May 1, 2009

1st bag free, $25 for a 2nd bag $15 for 1st bag, $25 for a 2nd bag

American

May 12, 2008, to June 14, 2008 June 15, 2008, to August 14, 2009 August 15, 2009, to February 1, 2010 After February 1, 2010

1st bag $15 for $20 for $25 for

free, $25 for a 2nd bag 1st bag, $25 for a 2nd bag 1st bag, $30 for a 2nd bag 1st bag, $35 for a 2nd bag

Continental

May 5, 2008, to October 6, 2008 October 7, 2008, to October 1, 2009 October 2, 2009, to January 8, 2010 After January 9, 2010

1st bag $15 for $20 for $25 for

free, $25 for a 2nd bag 1st bag, $25 for a 2nd bag 1st bag, $30 for a 2nd bag 1st bag, $35 for a 2nd bag

Delta

May 1, 2008, to July 30, 2008 July 31, 2008, to November 4, 2008 November 5, 2008, to July 15, 2009 July 16, 2009, to January 4, 2010 After January 5, 2010

1st bag 1st bag $15 for $20 for $25 for

free, $25 for a 2nd bag free, $50 for a 2nd bag 1st bag, $25 for a 2nd bag 1st bag, $30 for a 2nd bag 1st bag, $35 for a 2nd bag

Frontier

June 10, 2008, to September 12, 2008 September 13, 2008, to September 7, 2009 After September 8, 2009

1st bag free, $25 for a 2nd bag $15 for 1st bag, $25 for a 2nd bag $20 for 1st bag, $30 for a 2nd bag

United

February 4, 2008, to June 12, 2008 June 13, 2008, to May 13, 2009 May 14, 2009, to January 13, 2010 After January 14, 2010

1st bag $15 for $20 for $25 for

free, $25 for a 2nd bag 1st bag, $25 for a 2nd bag 1st bag, $30 for a 2nd bag 1st bag, $35 for a 2nd bag

US Airways

February 26, 2008, to July 8, 2008 July 9, 2008, to April 22, 2009 April 23, 2009, to August 26, 2009 After August 27, 2009

1st bag $15 for $20 for $25 for

free, $25 for a 2nd bag 1st bag, $25 for a 2nd bag 1st bag, $30 for a 2nd bag 1st bag, $35 for a 2nd bag

individual airlines, Table 3 also illustrates the impact of individual changes to baggage fees. For example, Delta’s imposition of a $15 fee for a first bag in November 2008 caused baggage revenue to increase to $102.8 million in the first quarter of 2009, a nearly 70% increase relative to the fourth quarter of 2008 and a 286.5% increase relative to the first quarter of 2008. While Table 3 clearly demonstrates that baggage fees have increased revenues for these airlines, it is less clear whether baggage fees impact firm pricing, something that we examine in the next several sections of this chapter.

Baggage Fees and Changes in Airline Ticket Prices

Table 3.

Revenues Generated from Baggage Fees in 2008 and 2009.

Company

2008 Baggage Revenues (Millions of $)

AirTran Alaska American Continental Delta Frontier United US Airways Total

183

2009 Baggage Revenues (Millions of $)

Q1

Q2

Q3

Q4

Q1

Q2

Q3

Q4

2.7 3.3 33.0 10.7 26.6 1.0 12.2 7.5 97

6.1 4.2 37.1 16.4 42.5 1.3 19.7 17.9 145.2

7.9 9.0 94.1 21.2 47.5 2.9 42.3 67.9 292.8

12.8 5.5 113.9 49.3 60.5 10.0 58.8 93.8 404.6

30.9 5.4 108.1 55.6 102.8 12.5 59.1 94.2 468.6

40.5 6.2 118.4 63.2 118.4 13.5 67.4 104.1 531.7

40.2 25.2 119.5 66.0 129.5 14.9 77.9 111.4 584.6

34.3 21.8 129.2 69.7 131.1 14.4 64.6 122.5 587.6

Total

175.4 80.6 753.3 352.1 658.9 70.5 402 619.3

DATA AND EMPIRICAL METHODOLOGY To estimate the impact of fuel costs and baggage fees on airline pricing, we use data collected from the U.S. Department of Transportation’s Airline Origin and Destination Survey, which represents a 10% sampling of airline tickets. These data contain information regarding ticket prices, the airport of origination, the destination airport, and other itinerary specific details. Because these data represent a 10% sample of all airline tickets, some represent unreasonable airfares if, for example, they were purchased using frequent flyer miles, which would mean that the reported ticket price was zero dollars. Therefore, we omit any unreasonable airfares and then aggregate the data by year, quarter and origin–destination to create quarterly average ticket prices for each origin–destination combination from 2006 to 2009. We then use these data to calculate year-over-year changes in the average quarterly ticket price for each origin–destination combination, leaving us with observations in 2007, 2008, and 2009. We focus on this particular period of time because, as was shown previously in Table 2, most of the baggage fees were introduced in 2008, so this time frame allows us to observe changes in ticket prices due to the introduction, and subsequent increases in, baggage fees. This period of time also allows us a great deal of variation in jet fuel prices to estimate the changes in ticket prices due to fuel cost changes. We further restrict our sample to the top 150 origin–destination combinations within the continental United States, as determined by the number of passengers traveling between the origin airport and destination airport as

184

KEVIN E. HENRICKSON AND JOHN SCOTT

reported in the U.S. Department of Transportation’s T-100 Domestic Segment data, plus any origin–destination observations where the origin and destination airports are both within 50 miles of the origin and destination of the ‘‘top 150’’ origin–destination combination. These additional observations are included because it is well established that airline customers are willing to substitute between airports, making ticket prices spatially dependent on ticket prices at adjacent airports (Daraban & Fournier, 2008; Dresner et al., 1996; Fournier et al., 2007; Morrison, 2001).5 Once the routes not fitting this definition are removed from the data, we are left with 9,656 observations on our dependant variable, the year-over-year change in the average quarterly ticket price for each origin–destination combination for 2007–2009. On the basis of the aforementioned literature regarding spatial price dependence in the airline industry, we model price changes as a spatial autoregressive (lag) process: P ¼ Xb þ rWP þ u

(1)

where P is our dependant variable, change in the average quarterly ticket price, and X is the vector of explanatory variables to be discussed below. The existence of rWP in Eq. (1) captures the impact of changes in the ticket prices charged by other firms on ticket prices and potentially causes ordinary least squares (OLS) to be biased due to an omitted variable. Specifically, WP is the interaction of the change in the average quarterly ticket price and W, the spatial weighting matrix. The spatial weighting matrix, W, is block diagonal, with each block representing a different quarter and year.6 In addition, all of the on diagonal elements of the spatial weighting matrix are set to zero to prevent each firm’s airfare from being regressed on itself, while the off diagonal elements are equal to 1/di,j, where di,j is based on the degree of contiguity between airfare i and airfare j. In particular, if airfares i and j share the same origin airport and destination airport, di,j takes a value of 1. Likewise, if airfares i and j do not share the same origin and destination airport, but do have origination and destination airports within 50 miles of each other, di,j takes a value of 2. Finally, di,j takes a value of 0 if fares i and j do not serve the same market, where a market is defined as all flights between an origin airport and a destination airport, including all airports within 50 miles of both the origin and destination. This structure of the spatial weighting matrix implicitly assumes that consumers are willing to substitute between airports as shown by Dresner et al. (1996); Morrison (2001); Fournier et al. (2007); and Daraban and Fournier (2008), but that airports are not perfect substitutes for one another, hence the smaller weight placed on routes out of different airports. In addition,

Baggage Fees and Changes in Airline Ticket Prices

185

it is assumed that any airports located more than 50 miles apart are not substitutes for the airport in question.7 Finally, this spatial weighting matrix is row standardized, making WP the weighted average change in the price of competing airlines, with r being the term to be estimated, which determines the impact of a changing in spatial competitors’ airfare on changes in airfare. To account for the impact of baggage fees on ticket pricing in Eq. (1), we include three separate variables, the first of which is the year-over-year quarterly change in baggage fees, which were obtained from newspapers and company press releases. We also note that Southwest Airlines has not instituted baggage fees and has spent a great deal of money on their ‘‘Bags Fly Free’’ advertising campaign, which means that the impact of baggage fees may differ on routes that Southwest serves. To capture this potential impact, we include change in baggage fees on routes served by Southwest, which is defined as our change in baggage fees variable, interacted with a dummy variable equal to 1 for routes in which Southwest offers flights. Additionally, because airlines are competing, we note that these baggage fees may enable Southwest to increase their prices as they have differentiated their product through their no baggage fees policy, as such we also include change in mean bag fees  Southwest Airlines, which is defined as the yearover-year change in the average bag fees for each route interacted with a dummy variable equal to 1 for Southwest Airlines. The other control variables determining changes in airfare are common to previous studies (e.g., Daraban & Fournier, 2008; Fournier et al., 2007), and include: the year-over-year quarterly change in jet fuel costs collected from the U.S. Energy Information Administration, and measured in cents; the year-over-year change in quarterly route/market concentration measured as the change in the Herfindahl index, where the Herfindahl index is computed using the airline’s portion of seats being flown on the route in question; the year-over-year quarterly change in market share at the route level for each firm, where each firm’s market share is defined as the proportion of seats they fly on each route; the year-over-year quarterly Change in seat capacity obtained from the U.S. Department of Transportation’s T-100 Domestic Segment data; the year-over-year quarterly change in departures offered, also obtained from the U.S. Department of Transportation’s T-100 Domestic Segment data; and the year-over-year quarterly change in on-time performance, which is obtained from the U.S. Department of Transportation’s airline on-time performance data.8 Summary statistics for each of these variables are presented in Table 4. Finally, because rWP is endogenous, Eq. (1) is estimated through instrumental variables using X and WX as instruments for WP as described by Anselin (1988).

186

KEVIN E. HENRICKSON AND JOHN SCOTT

Table 4. Summary Statistics. Variable

Mean

Average ticket price $165.00 Change in average ticket price $3.25 Weighted change in average ticket $3.38 price of spatial competitors Jet fuel costs $284.28 Change in jet fuel costs $5.38 Baggage fees $4.85 Change in baggage fees $4.03 On-time performance 75.89 Change in on-time performance 1.30 Herfindahl index 4,133.46 Change in Herfindahl index 134.68 Market share 29.94 Change in market share 4.22 Seat capacity 164,246.6 Change in seat capacity 2,324.32 Departures offered 500.00 Change in departures offered 18.50

Standard Deviation Minimum $57.90 $26.79 $20.19 $52.80 $75.64 $7.51 $6.61 5.66 6.12 1,847.17 741.45 23.50 360.36 193,498.2 11,357.2 367.23 92.10

Maximum

$44.44 $238.34 $141.64

$538.74 $275.75 $124.05

$181.40 $396.50 $181.50 $110.60 $0 $25.00 $0 $20.00 54.8 92.6 21.7 37.6 2,022.98 10,000 3,140.38 4,662.23 0.01 100 2,370.37 4,497.51 50 1,231,087 67,160 138,271 0 2,476 571 1,019

RESULTS The results of estimating Eq. (1) through OLS and instrumental variables are presented in Table 5. The OLS estimates are included to assess the stability of the parameter estimates, which are shown to suffer significantly in some cases from omitted variable bias. This is corrected in the spatial autoregressive model, which is also a better fit for the data; therefore, in what follows we focus on the spatial results in Table 5. The remainder of this section is broken into three subsections: the first subsection discusses the estimates on the market and product differentiation variables, while the second subsection analyzes the impact of changes in jet fuel costs on changes in ticket pricing, and the third subsection examines the impact of baggage fees on airfares. Estimates of Changes in Market and Product Differentiation Variables on Ticket Price Changes Focusing on the coefficient estimates of the market and product differentiation variables in Table 5, we find several results that coincide with the findings

Baggage Fees and Changes in Airline Ticket Prices

187

Table 5. Estimation Results.

Change in jet fuel costs Change in baggage fees Change in baggage fees on routes served by Southwest Change in mean bag fees  Southwest Airlines Change in Herfindahl index Change in market share Change in seat capacity Change in departures offered Change in on-time performance Weighted change in average ticket price of spatial competitors Constant

Adjusted R2 Observations

OLS

Spatial Autoregressive Model

0.129 (0.004) 0.725 (0.057) 0.657 (0.064) 0.814 (0.092) 0.002 (0.0003) 0.009 (0.001) 0.001 (0.0001) 0.042 (0.006) 0.007 (0.043)

0.016 (0.006) 0.243 (0.051) 0.004 (0.060) 0.793 (0.076) 0.0002 (0.0003) 0.002 (0.001) 0.0003 (0.00004) 0.030 (0.005) 0.004 (0.036) 0.856 (0.038) 0.070 (0.285)

2.502 (0.317) 0.20 9,656

0.46 9,656

Notes: The dependant variable is the year-over-year change in quarterly average ticket prices by origin-destination pair. Standard errors are in parentheses. Significant at 10%.  Significant at 5%.  Significant at 1%.

of the aforementioned airline pricing literature. First, increases in route/ market concentration, as measured by the Herfindahl index, and increases in a firm’s route specific market share both increase airfares, although this result is only significant for changes in an airline’s market share. Thus, as firms gain market power on a given route, they use this market power to their advantage and increase their airfares on the route. Likewise, the results in Table 5 indicate that increases in the number of departures an airline offers on a given route tend to increase their prices on

188

KEVIN E. HENRICKSON AND JOHN SCOTT

that route. Intuitively, an airline would offer more departures on a route if it determined that there is unmet demand for these extra departures. However, the fact that there is unmet demand on the route also implies that it hasn’t been profitable to offer these extra departures in the past; therefore, the airline will need to increase their fares on these routes to make the extra departures profitable. Our results also indicate that as firms decrease the number of seats on a route, holding the number of departures constant, they will increase their airfares on the route as there are fewer seats available to travelers. The coefficient on changes in the firm’s on-time performance is insignificant, indicating that firm’s do not alter their ticket prices based on their on-time performance, perhaps because they often have no control over this performance, as it is often determined by weather and/or airport delays. Finally, the estimate on the weighted change in average ticket prices of spatial competitors is positive and significant. This estimate indicates that fares are positively correlated on routes. That is, if a firm’s competitors on a route increase/decrease their fares, the airline will tend to follow suit and change their fares in the same direction. This result is consistent with airlines competing with one another as price cuts are met with price cuts. Also recall that the specification of this variable includes airlines at nearby airports, implying that this spatial competition spills over to other airports serving the same origin and destination cities.

Estimates of Changes in Jet Fuel Prices on Ticket Price Changes As mentioned previously, jet fuel prices have increased dramatically over the past several years, which have subsequently adversely impacted airline costs. The estimates presented in Table 5 indicate that these costs are passed along to travelers in the form of higher ticket prices, with a one cent increase in jet fuel costs increasing ticket prices by 1.6 cents. While this impact seems relatively benign, Fig. 2 shows the aggregate increase in yearly consumer expenditures on a typical airline’s route for various levels of jet fuel increases. Specifically, the average airline’s route in our data receives 50,944 passengers every quarter, and with each of these consumers spending $1.6 more for every $1 increase in jet fuel costs, consumers will spend approximately $320,000 more every year for their tickets on this one airline’s route. Aggregating this further to account for all airlines and all routes, it becomes evident that increases in jet fuel costs not only increase airline costs, but also can be expected to increase consumer expenditures on airfare by millions of dollars.

200

400

600

800

189

0

Estimated Increase in Yearly Consumer Expenditures for Airline Tickets (000s of Dollars)

Baggage Fees and Changes in Airline Ticket Prices

0

.5

1

1.5

2

Jet Fuel Price Increase

Fig. 2.

The Estimated Impact of Increases in Jet Fuel Prices on Yearly Consumer Expenditures for Airline Tickets on an Average Route.

Estimates of Changes in Baggage Fees on Ticket Price Changes Turning our attention to the impact of baggage fees on airfares, our results in Table 5 indicate that increases in baggage fees impact all consumers, whether they travel on an airline charging bag fees or not. Specifically, we find that for airlines charging baggage fees, a one dollar increase in bag fees leads to a $0.24 decrease in airfares. Thus, for a traveler who checks one bag, they can expect to pay $0.76 more for every one dollar increase in baggage fees. This result also shows how these airlines are able to use their baggage fees to offset the per-ticket revenue that they lose by charging lower airfares, which in turn enables them to be more competitive when it comes to selling tickets. The result for airfares on routes also served by Southwest Airlines is negative, implying that the direct competition from Southwest reduces ticket prices further as baggage fees increase, but this estimate is insignificant. Our results in Table 5 also indicate that baggage fees impact Southwest Airline’s pricing of their airfares. Specifically, we find that a one dollar increase in the average baggage fees charged by Southwest’s competitors on a route increases Southwest’s airfares by $0.73. That is, Southwest through their ‘‘Bags Fly Free’’ advertising campaign has been able to leverage this

140

150

160

170

180

190

KEVIN E. HENRICKSON AND JOHN SCOTT

Impact on Average Ticket Price (in dollars)

190

10

15

20

25

Baggage Fees Southwest Airlines

Fig. 3.

Bag Fee Charging Airlines

The Estimated Impact of Increases in Baggage Fees on Average Ticket Prices for Southwest Airlines and Baggage Fee Charging Airlines.

new found product differentiation into higher ticket prices in much the same way as our previous result demonstrated that ticket prices are positively correlated to those of their spatial competitors. Fig. 3 summarizes these three results by illustrating the impact of baggage fees on both the average airfare of fee charging airlines and on the average airfare of Southwest Airlines. Notice that in this figure, the two slopes are $0.76 and $0.73 respectively, showing that the increases in baggage fees over the past two years have increased the amount that a traveler will pay, regardless of their choice of airlines, by nearly identical amounts.

CONCLUSION While there are many studies examining airline pricing, we are the first, to our knowledge, to examine the impact on airline pricing of the introduction of service fees, such as the baggage fees introduced in 2008. These fees were introduced to help offset some of the increases in costs associated with the increasing price of jet fuel, while still allowing the airlines to remain competitive with low-cost carriers on airfares. Our results indicate that these baggage fees have allowed firms to lower their airfares to become more competitive, while still increasing their

Baggage Fees and Changes in Airline Ticket Prices

191

revenues. Specifically, we find that a one dollar increase in baggage fees causes the firms charging the baggage fees to reduce their ticket prices by $0.24, so that a passenger with one checked bag can expect to pay $0.76 more for every one dollar increase in baggage fees. These fees have also allowed Southwest Airlines to increase their ticket prices, as they have leveraged their ‘‘Bags Fly Free’’ advertising campaign into an element of their product differentiation. Our results indicate that travelers on Southwest can expect to pay $0.73 more for every one dollar increase in baggage fees charged by Southwest’s competitors on any given route. As such, even if bags fly free on Southwest, it appears that their customers don’t, as they end up paying approximately the same amount as flyers on bag fee charging airlines through higher ticket prices.

NOTES 1. These estimates were calculated using the U.S. Department of Transportation’s Form 41 Financial Data, in which each carrier reports their operating costs and fuel costs. 2. The jet fuel prices presented in Table 1 were collected from the U.S. Energy Information Administration. 3. Note that some airlines such as Delta, United, U.S. Airways, and Continental offer small, usually $2–$3, discounts off of the fees reported in Table 2 if the traveler pays their baggage fees online. 4. The baggage revenues were obtained from the U.S. Department of Transportation’s Schedule P-12 that breaks airline revenues into their various components. 5. Spatial pricing is not limited to the airline industry, but rather has been examined for: gasoline by Pinske, Slade, and Brett (2002); hamburgers by Kalnins (2003); college tuition by McMillen, Singell, and Waddell (2007); and sports tickets by Henrickson (2011). 6. Thus, the spatial weighting matrix has 12 blocks along the diagonal, one block for each quarter between the first quarter of 2007 and the fourth quarter of 2009. 7. Note that several alternative specifications of the spatial weighting matrix were examined and the results of this study are robust to these alternative specifications. 8. Note that many of the aforementioned studies include other explanatory variables such as distance, population, and income; however, because our dependent variable is the change in ticket prices rather than the absolute value of ticket prices, we don’t include these variables as there is little, if any, variation in these variables from year to year.

REFERENCES Anselin, L. (1988). Spatial econometrics: Methods and models. Boston, MA: Kluwer Academic Publishers. Borenstein, S. (1989). Hubs and high fares: Dominance and market power in the U.S. airline industry. RAND Journal of Economics, 20, 344–368.

192

KEVIN E. HENRICKSON AND JOHN SCOTT

Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the U.S. airline industry. Journal of Political Economy, 102, 653–683. Daraban, B., & Fournier, G. M. (2008). Incumbent responses to low-cost airline entry and exit: A spatial autoregressive panel data analysis. Research in Transportation Economics, 24, 15–24. Drakos, K. (2004). Terrorism-induced structural shifts in financial risk: Airline stocks in the aftermath of the September 11th terror attacks. European Journal of Political Economy, 20, 435–446. Dresner, M., Lin, J.-S. C., & Windle, R. (1996). The impact of low-cost carriers on airport and route competition. Journal of Transport Economics and Policy, 30, 309–328. Evans, W., & Kessides, I. (1993). Localized market power in the U.S. airline industry. The Review of Economics and Statistics, 75, 66–75. Fournier, G. M., Hartmann, M., & Zuehlke, T. (2007). Airport substitution by travelers: Why do we have to drive to fly? In: D. Lee (Ed.), Advances in airline economics: The economics of airline institutions, operations and marketing (Vol. 2, pp. 209–234). Amsterdam: Elsevier Ltd. Gerardi, K. S., & Shapiro, A. H. (2009). Does competition reduce price dispersion? New evidence from the airline industry. Journal of Political Economy, 117, 1–37. Goolsbee, A., & Syverson, C. (2008). How do incumbents respond to the threat of entry? Evidence from the major airlines. The Quarterly Journal of Economics, 123, 1611–1633. Guzhva, V. S., & Pagiavlas, N. (2004). US commercial airline performance after September 11, 2001: Decomposing the effect of the terrorist attack from macroeconomic influences. Journal of Air Transport Management, 10, 327–332. Henrickson, K. E. (2011). Spatial competition and strategic firm relocation. Economic Inquiry, forthcoming. Ito, H., & Lee, D. (2005). Assessing the impact of the September 11 terrorist attacks on U.S. airline demand. Journal of Economics and Business, 57, 75–95. Kalnins, A. (2003). Hamburger prices and spatial econometrics. Journal of Economics and Management Strategy, 12, 591–616. Mazraati, M. (2010). World aviation fuel demand outlook. OPEC Energy Review, 34, 42–72. McMillen, D. P., Singell, L. D., & Waddell, G. R. (2007). Spatial competition and the price of college. Economic Inquiry, 45, 817–833. Morrison, S. A. (2001). Actual, adjacent, and potential competition: Estimating the full effect of southwest airlines. Journal of Transport Economics and Policy, 35, 239–256. Pinske, J., Slade, M., & Brett, C. (2002). Spatial price competition: A semiparametric approach. Econometrica, 70, 1111–1153. Whinston, M. D., & Collins, S. C. (1992). Entry and competitive structure in deregulated airline markets: An event study analysis of people express. RAND Journal of Economics, 23, 445–462. Windle, R., & Dresner, M. (1999). Competitive responses to low cost carrier entry. Transportation Research Part E, 35, 59–75.

CHAPTER 9 FRAGMENTATION OF NORTH ATLANTIC AND TRANSPACIFIC AIR TRANSPORT MARKETS – WITHER THE HUBS? Andreas Knorr, Andreas Lueg-Arndt and Alexander Eisenkopf INTRODUCTION In their respective market outlooks, both Boeing and Airbus forecast strong growth in intercontinental passenger traffic until 2029. However, they differ substantially with respect to their assessment of the future development of airline (and alliance) networks. These deviating projections have, in turn, massively influenced their product range. Boeing, having long predicted a major growth in intercontinental point-to-point operations – based on the so-called fragmentation (dehubbing) hypothesis – has consistently opted for the development of the B787 (Dreamliner) family of midsized, and extremely efficient, wide-body aircraft. Airbus, on the contrary, is forecasting a substantial increasing demand for hub-to-hub traffic, which according to the company, will require airlines to purchase a large number of very large aircraft (VLA), especially its Airbus 380. Though both manufacturers did not put all their money where their mouths are – Boeing has reacted to the Airbus 380 challenge with an updated derivative of its Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 193–212 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003011

193

194

ANDREAS KNORR ET AL.

Boeing 747 flagship, the Boeing 747-8 Intercontinental, while Airbus is targeting its proposed Airbus 350 family against both the Boeing 787 and Boeing 777 – the fragmentation hypothesis remains one of the most controversial issues in the civil aviation community today. Regardless of which scenario will eventually turn out to be more realistic, either will impact tremendously on aircraft manufacturers, on the airlines’ route and fleet planning decisions as well as airport operators. This chapter consists of three parts. In the theoretical analysis, we will first define fragmentation and identify its various legal, technological, and economic drivers. Second, we will discuss the economics of hub-based operations compared to point-to-point services in intercontinental passenger services. Finally, in our empirical analysis, which is based on Official Airline Guide (OAG) data covering airline schedules over the 1982–2007 period, we will test the fragmentation hypothesis for both North Atlantic traffic (North America2geographical Europe) and transpacific markets (North America2Asia). In this context, our analysis will focus on the three scenarios for future network segmentation that were identified (and assessed) by Airbus in its Global Market Forecast published in 20071: hub-to-hub, hub-to-secondary (non-hub) airports and secondary-to-secondary airports (point-to-point routes, or non-hub-to-non-hub routes). In this context it might be obvious that all hub-to-hub services are operated as point-to-point services, too, notwithstanding this fact appears worth mentioning. Nevertheless, we will abide in the following by the standard definition that equates point-to-point (non-hub-to-non-hub) services with air links between secondary airports only.

FRAGMENTATION IN AVIATION In economics, the term fragmentation is generally used to describe a particular pattern in the international division of labor in which there is no trade in final goods. Instead, an international supply chain exists in which intermediate goods are produced by different manufacturers in different locations worldwide for final assembly somewhere else (global sourcing). In aviation, by contrast, the term fragmentation is mostly used to describe systemic inefficiencies that result in higher transaction and operating costs for airlines as well as external costs. Examples include:  The fragmentation of air traffic control systems, which has resulted in a very inefficient use of available air space capacity and, in turn, produced high congestion costs and substantial coordination costs.2

Fragmentation of North Atlantic and Transpacific Air Transport Markets

195

 The fragmentation of safety oversight standards responsibilities (which has led the United States and the European Union to ‘‘blacklist’’ a large number of foreign airlines, i.e. to suspend their traffic rights to/from the United States and/or the European Union).3  The fragmentation of international air transport markets that has led to a significant increase in the number of transatlantic and transpacific citypairs being served. Only this last variant of aviation-related fragmentation is linked to the subject of this chapter and will be explored further in the following. Typically, fragmentation is explicitly or implicitly equated with an increase of nonstop point-to-point services linking two secondary airports A and B. The supporters of the fragmentation hypothesis argue that, over time, pointto-point traffic will not only replace a significant number of today’s routine connecting flights from A to B via hub airport C and, as a result, give rise to a dehubbing (deconcentration) process at the airport level.4 Moreover, they hold that the rise in point-to-point services will also increase the demand for smaller and medium-sized (wide-body) aircraft at the expense of large and VLA such as the Boeing 747 and, particularly, the A380. Unsurprisingly, given the vastly different strategic thinking behind their respective flagship products, the (future) Boeing 787 and the Airbus 380, Boeing and Airbus have long been entrenched on opposite sides of the fragmentation debate.5

The Legal Prerequisites – Deregulation and Liberalization The legal prerequisite6 for the fragmentation of international air services was the gradual abolition – full or at least in part – of some of the numerous legal restrictions of market access, capacities, and frequencies, which have characterized aviation markets to this day. To begin with, almost all countries still bar foreign carriers from operating domestic revenue services (cabotage ban)7 and prohibit foreigners to own and/or to control domestic carriers.8 With respect to cross-border traffic, bilateral air service agreements9 negotiated by the two countries’ governments continue to strip carriers from third countries of the right to offer commercial services between the two countries.10 In the past three decades, however, some progress toward deregulation has been made. As for domestic air travel, policymakers in most industrialized countries have facilitated industry consolidation to some degree by relaxing market access, capacity, frequency, and pricing restrictions, following the

196

ANDREAS KNORR ET AL.

lead of the United States and their landmark 1978 Airline Deregulation Act.11 For airlines registered in the European Union, all such restrictions were lifted throughout the single market (plus Norway and Iceland) in January 1993 when the Third Liberalization Package took effect – de facto the first successful multilateral attempt ever to deregulate air transport markets (though only regionally).12 Finally, based on the 1979 International Air Transportation Competition Act, the US government began to export its deregulation philosophy to crossborder air links.13 Over the next few decades, under this so-called ‘‘open skies’’ approach14 numerous bilateral air service agreements were renegotiated. As a result, capacity and frequency controls were gradually abolished (or at least substantially lowered), more gateway points were included, multiple designations of airlines were permitted (although only airlines registered in either country were eligible), and governmental controls over fares were reduced (and mostly lifted). However, the massive restrictions on foreign ownership and the systematic discrimination of third-country carriers remained intact. At least, for the European Union and its member states, these rules of the game were fundamentally changed by the ‘‘open skies’’ judgment of the European Court of Justice on November 5, 2002.15 Arguing that all bilateral air service agreements of individual member states with a third country run afoul of fundamental single market rules for discriminating against those of EU-based airlines, which are registered in other member states, all bilaterals concluded by individual member states are being replaced by so-called horizontal agreements. They ensure that all traffic rights must be guaranteed to all other ‘‘community air carriers’’ as well. In practice, this gamechanging new approach has enabled Air France to offer (ninth freedom) services between London Heathrow and Los Angeles (now canceled), while a British Airways subsidiary – appropriately named Open Skies – currently serves the routes New York–Paris and Washington D.C.–Paris.

Technological Drivers Though technologically feasible at least from the East Coast to Ireland and the UK after the entry into service of Lockheed’s Super Constellation in the late 1940s, nonstop commercial transatlantic services remained the exception rather than the rule before the jet age began in the late 1950s; due to the much longer distances on transpacific routes (even from the West Coast to East Asia) large-scale nonstop operations were missing until the late 1980s when the Boeing 747–400 entered revenue service.16

Fragmentation of North Atlantic and Transpacific Air Transport Markets

197

Before 1969, the year of the original Boeing 747 variant’s – the Boeing 747–100 – first flight, the Boeing 707 and McDonnell Douglas’ DC8 family were the only jet aircraft available for nonstop intercontinental flights of up to around 8,000 kilometers (5,000 statute miles). Typically, seating around 140–210 passengers in a standard three-class configuration, the Boeing 747 as the world’s first wide-body aircraft offered nearly twice their passenger capacity (366 in a standard three-class configuration). From the mid-1970s, Boeing’s competitors began to fill that enormous capacity gap with three designs that typically accommodated roundabout 260 passengers each. While US manufacturers McDonnell Douglas and Lockheed opted for three-engined solutions (the DC10 and the L1011 Tristar, respectively), the (at that time) infant European Airbus consortium rather launched the first ever twin-engined wide-body in aviation history: the Airbus 300 (first flight: 1972), which was complemented by a smaller derivative, the Airbus 310, 10 years later. By that time, Boeing had reacted with the development of the Boeing 767 wide-body and a slightly smaller new single-aisle aircraft, the Boeing 757, both of which were capable of carrying some 200–230 passengers on US transcontinental routes. Today, Boeing and Airbus, the two remaining manufacturers of large commercial jet aircraft in the world, both offer a highly differentiated portfolio of very large (Airbus 380), large (Boeing 747, Boeing 777, Airbus 340–500, Airbus 340–600), and medium-sized (Boeing 767, Airbus 330, Airbus 340–300) wide-body aircraft for intercontinental flights; special versions of some of these aircraft are even capable of ultra-long range services of around 14,000 kilometers (8,700 statute miles). What is more, the ranges of latest variants of Boeing’s and Airbus’ extremely successful single-aisle aircraft Boeing 737 and Airbus 320 families – which seat between 120 and 190 passengers – have been pushed to some 6,000 kilometers (3,750 statute miles).

Economic Considerations The legal and technological drivers of fragmentation cannot be seen in isolation. Moreover, and obviously, both factors – per se and in combination – determine the economic viability of airline operations to a substantial degree. The development of ETOPS rules is a case in point. ETOPS is the acronym for Extended-range Twin-engine Operation Performance Standards.17 They regulate the farthest distance (in terms of flight time) an aircraft is permitted to stray away from the nearest available diversion airport en route. Initially set by the Federal Aviation Administration (FAA) at 60 minutes in the pre-jet era,

198

ANDREAS KNORR ET AL.

ETOPS rules were gradually extended to 90 minutes in 1985 to allow Trans World Airlines (TWA) to operate their Boeing 767–200 aircraft unrestricted on transatlantic services from their St. Louis hub. The statistically proven high reliability of modern jet engines has allowed safety oversight bodies to gradually broaden the geographical scope for commercial ETOPS operations even further. Today 180 minutes are the norm (with 207 minutes being the legal maximum and Boeing lobbying hard for 240 minutes), effectively freeing airlines from most commercial restrictions regarding intercontinental operations of twin-engined aircraft. As ETOPS rules never applied to four-engined aircraft, this has resulted in many airlines having substituted twinjets for four-holers (e.g., the Boeing 777–300 for the Boeing 747, both of which carry a similar number of passengers over a comparable range). Moreover, the development of the product ranges of aircraft manufacturers after – and owing to – deregulation provides strong evidence for the narrow interrelationship between (changes of) the regulatory regime, aircraft technology, and airline economics in civil aviation. Before deregulation, most major carriers in the United States (though not in Europe) did not base their operation on a hub-and-spoke system, offering a large number of point-topoint flights instead. Load factors were rather poor, but regulated fare levels were set in such a manner to guarantee airlines an average rate of return of 12% based on a 55% load factor.18 Accordingly, incentives to reduce costs, for example, by deploying more efficient aircraft, were small. However, in the years after deregulation took effect, airlines faced strong economic pressures to optimize their route network and to increase capacity utilization through a combination of price differentiation, loyalty programs and, most of all, to restructure their networks by bundling traffic flows and routing passengers via hub airports. For the airlines, the transition to a huband-spoke system offers many advantages (which have been discussed at length in the specialist literature).19 First and foremost, consolidating traffic flows through one or more hub airports permit airlines to serve a substantially larger number of city-pairs – many of which, especially long-haul cross-border air links,20 would not be economically viable as point-to-point services due to insufficient Origin and Destination (O&D) traffic – at higher frequencies and with a comparatively much smaller fleet of aircraft. This consolidation, in turn, translates not only into cost savings due to economies of scope as well as density owing to higher load factors and the larger route network. What is more, residents at major hub cities – including local businesses – reap the tangible and intangible benefits of (frequent, nonstop) access to a large variety of destinations.

Fragmentation of North Atlantic and Transpacific Air Transport Markets

199

The well-known drawbacks of hub-and-spoke systems include a certain preference among (time-sensitive) travelers for less time-consuming nonstop services (which also lower the risk of lost checked baggage), a peak-off-peak pattern with respect to ground capacity and airport staff utilization, lower average daily aircraft utilization, and costly delays in adverse weather conditions. Finally, hub-induced airport congestion might hinder the airline’s expansion – a restriction which, however, might be overcome by operating larger aircraft into slot-constrained airports (provided that ground facilities can cope with the additional passengers). Such are the advantages of hubbing that, contrary to common belief, even all major low-cost carriers (LCC) have long adopted, at least to some degree, a similar approach, which they refer to as either ‘‘focus cities’’ (Southwest Airlines) or ‘‘bases’’ (Ryanair) – the only difference being that LCCs normally do not officially schedule (and sell) connecting services through their ‘‘hubs.’’

EMPIRICAL FINDINGS Data Base (Limitations) and Methodology Our empirical research was based on OAG flight schedules data. The OAG data available to us covered the time period from 1979 to 2008 and contains variables based on published information on scheduled flights of more than 900 airlines at more than 3,600 airports worldwide. Variables include the departure airport, the destination airport, flight frequencies, aircraft type, and seat capacity for each flight and the number of stops during the flight. OAG data do suffer from a number of limitations, however. First, they only provide insight into scheduled flights and not into realized traffic flows. Given that we are interested in the structure of the aviation network, we do not consider this shortcoming to be much of a problem. Second, OAG data only registers scheduled services. We have deleted all-cargo flights from the data set and consider passenger flights only (we concentrated on directly operated flights; code-share flights were deleted). Finally, the OAG database only lists direct flights. Our analysis takes into account recorded data for scheduled flights measured in available seat kilometers (ASKs) for 1982 – when the first few liberal bilateral air service agreements on North Atlantic routes had entered into force – 1987, 1992, 1997, 2002, and 2007. For our purpose we first derived all city-pairs served and, additionally, the seat capacity offered.

200

ANDREAS KNORR ET AL.

Afterward, based on airline reports we identified hub and non-hub airports in the sample and calculated the number of hub-to-hub, hub-to-non-hub and point-to-point connections. Finally, we analyzed what aircraft types were operated on the city-pairs. In line with the purpose of our study we focused our analysis on all scheduled flights on all major trunk routes between North America and Europe and between North America and Asia, respectively. ‘‘North America’’ was defined to encompass both Canada and the USA,21 ‘‘Europe’’ included Denmark, France, Germany, Italy, the Netherlands, Norway (only Oslo-Gardermoen and the former airport Oslo-Fornebu), Portugal, Spain, Sweden, and United Kingdom and ‘‘Asia’’ comprised Australia, China (including Hong Kong), Japan, Korea, New Zealand, the Philippines, Taiwan, and Thailand.

Results North America2Europe Table 1 shows the number of North Atlantic city-pairs operated between 1982 and 2007. In the last year covered by our analysis, 170 city-pairs were served in total, which means that the number of routes has more than doubled over the last 25 years. The data reveals that the liberalization of air transportation on North Atlantic routes led to a substantial boost for air traffic overall, with hubto-hub and hub-to-non-hub traffic benefitting the most. The majority of the flights currently on offer are hub-to-hub routes – with a share of 74% in 2007, down from a 79% average over the entire observation period. Only 6% of the routes operated in 2007 were point-to-point connections bypassing hubs.

Table 1.

Number of City-Pairs for Different Route Types, North America2Europe. 1982

1987

1992

1997

2002

2007

Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Hub-to-hub Hub-to-non-hub Point-to-point Total

59 13 6 78

75 17 8 100

77 14 6 97

Source: OAG; own calculations.

79 14 7 100

110 18 3 131

84 14 2 100

111 22 5 138

80 16 4 100

115 27 5 147

78 19 3 100

126 34 10 170

74 20 6 100

Fragmentation of North Atlantic and Transpacific Air Transport Markets

Table 2.

201

Seat Capacity on Different Route Types, North America2Europe (mio. ASKs). 1982

1987

1992

1997

2002

2007

Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Hub-to-hub 8.35 82 11.87 83 16.46 84 19.08 87 21.82 88 24.81 82 Hub-to-non-hub 1.75 17 2.31 16 3.09 16 2.67 12 2.74 11 4.96 16 Point-to-point 0.14 1 0.19 1 0.03 0 0.21 1 0.14 1 0.52 2 Total 10.24 100 14.37 100 19.58 100 21.96 100 24.70 100 30.29 100 Source: OAG; own calculations.

As shown in Table 2 below, total seat capacity on North Atlantic routes has grown by 195% between 1982 and 2007. The growth of hub-to-hub traffic (197%) and hub-to-non-hub traffic (183%) has been similar. Capacity on point-to-point routes has risen by 271% over the same period. Although point-to-point routes account for the highest growth rate with regard to seat capacity, their market share can be neglected and hub-to-hub routes dominate the market. Over 80% of annually offered total seat capacity is operated on North Atlantic hub-to-hub routes. Our data shows no clear change in the share of the three categories of city-pairs over the whole period (hub-to-non-hub: B16%; point-to-point: B1%). A peculiarity we found in the data set (for which we did not produce a graph in the text but some separate tables in the Appendices 2 and 3) was the continuous downscaling of the (average) aircraft size over the observation period. The main underlying reason behind this trend appears to be the end of the Boeing 747’s long-standing de facto monopoly for long distance traffic. Accordingly, between 1982 and 2007 the share of the Boeing 747 declined from over 70% to 21% on hub-to-hub routes and from nearly 50% to a mere 11% on hub-to-non-hub routes. The market share loss on point-to-point relations was approximately 36%. Essentially, it was replaced by airlines by smaller aircraft types such as the Airbus 340, the Airbus 330 and, most of all, the Boeing 767 and the Boeing 777, which offered a better match of capacity with actual demand at, at least, comparable if not lower costs per available seat mile. North America2Asia In 2007, the transpacific market as defined above comprised a total of 108 city-pairs, a number that has more than tripled since 1982 as a result of strong economic growth in Asia and closer trade links between the two regions. In particular, it is noteworthy that the number of hub-to-hub

202

ANDREAS KNORR ET AL.

Table 3.

Number of City-Pairs for Different Route Types, North America2Asia. 1982

1987

1992

1997

2002

2007

Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Hub-to-hub Hub-to-non-hub Point-to-point Total

8 20 6 34

24 58 18 100

19 19 11 49

39 39 22 100

25 32 23 80

31 40 29 100

40 30 21 91

44 33 23 100

48 27 17 92

52 29 18 100

59 29 20 108

55 27 19 100

Source: OAG; own calculations.

connections has risen by a significant amount. As shown in Table 3 below, the share of hub-to-hub routes increased from 24% (8 routes in 1982) to 55% in 2007 (59 routes). In the same period the share of hub-to-non-hub links fell from nearly 60% to less than 30%, while the share of point-topoint connections has remained quite stable at roughly 20%. A different picture has to be painted, however, if the seat capacity offered for transpacific flights is considered (Table 4). By this measure, the share of both hub-to-non-hub traffic and point-to-point traffic decreased, while the share of hub-to-hub traffic increased from 30% in 1982 to 69% in 2007. A closer inspection of the data for the hub-to-non-hub segment reveals that much more capacity was offered on flights from hubs in Asia to non-hubs in North America (10 times higher) compared to the number of seats available from North American hub airports to non-hubs in Asia. In our view, this imbalance reflects the different market access strategies of Asian and US carriers. While the former have added an ever increasing number of US and Canadian spokes to their home bases, United Airlines and Northwest have traditionally used Tokyo Narita as their primary gateway not only in Japan but also as a gateway to much of East and Southeast Asia (last, but certainly not least, with the intention to fully exploit their generous fifth freedom rights to tap into the huge O&D market the Tokyo metro area represents for this traffic region). Similar to the development on North Atlantic routes, though much more slowly, the market share of the Boeing 747 decreased in all three route categories. The largest decrease so far was recorded on the point-to-point routes. Here, the market share effectively plummeted from a whopping 97% in 1982 to a mere 20% in 2007. In contrast to North Atlantic markets, where Airbus aircraft played an important role in this replacement process, on transpacific routes the Boeing 747 was nearly completely replaced by slightly smaller Boeing alternatives, in particular the Boeing 777 family.

Fragmentation of North Atlantic and Transpacific Air Transport Markets

203

Seat Capacity on Different Route Types, North America2Asia (mio. ASKs).

Table 4.

1982

1987

1992

1997

2002

2007

Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Abs. In % Hub-to-hub Hub-to-non-hub Point-to-point Total

1.49 2.88 0.56 4.93

30 58 12 100

3.32 3.81 0.69 7.82

42 5.98 47 9.00 56 9.03 64 11.16 69 49 4.87 38 5.47 34 4.33 30 4.38 27 9 1.89 15 1.63 10 0.86 6 0.74 4 100 12.74 100 16.10 100 14.22 100 16.28 100

Source: OAG; own calculations.

CONCLUSION Our analysis has demonstrated that fragmentation exists both on the North Atlantic and the transpacific markets – where it has begun about two decades later and has progressed much more slowly due to longer distances and less pervasive ‘‘open skies’’ policies. However, fragmentation has not been accompanied, let alone caused, a process of dehubbing. Quite the contrary, the strongest carriers’ primary hubs in all three traffic regions we considered appear to have benefitted the most from the increasing number of secondary cites overseas that have been added as spokes. This may have had two effects (the strength of which we lacked the data to measure): an extension of ‘‘regional’’ catchment areas which, especially in densely populated central Europe, now overlap more than ever before – thereby intensifying hub competition. By contrast, demand for pure point-to-point services remains very low due to the current mismatch of actual O&D traffic and available aircraft technology.

NOTES 1. Cf. Airbus (2007). 2. With its Single European Sky initiative, the European Commission has tried for years to eliminate the costly patchwork of 34 separate air traffic control systems in the European Union’s 27 member states but so far little progress has been made. 3. Cf. Knorr (2006). 4. Cf. Gellmann, Weber, Hamlin, and Aboulafia (2004); Leeham Co. LLC (2004); Boston Consulting Group (2004); Heymann (2006).

204

ANDREAS KNORR ET AL.

5. In its 2001 Current Market Outlook, Boeing kicked off the debate, predicting that the competitive pressures unleashed by deregulation and more liberal bilaterals would force airlines to offer more frequencies with smaller aircraft such as the Boeing 787 on intercontinental flights. Airbus, by contrast, has not ceased to point to the fact that the combination of growing urbanization and increasingly slotconstrained airports will translate into brisk demand for very large aircraft such as the Airbus 380 (cf. Airbus, 2007, p. 20ff; Boeing, 2001, p. 41ff). 6. For a more detailed discussion of the drivers of long-haul air route development see Weber and Williams (2001). 7. In legal terms, foreign carriers are refused the eighth and the ninth freedoms of the air (true cabotage and stand alone cabotage, respectively) – For a full explanation of the nine freedoms of the air cf. ICAO (n.d.). 8. Cf. Pilarski (2007, p. 193ff). 9. Cf. Doganis (2001, p. 19ff). 10. Except in the rather rare cases when third-country carriers were granted the fifth freedom right. 11. Technically, deregulation of domestic air transport in the USA began with the 1977 Air Cargo Deregulation Act. Cf. Doganis (2001, p. 24). 12. Although the Third Liberalization Package took effect in January 1993, the cabotage ban within the single market was in force until April 1997. 13. The trend setter, however, was the 1978 bilateral between the USA and the Netherlands. 14. For details see de Murias (1989). 15. Cf. European Court (2002). 16. Cf. Boberg and Choy (1988). 17. Cf. FAA (2008). – A powerful tool to calculate the impact of alternative ETOPS rules on the effective ranges of twinjets is available at http://gc.kls2.com/. 18. Cf. Bailey, Graham, and Kaplan (1985, p. 20ff). 19. For a comprehensive overview of the economics of hub-and-spoke systems in aviation cf. Gillen, Hinsch, Mandel, and Wolf (2001, p. 25ff). 20. Theoretical explanations are derived from standard gravity equations (cf. Neary, 2001). Moreover, significant border effects exist in international air transport. Cf. Klodt (2004). 21. A list of US-hubs can be found in Appendix 1.

REFERENCES Airbus. (2007). Flying by Nature. Global Market Forecast 2007–2026, Toulouse. Retrieved from http://www.airbus.com/fileadmin/documents/gmf/PDF_dl/00-all-gmf_2007.pdf Bailey, E., Graham, D., & Kaplan, D. (1985). Deregulating the airlines. Cambridge, MA: MIT Press. Boberg, K., & Choy, D. J. L. (1988). Emerging trends in trans-pacific air routes. Journal of Travel Research, 26(3), 15–23. Boeing. (2001). Current market outlook 2001. Seattle, WA. Retrieved from http://www.as777. com/data/manufacturer/forecast/boeing_2001.PDF

Fragmentation of North Atlantic and Transpacific Air Transport Markets

205

Boston Consulting Group. (2004). Airports – Dawn of a new era, n.p. Retrieved from http:// www.bcg.com/impact_expertise/publications/files/BCGAirportsDawnNewEra.pdf de Murias, R. (1989). The economic regulation of international air transport. London: McFarland & Co Inc. Doganis, R. (2001). The airline business in the 21st century. New York, NY: Routledge. European Court. (2002). Report on Case C-466/98. n.p. Retrieved from: http://eur-lex. europa.eu/LexUriServ/LexUriServ.do?uri ¼ CELEX:61998J0466:EN:HTML FAA. (2008, June 14). Advisory circular. Extended operations (ETOPS and polar operations), Washington, DC. Retrieved from http://rgl.faa.gov/Regulatory_and_Guidance_Library/ rgAdvisoryCircular.nsf/0/2e0f31985abd83ef8625746b0057fd06/$FILE/AC%20120-42B.pdf Gellmann, A., Weber, H. J., Hamlin, G. W., & Aboulafia, R. L. (2004, March). A shadow critical project appraisal: The A380 program. Updated July 2004. Retrieved from http:// www.leeham.net/filelib/050416-shadow.pdf Gillen, D. W., Hinsch, H., Mandel, B., & Wolf, H. (2001). The impact of liberalizing international aviation bilaterals: The case of the Northern German region. Aldershot, UK: Ashgate Publishing Limited. Heymann, E. (2006, August 16). The future of the hub strategy in the air transport industry. Deutsche Bank Research Current Issues, Frankfurt/Main. Retrieved from http://www.dbresearch.de/PROD/DBR_INTERNET_EN-PROD/PROD0000000000201837.pdf ICAO. (n.d.). Freedoms of the air. Montre´al. Retrieved from http://www.icao.int/icao/en/trivia/ freedoms_air.htm Klodt, H. (2004). Border effects in passenger air traffic. KYKLOS, 57, 519–532. Knorr, A. (2006). Will blacklists’ enhance airline safety? Discussion Paper No. 32. German Research Institute for Public Administration Speyer, Speyer. Retrieved from http:// www.foev-speyer.de/publikationen/pubdb.asp?reihen_id¼3 Leeham, Co. LLC. (2004, June 15). Airbus and Boeing continue their sniping. Commercial Aviation Report No. 12. Issaquah. Retrieved from http://leeham.net/filelib/ CAR_15_06_04pp_12-13_ScBag.pdf Neary, J. P. (2001). Of hype and hyperbolas: Introducing the new economic geography. Journal of Economic Literature, XXXIV, 536–561. Pilarski, A. M. (2007). Why can’t we make money in aviation? Aldershot, UK: Ashgate Publishing Limited. Research and Innovative Technology Administration. (RITA). (2009). Air Traffic Hubs 2009. U.S. Department of Transportation, Washington, DC. Retrieved from http://www.bts. gov/programs/geographic_information_services/maps/hub_maps/2009/html/map.html Skyguide. (n.d.). US Hub city map – Build connecting flights. n.p. Retrieved from http:// www.skyguideonline.com/reference/hub.html Weber, M., & Williams, G. (2001). Drivers of long-haul air transport route development. Journal of Transport Geography, 4, 243–254.

206

ANDREAS KNORR ET AL.

APPENDIX 1: LIST OF US-HUBS IN ALPHABETICAL ORDER City

State

Albuquerque Anchorage Atlanta Austin Baltimore Billings Boston Charlotte Chicago-O’Hare Chicago-Midway Cincinnati Cleveland Columbus Dallas Denver Detroit Guam Honolulu Houston-George Bush Houston-Hobby Indianapolis Kansas City Las Vegas Los Angeles Louisville Memphis Miami Milwaukee Minneapolis-St. Paul Nashville New York-JFK New York-La Guardia Newark Oakland

New Mexico Alaska Georgia Texas Maryland Montana Massachusetts North Carolina Illinois Illinois Kentucky Ohio Ohio Texas Colorado Michigan Hawaii Texas Texas Indiana Missouri Nevada California Kentucky Tennessee Fort Lauderdale Wisconsin Minnesota Tennessee New York New York New York California

Fragmentation of North Atlantic and Transpacific Air Transport Markets

207

APPENDIX 1: (Continued) City

State

Orlando Philadelphia Phoenix Pittsburgh Portland Raleigh-Durham Salt Lake City San Diego San Francisco Bay San Juan Seattle St. Louis Tampa Washington-Reagan Washington-Dulles

Florida Pennsylvania Arizona Pennsylvania Oregon North Carolina Utah California California Porte Rico Washington Missouri Florida Virginia Virginia

Source: Research and Innovative Technology Administration (RITA), U.S. Department of Transportation (2009); Skyguide (n.d.).

208

ANDREAS KNORR ET AL.

APPENDIX 2: NORTH AMERICA2EUROPE A. City-pair types Absolute values Hub-to-hub Hub-to-spoke Point-to-point

Percentages Hub-to-hub Hub-to-spoke Point-to-point

B. Capacity (million ASKs) Absolute values Hub-to-hub Hub-to-spoke Point-to-point

Percentages Hub-to-hub Hub-to-spoke Point-to-point

1982 59 13 6 78

1987 77 14 6 97

1992 110 18 3 131

1997 111 22 5 138

2002 115 27 5 147

2007 126 34 10 170

1982 0.76 0.17 0.08 1

1987 0.79 0.14 0.06 1

1992 0.84 0.14 0.02 1

1997 0.80 0.16 0.04 1

2002 0.78 0.18 0.03 1

2007 0.74 0.20 0.06 1

1982 8.35 1.75 0.14 10.24

1987 11.87 2.31 0.19 14.37

1992 16.46 3.09 0.03 19.58

1997 19.08 2.67 0.21 21.96

2002 21.82 2.74 0.14 24.7

2007 24.81 4.96 0.52 30.29

1982 0.82 0.17 0.01 1

1987 0.83 0.16 0.01 1

1992 0.84 0.16 0.00 1

1997 0.87 0.12 0.01 1

2002 0.88 0.11 0.01 1

2007 0.82 0.16 0.02 1

1987

1992

1997

2002

2007 0.08

0.05

0.08

C. Aircraft types (share W 5%) Hub-to-hub 1982 AIRBUS INDUSTRIE A330-300 AIRBUS INDUSTRIE A340 AIRBUS INDUSTRIE A340-300 AIRBUS INDUSTRIE A340-600 BOEING 747-400

0.08 0.07 0.09

0.19

0.21

Fragmentation of North Atlantic and Transpacific Air Transport Markets

209

APPENDIX 2: (Continued) Hub-to-hub BOEING 747 BOEING 767-300 BOEING 767 BOEING 777 BOEING 747 (MIXED CONFIG) BOEING (DOUGLAS) DC10 LOCKHEED L1011 TRISTAR LOCKHEED L1011 TRISTAR 500

1982 0.65

1987 0.59

0.06

0.09

0.12

0.16

1992 0.38 0.09 0.15

0.14

1997 0.27 0.17 0.10 0.07

2002 0.07 0.15

2007

0.29

0.07 0.08 0.20

2002

2007

0.06

0.06

0.20

0.11 0.19 0.16 0.11 0.07

0.11

0.07 0.09

Hub-to-spoke AIRBUS INDUSTRIE A310 AIRBUS INDUSTRIE A330-200 BOEING (DOUGLAS) DC10 BOEING 747 BOEING 747 (MIXED CONFIG) BOEING 747-400 BOEING 757 BOEING 767 BOEING 767-300 BOEING 777 LOCKHEED L1011 TRISTAR LOCKHEED L1011 TRISTAR 500

1982

Point-to-point AIRBUS INDUSTRIE A310 AIRBUS INDUSTRIE A330-200

0.06 1987 0.02

1992 0.14

1997

0.32

0.22

0.14

0.11

0.41 0.08

0.39 0.10

0.14

0.16

0.06 0.27 0.09

0.06 0.06 0.24 0.16

0.10

0.10

0.09

0.06

0.06

0.07

0.11

1982

1987

1992

1997

0.10 0.26 0.11

2002 0.08

2007

0.23

0.20

210

ANDREAS KNORR ET AL.

APPENDIX 2: (Continued) AIRBUS INDUSTRIE A340 Point-to-point BOEING 707/720 BOEING 747-400 BOEING 747 BOEING 757-200 BOEING 757 BOEING 767-300 BOEING 767 BOEING 747 (MIXED CONFIG) BOEING (DOUGLAS) DC10 MCDONNELL DOUGLAS DC8 (ALL 60/70 SERIES) LOCKHEED L1011 TRISTAR 500 BOEING (DOUGLAS) MD11

0.26 1982 0.11

1987

0.37

0.80

1992

1997

2002

2007 0.40

0.63 0.11 0.06 0.14

0.09 0.08

0.69

0.05

0.39 0.18 0.09

0.21 0.10

0.15

Fragmentation of North Atlantic and Transpacific Air Transport Markets

211

APPENDIX 3: NORTH AMERICA2ASIA A. City-pair types Absolute values Hub-to-hub Hub-to-spoke Point-to-point

Percentages Hub-to-hub Hub-to-spoke Point-to-point

B. Capacity (million ASKs) Absolute values Hub-to-hub Hub-to-spoke Point-to-point

Percentages Hub-to-hub Hub-to-spoke Point-to-point

1982 8 20 6 34

1987 19 19 11 49

1992 25 32 23 80

1997 40 30 21 91

2002 48 27 17 92

2007 59 29 20 108

1982 0.24 0.59 0.18 1

1987 0.39 0.39 0.22 1

1992 0.31 0.40 0.29 1

1997 0.44 0.33 0.23 1

2002 0.52 0.29 0.18 1

2007 0.55 0.27 0.19 1

1982 1.49 2.88 0.56 4.93

1987 3.32 3.81 0.69 7.82

1992 5.98 4.87 1.89 12.74

1997 9 5.47 1.63 16.1

2002 9.03 4.33 0.86 14.22

2007 11.16 4.38 0.74 16.28

1982 0.30 0.58 0.11 1

1987 0.42 0.49 0.09 1

1992 0.47 0.38 0.15 1

1997 0.56 0.34 0.10 1

2002 0.64 0.30 0.06 1

2007 0.69 0.27 0.05 1

1992

2007 0.05

0.07 0.39

1997 2002 0.06 0.05 0.19 0.06

0.43

0.66

0.51 0.22 0.06

C. Aircraft type (share W5%) Hub-to-hub 1982 1987 AIRBUS INDUSTRIE A340-300 BOEING (DOUGLAS) MD11 BOEING 747 0.79 0.65 BOEING 747 SP 0.20 0.17 BOEING 747-300/747-100/200 0.08 SUD BOEING 747-400 BOEING 777 BOEING 777-200

0.63 0.13

212

ANDREAS KNORR ET AL.

APPENDIX 3: (Continued) Hub-to-spoke AIRBUS INDUSTRIE A330-200 BOEING (DOUGLAS) DC10 BOEING (DOUGLAS) MD11 BOEING 737-800 BOEING 747 BOEING 747-400 BOEING 767 BOEING 767-300 BOEING 777 BOEING 777-200

1982 1987 1992

1997 2002

0.10

0.20

0.08

0.05

0.07

0.59 0.05

0.48 0.14 0.08

Point-to-point BOEING (DOUGLAS) DC10 BOEING (DOUGLAS) MD11 BOEING 727 BOEING 727-200 BOEING 737-800 BOEING 747 BOEING 747 SP BOEING 757 BOEING 767-400

1982 1987 1992 0.27

0.79

0.78 0.19

0.12

0.69

2007 0.06

0.05

0.07 0.38 0.15 0.05 0.11 0.05

0.05 0.12 0.23 0.09 0.08 0.11 0.10

1997 2002 0.44 0.26

2007

0.05

0.07

0.06 0.09

0.16

0.13

0.82

0.45

0.28

0.35 0.28

0.44 0.20 0.20 0.09

CHAPTER 10 ARE AIRPORTS TWO-SIDED PLATFORMS?: A METHODOLOGICAL APPROACH Marc Ivaldi, Senay Sokullu and Tuba Toru INTRODUCTION With the liberalization of air transport and the enlargement of air traffic, airports face insistent requests from airlines to perform and improve both service quality and cost efficiency. As a result, airport ownership, governance, and regulations are debated and sometimes have already been changed. Airport pricing under different governance structure is a central issue in this context. For long this question has been addressed in a framework where airports are considered as aviation service provider for the needs of airlines. However, almost everywhere around the world, airports provide both aeronautical and non-aeronautical services. Although aviation services are the main mission of airports, the revenues are coming from both sides: Airlines and passengers. As of 2009, the share of commercial revenues in airports’ total operating revenues has increased to 60% according to the Airport Financial Report of the Federal Aviation Administration. Under this environment, airports are not only setting the price level for their aviation services, but they decide on a price structure in which they apparently cross-subsidize aeronautical operations by non-aeronautical Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 213–232 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003012

213

214

MARC IVALDI ET AL.

revenues. Having two demand groups (i.e., airlines and passengers) who value each other’s existence make us consider airports as platforms that connect different types of users. In this chapter, we test a two-sided market theory in which airports are considered as platforms where airlines and passengers join to interact. In other words, airports internalize the network externalities arising from the two demands: Passengers are better off if there are more airlines and airlines are better off if there are more passengers. There are two main features of two-sided markets. First there exist externalities between the two end users of the platform. Strictly speaking, decision of one side to enter the market or not depends on the decision of the other side. So, the platforms have to ‘‘get both sides on the board’’. The most studied platform examples in the literature are: Credit cards, magazines, academic journals, or shopping malls. In the case of credit cards, a consumer wants to hold a card which is most widely accepted by retailers and a retailer would like to accept a card which is most widely used by consumers. In the magazine industry, firms would like to give adds to a magazine which has a large number of readers. On the other side of the market, readers may get either utility or disutility from the advertisements. The second noticeable aspect of two-sided markets is that the platform should be able to internalize these existing network externalities while deciding on its pricing scheme. For instance, platforms may discount prices on one side of the market to attract more agents in this side which would in return allow it to charge higher prices to the other side. As found by Kaiser and Wright (2006), readers are subsidized by the profit from advertising side in magazine industry, that is, the cover prices are discounted for readers to attract more readers, thereby to attract more advertisers. Following this literature, airports are candidates to be considered as two-sided platforms, that is to say, as markets with externalities in which they can cross-subsidize the two sides through the pricing structure. Obviously the end users are the airlines and the passengers. The formers’ demand depends on the aeronautical fees and the number of passengers using that airport. The latters’ demand for airports depends on the number of airlines serving at that airport, airline services and airport passenger service related features such as accessibility of airport, parking, or shopping (see Starkie, 2008). An airline would choose to operate at an airport which is more popular among passengers and passengers would enjoy an airport where they can access more airlines and more destinations, as well as a wide range of shops and restaurants, and convenient parking and transportation facilities. Airports earn revenues from both sides and determine the prices of airport services used by both sides. Although the airport charges the airlines explicitly with their agreements and negotiations as in Starkie (2008), the case is different for the passengers.

Are Airports Two-Sided Platforms?

215

Mainly, the airport has two sources of pricing for the passengers. One is directly taken as airport taxes at the stage of ticket sale (through airlines). The other can be deduced via the non-aviation facilities that the airport serves such as parking, restaurants, or stores. Considering the previous literature on airports, papers by Basso (2007), Brueckner (2002), or Starkie (2008) either make a partial equilibrium model where the airline market is not modeled formally or assume a vertical relation between the airlines and airports taking the passengers as final consumers. In addition to this, Gillen (2009) points out that in the last decades, airports have gone through a transition both because of privatization of the industry and increasing importance of commercial revenues. Given this transition and the structure that we have defined it is indispensable to look at airports as two-sided platforms. In this chapter, we provide a methodology to test whether airports are two-sided platforms where airlines and passengers are the two end users. After developing the structural model, we look for an empirical evidence for two-sidedness by using data on U.S. airports. We begin with a monopoly platform and derive the demand equations of passengers and pricing equation of airlines which are estimated simultaneously. To check the structure of the market, we examine the significance of the externality effect parameters. That is to say, we check if the number of flights is significant in passengers’ demand for an airport. While payment cards, shopping malls, academic journals, and magazines have all received a considerable attention as two-sided platforms, the airport industry has not yet been analyzed through this approach. The outcome of the test we perform has important policy applications. Within a framework considered by aforementioned literature that neglects the two-sidedness of airports, the public policy on airport could be misled a suboptimal if two-sidedness is relevant for airport because, as explained by Rochet and Tirole (2003), this feature has striking implications on pricing policy. For instance, prices can be below marginal costs in two-sided markets. For that reason, the methodology proposed in this chapter should contribute to the guidance for the public policy of airport pricing. Our test is a first step in an extensive research agenda. First, our model can be extended to the case of competition between airports, in which we consider the case of competitive bottlenecks, where the airlines do multihoming while the passengers do single-homing. Moreover, the competition between two asymmetric airports can be analyzed under this setting. Airports can also be examined for the optimal platform design like in Hagiu and Jullien (2007), which in turn can increase its profits by pricing optimally the stores. Besides all these, the structural parameters obtained from empirical examination can be used for policy issues.

216

MARC IVALDI ET AL.

The chapter is structured as followed. In the first section, we present the related literature on airports and two-sided markets. The second section explains the theoretical model in which passenger demand, airline demand, and airport pricing scheme are introduced. The third section describes the data and provides some descriptive analysis while the fourth section describes the empirical model and estimation procedure. The fifth section contains the empirical results. Finally, we conclude in the sixth section.

RELATED LITERATURE Although many theoretical and empirical articles deal with the economics of airports, none considers them as two-sided markets. Likewise, in spite of growing literature on two-sided markets, airport industry has never been considered under this approach. Our chapter is indeed related to the literature on the airport industry and the literature on two-sided markets. Economic studies on airport industry generally focus on pricing, capacity, congestion, and regulation issues. Previous studies that look at the question of airport pricing from a theoretical point of view such as Basso (2007) and Basso and Zhang (2008) or from an empirical angle like Gagnepain and Marin Uribe (2005, 2006), consider that the airport–airline–passenger relationship is vertically integrated, taking passengers as final consumers. In other words, demand for airport services is a derived demand which comes from the necessity of the product of airlines (air transport demand) so that they consider airlines as intermediaries. In our setting alternatively, airlines and passengers are the two end users which use the platform, namely the airport, to interact between on each other. However, as in these previous empirical papers, we also assume below that airlines maximize profit at each route they serve to account for airline and route characteristics. In addition to this, we allow passenger demand to depend on airport (i.e., origin) characteristics in our estimation which enables us to capture two-sided externalities. There are other papers on airport economics related to ours. Berry (1990) mentions that when passengers are choosing an airline, they consider if the airline has a dominant position at an airport in terms of flight frequency, as well as some other airline characteristics (e.g., frequent flyer programs, travel agent commission overrides). Similarly, in our model below the passenger demand depends not only on the flight frequency of airline itself, but also on the total frequency at the airport. Hess and Polak (2006) and Pels, Nijkamp, and Rietveld (2003) analyze the choice of airport in London and San Francisco Bay Area, respectively, using a nested logit specification

Are Airports Two-Sided Platforms?

217

for airport choice in which some route specific effects are included as explanatory variables although they do not measure the network effects of airlines by their approach. We also use a nested logit model below but we address the interrelations between the airlines and passengers. As airports have had a monopoly position for many years, they were subject to the regulation of aeronautical charges. Especially, two price-cap regulations, namely single till and dual till, are opposed. In single-till approach the price-cap formula for aeronautical charges includes revenues derived from both aeronautical and non-aeronautical activities, while in dualtill approach only the revenues from aeronautical activities are taken into account. The advocates of dual till, claim that regulation should concentrate on activities which are characterized by a natural monopoly, thus revenues from commercial activities should not be included in the formula (see Beesly, 1999). On the other hand, there is a strong complementary between the aeronautical and non-aeronautical activities (see Starkie & Yarrow, 2009). Some recent papers, like Zhang and Zhang (2010), study the airport decision on pricing and capacity both under single-till and dual-till approaches. Currier (2008) looks at a price-cap regulation of airports and proposes a pricecapping scheme which yields Pareto improvements compared to the status quo regardless of single-till or dual-till regulation. Czerny (2006) points out that single-till regulation is more welfare enhancing at non-congested airports compared to dual till. Here we do not investigate the impact of different regulations. However, our approach, as it is more realistic, should help finding the adequate model to measure the effectiveness of regulation. This chapter is also related to the two-sided markets literature which has been grown substantially in the last decade. The theoretical side, Rochet and Tirole (2002), Rochet and Tirole (2003), and Armstrong (2006) are the kernels. Rochet and Tirole (2002) focus on credit cards, and the other two articles consider the platform competition in two-sided markets in a general setting. Rochet and Tirole (2003) point out that a market can be considered as two-sided if there are network externalities between the two sides, and if the platform choose a price structure not only a price level for its service, that is, it decides on a pricing strategy which depends on the externalities between the two sides. Although airports fit to this definition, none of the aforementioned papers refers to airports as an example of two-sided platform. Additionally, Hagiu and Jullien (2007) consider platform design for new economies like Yahoo!, eBay, Amazon, Google, or platforms like shopping malls. As airports are providing non-aeronautical facilities such as shops and restaurants to passengers, they can also be studied in this dimension in the future.

218

MARC IVALDI ET AL.

Concerning empirical studies in two-sided markets, we have examples on media (Argentesi & Filistrucchi, 2007; Kaiser and Wright, 2006), yellow pages (Rysman, 2004), and academic journals (Dubois, Hernandez-Perez, & Ivaldi, 2007). The two-sidedness in these industries are proved empirically. Similarly we aim to provide a methodology to estimate a two-sided market model for airports and present preliminary results.

MODEL This chapter presents the methodology to test that airports can be considered as two-sided platforms. To do so we combine aspects of the two-sided market theory with transportation models. Our model assumes the airport as a monopoly platform in which airlines and passengers join to interact. Thus, the industry is composed of a monopoly airport, J airlines, and I passengers. We define the market as the set of nonstop directional origin–destination (O–D) routes. In this section we present the structural model to be estimated. First, we derive the transport demand equation of passengers then the pricing and frequency equations that define the airlines’ strategies are derived. Finally, we describe the airport’s program.

Passenger Side Each passenger i, i ¼ 1, y, I, in a given origin decide to use airport or ‘‘not to use’’ which represents the outside option, a ¼ 0, including both ‘‘not to travel’’ and ‘‘other modes of transport’’. Then, each passenger chooses a destination d, d ¼ 1, y, D. After choosing the destination, each passenger chooses an airline j, j ¼ 1, y, Jd to travel with and JdrJ because the number of airlines in each destination can be different. Given this choice structure, we adapt a nested logit type demand model (See Appendix for the choice tree.). Note that including first the choice of using airport a or not using it allows us to extend the model to competing platforms (airports) easily. As in that case, the choice would comprise other airport(s). The indirect utility level achieved by passenger i from choice of airline j on destination d from airport a is given by: U iadj ¼ V adj þ iadj

(1)

Are Airports Two-Sided Platforms?

219

where iadj is the consumer specific unobservable effects specified as follows: iadj ¼ nia þ ð1  s1 Þniad þ ð1  s2 Þniadj

8i ¼ 1; . . . ; I

(2)

where s1 and s2 are parameters to be estimated. The error terms nia , niad , and niadj are common to all products in airport a, airport-destination ad, and airport-destination-airline adj, respectively. In this specification, s1 and s2 show the within group correlation of unobserved utility where the former shows the substitutability of destinations from airport a and the later is the substitutability of airlines in the same origin destination subgroup. The higher the s1 and s2 are, the more substitutable the products are. Let Vadj be the mean utility level of using airline j to destination d from airport a which is specified as: V adj ¼ X adj b þ bf f a  ap~adj þ xadj

(3)

where Xadj is a vector of observable airport, destination, and airline characteristics, xadj is an error term, fa is the airport capacity measured as the sum P flight frequencies of all airlines operating at airport a, that is, P of f a ¼ ð d j f adj Þ, p~adj is airline j’s effective price defined as ! g p~adj ¼ padj þ pc þ pffiffiffiffiffiffiffi (4) fadj where Padj is the price of airline j on destination d from airport a, pc is the average price of airport facilities for passengers, and fadj is the frequency of airline j from airport a to destination d. Note that the bs, a, and g are parameters to be estimated. pffiffiffiffiffiffiffi The term g= f adj in Eq. (4) is the consumer’s cost of schedule delay (the difference between the passenger’s preferred departure time and the actual departure time). A passenger’s schedule delay is inversely proportional to the frequency, assuming that desired departure times are uniformly distributed and an airline groups some of its departure times (Richard, 2003). We normalize the mean utility from outside option a0 to 0, that is, V0 ¼ 0. Following Berry (1994), the share of passengers using airline j in a market is given by: lnðsadj Þ ¼ X adj b  ap~adj þ s2 lnðsjja;d Þ þ s1 lnðsdja Þ þ lnðs0 Þ þ xadj

(5)

220

MARC IVALDI ET AL.

where sjja;d designates the share of airline j within the nest d from airport a, sdja shows the share of destination d from airport , and s0 is the probability of choosing the outside option. The market shares are measured as: qadj (6) sj ¼ M sjja;d ¼ P sajd ¼ P

P

qadj j2J ad qadj

(7)

j2J ad qadj

d2Da

P

(8)

j2J ad qadj

where M is the total market size. Additionally, Jad is the total number of airlines operating on route ad and Da is the total number of destinations from airport a. Airline Side Each airline j, j ¼ 1, y, J, sets its fare (padj) and frequency (fadj) which maximizes its profit Padj on each route. The profit maximization problem of airline j on route ad is written as: maxpadj f adj Padj ¼ ðpadj  cqadj Þqadj  pa f adj  F adj

s:t:

Padj  0

(9)

where pa is the aeronautical fees charged by the airport a per flight (departure) and cqadj is the marginal cost per passenger of airline j for route ad. Since we do not observe this marginal cost, we need a specification based on observable cost shifters. Then, we posit a simple specification for cqadj as: cqadj ¼ l0 þ l1 Zadj þ uadj

(10)

where Zadj is the vector of cost shifters that includes airline, destination, and airport specific variables, and uadj is the error term. Then, the optimal levels of price and frequency are given by: padj ¼ cqadj þ

1 afð1Þ=ð1  s2 Þð1  sjjad Þ þ 1=ð1  s1 Þsjjad ð1  sdja Þ þ ðsjjad sdja  sadj Þg (11) f adj

  gqadj 2=3 ¼ 2pa

(12)

Are Airports Two-Sided Platforms?

221

Note that the price of product adj is equal to the marginal cost of product adj plus a markup term. The markup term decreases in increasing substitutability among the products in a given nest. Moreover, Eq. (11) shows that higher market shares lead to higher prices. If an airline has a dominant position at an airport, then it can use its market power to charge higher prices (see Borenstein & Rose, 1994). Eq. (12) shows that the optimal level of frequency depends on the number of passengers, their valuation of waiting time captured by g and aeronautical fee charged by airport. When airlines are choosing their optimal frequency, they take into account not only passenger demand and aeronautical fees but also passengers’ valuation of waiting time.

Airport Pricing We present now the platform’s, namely the airport’s, problem. Assume that the airport is privately owned and thus maximizes its profits. Given the total number of flights fa and the total number of passengers departing from airport a, qa the equilibrium aeronautical fee per departure, pa, and the concession fee per passenger pc are solutions of the following maximization problem. (13) maxpa ;pc P ¼ ðpc  cc Þqa þ ðpa  ca Þf a  F a P P where qa ¼ d j qadj and Fa is the fixed cost of airport a. We also assume that, each airport is run at full capacity. Therefore, the total number of departures from airport a, fa, is exogenous in the short run. Since the airport capacity is measured as the total number of flights at an airport which is itself the sum of frequencies of each airline operating at that airport, airlines are competing in frequency in the short run. Here we do not address the long-run problem faced by the airport authority which plans to increase its capacity by building extra terminals and/or runways in the long run where fa would be treated as an endogenous variable. Under the assumption of exogeneity of the total number of flights at an airport, the first order conditions are given as follows: X X @qadj @f adj @P  fa ¼ 0 ¼ ðpc  cc Þ @f adj @pa @pa j d X X @qadj @P X X qadj þ ðpc  cc Þ ¼0 ¼ @pc @pc j j d d

(14)

(15)

222

MARC IVALDI ET AL.

From Eqs. (14) and (12) the optimal aeronautical fee of airports is obtained as: " #3=5 1 3 f a  (16) pa ¼ P P 2 4 ðpc  cc Þb2=3 ð d j ð@qadj =@f adj Þqadj Þ

From Eq. (15), the optimal concession fee for passengers is given by the following formula: P P P P p cc d j @qadj =@pc  d j qadjc  P P pc ¼ P P (17) d ja d j @qadj =@pc 

As mentioned before, the airport internalizes the two-sided network externalities between passengers and airlines when it is deciding on its price structure. However, as fa is fixed in the short run, the main two-sidedness effect conveyed through the optimal aeronautical fee in Eq. (16) since the airport authority must internalize the effect of aeronautical fees on passengers. It takes into account the fact that pa does not only affect the demand of airlines but also the demand of passengers for the airport. Since the passengers do care about the flight frequency of airlines, this interaction further affects the demand of airlines.1 Note that the resulting effect depends on the price elasticities for passengers and airlines, and the magnitude of externalities. Hence, for the policy point of view, when airports are analyzed as two-sided platforms instead of vertically integrated institutions, where the passengers are final consumers, the discussion on the difference between single-till or dual-till price-cap regulation becomes meaningless since in a two-sided market setting, the airport can clearly do cross-subsidization between the two sides, and it is the single-till price-cap regulation that can capture this cross-subsidization.

DATA This study uses mainly four sources of data. The airline industry data are drawn from the Airline Origin and Destination Survey Ticket (DB1B-Ticket) provided by the U.S. Bureau of Transport Statistics (BTS) and available on its web site. The DB1B-Ticket survey is a 10% sample of airline tickets from large carriers in United States and comprises detailed information on ticket fares, itinerary (origin, destination, and all connecting airports), the ticketing and operating carrier for each segment, and the number of passengers

Are Airports Two-Sided Platforms?

223

traveled on the itinerary at a given fare. To construct our working sample we extracted from the DB1B-Ticket dataset the record corresponding to the third quarter of year 2006 during which, for the first time after 2000, the U.S. airline industry experiences a positive aggregate net profit of $3.04 billions excluding restructuring and bankruptcy costs (see ATA, 2007). The flight frequency data is constructed from airline on-time performance data of BTS which contains the number of nonstop domestic flights by major carriers. To match two datasets, we use only nonstop flights in our estimation which also allows us to get rid off connecting airport effects. It is worthwhile to note that to measure the impact of frequency on passenger demand, we need to use actual frequencies instead of gathering them from DB1B data, which is the 10% sample of airline tickets. In return this leads to discrepancy between the frequency obtained from on-time data and the number of passengers coming from DB1B data. For that reason, although we derive the demand of airlines for airport (i.e., the frequency), we cannot include it in the estimation procedure. The airport data is constructed from the Airport Financial Data and the Airport Data (5010) published by the Federal Aviation Administration (FAA), which give detailed information about airports’ aeronautical and commercial activities as well as the facilities. Moreover, some of the airport characteristics, such as the number of parking and the number of concession contracts, are gathered directly from the airports. Finally, the demographics data is obtained from U.S. Census Bureau. As is the assumption of monopoly airport, we consider the 31 hubs among the top 50 U.S. airports (in terms of number of enplanements). After gathering the concession and parking data, we are left with nine U.S. hub airports that represent 42.1% of 31 hubs (see Table 1). In Table 2, we present the shares of aeronautical and nonaeronautical revenues in airports’ total operating revenue. Table 3 shows the airlines and their frequencies in our data. In the end of matching all four datasets together, our dataset contains 696 observations. We define a product as a directional trip between origin and destination airports. This allows us to capture the origin airport and city characteristics in passenger demand. In total we have 498 origin–destination pairs (see Table 4). The market size is measured by the population size at origin. Our nested logit demand specification necessitates to have a common market size for all the products from the same origin airport. For that reason, we do not relate destination population to market size. After all, since we are working on quarterly data, our market size specification gives reasonable market shares. In DB1B dataset, there are three types of carriers defined: Operating carrier, ticketing carrier, and reporting carrier. For more than 80% of

224

MARC IVALDI ET AL.

Table 1.

Passengers Boarded at the Nine U.S. Airports.

Airport

Code

City

Hartsfield-Jackson Atlanta International Chicago O’Hare International George Bush Intercontinental Minneapolis-St.Paul International/Wold-Chamberlain John F. Kennedy International San Francisco International Salt Lake City International Baltimore/Washington International Thurgood Marshall Dulles International

ATL Atlanta

GA

40.56

37593

463644

ORD Chicago IAH Houston MSP Minneapolis

IL TX MN

34.54 19.61 17.13

42829 42701 44975

448949 284128 208952

JFK SFO SLC BWI

NY CA UT MD

14.97 13.91 10.28 10.02

50084 59440 36210 44658

173344 153800 150628 119487

DC

9.55

53401

151788

New York San Francisco Salt Lake City Baltimore

IAD Washington

State Passenger Capita No. of (Million) ($) Departures

Top 31 Hub airports United States all airports

458.69 691.17

Note: Values are sorted by the number of passengers (2006).

Table 2. Airport

ATL ORD IAH MSP JFK SFO SLC BWI IAD

Aeronautical and Non-Aeronautical Revenues for the Nine U.S. Airports. Aeronautical Revenue (million$)

Share

Non-Aeronautical Revenue (million$)

Share

53.17 340.26 230.73 87.42 553.78 259.01 41.70 69.66 137.45

0.252 0.687 0.738 0.585 0.781 0.647 0.512 0.579 0.703

158.02 155.23 81.74 62.08 155.79 141.18 39.80 50.77 56.86

0.748 0.313 0.262 0.415 0.219 0.353 0.488 0.421 0.292

tickets in the database, the three carriers are identical. Hence we use the ticketing carrier to identify the airline. As it is mentioned before, we use only nonstop flights in order to match airline industry data with flight frequency data. The number of products at each origin airport is given in Table 3. Our product which is defined as origin-destination-airline (adj) appears several

Are Airports Two-Sided Platforms?

Table 3.

225

List of Airlines.

Airline

Code

Frequency

Percent

American Airlines Alaska Airlines JetBlue Airways Continental Airlines Delta Airlines Atlantic Southeast Airlines Frontier Airlines AirTran Airways Hawaiian Airlines Northwest Airlines Sky West Airlines United Airlines US Airways Southwest Airlines

AA AS B6 CO DL EV F9 FL HA NW OO UA US WN

82 6 10 71 172 1 5 56 1 108 11 127 29 17

11.78 0.86 1.44 10.20 24.71 0.14 0.72 8.05 0.14 15.52 1.58 18.25 4.17 2.44

696

100.00

Total

Table 4. Origin Airport

The Number of Markets. Frequency

Percent

ATL BWI IAD IAH JFK MSP ORD SFO SLC

102 29 29 57 25 86 75 40 55

20.48 5.82 5.82 11.45 5.02 17.27 15.06 8.03 11.04

Total

498

100.00

times in DB1B survey. Therefore, the price of product, padj, is computed as ratio of the sum of fares to the sum of passengers for the same products.

EMPIRICAL MODEL In the second section, we derive passenger demand (Eq. (5)), airline demand (Eq. (12)), and airline pricing (Eq. (11)) equations. The solution of these

226

MARC IVALDI ET AL.

three equations gives us the equilibrium solution in prices and frequency. In other words, the number of passengers, the airlines’ frequencies, and the ticket prices are determined simultaneously. The arguments of the passenger demand are the following: ticket fares, commercial fee, frequency of airline, total frequency at airport, distance, origin airport characteristics, airline characteristics, origin and destination demographics, and dummy variables.2,3 We introduce also a hub dummy, which is equal to 1 if the origin airport is a hub for the airline offering the product and 0 otherwise, and carrier dummies.4 The airline demand links the optimal frequency to the equilibrium number of passengers, up to a stochastic disturbance term which represents measurement errors. The marginal cost defined by Eq. (10), which enters the optimal price equation of each airline contains input price indices, a measure of distance, the network size of the airline, carrier dummies, and an origin– destination hub dummy which is equal to one if either the origin or the destination airport is a hub for the airline.5 The system of simultaneous equations is estimated by means of the Generalized Method of Moments (GMM).6 The econometric problem that we face is the endogeneity of market shares, price, and frequency. The classical solution to this problem is to estimate three equations jointly by using instruments which are orthogonal to the unobservables in all three equations. So, in addition to exogenous variables of the model, we construct BLP type instruments (see Berry, Levinsohn, & Pakes, 1995). These are the number of available products at origin, the number of competitors on the same route, the percentage of direct flights in a route, the average price of airline’s other products in the same market, the average price of airline’s products in the other markets, and the per capita income at the destination. Nonetheless, the predicted frequencies are well below the actual ones. This is because the number of passengers in our data is 10% sample of airline tickets while the frequencies are the total number of nonstop flights. Thus, for some observations, frequency is much larger than the number of passengers. If only the DB1B data is constructed proportionate to the presence of airlines at the airport, one solution would be to scale either the number of passengers or the number of frequencies. However, an airline with many operations at a given airport may have been underrepresented in DB1B. In other words, the DB1B is not a homogenous representation of the whole survey. After all, we estimate simultaneously only the passenger demand and the pricing equations keeping frequency of airlines, (fadj) endogenous in these two equations.

Are Airports Two-Sided Platforms?

227

ESTIMATION RESULTS Our estimation results are presented in Tables 5 and 6. All the estimated parameters have expected signs and most of them are significant. According to nested logit model, products should have higher substitutability in the lower nests. Note that, s2 is, as expected in our model, higher than s1 which validates the nesting structure. Since s2 is estimated to be 0.991, we can conclude that the airlines flying to the same destination from the same origin are highly substitutable. They are both significantly different than zero; the simple logit model is therefore rejected against the nested logit model. Moreover, the marginal utility of income, a, is estimated at 0.0075 and is significant. Any rise in ticket fares, commercial fees or waiting cost leads to a decrease in passenger demand. Moreover, g, which is the coefficient of the schedule delay, is also found to be positive and significant. Hence, the passengers prefer to fly with a carrier with more frequent departures because it means that they could catch a flight as close as possible to their desired departure time.

Table 5.

Estimation Results for the Passenger Demand Equation. Demand Function

Variable CONSTANT PARKCOMM MILESFLOWN DISTRICT fa POPDESTINCOME POPDEST DIRECTNETWORKSIZE (INCOME)2 OHUB B6 DL WN OTHERTRAD OTHER PRICE WAITINGCOST ln(sj|ad) ln(sd|a)

Parameter

Estimate

Standard Error

t-Stat

p-Value

b0 b1 b2 b3 b4 b5 b6 b7 b8 b9 b10 b11 b12 b13 b14 a g s2 s1

3.33381 0.000077 0.000239 0.05804 0.00001 3.25E-13 2.71E-8 0.00394 1.33E-10 0.004963 0.7857 0.411628 0.42061 0.132294 0.16102 0.007564 0.022712 0.991065 0.823382

0.6570 0.000016 0.000127 0.0167 3.373E-6 1.44E-12 6.558E-8 0.00160 1.19E-10 0.1498 0.4645 0.1708 0.3844 0.1128 0.1936 0.00409 0.0101 0.1209 0.1075

5.07 4.85 1.89 3.48 4.09 0.23 0.41 2.46 1.12 0.03 1.69 2.41 1.09 1.17 0.83 1.85 2.25 8.2 7.66

o.0001 o.0001 0.0596 0.0005 o.0001 0.8220 0.6799 0.0140 0.2643 0.9736 0.0912 0.0162 0.2743 0.2412 0.4059 0.0650 0.0251 o.0001 o.0001

228

MARC IVALDI ET AL.

Table 6.

Estimation Results for the Airline’s Cost Function. Cost Function

Variable CONSTANT MILESFLOWN ODHUB LABOURPRICEINDEX FUELPRICEINDEX NETWORKSIZE B6 DL WN OTHERTRAD OTHER

Parameter

Estimate

Standard Error

t-Stat

p-Value

l0 l1 l2 l3 l4 l5 l6 l7 l8 l9 l10

2754.77 0.031634 13.6194 75.05664 1179.762 0.005241 231.54 196.356 234.176 80.9272 234.421

1058.9 0.00610 11.9634 25.9398 452.4 0.0119 58.9271 84.6498 123.4 40.0814 79.5793

2.60 5.18 1.14 2.89 2.61 0.44 3.93 2.32 1.90 2.02 2.95

0.0095 o.0001 0.2553 0.0039 0.0093 0.6604 o.0001 0.0207 0.0582 0.0439 0.0033

Concerning our main parameters of interest, we found empirical evidence of two-sidedness. One aspect is that passengers do care about the airport facilities as number of parking and commercial areas are found to be significant. Another aspect is that both the number of flights of the airline and the total frequency at the airport are significant in demand of passengers. If an airline raises its frequency on a given route, it results in an increase in passenger demand through decreasing waiting cost. In addition to this, an increase in total frequency at airport would reduce passenger demand via congestion effect. Consequently, a change in aeronautical fees would not only lead to a change in airlines’ demand but also passenger demand. Similarly, a change in concession fees would affect passengers, and then airlines through network effects.

CONCLUSION This chapter has developed a methodology to analyze airports under a twosided market setting with the available database. Starting with a monopoly platform, we derive the demand equation of passengers and pricing equation of airlines which are then estimated simultaneously. After explaining the framework, we specify the empirical model which allows us to assess network effects. We find empirical evidence about the two-sidedness of airports through the significant coefficients of flight frequencies and airport characteristics. Moreover, the pricing scheme of airports shows that they

Are Airports Two-Sided Platforms?

229

can cross-subsidize the two sides with respect to their elasticities. The chapter as a whole is a contribution to air transport literature since airports have been considered as two-sided platforms neither theoretically nor empirically. The topic is very fruitful for the future work. First, our model can be easily extended to the case of competition between airports. Moreover, airports can also be examined for the optimal platform design, which in turn can increase its profits by pricing the stores optimally. Besides all these, the correct market definition is crucial for regulators so the structural parameters obtained from empirical examination can be used for policy issues. Finally, the debate of single till versus dual till can be reconsidered under the structure provided in this chapter.

NOTES 1. If one assumes that the airport capacity fa is endogenous, Eq. (17) would also show the impact of pc on airlines. In this case, an increase in concession price would decrease the demand of passengers for the airport and it would also decrease the demand of airlines by two-sided network externalities. At the same time the airport authority could compensate this negative effect by decreasing aeronautical fees to attract more airlines, and thus to attract passengers. 2. ffiffiffiffiffiffiffi The p ffi origin income is used in front of to capture dollar valuation of waiting cost 1= f adj . 3. The number direct destinations of airline from a given airport is named as DIRECTNETWORKSIZE in Table 5. 4. We include American Airlines (baseline dummy), JetBlue Airways, Delta Airlines, Southwest Airlines, a group of other traditional airlines (Northwest Airlines, US Airways, United Airlines, and Continental Airlines) and a dummy for the rest. 5. The number of total destinations from the given airport represented by NETWORKSIZE includes stop and nonstop flights. 6. We also perform the nonlinear three stages least squares estimation, and our results remain similar which indicates that our estimations are robust.

REFERENCES Argentesi, E., & Filistrucchi, L. (2007). Estimating market power in a two-sided market: The case of newspapers. Journal of Applied Econometrics, 22, 1247–1266. Armstrong, M. (2006). Competition in two-sided markets. RAND Journal of Economics, 37, 668–691.

230

MARC IVALDI ET AL.

ATA. (2007). Annual Report. Accessed on August 2010. Retrieved from Air Transport Association website: http://www.airlines.org/Economics/ReviewOutlook/Pages/Annual EconomicreportsoftheUSAirlineIndustry.aspx Basso, L. (2007). Airport deregulation: Effects on pricing and capacity. Technical Report. Sauder School of Business, University of Bristish Columbia, Vancouver, Canada Basso, L. J., & Zhang, A. (2008). On the relationship between airport pricing models. Transportation Research, 42, 725–735. Beesly, M. (1999). Airport regulation. London: Regualting utilities: A new era? Institute of Economic Affairs. Berry, S. (1994). Estimating discrete choice models of product differentiation. RAND Journal of Economics, 25, 242–262. Berry, S., Levinsohn, J., & Pakes, A. (1995). Automobile prices in market equilibrium. Econometrica, 63(4), 841–890. Berry, S. T. (1990). Airport presence as product differentiation. American Economic Review, 80(2), 394–399. Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the U.S. airline industry. Journal of Political Economy, 102(4), 653–683. Brueckner, J. K. (2002). Airport congestion when carriers have market power. American Economic Review, 92(5), 1357–1375. Currier, K. (2008). Price cap regulation of airports: A new approach. Economics Bulletin, 12(8), 1–7. Czerny, A. I. (2006). Price-cap regulation of airports: Single-till versus dual-till. Journal of Regulatory Economics, 30(1), 85–97. Dubois, P., Hernandez-Perez, A., & Ivaldi, M. (2007). The market of academic journals: Evidence from data on French libraries. Journal of the European Economic Association, 5(2–3), 390–399. Gagnepain, P., & Marin Uribe, P. L. (2005, May). Alliances in the air: Some worldwide evidence. CEPR Discussion Papers No. 5063. CEPR, London. Gagnepain, P., & Marin Uribe, P. L. (2006). Regulation and incentives in European aviation. Journal of Law and Economics, 49(1), 229–248. Gillen, D. (2009). The evolution of airport business: Governance, regulation and two-sided platforms. Technical Report. Sauder School of Business, University of British Columbia, Vancouver, Canada. Hagiu, A., & Jullien, B. (2007, August). Designing a two-sided platform: When to increase search costs? IDEI Working Papers No. 473, Institut d’conomie Industrielle (IDEI), Toulouse. Hess, S., & Polak, J. W. (2006). Exploring the potential for cross-nesting structures in airport-choice analysis: A case-study of the Greater London area. Transport Research, 42(2), 63–81. Kaiser, U., & Wright, J. (2006). Price structure in two-sided markets: Evidence from the magazine industry. International Journal of Industrial Organization, 24(1), 1–28. Pels, E., Nijkamp, P., & Rietveld, P. (2003). Access to and competition between airports: A case study for the San Francisco Bay area. Transportation Research Part A: Policy and Practice, 37(1), 71–83. Richard, O. (2003). Flight frequency and mergers in airline markets. International Journal of Industrial Organization, 21(6), 907–922. Rochet, J.-C., & Tirole, J. (2002). Cooperation among competitors: Some economics of payment card associations. RAND Journal of Economics, 33, 549–570.

Are Airports Two-Sided Platforms?

231

Rochet, J.-C., & Tirole, J. (2003). Platform competition in two-sided markets. Journal of the European Economic Association, 1, 990–1029. Rysman, M. (2004). Competition between networks: A study of the market for yellow pages. Review of Economic Studies, 71(2), 483–512. Starkie, D. (2008). The airport industry in a competitive environment: A United Kingdom perspective. Technical Report No. 2008, OECD and International Transport Forum, Paris, France. Starkie, D., & Yarrow, G. (2009). Single till aproach to the price regulation of airports. Technical Report. The UK Civil Aviation Authority, London, UK. Zhang, A., & Zhang, Y. (2010). Airport capacity and congestion pricing with both aeronautical and commercial operations. Transport Research, 44, 404–413.

232

MARC IVALDI ET AL.

APPENDIX: A NESTED CHOICE STRUCTURE

CHAPTER 11 THE SIZE AND GROWTH OF AIRPORTS Dan Mahoney and Wesley W. Wilson INTRODUCTION Over the past 50 years, air travel in the United States has increased from approximately 33 million passengers in 1960 to over 607 million passengers in 2007 (National Transportation Statistics, 2011, Table 1–40). This is over an 18-fold increase in air travel in the past five decades. Over that same time period, the number of airports increased modestly, from 15,161 in 1980 to 19,750 in 2009. The number of those airports serving public commercial traffic is even smaller, and has declined from 730 airports in 1980 to 559 in 2009 (National Transportation Statistics, 2011, Table 1–3). Together, these two facts point to phenomenal growth among airports (measured by the number of passenger trips). The presence of an airport in a region and the presence of multiple airports, of course, not only offer mobility to local residents but also open access to traveling passengers located at other areas. They also provide for tremendous economic development. As stated by the United Nations Institute for Training and Research (UNITAR, 2011), airports have become critical engines for economic growth at nation, regional, and local levels. Enhancing the connectivity, quality of services and competitiveness of airports can create jobs and spur the economic growth of the surrounding region.1

Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 233–273 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003013

233

234

DAN MAHONEY AND WESLEY W. WILSON

This is underscored by Weisbrod, Reed, and Neuwirth (1993) who note not only the economic development of the region but also the role of economic development in rationalizing investments in airports. As business markets become national and international in scale, airports are increasingly being viewed as catalysts for economic development. Their ability to create jobs and attract new business is being used in many locations as a justification for public investments in new airport construction and expansion.

For example, since the early 1970s, the area surrounding the Dallas/Fort Worth Airport has experienced tremendous growth. As noted by Stein (1991), in the early 1970s, the 35 miles between Dallas and Fort Worth was largely undeveloped, but today there are scores of offices, hotels, shopping centers, which according to Stein (1991-2), were generated by the presence of the DFW airport. Brueckner (2003) analyzes the link between airline traffic and employment and finds strong positive relationships. This is accomplished both by attracting new firms to urban areas and by expansion of outputs by existing firms. Throughout the United States, there are a number of rapidly growing airports, but there are also examples of airports that have slow or even negative growth. For example, the rapid growth in population and economic growth in the Redmond/Bend area of central Oregon have fueled investment in the local airport. The current airport terminal was built in 1993 and served 72,000 boardings. In 2008, there were almost 190,000 boardings – an increase of about 163%.2 The airport in Bellingham, Washington, located between Seattle and Vancouver has also experienced tremendous growth in the range of 17% to 24% each year, and in 2007 it was 75%.3 Fueling the growth at this airport are low prices to vacation spots, and investments in the airport that can handle larger airplanes. While there are lots of examples of growth, some airports have declined. Indeed, many airports in the United States have experienced recent declines. For example, Tucson International the number of emplaned passengers is falling due to recent declines in the economy and the cessation of service by JetBlue, ExpressJet and Aeromexico, while others dropped nonstop destinations.4 Tremendous growth in airports such as these has intensified the competition between airports. Forsyth (2006a) discusses cases in which there is an excess demand that fuels secondary airports, and also the case where there major airport has ample capacity. He finds that there are both positive and negative features of competition amongst major and secondary airport. Certainly, if competition tempers excessive prices and/or inefficiency competition is beneficial, however, airports require substantial sunk

The Size and Growth of Airports

235

cost investments, which may necessitate higher prices to recoup those costs. If such is the case, the resulting competition can be detrimental to economic efficiency, especially if nearby airports are only able to offer lower prices because of subsidies. In this chapter, we examine the size and growth of airports. The size of airports is developed in terms of economic principles such as incomes, populations, etc. of the United States and of the local area in which an airport is located as well as the type of service, that is, low-cost, the connectivity of the airport (nonstop destinations served), and the presences of alternative airports and their attributes. On this latter, Barbot (2009) points to the choices passengers make in terms of the costs of travel to each of the airports, as well as the fares that airlines charge. In our model, we include incomes and population, but also the presence of an alternative airport and the attributes of the alternative airport. Our measures of incomes, population, etc. are examined for a variety of catchment areas, and generally, we find that each of these attributes has statistically important and sizable effects on airport size. We also include measures of the airport’s attributes. These include its connectivity (the number of nonstop destinations that are offered from the airport), whether it is a ‘‘hub,’’ and the percentage of low-cost traffic. Each of these have the expected effect are statistically important. That is, each of these positively and statistically impact passenger travel. The ‘‘presence’’ of an alternative airport depends on the distance to the nearest airport and, as distances to the rival increases, passengers at the airport increase, underscoring the effects of alternative airports in attracting passengers. However, the attributes of the alternative airports are also important. We measure these with the connectivity of the alternative and the percentage of low-cost passenger service. As discussed below, these results are robust across a wide spectrum of empirical strategies. While size is a central feature of the chapter, we also calculate and present measures of growth; there are tremendous differences in growth across airports, and we examine how the growth and size of an airport are interrelated. Our general findings are that there is a positive relationship between growth and airport size, but we also find many idiosyncrasies especially among the smaller airports. As noted above, the growth of Bellingham and Dallas/Fort Worth is quite clear, but there are also airports that have experienced negative growth over time. We present these data and then estimate growth rates by airport. These are presented along with measures of airport size. The primary source of data in our examination is the Airline Origin and Destination Survey (DB1B) from the Bureau of Transportation Statistics.

236

DAN MAHONEY AND WESLEY W. WILSON

These data reflect a 10% sample of the ticket prices and attributes for passengers’ trips delineated by originating and terminating airports and airline. The data are quarterly from 1993 to 2009 (68 quarters). In these data, there are a total of 746 airport origins. We exclude airports in the data that are outside of the lower 48 states, which reduces the number of airports in the data to 527. These airports form an unbalanced panel data set of 28,171 airport-quarters. Most of the airports are relatively small; however, a relatively few number of extremely large airports handle the overwhelming majority of all passenger traffic. The largest airports tend to be located in the most populous and higher income counties in the United States. The top 100 airports, ranked by size, account for over 92% of airport origins, which means the remaining 427 airports account for only 8% of origins. The air traffic data are combined with the National Transportation Atlas Database (NTAD), which contains latitudes and longitudes and county-level FIPS codes. The result allows pair-wise distances to be calculated between all combinations of airports in the data, which are used to identify nearest alternative airport. Furthermore, the county codes allow a connection to Census data that provide incomes and populations for local areas. In developing this chapter, we describe some of the industrial organization literature on the determinants of size and growth as well as applications to airline markets. This review is augmented with a recent and evolving literature on the role of competition among airlines at airports and adjacent or neighboring airports in affecting rates and air travel outcomes. In third section, we describe the data. Fourth section describes a model of airport sizes and the role of local markets. In this section, we also provide a number of descriptive tables and figures that illustrate the size and shares of airport traffic. These lead to the presentation of results across a broad spectrum of alternative estimation strategies, which involve a base model with standard OLS regressions. These are augmented by various random and fixed effect models. While the results are relatively similar, a Hausman tests suggests that fixed effect models are preferred. Since some of the key variables do not vary over time, for example, distance to rival, whether an airport is a hub or not, we used a Hausman-Taylor estimator, which allows for estimation of the coefficients on a time constant variables while controlling for fixed effects. In fifth section, we note that most of the variables in the model are time dependent and estimate growth of airports using a range of techniques. The descriptive statistics point to highly variable growth rates, which vary dramatically across airports, especially among the smaller airports. They also

The Size and Growth of Airports

237

point to median growth rates that are larger for the larger airports than for smaller airports. In addition, while most of the airports are growing, there are sizable numbers that are not. Given the wide range of growth rates, a model of airport growth is applied to each airport (for which data were available for the range of quarters) to estimate long-term growth rates by airport. The results are compared across size of airport. The results point to a strong and positive relationship between growth and airport size.

LITERATURE There is a long history of economic research that examines the size and growth of firms in an industry. Yet, only a handful examines the size and growth of airports. There is in contrast a large and growing literature on the prices that consumers face for air travel and the role of airports in establishing those prices. Each is discussed in this section.

Size of Airports At least since the time of Adam Smith, economists have examined the size of firms and the role of the market. Adam Smith (1776) described the size of the market and its role in limiting the degree of specialization. Since then the literature has evolved along three lines – the technological, hierarchical, and vertical integration views of firm size. In the technological view, the size of the firms is completely determined by the technology, and, in particular, the minimum efficient scale of a firm. Larger firms are expected in industries with sizable economies of scale. This, of course, necessarily involves the ‘‘size of the market.’’ Indeed, in a standard model, firm average costs are U-shaped and markets are competitive, long-run equilibrium is established with P ¼ MC ¼ AC, which yields the size of the firm (Q). The number of firms then is simply D ¼ D(min(AC))/Q. And, as the quantity attached to min(AC) increases, that is, the level of scale economies increases, the size of firms in the industry naturally increases. Stigler (1968), Ruffin (1971), Martin (2002), and others point to rivalry and the relationship between measures of market structure. For example, a Cournot model can be used to derive a relationship between market performance and the number of firms in the market. Such a model can also be used to reflect the size distribution of firm measured by the Herfindahl index. While this model is very similar to the standard model,

238

DAN MAHONEY AND WESLEY W. WILSON

it does point to the notion that as the number of firms in an industry increases, performance is enhanced. In the case of airports, it is most assuredly the case that airport size depends on the size of the market and the degree of scale economies. The size of the market depends on incomes, population and costs to access the market (Blackstone, Buck, & Hakim, 2006) as well as the options consumers have to rival markets. In Morrill (1974) and Hsu and Wu (1997), markets are defined in terms of circular catchment areas. Hsu and Wu use a gravitytype model to show generate OD demands based on such a model that includes income, population, socio-demographic variables. In terms of scale economies, there are a few studies that point to economies being exhausted with about three million passengers (Doganis, 1992), while Jeong (2005) point to about 3–5 million passengers. While not substantially large, Oum and Fu (2008) point to the costs of establishing a new airport as the primary source of market power. These costs include the decades to plan, review, and construct a new airport. In the organizational and vertical integration veins of the firm size literature, productivity activities must be coordinated. In Coase (1937) firms exist as a profitable alternative to transacting in the market place. Alchian and Demsetz (1972) point to the limits of firm size from coordination. The basic idea is that as firm sizes grow, it is more difficult to monitor inputs and costs tend to rise. Indeed, firms/organization can be thought of as a profitable alternative to contracting in the market. Another approach is taken by Williamson (1967) who provides an explicit model of hierarchical control. In his model, as firms grow, the firms choose more levels of hierarchy, but that greater levels of hierarchy can result, ultimately, in a loss of control that limits the size of the organization. Rosen (1982) describes a similar model where there are managerial scale economies, but also the loss of control in managing inputs, which again limits the size of firms. While such models are developed in the context of profit-maximizing firms, it is likely the case that airport management can become more complicated and monitoring costs may rise, and limit the size of an airport. In terms of the topic of this chapter, airport size is likely influenced to some degree by each of these theories. However, market size and the presence of competitors are central to the analysis. Indeed, as demonstrated later, the largest airports by these criteria are located in the most populous areas of the United States. However, the ‘‘market’’ per se is not well defined. Specifically, the demand for a given airport’s services is a derived demand. It is the aggregation of demand functions for airline services.5 In most airports, there are a multiplicity of airlines which compete for customers

The Size and Growth of Airports

239

who must simultaneously decide both an airline and an airport to use. The airlines compete largely on price and flight options, while the airport-specific competition is largely geographic. Hence, the model we use to explain airport size rests heavily on the spatial distribution of population around the airport, as well as the presence and location of airports nearby.

Growth of Firms and Airports A second focus of this chapter is on the growth of airports. In general, rising income and population fuel airport growth, but there is also a related and long history on firm growth. Gibrat’s Law, sometimes called the Law of Proportionate Growth, was developed in the seminal article by Gibrat (1931). Gilbrat’s law states that firm size and growth rate are independent from each other, and as a consequence of this, firm size will follow a lognormal distribution. That is, even if growth rates are independent of firm size, concentration in the market evolves from the process. Furthermore, the greater the variance in growth rates, the greater (faster) is the evolution. Since Gibrat first introduced his theory, there has been a litany of studies that test the law. Early studies (e.g., Hymer & Pashigan, 1962; Simon & Bonini, 1958) tended to find growth rates were independent of size while later studies (Audretsch, 1991; Dunne & Hughes, 1994; Evans, 1987; Mansfield, 1962) tend to find that firm growth is negatively correlated with size. There are also a stream of studies that attempt to explain the findings. Mansfield (1962) explains the negative relationship in terms of sample selection, that is, small firms with negative growth exit the industry and are not in the sample. Related, Audretsch (1991) models entrants into an industry at less than minimum efficient scale and must innovate and grow to survive. Ijiri and Simon (1967) introduce an ‘‘impetus for growth.’’ That is, firms innovate, hire a new management team, etc. These factors lead to growth sustained over multiple time periods. Dunne and Hughes (1994) focus on the changes over time in the size distribution of firms. They hold that the birth and death of business relate to the pattern of growth rates. While tendency toward concentrated structures exists, it can be offset by high birth rates of firms and higher growth rates of smaller firms. And as shown by Jovanovich (1982), Dunne, Roberts, and Samuelson (1988) if the variance of growth is lower due to learning of cost/efficiency by larger firms. In Cabral (1995) capacity and technological choices have sunk costs, since small firms are more likely to exit, they invest at a lower rate and thus experience higher expected growth rates.

240

DAN MAHONEY AND WESLEY W. WILSON

In a recent but quite different study, Pagano and Schivardi (2003) find a positive relationship between firm size and growth. They explain the finding in terms of innovative activities. In particular, larger firms have a greater propensity to take advantage of increasing return to research and development. This leads to higher growth rates by larger firms. In a related work, Bottazzi and Secchi (2006) explain the growth rates of firms with a model in which firms’ ability to grow increases as a function of the number of previous opportunities for growth undertaken. While much of the size and growth literature focuses on describing the size distribution of firms in a market and growth rates, there is much that can be gleaned that is useful for this study. There are a number of key points. First, the market definition of industries typically abstract from spatial considerations. That is, most of the studies examine panels of firms that produce a common product without regard to location. While the size of the market, that is, measured by income and population in the area of an airport and the presence of and distance to an alternative airport are central to the size of a given airport. Furthermore, it does not seem likely that an airport located in another part of the country has a sizable influence on a given airport’s size. Second, much of the literature on size focuses on characterizing the distribution of firm sizes in a market, but not the determinants of size. Third, explanations of growth tend to focus on selection issues, strategies that firms have upon entry and/or strategies to grow. These tend to follow profit-maximizing principles, while most airports are publically owned. Anecdotally, several instigators of growth include the establishment of a hub for a major airline, the expansion of the runways, and major renovation or construction of a new terminal. In general, these tend to support a positive relationship between airport size and growth. One such example of an airport experiencing dramatic growth occured in 2001 when JetBlue Airline initiated service to Long Beach Airport and began using it as a hub. This fueled tremendous growth in the airport, as the number of originating passengers tripled (in our data). In contrast, in the past decade, the size of the St. Louis airport has fallen by two-thirds, owing to the TWA bankruptcies and the 9/11 attacks.

Airports and Airline Pricing Because demand for airport services is derived from demand for airline services, it would be impossible to properly analyze the growth of airports without understanding how the airline industry market functions. The United

The Size and Growth of Airports

241

States airline industry was deregulated in 1978, and subsequently, there have been many papers trying to better understand the nature of pricing and competition in the industry. Beginning with the work of Graham, Kaplan, and Sibley (1983), it was identified that airfare is not independent of market concentration. Airlines possess varying degrees of market power, and this market power is one of the driving determinants of pricing. Two of the primary sources of airline market power are the combination of large barriers to entry, along with economies of scale. Both of these effects manifest themselves through the airline practice of hubbing. When airlines establish hub-and-spoke networks, they are able to lower costs through greater capacity utilization; however, establishing a hub at an airport necessitates controlling a large share of the airport’s total traffic. The fact that increased airport presence by a particular airline manifests itself both higher prices and lower costs makes welfare analysis particularly difficult. Identifying market power, and its subsequent effects on airfare is one of the most studied topics in the airline industry. Borenstein (1989) estimates an airline pricing equation including measures of the market concentration among both the route, as well as market concentration at the origin and destination airports. Borenstein finds that a carrier’s share of both route and total airport traffic has significant effects on pricing. While it is expected that airlines with a greater share of route traffic are able to charge higher prices as a result of their market power, it is less apparent why the airline’s overall presence should influence pricing on a particular route. One possible explanation is the prevalence of frequent flyer programs; reward programs for customers who do repeated business with a particular airline. When frequent flyer programs are present, customers may prefer to do business with the airline that offers the most flight options from their local airport, as their airline decision depends on both the current flight as well as expected future flights. Other potential explanations include travel agent commission override bonuses, which pay travel agents for directing a specified level of traffic to a particular airline. There may also be common advertising costs for an airline in a local market. The vertical relationship between airports and airlines is the subject of the paper by Oum and Fu (2008). Airport revenues come from two sources. The first source is charges for aeronautical services. These include take-off and landing fees, terminal rental, aircraft parking, and other such services directly related to the facilitation of flights. The second source of airport revenue comes from non-aeronautical services, such as parking, concessions, office rental, and other commercial uses of airport land. For these services, airports possess significant market power, since price elasticity of demand is

242

DAN MAHONEY AND WESLEY W. WILSON

very low. Several key factors determine airport market power. The first is airport capacity relative to demand. In most of the United States, Europe, and Asia, air traffic demands have been increasing by approximately 5% per year, and airport infrastructure has not kept up with this growth. The second is regional airport competition; when multiple airports serve the same metropolitan area, market power among both airlines will be reduced, so long as these airports do not share common ownership. The share of connecting passengers is also an important determinant of airport market power. While local traffic is relatively inflexible, both passengers and airlines are free to choose between different hub airports. Because of the intertwined relationship between airports and airlines, it may often be beneficial to adopt some level of integration between the two. These relationships may serve to guard against risk, internalize demand externalities, or gain a competitive advantage over other airports and airlines. This integration may take several forms. Airlines may own shares in the airport, or may engage in long-term contracts to guard the airport against risk; in exchange offering the airline favorable rates. Airport–airline relationships often serve to strengthen the position of the airport’s dominant carrier who is best able to negotiate favorable terms with an airport. These long-term contracts can create a barrier to entry for new firms in the market. Ciliberto and Williams (2010) investigate the role of these arrangements in terms of the ‘‘hub premium’’ – the difference in between fares to or from airports where major airlines have hubs relative to comparable trips that do not originate or terminate at a hub airport. Estimating a log-linear pricing specification, Ciliberto and Williams find that the hub premium is present, and increasing in the fare. Unconditionally, they find the hub premium to vary from approximately 10% at the 10th percentile of fare distribution to 20% at the 90th percentile of fare distribution. The apparent hub premium decreases in magnitude when controls for barriers to entry, and airport congestion are added to the model. The hub premium also decreases with the presence of low-cost carrier Southwest Airlines is present, suggesting that increased competition may eat away at the markup. Airport congestion and airport barriers may explain a significant portion of markup power, as represented by the hub premium; however, they only account for approximately 50% of the observed hub premium. The remaining 50% may be attributable to the hubbing market power factors outlined by Borenstein (1989). The relationship between barriers to entry and airport congestion is the subject of a paper by Dresner, Windle, and Yao (2002). They examine several barriers, including slot controls, gate constraints, and gate utilization

The Size and Growth of Airports

243

during peak operating periods. They estimate both a choice model for the entry decision, as well as a standard regression on passengers and yield (defined as average price per passenger-mile). Their findings indicate that all three variables have a statistically significantly positive effect on yield. Only one barrier, gate utilization during peak operating periods, had a significant effect on airline entry into a market. Their results are indicative that although contracts between airports and dominant airlines may correlate with greater market power, unless the airport is capacity-constrained, these contracts will not be able to inhibit new entry. Another concern associated with airport congestion is the costs imposed by an airline’s flight due to congestion. Although weather is the single largest source of delays in the U.S. airline industry, in most cases ‘‘volume’’ delays, caused by traffic exceeding airport capacity, are the second-largest sources of delay. Brueckner (2002) considers the effects of congestion pricing in the airline industry and compares it to the results of the roadpricing literature. Contrary to road-pricing, in the airline industry, firms with market power will internalize some of the congestion costs of their own flights. In the case of the monopolist, the congestion costs will be fully internalized. In the case of an oligopoly, the firms will internalize the portion of the congestion costs imposed on themselves. Pels and Verhoef (2004) derive a similar model of congestion costs with market power and, like Brueckner, find that a naı¨ ve congestion toll will be too large, and may actually be welfare-reducing. Their model also incorporates regulator coordination issues, particularly in the case where origin and destination airports are located in different countries and subject to differing regulatory agencies. Without coordination, the incentive to reduce tolls to the optimal level will be disproportionately reduced, leading to an inefficient outcome. One factor critical to airport congestion is the choice of airplane size. As the number of runways, gates, and departure times are fixed in the shortterm, larger airplanes may be the only way to increase passenger volume. Wei and Hansen (2005) estimate a nested logit model to study the relationship between aircraft size, service frequency, seat availability, airline fares, and market share. They find that airlines can realize higher returns from increasing flight frequency compared to utilizing larger aircraft. Although there may be cost-savings associated with a larger aircraft, holding other factors constant, passengers do not display preference for a particularly sized aircraft. Instead, passengers display a preference for greater choice in departure time. In this case, the airlines will choose to fly airplanes that are smaller than those that would minimize the cost per passenger-mile.

244

DAN MAHONEY AND WESLEY W. WILSON

Related to airport congestion, one critical issue to understand is the optimal market size of a city-pair route at an airport. As airport market size increases, unit operating costs decrease as airlines are able to use larger aircraft filled to greater capacity. A larger airport, however, may face greater delays as it encounter capacity constraints. As the airport increases its market size, the average airport access costs rise, as customers must travel from further away. Hsu and Wu (1997) attempt to model this problem and solve for the optimal airport market size using linear programming techniques. Using hypothetical estimates of various parameters, they find that airports will operate more efficiently in markets with greater population density. Cities with greater per-capita income allow an airport to serve a larger market size, along with a larger market area. Finally, they find that stability among passenger demand will allow airports to operate more efficiently. Complementary to the question of which market patronizes an airport, there is also the question of which airport or airports serve a particular market. Forsyth (2006b) outlines several of the potential issues when a city’s dominant airport faces competition from smaller, fringe airports. Most major cities feature a single dominant airport, located either within, or near the city limits. More recently, there has been growth in secondary airports, which has been associated with the growth of low-cost carriers (LCCs). The secondary airports are often less convenient for consumers, and so they compete largely on price; appealing to the more price-sensitive consumers who are willing to sacrifice some of the benefits of flying with the larger, full service carriers (FSCs). When the LCCs at fringe airports enter the market, it may or may not improve overall efficiency in the market. In the case when a major airport has excess capacity, and the markup above marginal cost is designed to cover the airport’s substantial sunk costs, the airlines may not be able to adjust their pricing to appropriately compete, and an inefficient allocation will be realized. Inefficient allocations may also arise if the secondary airports are receiving subsidies. Conversely, if the secondary airports and the LCCs cost advantage are through greater efficiency, competition in the market will have a positive effect. Morrison (2001) attempts to directly estimate some of the gains offered by LCCs operating out of regional airports. In a study commissioned by Southwest Airlines, he looks at the effects of Southwest’s competition on the U.S. airline industry. When considering the effect of a LCC, such as Southwest, competition may come by the LCC serving the same route in question as the major carriers, or it may come by the LCC serving some combination of the same or adjacent airports. Estimating the effects of

The Size and Growth of Airports

245

Southwest Airlines on fares, for a single year (1998), Morrison finds that competition from Southwest resulted in $12.9 billion in savings, $3.4 of which from Southwest’s own fares, while the remaining savings came from other airline’s lower fares. The cost-savings are greatest when Southwest serves the same route in question as the FSCs, however, even when Southwest doesn’t serve the market in question, but has a presence at either of the endpoints (or their adjacent airports), the threat of entry results in a statistically significant decline in average airfare. Brueckner, Lee, and Singer (2010) offer a comprehensive evaluation of competition and airline pricing. They estimate the model allowing for in market, adjacent competition as identified by Morrison (2001). Unlike Morrison (2001), they consider not only LCC competition from adjacent airports but also legacy carrier competition from adjacent airports. The second contribution of the paper is to distinguish between competition from nonstop flights, and competition from connecting flights. Brueckner, Lee, and Singer find that in-market competition from LCCs contributes to lower fares significantly more than competition from legacy airlines. This pattern extends to adjacent competition from LCCs. They find that in many cases, adjacent airport competition from legacy carriers has no effect on airfare. This result holds for competition among both nonstop flights, as well as connecting flights. Although market power plays a significant role in pricing differences between flights, it may also play a critical role in pricing within flights. Borenstein and Rose (1994) examine price dispersion within a particular route. They find that on average, the expected difference in price paid between two passengers on given route will be 36% of the mean fare for that route. Borenstein and Rose consider two types of elasticities, an industry elasticity of demand, and a cross-elasticity of demand between specific brands. Discrimination based on the first type of customer elasticity (industry elasticity of demand) increases and market concentration increases. Conversely, discrimination based on the second type (crosselasticity of demand) will decrease as market concentration increases. Their findings indicate that greater price dispersion is associated with a greater number of competitors in the market. Greater flight frequency is found to be associated with lower price dispersion, and airport dominance by a particular airline increases price dispersion in its own flights. Price dispersion also tends to be lower on routes typically traveled by tourists. Another study of price dispersion by Bilotkach (2006) attempts to isolate the sources of such price dispersion. Focusing only on economy fares in the London–New York market, Bilotkach attempts to quantify the two leading

246

DAN MAHONEY AND WESLEY W. WILSON

candidates for price dispersion: capacity constraints with uncertain demand, and price discrimination. Bilotkach attempts to differentiate between business travelers and leisure travelers and finds significant cross-airline variation in fares offered to business travelers; however no substantial variation was observed for leisure travelers. He finds that short-term capacity constraints are responsible for much of the observed price dispersion, while the contribution of price discrimination cannot be ruled out. One lingering issue in the airline industry is that of bankruptcy. Within the past 25 years, almost all the largest carriers in the United States have filed for protection under Chapter 11. With such a large share of the market involved in bankruptcy proceedings at one point or another, it is crucial to understand the effects that bankruptcy has on pricing for both the bankrupt airline and its rivals. Borenstein and Rose (1995) offer and overview the issues surrounding bankruptcy in the airline industry. They identify three primary channels through which bankruptcy may alter airline pricing behavior. The first channel is the direct effects of costs and demand. Through bankruptcy proceedings, the airline may be able to renegotiate existing contracts to achieve a better rate and lower costs. There may also be demand effects, as consumers may perceive the bankrupt airline to be of lower quality of higher risk. Both effects serve to lower the prices offered by the airlines. The second channel is one of time discounting; bankrupt airlines may discount the future more heavily as the probability of their future existence is lowered. This may potential raise prices, if low fares are viewed as an investment in the future, or it may lower prices if airlines are engaged in collusive behavior, and increased discounting induces a deviation from cooperative behavior. The final effect is one of strategic interaction. Bankrupt firms may alter the risk firms take on, leading to more or less aggressive pricing behavior. Bankruptcy may also induce predatory pricing from rival firms, who may capitalize on an opportunity to drive a competitor from the market. Borensteain and Rose find that airlines lower fares an average of 5%–6% before filing for bankruptcy, and experience a decline in market share. Competitors’ prices increase in response to the bankruptcy filing. They argue that this evidence is consistent with a demand effect, as consumers shy away from flying with bankrupt firms in favor of their rivals. Barla and Koo (1999) also examine the effects of bankruptcy on airline pricing. They also find that airlines lower fares on and around the filing of Chapter 11 bankruptcy protection. They find fares lower by slightly more than 2% and argue that this is driven largely by the approximately 4% decrease in costs associated with bankruptcy. In contrast to Borenstein and Rose (1995),

The Size and Growth of Airports

247

Barla and Koo find that competitors prices fall by slightly more than 4% in response to a bankrupt rival. They also find a correlation between rival pricing and the ultimate survivability of the bankrupt firm – greater price cuts by rivals adversely affects survival. The prevalence of bankruptcy, along with the response of the rivals may be a significant contributor to the overall profitability struggles of the industry. Berry (2010) attempts to identify the source of the airline bankruptcies in the past decade. These financial difficulties come despite the fact that total passenger-miles has been trending upward, along with the average load factor – both should correspond to profitability. What Berry finds it that passenger demand sensitivity has increased dramatically compared to the late 1990s. Compared to the late 1990s, in 2006, price elasticity of air-travel demand increased by 8%. Passenger preference for direct flights increased, and the connection semi-elasticity had risen by 17%. The increased price sensitivity, combined with increased competition from LCCs explains more than 80% of the reduction in airline profits. Combined with risen costs, due to such factors as rising fuel prices, airline profitability has declined significantly in the past decade. Particularly with the increase in hubbing in recent years, airports and airlines can form close partnerships with each other. The airlines have the ability to bring a tremendous amount of traffic to a particular airport, while the airports, as the gateways to the sky, can help protect market power for the airlines. As airports themselves are constrained by the number of terminals and runways, when a single airline possesses sufficiently large control over that airport’s services, it may be able to deter competition. Subsequentially, if rival airlines wish to compete, they must often seek out an adjacent airport to locate their operations. Often it is the low-cost carriers who are more recent entrants into the market, which must rely on alternative airports. Thus, one of the crucial factors in airport growth will be the presence of LCCs, either at the airport itself or at a nearby rival airport.

DATA SOURCES The primary source of data for this study is the United States Department of transportation’s Airline Origin and Destination Survey (DB1B). This is a 10% sample of airline tickets from reporting carriers and is provided quarterly from 1993 through 2010. The data are voluminous, ranging from a minimum of 2.74 million observations to a maximum of over 4.5 million. Each record contains detailed information on the trip, including the origin

248

DAN MAHONEY AND WESLEY W. WILSON

and destination airports, along with detailed information on ticket prices, passengers, air carrier(s), flight distance(s), routing information, as well as geographic and itinerary information. We combined these data with Airport geographic information from the NTAD 20096 and with income and population from the U.S. Department of Census.7 We also used a measure of U.S. per capita income and a U.S. quarterly price deflator for all goods. Each of these were downloaded from the Federal Reserve Economic Data available the St. Louis Federal Reserve Bank’s website.8 Because of unavailable information for incomes and population for 2010, we excluded the 2010 data, and we removed outliers in the data using commonly used screens in the literature. In particular, we dropped observations with less than $25 or fares greater than $5,000. Observations marked as bulk fare were also removed, along with itineraries reporting a change in ticketing carrier, or flights fewer than 26 miles.9 These data were aggregated by origin, giving the total number of passengers that originate from a given airport.10 These data were matched to the NTAD data. In our OD data, there were 729 unique origin airport codes, and in the NTAD, there were 19,498 different airports. From the former, we include in this analysis only the 530 airports in the 48 contiguous states that appear in the data. In the NTAD data, there are not only public airports, but also ballonports, gliderports, private airports, seaplane bases, and ultralight airports. The OD data have origination and destination airports. These are recorded using three-digit International Air Transport Association Codes (IATA), while the NTAD data use codes assigned by the Federal Aviation Administration (FAA). In most cases, the two codes align well, however, some airports were relabeled to facilitate the merge, and all but three airport codes were successfully merged. The three that were not merged were area codes, that is, WAS, NYC, and CHI, representing all airports in the Washington, DC, New York City, and Chicago areas. Since these markets are not well defined, they were removed from the analysis. After these steps, there are 527 unique airports spanning 68 quarters from 1993 to 2009. These represent all origins in the OD data (except WAS, NYC, and CHI) that appear in the data at least once in the 68 quarters. Of these airports, there are a total of 348 that are present in all 68 quarters, and the top 137 ranking airports each have 68 quarters and they account for over 96% of origins. The quarterly data were then merged with data from the U.S. Census that allowed estimates of county and mean household income by year. Specifically, Census provides these data with state and country FIPS codes and script descriptors. These were converted to five-digit FIPS codes. Using the latitudes and longitudes of airports observed in the airport attribute file, we identified counties with 20, 40, 60, y, 200 mile bands (as the ‘‘crow flies’’ distances) and

The Size and Growth of Airports

249

then calculated the total population and total income for each band by airport. These are then merged into the OD data.

AIRPORT SIZES In this section, we first describe the data in terms of airport size, and then evaluate the size of the airport in terms of market size and competition from other airports. Fig. 1 presents a map of all U.S. airports identified by size. Here the size of the airport is measured by the total number of originating passengers over the entire length of the sample. The size of the circles corresponds to the size of the airport, with passengers measured in millions. Though the largest airports are located near the major population centers there is relatively dense coverage throughout the entirety of the lower 48 states.11 Furthermore, examination of Fig. 1 reveals that among the largest airports, most appear to have sizable alternatives in the vicinity, offering multiple options to passengers. In Fig. 2, we present a histogram of airport size (in passengers) and in Fig. 3 a histogram of market shares. Most airports are relatively small, with

Fig. 1.

Airport Sizes and Location.

DAN MAHONEY AND WESLEY W. WILSON

40 0

20

Percent

60

80

250

0

50

100

150

200

Total Estimated Passengers 1993-2009 (mil)

Total Estimated Passengers 1993–2009.

40 0

20

Percent

60

80

Fig. 2.

0

1

2

3

4

Market Shares

Fig. 3.

Shares (%).

well over 80% serving fewer than passengers 1 million over the time period. As observed in Fig. 1, there are several very large airports that make up a large percentage of total air travel. Indeed, the 95 percentile has less than a 1% share of total originating passengers. Of the 726 airports in the data, only 32 have a total share greater than 1%.

The Size and Growth of Airports

251

Table 1 presents the size, share, and cumulative share of the top 40 airports, while in Fig. 4, we present the cumulative market shares for all the airports. The largest airport in terms of origins is O’Hare International Airport, in Chicago, which has a share of 3.39, followed by McCarran International Airport, in Las Vegas, with a share of 3.27. More generally, the top 20 airports account for 47%; the top 50 account 78%; and the top 100 for over 93%.

Determinant of Market Size Standard models of industrial organization determine the number and size of firms in terms of the size of the market, the structure of costs, and the type of rivalry amongst firms. Specifically, let P(Q) represent a demand function, and for the present illustrative case, consider it linear i.e., ! X D Qi PðQÞ ¼ aðX Þ  b i

The size of the market is reflected by the intercept a(XD), and it depends on a set of demand drivers, for example, population in the market, income. In the standard model, firms are profit-maximizing and there is some form of competition, for example, Cournot. Firms then each maximize profits, given the outputs of their rivals. The equilibrium is established by solving all of the firms’ first-order conditions simultaneously. The result in this simple model has all firms receiving the same price, and cost differences explain differences in output (size) across firms. The solution for two firms with constant costs (ci) is given by the intersection of best-response functions. And, the equilibrium quantities for firms (1 & 2) and price is given mathematically by:   1 1 1 aðX D Þ  2c1 þ c2 ; Q2 ¼ aðX D Þ  2c2 þ c1 ; P ¼ aðX D Þ þ c1 þ c2 Q1 ¼ 3b 3b 3 From this simple model, prices increase in the size of the market and the costs of the two firms. Additionally, if the costs of one firm falls (given the other), output increases. Finally, the low-cost firm, for example, firm 1, has greater output than the high cost firm. These are natural results common to this type of model. Since airports are generally publically owned, the use such a model is tenuous in this case, but the airlines do compete, and the model points directly to important factors that subsequently affect airport size, in

252

DAN MAHONEY AND WESLEY W. WILSON

Table 1.

Leading 40 Airport Origins, Passengers (Mil), Share, and Market Share.

Airport

Airport Name

Passengers

Share

Cumulative

ORD LAS LAX ATL MCO PHX LGA DFW DEN BOS SEA SFO EWR BWI DTW SAN PHL FLL TPA MSP DCA MDW OAK IAH STL SJC PDX MCI SLC SMF MIA SNA MSY JFK IAD CLE HOU RDU BNA IND

CHICAGO O’HARE INTL MC CARRAN INTL LOS ANGELES INTL HARTSFIELD-JACKSON ATLANTA ORLANDO INTL PHOENIX SKY HARBOR INTL LA GUARDIA DALLAS/FORT WORTH INTL DENVER INTL LOGAN INTL SEATTLE-TACOMA INTL SAN FRANCISCO INTL NEWARK LIBERTY INTL BALTIMORE/WASHINGTON INTL DETROIT METROPOLITAN SAN DIEGO INTL PHILADELPHIA INTL FORT LAUDERDALE/HOLLYWOOD TAMPA INTL MINNEAPOLIS-ST PAUL INTL RONALD REAGAN WASHINGTON CHICAGO MIDWAY INTL METROPOLITAN OAKLAND INTL GEORGE BUSH –HOUSTON LAMBERT-ST LOUIS INTL NORMAN Y. MINETA SAN JOSE PORTLAND INTL KANSAS CITY INTL SALT LAKE CITY INTL SACRAMENTO INTL MIAMI INTL JOHN WAYNE AIRPORT LOUIS ARMSTRONG NEW ORLEANS JOHN F KENNEDY INTL WASHINGTON DULLES INTL CLEVELAND-HOPKINS INTL WILLIAM P HOBBY RALEIGH-DURHAM INTL NASHVILLE INTL INDIANAPOLIS INTL

188.97 188.55 181.30 179.46 159.81 151.53 142.71 139.89 130.08 122.40 120.47 119.42 117.15 103.73 100.71 100.29 99.97 98.35 98.16 93.34 91.72 83.04 79.29 77.68 74.24 69.70 69.44 67.77 62.63 58.85 58.33 57.48 56.88 53.67 52.98 52.78 50.25 50.06 49.80 49.18

3.39 3.39 3.26 3.22 2.87 2.72 2.56 2.51 2.34 2.20 2.16 2.14 2.10 1.86 1.81 1.80 1.80 1.77 1.76 1.68 1.65 1.49 1.42 1.40 1.33 1.25 1.25 1.22 1.12 1.06 1.05 1.03 1.02 0.96 0.95 0.95 0.90 0.90 0.89 0.88

3.39 6.78 10.04 13.26 16.13 18.85 21.41 23.93 26.26 28.46 30.62 32.77 34.87 36.74 38.54 40.35 42.14 43.91 45.67 47.35 48.99 50.49 51.91 53.30 54.64 55.89 57.14 58.35 59.48 60.54 61.58 62.62 63.64 64.60 65.55 66.50 67.40 68.30 69.20 70.08

The Size and Growth of Airports

60 40 0

20

Cumulative Shares

80

100

253

0

100

Fig. 4.

200

300 Rank

400

500

Cumulative Shares.

particular, the size of the market and the presence of competitors. When firms engage in spatial competition, both income and population density remain factors of demand; however, transportation costs to and from the airport become relevant. Hence, airport size is determined by traditional demand characteristics such as income and population, but will also depend on the presence and location of competing airports, as well as cost characteristics for the airports based upon the airlines that serve them. These factors are incorporated in our model given by: X dkit X kit þ it Qit ¼ bi þ at þ k

In this expression, there are a number of issues, relating to measurement, variables, and estimation discussed below.

Data For the purposes of estimation, we use the quarterly OD data, aggregated over origin. Hence, the unit of observation in the analysis is the total

254

DAN MAHONEY AND WESLEY W. WILSON

estimated number of originating passengers in time t. The data are quarterly, and the sample provides an unbalanced panel of airports over time. As discussed above, there are 527 airports in the data, but those range in number of observed quarters from 1 to 68. There are a total of 312 present for all 68 quarter. The top 137 airports are each present for 68 quarters, and collectively account for over 96% of the total origins. The dependent variable in this section is the natural log of the number of passengers that originates in time t [ln(PASSit)] from airport i. Following the earlier discussion, passengers are a function of demand, cost, and competitiveness variables. The specifications also include controls for seasonality (quarterly dummies), a trend for omitted influences correlated with time, and a set of dummy variables to capture the effects of 9/11. As direct demand side variables for the size of the market, we include the population of the county in which the airport is located. We also include two measures of income. Because the origin-destination data in the survey cannot distinguish between passengers originating from a particular locale from passengers whose flight is the return leg of a round-trip flight, it was necessary to include a measure of income that was specific to a particular airport’s location, but also a measure of income in the rest of the country. To measure the effect of income for passengers that do not live in the vicinity of the airport, we use real U.S. per capita income.12 To capture the income of originating passengers that live in the area, we use the average household real income in counties in which the specific airport is located. For both income and the population measure, we used data from the county in which the airport is actually located. We considered the effect of neighboring counties, identifying those within 20, 40, y, 200 miles. Comparing results based on alternative mean income and population measure, we found in both cases, the highest R2 resulted from using the only data for the county in which the airport was located. The expected signs of population, real U.S. income per capita, and real income for the county in which the airport is located are each positive. We also included various airport attributes. These include a dummy indicating whether the airport is hub, the percent of passengers that were handled by ‘‘low cost’’ carriers, and the total number of nonstop destinations offered by the airport. The former are well established in the literature, while the number of nonstop destinations is included a measure of the airport’s connectivity. The expected sign of each of these variables is positive. A set of time-related variables are also included. First, a trend is included to reflect variables that are omitted from the model that are correlated with

The Size and Growth of Airports

255

time. Second, the model needs to take account of the effects of 9/11. This event occurred in the last month of quarter 3 of 2001, and since it only affected three weeks of quarter 3, the majority of the effects manifested themselves in quarter 4 of 2001. We include a dummy variable for quarter 3, and a nonlinear adjustment for the remaining effect. Specifically, we defined a hyperbolic variable 1/(trend-35) to account for the effects. This allows for a large initial effect (in magnitude) that dissipates with time. Finally, we include a set of variables to reflect the effects of a rival airport. Various measures for the rival airport were considered; however, in the reported results, we use distance to the nearest airport, the number of nonstop destinations, and the percentage of low-cost passengers. The effect of distance is expected to be positive, while the number of nonstop destinations and the percentage of low-cost passengers are expected to be negative. Table 2 presents descriptive statistics of the variables in the model. The mean number of originating passengers per quarter is about 200,000. There is a significant range from only 10 passengers to nearly 4 million passengers. Of course, it is unlikely that an airport only originates 10 passengers in a three month period. The reason for a number so small is that the number of passengers is the sum of passengers observed in the data. The data are the result of a 10% sample. Hence, small airports are not likely to show up in the data, and if they do, the number of passengers is relatively small. We multiplied the total number of passengers in the sample by 10 to approximate the total distribution of passengers, but, there were total of 670 cases in which there was only one passenger originating in a quarter, which is unlikely to represent

Table 2.

Descriptive Statistics. N

Mean

Std. Dev.

Min

Max

Passengers Real income per capita (US) Real household income (county) Population (county) Origin-Hub Percent of low-cost passengers Number of nonstop destinations

28,171 28,171 28,158 28,170 28,171 28,171 28,171

197,651 29,335 43,641 428,588 0.07 0.11 18.37

476,138 2,422 8,641 1,035,043 0.263 0.226 29.969

10 25,311 24,225 4348 0 0 0

3,752,370 32,933 105,157 9,828,221 1 1 192

Closest airport attributes Distance to closest airport Percent of low-cost passengers Number of nonstop destinations

28,171 28,171 28,171

75.40 0.08 11.41

37.405 0.210 24.724

6 0 0

225 1 192

Variable

256

DAN MAHONEY AND WESLEY W. WILSON

80 70 60 50

qpass

90

100

the true passenger-flow. On the other side, the largest single observation was 3.7 million estimated origins in Las Vegas, for quarter 2 of 2006. Also, of interest, is the trend in total passengers. Fig. 5 presents the total number of passengers in the data over time (in millions). It is clear from Fig. 5 that airport originations have been growing and growing substantially over time. In quarter 1 of 1993, there were slightly more than 50 million origins (estimated), and that number rises to a peak in quarter 2 of 2007 with more than 101 million origins (estimated). But, there are a number of other factors. First, there are obvious seasonality factors to consider. Breaking the results down by quarter reveals that quarter 2 tends to be the highest travel quarter, while quarter 1 is the lowest. Second, beyond seasonality, there is a substantial reduction in passenger origins that took place in the aftermath of 9/11. Indeed, in quarter 2 of 2001 (April–June), traffic had reached over 93 million (estimated). In quarter/3 (July–September), traffic fell to about 83 million (estimated) about 11%. In quarter 4 (October–December), traffic fell to about 75 million-about 10% lower than quarter 3 and about 20% lower than quarter 2. While some of this is due to normal seasonality, it is clear from inspection that 9/11 had a sizable effect. But, it is also clear that over the course of a few years, there was a strong rebound. Over the last few years of the data, there has also been a noticeable decline in traffic. This is likely due to the recession, and so it is captured in the income variables.

0

20

40

60

trend

Fig. 5.

Total Passengers (Mil) over Time.

80

The Size and Growth of Airports

257

Rival airports are accounted for empirically by the distance to the nearest airport and its attributes. On average the closest airport is 75 miles away, but there is a tremendous range (see Fig. 1). The minimum value is about 6 miles. This reflects ‘‘as the crow flies distance’’ from the East 34th Street Airport in NY to the Port Authority-Downtown Manhattan Airport.13 While not sizable, it points to the density of airports on the East Coast. In contrast, the maximum distance is about 225 miles. This is from Del Rio International Airport in Del Rio, Texas to San Angelo RGNL/Mathias Field in San Angelo, Texas. Controlling for all else, the expected sign on distance is positive to reflect a greater area over which to attract passengers. Two other factors are included to reflect the attractiveness of the rival. These are the percentage of low-cost passengers handled by the airport and the number of nonstop destinations of the airport. The former is expected to have a negative effect on airport quantities as the rival airport has a more price attractive service. The latter is also expected to have a negative influence in that the number of nonstop destination served is not only a size of alternative airport effect but also a measure of connectivity. In estimating the model there are many different approaches that could be used. We begin with a simple OLS model and then add a series of alternative estimation approaches, presented in Table 3. The results in columns 1 and 2 present a standard OLS estimation, with column 2 include standard error clustering (by origin). The R2 is quite high suggesting that the model fits the data well. The estimates themselves are identical; however, clustering the errors by origin allows the errors for an airport to be correlated through time. This correction yields a marked and sizable reduction in the standard errors, and differences in the outcomes of related hypothesis tests. Generally, the results are consistent with priors. Estimated coefficients on both measures of income are positive as is the population of the locale area. The attributes of the airport are either not statistically important or are the correct (with priors) sign. Whether or not the airport is hub is not statistically important in either specification. The percent of low-cost passenger travel is the wrong sign (negative) and is statistically significant in column 1, but not in column (2). The number of destinations is positive (as expected) and statistically significant in both specifications. The coefficients on distance to and the attributes of the nearest alternative airport are each of the correct sign, and all statistically significant in the OLS column, while only the amount of low-cost passenger travel is statistically important in the clustered column. Finally, the results point to negative trends14 and seasonality. The variables reflecting 9/11 are, at best, mixed, and generally not as expected.

258

DAN MAHONEY AND WESLEY W. WILSON

Table 3. Coefficient Estimates – OLS and Random Effects. Variables

Real income/capita (US) Real household income (local county) Population (local county) Trend Quarter 2 (dummy) Quarter 3 (dummy) Quarter 4 (dummy) Origin-Hub (dummy) Percent low cost Nonstop destinations Trend ¼ 35 (dummy) 911–1/(time since 9/11) Distance to alternative airport Percent low-cost (alternative airport) Nonstop destinations (alternative airport) Constant Observations R-squared Number of airports

(1)

(2)

(3)

(4)

Ln(Pass)

Ln(Pass)

Ln(Pass)

Ln(Pass)

11.441 (0.585) 0.263

11.441 (0.865) 0.263

8.520 (0.340) 1.119

8.520 (0.787) 1.119

(0.046) 0.402 (0.008) 0.032 (0.002) 0.091 (0.021) 0.146 (0.022) 0.054 (0.021) 0.010 (0.034) 0.205 (0.037) 1.423 (0.007) 0.015 (0.064) 0.084 (0.055) 0.122

(0.232) 0.402 (0.048) 0.032 (0.004) 0.091 (0.014) 0.146 (0.014) 0.054 (0.013) 0.010 (0.134) 0.205 (0.156) 1.423 (0.038) 0.015 (0.035) 0.084 (0.050) 0.122

(0.097) 0.701 (0.040) 0.028 (0.001) 0.138 (0.012) 0.172 (0.012) 0.102 (0.012) 2.607 (0.254) 0.341 (0.037) 0.630 (0.011) 0.006 (0.037) 0.107 (0.032) 0.793

(0.260) 0.701 (0.100) 0.028 (0.003) 0.138 (0.013) 0.172 (0.015) 0.102 (0.012) 2.607 (0.278) 0.341 (0.178) 0.630 (0.050) 0.006 (0.021) 0.107 (0.033) 0.793

(0.015) 0.531

(0.077) 0.531

(0.093) 0.165

(0.179) 0.165

(0.040) 0.027

(0.200) 0.027

(0.035) 0.073

(0.127) 0.073

(0.006) 117.675 (5.930) 28,158 0.815 526

(0.030) 117.675 (8.763) 28,158 0.815 526

(0.011) 102.684 (3.489) 28,158

(0.054) 102.684 (8.078) 28,158

526

526

Notes: Standard errors in parentheses. The income measures, population, distance to other airport, and number of nonstop destinations are each measured in natural logs.po0.01, po0.05, po0.1.

The Size and Growth of Airports

259

While generally supportive of the tenets of airport size, there are some anomalies in the results that may be the result of unobserved influences that are airport specific. There are two general approaches to incorporating unobserved influences-random and fixed effect models. In a random effect model, there is heterogeneity across airport origins in the intercept, but differences are taken to be random and not correlated with the variables in the specification. If the unobserved influences are correlated with the explanatory variables, then omission of the heterogeneity can yield significantly biased coefficient estimates. In columns 3 and 4, the random effect model is presented without and with clustering. A cursory inspection of the estimates suggests marked differences from columns (1) and (2). First, the coefficients on real income and populations remain of the correct sign and statistically significant. However, their magnitudes are sizably different. The trend and quarterly dummy variables also remain of the same sign and remain statistically important. The attributes of the airport (Hub, percentage of low-cost passengers and number of direct destinations) are heavily influenced by the inclusion of random effects with and without clustering. Furthermore, the introduction of individual effects changes the sign and statistical import of the percentage of low-cost passengers. Clearly, it is very important to model individual effects in this model. Reinforcing this are the effects of 9/11. In columns (1) and (2) the results are either zero or positive, while in both columns (3) and (4), the effects of 9/11 in quarter 4 of 2001, are about an 11% reduction. Of course, owing to the specification of 9/11, the effect dissipates with time. Finally, the results on the nearest rival airport are perhaps, not as pronounced. Specifically, the magnitudes of distance from airport along with the percentage of low-cost passengers are each smaller in magnitude than in columns (1) and (2), and when the errors are clustered, it lacks statistical significance. The number of nonstop flights offered by airlines serving the rival airport is, again, negative, but when errors are clustered [column (4)], the results are no longer statistically important. A central result in comparing the specifications is that the coefficients appear to be sizably influenced by the modeling of individual effects in the intercept and the clustering of errors. We present in column (1) of Table 4, an alternative model wherein there are robust standard errors with autoregressive errors. In this case, the data were limited to observations that appeared without gaps in the series. This reduced the sample to 361 airports that, by and large, tended to be the largest airports in the sample. This points, again, to significant differences in some of the parameters and differences in the qualitative results. First, while incomes still have a positive effect, the effects are much smaller than for the other random effect models.

260

DAN MAHONEY AND WESLEY W. WILSON

Table 4. Coefficient Estimates for a Fixed Effect Model with AR and a Random Coefficients Model. Variables

Real income/capita (US) Real household income (local county) Population (local county) Trend Quarter 2 (dummy) Quarter 3 (dummy) Quarter 4 (dummy) Origin-Hub (dummy) Percent low cost Nonstop destinations Trend ¼ 35 (dummy) 911–1/(Time since 9/11) Distance to alternative airport Percent low-cost (alternative airport) Nonstop destinations (alternative airport) Constant Observations Number of airports

(1)

(2)

Ln(Pass)

Ln(Pass)

2.282 (0.333) 0.630 (0.135) 0.970 (0.082) 0.003 (0.002) 0.148 (0.011) 0.162 (0.015) 0.110 (0.009) 2.057 (0.258) 0.059 (0.308) 0.264 (0.043) 0.095 (0.012) 0.125 (0.012) 0.566 (0.156) 0.277 (0.097) 0.067 (0.041) 34.969 (3.747) 22,602 361

9.611 (0.272) 0.804 (0.103) 1.041 (0.081) 0.038 (0.002) 0.147 (0.009) 0.183 (0.009) 0.111 (0.009) 3.131 (0.591) 0.166 (0.124) 0.385 (0.026) 0.021 (0.027) 0.154 (0.031) 0.876 (0.182) 0.209 (0.036) 0.047 (0.027) 114.313 (3.025) 28,158 526

Notes: Standard errors in parentheses. The income measures, population, distance to other airport, and number of nonstop destinations are each measured in natural logs.po0.01, po0.05, po0.1.

The Size and Growth of Airports

261

The trend and quarterly dummies are about the same magnitude, but the trend is no longer statistically significant. The effects of the airport attributes, however, appear to be somewhat different. While the effect of HUB is about the same, the effect of the percentage of low-cost passenger origins is no longer statistically important. And, the number of nonstop destinations is much smaller in magnitude. The effects of 9/11 are consistent with the other RE models, as are the attributes of the rival airport. Comparisons of the OLS versus the RE models point to considerable variation among the coefficients on the right-hand side variables. The last set of results resulted in significant reductions of airports, and so there may be issues to differences across airports used in the various subsets of data. Both cases raise an issue of slope heterogeneity across the airports in the data. To examine this issue, we use a random coefficients model, that is, selected slope coefficients are allowed to be random in addition to the intercept. This allows us to assess the degree to which and which coefficients are affected.15 The results suggest that here are sizable differences in the coefficients, and the likelihood ratio test statistic (the chi-square value) was quite large 30,976. The results are presented in Table 4 column (2). The coefficient estimates are similar to those of the RE models in Table 4, but with much more precise standard errors. In the results, all of the coefficients are statistically significant (at conventional levels) except for the percentage of low-cost passengers at the airport, and the effect of 9/11 in quarter 3 of 2001. Hence, real incomes both in the nation and in the local area have positive and statistically important effects as does population. The trend is decidedly negative throughout most of the specifications as well as this specification. The quarterly dummies are positive and significant through as well as this specification. If an airport is a HUB and offers a lot of nonstop direct traffic, it attracts more passengers. The effect of 9/11 was negative and about 15% in quarter 4 of 2001, dissipating over time. Rival airports and attributes have a strong and negative effect on airport origins (as expected). In particular, passengers fall as alternative airports are closer. This effect is reinforced by whether or not the alternative offers a high degree of low-cost airline service and a large number of nonstop destinations. The results of the last specification point to considerable heterogeneity of the coefficients. Presented in Table 5 are the estimated means of the coefficients, their estimated standard deviations, and associated standard errors. In all cases reported, there are sizable standard deviations. Indeed the estimated standard deviations for each of the random coefficients are statistically significant and point to sizable differences in the estimated

262

DAN MAHONEY AND WESLEY W. WILSON

Table 5.

Random Coefficients Estimates. Mean

Trend Percentage low cost Number nonstop destinations 9/11 (1/time since 9/11) Number of nonstop destinations – alternative airport Constant

SEMean

Std. Dev.

SESD

0.038 0.166 0.385 0.154 0.047

0.002 0.124 0.026 0.031 0.027

0.027 1.609 0.457 0.398 0.452

0.001 0.106 0.021 0.039 0.024

114.313

3.025

2.262

0.083

coefficients. To illustrate, the distribution for the mean value of the effects of 9/11 and the trend are presented in Figs. 6 and 7.16 Each of these figures point to substantial differences in the random effects across observations. While the bulk of the draws point to trends that are decreasing and negative 9/11 effects, it is clear that there is a wide range of effects that differ across airports.17 The other coefficients follow similarly. Generally, the results point to effects these two variables (and the others as well) that are both positive and negative. This is simply the result of the random effects modeled as normal random variables. The important concept is that these results and the more general test of whether or not there are random coefficients point to sizable differences across airports. Two final sets of results are presented. In these results, we consider ‘‘fixed effects’’ rather than ‘‘random effects.’’ Generally, in estimating random effect models, the unobserved effect is taken to be uncorrelated with the right-hand side variables. In a fixed effect model, the effects are correlated with the result that if random effects are used instead of fixed effects, biased coefficient estimates result. In column (1) of Table 6, we present a model with fixed effects. Columns (1) and (2) are the fixed effect analogs to the random effect models presented in columns (3) and (4) of Table 4, that is, fixed effect models without and with error clustering. Again, the clustering only affects the standard errors reported and, therefore, the statistical tests. Inspection of these columns suggests that the estimates from the fixed effect models are quite similar to those of the random effect specifications. The statistical tests do point to significant differences and, therefore, fixed effects should be used.18 A glaring problem with the FE specifications is that and time-invariant attributes cannot be identified. Hausman and Taylor (1981) develop an estimator for such cases. In essence, there are two types of variables – those correlated with the individual effects and those that are not. If correlated with the individual effect, fixed effects apply, but is not correlated, then

The Size and Growth of Airports

0

5

Density

10

15

263

–.15

–.1

–.05

0

.05

.1

.15

Trend Coefficient

Density of Random Coefficients for Trend.

.6 .4 .2 0

Density

.8

1

Fig. 6.

-2

-1

0

1

9/11 Dummy Coefficient

Fig. 7.

Density of Random Coefficients for 9/11.

2

264

DAN MAHONEY AND WESLEY W. WILSON

Table 6.

Fixed Effects without and with Clustering and Hausman– Taylor Coefficient Estimates.

Variables

Trend ¼ 35 (dummy) Real Income/Capita (US) Real Household Income (local) Population (local County) Trend Quarter 2 (dummy) Quarter 3 (dummy) Quarter 4 (dummy) Percent_Lowcost Nonstop Destinations Time since 911-1/(time since 911) Percent Lowcost (alternative airport) Nonstop Destinations (alternative airport) Origin-Hub (dummy) Distance to alternative airport Constant Observations R2 Number of ooo

(1)

(2)

(3)

lpass

lpass

lpass

0.004 (0.036) 8.362 (0.336) 1.296 (0.102) 0.136 (0.080) 0.027 (0.001) 0.140 (0.012) 0.173 (0.012) 0.104 (0.012) 0.346 (0.037) 0.570 (0.011) 0.103 (0.032) 0.158 (0.035) 0.047 (0.011)

0.004 (0.021) 8.362 (0.791) 1.296 (0.293) 0.136 (0.319) 0.027 (0.003) 0.140 (0.014) 0.173 (0.016) 0.104 (0.012) 0.346 (0.183) 0.570 (0.053) 0.103 (0.033) 0.158 (0.129) 0.047 (0.059)

92.157 (3.470) 28,158 0.135 526

92.157 (8.314) 28,158 0.135 526

0.003 (0.036) 8.367 (0.335) 1.295 (0.102) 0.134 (0.080) 0.027 (0.001) 0.140 (0.012) 0.173 (0.012) 0.103 (0.012) 0.346 (0.037) 0.571 (0.011) 0.103 (0.031) 0.158 (0.035) 0.048 (0.011) 5.406 (0.620) 2.753 (0.300) 104.501 (3.688) 28,158 526

Notes: Standard errors in parentheses. The income measures, population, distance to other airports, and the number of nonstop destinations are each measured in natural logs. A,, and indicate statistical significance at the 99, 95, and 90 level.

The Size and Growth of Airports

265

random effects can be used. In the case where there are time invariant variables (in our case, the airport hub indicator and distance to the nearest airport are both time-invariant), then these effects can be identified with the Hausman and Taylor estimator under the maintained hypothesis that the associated variables are not correlated with the unobserved effect. The results of this procedure are presented in column (3) of Table 6. The results are quite similar to results previously discussed. The random effect models (columns 3 and 4 of Table 4) and the fixed effect models (columns 1 and 2 of Table 6) are very similar in terms of the sign and magnitudes of the coefficients with few exceptions. Further, they differ only modestly in terms of statistical inference. The Hausman-Taylor model allows for the inclusion of fixed effects and allows for identification of the coefficients on HUB and distance to alternative airport. These results are quite different from those of Table 4. Indeed, the coefficient on HUB is positive and precisely estimated as is the coefficient on distance.

AIRPORT GROWTH As discussed earlier, growth of firms is often a topic of interest, and there is a vast literature that seeks to explain the growth of firms. These models focus on the relation of growth rates to firm size, and testing for whether or not firm size matters in determining growth rates. In the last section, airport sizes are driven by real incomes, population, and other variables. In this section, we focus on growth and airport size. We examine growth rates by defining growth as the change from the previous time period, that is, [growth ¼ (Passt/Passt11)1]100. There are several extreme values in this calculation, but the 10th, 50th, and 90th percentiles are 26, .2, and 39.1, while the minimum values as 99.8 and the maximum was 24,433. In each case, the growth rates, as expected, are heavily influenced by seasonality. To control for this, growth rates were recalculated using the annual growth [growth ¼ (Passt/Passt41)1]100. Using this measure, there remain extreme values, for example, GilletteCampbell County Airport had a 21,300% growth rate in quarter 4 of 1996. In this quarter, there were 2140 estimated originations whereas in the same quarter in 1995, there were only 10. The result is an astronomical growth rate. The distribution of growth rates (below 50 and above 50) is presented in Fig. 8,19 and in Table 7, different percentiles are presented. As shown, the growth rates are centered slightly above zero, with a median value of 1.01. This means that for the median airport, passengers originations are growing

DAN MAHONEY AND WESLEY W. WILSON

.02 0

.01

Density

.03

.04

266

–50

0

50

Annual Growth Rates

Fig. 8.

Annual Growth Rates.

Table 7. Four-Period Moving Average Growth Rates. Centile 5 10 25 50 75 90 95

Growth Rates 31.47 21.43 8.80 1.01 10.86 28.09 51.17

about 1.01% per quarter. However, there are some airports that have serious declines. Indeed, about 44% of the airport quarters have negative growth, while about 56% have positive growth rates. To examine the relationship between airport growth and airport size, we divided the data into 10 groups based on ranks of airport size, and present the median growth rates for each group over the 68 quarters in Table 8; these are presented graphically in Fig. 9. Median rates were used instead of mean values owing to the extreme values observed in the raw data.

The Size and Growth of Airports

Table 8.

267

Median Growth Rates and Airport Size.

Group

Median Growth 3.09 2.62 1.35 1.24 0.40 0.84 0.45 0.81 3.55 1.41

0 –4

–2

mean_ma4

2

4

1 2 3 4 5 6 7 8 9 10

0

2

4

6

8

10

Group

Fig. 9.

Median Growth Rates and Airport Size.

The results point to positive correlation between median growth rates and airport size. Indeed, the median values decline from 3.09 for the largest group to 1.41 for the smallest airport group. As a final examination of airport growth, we used a simple model to estimate long-term growth rates for each airport. Specifically, we regressed

268

DAN MAHONEY AND WESLEY W. WILSON

20 10 0

Density

30

40

ln(Passengersit) on a time trend, quarterly dummies, and the hyperbolic dummy for 911, that is, 1/(quarters since 911). We note that the right-hand side variables in the earlier regressions are generally changing with time and some may be influenced by the quarterly and 9/11 dummy variables. Hence, the trend estimated incorporates the effects of these variables. In performing the regressions, we used only the 312 airports for which data were available all 68 quarters and performed the estimation by airport. We use a correction for serial correlation in all cases. The result gives a unique set of coefficient estimates for each airport, and, in particular on trend for each of the 312 airports. A histogram of the coefficients on trend is presented in Fig. 10. There are 217 with estimated positive effects, and 95 with negative growth rates. The average value is a .05% growth rate, pointing to little or no growth. However there is a range of growth from .051 to .074, pointing to a minimum growth of 5.1% to a maximum of 7.4%. The results point to dramatic differences in the long-term estimated growth rates. In each case, there is not just a trend coefficient, but also a standard error. A test of significance in each case, results in 198 (64%) estimated trend coefficients that are statistically different from zero. In examining the effects of size on growth, we simply regressed the airport specific trends (corrected for heteroskedasticity) on the size rank of the

–.05

0

.05 Trend

Fig. 10.

Trend Estimates.

.1

The Size and Growth of Airports

–.05

0

.05

.1

269

0

100

200

300

ranks Regression

Fig. 11.

Estimated Growth Rates

Firm Growth and Rank.

airport. The results point to a negative relationship between the estimated trend coefficient and the size rank of the airport and is given by: bti ¼ :0086 :0000237 ranki ð7:44Þ

ð2:35Þ

R2 ¼ :25

While the coefficient estimates are small in magnitude, this is due to the measurement of the dependent variable (it ranges in value from .051 to .074). The overall fit is strong and points to a positive relationship between airport size and airport growth. As is evident in Fig. 11, the dispersion of the predicted values from the individual regressions (the ‘‘dots’’) from the estimated relationship with airport rank grows with rank.

SUMMARY AND CONCLUSIONS This chapter presents an analysis of airport size and growth. Following a long literature from industrial economics and a limited amount of literature directed at airports, airport size is developed in terms of market principles such as the size of the market, cost variables and rivalry. The size of the market is captured in incomes and population. The costs of traveling and the attractiveness of an airport are captured in variables such as whether it is a

270

DAN MAHONEY AND WESLEY W. WILSON

hub, its connectivity, and the percentage of low-cost airline traffic emanating from the airport. Rivalry is captured by the distance to the nearest alternative airport, its connectivity, and its percentage of low-cost airline traffic. The model was estimated with a myriad of different estimation techniques, and virtually all results point to the import of each of the variables, but also to the heterogeneity of airports in the form of both intercept and slope. Naturally, larger airports tend to be located in highly populated areas with large incomes. If the airport is a hub, it tends to be of a larger size. If the airport is highly connected, that is, the number of destinations served nonstop, it tends to be of a larger size. Finally, low-cost service tends to fuel larger airport sizes. However, alternative airports play an important role. As distance to alternative airports falls (costs of access are lower), airport size also falls. Furthermore, if the alternative airport is highly connected and/or has a high level of low-cost airline service, airport sizes tend to fall. Airport growth was another factor considered. Different airports have different growth rates, which can be positive or negative, but tend to be positively correlated with airport size, that is, the higher growth rates tend to be associated with the larger airports. This may be the result of better access to capital, for example, the addition or renovation of a terminal, the addition of a runway, require enormous capital and sunk costs which may be better borne by the larger airports. However, there is also tremendous volatility in growth rates especially among smaller airports. This may be due to a variety of factors. One such factor might be less diversified traffic patterns. Understanding these sources of growth differences is an important area of future research.

NOTES 1. www.unitar.org/event/cifal-atlanta-management-economic-development-carib bean. 2. The Bulletin, April 2, 2011. Retrieved from http://www.bendbulletin.com/ Local/State. 3. King News. Retrieved from http://www.king5/news/local. Accessed on April 28, 2010. 4. http://www.TownNews.com, Inside Tucson Business, AZ (http://www.inside tucsonbusiness.com/news/inside_buiness_travel). 5. Morrill (1974) and Hsu and Wu (1997) model the catchment area with circular market areas, with the airport in the center of the circle. 6. http://www.bts.gov/publications/national_transportation_atlas_database. 7. The income and population statistics are available annually from Census at http://www.census.gov//did/www/saipe/county.html. The data are available for 1993, 1995, and 1997–2009 at the county level. Population is inferred from the

The Size and Growth of Airports

271

population below the poverty line and the percentage of population below the poverty line. The missing data for population and income for the years 1994 and 1996 were extrapolated by regressing the log of population and income on a trend for each county in the sample (a total of 3,141 separate regressions). 8. The St. Louis website is http://research.stlouisfed.org/fred2/. The specific series were A229RC0 for per-capita income and USACPIALLQINMEI (2005 base year) for the CPI. 9. These filters were designed to catch data errors, and other outliers in the data, and such exclusions are consistent with other works in the literature. While, perhaps, not central to our analysis, these screens are common in the literature on pricing, and we include the screens for consistency. 10. The data do not allow us to identify whether the passenger lives in the area or not or whether the origination is a return flight to a ‘‘home’’ destination. It is used simply as a proxy for the level of traffic that originates in an area. 11. The data also include points in Alaska and Hawaii, but these were omitted from the map. 12. The data were taken from Federal Reserve Economic Data. The specific series is A229RC0. The website is http://research.stlouisfed.org/fred2/ 13. This is a heliport that shows up in the OD data. 14. This is not to say that the market is not growing. Indeed, there are sizable increases in population and incomes over time. This result simply suggests that there are omitted influences that have negative impacted the number of origins over time. 15. The data did not permit all the coefficients to be random. By inspection, we limited the number of random coefficients to the trend, percentage of low-cost passengers, and the number of nonstop destinations for both the airport and the closest airport. Once estimated, other coefficients were introduced, but generally, the standard deviations were quite small. 16. To develop these figures, we used the estimated coefficient and standard deviations reported in Table 5, and then drew 100,000 observations from a normal distribution. 17. While the results point to 9/11 effects that are not negative, in some airports, the results are very marginal, while in other airports the results are very negative and strong. This is p. 18. The Hausman test is framed for differences in the estimates and is (bFEbRE)T[Var(bFE)var(bRE)]1(bFEbRE). While by inspection the beta’s are quite similar, it may be that the covariances are also quite similar with the result that the chi-square statistic is large. 19. There were 23 of 21,216 observations below 50 and 1513 observations above 50. This represents .11% of the sample on the low side and 7% on the high side that are excluded from the Fig. 8.

REFERENCES Alchian, A. A., & Demsetz, H. (1972). Production, information costs, and economic organization. The American Economic Review, 62(5), 777–795. Audretsch, D. B. (1991). New-firm survival and the technological regime. The Review of Economics and Statistics, 73(3), 441–450.

272

DAN MAHONEY AND WESLEY W. WILSON

Barbot, C. (2009). Airport and airlines competition: Incentives for vertical collusion. Transportation Research Part B: Methodological, 43(1), 952–965. Barla, P., & Koo, B. (1999). Bankruptcy protection and pricing strategies in the US airline industry. Transportation Research Part E, Logistics and Transportation Review, 35(2), 101–120. Berry, S. T. (2010). Tracing the woes: An empirical analysis of the airline industry. American Economic Journal Microeconomics, 2(3), 1–43. Bilotkach, V. (2006). Understanding price dispersion in the airline industry: Capacity constraints and consumer heterogeneity. In D. Lee (Ed.), Advances in airline economics (Vol. 1, pp. 329–345). Elsevier. Blackstone, E., Buck, A., & Hakim, S. (2006). Determinants of airport choice in a multi-airport region. Atlantic Economic Journal, 34(3), 313–326. Borenstein, S. (1989). Hubs and high fares: Dominance and market power in the U.S. airline industry. The Rand Journal of Economics, 20(3), 344–365. Borenstein, S., & Rose, N. (1995). Bankruptcy and pricing behavior in U.S. airline markets. The American Economic Review, 85(2), 397–402. Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the U.S. airline industry. Journal of Political Economy Chicago, 102(4), 653–683. Bottazzi, G., & Secchi, A. (2006). Explaining the distribution of firm growth rates. RAND Journal of Economics, 37(2), 235–256. Brueckner, J., Lee, D., & Singer, E. (2010). Airline competition and domestic U.S. airfares: A comprehensive reappraisal. Working Paper. Brueckner, J. K. (2002). Airport congestion when carriers have market power. The American Economic Review, 92(5), 1357–1375. Brueckner, J. K. (2003). Airline traffic and urban economic development. Urban Studies, 40(8), 1455–1469. Cabral, L. (1995). Sunk costs, firm size and firm growth. The Journal of Industrial Economics, 43(2), 161–172. Ciliberto, F., & Williams, J. (2010). Limited access to airport facilities and market power in the airline industry. Journal of Law and Economics, 53(3), 467–495. Coase, R. (1937). The nature of the firm. Econometrica, 4(16), 386–405. Doganis, R. (1992). The airport business. London: Routledge. Dresner, M., Windle, R., & Yao, Y. (2002). Airport barriers to entry in the US. Journal of Transport Economics and Policy, 36(3), 389–405. Dunne, P., & Hughes, A. (1994). Age, size, growth and survival: UK companies in the 1980s. The Journal of Industrial Economics, 42(2), 115–140. Dunne, T., Roberts, M. J., & Samuelson, L. (1988). Patterns of firm entry and exit in U.S. manufacturing industries. The RAND Journal of Economics, 19(4), 495–515. Evans, D. S. (1987). Tests of alternative theories of firm growth. The Journal of Political Economy, 95(4), 657–674. Forsyth, P. (2006a). Airport competition an: Regulatory issues and policy implications. In D. Lee (Ed.), Advances in airline economics: Competition policy and antitrust (pp. 347–368). Amsterdam: Elsevier. Forsyth, P. (2006b). Competition between major and secondary airports-implications for pricing, regulation and welfare. Presented at the 2006 ATRS World Conference, Air Transport Research Society, Nagoya, Japan. Gibrat, R. (1931). Le ine´galite´se´conomiques. Paris, France: Librairie du RecueilSirey.

The Size and Growth of Airports

273

Graham, D. R., Kaplan, D. P., & Sibley, D. S. (1983). Efficiency and competition in the airline industry. The Bell Journal of Economics, 14(1), 118–138. Hausman, J. A., & Taylor, W. E. (1981). Panel data and unobservable individual effects. Econometrica, 49(6), 1377–1398. Hsu, C., & Wu, Y. (1997). The market size of a city-pair route at an airport. The Annals of Regional Science, 31(4), 391–409. Hymer, S., & Pashigan, P. (1962). Firm size and rate of growth. The Journal of Political Economy, 70(6), 556–569. Ijiri, Y., & Simon, H. A. (1967). A model of business firm growth. Econometrica, 35(2), 348–355. Jeong, J. H. (2005). An investigation of operating costs of airports: Focus on the effects of output scale. Master’s thesis, University of British Columbia, Vancouver, BC. Jovanovich, B. (1982). Selection and the evolution of industry. Econometrica, 50(3), 649–670. Mansfield, E. (1962). Entry, Gibrat’s law, innovation, and the growth of firms. The American Economic Review, 52(5), 1023–1051. Martin, S. (2002). Advanced industrial economics (2nd ed.). Oxford: Blackwell. Morrill, R. L. (1974). The spatial organization of society. North Scituate, MA: Duxbury Press. Morrison, S. A. (2001). Actual, adjacent, and potential competition: Estimating the full effect of southwest airlines. Journal of Transport Economics and Policy, 35(2), 239–256. National Transportation Statistics. (2011). US department of transportation. Retrieved from http://www.bts.gov/publications/national_transportation_statistics/ Oum, T. H., & Fu, X. (2008). Impacts of airports on airline competition: Focus on airport performance and airport-airline vertical relations. Paris: Joint Transport Research Centre of the OECD and the International Transport Forum. Pagano, P., & Schivardi, F. (2003). Firm size distribution and growth. The Scandinavian Journal of Economics, 105(2), 255–274. Pels, E., & Verhoef, E. T. (2004). The economics of airport congestion pricing. The Journal of Urban Economics, 55(2), 257–277. Rosen, S. (1982). Authority, control, and the distribution of earnings. The Bell Journal of Economics, 13(2), 311–313. Ruffin, R. J. (1971). Cournot oligopoly and competitive behavior. The Review of Economic Studies, 38(4), 493–502. Simon, H. A., & Bonini, C. P. (1958). The size distribution of business firms. The American Economic Review, 48(4), 607–617. Smith, A. (1776). The wealth of nations. London: Methuen and Co., Ltd. Stein, P. B. (1991). The price of success: Mitigation and litigation in airport growth. Journal of Air Law and Commerce, 57(2), 413–556. Stigler, G. J. (1968). The organization of industry. Homewood, IL: R. D. Irwin. Wei, W., & Hansen, M. (2005). Impact of aircraft size and seat availability on airlines demand and market share in duopoly markets. Transportation Research: Part E, Logistics and Transportation Review, 41(4), 315–327. Weisbrod, G. E., Reed, J. S., & Neuwirth, R. M. (1993). Airport area economic development model. Paper presented at the PTRC International Transport Conference, Manchester, England. Williamson, O. E. (1967). Hierarchical control and optimum firm size. The Journal of Political Economy, 75(2), 123–138.

CHAPTER 12 WHERE DOES AIRPORT NOISE FALL? EVIDENCE FROM ATLANTA Jeffrey P. Cohen and Cletus C. Coughlin INTRODUCTION Airport noise is an undesirable consequence of arriving and departing flights. Much research effort has focused on how such noise affects the prices of houses located nearby and consistently finds that more noise is associated with lower housing prices.1 On the other hand, few studies have examined the determinants of airport noise. Sobotta, Campbell, and Owens (2007) is a notable example of a study focused on the determinants of airport noise. They regress airport noise, expressed as a qualitative dependent variable, on various independent variables, including the percentage of the neighborhood population that is Hispanic. They find that households in neighborhoods with greater Hispanic population were subjected to higher noise levels than households in other neighborhoods.2 One might wonder, however, whether a closer look might reveal some substantial differences across geographic locations. Such spatial heterogeneity could occur in the impacts of demographic variables, as well as other spatial variables including distance from the airport, on the probability of greater noise exposure.

Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 275–295 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003014

275

276

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

The importance of addressing spatial effects has become clear in recent studies of airport noise (Cohen & Coughlin, 2008). In the present study, we focus on spatial heterogeneity in the context of the determinants of the geographic distribution of airport noise. We postulate that there is substantial geographical variation in the determinants of airport noise, and that ignoring such heterogeneity can produce misleading views of where noise (from a geographic perspective) falls on different racial and ethnic groups. Beyond incorporating spatial heterogeneity, our contribution includes several innovations directly relevant to the analysis by Sobotta et al. (2007).3 First, we order the dependent variable with three categories ranging from the least noisy to the greatest noisy area. The three categories, based on yearly day-night sound levels (DNL) are: (1) buffer zone – houses are located in a less the 65 DNL zone (i.e., less than 65 dB); (2) 65 DNL zone (i.e., 65 up to 70 dB); and (3) 70 DNL zone (i.e., 70 up to 75 dB).4 In addition to estimating a standard ordered probit model, following McMillen and McDonald (2004), we estimate ordered probit locally weighted regressions (OPLWR). This estimation approach allows us to explore the issue of spatial heterogeneity in the context of the determinants of airport noise, which to our knowledge has not been examined previously.5 OPLWR is a more tractable approach than parametric estimation approaches such as a spatial ordered probit model. It also allows for heterogeneity in each individual parameter estimate by obtaining a separate parameter estimate for each data point. One might anticipate that because our dataset is limited to those sales near the airport spatial heterogeneity is likely to be unimportant. Such an expectation is not supported by our results. We find notable differences in parameter estimates for different houses in our sample with the OPLWR estimates. In particular, the sign on the coefficient for each explanatory variable contains some positive and some negative values. Also, the mean of the magnitudes of the coefficients for some of other explanatory variables are larger with the OPLWR model, while for other coefficients the mean is smaller. These differences between the OPLWR and the ordered probit results imply that focusing exclusively on an ordered probit model for the determinants of noise can lead to biased estimates in our context due to ignored heterogeneity among individual houses in our sample. Prior to providing details on our equations and results, we provide an overview of our dataset. Next, we focus on the standard ordered probit

Where Does Airport Noise Fall? Evidence from Atlanta

277

model and the results. This is followed by details on the ordered probit locally weighted regressions. A discussion of our key findings completes the chapter.

DATA We use data on airport noise levels surrounding the Atlanta airport in 2003. The airport noise contours were obtained from the Atlanta Department of Aviation, and are the same noise contours used by Cohen and Coughlin (2008). For 508 houses near the Atlanta airport that were sold in 2003, we purchased housing sales prices and characteristics data from First American Real Estate. These data include house sale price as well as detailed housing characteristics such as the number of bedrooms, bathrooms, fireplaces, stories, and the lot size. Table 1 contains definitions of the variables in our regressions and Table 2 presents the descriptive statistics for the sales prices and characteristics of the data from 2003. Approximately 29 percent of our observations fall in the 65 DNL zone, about 4 percent fall in the 70 DNL zone, and the remainder are in a ‘‘buffer zone’’ extending 0.5 miles outside of the 65 DNL zone. See Fig. 1 for a plot of the locations of the houses that were sold in 2003 on the contour maps. The houses are located in either Fulton County or Clayton County. In terms of cities, the houses are located in Atlanta, College Park, Conley, East Point, Forest Park, and Hapeville. The average house sold for approximately Table 1.

Variable Definitions.

Name

Definition

Noise

Ordered categorical variable with three noise levels for houses in the buffer zone (least noise), 65 decibel day-night sound level noise contour, and 70 decibel day-night sound level noise contour. Distance in miles from house to airport (in natural logs). Age of house (in natural logs). Percentage of houses in the neighborhood in which a house was sold with a black head of household. Percentage of houses in the neighborhood in which a house was sold with a Hispanic head of household. Median household income in the neighborhood in which a house was sold.

DistanceLog AgeLog B1kHH00 HispHH00 MedHHInc00

278

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Table 2.

Summary Statistics – 508 Observations. Count

Percentage

343 146 19 49 147 60 66 136 50

67.5 28.7 3.7 9.6 28.9 11.8 13.0 26.8 9.8

1 story 2 or more stories

425 83

83.7 16.3

2 3 4 5

or less bedrooms bedrooms bedrooms or more bedrooms

138 258 99 13

27.2 50.8 19.5 2.6

1 bathroom 2 bathrooms 3 or more bathrooms

246 151 111

48.4 29.7 21.9

0 or 1 fireplace 2 or more fireplaces

494 14

97.2 2.8

House House House House House House House House House

sales sales sales sales sales sales sales sales sales

in in in in in in in in in

the buffer zone – 2003 contours 65 db zone – 2003 contours 70 db zone – 2003 contours Atlanta College Park Conley East Point Forest Park Hapeville

Price (dollars) Distance (miles) Acres Age (years) B1kHH00 (percent) HispHH00 (percent) MedHHInc (hundreds of dollars)

Mean

Range

128,442 3.29 0.37 39.85 56.96 8.64 319.4

32,378–460,500 1.06–6.06 0.03–3.88 0–100 0–97.5 0–30.1 116.7–606.3

$128,400, contained about 3 bedrooms and 1.78 bathrooms, and was located on a lot of 0.37 acres. Block group data on demographics, including percent black, percent Hispanic, and median income, were obtained from the 2000 U.S. Decennial Census. Because the demographic information was from the year 2000 while the noise levels were based on estimates in 2003, it seems reasonable to postulate that previous demographics may have influenced 2003 noise levels.

Where Does Airport Noise Fall? Evidence from Atlanta

Fig. 1.

279

The Location of Houses in the Sample.

ORDERED PROBIT MODEL The first model we estimate, a standard ordered probit (OP) model, is as follows: Noise ¼ f ðX; uÞ

(1)

where Noise is a categorical variable for a house sold in one of the three noise level groupings described above, ordered from least to most noisy; X represents a set of variables measuring: (1) the age of the house in logs – AgeLog; (2) the distance in logs from the house to the airport – DistanceLog; (3) the percentage of the houses in the neighborhood in which the house was sold with a black head of household – BlkHH00;(4) the percentage of houses in the neighborhood in which the house was sold with a Hispanic head of household – HispHH00; and (5) the median household income in the neighborhood in which the house was sold – MedHHInc00, and u is an error term with a normal distribution with zero mean and constant variance.

280

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

ORDERED PROBIT RESULTS The results produced by estimating Eq. (1) by ordered probit are presented in Tables 3 and 4. The results in Table 3 indicate that all the variables are statistically significant. The results in Table 3 must be transformed before interpreting them as marginal effects (see Greene, 2003). Because there are three categories for the dependent variable, each can be ordered on a line segment under the normal distribution curve, and the width of each subsegment would depend on the frequency of the observations for each noise level. The probability of each value of the dependent variable is the area under the curve between the boundaries of each particular subsegment. The marginal effects of an increase in an exogenous variable on the predicted probabilities of each possible value of the dependent variable can be assessed in the context of a normal distribution that shifts in response to the change in the exogenous variable. This shift leads to a different area under the normal distribution for each of the three possible outcomes.

Table 3.

Estimation Results (1u).

Variable

Ordered Probit

AgeLog

0.178 (4.43) 0.548 (2.89) 0.029 (8.07) 0.034 (3.16) 0.003 (3.92) 1.7 2.14 312.46 133.47 0.00 0.18 508

DistanceLog B1kHH00 HispHH00 MedHHInc m Const Log likelihood LR w2 (5) ProbWw2 Pseudo R2 Observations

Notes: z-statistics are in parentheses. Dependent variable is an ordered, categorical noise variable with three noise levels starting from least noise (lowest level).  Denotes significance at the 5 percent (two-tailed) level.

Where Does Airport Noise Fall? Evidence from Atlanta

Table 4.

281

Partial Derivatives (z-Statistics) – Ordered Probit.

Variable

Buffer Zone

65 DB

70 DB

AgeLog

0.062 (4.40) 0.189 (2.90) 0.010 (8.19) 0.012 (3.18) 0.001 (3.91)

0.056 (4.26) 0.171 (2.87) 0.009 (7.52) 0.010 (3.14) 0.001 (3.83)

0.006 (2.97) 0.018 (2.31) 0.001 (3.46) 0.001 (2.43) 0.00009 (2.75)

DistanceLog B1kHH00 HispHH00 MedHHInc

When there is a positive relationship between the dependent variable and the exogenous variable causing the shift, there will be less area under the normal curve for the lowest outcome (noise less than 65 dB), so this probability will decrease. For the largest outcome (noise greater than 70 dB), the area under the normal curve will increase, so the probability that a house is exposed to noise greater than 70 dB increases. The outcome of an increase in an exogenous variable on the area in the middle range (65 up to 70 dB) is ambiguous, as the probability of being in this noise range may either increase or decrease. After transforming the results in Table 3, an examination of Table 4 reveals that the marginal effects are negative and significant in the buffer zone (noise less than 65 dB) for the black (BlkHH00), Hispanic (HispHH00), and income (MedHHInc) variables. Because of their positive coefficients in Table 3, increases in any of these three exogenous variables (i.e., larger neighborhood percentages of black and Hispanic heads of households and higher neighborhood median income) will shift the entire probability distribution to the right, which decreases the probability of being in the buffer zone.6 Meanwhile, the variables for the age of the house (AgeLog) and distance from the airport (DistanceLog) are negative and statistically significant. Because of their negative coefficients in Table 3, the positive sign for the buffer zone partial derivatives in Table 4 reflects the fact that increases in these explanatory variables shift the buffer zone probability distribution to the left. Thus, higher values of these variables (i.e., older houses and houses farther from the airport) increase the probability of being in the buffer zone. For each explanatory variable, the signs of the marginal effects for the buffer zone and the most noisy (noise greater than 70 dB) part of the probability distribution are opposite each other, and the interpretations for houses in the most noisy zone follow accordingly.

282

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

We also examine the marginal effects for the 65 up to 70 dB noise contour. For percent black and Hispanic households, the signs of their marginal effects imply that for the average house in the 65 up to 70 dB zone, higher percentages of either of these populations in the neighborhood leads to a higher probability that houses in the neighborhood will be exposed to 65 up to 70 dB of noise. A similar finding holds for median household income – for the average house in the 65 up to 70 dB zone, higher household income in the neighborhood leads to a higher probability of exposure to 65 up to 70 dB of noise. On the other hand, the age and distance marginal effects are negative and significant for the 65 up to 70 dB dependent variable. Larger values of either age of a house or distance from the airport lead to a lower probability that a house is exposed to 65 up to 70 dB of noise.

ORDERED PROBIT LOCALLY WEIGHTED REGRESSIONS: LOCALLY WEIGHTED MAXIMUM LIKELIHOOD It is possible that some of our variables affect the probability of a given level of airport noise nonlinearly. In other words, the neighborhood characteristics of different houses may have different impacts on the probability of a given level of noise exposure. A standard ordered probit model does not adequately account for such nonlinearities because the parameter estimates are constrained to be equal across all data observations. Thus, ignoring the spatial heterogeneity in the parameter estimates can lead to inaccuracies in interpretation of the magnitude and direction of the distance and the demographic variables on the probability of greater noise. McMillen and McDonald (2004) propose an estimation approach that allows for heterogeneity, which we call ordered probit locally weighted regressions (OPLWR).7 They specify a ‘‘pseudo log-likelihood function’’ to estimate a separate set of parameters for each observation, and they call this a locally weighted ordinal probit pseudo log-likelihood function. For the case where there are three possible ‘‘regimes’’ in the ordered probit, the pseudo log-likelihood function is: Sj wij ½D0j log Fðb0i X j Þ þ D1j ½log Fðmi  b0i X j Þ  log Fðb0i X j Þ þ D2j log Fðmi þ b0i X j Þ; i; j ¼ 1; 2; . . . ; 508

ð2Þ

where F() is the standard normal cumulative density function; bi is the parameter vector for observation i; D0j , D1j, and D2j are dummy variables

Where Does Airport Noise Fall? Evidence from Atlanta

283

taking the value of 1 if observation j is either 0, 1, or 2, respectively, and 0 otherwise; mi is a parameter for observation i; and wij is the weight that house j has on house i. The weight structure is somewhat different than for typical spatial econometric weighting matrices. One possibility, which we use in our analysis, relies on the ‘‘Gaussian function’’, and is represented as:   d ij (3) wij ¼ f si b where f is the standard normal (Gaussian) density function; dij is distance (as the crow flies) between house i and house j; si is the standard deviation of the distances between house i and all other houses j; and b represents the ‘‘bandwidth’’.8 Many locally weighted regression applications have used the Gaussian function. The determination of the bandwidth tends to be more important than the choice of the weighting function. For example, the results in Thorsnes and McMillen (1998) are essentially invariant to choosing among several different weighting functions. McMillen and McDonald (2004) suggest the ‘‘cross-validation’’ approach for selecting the appropriate bandwidth. This approach consists of estimating the OPLWR model for several different bandwidths (and setting wii ¼ 0), and choosing the bandwidth for which the pseudo-likelihood function is maximized. In the present context, we estimated the pseudo-likelihood model for bandwidths of 0.4, 0.6, 0.8, and 1.0. Cross-validation implied that b ¼ 0.4 was the preferred bandwidth.

ORDERED PROBIT LOCALLY WEIGHTED REGRESSIONS: RESULTS Table 5 contains results for the OPLWR estimations, based on the preferred bandwidth of b ¼ 0.4. Prior to examining the results for specific variables, we summarize some of our findings. While some similarity in terms of the signs (e.g., the matching of the signs using OP with the averages signs using OPLWR) and magnitudes of the estimated coefficients (e.g., the magnitude of the coefficient for distance using OP is roughly equal to the average of the coefficients using OPLWR) exists, most noteworthy is that substantial heterogeneity is found. For each variable, the estimated coefficients for the OPLWR model exhibit positive as well as negative values. For every

284

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Table 5. Variable

Ordered Probit Models for Aviation Noise. Standard Ordered Probita

AgeLog

0.178 (4.43)

DistanceLog

0.548 (2.89)

B1kHH00

0.029 (8.07)

HispHH00

0.034 (3.16)

MedHHInc

0.003 (3.92)

m

1.7

Const

2.14

Log likelihood Observations

312.46 508

Locally Weighted Ordered Probitb 0.422 (0.299) [1.613, 0.233] 4.460 (2.420) [9.600, 0.109] 0.056 (0.039) [0.129, 0.104] 0.017 (0.056) [0.239, 0.122] 0.013 (0.014) [0.003, 0.042] 2.78 (1.00) [ 1.27, 5.37] 0.981 (5.234) [6.802, 27.489] 6855.42 508

a

Parameter estimates with z-statistics in parenthesis. The average of the 508 parameter estimates for the variable is listed on the first of the three lines, the standard deviation in parenthesis is on the middle line, and the range of parameter estimates in brackets is provided on the third line. The log-likelihood value is the sum of the log likelihoods for the 508 regressions. Bandwidth ¼ 0.4.

b

explanatory variable, there are at least some houses for which the estimated coefficients differ substantially between the OP and OPLWR models. Turning to the results for specific variables, the mean estimate from the OPLWR for the household income (MedHHInc00) is roughly four times the magnitude of the coefficient estimate from the OP. While there are some negative values for some houses, the majority of the houses have positive income coefficients. Thus, the qualitative insights associated with this variable are similar across the two estimation procedures. Results associated with the age explanatory variables suggest the additional insights and value provided by OPLWR. The mean OPLWR estimate for the age variable (AgeLog) is more than double the coefficient

Where Does Airport Noise Fall? Evidence from Atlanta

285

estimate of the OP. The range of estimates, which contains mostly negative values, is much larger than the distribution suggested by the OP results. For the variable measuring the percentage of houses in the neighborhood in which a house was sold with a black head of household (BlkHH00), the mean from the OPLWR is roughly twice as large as the coefficient estimate of the OP. The range of the OPLWR results includes some negative values, but the vast majority of the houses have positive values. The Hispanic variable (HispHH00) demonstrates a substantial amount of heterogeneity, with a notable mix of both positive and negative coefficients. The mean OPLWR is about half the magnitude, and the same sign as, the OP coefficient estimate. Thus, the OPLWR approach, compared with the OP estimates, adds explanatory power with respect to the Hispanic variable. For the distance variable (DistanceLog), the mean OPLWR has the same sign but is about eight times the magnitude of the coefficient estimate of the OP. Moreover, the range includes mostly negative but only a handful of positive values. In general, distances closer to the airport imply that houses are subjected to more rather than less noise.

WHERE DOES THE NOISE FALL? A GRAPHICAL VIEW The preceding discussion summarizes our results, but provides virtually no geographic perspective. Now we attempt to increase the insights relating to the effects of our independent variables by adding some geographic meat. Using a picture identifying the noise contours surrounding the Atlanta airport and the locations of houses in our sample, the estimates for each independent variable are presented in two ways. In the top panel of each figure, the locations of positive coefficients (in red) are distinguished from the negative coefficients (in black). In the bottom panel, the coefficients are grouped by quintiles with the first quintile containing the smallest estimates. Let’s start by examining the estimates for the distance variable (DistanceLog). For the model with the b ¼ 0.4 bandwidth, the 505 houses with negative coefficients on the distance variable are plotted in black in Fig. 2a, while the three houses with positive coefficients for the distance variable are in red. For the ‘‘red’’ houses, moving closer to the airport (i.e., the value of the distance variable declines) increases the probability of those houses being in the buffer zone. While there are few red houses, their sign may be a result of the fact that just to the north of the airport, the 65 and 70

286

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Fig. 2a.

Distance Coefficients and Location of Houses.

DNL zones are very thin. For the houses shaded in ‘‘black’’, moving closer to the airport lowers the probability of those houses being in the buffer zone. Also, for houses not located directly east or west of the airport, moving closer to the airport may or may not lead to higher noise levels. For example, for the ‘‘red’’ shaded houses located north of the airport in the buffer zone, moving further from the airport in the southeasterly direction can put them in the 65 DNL zone and, thus, subjected to more noise. Overall, however, the negative relationship between the estimated coefficients and distance (in logs) is strong. Generally speaking, the farther the house from the airport, the more likely the house is in the buffer zone (is subjected to the lowest noise level).9 This statement is corroborated by Fig. 2b. The smallest coefficient estimates (i.e., houses identified by white dots) tend to be farthest from the airport, while the largest coefficient estimates (i.e., houses identified by red dots) tend to be nearest the airport. The results for the variable measuring the percentage of houses in the neighborhood in which a house was sold with a Hispanic head of household (HispHH00) exhibit much heterogeneity. While the majority of estimates are positive (319 or 63 percent), a substantial number (189 or 37 percent) of the estimates are negative. The positive and negative values in Fig. 3a do cluster. Generally speaking, as shown in Fig. 3b, the smallest values are located west of

Where Does Airport Noise Fall? Evidence from Atlanta

Fig. 2b.

Distance Coefficient Quintiles and Location of Houses.

Fig. 3a.

Hispanic Coefficients and Location of Houses.

287

288

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Fig. 3b.

Hispanic Coefficient Quintiles and Location of Houses.

the airport, while the largest values are located southeast of the airport. When we calculate the correlation between the coefficient estimates and HispHH00, we find a positive association.10 This means that for these largest values an increase in the Hispanic percentage tends to be associated with a reduced probability of being in the buffer (i.e., more likely to be subjected to a noisier area). We also find heterogeneity in the parameter estimates for the variable measuring the percentage of houses in the neighborhood in which a house was sold with a black head of household (BlkHH00). However, the vast majority of estimates are positive (463 or 91 percent). The few negative values (45 or 9 percent) are clustered and dominate the southwest portion of our map in Fig. 4a. Because the majority of estimated coefficients are positive, an increase in the level of this variable is generally associated with a reduced probability of remaining in the buffer zone. Also, contrary to the findings for our Hispanic-related variable, the correlation between the coefficient estimates and BlkHH00 is negative.11 Thus, as suggested by Figure 4b, the very largest percentages of this variable tend to be associated with the smallest (i.e., negative) coefficients. This means that for these largest values an increase in the black percentage tends to be associated with an increased probability of being in the buffer (i.e., less likely to be subjected to a noisier area).

Where Does Airport Noise Fall? Evidence from Atlanta

Fig. 4a.

Fig. 4b.

Black Coefficients and Location of Houses.

Black Coefficient Quintiles and Location of Houses.

289

290

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Fig. 5a.

Fig. 5b.

Income Coefficients and Location of Houses.

Income Coefficient Quintiles and Location of Houses.

Where Does Airport Noise Fall? Evidence from Atlanta

Fig. 6a.

Fig. 6b.

Age Coefficients and Location of Houses.

Age Coefficient Quintiles and Location of Houses.

291

292

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

Turning attention to the income variable, the 36 negative values are clustered slightly northeast of the airport in Fig. 5a. Nearly all the negative values are in the buffer zone. The majority (472 or 93 percent) of estimated coefficients are positive. Thus, somewhat surprisingly, an increase in income is generally associated with a reduced probability of remaining in the buffer zone. As suggested by Figure 5b, the correlation between the coefficient estimates and income levels is, at most, weakly positive.12 Still, we find that for these highest income levels, an increase in income is associated with a reduced probability of being in the buffer zone. Finally, the clustering of positive values of the age variable occurs northeast of the airport in Fig. 6a. The vast majority (457 or 90 percent) of the estimated coefficients are negative. Thus, as suggested by Figure 6b, an increase in age is generally associated with an increased probability of being in the buffer zone. There is no obvious relationship between the estimated coefficients and income levels.

CONCLUSION The findings of a vast degree of heterogeneity with the OPLWR approach contrast with those from the OP estimation, so it is clear that exploring heterogeneity in different neighborhoods generates additional insights in assessing where the noise falls are masked in the OP model estimates.13 One implication is that standard ordered probit in the present case generates misleading and biased estimates due to the ignored heterogeneity among individual houses. This implication arises despite the fact that our analysis is restricted to a relatively small geographic area near the Atlanta airport. One might reasonably expect spatial heterogeneity to become even more pronounced for larger geographic areas. The Hispanic population variable exhibits a notable amount of heterogeneity in the sign of the coefficients. The impact of Hispanic population on airport noise varies in sign depending on geographic location. For the preferred bandwidth of 0.4, the estimated coefficient tends to be negative for the majority of the houses located west of the airport and positive for most houses east of the airport. We find heterogeneity depending on the composition of neighborhoods. The use of OPLWR is especially well-suited to identify such heterogeneity. In contrast, it is not possible to generate such detailed insights in an ordered probit model, so the OPLWR model enhances the interpretative potential by generating different parameter estimates for each house in our sample.

Where Does Airport Noise Fall? Evidence from Atlanta

293

By providing detailed geographic content, the OPLWR approach enables us to thoroughly answer the question: ‘‘Where does airport noise fall?’’

NOTES 1. See Cohen and Coughlin (2008, 2009) for numerous references. 2. This finding led them to conclude that those with Hispanic ethnicity incurred an environmental injustice. Environmental justice is not an issue that we can address effectively with our dataset. We lack sufficient data to assess whether a particular racial or ethnic group moved to a noisy neighborhood or airport noise encroached on a group to a disproportionate degree. Thus, we reach no conclusions as to whether some groups are affected unfairly by the decisions of others concerning airport noise. 3. In our study, we also considered confronting the possibility of simultaneity between housing prices and noise. In addition to the standard relationship of noise affecting housing prices, it is possible that housing prices affect noise. Airport authorities may choose to direct flights so as to distribute relatively more noise over relatively less expensive houses. This may be done for economic reasons, one of which is that compensation for harm might be less for lower-valued houses. Political reasons may also be operative as those living in less valuable houses may lack the political power to resist higher noise levels. We considered estimating an equation in which airport noise is a function of the instrumented housing prices, demographic variables, and other variables. This second equation would be estimated by ordered probit because airport noise is a qualitative dependent variable. But since the ordered probit is a nonlinear equation, we could not be sure that the parameters of the simultaneous system would be identified, so we opted to not pursue the simultaneous equations approach. 4. The measure of noise, the yearly day-night sound level (DNL), is a standard measure of noise used by the Federal Aviation Administration. A DNL of 65 decibels is the Federal Aviation Administration’s lower limit for defining a significant noise impact on people. At 65 decibels and above, individuals experience the disruption of normal activities, such as speaking, listening, learning, and sleeping. As a result, such noise levels are viewed as incompatible with residential housing. 5. McMillen and Redfearn (2010) and Carruthers and Clark (2010) estimate locally weighted regressions in the context of a hedonic housing price framework. But we are unaware of any studies that attempt to implement locally weighted regressions to assess where noise falls on different groups of people. 6. Using different estimation methods and a different model, Sobotta et al. (2007) find, similar to our result, that increased Hispanic percentages are significantly associated with more noise. While they find a positive association between higher ‘‘non-white’’ percentages in a neighborhood and more noise, the relationship is not statistically significant. Finally, they find a positive, statistically significant association between the percentage of households at or below the poverty rate in a neighborhood and more noise. Contrary to expectations, but somewhat similar to our results, they also found a positive association between the percentage of

294

JEFFREY P. COHEN AND CLETUS C. COUGHLIN

high-income households and more noise. However, this association was not statistically significant. 7. See Fotheringham, Brunsdon, and Charlton (1998, 2002) for general background on locally weighted regressions. 8. See Thorsnes and McMillen (1998) and McMillen and McDonald (2004) for details on the Gaussian function. 9. The Pearson coefficient is 0.80 and the Spearman rank-order coefficient is 0.84. 10. The Pearson coefficient is 0.44 and the Spearman rank-order coefficient is 0.56. 11. The Pearson coefficient is 0.43 and the Spearman rank-order coefficient is 0.32. 12. The Pearson coefficient is 0.002 and the Spearman rank-order coefficient is 0.14. 13. Due in part to this heterogeneity, we are unable to make any general statements about the presence of environmental justice (or injustice) with respect to airport noise in Atlanta. This is because the heterogeneity implies no clear pattern in the effects of demographics on noise levels, particularly for the Hispanic-related variable.

ACKNOWLEDGMENTS The authors thank Lesli Ott and David Lopez for excellent research assistance and the Atlanta Department of Aviation for noise contour data. We thank meeting participants at the Federal Reserve System Committee on Regional Economic Analysis, especially Anil Kumar, for their comments. We also benefitted from comments by participants at the Lincoln Institute of Land Policy Conference on Urban Economics and Public Finance in May 2010. Wim Vijverberg and Nancy Lozano-Gracia also provided helpful comments on earlier versions of the paper. The views expressed are those of the authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors.

REFERENCES Abramowitz, A. D., & Brown, S. M. (1993). Market share and price determination in the contemporary airline industry. Review of Industrial Organization, 8(4), 419–433. Carruthers, J., & Clark, D. (2010). Valuing environmental quality: A space-based strategy. Journal of Regional Science, 50, 801–832. Cohen, J. P., & Coughlin, C. C. (2008). Spatial hedonic models of airport noise, proximity, and housing prices. Journal of Regional Science, 48, 859–878. Cohen, J. P., & Coughlin, C. C. (2009). Changing noise levels and housing prices near the Atlanta airport. Growth and Change, 40, 287–313. Fotheringham, A., Brunsdon, C., & Charlton, M. (1998). Geographically weighted regression: A natural evolution of the expansion method for spatial data analysis. Environment and Planning A, 30, 1905–1927.

Where Does Airport Noise Fall? Evidence from Atlanta

295

Fotheringham, A., Brunsdon, C., & Charlton, M. (2002). Geographically weighted regression. Chichester, UK: John Wiley and Sons. Greene, W. (2003). Econometric analysis. Upper Saddle River, NJ: Prentice Hall. McMillen, D. P., & McDonald, J. F. (2004). Locally weighted maximum likelihood estimation: Monte Carlo evidence and an application. In L. Anselin, R. J. G. M. Florax & S. J. Rey (Eds.), Advances in spatial econometrics (pp. 225–239). New York, NY: Springer. McMillen, D. P., & Redfearn, C. (2010). Estimation and hypothesis testing for nonparametric hedonic house price functions. Journal of Regional Science, 50, 712–733. Sobotta, R. R., Campbell, H. E., & Owens, B. J. (2007). Aviation noise and environmental justice. Journal of Regional Science, 47, 125–154. Thorsnes, P., & McMillen, D. P. (1998). Land value and parcel size: A semiparametric analysis. Journal of Real Estate Finance and Economics, 17, 233–244.

CHAPTER 13 COMPETITION ON THE BASIS OF SAFETY? Ian Savage INTRODUCTION Safety is arguably the most important ‘‘quality’’ attribute of commercial aviation, yet it rarely figures into overt interfirm rivalry. Usually, airlines do not even allude to their safety record vis-a`-vis rivals in their advertising and press statements. Moreover, statistical analysis by independent parties usually indicates that peer airlines within the same geographic region and segment of the industry have indistinguishable safety records (Barnett, 2010). Of course, airlines and their trade associations are never shy about touting the continued decade-over-decade improvement in safety, and the lower level of risk inherent in air travel compared with land-based forms of transportation. However, this chapter does not focus on time series or crossmodal comparisons, but rather it considers cross-section differences between individual airlines. The lack of differentiation between airlines might seem to be somewhat surprising. One might imagine that some airlines might find it profitable to seek to gain a market advantage over their rivals by offering exceptionally high, or low, levels of safety. Indeed, as we will describe in the first part of the chapter, microeconomic theory predicts that firms will prefer to offer a different level of quality from their rivals because it reduces the intensity of price competition, and allows for the earning of supernormal profits. Pricing Behavior and Non-Price Characteristics in the Airline Industry Advances in Airline Economics, Volume 3, 297–323 Copyright r 2012 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 2212-1609/doi:10.1108/S2212-1609(2011)0000003015

297

298

IAN SAVAGE

Clearly, airlines have generally not elected to adopt such a strategy. Why is this? The chapter answers this question by describing the conditions that are necessary for a market outcome characterized by a diversity of safety offerings, and critically evaluating whether these conditions hold in practice. We will show that some prevalent market failures, including poor information, lead to a lack of incentives for airlines to diverge from the pack in their safety offerings. Airlines conclude that passengers would not notice that a greater-than-average level of safety is being offered, and thus would be unwilling to pay a higher price to obtain it. That said, there is often a fear that poor information flows might lead some airlines to, either intentionally or unintentionally, exploit passengers by providing less safety in the short run while still representing themselves to passengers as providing mainstream levels of safety. The chapter describes the theoretical and empirical literature dealing with this type of behavior. The main thrust of the chapter is to provide a theoretical and empirical understanding of why there is usually minimal safety differentiation. However, this is not always the case. The chapter concludes by pointing out that there are, at least, two examples of types of markets where there is overt rivalry that involves actual or imagined safety differences.

A MODEL OF SAFETY RIVALRY To motivate and frame the discussion, let us consider a stylized model of safety determination and safety rivalry. The aviation industry will be taken to be composed of a network containing many different city-pair markets, which we will call ‘‘routes.’’ Individual routes will be denoted by the subscript j. The industry as a whole has multiple airlines competing within it. Individual airlines will be denoted by the subscript k. In practice, not every airline serves every route and there are usually few rivals on each route. Indeed, in some ‘‘thin’’ markets there may be only one airline offering service. The implication is that the most applicable economic models to consider are those describing oligopoly or monopoly.

Airline Passengers There are numerous potential consumers of airline service. We will denote individual consumers by subscript i, and indicate the subset of these consumers who might travel on each route by the subscript ij. We will

Competition on the Basis of Safety?

299

assume that each consumer buys at most one unit of travel (a trip on a given route) in each time period. For the sake of illustration, that time period can be taken to be one calendar day. In deciding whether to take an airline trip on route j, and deciding which airline to patronize (in the event that there is a choice), the consumer takes the following variables into account: Bij

pjk Fjk

aj(1Sk)

Tjk

The net benefit from taking the trip by air relative to undertaking the trip by the next best alternative mode. If there is no practical alternative mode, this is simply a measure of the utility obtained by the consumer at the trip’s destination relative to the next best use of the consumer’s time. This is independent of the specific airline used. The fare charged by the various airlines. The vector of departure times offered across the day by the various airlines. Consumers will have a preferred departure time, and will gravitate (all other things being equal) to the airline that offers the least ‘‘schedule delay’’ or deviation from their preferred time (see Panzar, 1979). The probability of an incident on a given flight. We will argue in a few moments that each airline will determine an underlying level of safety (Sk) that will be common to every route that it serves. In addition there may be some route-specific safety factors, such as flying in difficult terrain or serving a poorly equipped airport, that are common to all airlines serving the route. This exogenously determined route-specific factor (aj) can be thought of as a proportionate deviation from the risk on the ‘‘average’’ route. A vector of other quality attributes (such as seat pitch and inflight amenities). While many of these quality attributes will be the same for an individual airline on each route its serves, such as the nature of its frequent-flier program or the responsiveness of its customer service, others will vary by route due to considerations of whether it is a short-haul or long-haul flight.

Frequency, safety and aspects of quality are examples of ‘‘vertical’’ product attributes in that all passengers would agree that ‘‘more is better,’’ but they may vary in their strength of taste for these attributes. We will specify a convenient separable, quasi-linear form for a utility function.

300

IAN SAVAGE

Specifically, presuming that the consumer i decides to travel, and patronizes airline k on route j, she will have a utility from taking the trip given by U ijk ¼ Bij þ bij F jk þ gij T jk  yij aj ð1  S k Þ  pjk

(1)

where b and g indicate a consumer’s individualized valuation of frequency and other quality attributes, respectively, and y measures the individualized (dis)utility from involvement in a crash. The consumer will evaluate their utility using Eq. (1) for every airline serving route j, and select to travel with the airline that provides the greatest net utility. In the event that the utility given by Eq. (1) is negative for all airlines offering service, the consumer will decide not to undertake an airline trip and either take the next best alternative mode or make the next best alternative use of their time. We will define the demand that each airline receives on a route as qjk. The nature of the determination of demand will be given in the next section but one.

Airline Costs Each airline has a total cost function for its entire network defined as follows: X TC k ¼ cðqjk ; F jk ; Sk ; T jk Þ (2) j

Total cost is assumed to be increasing in q, F, S, and T. There is a positive marginal cost of serving an additional passenger, running an additional flight or increasing amenities. The safety variable deserves some additional analysis. Safety-related costs are incurred at two points. First, the airline has to incur the cost of the ‘‘preventive effort’’ undertaken to ensure a given level of safety. Preventive effort has to be expended each time period. Airlines can reduce the probability of a crash by hiring higher-quality pilots, providing them with more training, ensuring that they do not become fatigued, purchasing more safety features for their aircraft, and making sure that the aircraft are properly maintained. In practice, economies of scope and scale in the production of safety and the interoperability of equipment and staff between routes mean that an airline has to decide on the level of safety at a network level and not on an individual route level. Consequently, we will be just defining Sk and not Sjk (remember that any idiosyncratic exogenous route-level safety factors

Competition on the Basis of Safety?

301

common to all airlines are measured by the variable aj). We will define the unit per-flight-departure safety prevention costs as an increasing convex function of the level of safety, and denote it as e(Sk). The second point at which costs may be incurred is when a crash occurs. These ‘‘crash costs’’ include (from the perspective of the airline) aircraft damage, lost staff productivity and staff medical expenses, and all exogenously defined legal responsibility for compensating injured passengers, cargo shippers, and any bystanders on the ground. In addition, there will be the discounted value of future financial consequences such as increased insurance premiums, reduced revenues if some passengers shun the airline, and any increased capital costs arising because the airline’s stock is less attractive.1 We will denote the amount of damages in a typical crash as D. (Because a crash will occur at a random location, these damages will be based on an average passenger load factor and will affect an average number of bystanders on the ground.) For the sake of clarity of discussion, we will assume that while the probability of a crash occurring, (1S), is endogenously determined, the consequences are exogenously determined. In reality, airlines can also choose to invest to mitigate the consequences in the event that an untoward incident occurs. Phillips and Talley (1992) empirically show that investments in pilot training and safety equipment can reduce the expected severity of a crash. For any given flight on a route with average exogenous safety factors (i.e., aj ¼ 1), an airline will have an expected safety cost, denoted by ESC, given by ESCðSÞ ¼ eðSÞ þ ð1  SÞD

(3)

This function and its components are illustrated in Fig. 1. The level of safety is shown on the horizontal axis, with low levels of safety at the lefthand end and high levels of safety at the right-hand end. As a consequence of the linear nature of the crash cost function and the convex nature of the prevention cost function, the expected safety cost function takes on a ‘‘U’’ shape. Expected costs are minimized at Smin. This point is defined by the first-order condition: @ESC ¼ e0 ðSÞ  D ¼ 0 @S

(4)

At the cost-minimizing level of safety, the slope of the prevention cost curve is equal to the damages suffered in a crash (the latter is the slope of the expected crash cost function in Fig. 1). Rational airlines would never offer safety of less than Smin because they could simultaneously lower their costs and increase their attractiveness to passengers by offering a higher level of safety. Therefore, the

302

IAN SAVAGE Cost ESC(S)

e(S)

D

Safety (S) 0

Fig. 1.

S min

1

Safety Costs for a Representative Flight on a Given Airline.

only part of the cost function for safety that would normally concern us is the portion that is upward sloping in the range SminrSr1. Consumer Choice This section will illustrate the nature of a consumer’s choice. To do so, let us consider the case where there are just two airlines serving a route. One airline, that we will denote by the subscript k ¼ H, offers a high level of safety and the other, denoted by the subscript k ¼ L, offers a lower level of safety. For simplicity, let us assume that we are dealing with a typical route, so that aj ¼ 1. We will not concern ourselves at this point with the industrial organization aspects of why there are just two airlines on this route, and how they selected their respective levels of safety and prices. The choice of safety level and price is the focus of the next section of the chapter. We will just specify that pjHWpjL. This assumption makes sense because we know that the high-safety airline incurs greater costs than the lower safety airline, and moreover the lower safety airline would be unable to attract any passengers if it offered worse safety and a higher fare. It is clear that we have a multidimensional problem in that the airlines are not just competing on the basis of safety but also in terms of the frequencies they offer, and the other quality attributes. As this is a chapter dealing with safety, we will assume for the sake of illustration that both airlines are

Competition on the Basis of Safety?

303

offering the same set of departure times, and identical nonsafety-related service quality. By doing so we can simplify Eq. (1) by defining a new variable, vij, that represents the valuation of the nonsafety aspects of the trip vij ¼ Bij þ bij F jk þ gij T jk If a consumer decides not to fly, denoted by subscript k ¼ 0, he will earn a utility of U ij0 ¼ 0

(5)

If he flies on the high-safety airline, his utility will be U ijH ¼ vij  gij ð1  S H Þ  pjH

(6)

and if he flies on the lower safety airline, his utility will be U ijL ¼ vij  yij ð1  S L Þ  pjL

(7)

max  and distributed according to a known We define that vj 2 ½vmin j ; vj max  is also function. The valuation of crash involvement yj 2 ½ymin j ; yj distributed with a known function, which will be taken to be independent of the function describing the distribution of vj. Consequently the set of possible consumers looking for airline service on route j is represented by the rectangular space illustrated in Fig. 2. Consumers located in the top righthand corner of the rectangular space have a high valuation of the nonsafety

vjmax

III

II

IV vj0H( ) I

vj0L )

V

VI vjmin j

Fig. 2.

min jLH

j

max

Consumer Choice Between Not Traveling by Air, and Traveling with High or Low Safety Airlines on Route j.

304

IAN SAVAGE

aspects of the trip and have a high valuation of safety (which is to say they incur a relatively large disutility from involvement in a crash). We can define the point where consumers are indifferent between the high- and low-safety airlines by setting Eq. (6) equal to Eq. (7). Rearranging this equality produces the critical value, yjLH, for which consumers are just indifferent: pjH  pjL (8) yjLH ¼ S jH  S jL Consumers with larger values of yj prefer to travel with the high-safety airline rather than that the low-safety airline, regardless of their value of vj. Equating Eqs. (5) and (6) and rearranging terms yields the following boundary condition characterizing the values of vj and yj for which consumers are indifferent between traveling with the high-safety airline and not flying: vj0H ðyÞ ¼ pjH þ ð1  SH Þyj

(9)

For any given value of yj, consumers with a vj greater than vj0H(y) prefer to travel with the high-safety airline rather than not flying. Similarly, we can define vj0L ðyÞ ¼ pjL þ ð1  S L Þyj

(10)

For any given value of yj, consumers with a vj greater than vj0L(y) prefer to travel with the low-safety airline rather than not flying. These boundary conditions are plotted in Fig. 2.2 We can now determine the consumption choices of individual consumers whose (vj,yj) values lie in different regions of the figure. Consumers in regions III and IV fly with the high-safety airline. Consumers in regions I and II travel with the low-safety airline. Consumers in regions V and VI choose not to fly. The implication is that provided the conditions are met for this market to operate, both high and lower safety airlines can coexist in the same market, and both can receive positive demand. Indeed, one could extend this model so that three or more different safety levels may exist on the same route.

AIRLINES’ CHOICE OF SAFETY AND PRICE Let us now explore how airlines might select their level of safety, and the prices that they charge. While it is not very realistic, the natural starting point is the benchmark case of the safety choices and prices that might exist

Competition on the Basis of Safety?

305

in a first-best world. We will then consider more realistic models of interfirm rivalry. We will contrast cases where the airlines provide similar level of safety, and cases where they differentiate their product. To compare these various market structures in the clearest possible way, we will make three simplifications. First, we will take it as understood that we are focusing on just one route, which will allow us to drop the subscript j. Second, all consumers will have identical values of vi, which we will just denote as v. Third, we will assume that the N total consumers are distributed in a uniform distribution over the range [ymin, ymax]. In the first-best world, every single level of safety in the range SminrSr1 is made available, and the price per passenger is driven to cost. We denote the latter as cq(S), where the cost per passenger is an increasing convex function in S. Each consumer’s utility will be given by U i ¼ v  yi ð1  SÞ  cq ðSÞ

(11)

The consumer will only choose to fly if Eq. (11) is nonnegative. Let us assume at this point that v is sufficiently large that not only do all consumers purchase at first best but also all do so when, as is described later in the section, airlines price above cost. The first-order condition indicates that the consumer’s utility maximizing, or ‘‘preferred,’’ level of safety, which we will denote as S i , is defined when the marginal cost of providing additional safety equals that consumer’s value of y. It is worth emphasizing that this is a consumer’s preferred safety level under the assumption that price is driven to cost. The first-order condition will produce a monotonic mapping from various consumers’ yi to their value of preferred safety level S i . Consequently consumers will be uniformly distributed in the range [Smin, Smax]. We will take this to be contained within the range [Smin, 1], which is the feasible range the airlines would be prepared to provide. While interesting as a reference point, this is clearly not of practical relevance to commercial aviation that is a lumpy product in that a large number of passengers share the same flight.3 Therefore, not every consumer will be provided with a level of safety that is tailored to their exact tastes. At best, we will be in a world of monopolistic competition described by Dixit and Stiglitz (1977). In their model, there is a fixed cost associated with production that leads to less variety of products than exists at first best. This would seem to be particularly relevant to the airline industry where there is a sunk cost associated with operating a flight with a negligible marginal cost for each additional passenger conveyed. In practice, we find that there are even fewer firms in the airline industry than is typically consistent with a model of monopolistic competition. The airline industry is more closely characterized by oligopoly models. Let

306

IAN SAVAGE

us contrast some duopoly models. Initially, we will look at models that have minimal (i.e., zero) safety differentiation, and then we will look at models that predict a large amount of differentiation. Models of minimal differentiation might be associated with conditions in the pre-deregulation airline industry when price was determined exogenously and was common to both airlines. The famous model of Hotelling (1929) has duopoly firms competing to obtain the largest market share. In this model, airlines do not compete on price but try to attract passengers by providing a level of safety that is closer to an individual passenger’s tastes than the level of safety offered by their rival airline. The dynamics of this model are that each airline is attracted closer and closer to the middle of the distribution of passenger tastes in an attempt to ‘‘steal’’ passengers from the other airline. The ultimate equilibrium is for both airlines to locate exactly in the middle of the distribution, which we might designate as safety level SM, and they share all the traffic equally between them. To extend this example, let us consider a market in which price is now endogenous but both airlines have the same exogenously predetermined safety level SM. The airlines now compete vigorously on price in a Bertrand fashion, and price gets driven down to cost, and neither airline earns a profit. If we now allow both safety level (in the first stage of a game) and price (in the second stage of a game) to be endogenous, we find that the duopolists have strong incentives to diverge in their safety offerings in order to lessen price competition (see Shaked & Sutton, 1982). Let us look at the second stage of the game after the two airlines have differentiated their product and offer safety levels SH and SL. Assuming that v is sufficiently large that all consumers decide to fly, the dividing point in the uniform distribution of consumers between ymin and ymax that will determine the market share of the two airlines is given by Eq. (8). Therefore, the profit function for the highsafety airline is given by pH ¼ ðpH  cq ðS H ÞÞ

ðymax  ðpH  pL Þ=ðSH  SL ÞÞN ðymax  ymin Þ

(12)

The first part of the term on the right-hand side is the price-cost margin, and the second part of the term is the demand for this airline indicated by the total market size, N, multiplied by the proportion of the consumers whose taste in safety is greater that the indifference point given by Eq. (8). There is a similar profit function for the low-safety airline given by pL ¼ ðpL  cq ðS L ÞÞ

ððpH  pL Þ=ðS H  S L Þ  ymin ÞN ðymax  ymin Þ

(13)

Competition on the Basis of Safety?

307

We will make the normalization that N ¼ 1, and (ymaxymin) ¼ 1. Both airlines maximize their profits with respect to the price that they charge. Differentiating Eqs. (12) and (13) with respect to an airline’s own price, and making a Cournot-style assumption that the other airline will not change its price in response to a price change by the first airline, produces a pair of reaction functions. The intersection of these reaction functions defines a Nash equilibrium in prices. The equilibrium price charged by the high-safety airline is given by 1 1 pH ¼ ð2cq ðSH Þ þ cq ðS L ÞÞ þ ð2ymax  ymin ÞðS H  S L Þ 3 3 and the price of the low-safety airline is given by 1 1 pL ¼ ðcq ðS H Þ þ 2cq ðS L ÞÞ þ ðymax  2ymin ÞðSH  SL Þ 3 3 Some additional manipulation of these two price definitions produces a number of powerful conclusions. The first is, as one would hope, the price charged by the high-safety airline is higher than that of the low-safety airline (if this was not the case, the low-safety airline would receive zero demand). Second, both airlines make positive profits, so both of them are better off than offering an identical SM and competing vigorously on price. Third, the price-cost margin for the high-safety airline is larger than that for the lowsafety airline. Fourth, as indicated by the final parenthetical terms in both price equations, the profit margins for both airlines get larger the more differentiated their products. This implies that in the first stage of the game where the airlines choose their safety level, they should differentiate their products as much as is possible.

CONDITIONS FOR A DIFFERENTIAL SAFETY LEVEL TO EXIST The implication of the previous section is that airlines should attempt to differentiate themselves as much as possible from their rivals because it softens the price competition between them. In reality, the stereotype of commercial aviation is that all the mainstream airlines seem to be offering similar levels of safety. Why is this? For the market to produce differentiation, it has to meet certain conditions. Let us list these conditions and assess their validity for commercial aviation.

308

IAN SAVAGE

A Diversity of Consumer Tastes for Safety The model described in the previous section requires a sufficient amount of consumer heterogeneity. Specifically, ymax has to be approximately twice as large as ymin. The low-safety airline will only make nonnegative profits if ymax  2ymin 

CL  CH SH  SL

(14)

One will note that the numerator of the right-hand side is negative, so it is possible that ymax could be less than twice as large as ymin, but only by the amount given by the ratio on the right-hand side. If there is insufficient diversity, the low-safety airline will make negative profits and withdraw from the market. Expressed in a more intuitive way, in the event that the range [ymin, ymax] is very small, or even if the range was larger but the preferences of nearly all of the consumers were not uniformly distributed but were tightly packed together, then we would expect that airlines would offer a similar level of safety. Airlines will locate where the passengers are, and if the passengers all have broadly similar valuations, then airlines will tend to offer broadly similar levels of safety. To my knowledge there are no papers or reports that have conducted surveys to measure interpassenger variation in the value placed on safety, or conducted revealed or stated preference tests from which one could infer such a variation. However, one would imagine that the variation in risk/ price trade-offs that consumers display in their purchases of other products carries over to the airline markets. Some consumers willingly purchase less safety features for their automobiles, or purchase older models with less safety features, or decide to replace their tires less often in order to save money. Some consumers prefer to ride bicycles without helmets while others do not. Some people jaywalk when crossing the street, while others walk a longer distance to cross the road at a recognized crosswalk. Therefore, I would imagine that a similar variation exists for commercial aviation.

Consumers Know the Safety Level(s) Offered by Airlines The general requirement for most economic models to function correctly is that consumers need to be fully informed when they make their consumption choices. For some aspects of the airline product, such as fare, schedules, seat pitch, and in-flight amenities, passengers can quite easily

Competition on the Basis of Safety?

309

collect information so as to make informed choices. In contrast, the underlying probability of a crash for a particular airline is less easy to observe. Indeed this probability is difficult for even safety professionals to know (see, for example, the paper by Chang and Yeh (2004) which attempts to define an index to measure safety). We will argue later that even an individual airline really cannot quantify its own safety posture in any definitive sense, although it can make a better informed judgment than can passengers. The asymmetry of safety information between airlines and passengers is a classic market failure. Of course, one would not expect passengers will be totally uninformed. The industry’s and an individual airline’s safety mishaps are public knowledge and thus passengers can get information from the press and other sources including official government databases. Consequently, even somebody who has never flow before can form some knowledge. In economic parlance, aviation is ‘‘search good’’ rather than an ‘‘experience good’’ whose quality can only be assessed by consumption (Nelson, 1970). While one would imagine that frequent fliers do collect observations and experiences to update their safety perceptions, the fact that safety is a probabilistic attribute means that it cannot be assessed on every trip (Shapiro, 1982). Indeed the trip on which you learn that your chosen airline displays a low level of safety may be your last. There is a small literature that empirically investigates whether an airline’s safety performance is correlated with other quality attributes that may be more readily observed. Rhoades and Waguespack (1999) were unable to find a correlation between complaints about service made to the federal Department of Transportation and a broadly defined measure of safety for nine major US airlines for the period between 1991 and 1997. However, they did find a positive correlation – more complaints about service are associated with more safety incidents – for a sample of 21 regional airlines over the same time period (Rhoades & Waguespack, 2000). Martin and Roma´n (2010) conclude that there is such a correlation for even the largest 10 airlines in the United States, but their data is for only one year, and that year was the tumultuous 2001. It is beneficial to separate consideration of passengers’ knowledge of safety level into two parts. The first is whether passengers can form an accurate evaluation of the general safety level of the industry. The second is whether they can differentiate between the safety offered by various airlines within the industry. There has been a literature dating back more than 30 years that explores how people form perceptions of risk. In particular, psychologists have found

310

IAN SAVAGE

two systematic biases (Lichtenstein, Slovic, Fischhoff, Layman, & Combs, 1978). The first, termed primary bias, is the tendency to overestimate infrequent causes of death and to underestimate more frequent causes. The ‘‘crossover’’ point where perception and reality are closest is for risks that claim 300–500 per 100 million Americans a year, such as syphilis and hypothermia. The fatality count for commercial aviation is considerably smaller. The upward primary bias is compounded by a secondary bias. Hazards with an upward secondary bias are generally dramatic and sensational, whereas hazards with downward secondary bias tend to be unspectacular events, that claim one victim at a time. The receipt of information from the media was also found to influence the judged frequency, and the frequency of newspaper reporting usually has little to do with the actual frequency of that risk in the community (Combs & Slovic, 1979). Barnett (1990) illustrates that aviation crashes generate a disproportionate level of coverage by the media relative to other common risks in society. The consequences of this general tendency to believe that aviation crashes happen more frequently than they actually do can be examined in a simple model of monopoly. The general result is that the upward misperception will lead airlines to provide less than the optimal level of safety. This somewhat counterintuitive result emerges because passengers do not fully incorporate into their demand functions the ‘‘benefits’’ of the preventive actions taken by the airlines (Spence, 1977). However, there is another effect that works in the other direction. Psychometric researchers have also asked respondents to rate acceptance of risks based on various characteristics. Because most of these characteristics are collinear with each other, factor analysis has been used to boil these down to two major factors. The first is whether the probability and consequences of a risk are known in advance and generally understood. This is referred to as the ‘‘unknown factor.’’ The second is that certain types of risk engender ‘‘dread.’’ The dread factor is an amalgam of various risk attributes including whether the victim is exposed involuntarily, whether the outcome is likely to be fatal, whether the risk involves a nasty drawn-out form of death, and whether the consequences can be mitigated by the diligence or skill of the victim when a risky situation occurs. Researchers have found that the higher the unknown or dread rating of a risk, the more that society is intolerant of it. In Fischhoff, Slovic, Lichtenstein, Read, and Combs’s (1978) analysis of 30 common risks, commercial aviation was found to have a close to average unknown score, but the largest dread score. Indeed the dread score was considerably larger than that assessed for handguns, motorcycles, or

Competition on the Basis of Safety?

311

working in the construction industry. The extremely high dread score resulted from respondents feeling that an aviation crash was certain to have fatal consequences, and that the victim could not personally mitigate or escape from the any dangers that arise. Clearly, being strapped into a seat in a confined aluminum tube miles above the earth is considered to be very disconcerting in the event that an emergency occurs. The implication is that the values of y for aviation will tend to be very much larger than is the case for, say, automobile travel. Consequently, we would expect that the market would work to provide very high levels of safety in this industry. So, perhaps it is not surprising that aviation is statistically the safest mode of transportation. The situation becomes more complicated in competitive markets with product differentiation where passengers not only have to have some idea of the general safety level of aviation, but also have to be able to distinguish between the safety records of individual airlines. Theoretical models show that in the extreme, if passengers were unable to distinguish on the basis of safety between airlines, then no airline would choose to supply a high-safety service because passengers would not recognize the service and would be unwilling to pay a higher price to obtain it (Akerlof, 1970). In effect, there will be a ‘‘race for the bottom’’ as all airlines reduce costs by moving to offer Smin. This extreme situation does not seem to apply in practice, however. Not all airlines seem to offer the bare minimum in terms of safety. In part this may be because while passengers may not be able to differentiate between many airlines due to a lack of information, they can recognize a ‘‘notorious’’ airline and act accordingly. Klein and Leffler (1981) argue that consumers are aware of the underlying costs of production and can thus calculate a ‘‘quality assuring price.’’ Firms charging less than this amount cause consumers to suspect that a shoddy product is being offered. An alternative explanation is advanced by Rogerson (1983) who argues that firms with poor quality services tend to have more dissatisfied customers than firms offering high-quality service, and ‘‘word of mouth’’ may signal to other consumers which firms are notoriously bad. It would be an interesting research project to determine how and why an airline may acquire a notorious reputation. Is there a ‘‘tipping point’’ or a critical number of crashes that has to occur in a given period of time for an airline’s reputation to change? Are there certain types of information that might be revealed that are particularly damaging to a reputation? There is some literature that empirically investigates whether the information provided by a crash is incorporated into demand decisions. Borenstein and Zimmerman (1988) using US data from 1960 to 1985 find

312

IAN SAVAGE

that crashes do lead to a decline in demand for the crash-involved airline, but that the effect is temporary and equivalent to about 10–15% of one month’s demand volume.4 Wong and Yeh (2003) found similar results using data from Taiwan for the period from 1981 to 1999. The monthly demand loss was about 22% with a duration of two-and-a-half months on average. Interestingly, while some of the lost passengers may decide to travel on rival airlines, there is also a dampening effect on the demand for air travel in general with the demand for other airlines declining temporarily by 5%. Of course, set against these general findings of limited demand response, one could point to specific examples of airlines with notoriously poor records having been forced to contract or exit the market, or passengers’ shunning specific models of aircraft. It is attractive to argue that passengers can probably identify notoriously poor airlines, but they are probably less sure about the safety rankings of more mainstream airlines. This is primarily because crashes occur with (fortunate) rarity for individual airlines, making recent press coverage an unreliable indicator for making demand decisions. Moreover, those crashes that do occur often are caused by a bizarre set of circumstances that makes it unclear whether passengers should infer that the airline was at fault or instead blame the weather, an ‘‘act of God,’’ or pure bad luck. More definitive interoperator safety data, such as the number of pilot deviations or the number of engines shut down in flight, can be difficult to obtain and interpret. Information on the underlying determinants of safety such as staff training and maintenance procedures is generally not available. Even if they were, the link between the magnitude of these inputs and actual safety performance is a mystery even to safety professionals. The situation is made worse because it is regarded as somewhat unseemly for airlines to advertise that their crash rate is better than that of their competitors. It may also be counterproductive in that highlighting an essentially negative aspect of aviation may reduce the demand for all airlines. At best airlines have to use code words to communicate to potential passengers that they are supplying a premium service. Examples are highlighting the experience of their mechanics, or indicating that they offer high quality in other attributes of service and hoping that this reputation will also be inferred concerning their safety performance. The problems in effectively communicating information to passengers probably help explain why most mainstream American airlines have statistically indistinguishable safety records. Airlines would be unwilling to offer a higher level of safety if they believe that passengers would not appreciate that a higher level of safety is being provided, and hence would be unwilling to pay higher prices to compensate for the higher costs.

Competition on the Basis of Safety?

313

Consumers Correctly Process Safety Information Even if passengers were fully informed, they may not make rational choices. The unpleasant consequences of a crash may cause even fully informed passengers to downplay the probability of a crash. Calabresi (1970) explains this ‘‘it will not happen to me’’ behavior as the ‘‘Faust’’ attitude whereby people are myopic when making a choice between a lower price now and an increased probability of death or injury later. Oi (1973) explains the phenomenon as consumers with a very high time preference who in retrospect regret their choices because their retrospective time preference is different. It may well be that cognitive dissonance reinforces the tendency for mainstream airlines to offer similar safety levels because passengers will ignore the lower risk offered by a safer airline, but react negatively to the higher prices that they charge.

No Complications from Multi-Attribute Competition In the model described earlier in the chapter, rivalry between airlines occurs in several quality dimensions in addition to safety. Airlines compete on the basis of their schedule, and a whole host of other attributes including inflight amenities, seat pitch, frequent-flier benefits, ticketing policies, and customer service. It is tough to describe a unique equilibrium when airlines can compete in so many dimensions. Moreover, because passengers can readily compare schedules and amenities, it would seem that airlines would find it more effective to differentiate themselves in these dimensions, and hence blunt price competition, rather than in the rather amorphous safety dimension. This is yet another reason why we may not find much differentiation in terms of safety between mainstream airlines.

Airlines Know Their Production Functions The theoretical analysis assumes that firms can definitively decide how much safety they wish to produce, and that a given amount of preventive effort will lead to a given safety performance. There is a growing literature on how safety is produced, with aviation professionals taking a leading role, that suggests that simplistic relationships between maintenance expenditures or staff training and safety outcomes are not sufficient. Safety outcomes depend not only on these direct inputs to safety but also on the layers of

314

IAN SAVAGE

‘‘defenses’’ that companies build into their systems to protect against naturally occurring human and environmental errors (Maurino, Reason, Johnson, & Lee, 1995). It also depends on the ‘‘safety culture’’ that senior management creates throughout the whole organization. The work of Professor James Reason argues that most accidents in high technology systems do not occur as a result of only one failure, but are caused by a whole chain of events. Even if one link in the chain can be broken, the hazardous circumstance may not result in an accident. Furthermore every ‘‘defense’’ against accidents usually has some, known or unknown, flaw. One could think of this as a hole in a slice of Swiss cheese. For an accident to occur, all of the holes in the multiple slices of cheese, that represent the multiple defenses, must be lined up. The implication is that the production of safety is somewhat of a black box. Firms can only make decisions on the inputs to safety, and the number of defenses put in place (i.e., the number of slices of cheese), but the resulting safety level is somewhat unpredictable.

Regulations Do Not Truncate the Range of Safety Offerings This chapter deals with the competitive process and is not a treatise on the many justifications for safety regulation. For our purposes, it is sufficient to note that a minimum safety standard may curtail the range of safety options that airlines could offer to knowledgeable and willing passengers. Depending on how binding the minimum standard is on the distribution of consumer tastes and the commercial decisions of airlines, it is possible that minimum standards may work to compress the range of safety choices offered. In effect, we may observe less diversity because regulation is truncating the lowest part of the safety distribution, and some – but not all – passengers in the truncated segment decide to purchase greater safety than they would like at a higher price. As an aside, there may be an interesting social dynamic at work. In a vertically differentiated safety market, high-safety choices will be available at premium prices, whereas low-safety choices will be available at discounted prices. Less well-off members of society may only be able to afford to patronize airlines with very poor safety records. Even though these people may make that choice in a fully informed way, society may paternalistically decide that this is not right and that these travelers face ‘‘too much’’ risk. However, if society implements a minimum safety standard that increases the safety offered by low-safety airlines, at least some passengers will be priced out of the market. Implicitly, society is indicating that it would prefer that some consumers not travel rather than face inordinate risks.

Competition on the Basis of Safety?

315

WHY SOME AIRLINES MAY DEVIATE DOWNWARDS FROM THE PACK If passengers have imperfect information on safety, it is clear that no rational airline would wish to provide a higher level of safety than mainstream airlines because it would not be able to convince passengers to pay a higher price to compensate for the greater costs that the higher safety level entails. However, on the flip side there would be an incentive to take advantage of passengers by providing less safety. For example, an airline that had previously offered similar safety to its mainstream rivals could make cost savings in its preventive efforts, but would not suffer reduced revenues because it could masquerade as a mainstream-safety airline, and continue to charge the regular price. The incentives to engage in this kind of behavior are even stronger because the costs of prevention are borne in the present, whereas the effects of crashes occur at randomly defined points in the future. Even an airline that becomes very careless may not suffer a visibly increased crash rate for several years. In the interim the airline can earn excess profits, which will cease only when it incurs the costs of crashes and/or when its passengers find out and either shun the airline or demand a lower price. There is an extensive theoretical and empirical literature concerning such ‘‘cheating.’’ In our discussion, we will consider two types of cheating. The first is premeditated, or avaricious, cheating, and the second is involuntary cheating that occurs when a firm does not intend to cheat the customer but does so because of inexperience or a lack of knowledge of its production function. One should note that this is not technically competition on the basis of safety, because the airline that engages in this behavior wants passengers to believe that it is still providing the same level of safety as the other mainstream airlines, and does not cut its price as it wishes to earn a price premium over cost. Avaricious Cheating The most common explanation of avaricious cheating is that the airline is close to bankruptcy. The airline reasons that it can save on prevention costs now and declare bankruptcy to protect itself against the cost of crashes later on.5 A less callous explanation may be that a financially distressed airline hopes that cost economies can prolong its life until better times come along. Other explanations are that an airline might feel that it needs a short-term financial boost to improve its stock price or make itself attractive to a potential purchaser, or to prove to stockholders that a recent merger had led to the promised cost savings.

316

IAN SAVAGE

There has been a small but influential literature investigating whether there is an empirical link between financial condition and accident experience.6 An early paper by Golbe (1986) was updated and expanded by Rose (1990). Rose’s work used data for 35 large scheduled airlines in the United States over the period 1957 to 1986. She found a negative relationship that was statistically significant at the 10% level: a larger operating margin implied lower accident rates. A decrease in financial performance from average to one standard deviation below average was estimated to increase the accident rate of the airline by 7.5%. When categorized by size of firm, it would appear that the profitability–safety relationship only held for middle-sized and small airlines. There was no statistical relationship for large airlines. Rose’s work was updated by Raghavan and Rhoades (2005) using data on 12 major airlines and 18 regional airlines in the post-deregulation period from 1980 to 2002. Their results were remarkably similar to Rose’s. Overall, there was a negative relationship between profitability and safety that was significant at the 5% level. When the data set was divided into major airlines and regional airlines, the sign of the relationship still held for both groups, but it was statistically significant at the 10% level for the smaller regional airlines, and statistically insignificant for the major airlines. Dionne, Gagne´, Gagnon, and Vanasse (1997) conducted a study of 120 Canadian airlines of various sizes using pre-deregulation quarterly data from 1976 to 1987. The definition of accidents was broader than that used by Rose. They, like Rose, found that there was not a statistically significant relationship between operating margin and safety for large airlines. However, for smaller airlines there was a strong positive relationship: a larger operating margin implied higher accident rates! The authors then explored alternative measures of financial health, and found more intuitive results. Among the smaller airlines, higher maintenance expenditures were found to reduce accident rates in a statistically significant way. However, this relationship was not found for larger airlines. The use of data on debt to equity ratios provided additional insights. Among the smaller airlines a higher debt to equity ratio (when equity was positive) resulted in greater safety. In addition, airlines with large debts and negative equity (when measured with a lag) were found to have worse safety than those with positive equity.

Involuntary Cheating The fact that the production function for safety is ill defined means that there is the possibility that some incumbent airlines may involuntarily

Competition on the Basis of Safety?

317

deviate from the safety level offered by other airlines. Nance (1986) argued that the pressures for airlines to trim costs in the period after deregulation in the United States in 1978 led to decisions that could have led to safety reductions. Nance argues that airline vice presidents of maintenance or operations would have a hard time resisting requests for cost reductions because the link between a particular cut and an increase in risk is tough to quantify. In James Reason’s terminology, what is the safety consequence of saving money by removing one of the layers of Swiss cheese? A second group that may involuntarily offer lower safety is new entrants. While these entrants may aspire to provide similar safety to mainstream incumbent airlines, entrants may undertake too little prevention in the present and regret it when crashes and adverse customer reaction occur in the years ahead. While the motivation for some of this behavior might be avaricious, it is more likely to be attributed to inexperience. The complexity of the safety production function only makes this problem worse. This is a very real concern given that there has been considerable new entry since economic deregulation. Investigations show that the new jet airline entrants of the early 1980s in the United States were not noticeably worse than established airlines (Kanafani & Keeler, 1989), but that was not true of the cohort of entrants in the early 1990s (Savage, 1999). Of course, passengers may suspect that new entrants might provide a lower level of safety than established airlines. Shapiro (1983) presents a model where all new entrant firms are regarded by consumers as providing low quality in their first period of operation. Consequently, all entrants must initially price in a way that is consistent with the prices charged by lowquality firms. If an entrant is truly providing high quality, then consumers will recognize this after a period of time. As soon as the firm gains a reputation for high quality, it can then charge a price consistent with high quality. Of course, the price of high quality will have to contain a mark up over costs because the firm has to recoup, over time, its investment in pricing below costs in its initial period of operation.

WHERE SAFETY DIFFERENTIATION DOES OCCUR The main thrust of this chapter is to provide a theoretical underpinning to explain why most mainstream airlines offer similar safety, and why rivalry on the basis of safety is largely absent. However, one should not assume that it is totally absent. There are two examples that come to mind that feature

318

IAN SAVAGE

rivalry between airlines that includes safety differences which are readily apparent to potential passengers. The first is intercontinental service. Using data from 2000 to 2007, Barnett (2010) states in his abstract that: The safest nations are the traditional first-world countries (e.g., Canada, Japan), with a death risk per flight of about 1 in 14 million. Next safest are those developing-world nations that have either have recently attained first-world status (e.g., Singapore, South Korea) or are classified by experts as newly industrialized (e.g., Brazil, China). Their aggregate death risk per flight was about 1 in 2 million. The least safe nations statistically are remaining developing-world countries, with a death risk per flight of about 1 in 800,000.

Barnett finds that any differences within the various groups, even those within the developing nations group, were statistically insignificant, but that the differences between these three groupings are highly significant. The reason for the safety differential is clear. In developing countries, risks of all types (from disease, workplace risks, and common accidents) are much higher than in the first world, and it is natural to expect that the risks from flying are similarly higher. In markets where the route networks of airlines from these different groups overlap, safety is clearly a differentiating factor in the rivalry. The second concerns the introduction of regional jets in the past 15 years that has blurred the market boundary between mainline jet airlines and regional airlines. Until the mid 1990s, certain markets were served almost exclusively by turbo-prop aircraft operated by regional or commuter airlines, and other markets were the exclusive preserve of traditional large jet service. Now this is not the case. A couple of parochial examples can illustrate this point. Delta Air Lines recently entered the market between Chicago O’Hare and New York LaGuardia, a market long dominated by United Air Lines and American Airlines. It did so using a regional affiliate, Shuttle America, using Embraer 175 regional jets. United and American Airlines provide service with their own staff and equipment using (as of January 2010) primarily Airbus A320-family and McDonnell Douglas MD80 jets, respectively. In the market between Chicago O’Hare and Washington Reagan National, United and American Airlines are longtime duopolists. United still primarily operates this service with their own Airbus A320-family and Boeing 757 equipment. American Airlines, however, now splits its flights roughly equally between directly operated flights using Boeing 737–800s and flights operated by its regional sister company, American Eagle, using Bombardier CRJ700 regional jets. While the public may regard regional jets as inferior in tangible ways such as baggage space

Competition on the Basis of Safety?

319

and headroom, there is a perception (rightly or wrongly) that smaller aircraft are less safe and that the pilots of regional airlines are less experienced than their large-jet counterparts and work longer hours.7 Both of these types of markets provide the opportunity for interesting research studies. Studies could investigate a number of questions. Do the passengers on these routes take into account any safety differences in their airline choice decisions? Does the operator of the (actual or perceived) lower safety option have to offer a lower fare? What is the consequent effect on market share?

CONCLUDING REMARKS Aviation safety is a very emotive subject. The media disproportionately covers aviation crashes relative to other risks in society. In part this is because aviation crashes, unlike most highway crashes, are multiple fatalities events. In part, it is explained by psychometric studies that have shown that the uncontrollability of the situation (from the perspective of the passenger) when something untoward occurs leads to a feeling of ‘‘dread’’ and a greater social intolerance of safety lapses. One would imagine that airlines might cater to these concerns by competing strongly with each other by touting their safety credentials. However, in practice, mainstream airlines seem to offer statistically indistinguishable safety, and do not overly discuss safety in their advertising. Microeconomic theory argues that that is not a profit-maximizing strategy. In general, firms should try to differentiate their products as much as possible so as to lessen direct price competition and allow for the possibility of earning supernormal profits. This chapter lays out the standard oligopoly models that support maximal differentiation. Of course, such models have a variety of underlying assumptions. The chapter discusses each of these assumptions and whether they hold in practice. Clearly airlines would only offer differential safety levels if some passengers preferred high levels of safety, whereas others would be satisfied with a slightly lower level of safety that is supplied at a lower price. If all passengers had roughly similar tastes, then we would optimally expect to see minimal differentiation. While I personally do think that such a variability of tastes does exist, I cannot point to any empirical studies that provide any statistical support to confirm or refute this. Perhaps the classic market failure is that passengers are not well informed about safety offerings, and internal cognitive processes may further hinder

320

IAN SAVAGE

any rational decision making. In these circumstances, airlines would not have any incentive to provide safety levels greater than those offered by rival airlines because passengers would not fully incorporate the greater safety levels into their decision making and would be unwilling to pay a higher fare to compensate for the greater safety investment. It is perhaps not surprising that airlines prefer to differentiate their products in more tangible ways such as schedule and frequent-flier privileges rather than competing on the basis of safety. In some classic economic models, failures in information flows can lead to all firms offering the lowest possible levels of quality because no firm has any incentive to invest in quality because they are not rewarded for it. This does not appear to be the case for commercial aviation, where the general level of safety is extremely high. The implication is that while passengers may not be able to differentiate between mainstream airlines, they can recognize a ‘‘notorious’’ airline and act accordingly. The concept of notoriety would appear to present a substantial opportunity for empirical research. Are there examples of airlines that have obtained a notorious reputation for poor safety? What was the mechanism by which this reputation was acquired? Does there need to be a critical number of incidents in a given period before a ‘‘tipping point’’ occurs and an airline obtains a notorious reputation? Is the occurrence of crashes the most important element in determining a reputation, or are there other types of information that might become available that are more damaging? Finally, while it is generally true that most airlines offer similar safety to their peers in the same geographic region and market segment, there are some markets in which airlines with real or perceived safety differences are pitted against each other. One example is on routes between developed and developing world countries. Another is on routes where regional jet service competes with mainline jet service. These markets provide empirical researchers with the opportunity to investigate how passengers choose between the rival airlines and the consequences for pricing and market share. In summary, there is a considerable agenda that empirical researchers may wish to tackle.

NOTES 1. The effect of crashes on demand will be discussed later in the chapter. For evidence on the, seemingly limited, long-term effect on stock prices, see Chance and Ferris (1987), Mitchell and Maloney (1989), the review by Rose (1992), Bosch, Eckard, and Singal (1998), and Kaplanski and Levy (2010).

Competition on the Basis of Safety?

321

max 2. For the diagram to look like Fig. 2, we are assuming that ymin . j oyjLH oyj 3. This model may, however, be ideal for thinking about the safety of general aviation. 4. This literature does suffer from the problem that a crash-involved airline may have to temporarily reduce its fares that would mean that the demand response to a crash may be understated. 5. For theoretical treatments, see Bulow and Shoven (1978), Golbe (1981), Klein and Leffler (1981), and Shapiro (1982). 6. There is always an element of ‘‘chicken and egg’’ with the empirical literature, in that it is not always clear whether the poor financial conditions lead to a larger number of crashes, or vice versa, or some combination of the two. 7. The latter was the subject of a National Transportation Safety Board investigation of the crash of Colgan Air flight 3407 near Buffalo, New York on February 12, 2009. The aircraft involved in this accident was a turbo-prop aircraft, operated by a regional affiliate of Continental Airlines. While Colgan Air only operates turbo-prop aircraft, it is part of a larger undertaking that also operates regional jets.

REFERENCES Akerlof, G. A. (1970). The market for ‘lemons’: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 3, 488–500. Barnett, A. (1990). Air safety: End of the golden age. Chance: New Directions in Statistics and Computing, 3, 8–12. Barnett, A. (2010). Cross national differences in aviation safety records. Transportation Science, 44, 322–332. Borenstein, S., & Zimmerman, M. (1988). Market incentives for safe commercial airline operation. American Economic Review, 78, 913–935. Bosch, J. C., Eckard, E., & Singal, V. (1998). The competitive impact of air crashes: Stock market evidence. Journal of Law and Economics, 41, 503–519. Bulow, J., & Shoven, J. (1978). The bankruptcy decision. Bell Journal of Economics, 9, 437–456. Calabresi, G. (1970). The cost of accidents: A legal and economic analysis. New Haven, CT: Yale University Press. Chance, D. M., & Ferris, S. P. (1987). The effect of aviation disasters on the air transport industry: A financial market perspective. Journal of Transport Economics and Policy, 21, 151–165. Chang, Y., & Yeh, C. (2004). A new airline safety index. Transportation Research Part B: Methodological, 38, 369–383. Combs, B., & Slovic, P. (1979). Newspaper coverage of causes of death. Journalism Quarterly, 56, 837–843, 849. Dionne, G., Gagne´, R., Gagnon, F., & Vanasse, C. (1997). Debt, moral hazard and airline safety: An empirical evidence. Journal of Econometrics, 79, 379–402. Dixit, A., & Stiglitz, J. E. (1977). Monopolistic competition and optimum product diversity. American Economic Review, 67, 297–308. Fischhoff, B., Slovic, P., Lichtenstein, S., Read, S., & Combs, B. (1978). How safe is safe enough? A psychometric study of attitudes towards technological risks and benefits. Policy Sciences, 9, 127–152.

322

IAN SAVAGE

Golbe, D. L. (1981). The effects of imminent bankruptcy on stockholder risk preferences and behavior. Bell Journal of Economics, 12, 321–328. Golbe, D. L. (1986). Safety and profits in the airline industry. Journal of Industrial Economics, 34, 305–318. Hotelling, H. (1929). Stability in competition. The Economic Journal, 39, 41–57. Kanafani, A., & Keeler, T. E. (1989). New entrants and safety. In L. N. Moses & I. Savage (Eds.), Transportation safety in an age of deregulation. New York, NY: Oxford University Press. Kaplanski, G., & Levy, H. (2010). Sentiment and stock prices: The case of aviation disasters. Journal of Financial Economics, 95, 174–201. Klein, B., & Leffler, K. E. (1981). The role of market forces in assuring contractual performance. Journal of Political Economy, 89, 615–641. Lichtenstein, S., Slovic, P., Fischhoff, B., Layman, M., & Combs, B. (1978). Judged frequency of lethal events. Journal of Experimental Psychology: Human Learning and Memory, 4, 551–578. Martin, J. C., & Roma´n, C. (2010). Evaluating the service quality of major air carriers: A DEA approach. International Journal of Applied Management Science, 2, 351–371. Maurino, D., Reason, J., Johnson, N., & Lee, R. B. (1995). Beyond aviation human factors: Safety in high technology systems. Aldershot, UK: Ashgate. Mitchell, M. L., & Maloney, T. M. (1989). Crisis in the cockpit? The role of market forces in promoting air travel safety. Journal of Law and Economics, 32, 329–355. Nance, J. J. (1986). Blind trust. New York, NY: William Morrow. Nelson, P. (1970). Information and consumer behavior. Journal of Political Economy, 78, 311–329. Oi, W. Y. (1973). The economics of product safety. Bell Journal of Economics and Management Science, 4, 3–28. Panzar, J. C. (1979). Equilibrium and welfare in unregulated airline markets. American Economic Review, 69, 92–95. Phillips, R. A., & Talley, W. K. (1992). Airline safety investments and operating conditions: Determinants of aircraft damage severity. Southern Economic Journal, 59, 157–164. Raghavan, S., & Rhoades, D. L. (2005). Revisiting the relationship between profitability and air carrier safety in the U.S. airline industry. Journal of Air Transport Management, 11, 283–290. Rhoades, D. L., & Waguespack, B. (1999). Better safe than service?: The relationship between service and safety quality in the U.S. airline industry. Managing Service Quality, 9, 396–401. Rhoades, D. L., & Waguespack, B. (2000). Judging a book by it’s cover: The relationship between service and safety quality in U.S. national and regional airlines. Journal of Air Transport Management, 6, 87–94. Rogerson, W. P. (1983). Reputation and product quality. Bell Journal of Economics, 14, 508–516. Rose, N. L. (1990). Profitability and product quality: Economic determinants of airline safety performance. Journal of Political Economy, 98, 944–964. Rose, N. L. (1992). Fear of flying? Economic analysis of airline safety. Journal of Economic Perspectives, 6, 75–94. Savage, I. (1999). Aviation deregulation and safety in the United States: The evidence after twenty years. In M. Gaudry & R. Mayes (Eds.), Taking stock of air liberalization. Norwell, MA: Kluwer Academic Publishers.

Competition on the Basis of Safety?

323

Shaked, A., & Sutton, J. (1982). Relaxing price competition through product differentiation. Review of Economic Studies, 49, 3–13. Shapiro, C. (1982). Consumer information, product quality, and seller reputation. Bell Journal of Economics, 13, 20–35. Shapiro, C. (1983). Premiums for high quality products as returns to reputations. Quarterly Journal of Economics, 98, 659–680. Spence, A. M. (1977). Consumer misperceptions, product failure and producer liability. Review of Economic Studies, 44, 561–572. Wong, J., & Yeh, W. (2003). Impact of flight accident on passenger traffic volume of the airlines in Taiwan. Journal of the Eastern Asia Society for Transportation Studies, 5, 471–483.