Investor Behavior in the Market for Bank-issued Structured Products [1 ed.] 9783896446964, 9783896446961, 9783896736963

This work investigates the trading behavior of retail investors in the largest European market place for bank-issued str

98 25 1MB

German Pages 188 [189] Year 2015

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

Investor Behavior in the Market for Bank-issued Structured Products [1 ed.]
 9783896446964, 9783896446961, 9783896736963

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Studienreihe der Stiftung Kreditwirtschaft an der Universität Hohenheim

Sebastian Schroff

Investor Behavior in the Market for Bank-issued Structured Products

Verlag Wissenschaft & Praxis

Investor Behavior in the Market for Bank-issued Structured Products

Studienreihe der Stiftung Kreditwirtschaft an der Universität Hohenheim Herausgeber: Prof. Dr. Hans-Peter Burghof

Band 51

Sebastian Schroff

Investor Behavior in the Market for Bank-issued Structured Products

Verlag Wissenschaft & Praxis

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.dnb.de abrufbar.

D100 ISBN 978-3-89673-696-3 © Verlag Wissenschaft & Praxis Dr. Brauner GmbH 2015 D-75447 Sternenfels, Nußbaumweg 6 Tel. +49 7045 930093 Fax +49 7045 930094 [email protected] www.verlagwp.de

Alle Rechte vorbehalten Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Druck und Bindung: Esser printSolutions GmbH, Bretten

5

Contents List of Figures

7

List of Tables

9

List of Abbreviations

13

1 Introduction 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Research Outline . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . .

15 15 17 20

2 Retail Investor Behavior 2.1 Noise Traders and Market Efficiency 2.2 Trading Motives . . . . . . . . . . . 2.3 Investor Psychology . . . . . . . . . 2.4 Welfare Evaluation . . . . . . . . . . 2.5 Do Retail Investors Move Markets? .

. . . . .

23 23 26 31 38 41

Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47 48 52 55

4 Data & Methodology 4.1 Retail Investor Trading . . . . . . . . . . . . . . . . . . . . 4.2 Financial Market Information . . . . . . . . . . . . . . . . . 4.3 Methodological Approach . . . . . . . . . . . . . . . . . . .

59 59 64 66

5 Retail Investor Information Demand 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .

69 69

3 The 3.1 3.2 3.3

. . . . .

. . . . .

Market for Bank-Issued Structured Market Design . . . . . . . . . . . . . . Product Design . . . . . . . . . . . . . . Pricing and Complexity . . . . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

6

Contents 5.2 5.3

5.4

5.5

Literature Review . . . . . . . . . . . . . . . . . . . . Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Information Demand and Information Supply . 5.3.2 Retail Investor Trading and Market Data . . . Empirical Results . . . . . . . . . . . . . . . . . . . . . 5.4.1 Time Series Properties of Information Demand Supply . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Correlation and Causality . . . . . . . . . . . . 5.4.3 Trading Activity . . . . . . . . . . . . . . . . . 5.4.4 Order Submission Strategies . . . . . . . . . . . 5.4.5 Positioning . . . . . . . . . . . . . . . . . . . . 5.4.6 Return Predictability . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . .

6 Media Sentiment and Leveraged Trading 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . 6.2 Literature Review . . . . . . . . . . . . . . . . . 6.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 News Data . . . . . . . . . . . . . . . . . 6.3.2 Trading Data . . . . . . . . . . . . . . . . 6.4 Empirical Results . . . . . . . . . . . . . . . . . . 6.4.1 Trading Intensity around News . . . . . . 6.4.2 Order Submission Strategies around News 6.5 Conclusion . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . and . . . . . . . . . . . . . . . . . . . . .

72 74 74 77 81 81 86 89 92 95 98 99

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

103 103 105 107 107 113 114 114 123 127

7 Leveraged Trading and Earnings Announcements 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 7.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Trading Data . . . . . . . . . . . . . . . . . 7.2.2 Earnings Surprise and Abnormal Returns . 7.3 Methodology . . . . . . . . . . . . . . . . . . . . . 7.3.1 Risk Appetite . . . . . . . . . . . . . . . . . 7.3.2 Positioning . . . . . . . . . . . . . . . . . . 7.4 Empirical Results . . . . . . . . . . . . . . . . . . . 7.4.1 Trading Activity . . . . . . . . . . . . . . . 7.4.2 Correlated Trading . . . . . . . . . . . . . . 7.4.3 Risk Appetite . . . . . . . . . . . . . . . . . 7.4.4 Positioning . . . . . . . . . . . . . . . . . . 7.4.5 Predictive Capabilities . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

129 129 132 132 133 135 135 138 139 139 142 145 152 156

Contents

7.5

7

7.4.6 Post-event Trading . . . . . . . . . . . . . . . . . . . 158 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

8 Conclusion 163 8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 8.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 8.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 Bibliography

167

9

List of Figures 1.1

Structure of the Thesis . . . . . . . . . . . . . . . . . . . . .

20

3.1 3.2 3.3 3.4

Monthly Number of Exchange Listed Products in Germany Market Volume of Structured Products in Germany . . . . Payoff Structure of Investment Products . . . . . . . . . . . Payoff Structure of Leverage Products . . . . . . . . . . . .

50 51 53 54

5.1

Illustration of Information Demand and Supply . . . . . . .

78

6.1 6.2 6.3 6.4

Number of Intraday News . . . . . . . . . . . . Sentiment Distribution of Intraday News . . . News Characteristics by Trading Hour . . . . . News Characteristics by Weekday . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

110 111 113 114

11

List of Tables 4.1 4.2

Illustration of Product Master Data . . . . . . . . . . . . . Illustration of Customer Trading Data . . . . . . . . . . . .

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9

Descriptive Statistics of Information Demand and Supply . 76 Descriptive Statistics of Trading in Structured Products . . 80 Variable Definitions . . . . . . . . . . . . . . . . . . . . . . 82 Time Series Properties of Information Demand and Supply 84 Correlations and Granger Causalities of Information Demand and Supply . . . . . . . . . . . . . . . . . . . . . . . . 88 Information Demand and Trading Activity . . . . . . . . . . 91 Information Demand and Order Submission Strategies . . . 94 Information Demand and Positioning . . . . . . . . . . . . . 97 Information Demand and Returns . . . . . . . . . . . . . . 100

6.1 6.2 6.3 6.4 6.5

Illustration of Dow Jones Newswires Data . Descriptive Statistics of News . . . . . . . . Descriptive Statistics of Trading in Leverage Leveraged Trading Intensity around News . Order Submission Strategies around News .

7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8

Decriptive Statistics . . . . . . . . . . . . . . . . . Trading Activity . . . . . . . . . . . . . . . . . . . Absolute Imbalances . . . . . . . . . . . . . . . . . Abnormal Risk Appettite . . . . . . . . . . . . . . Predictive Capabilities of Abnormal Risk Appetite Abnormal Imbalances . . . . . . . . . . . . . . . . Predictive Capabilities of Abnormal Imbalances . . Post-event Trading . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . Products . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

62 63

. . . . .

. . . . .

. . . . .

109 112 115 119 124

. . . . . . . .

. . . . . . . .

. . . . . . . .

134 141 144 147 149 154 157 159

13

List of Abbreviations ADF AIC AMEX CFD CAR DAX DJIA DJNS EPS EUSIPA EUWAX GDP HFT IBES ID IQ IS ISIN MDAX NYSE OTC PP RA RIC RNSE

Augmented Dickey-Fuller Aikaike Information Criterion American Stock Exchange Contract for Difference Cumulative Abnormal Return German Stock Market Index Dow Jones Industrial Average Dow Jones Newswire Service Earnings per Share European Structured Investment Products Association European Warrant Exchange Gross Domestic Product High Frequency Trading Institutional Brokers Estimate System Information Demand Intelligence Quotient Information Supply International Securities Identification Number German Midcap-Index New York Stock Exchange Over-The-Counter Phillips Perron Risk Appetite Reuters Instrument Code Reuters NewsScope Sentiment Engine

14 RV SIC SIRCA SIX TecDAX WSJ

Realized Volatility Schwarz Information Criterion Securities Industry Research Centre of Asia-Pacific Swiss Stock Exchange German Technology Companies Index Wall Street Journal

15

Chapter 1

Introduction Noise makes financial markets possible, but also makes them imperfect. Fisher Black (1986)

1.1

Motivation

Retail investors are currently facing challenging times. The global financial crisis, followed by the European sovereign debt crisis, has left many retail investors with little understanding and confidence in financial markets and their institutions.1 On top of that, retail investors are confronted with a steadily growing flood of information. This makes it difficult for them to keep track of financial market events and process relevant information. At the same time, households in industrialized countries would be well-advised to build private capital stock to ensure adequate pension provisions.2 However, persistently low interest rates in recent years and the 1

2

See, for example, Dorn and Weber (2013) who document that recent financial turmoils have contributed to a loss of confidence in financial intermediation and lower stock market participation rates as a result. See, for example, the OECD Social, Employment and Migration report by Antolin and Whitehouse (2009).

16

1.1. Motivation

resulting scarcity of traditional investment opportunities, makes this a difficult task. More generally, retail investors are increasingly required to deal with financial market topics, while the environment for this is becoming more complex and demanding. Hence, the behavior of retail investors is of major importance to financial research, policy makers, and financial institutions alike, since solutions for the challenges at hand require a thorough understanding of what drives financial market decisions of retail investors. To contribute to this field, my thesis analyzes the behavior of retail investors in the German market for structured financial products. These bank-issued structured products are securitized derivatives, specifically designed to grant retail investors access to a broad range of risk-return combinations in various underlyings. They range from conservative investment certificates to high-risk leverage products. Legally, all these products are bearer bonds payable by the issuer to the investor. With more than 1 million exchange-traded products and an outstanding market volume of EUR 100 billion in 2013, structured products play an important role in the asset allocation of German households: 2% of all monetary assets in Germany are invested in structured products.3 Previous work on structured products predominantly deals with the question, whether issuers fairly price their products.4 The question, how retail investors trade with structured products and whether they follow systematic behavioral patterns has not been addressed comprehensively. This is despite the fact that these products provide an ideal market environment to study the behavior and the trading motives of retail investors. The extensive universe of different products allows retail investors to implement much more diverse and sophisticated trading strategies as this is the case with traditional stock market investments. They enable retail investors to set up trading strategies that match their individual expectations at any given point in time. For example, it is not only possible to benefit from rising security prices, but also from sideways movements or falling prices. From a research perspective, this provides for a much more differentiated 3 4

Total monetary assets of German households amount to EUR 4,991.6 billion in Q1/2013. For details, see Deutsche Bundesbank (2013), p. 55. See, for example, Baule et al. (2008), Wilkens and Stoimenov (2007), and Entrop et al. (2009).

Chapter 1. Introduction

17

evaluation of retail investor behavior as would be possible in traditional stock markets. The particular product choice of an investor can shed light on the underlying trading motivation, market expectation and risk preference. For example, an investor who enters a position in a highly leveraged structured product clearly has a different motivation than an investor who buys an investment product with a conservative payoff structure.

1.2

Research Outline

I analyze trading behavior in structured products based on data from the EUWAX trading segment of Boerse Stuttgart, the largest market place for these products in Europe in terms of exchange turnover and number of listed products.5 I particularly focus on the effect of financial market information on retail investor trading behavior. The investigation, which kind of information retail investors take into account for their trading decisions and how they react to new information, can reveal systematic trading patterns and shed light on potential behavioral biases. Generally, research on the question how information translates into trading decisions is more relevant than ever since the universal internet adoption in recent years has greatly changed the financial information landscape. Nowadays, even retail investors have a wealth of information on almost any financial market topic conveniently available at their fingertips. Most likely, this has greatly affected information acquisition, information processing and financial decision-making. This raises the question, how retail investors use online sources as basis for their investment decisions. I contribute to this field with my first research question: Research Question 1: How does internet search volume on financial market topics affect the trading behavior of retail investors in structured products?

5

For details, see the EUSIPA Q2/2013 market report http://www.eusipa.org/images/inhalt bilder/13 09 02 EUSIPA Market Report Q2 2013.pdf.

at

18

1.2. Research Outline

I empirically analyze how the information demand of retail investors, as measured by investing-related Google search volume, affects various aspects of trading behavior in structured products. Specifically, I focus on its impact on order submission strategies, trading positions and returns of the underlying assets. I differentiate between stock-specific information demand and market-wide information demand. Further, I analyze how these two categories of information demand are related to different trading motives of retail investors (‘short-term speculation’ vs. ‘long-term investing’). Overall, the analysis aims to contribute to the question, how informationally efficient retail investors trade with structured products. Following up on the first research question, the rapid technological developments in information and communications technology have also greatly influenced the way how new information is disseminated to traders and at which speed it can be translated into trading decisions. News feeds and microblogging websites, such as twitter, allow retail investors to constantly access financial market information in real-time via mobile devices. Accordingly, the starting point of my second research question is the notion that the impact of information arrival on investor behavior has most likely undergone drastic changes in recent years. I intend to contribute to this field, by analyzing the following research question: Research Question 2: How does new information affect the intraday trading behavior of retail investors in leverage products? Specifically, I investigate whether trading activity in short-term oriented leverage products increases around the arrival of firm-specific news announcements. Further, I analyze whether retail investors have an increased desire for immediate order execution and intensified safety precautions, by implementing intelligent order types in response to new information. In this context, I place particular weight on the question, whether retail investors’ reactions differ, depending on the sentiment of news (positive, neutral, negative) and the type of news (e.g. analyst report vs. ad-hoc publication). The last research question focuses most strongly on behavioral trading patterns and the question, how well retail investors can anticipate new information and corresponding capital market reactions. Whereas the first

Chapter 1. Introduction

19

two research questions focus on information acquisition and the timely reaction to information, the third research question investigates the information content of retail investors’ risk preferences and their speculative bets. To do this, I analyze the following research question: Research Question 3: Do retail investors assume higher risks and speculate skillfully around earnings announcements? Specifically, I first investigate whether retail investors deliberately speculate on quarterly earnings announcements. Next, I analyze whether retail investors systematically enter similar trading positions around earnings announcements. Based on that, I test whether retail investors can correctly anticipate earnings surprises and the corresponding capital market reaction. In this context, I further investigate whether retail investors deliberately assume higher risks, by trading more highly leveraged products and whether such behavior is motivated by an informational advantage. Lastly, I investigate potential post-announcement patterns of retail investor trading. Overall, the analysis aims to identify behavioral biases and cognitive limitations that could be mirrored in retail investors’ trading decisions. Generally, this can provide insights on the question how sophisticated retail investors act on capital markets.

20

1.3. Structure of the Thesis

1.3

Structure of the Thesis

This thesis consists of 8 chapters. An overview of the structure is given in Figure 1.1. Chapter 2 provides an overview of related work on the behavior of retail investors. The focus is on typical trading motives and psychological explanations of systematic behavioral patterns. Further, I discuss whether retail investors can affect financial markets and how their trading decisions influence their welfare. Chapter 3 outlines the institutional design and the functioning of the market for bank-issued structured products. I explain essential characteristics of the structured product types used in the empirical analysis. Based on this, I discuss previous work that deals with the pricing and complexity of structured products. Chapter 1: Introduction Chapter 2: Retail Investor Behavior Chapter 3: The Market for Bank-Issued Structured Products

Empirical Analysis

Chapter 4: Data & Methodology Chapter 5:

Chapter 6:

Chapter 7:

Retail Investor Information Demand

Media Sentiment and Leveraged Trading

Leveraged Trading around Earnings Announcements

Chapter 8: Conclusion

Figure 1.1: Structure of the Thesis Chapter 4 gives an overview of the various data sets used in the thesis. The key data set consists of customer trading data and product master data from the EUWAX trading segment at Boerse Stuttgart during the sample period from April 2009 to November 2012. Further, I describe

Chapter 1. Introduction

21

the additional data sources that are used to derive the financial market information data sets used in the analysis. Lastly, the chapter outlines the methodological approach adopted to answer the research questions. The following three chapters consist of the empirical analyses, which are the center piece of the thesis. Chapter 5 deals with the first research question and analyzes how the information demand of retail investors affects their trading decisions in structured products.6 I first provide an overview of previous research on information demand on financial markets. The following section provides descriptive statistics of the information data sets and retail investor trading data. The main section reports the results on the intertemporal relationship between information demand and information supply and the effect of information demand on different aspects of retail investor trading behavior. Further, I study the asset pricing impact of information demand on the returns of the underlying stocks. The last section summarizes the main empirical findings. Chapter 6 focuses on the second research question and investigates the intraday relationship between the release of new information and speculative retail investor trading in leverage products.7 The first section provides an overview of previous research on investor behavior around news. Next, I describe my news and trading data. The main part of the chapter reports the empirical results of my analysis on information arrival and trading behavior in leverage products. The last section concludes. In Chapter 7, I address the third research question and analyze the risk preferences and the trading positions of retail investors in leverage prod-

6

7

This chapter is based on the joint paper “Retail Investor Information Demand: Speculating and Investing in Structured Products” with Hans-Peter Burghof and Stephan Meyer. The paper was presented at the 30th International French Finance Association Conference (AFFI), the 11th INFINITI Conference on International Finance, and the 2013 Asian Finance Association Annual Conference (AsianFA), and also accepted at the 53rd Southern Finance Association Annual Conference (SFA) and the 26th Australasian Finance and Banking Conference (AFBC). The paper has been accepted for publication in the European Journal of Finance. This chapter is based on the joint paper “Media Sentiment and Leveraged Retail Investor Trading” with Hans-Peter Burghof and Michael Siering. The paper was presented at the 26th Australasian Finance and Banking Conference (AFBC).

22

1.3. Structure of the Thesis

ucts in the context of quarterly earnings announcements.8 First, I provide descriptive statistics of the data sets used in the empirical analysis. Next, I describe the methodology adopted to measure retail investor risk appetite and retail investor imbalances. The main section reports the results on the different aspects of pre- and post event retail investor trading. Specifically, I investigate whether retail investors speculate on the outcome of earnings announcements, trade in the same direction, can correctly anticipate the capital market reaction to earnings announcements, and exhibit systematic post-event trading patterns. The last section concludes. Chapter 8 summarizes the main findings of the empirical analysis and discusses their implications. Finally, I give an outlook on future research questions in this field.

8

This chapter is based on the joint paper “Individual Investor Trading in Leverage Products - Risk Appetite and Positioning around Earnings Announcements” with Hans-Peter Burghof and Stephan Meyer. The paper was presented at the 62nd Midwest Finance Association Annual Conference (MFA), and the 16th Swiss Society for Financial Market Research Annual Conference (SGF), and also accepted at the 49th Eastern Finance Association Annual Conference (EFA), the 52nd Southwestern Finance Association Annual Conference (SWFA), and the 26th Australasian Finance and Banking Conference (AFBC).

23

Chapter 2

Retail Investor Behavior People who trade on noise are willing to trade even though from an objective point of view they would be better off not trading. Perhaps they think the noise they are trading on is information. Or perhaps they just like to trade. Fisher Black (1986)

2.1

Noise Traders and Market Efficiency

The debate how informationally efficient financial markets are continues to remain controversial and a final consensus has not been reached up to this point. A central aspect of this debate is the question, whether the sentiment of market participants that is not justified by fundamental information has the ability to affect market prices. This concept of trading on ‘noise’ was first introduced by Black (1986). The starting point of his argument is that very little trading would take place on stock markets in a world of rational expectations. The theoretical foundation of this so called ‘no-trade theorem’ is given by Milgrom and Stokey (1982) and Tirole (1982). The intuition behind their reasoning is that a rational investor should be suspicious that another rational investor, who offers to trade with

24

2.1. Noise Traders and Market Efficiency

him, does so because he is better informed. Further, as documented by Subrahmanyam (1991) and Gorton and Pennacchi (1993), non-speculative trading should not take place in individual securities, but in well-diversified portfolios such as stock index futures, due to lower adverse selection costs. However, given the massive amounts of trading volume on global financial markets, a world of rational expectations does not reflect empirically observed realities adequately.1 Based on this, Black argues for the existence of another type of trader, who does not act rationally. He contrasts noise from information and describes those market participants as uninformed traders, who “trade on noise as if it were information” and do not base their decisions on fundamentals.2 De Long et al. (1990a) are first to argue theoretically that the trading decisions of some investors are driven by sentiment that is not backed by relevant information. They demonstrate that these noise traders have the ability to affect market prices, because rational arbitrageurs have only limited abilities to force prices back to fundamental values.3 Shleifer and Vishny (1997) highlight that limits to arbitrage exist because betting against noise traders is costly and risky. For example, as famously stated by Keynes, “markets can remain irrational longer than you can stay solvent”, which eventually can force rational arbitrageurs to liquidate positions and realize losses. Recent stock market history provides numerous examples that financial markets can indeed remain in a state of irrational exuberance for extended periods of time (e.g. the dot-com bubble or the U.S. subprime crisis) and that arbitrage trading strategies can pose enor1

2 3

For example, the World Federation of Exchanges reports a yearly global exchange trading volume of USD 49 trillion (9.78 billion trades) in equities alone. See http://www.worldexchanges.org/files/statistics/pdf/2012%20WFE%20Market%20Highlights.pdf for details. Black (1986), p. 529. Milton Friedman famously argued that irrational noise traders cannot survive in competitive markets, because they keep losing money to rational arbitrageurs (Friedman, 1953). Consequently, they would disappear and not affect asset prices for extended periods of time. Figlewski (1979) softens Friedman’s reasoning and shows that it can take a very long time until noise traders loose all their wealth. In contrast, De Long et al. (1991) provide a framework, in which noise trader can survive in the long run if they hold portfolios with higher growth rates. More recently, Kogan et al. (2006) show that noise traders can even affect asset prices if they do not survive in the long run.

Chapter 2. Retail Investor Behavior

25

mous risks (e.g. the prominent collapse of the hedge fund Long-Term Capital Management). In the more recent literature, there is a growing consensus that noise trading can induce substantial price movements. Therefore, the discussion has shifted from the question, whether noise traders can affect financial markets, to the question how noise can be measured empirically and how the financial market impact can be quantified.4 Baker and Wurgler (2007) treat the origin of noise as exogenous and find for the U.S. market that investor sentiment affects both, individual stocks and the stock market as a whole. Giving support to the theoretical reasoning of previous work, the effects are found to be most pronounced for those stocks that are difficult to arbitrage (e.g stocks with a low market capitalization, highvolatility, etc.). Schmeling (2009) analyzes the effect of investor sentiment internationally and finds that sentiment negatively predicts stock prices across 18 industrialized countries. He confirms that the market impact of sentiment is strongest for stocks with limited arbitrage opportunities. Da et al. (2015) propose a novel measure of investor sentiment based on the Google search volume related to household concerns and also find that sentiment drives market prices. The effects are again most pronounced for stocks that are attractive to noise traders and difficult to arbitrage by rational investors. Finally, Bloomfield et al. (2009) find in an experimental setting that uninformed traders act as noise traders, who trade contrary to recent price fluctuations. Their trading activity is found to be beneficial to markets by increasing liquidity, but adversely affects the ability of market prices to incorporate new information. Finally, they find that a reduction of noise trading on financial markets does not increase the informational efficiency of market prices. Retail investors are usual suspects for being noise traders since they typically lack the sophistication and resources of professional investors. In fact, a large body of literature models retail investors investors as uninformed traders.5 To answer the question, whether retail investors are noise traders, it is necessary to take a detailed look at their trading motives, potential behavioral biases and their investment performance. 4 5

See Baker and Wurgler (2007), p. 130. See, for example, De Long et al. (1990a) and Barberis et al. (1998).

26

2.2

2.2. Trading Motives

Trading Motives

Under the assumption of rational expectations, investors only trade, if the costs incurred by trading are offset by the marginal benefits (that is, expected returns). Based on this reasoning, Grossman and Stiglitz (1980) argue that an informational advantage that rests on private information provides strong incentives to trade. Naturally, this trading motive applies to retail investors and professional investors alike. In addition, various other trading motives exist that can induce rational retail investors to trade. For example, following the life-cycle hypothesis of Ando and Modigliani (1963), retail investors might want to smooth consumption across their lifetime by investing their savings. Further, rebalancing portfolios to maintain the desired asset allocation, hedging against risks, or liquidating positions due to liquidity needs can be aligned with rational expectations. Another apparent trading motive for retail investors is tax considerations. Grinblatt and Keloharju (2004) and Barber and Odean (2004) show that taxes indeed play an important role in the trading decisions of retail investors (e.g. realizing losses towards the end of the year). However, as convincingly argued by Dorn and Sengmueller (2009) and Barber et al. (2009), the massive amounts of trading volume attributable to retail investors cannot be explained sufficiently by rational trading motives. Accordingly, behavioral explanations are required to comprehensively address the question, why people trade. Lately, various behavioral trading motives have been identified as important drivers of retail investor trading in the literature. Overconfidence The most prominent and well-established behavioral explanation for excessive retail investor trading is overconfidence. The intuition is that people commonly overestimate their own abilities and/or the precision of their own knowledge relative to others. Such beliefs are a widespread phenomenon in different aspects of life. For example, Svenson (1981) asks people in the U.S. to rank their own driving skills, relative to all other drivers. The striking result is that 93% of the participants judge themselves to have above-average skills (which, of course, can only be true for 50% of all drivers). In a similar setting, Cooper et al. (1988) find that

Chapter 2. Retail Investor Behavior

27

entrepreneurs greatly overestimate their chances of success relative to the prospects of similar businesses.6 Various theoretical papers have addressed investor overconfidence. Generally, these models rest on the assumption that investors cannot accurately assess their investment skills, in the sense that they think they are better informed than they actually are. For example, Gervais and Odean (2001) develop a multi-period model, in which traders ‘learn to be overconfident’. In the first period, the trader is unaware of his own abilities. In the following periods, he infers his ability from past success or failure. However, in contrast to models with true Bayesian updates such as, for example, Sharpe (1990) for long-term lending relationships, the trader suffers from a self-attribution bias. That is, he takes too much credit for his success (and too little for his failure). The resulting overconfidence induces the investor to trade aggressively, which in turn causes increased trading volume and market volatility, as well as lower expected returns for the overconfident investor. The survival of overconfident traders is not threatened in the model, because overconfident traders are wealthy due to their success in early periods. In sum, being successful fuels traders overconfidence, but overconfidence is not beneficial for traders wealth. Daniel et al. (1998) analyze overconfidence in a theoretical model and show that it induces investors to overreact to private information and underreact to public information. Further, Odean (1998) illustrates that overconfidence severely affects asset prices. The specific effect on asset prices varies, depending on which agents are overconfident and how information is disseminated. Further, the model shows that overconfidence leads to increased trading and reduces the welfare of overconfident agents.7 The theoretical predictions are largely in line with empirical work on overconfidence. Odean (1999) investigates 10,000 randomly selected customers at an U.S. discount broker and tests whether they trade excessively. He excludes trades that could be driven by rational motives, such as portfolio rebalancing, liquidity needs or tax-saving considerations and 6 7

See Moore and Healy (2008) for a more detailed description of overconfidence and an overview of the overconfidence literature in different social aspects. Further, theoretical literature on overconfidence includes Kyle and Wang (1997), Scheinkman and Xiong (2003), and Peng and Xiong (2006).

28

2.2. Trading Motives

nonetheless finds that retail investors trade excessively. Since the gains from trading are unable to cover the costs that arise from trading, he argues that overconfidence is a suitable explanation for excessive retail investor trading. Similarly, Barber and Odean (2000) analyze the trading activity of more than 60,000 households in the U.S. and argue that overconfidence can explain the large trading activity that is unlikely to stem from rational considerations. Based on previous psychological research that finds that men are more prone to suffer from overconfidence, Barber and Odean (2001) test, whether trading activity differs by gender. They find that the average equity turnover rate is around 1.5 times higher for men than for women. Since the resulting returns do not imply skillful trading, they again attribute excessive trading to overconfidence. Dorn and Huberman (2005) combine portfolio data of German investors with survey responses given by these investors. They find that investors with an above average level of self-perceived knowledge have a higher portfolio turnover. Glaser and Weber (2007) adopt a similar methodology and find that investors, who wrongly overestimate their abilities in terms of trading skills and past performance, trade more frequently. Finally, Grinblatt and Keloharju (2009) analyze the trading data of all households in Finland. They find that even after controlling for a great number of socioeconomic characteristics, overconfident investors trade more. Sensation-Seeking An alternative (non-competing) explanation for excessive trading by retail investors is already addressed in Black’s quote preceding this chapter. In plain words, retail investors might simply ‘like trading’. To put it more formally, retail investors might not only be motivated by the pecuniary benefits of trading such as expected returns, but also derive direct utility from the sensation offered by trading the stock market. A growing body of literature follows this line of thought and identifies entertainment motives as important driver of retail investor trading. Hoffmann (2007) performs a survey on Dutch retail investors, in which they are asked to indicate why they actively participate in stock markets. The trading motive “it is a nice free-time activity” is ranked second, just below “the potential for financial gains” but above “saving for retirement”.8 Dorn and Sengmueller (2009) match survey data with the trades of Ger8

Hoffmann (2007), p. 83.

Chapter 2. Retail Investor Behavior

29

man retail investors. In the survey, they ask investors to what extent they agree with the following statements: (1) “I enjoy investing”, (2) “I enjoy risky propositions”, (3) “Games are only fun when money is involved” and (4) “In gambling, the fascination increases with the size of the bet”.9 Retail investors who indicate agreement with these statements tend to trade more. Further, the impact of entertainment trading is economically significant, with trading activity being twice as high for retail investors who take pleasure from trading. Grinblatt and Keloharju (2009) use speeding tickets as a proxy for sensation-seeking, based on the notion that people who speed are generally more likely to derive pleasure from thrilling activities. They find that variations in speeding tickets can be positively related to variations in trading activity. For robustness, they also use sports-car ownership as proxy for sensation-seeking and obtain corresponding results. Gambling Another behavioral motive to explain trading in financial markets is gambling. That is, people (ab)use financial markets to live out their aspiration to be rich, by entering bets that potentially offer “dream values”.10 Lately, numerous papers have identified such gambling motives as important driver of retail investor trading. Shefrin and Statman (2000) highlight the role of gambling in a behavioral portfolio theory model, in which people combine a downside protection (an equity participation note with capital protection) with an upside potential (a ‘lottery ticket’) in their portfolio.11 Barberis and Huang (2008) show in a theoretical model that investors, who place too much weight on low-probability events (overweighting the tails of a distribution), highly value securities with positively skewed expected returns (frequent small losses and few extreme gains). Hence, they have a strong desire to invest in stocks that have similar characteristics than a lottery ticket. Kumar (2009) builds on the theoretical findings of Barberis and Huang (2008) and empirically analyzes ‘who gambles in the stock market’, by trading lottery-type stocks. He describes gambling as taking on high risks, without being rewarded by correspondingly high returns. Because investors 9 10 11

Dorn and Sengmueller (2009), p. 595. Statman (2002), p. 16. Generally, behavioral portfolio theory differs from modern portfolio theory in the sense that investors are not always risk averse and do not consider their portfolio as a whole. For details, see Shefrin and Statman (2000).

30

2.2. Trading Motives

overweight tail events, they are willing to accept this mismatch of risk and return. Lottery-type stocks are defined as those stocks that have low absolute prices, high idiosyncratic volatility and low idiosyncratic skewness. Kumar finds that, unlike professional investors, retail investors prefer to trade these kind of stocks. The gambling demand is particularly high during bad economic times. Also, various socioeconomic and psychological factors affect the proclivity to gamble.12 Han and Kumar (2013) provide further empirical evidence that stocks intensively traded by retail investors have lottery-like features. Further, these stocks are particularly attractive to retail investors with a strong propensity to gamble on stock markets. Several recent papers have adopted a substitution approach to identify gambling-driven trading. The starting point of these papers is the assumption that gambling on stock markets competes with other forms of gambling (e.g. lotteries, casinos). If this is the case, gambling-motivated trading should decrease (increase) if alternatives become more (less) attractive. Dorn et al. (2012) provide empirical evidence that supports this rationale for retail investors in the U.S. and Germany. All else being equal, the trading volume attributable to small trades significantly drops when the jackpot in the multi-state lotteries Powerball and Mega-Millions increases.13 In Germany, the trades at a discount broker show that retail investors are less likely to trade during weeks of high national lottery jackpots. The documented effects are found to be more pronounced for lottery-type securities. Further, less educated and male investors are more strongly affected. Similarly, Barber et al. (2009) find for the Taiwanese market that trading activity drops by 25% following the introduction of a government-sponsored lottery. Further, Gao and Lin (2012) provide results for the Taiwanese market that correspond to the findings of Dorn et al. (2012). Retail investor trading volume is found to decline around lotteries with exceptionally high jackpots. 12

13

Specifically, the gambling activity in the U.S. is most pronounced for poor, young men from urban areas, minorities, Catholics and Republicans. For details, see Kumar (2009), pp. 1910. In times of high frequency trading (HFT), which has brought down the average trade size, trade size does not allow for a meaningful classification of trade characteristics anymore. However, given the observation period from 1998 to 2004, small trade sizes can reasonably be used to proxy for retail investor order flow.

Chapter 2. Retail Investor Behavior

31

In sum, the literature on retail investors’ trading motives strongly suggests that behavioral factors such as overconfidence, sensation-seeking and gambling proclivities play an important role in explaining trading activity on financial markets.

2.3

Investor Psychology

The previous section has shown that the trading activity of retail investors is often driven by non-rational considerations. Building on this finding, the focus of this section is to shed more light on the question which role heuristics and cognitive biases play for retail investors’ trading decisions.14 Disposition Effect The disposition effect is one of the most well documented phenomenon driving the trading behavior of retail investors. It was first discovered by Shefrin and Statman (1985) and describes the tendency to sell winning stocks too early and hold on to loosing stocks for too long. Odean (1998) empirically tests the disposition effect on the trading records of 10,000 retail investors at a discount broker in the U.S. He finds that retail investors have a strong tendency to sell ‘winners’ and hold on to ‘losers’. Their behavior cannot be explained by rational trading motives, such as informed trading, since past winners continue to outperform past losers after they have been sold by retail investors. Further, neither portfolio rebalancing needs nor tax-considerations drive the results. Shapira and Venezia (2001) find a disposition effect for Israeli investors and show that as a result, stock holding periods are generally lower (higher) for winners (losers). For the Finish market, Grinblatt and Keloharju (2001b) also find strong evidence that retail investors are reluctant to realize losses. Feng and Seasholes (2005) document a disposition effect among Chinese retail investors and show that it decreases with investor sophistication and trading experience. Seru et al. (2010) also highlight the role of ‘learning from trading’ and find that the disposition effect decreases with 14

The focus of this section is on biases that are of particular relevance to the empirical analysis performed in this thesis. It does by no means attempt to provide a complete overview of all heuristics and cognitive biases, which is beyond the scope of this thesis.

32

2.3. Investor Psychology

trading experience. Grinblatt et al. (2012) also report a disposition effect and document that it decreases with retail investors’ IQ. This finding supports the notion that cognitive abilities are an important determinant of investor behavior. Finally, Barber et al. (2007) report a substantial disposition effect in the Taiwanese market, with retail investors being four times more likely to sell winners than losers. Also, they show that retail investors are much more subject to the disposition effect than professional investors. This difference in the behavior of professional and retail investors is further backed by Brown et al. (2006), Chen et al. (2007), and Choe and Eom (2009). Altogether, the literature provides strong support that retail investors suffer from a disposition effect. The finding that the effect is most pronounced for unsophisticated retail investors with little experience, rather than professional investors, suggests that it stems from a lack of cognitive abilities and/or financial literacy. However, the question which specific heuristics and cognitive biases drive the disposition is subject of ongoing research. Shefrin and Statman (1985) and Grinblatt and Keloharju (2001b) argue that the disposition effect is an application of the seminal prospect theory, originally developed by Kahneman and Tversky (1979). In various experiments, Kahneman and Tversky find that people value gains and losses differently. For example, in Tversky and Kahneman (1981), they asked people to choose from a sure gain of USD 240 or a risky payoff with a 25% chance to gain USD 1,000 or nothing with a 75% chance. Most people (84%) decide for the sure gain, which implies that we are less likely to take on risks for additional gains. In contrast, when asked whether they would prefer a sure loss of USD 750 to a 75% chance of losing USD 1,000 and a 25% chance of losing nothing, most people (87%) opt for the second option. This, in turn, indicates that people are more likely to assume risks to avoid losses.15 Applied to financial markets, their findings support the notion that investors are subject to ‘loss aversion’.16 15 16

See Tversky and Kahneman (1981), p. 454. Loss aversion describes the tendency to prefer avoiding losses to making gains. For details on the effects of loss aversion on financial markets, see, for example, Benartzi and Thaler (1995) and Thaler et al. (1997).

Chapter 2. Retail Investor Behavior

33

However, there is an ongoing debate, particularly in the theoretical literature, whether prospect theory actually predicts the disposition effect. Kaustia (2010) finds that prospect theory is unlikely to explain the disposition effect since it can induce investors to hold on to both, winners and losers. Barberis and Xiong (2009) develop a theoretical model, in which investors are assumed to have prospect theory preferences. They show that prospect theory does not necessarily result in a disposition effect. In fact, prospect theory can even result in a reversed disposition effect. In contrast, Henderson (2012) illustrates by means of an optimal stopping model, in which investors voluntarily sell loser, that prospect theory implies a disposition effect.17 Hens and Vlcek (2011) theoretically show that investors with prospect theory preferences would refrain from trading in the first place. Based on this, they argue that prospect theory can explain the disposition effect ex-post, after the investment has been made, but not ex-ante. Li and Yang (2013), provide a general equilibrium model on the question and find that the underlying link between prospect theory and the disposition effect is a time-varying risk-attitude story.18 Finally, Yao and Li (2013) propose a model in which two types of investors interact: investors with prospect theory preferences and investors with standard constant risk aversion. Investors with prospect theory preferences tend to suffer from the disposition effect and act as contrarians, by implementing negative feedback strategies. In the following, I will address to what extend such contrarian behavior can describe retail investor trading. Contrarian Behavior The literature widely agrees on the notion that retail investors have a tendency to act as short-term contrarian traders.19 That is, they buy stocks after price drops and sell stocks after price increases. Choe et al. (1999) use daily imbalance measures to identify ag17

18

19

Optimal stopping theory deals with the problem of choosing the right timing for a particular action, to maximize (minimize) an expected payoff (cost) based on sequential random variables. A famous application of optimal stopping is the so called ‘secretary problem’. For details, see Ferguson (1989). Specifically, they argue that investors with prospect theory preferences are loss averse (that is, they are more sensitive to losses than to gains) and display a diminishing sensitivity (that is, they are risk-seeking for losses and risk-averse for gains). The link between prospect theory preferences and the disposition effect thus depends on the interaction between loss aversion and diminishing sensitivity. For further details, see Li and Yang (2013), p. 717. See Kaniel et al. (2008).

34

2.3. Investor Psychology

gregate trading patterns of different investor groups. They find that Korean retail investors act as negative feedback traders (contrarian traders), whereas professional investors are positive feedback traders (momentum traders). Jackson (2003) also finds contrarian short-term trading patterns of retail investors in the Australian market. Grinblatt and Keloharju (2000) and Grinblatt and Keloharju (2001b) analyze Finish data and provide evidence that retail investors display both, short-term and long-term contrarian tendencies. They also find that negative past returns affect buy-sell decisions more strongly than positive past returns. For the U.S. market, Goetzmann and Massa (2002) find that retail investors investing in index funds are more often contrarian traders instead of momentum traders. Particularly, active traders are typically contrarian traders, whereas less active traders tend to be momentum traders. Griffin et al. (2003) provide further evidence for the U.S. and report that retail investors trading NASDAQ 100 stocks intensively sell those stocks that have outperformed the market on the previous trading day. Finally, Kaniel et al. (2008) analyze retail investor trades at the New York Stock Exchange (NYSE). They find intense buying (selling) of retail investors for stocks that have experienced large price decreases (increases) in the previous month. They further show that this contrarian behavior positively predicts abnormal returns, which they attribute to a compensation for providing liquidity to professional investors that demand immediacy. Whereas empirical observations provide strong evidence that retail investors tend to act contrarian on financial markets, the answer to the question which motives drive this behavior is less clear. One potential explanation is that retail investors believe in mean-reversion of security prices. Hence, after periods of rising prices, they expect prices to revert to previous levels.20 A closely related explanation is the ‘Gambler’s Fallacy’, initially proposed by Tversky and Kahneman (1971). It describes the mistaken belief that deviations from what happens on average will be corrected in the short-term. As Tversky and Kahneman put it in the context of playing roulette: “After observing a long run of red on the roulette wheel, for example, most people erroneously believe that black will result in a more 20

See Odean (1998) for empirical evidence and Andreassen (1988) for experimental evidence on mean-reverting beliefs.

Chapter 2. Retail Investor Behavior

35

representative sequence than the occurrence of an additional red.”21 Applied to stock markets, investors might wrongly assume that, for example, after a stock has exhibited a succession of negative returns, the likelihood of subsequent positive returns is increased. If retail investors act on such beliefs, the result would be a contrarian trading strategy. Hence, the empirically observed contrarian trading activity might be motivated by the Gambler’s Fallacy. Poorly Diversified Portfolios To minimize idiosyncratic risk, retail investors would be well-advised to hold diversified portfolios. However, empirical evidence predominantly suggests that retail investors hold very poorly diversified portfolios. Barber and Odean (2000) report that between 1991-1996 retail investors in the U.S., on average, only hold four stocks in their portfolio. Building on the same data set of 78,000 household portfolios, Goetzmann and Kumar (2008) analyze portfolio choices in more detail. They confirm that retail investors are poorly diversified. The level of under-diversification is more pronounced for young investors, low-income investors, less-educated investors, and investors who are less financially sophisticated (as measured by various proxies, such as investment experience or foreign portfolio holdings). Further, overconfident investors and investors with a preference for high volatility and skewed returns display lower levels of diversification. Whereas the average number of stocks in investors’ portfolios increases during the sample period, no actively improved portfolio composition, by means of adding less correlated stocks, is observed. Generally, the high idiosyncratic risks assumed by retail investors are not compensated by equally high returns and thus negatively affect their wealth. Mitton and Vorkink (2007) develop a model in which investors have heterogeneous preferences for stocks with skewed returns. They predict that this allows investors to hold poorly diversified portfolios in equilibrium. They test their model predictions based on the same data as Barber and Odean (2000) and argue that the small number of portfolio holdings can be attributed to the desire for skewness exposure. In addition, they show that the stocks held by badly diversified investors have higher idiosyncratic skewness than those held by better diversified investors. Lastly, Grinblatt et al. (2011) show that retail investors with a 21

Tversky and Kahneman (1974), p. 1125.

36

2.3. Investor Psychology

higher IQ tend to be better diversified (by holding more stocks and mutual funds) and realize higher Sharpe ratios.22 These findings again highlight the importance of cognitive skills for investment success. Another commonly observed phenomenon is that retail investors intentionally invest a large fraction of their retirement savings in their employers’ stock. A famous example that highlights the potentially severe consequences of such behavior, is the Enron scandal at the beginning of the century. On average, Enron employees had allocated 62% of their retirement plan holdings to company stock in 2000.23 When the company went bankrupt in 2001, employees lost substantial amounts of their retirement savings. While the fraction of U.S. pension plan funds allocated to company stock has dropped significantly in recent years, many investors continue to remain highly exposed.24 Empirical evidence on what drives the excessive fund allocation of retail investors to company stock is reported by Benartzi (2001), Huberman (2001) and Huberman and Sengmueller (2004). Diversification of retail investors’ portfolios is often impeded by a preference to hold familiar stocks. As documented by numerous studies, retail investors tend to avoid foreign stocks (‘home bias’) and hence miss out on the benefits of cross-border diversification.25 Although there has been a trend towards more international diversification, investors’ portfolios remain heavily biased towards domestic stocks in many countries around the globe.26 The same bias also applies to domestic holdings, where investors typically prefer local stocks (‘local bias’). Ivkovic and Weisbenner (2005) and Seasholes and Zhu (2010) find that geographically near stocks 22 23 24

25 26

The Sharpe ratio measures the risk-adjusted excess performance of an investment. For details, see Sharpe (1994). For details, see the Congressional Research Service Report on the matter at http://fpc.state.gov/documents/organization/9102.pdf. For details, see the 2011 Employment Benefit Research Institute report at http://www.ebri.org/pdf/briefspdf/EBRI IB 12-2012 No380.401k-eoy2011.pdf, which reports a drop from 19% in 1999 to 8% in 2011. See Karolyi and Stulz (2003) and Lewis (1999) for overviews of the extensive literature on the home bias phenomenon. See, for example, Solnik and Zuo (2012), p. 283, who report that foreign holdings account for 35% of investors’ portfolios in developed markets and only 5% in emerging markets.

Chapter 2. Retail Investor Behavior

37

are overrepresented in the portfolios of U.S. retail investors. Grinblatt and Keloharju (2001a), Massa and Simonov (2006), Baltzer et al. (2013) provide similar findings for Finland, Sweden and Germany. Taken together, retail investors are subject to various biases, such as preferences for skewed returns and a desire to hold familiar stocks, which prevents them from holding well-diversified portfolios. Further, limited cognitive abilities and a lack of financial sophistication can adversely affect portfolio diversification. Limited Attention Retail investors’ primary occupation is typically not trading. As a result, they tend to have limited amounts of attention they can direct at stock markets. As a consequence, they are more likely to miss relevant information and thus be at an informational disadvantage in comparison to professional investors. In the light of an ever increasing flood of information, filtering relevant information is becoming more challenging. This is again particularly the case for retail investors, who typically have less resources (e.g. access to detailed financial data services that assist in information gathering and processing) than professional investors. Recent empirical evidence supports the view that investors miss important financial market events, because of distraction. Hirshleifer et al. (2009) find that the market reaction to earnings surprises is smaller and the post earnings announcement drift is more pronounced for firms that issue earnings announcements on the same day as many other firms. Adopting a similar approach, Dellavigna and Pollet (2009) find that the response to earnings announcements issued on Fridays is weaker and the following drift is more pronounced, because attention is limited just before the weekend. Barber and Odean (2008) believe that the limited attention retail investors can devote to the stock market poses a huge problem when making investment decisions. They argue that retail investors have a tendency to invest in those stocks that first catch their attention (e.g. in the news or via social media channels), because they do not follow a systematic stock selection process. In support of this notion, the results of their empirical analysis show that retail investors heavily buy attention-grabbing stocks. Seasholes and Wu (2007) confirm that retail investors are more likely to buy stocks that exhibit attention-grabbing events and document that pro-

38

2.4. Welfare Evaluation

fessional investors systematically take advantage of attention-driven retail investors. A more detailed discussion on (limited) attention on financial markets is given in Chapter 5, which empirically analyzes attention effects of retail investors in the market for structured products.

2.4

Welfare Evaluation

The previous sections have shown that retail investors often trade for irrational reasons and suffer from a broad range of cognitive limitations and biases. The question is, how these ‘shortcomings’ translate into welfare gains or losses that arise from stock market participation. The answer to this question is not straightforward. One way of answering it would be to simply look at the trading performance of retail investors (after transaction costs), to determine if they lose or gain money from trading. However, such a narrow approach would not address the question adequately, since welfare ultimately depends on the individual utility functions of retail investors. Hence, factors such as additional costs associated with trading (e.g. opportunity costs caused by information gathering, processing and decisionmaking) and non-pecuniary benefits (e.g. excitement and entertainment) also need to be taken into account to provide a complete picture. However, because of difficulties to capture (or model) individual utility functions of retail investors accurately, most studies simply use performance measures as proxy for welfare. Aggregate Performance Much of the empirical evidence implies that retail investors perform poorly on aggregate. Barber and Odean (2000) find for the U.S. that the average retail investor underperforms the market by 1.5% per year. The annual underperformance of those investors that trade most actively during the sample period from 1991-1996 is even more pronounced with 6.5%. Based on their findings, Barber and Odean conclude that retail investors trade on noise and hence, ‘trading is hazardous to their wealth’. Coval et al. (2005) analyze the same data set from a different perspective and report that only about 5% of retail investors can beat the market. Put differently, the overwhelming majority of retail investors performs poorly. Barber and Odean (2008) also document economically

Chapter 2. Retail Investor Behavior

39

large aggregate losses of individual investors for the Taiwanese stock market. Specifically, retail investors underperform the market by 3.8% on an annual basis. The losses are equivalent to 2.2% of GDP or 2.8% of peoples’ income. Further, nearly all losses incurred by retail investors arise from aggressive orders and therefore deliberately placed trades. In contrast, professional investors outperform the market by 1.5% per year and both aggressive and passive orders are profitable.27 Similar evidence of poor gross returns earned by retail investors is reported by Grinblatt and Keloharju (2000) for the Finish market. The findings for alternative asset classes and trading segments provide similar results. Bauer et al. (2009) analyze retail investor option trades at a discount broker in the Netherlands from 2000-2006. They find that retail investors incur substantial losses from trading options (raw monthly returns of -1.8%), that greatly exceed the losses from trading stocks (raw monthly returns of -0.6%). They attribute the poor trading performance to bad market timing and high trading costs. Further, hedging considerations only play a negligible role and therefore entertainment and gambling motives are mentioned as the most likely drivers of option trading by retail investors. For the German market of bank-issued structured products, the findings of Meyer et al. (2014) show that retail investor trades in shortterm oriented leverage products are characterized by poor risk-adjusted and predominantly negative returns. Similar findings are documented for more conservative structured products by Entrop et al. (2014). Cross-Sectional Performance Variation Looking at the aggregate performance provides important insights on the ‘average’ retail investor. However, it cannot shed light on the question whether there is a variation in performance depending on personal characteristics. Whether this is the case and which characteristics are important for being a ‘better investor’ will be addressed in the following. First evidence that characteristics exist, which make people trade more successfully, are analyzed on an abstract level by Coval et al. (2005). Their analysis reveals that performance among individual investors is persistent. That is, investors that have performed well in the past continue to per27

For details on the order classification, see Barber and Odean (2008), pp. 615.

40

2.4. Welfare Evaluation

form well in the future. Although, after transaction costs, even better investors exhibit negative returns, the results suggest that trading skills vary considerably between investors. Following a similar approach, Barber et al. (2013) investigate performance differences of Taiwanese day traders. Specifically, they rank investors according to their performance in one year and analyze their performance in the consecutive year. Their findings show that the 500 best performing traders outperform their peers by 0.6% per day. With daily abnormal returns of around 0.5%, the best traders earn such high returns that they are likely to be profitable even after covering transaction costs. Given this magnitude of daily abnormal returns, the findings provide strong support that some traders have superior trading skills. Recent findings suggest that cognitive abilities can contribute to explaining variations in retail investor performance. Grinblatt et al. (2012) document that Finish investors with a high IQ outperform investors with a low IQ by 2.2% per year. Korniotis and Kumar (2013) argue that the trading decisions of ‘smart’ investors are likely to be informed, whereas the trading decisions of ‘dumb’ investors are more likely to be driven by behavioral biases.28 In their analysis, they use a demographics based proxy for ‘smartness’ that includes factors such as education, wealth and income.29 Smart investors are found to overperform dumb investors by about 3% per year on a risk-adjusted basis. Further, the returns after transaction costs largely correspond to market returns. This implies that smart investors trade skillfully, but only well enough to cover transaction costs. Dumb investors significantly underperform the index, which is equally attributed to poor stock-selection abilities and transaction costs. In a related paper, Korniotis and Kumar (2009) investigate the relationship between age and investing performance. The underlying motivation is that age could either positively affect performance because of greater experience or negatively affect performance because of declining cognitive abilities. They find that deteriorating cognitive abilities dominate positive effects of in28

29

Korniotis and Kumar (2013) put forward various reasons why smart investors tend to be better informed: Better access to information, better information gathering and processing skills, better learning abilities, better analytical abilities, more able to adapt. For details, see Korniotis and Kumar (2013), p.3. For details, see Korniotis and Kumar (2013), p. 246.

Chapter 2. Retail Investor Behavior

41

creased investment knowledge. Risk-adjusted returns are 3-5% lower for older investors than for the average investor. Barber and Odean (2001) focus on gender differences in the U.S. They argue that men are more prone to being overconfident and therefore likely to trade excessively. This in turn adversely affects their performance. Their empirical findings are in line with their predictions: Men trade more excessively and perform worse as a result. Further, neither men nor women are found to trade skillfully and exhibit poor overall performances. Dorn and Huberman (2005) provide similar findings for the German market. In contrast, no gender specific differences in performance are found by Feng and Seasholes (2008) for the Chinese market. In sum, the evidence predominantly suggests that trading is detrimental to retail investors’ wealth. On aggregate, retail investors seem unable to beat the market. Only a very small minority of smart traders is capable of outperforming the market. However, after transaction costs, the monetary benefits of speculative trading are generally limited. Nonetheless, this does not justify jumping to the conclusion that trading also reduces investor welfare. As suggested by Black (1986), this question can only be answered if non-pecuniary benefits of trading are incorporated in the utility function of retail investors. Since adequately doing this poses a great challenge to both, theoretical and empirical research, the question how trading affects retail investors welfare remains unanswered in the literature as of now.

2.5

Do Retail Investors Move Markets?

The previous sections have shown that retail investors tend to base their trades on irrational motives, are subject to a broad range of cognitive limitations and behavioral biases, and tend to loose money from trading. In sum, this supports the notion that retail investors are noise traders. As outlined in the first section of this chapter, noise traders can affect financial markets and asset prices substantially. Based on this starting point, this section addresses the question, whether and how retail investors move markets. Given that retail investor trades are typically small, they can only affect security prices if they systematically point in the same

42

2.5. Do Retail Investors Move Markets?

direction (‘correlated trading’). If this necessary condition is not met, retail investor trades cancel each other out and do not affect financial markets. Correlated Trading Barber et al. (2009) analyze retail investor trades at two different brokerages in the U.S. and find for both data sets that their trading activity is highly correlated and persistent. Psychological biases such as the disposition effect, limited attention and representativeness heuristic are identified as drivers behind correlated trading.30 In contrast, explanations that have been put forward for institutional herding are not found to apply to retail investors.31 Kumar and Lee (2006) provide additional evidence that the trades of U.S. retail investors are systematically correlated. Dorn et al. (2008) analyze whether retail investors at a German discount broker tend to be on the same side of the market on a given day, week, month or quarter. They find that retail investors’ trades are generally correlated, with 57% of investors trading in the same direction in a typical stock and quarter (by chance it would be 50%). Further, they distinguish between speculative trading (driven by perceived information about future price movements) and non-speculative trading (motivated by investing savings, liquidity needs, and risk management). For example, trades that stem from automatic saving plans that gradually invests in a specific stock or fund, are classified as non-speculative. Since non-speculative trades are likely to be either explicitly (e.g. pre-specified investment dates for automatic savings plans) or implicitly (e.g. paycheck clustering at monthend) coordinated, they can be expected to be correlated. Interestingly, however, the findings suggest that correlated trading is substantially (up to 50%) higher for speculative trading than for non-speculative trading. This strongly implies that correlated trading is predominantly driven by retail investors placing similar bets at the same time.32 Further, Dorn 30

31 32

The concept of ‘representativeness heuristic’ was first proposed by Kahneman and Tversky (1972) and describes the cognitive bias that people determine the probability of an event based on beliefs or previous experiences and hence fail to make accurate predictions. See Tversky and Kahneman (1982) for an illustrative example (‘taxicab problem’). Institutional herding is found to be driven by momentum trading (Nofsinger and Sias, 1999) and information inference from each others trades (Sias, 2004). See Dorn et al. (2008), p. 886.

Chapter 2. Retail Investor Behavior

43

et al. (2008) find that correlated trading activity is higher when more retail investors trade and when market-wide trading volume is higher. Also, correlated trading tends to be higher in large capitalized stocks. The correlated trading activity is not found to be motivated by an informational advantage on behalf or retail investors. Market Impact Given that retail investor trading is correlated, they can affect financial markets in various ways. First, they can play a role in price formation by supplying or demanding liquidity to other market participants. Second, they can drive prices in the direction of their trading decisions - even if they are noise traders (as predominantly suggested by their trading performance) - as long as there are limits to arbitrage and rational investors cannot sufficiently bet against them. Kumar and Lee (2006) provide evidence that the correlated trades of retail investors can explain co-movements of stocks that are intensively traded by retail investors (e.g. stocks with a small market capitalization, low absolute prices, low institutional ownership). In line with the predictions of ‘noise trader’ models, the effect is particularly strong for stocks that are difficult to arbitrage. Taken together, this suggests that retail investors’ trading activity affects return movements. Barber et al. (2009) analyze data from brokerage accounts as well as retail investor trades at NYSE, AMEX and NASDAQ (proxied by small trade size). First, they confirm that retail investors systematically trade in the same direction. Second, they find that retail imbalance measures (that is, their net buying/selling) positively predict cross-sectional stock returns in the short-run (weekly and monthly), but negatively in the longrun (yearly). Similar to Kumar and Lee (2006), the results are strongest for difficult-to-arbitrage stocks and those most actively traded by retail investors. In sum, the findings provide strong support to the notion that retail investors affect stock prices. Following similar methodology, Hvidkjaer (2008) backs the storyline of Barber et al. (2009), by showing that U.S. stocks intensely bought by retail investors outperform in the short-run and subsequently underperform in the long-run. Kaniel et al. (2008) analyze retail investor trading from 2000-2003 in the cross-section of stocks traded on NYSE. They find that intense retail

44

2.5. Do Retail Investors Move Markets?

investor buying (selling) is followed by positive (negative) short-term abnormal returns. They argue that these patterns can be explained by a compensation for providing liquidity to professional investors who demand immediacy. Dorn et al. (2008) differentiate between market orders and limit orders submitted by retail investors. They argue that this differentiation is essential to understanding the relationship between retail investor trading and stock prices.33 Market order imbalance measures positively predict future returns, which is attributed to persistent speculative price pressure. Limit order imbalances also predict positive returns, which can be interpreted as a compensation for liquidity provision of executed limit orders. In sum, albeit for different reasons, both order types are shown to affect stock prices. Kaniel et al. (2012) find that retail investors positively predict stock returns. Specifically, they find that stocks intensely bought by retail investors in the 10 days prior to earnings announcements outperform those intensely sold by 1.5% in the two days around the announcement. They attribute these abnormal returns in equal shares to informed trading and liquidity provision. Kelley and Tetlock (2013) find that daily order imbalances of retail investors predict positive abnormal stock returns at the monthly horizon. They find that only market orders correctly predict firm news and conclude that aggressive retail investor trading is informed, whereas passive retail investor trading provides liquidity when it is scarce. Both actions are found to have a positive impact on financial market efficiency. Han and Kumar (2013) examine the asset pricing impact of speculative retail investor trading in U.S. stocks. They provide evidence that stocks predominantly traded by retail investors tend to be overpriced and exhibit negative abnormal returns. This mispricing is not reduced by rational investors since the lottery-like characteristics (high skewness, high volatility, low prices) of stocks with a high proportion of retail traders make arbitrage costly. Hence, speculative retail trading is identified as an important driver of asset prices. Finally, Foucault et al. (2011) show that the trading activity of retail investors has a positive effect on stock market volatility. 33

The reason for this is that stale limit orders are often executed mechanically and hence tend to be correlated. For further details on this effect, see Linnainmaa (2010).

Chapter 2. Retail Investor Behavior

45

Based on this finding, they argue that retail investors are noise traders who do not trade on fundamentally relevant information. Collectively, the findings imply that the systematically correlated behavior of retail investors moves markets. The documented effects on liquidity, volatility and returns clearly highlight that retail investor sentiment plays an important role for price formation and asset prices. Hence, a detailed understanding of what exactly drives retail investor behavior is an important piece to a better understanding of how financial markets function. Further, understanding how behavioral biases and emotions affect markets plays an important role for questions of financial market regulation and ultimately financial stability.

47

Chapter 3

The Market for Bank-Issued Structured Products A major source of objection to a free economy is precisely that it ... gives people what they want instead of what a particular group thinks they ought to want. Underlying most arguments against the free market is a lack of belief in freedom itself. Milton Friedman (2000)

This controversial quote by Milton Friedman is at the center of a current debate in many countries on how to regulate structured financial products distributed to retail investors. Should there be unrestricted access for retail investors to all kinds of structured products drafted by investment banks or is there a need to ‘protect’ retail investors from certain product types? Following the reasoning of Merton (1990), the fundamental argument in favor of a ‘free market’ for structured products is that they provide a useful extension to traditional capital markets, by broadening the range of investment opportunities. The basic argument for a more interventionist approach, including product bans, is that ‘irrational’ retail investors need

48

3.1. Market Design

to be protected from too complex structured products, which they are unable to understand. Against this background, the focus of this chapter is to introduce basic aspects of the market for structured products, such as the market design and important product characteristics. Further, I outline why the market for structured products provides a uniquely well-suited environment to study the behavior of retail investors. Finally, the chapter gives an overview of related work that deals with pricing and complexity issues of structured products.

3.1

Market Design

Structured products are securitized derivatives issued by investment banks. According to the World Federation of Exchanges, a structured product is a “tradable financial instrument designed to meet specific investor needs and to respond to different investment strategies, by incorporating special, non-standard features”.1 The redemption value of structured products is linked to the performance of one or more underlying assets in a prespecified manner. In other words, the issuer contractually commits to a payoff, which depends on a pre-defined set of circumstances. Legally, structured products are debt obligations (‘bearer bonds’) payable by the issuer to the investor. If the issuer is unable to service its debt, the retail investor may suffer a total loss (default risk ).2 Structured products are specifically designed for retail investors and grant them access to complex option positions. Due to high commission and transaction costs, a direct participation in option markets is unfeasible for most retail investors. Hence, structured products provide an extension to traditional capital markets. They offer a large variety of risk-return profiles and therefore facilitate retail investors’ access to sophisticated trading 1 2

http://world-exchanges.org/statistics/statistics-definitions/securitized-derivatives. This is what happened to many retail investors in Germany when Lehman Brothers filed for bankruptcy in 2008. For details, see for example http://www.ftd.de/finanzen/derivate/:bgh-urteil-lehman-opfer-gehen-leeraus/60109513.html.

Chapter 3. The Market for Bank-Issued Structured Products

49

strategies. From a research perspective, the large product universe provides for a detailed analysis of retail investor behavior. First, the specific product choice that is reflected in each order implicitly provides information about the trading motive of the investor behind the order. For example, an order for a highly leveraged product is likely to be motivated by the desire to speculate (or gamble) on short-term price fluctuations in the underlying instrument. In contrast, an investor who submits an order for an investment product with a much more conservative payoff profile probably intends to invest money and benefit from medium to long-term price movements of the underlying. Also, structured products can provide a more complete picture regarding investors market expectations, because they allow for an accurate implementation of the desired risk-return profile. For example, investors who expect market prices for a specific underlying to decline greatly can buy a product that benefits strongly from falling prices. As argued by Barber and Odean (2008), retail investor trades in stocks are biased because of an imbalance in buying and selling opportunities. Specifically, retail investors can buy all available stocks to implement positive market expectations, but can only sell the stocks in their portfolio to implement negative market expectations. This limitation does not apply to structured products. More generally, trading in structured products allows to investigate questions that cannot be addressed when looking at retail investor trading in ordinary stocks. The empirical analysis of this thesis focuses on the German market of structured products, which is the most established national market of its kind in Europe. In Germany, structured products play an important role in the asset allocation of retail investors: In 2012, the money invested in structured products amounted to EUR 100.7 billion, which is equivalent to 13.8% of investment fund holdings and 2% of all monetary household assets in Germany.3 Generally, structured products are more widely used in European countries than in other markets. With a global market share of 64%, the European market for structured products is by far the largest 3

Total monetary assets of German households amount to EUR 4,991.6 billion in Q1/2013. For details, see Deutsche Bundesbank (2013), p. 55. Total public investment fund holdings amount to EUR 729.9 billion. For details, see the numbers provided by the German Investment Funds Association at http://www.bvi.de/statistik/.

50

3.1. Market Design

worldwide. Still lacking behind in absolute size, markets in the U.S. and Asia are developing fast, with growing global market shares.4

Number of Structured Products

The product universe available to retail investors in Germany has grown rapidly from less than 200,000 products in 2007 to more than 1,000,000 products in 2013 (see Figure 3.1).5 1,000,000 800,000 Leverage Products

600,000 400,000

Investment Products

200,000 0 Jan 07

Jan 08

Jan 09

Jan 10

Jan 11

Jan 12

Jan 13

Figure 3.1: Monthly Number of Exchange Listed Products in Germany This figure shows the monthly number of structured products listed in Germany from January 2007 to June 2013.

These product are issued by 16 different investment banks, with Deutsche Bank (16.5%) and Commerzbank (15.9%) holding the largest market shares.6 4 5

6

For details, see http://www.structuredretailproducts.com. The large product universe is a specific feature of the German market. In Switzerland, the second biggest market in Europe in terms of turnover (56% of the exchange turnover in Germany), the number of listed products is considerably smaller (with 36,054 only 3.7% of the German market). For details, see http://www.svsp-verband.ch/download/news content/831 market report sspa july 2013.pdf. An explanation for this deviation is the different fee structure for new listings in the two countries. For example, at the EUWAX trading segment, issuing a new product costs EUR 250, but only up to 5,000 structured products per issuer a year. Each additional product exceeding 5,000 only costs EUR 0.6. (for details, see http://www.boersestuttgart.de/media/dokumente/regelwerke/eng/Fee%20Schedule.pdf). At SIX Exchange in Switzerland, the basic charge per issued structured product is CHF 3,000 (for details, see http://www.six-exchange-regulation.com/admission manual/10 01-LOC en.pdf). Specifically, Bayern LB, BNP Paribas, Commerzbank, Deutsche Bank, DZ BANK, Goldman Sachs, Helaba, HSBC Trinkaus, HypoVereinsbank, LBB, LBBW,

Chapter 3. The Market for Bank-Issued Structured Products

51

Interestingly, the strong product growth has not been accompanied by likewise growth in market volume. As illustrated by Figure 3.2, the market has not reached the same levels as before the global financial crisis and has been fluctuating around EUR 100 billion. In Germany, structured prod-

Market Volume (Billion EUR) Structured Products

140 120 100 80 60 40 20 0 Jan 07

Jan 08

Jan 09

Jan 10

Jan 11

Jan 12

Jan 13

Figure 3.2: Market Volume of Structured Products in Germany This figure shows the monthly market volume of structured products in Germany from January 2007 to June 2013.

ucts can be traded by retail investors in the same way as stocks. Through their brokerage accounts, retail investors can submit orders, either to regulated exchanges or over-the-counter (OTC). However, the underlying market design differs between traditional stock markets and structured products: Whereas on order-driven stock, option and future markets, prices are driven by supply and demand, the price formation of structured products follow a different mechanism. Trading is quote-driven, with market makers (usually the issuer) continuously providing liquidity by quoting bid and ask prices for their products. The reason for this is that due to the large number of heterogeneous products, liquidity would be scarce for each single product in an order-driven market and market functioning could not be guaranteed at all times. An important implication of this market design is that the quoted prices of structured products directly depend on the pricing models adopted by the respective issuer. Further, arbitrage opportunities NORD/LB, Royal Bank of Scotland, Soci´ et´ e G´ en´ erale, UBS und WGZ BANK. The market shares are as of June 2013 and obtained from http://www.deutscherderivate-verband.de/DEU/Statistiken/Marktanteile.

52

3.2. Product Design

virtually do not exist on the market for structured products. Since the counterparty of the investor is always the issuing investment bank, short selling of structured products by investors is impossible. Hence, arbitrage opportunities would only arise if issuers ‘underprice’ their products, as measured by their replication portfolio, which seems highly unlikely.

3.2

Product Design

The specific risk-return profiles of structured products depend on a variety of design characteristics, such as the issuer, the underlying (e.g. stocks, indices, bonds, commodities, currencies), the time-to-maturity, the strike price, and the subscription ratio.7 Generally, there are no strict standardization rules in the market. Hence, there is a large heterogeneity of products, ranging from plain vanilla warrants to more exotic structured products with a high degree of complexity. For this reason, structured products are unattractive to professional investors, who typically value standardized securities. As a result, structured products (almost) exclusively attract order flow of retail investors. Generally, the German market for structured products can be divided into investment products and leverage products.8 Investment Products Investment products typically have risk-return profiles similar to the underlying or slightly more risky or conservative than the underlying. Depending on their specific design, they allow investors to benefit in sideways markets, as well as rising or falling markets. Investment products typically cater for medium to long-term investment strategies and are often used by retail investors in the context of savings plans.9 The most popular types of investment products are discount certificates and 7 8 9

The purpose of the subscription ratio is to scale absolute prices to investor-friendly levels. For details, see the product classification of the German Derivatives Association at http://www.derivateverband.de/ENG/Statistics/MarketVolume. With Cortal Consors, Comdirect, S-Broker and DAB Bank all leading retail brokerages offer saving plans on investment products. For details, see the respective brokerage websites.

Chapter 3. The Market for Bank-Issued Structured Products

53

bonus certificates, with market shares of 38.2% and 42.0%, respectively.10 All following analyses in this thesis are restricted to these two types of investment products.11 Figure 3.3 depicts their payoff profiles. Discount certificates enable retail investors to obtain exposure to the underlying at a price below the price of the underlying. However, in return for this advantage, profit opportunities are capped. Discount certificates can be replicated by a long position in the underlying and a short position in a call option on the underlying, where the strike price equals the cap of the discount certificate. Bonus certificates offer a pre-defined redemption Profit

Profit

Profit Barrier

Cap

0

Strike

0

0 Strike

Discount Certificate Underlying Loss

Bonus Certificate Underlying Loss

(a) Discount Certificate Profit

Loss

(b) Bonus Certificate

Figure 3.3: Payoff StructureProfit of Investment Products

Profit

This figure depicts the payoff structure of warrants and knock-out products with regard to the price of the underlying. 0

0

at maturity, above the price of the underlying at issuance (the bonus). However, if the underlying breaches a pre-defined threshold that is below the underlying price at issuance (the barrier), the redemption value equals Loss Loss the value of the underlying at maturity. Bonus certificates can be replicated by a long position in the underlying and a long position in a down-andout-put option. 10 11

As of 06/2013 in terms of number of products. For details, see http://www.derivateverband.de. More complex investment products include, for example, express certificates, capital protection certificates or twin-win certificates. For details on these products, see http://www.svsp-verband.ch/home/produkttypen.aspx.

0

Loss

54

3.2. Product Design

Leverage Products Leverage products magnify price fluctuations of the underlying. That is, the price of the leverage product changes disproportionately more strongly relative to the underlying. This allows investors Profit Profit to enter leveraged positions with little Profit invested money and facilitates tradProfit ing strategies with aggressive risk-return profiles. The group of leverage products consists of two distinct product types: Warrants and knock-out products. Warrants are bank-issued plain vanilla options. The key differ0 0 0 0 ence of knock-out products, in comparison to warrants, is the additional feature of a so called ‘knock-out barrier’: If the price of the underlying hits a fixed designated threshold, the product becomes worthless imme12 The payoff profiles of the leverage products analyzed in this diately. Loss Loss LossLoss thesis are displayed in Figure 3.4. Irrespective of this difference, both Profit Profit

Profit Profit

0

0

0 Call Strike

0

Put Strike

Call Warrant Put Warrant Underlying

LossLoss

(a) Warrant

Call Knock-Out Put Knock-Out

Call Knock-Out Put Knock-Out Underlying

LossLoss

(b) Knock-Out Product

Figure 3.4: Payoff Structure of Leverage Products This figure depicts the payoff structure of warrants and knock-out products with regard to the price of the underlying.

product types have in common that they allow investors to enter leverage positions with little invested money and thus cater for similar investment strategies and trading set-ups. First, both product types allow investors to benefit from rising (‘call warrants’ or ‘long knock-out products’), as well as 12

Whereas warrants can be priced using traditional option pricing models such as Black and Scholes (1973), knock-out products are equivalent to one-sided barrier options and can be priced with the Rubinstein and Reiner (1991) methodology for down-and-out puts and down-and-out calls for short and long knock-out products, respectively. For details, on the pricing of knock-out products, see Wilkens and Stoimenov (2007) and Entrop et al. (2009).

Chapter 3. The Market for Bank-Issued Structured Products

55

falling prices (‘put warrants’ or ‘short knock-out products’) of the underlying. Second, retail investors commonly use both leverage product types for short-term trading strategies.13 It is important to note that (put) warrants are well qualified to hedge positions. However, as documented by Schmitz et al. (2009), this is potentially only the case in less than 1% of trades and therefore hedging can be ruled out as an important motivation for trading warrants.14

3.3

Pricing and Complexity

Pricing Most of the literature on structured products has focused on their pricing. That is, the question whether prices offered by issuers exceed their theoretical ‘fair’ values and contain significant product margins. It is important to note that product margins in itself are not a negative feature of structured products. As in any other market, a ‘free lunch’ does not exist. Investors are required to pay a price for the service offered by issuers - that is, providing retail investors access to structured products. This compensation at least needs to cover the issuers’ costs (e.g. for product development, structuring, listing, distribution, hedging).15 Hence, to evaluate whether issuers overcharge retail investors, relative overpricing (with regard to other issuers) and the absolute levels need to be assessed in combination. For the German market for structured products, Baule et al. (2008) estimate premiums of discount certificates with individual stocks as un13

14

15

See Entrop et al. (2012), who report mean holding periods of 1.17 trading days for knock-out products and 4.14 trading days for warrants. This implies that the typical trading horizon for leverage products is slightly shorter than for warrants. It nonetheless seems plausible to consider them together, since an investor buying either a knock-out product or a warrant most likely has the same intention: To enter a speculative leveraged position. That is, ‘direct hedging’ by buying a put warrant on an underlying that is at the same time held in the portfolio. Schmitz et al. (2009) further argue that the short average holding periods for put warrants (6 days) and the very low proportion of put warrant trades in all warrant trades (6%) with individual stocks as underlying, provide additional evidence against hedging as trading motivation. For details, see Schmitz et al. (2009), p. 13. In this context, it is important to also take into account the offsetting benefit of structured products from an issuer’s perspective: To enter a debt obligations without having to pay interest rates.

56

3.3. Pricing and Complexity

derlying. They find that margins are low and have decreased over time, which they attribute to increasing competition among issuers. Specifically, based on a structural model, the margins are between 0.67% to 2.27%.16 Wilkens and Stoimenov (2007) study the pricing of knock-out products and find evidence of substantial overpricing, with an average of 4.26% for long certificates and 7.13% for short products across issuers. Further, they document decreasing premiums over products’ lifetimes, which tends to favor the issuer at the expense of the investor (life cycle hypothesis).17 Entrop et al. (2009) study open-end knock-out products and confirm the effect of increasing overpricing over time. In line with the reasoning of Baule et al. (2008), they expect lower margins in the future due to competitive pressure. Some circumstantial evidence that competition is intense and product margins do not necessarily translate into correspondingly high issuer profits is provided by the market exit of some issuers with considerable market share.18 Henderson and Pearson (2011) study the offering prices of a specific type of structured product, distributed to retail investors in the U.S.19 They find that these products, on average, contain an 8% premium on their fair market prices. Bergstresser (2008) analyzes a broader sample of structured notes in the U.S. and finds premiums of similar magnitude. Finally, Benet et al. (2006) focus on reverse-exchangeable securities on the U.S. market. They find that product prices substantially deviate from ‘fair’ valuations, with product premiums of 3-5% at issuance. In sum, structured products are commonly overpriced, but the magnitude of overpricing is different depending on the specific product type and the respective market. On the German market, characterized by a high level of competition, structured products with a relatively low level of complexity seem to be less 16 17

18

19

For details, see Baule et al. (2008), p. 392. Issuers benefit from diminishing overpricing, because investors tend to invest in structured products at the beginning of their lifetime and tend to liquidate positions towards the end of their lifetime. In Germany, the 8th largest issuer in terms of market share (Royal Bank of Scotland) is currently in the process of selling their structured product division. In 2012, Macquarie Bank, who had taken over the structured products division from Sal. Oppenheim in 2010, left the market. Specifically, they analyze so called SPARQS (Stock Participation Accreting Redemption Quarterly-pay Securities) which are medium term notes whose payoff depends on an underlying stock or index.

Chapter 3. The Market for Bank-Issued Structured Products

57

affected by overpricing. How product premiums translate into issuer gains continues to remain unanswered, since the costs structure of the issuers is not elusive. Complexity Since structured retail products were first introduced, ongoing financial innovation has greatly changed the landscape of the market. Issuers have continuously drafted products with new risk-return profiles. This development gives rise to the question introduced at the onset of the chapter: Have structured products become too complex to be understood by retail investors? Contributing to this question, Carlin (2009) propose a theoretical model, to show that overpricing of structured retail products stems from pricing complexity. Their argument is that firms deliberately increase product complexity because it prevents retail investors from assessing whether they are priced fairly or not. As a result, complexity can contribute to a higher market power of the issuer. Therefore, as shown by their theoretical model, an increase in competition induces issuers to increase product complexity. Given the compelling empirical evidence for overpricing, their findings support the reasoning that high levels of complexity indeed prevent retail investors from sufficiently understanding structured products. Bernard et al. (2011) focus on the complexity of a particular type of path-dependent structured product (locally-capped, globally-floored contracts), which are distributed to retail investors in the U.S. market.20 They show that rational retail investors should prefer less complex products, because path-dependent securities do poorly in maximizing expected utility.21 Based on the empirically observed demand for complex products, they argue that retail investors deviate from rational considerations, because they overweight the probability of large positive expected returns in the maximum payoff state. This finding corresponds to the findings of Barberis and Huang (2008) and Kumar (2009) that retail investors prefer skewed returns, which offer large expected returns with a low probability. 20 21

For details on the product design, see Bernard et al. (2011), p. 75. The reason is that for each payoff distribution of path-dependend securities, there exists a replication strategy that generates the same distribution at a lower cost. For details, see Bernard et al. (2011), p. 80.

58

3.3. Pricing and Complexity

Further, Dorn (2012) analyze whether there are too many product options available and as a result, investors fail to pick the most attractively priced products among similar substitutes. Their empirical analysis of retail warrant trades in the German market suggests that they make poor product choices: The warrants actually chosen by retail investors substantially underperform alternative warrants with a similar price elasticity to changes in the underlying. Dorn suggests that different behavioral search heuristics can explain the poor performance. Rather than picking products based on their pricing, retail investors (1) tend to prefer products with low nominal prices, (2) prefer persuasively advertised products and (3) suffer from a ‘status quo’ bias.22 In sum, it can be concluded that structured products often include substantial product premiums. On the well-established German market, the evidence predominantly suggests that product premiums have decreased over time, because of competitive price pressure. Further, the findings imply that retail investors have difficulties dealing with complex structured products, which negatively affects decision-making.

22

The status quo bias has first been documented by Samuelson and Zeckhauser (1988). Applied to structured products, it describes the empirical observation that investors are likely to stick with a particular product/issuer after they have once decided for it.

59

Chapter 4

Data & Methodology 4.1

Retail Investor Trading

The main data set used in this thesis contains more than 12.5 million orders submitted by retail investors in the EUWAX trading segment of Boerse Stuttgart between April 2009 and November 2012.1 With more than 1 million listed securities and a monthly trading volume of EUR 2.2 billion as of August 2013, it is the largest European trading segment for structured products. Trading in the EUWAX segment follows a hybrid market model that combines the benefits of an electronic trading system with the expertise of human market makers (so called quality liquidity providers). Besides providing additional liquidity to the market, the function of the quality liquidity provider is to perform plausibility checks on submitted orders and issuer quotes transmitted to EUWAX. The additional liquidity provision increases the probability that orders are executed at a better price than the one quoted by the issuer.2 The quality assurance offered by quality liq1 2

I thank Boerse Stuttgart for providing the data. This results in the ‘best price principle’, which guarantees that orders are always executed at least at the issuer quote. Potentially lower execution price can be achieved by bundling the available liquidity in the EUWAX trading segment.

60

4.1. Retail Investor Trading

uidity providers can prevent costly partial order executions and mistrades, due to missing or inaccurate quotes. To ensure a high level of investor protection, trading at EUWAX follows a set of pre-defined trading rules, such as equal market access, pre- and post-trade transparency and an independently acting market surveillance.3 To provide retail investors with a high level of flexibility when trading structured products, they can submit different order types to EUWAX. Generally, investor can choose between limit orders and market orders. Investors face a trade-off when deciding between these order types: Limit orders enter the order book and are executed when and if the entered limit is hit. They are associated with uncertainty regarding their execution, which only happens if the quoted price moves in the desired direction and the order can be executed against a counterparty (here, the issuer or the quality liquidity provider). Hence, investors face non-execution or delayed execution risk. In contrast, market orders are always executed immediately for the best available price. However, investors have to pay a price for this immediacy: Orders might be executed at a different price as desired (this particularly applies to fast-moving markets with volatile prices). Further, the EUWAX trading segment allows retail investors to submit additional order types such as stop-loss or stop-buy orders, which enhances the opportunity to implement dynamic trading strategies.4 The EUWAX trading segment provides an ideal environment to study the behavior of retail investors in structured products. First, the trading segment exclusively attracts order flow of retail investors.5 Hence, I do not rely on questionable proxies, such as small trade size, to infer retail investor order flow.6 Further, the trading data not only contains orders from a specific broker (and thus a specific clientele), but all orders routed 3 4

5 6

For details on the EUWAX rules, see §§44-47 of the Exchange Rules, accessible at http://www.boerse-stuttgart.de/files/exchange rules 01072013.pdf. An overview of possible order types is available at http://www.boersestuttgart.de/de/toolsundservices/ordertypen/dieordertypenderboersestuttgart.html. Further details, on the order types analyzed in this thesis will also be given in Section 5.4.4. Nonetheless, to account for the theoretical possibility of professional traders in my sample, I eliminate all trades with a size larger than 250,000 EUR. The average trade size has come down substantially since the introduction of high frequency trading, because the underlying algorithmic trading systems commonly

Chapter 4. Data & Methodology

61

to the EUWAX segment from different brokers. Hence, in all aspects of the thesis, I can draw meaningful conclusions about the general population of retail investors who trade structured products. In addition, I know for each trade whether it is buyer-initiated or seller-initiated and thus do not have to rely on disputable classification algorithms such as Lee and Ready (1991).7 The empirical analysis of retail investor trading in the EUWAX segment relies on two data sets: Product master data and Customer trading data, which are illustrated in Table 4.1 and Table 4.2, respectively. Product master data contains all product-specific characteristics such as the ISIN, ISIN underlying, product type, product sort, underlying type, option type, currency, subscription ratio, strike price, issuer, issue date, and expiration date. This data set can be matched with customer trading data. This second data set contains information about all orders routed to the EUWAX trading segment in the sample period. Each order is assigned an individual order number and a millisecond timestamp. Further, every submitted order contains ISIN, trade direction, quantity, limit and limit type. If the order is executed, a second record with the same order number is generated. This record also includes a millisecond timestamp as well as information about trade price, trade quantity and trade value. Based on the passing time between order submission and order execution, it is possible to determine whether submitted orders are executed and how much time passes before they are executed. This allows for a distinction between limit orders and marketable orders. In all further analysis, I classify orders as marketable orders if they are executed within 10 seconds after submission. That is, marketable orders contain both, market orders and marketable limit orders. The reason for this methodology is that market orders without a limit and very aggressive limit orders are commonly used interchangeably because they reflect

7

split orders. Hence, trade size does not serve as a reasonable proxy for retail order flow anymore. The Lee and Ready (1991) classification algorithm is commonly used in the literature. It compares trades with the mid-quote and, if the results are not conclusive, with the previous trade. However, various studies have questioned the accuracy of the approach. For example, Theissen (2001) argues that it overstates discrepancies between buys and sells.

ISIN ISIN Underlying Instrument Product Type Product Sort Type Underlying Option Type Currency Subscription Ratio Strike Issuer Issue Date Expiration Date

DE000CG5KWL8 DE0008404005 Allianz SE Leverage Product Knock-Out Stock Call EUR 0.10 68.76 Citigroup 2009/08/06 Open End

Product I DE000CM8QFB7 DE0005557508 Deutsche Telekom AG Investment Product Bonus Certificate Stock Call EUR 1.00 N/A Commerzbank 2009/06/03 2010/06/17

Product II

DE000DB3DK64 DE0007500001 ThyssenKrupp AG Leverage Product Warrant Stock Call EUR 0.10 20.00 Deutsche Bank 2009/05/04 2010/12/15

Product III

This table illustrates the information that is contained in the EUWAX master data set and used in the empirical analysis of the thesis. It provides examples of the data for three structured products, which are used in the following empirical analysis.

Table 4.1: Illustration of Product Master Data

62 4.1. Retail Investor Trading

Order Number Submission Time Execution Time ISIN Trade Direction Quantity Limit Limit Type Trade Price Trade Quantity Trade Value Classification

1005073805131 2010/05/07-11:47:03.830 2010/05/07-16:00:59.930 DE000CG5KWL8 Buy 1,000 0.98 N/A 0.98 1,000 980 Limit Order

Order I 906307301246 2009/06/30-12:27:47.620 2009/06/30-12:27:47.680 DE000CM8QFB7 Sell 2,000 0.00 N/A 8.69 2,000 17,380 Marketable Order

Order II

1003084300175 2010/03/08-07:34:37.710 2010/03/09-09:46:07.420 DE000DB3DK64 Sell 6,000 0.57 Stop Loss 0.57 6,000 3,420 Stop Order

Order III

This table illustrates the information that is contained in the EUWAX trading data set and used in the empirical analysis of the thesis. It provides examples for three orders and the order type classification that is derived for the empirical analysis.

Table 4.2: Illustration of Customer Trading Data

Chapter 4. Data & Methodology 63

64

4.2. Financial Market Information

the same motivation for immediate execution.8 A consolidated data set that contains all customer trades and corresponding product master data is used in all following empirical analyses.

4.2

Financial Market Information

For the empirical analysis in this thesis, I combine the retail investor trading data sets with various sources of financial market information data. This section provides an overview of these data sources. Google Trends Google Trends is used in this thesis to capture the investing related information demand of retail investors towards listed companies and the German stock market as a whole. This publicly accessible service by Google Inc. allows to capture historical search volume, directed at particular search terms, relative to the total search volume on Google. An additional feature is the ability to regionally restrict the origination of the captured search volume, for example by country. Generally, Google Trends can be used to derive a direct measure of how much interest the public displays for particular topics in various aspects of life. Choi and Varian (2009) are first to assess the forecasting ability of Google Trends in an economic context. They show that Google search volume can be used to ‘predict the present’, because it “may be correlated with the current level of economic activity in given industries and thus may be helpful in predicting the subsequent data releases”.9 Specifically, they adopt search volume to predict lagged releases of retail sales, automotive sales, home sales and number of tourist traveling to a specific destination. In the following, numerous empirical studies have applied Google search volume data to investigate economic relationships. Goel et al. (2010), for example, provide evidence that search volume can be used to predict consumer behavior such as box-office revenues or video game sales. Other economic variables that have been analyzed include labor markets (McLaren 8

9

The required execution time of up to 10 seconds seems reasonable since market orders that are routed to human market makers (in the hybrid trading system of Boerse Stuttgart this is the case for roughly 30% of all orders) can take a couple of seconds to execute. Choi and Varian (2009), p.1.

Chapter 4. Data & Methodology

65

and Shanbhogue, 2011; Askitas and Zimmermann, 2009; D’Amuri, 2009; Fondeur and Karame, 2013), and housing markets (Wu and Brynjolfsson, 2013). Phenomena unrelated to economic variables that have been successfully related to internet search queries include influenza epidemics (Ginsberg et al., 2009; Althouse et al., 2011) and various other diseases (Pelat et al., 2009). Lexis Nexis Lexis Nexis is used in this thesis to capture historical time series on the media coverage of my sample companies in leading public news sources. This data is used to proxy for the information supply that is available to retail investors in Germany on particular companies. Lexis Nexis, as the world’s largest electronic database for public records related information, is well-suited for this purpose. It aggregates news from a large abundance of publicly available sources and allows to search the database for news articles on particular search terms. In addition, it allows to restrict the sources that are taken into account, which makes it possible to only focus on nationwide news providers. This is an important feature, because the inclusion of regional news providers could induce biases in the information supply.10 Further, it allows to include both, print and online sources, which allows for a complete picture of the information supply typically available to retail investors. Lastly, duplicates can be eliminated from the analysis. In economic research, Lexis Nexis has been adopted by various studies to proxy for media coverage in different contexts, such as financial markets (Fang and Peress, 2009) and politics (Faccio et al., 2006). Dow Jones Newswires Dow Jones Newswires data is used in this thesis, to analyze the intraday market reaction of retail investors to public information arrival. As a leading real-time financial news provider with more than 600,000 subscribers, Dow Jones Newswires serves as a highly representative source for intraday financial news from a broad range of different sources. Specifically, the service is based on a global reporting network of more than 2,000 journalists and publishes close to 20,000 daily news items, covering asset markets worldwide. Further, it includes content from sources such as The Wall Street Journal, Barron’s and SmartMoney. Each Dow Jones Newswires message is tagged with a timestamp and the 10

For example, by a disproportionately high coverage by news providers located in close proximity to company headquarters.

66

4.3. Methodological Approach

companies that are covered in the respective message. This allows me to obtain complete historical records of public information arrival on the sample companies. In financial research, Dow Jones Newswires has been adopted successfully by papers such as Tetlock (2007) and Tetlock et al. (2008). IBES Estimates Earnings Announcements The Institutional Brokers Estimate System (IBES) is employed in this thesis to compute earnings surprises - that is, differences between analyst earnings forecasts and actual earnings. This data is used to analyze how retail investors trade around earnings announcements. IBES is a global database of analysts’ earnings forecasts for listed companies. It covers 40,000 companies in 80 countries and collects earnings estimates from analysts affiliated with 900 research firms. IBES has been the most widely used data source for research that analyzes the financial market impact of earnings announcements and analyst forecasts.11

4.3

Methodological Approach

All following empirical analyses in this thesis are designed to evaluate the aggregated behavior of retail investors in the market for bank-issued structured products. Hence, in all aspects of this thesis, the focus is to analyze the behavior of the general population of retail investors, as opposed to the individual behavior of particular retail investor. Lately, research on investor behavior has more and more focused on the question how personal characteristics systematically influence behavior.12 Whereas, this growing body of literature can provide detailed insights into retail investors’ particular investment decisions and motives, it can provide less insights on the financial market impact. For example, retail investors as a whole cannot affect financial markets in any way, if the systematic behavior of one investor subgroup (characterized by certain personal features) is com11 12

For recent examples, see Roll et al. (2010), Kaniel et al. (2012) and Barber et al. (2013). See, for example, Kumar et al. (2011) or Addoum et al. (2013).

Chapter 4. Data & Methodology

67

pensated by opposing actions of another subgroup (with different personal characteristics). Further, the aggregation assumption inherently applies to all theoretical models in the field of behavioral finance, which build on representative irrational agents or the existence of similarly acting irrational investors. Further, the systematic nature of many behaviorally induced biases suggests that they should persist, on aggregate, across investors. However, since a large body of literature on behavioral biases is based on experimental settings, testing whether these biases actually continue to be visible in real-life investment decisions after aggregation across a large population of investors, is important for their universal validity.13 More generally, as argued by Frey and Gallus (2013), the aggregated perspective of behavioral biases, may provide different results, because individuals might react differently when they face institutional constraints and interact with others. As a result, behavioral patterns documented in experimental settings, might either disappear or become even more pronounced. For these reasons, a macro-level approach to investor behavior, as adopted in this thesis, is better suited to shed light on the question, how retail investors can systematically affect financial markets. Put differently, the aggregation assumption underlying this thesis allows to analyze whether deviations from rational behavior are sufficiently systematic to persist across the general population of retail investors trading structured products. This implies that the empirical findings presented in the following chapters are ‘on aggregate’ valid for all types of retail investors trading structured products. Further details, that go beyond the general methodological approach adopted in this thesis, are provided in the empirical chapters. The reason is that the specific methodologies adopted to address the respective research questions differ considerably.

13

See Jackson (2003), p. 2.

69

Chapter 5

Retail Investor Information Demand Attention is a scarce cognitive resource. Daniel Kahnemann (1973)

5.1

Introduction

The hypothesis that financial markets are informationally efficient rests on the assumption that prices respond immediately when new information becomes available. However, such rapid incorporation of information into security prices requires two things: First, investors need to have access to all relevant information (information supply). Second, investors need to pay attention to this available information (information demand ) and include it in financial decision-making. Given a continuously growing information flow on financial markets, this seems questionable. In fact, as stated by Kahnemann (1973), in today’s information economy attention has become ‘a scarce cognitive resource’. This applies to all kind of investors, but in particular to retail investors, who typically lack time and resources. Consequently, information is not fully incorporated into asset prices until all

70

5.1. Introduction

investors pay attention to it. Previous research has primarily focused on the role of information supply in financial markets. However, neglecting the other side of the coin (information demand) can only provide incomplete insights into the relationship between information transmission and financial markets. In this chapter, I therefore investigate interlinkages between information demand and financial markets and specifically focus on the information demand (as measured by investing-related Google search volume) and the trading behavior of retail investors. Da et al. (2011) are first to propose aggregated internet search volume on Google as a direct measure of investor attention/information demand.1 This measure seems well-suited for various reasons. First, the internet has become the primary source of freely available information in all aspects of life and is thus a resourceful medium to measure public interest. In addition, search engines have become the natural starting point of information collection and Google continues to be the dominant player with a market share of 96% in Germany as of January 2012.2 For this reason, Google searches almost perfectly mirror the collective search activity of the general population. Second, search volume is superior to alternative measures since it requires a pro-active behavior of internet users and therefore only captures actual interest in a topic. When retail investors are interested in investing in a company, they might gather additional information about the company via search engines. In contrast, professional investors are more likely to use more sophisticated information sources of financial data providers such as Bloomberg. Thus, internet search volume enables me to directly and reliably capture the information demand of retail investors. I introduce a novel keyword methodology, to obtain the information demand from Google that specifically takes into account how retail investors search for investing-related information. In a first step, I provide a detailed analysis of the previously largely unexplored relationship between the information demand and information supply available to retail investors. I measure information supply as the 1 2

I use investor information demand and investor attention interchangeably throughout the thesis. Pew Internet survey 2012 (http://www.pewinternet.org) finds that 91% of online adults use search engines. Google market shares as reported by comScore (http://www.comscore.com).

Chapter 5. Retail Investor Information Demand

71

number of news mentions for individual firms and the stock market as a whole, in media sources typically available to retail investors. On the firmlevel, information demand and information supply are positively correlated, whereas on the market-level, I find a negative correlation. The intertemporal analysis shows that causality tends to run from information demand to information supply, on the stock-level as well as the market-level. Thus, information discovery often takes place via alternative information channels, before it is incorporated in traditional news media. To analyze the relationship between retail investor information demand and retail investor behavior, I combine information demand with customer trades in structured products at the EUWAX trading segment. In the analysis, I differentiate between effects on short-term speculative trading and more long-term oriented investing. Further, I study order submission strategies and positioning of retail investors. I find a significant positive relationship between firm-specific information demand and both, contemporaneous and future speculative trading activity. No such relationship exists for long-term oriented investing activity. This implies that investing either relies less on information gathering online or that it is less information-driven. Market-wide information demand leads to subsequently lower trading activity in structured products and more aggressive trading. This particularly applies to speculative trading. Further, high market-wide information demand is associated with higher order uncertainty. I find no effect of firm-specific information demand on retail investor positioning in leverage products. Market-wide information demand is associated with ‘short interest’ expressed by retail investors. I argue that this behavior stems from increased uncertainty in the overall market development. I do not find retail investor information demand to exert an upward price pressure on stock prices. In all analysis, the effect of market-wide information demand on retail investor trading is more pronounced than firm-specific information demand. This indicates that trading behavior in bank-issued structured products is more influenced by overall market developments than by company fundamentals. As a result, the informational efficiency of retail investor trading in structured products seems limited.

72

5.2. Literature Review

The remainder of this chapter is structured as follows: Section 5.2 provides an overview of previous research on information demand on financial markets. Section 5.3 provides descriptives of the information data sets and retail investor trading data. In Section 5.4, I report the results on the intertemporal relationship between information demand and information supply, the effect of information demand on different aspects of retail investor trading behavior, as well as the asset pricing impact on the returns of the underlying stocks. Section 5.5 concludes.

5.2

Literature Review

Merton (1987) and more recently, Peng and Xiong (2006), and Barber and Odean (2008) provide theoretical frameworks of the relationship between information demand and stock markets. In a model of capital market equilibrium, Merton (1987) argues that investors are incompletely diversified, because they are not aware of all available securities. Hence, less-known firms need to compensate investors with higher returns. Based on this reasoning, an increase in investor information demand leads to positive price pressure in the short-run and lower returns in the long-run. Based on the notion that investors are generally attention-constrained, Peng and Xiong (2006) show that investors tend to allocate more attention to marketwide and sector-wide information than to firm-specific information. Markets/sectors with higher investor attention allocated to firm-specific information are informationally more efficient since stock prices contain more information about fundamentals. Barber and Odean (2008) build on Merton (1987) and argue theoretically that investor attention temporarily exerts upward pressure on security prices. The rationale behind this is that retail investors can implement positive market expectations on any company by buying their stocks, whereas negative market expectations can only be transformed into trading decisions for stocks already held, due to the inability to sell short. As a result of this imbalance in buying and selling opportunities, retail investors are net buyers of stocks. Testing theories of investor attention empirically remains challenging since direct measures are largely unavailable. Thus, several indirect mea-

Chapter 5. Retail Investor Information Demand

73

sures have been proposed as proxies. Barber and Odean (2008), and Gervais (2001) use high trading volume and extreme returns as proxies, based on the reasoning that both are associated with the arrival of information and thus generate interest on behalf of investors. Both studies find a temporary upward price pressure on stock returns. Fang and Peress (2009) use media coverage as proxy, arguing that events which stir high information demand are usually also newsworthy and covered by the media. They support the theoretical findings of Merton (1987) and document higher returns for stocks with no media coverage, in comparison to stocks with high media coverage.3 All these previously used proxies inherently assume that investors pay attention to the available supply of information. However, this is only the case, if, for example, investors actually read newspaper articles about a company. For this reason, their ability to serve as an accurate proxy for investor information demand is limited. To overcome the shortcomings of such proxies, internet search queries are adopted in a growing number of different contexts to measure investor attention. Da et al. (2011) use it to measure general investor attention and find evidence of a short-term upward price pressure on stock markets, which is subsequently reversed in the long-term. Bank et al. (2011) also find search queries to predict temporarily higher returns and improved liquidity. Fink and Johann (2013) also document a positive effect on liquidity, with effects being particularly pronounced for small firms. Vlastakis and Markellos (2012) show that information demand is positively related to trading activity and volatility. Further, they document that market-wide information demand has a more pronounced impact than firm-specific information demand. Da et al. (2011) and Drake et al. (2012) investigate investor information demand around earnings announcements and show that it can predict earnings surprises and returns following the announcements.

3

Barber and Odean (2008) also include media coverage in their analysis. Other indirect measures of investor attention include advertising expenses, analyst coverage, institutional holdings and price limits.

74

5.3. Data

5.3 5.3.1

Data Information Demand and Information Supply

To measure information demand, I employ the volume of internet search queries as provided by ‘Google Trends’.4 The search volume for a particular search term is not reported in absolute terms, but relative to the total search volume in the specified period. Accordingly, a positive time trend in search volume due to increasing search engine usage is eliminated, which allows for comparison of the resulting time series over time. Further, search volume is normalized to a range of 0 to 100, whereas 100 represents the highest relative search volume during the period and 0 implies a search volume below a certain threshold.5 The keyword methodology is essential for the information content of the extracted time series. To capture investors’ information demand expressed for individual securities, two different keyword methodologies have been adopted in empirical studies. One is to identify search volume by company ticker symbol, the other is to use ordinary company names. For example, Da et al. (2011) use stock ticker symbols. However, the ability of this approach to capture the information demand of retail investors is severely limited, since retail investors typically do not use ticker symbols when searching on Google.6 Bank et al. (2011) employ a broader approach and proxy investor information demand by using ordinary company names. The problem associated with this approach is that searching for a company name can be entirely unrelated to investing. In fact, most of the search volume related to the term ‘BMW’ is likely to be attributable to interest in the products of the company - that is, cars for this particular example. Further, the company name can have multiple meanings (e.g. for the German-based company ‘MAN’ or the U.S.-based company ‘Apple’).

4 5 6

Google Trends can be accessed at http://www.google.com/trends. Further methodological details are not provided by Google. This is supported by the fact that no search volume is reported by Google Trends for any of the stocks in my sample when I use ISIN, RIC or Bloomberg ticker symbol, respectively, as keyword.

Chapter 5. Retail Investor Information Demand

75

For this reason, I introduce a novel approach that specifically focuses on the information demand of retail investors. I use the company names as reported by Thomson Reuters Datastream and add the German equivalent for the term ‘stock’ (that is, ‘Aktie’) to the search query. I drop the term that identifies the legal form of the firm (e.g. ‘AG’ or ‘SE’) since they are typically not included in the search query.7 In addition, I take into account that investors might use different names for one and the same company, by using the most widespread name as indicated by Google Trends.8 Finally, I restrict the search volume to queries submitted from Germany, since I analyze retail investor trading on the German market and require corresponding information demand data. To obtain search volume directed at the German stock index DAX, the proxy for market-wide information demand, I proceed analogously and use the keyword ‘dax’.9 As a sample, I include all stocks listed on the stock indices DAX, MDAX and TecDAX and focus on the sample period from April 2009 to March 2012.10 I adopt this recent sample period, because more missing values are reported in prior years, due to insufficient search volume. For each stock, I manually attempt to obtain search volume (information demand ). Depending on the magnitude of search volume during the specified period, Google reports weekly, monthly or no data at all. In my analysis, I focus on the stocks with weekly data availability (that is, the stocks with the highest information demand during the sample period), which is given for 25 stocks and the index.11 Descriptive statistics, along with the used search queries, are reported in Table 5.1. The second important variable in my analysis is a measure of retail investor information supply. To determine the impact of information demand on retail investors, it is crucial to control for potential effects of 7 8 9

10 11

I repeat the keyword search, including the legal forms and only obtain monthly search volume or no observations at all due to a lack in search volume. E.g. to measure information demand for the company ‘Deutsche Telekom’, I use the search term ‘Telekom’ which is used more often. Potential alternative search queries as proxy, such as ‘Aktien’ (stocks) or ‘Aktienkurse’ (stock prices) are highly correlated with ‘dax’, but receive substantially less search volume. Hence, ‘dax’ seems to be the most suitable and robust proxy for market-wide information demand. The three indices include the 110 largest German companies. Each week ranges from Sunday to Saturday.

Search Query

”siemens aktie” ”vw aktie” ”sap aktie” ”eon aktie” ”basf aktie” ”daimler aktie” ”telekom aktie” ”bayer aktie” ”allianz aktie” ”deutsche bank aktie” ”bmw aktie” ”rwe aktie” ”eads aktie” ”post aktie” ”thyssenkrupp aktie” ”man aktie” ”commerzbank aktie” ”infineon aktie” ”wacker chemie aktie” ”lufthansa aktie” ”aixtron aktie” ”sky aktie” ”solarworld aktie” ”q-cells aktie” ”nordex aktie”

”dax”

Security

Siemens AG Volkswagen AG SAP AG E.ON SE BASF SE Daimler AG Dt. Telekom AG Bayer AG Allianz SE Deutsche Bank AG BMW AG RWE AG EADS N.V. Dt. Post AG ThyssenKrupp AG MAN SE Commerzbank AG Infineon Tech. AG Wacker Chemie AG Dt. Lufthansa AG AIXTRON SE Sky Dt. AG SolarWorld AG Q-Cells SE Nordex SE

Stocks pooled DAX

26.67 19.46

24.41 25.83 27.17 26.17 26.90 36.25 29.18 25.94 38.83 42.32 15.66 28.85 23.91 29.35 25.15 27.85 27.17 40.80 11.25 32.26 35.74 23.85 10.30 20.53 11.19 18.66 9.06

12.66 12.67 26.39 13.02 14.30 15.20 11.05 16.45 18.42 14.45 11.15 17.63 26.42 28.21 28.13 21.84 23.71 14.37 27.53 15.70 25.95 24.86 10.81 20.51 15.06 1.41 5.28

2.04 1.82 0.58 2.18 1.30 1.10 2.41 0.85 1.19 0.98 3.27 1.28 0.77 0.35 0.56 0.10 1.34 1.00 2.13 1.07 -0.12 0.92 4.55 0.74 2.77

Information Demand Mean Std. Skew.

4.80 40.97

8.14 7.42 -0.40 8.78 3.26 1.87 10.99 3.70 1.14 1.22 20.90 1.93 -0.22 -1.04 -0.98 -0.54 0.69 2.06 2.79 2.85 -0.85 0.26 32.90 0.58 12.47

Kurt.

79.49 46.59

51.88 49.05 97.13 49.78 53.16 41.94 37.87 63.40 47.44 34.14 71.18 61.11 110.49 96.11 111.87 78.44 87.27 35.22 244.74 48.66 72.62 104.26 104.96 99.90 134.59

CV.

43.23 278.55

102.66 46.87 16.71 29.29 15.17 22.51 40.90 24.76 112.31 152.26 31.55 31.56 32.73 26.11 14.76 9.59 175.61 13.99 3.84 85.28 15.51 14.21 22.07 24.95 15.68

25.41 97.69

48.91 27.47 16.08 23.38 15.31 14.46 24.32 18.75 38.03 58.20 18.09 27.57 26.74 19.13 15.94 11.64 75.35 15.11 6.30 49.05 15.38 16.18 17.70 19.47 16.80

1.83 0.46

1.06 0.91 1.90 2.49 1.95 0.82 1.77 1.73 0.85 0.81 1.46 2.03 2.01 2.03 2.35 2.60 0.49 1.79 2.68 1.81 2.25 2.75 2.21 1.88 3.17

Information Supply Mean Std. Skew.

4.81 -0.62

0.83 1.11 5.19 8.58 3.82 0.29 5.30 3.60 0.51 0.01 2.78 5.05 5.10 5.18 6.70 8.82 -0.38 3.40 7.36 4.86 5.20 10.60 6.23 4.15 15.88

Kurt.

81.38 35.07

47.64 58.62 96.24 79.81 100.92 64.27 59.46 75.70 33.86 38.23 57.32 87.34 81.70 73.28 108.05 121.34 42.91 108.02 164.01 57.51 99.16 113.82 80.21 78.02 107.14

CV.

This table reports descriptives statistics for weekly raw information demand (ID) and raw information supply (IS) for all sample stocks and the German stock market index DAX. It also reports the search queries used to extract information demand from Google Trends. Information supply shows the number of news articles in leading print and online media, mentioning the respective company or index, as reported by the news aggregation database LexisNexis. ‘Skew.’ reports skewness, ‘Kurt.’ kurtosis, and ‘CV.’ the coefficient of variation.

Table 5.1: Descriptive Statistics of Information Demand and Supply

76 5.3. Data

Chapter 5. Retail Investor Information Demand

77

information supply in the analysis. To do this, I deliberately do not resort to news feeds of financial data providers, but use information sources typically available to retail investors. Thereby, I ensure that the information supply measure corresponds to the information demand measure. Specifically, I obtain news data from LexisNexis, the world’s largest electronic database for public records related information.12 It aggregates news from a large abundance of sources and allows me to extract the number of news items mentioning the respective company. I use the full company name (including the legal form) as search query and count the number of mentions for each week and each sample stock. I include articles published in print versions as well as online versions of the leading public news sources in Germany.13 To avoid biases in the number of news items, I do not take into account duplicates. To measure market-wide information supply, I proceed analogously and use ‘dax’ as search keyword. As an example, Figure 5.1 provides the weekly time series of information demand and information supply for Germany’s biggest telecommunications company Deutsche Telekom AG during the sample period. Retail investor information demand was highest when the company announced that it would sell its U.S. operation to AT&T in March 2011. Other companyspecific events that caused spikes in retail investor information demand are annual shareholder meetings, earnings announcements and the failure of the U.S. operations sell-off. Further, the differences between information demand and information supply show that although they exhibit similar patterns, some events are more relevant to information demand, whereas other events are more relevant to information supply.

5.3.2

Retail Investor Trading and Market Data

To analyze aggregated retail investor behavior, I study trading in investment products and leverage products in the EUWAX trading segment.14 I use the two distinct product categories to differentiate between ‘investing’ 12 13 14

http://www.lexisnexis.de/. For example, Handelsblatt, Financial Times Deutschland, Spiegel Online, Die Zeit, FAZ, Boersen Zeitung, etc. In total, I include 30 different sources. For details on the data set, see Section 4.1 of this thesis.

78

5.3. Data

100

Information Demand (Google Trends)

Information Demand and Supply

Deutsche Telekom Surges on $39 Billion U.S. Sale to AT&T

Information Supply (Lexis Nexis)

90

Mar 21, 2011 Bloomberg:

80

April 30, 2009:

70

May 3, 2010:

60

Deutsche Telekom Annual Shareholders’ Meeting

Deutsche Telekom Annual Shareholders’ Meeting

Aug 4, 2011 Bloomberg:

Aug 31, 2011 Bloomberg:

Deutsche Telekom Quarterly Profit Declines on TMobile USA

U.S. Files Antitrust Complaint to Block AT&T, T-Mobile Deal

50 40 30 20 10 0 Apr 2009

Sep 2009

Feb 2010

Jul 2010

Dec 2010

May 2011

Oct 2011

Mar 2012

Figure 5.1: Illustration of Information Demand and Supply This figure reports the time series of weekly aggregate search volume in Germany for the search term ”telekom aktie” (information demand). It also shows the corresponding weekly aggregate number of news articles (information supply), as well as descriptions of the firm-specific events during the weeks with the highest information demand. Search volume is scaled to the total search volume during the observation period from April 2009 to March 2012 and normalized to a range from 0 to 100. Information supply is also scaled to a range from 0 to 100.

Chapter 5. Retail Investor Information Demand

79

and ‘speculating’. Trades in more long-term oriented investment products are used as proxy for investing-related trading motives, whereas trades in short-term oriented leverage products are used to proxy for speculative trading motives.15 As product underlyings, I focus on all stocks, for which I could acquire information demand and information supply data.16 Corresponding to information demand, the trading data covers the 36 month period from April 2009 to March 2012 (156 trading weeks). Table 5.2 reports descriptive statistics. The number of leverage products in my sample is 148,820, whereas investment products account for 41% and leverage products for 59%. More than 73% of leverage products in my sample are calls, whereas 27% are puts. The total trading volume during the sample period is EUR 26.33 billion, whereas 80% is in investment products and 20% is in leverage products. The largest share of trading volume is in calls, which account for EUR 25.56 billion or 97% of the trading volume. The average trade size in the sample is EUR 42,649 for investment products and EUR 5,110 for leverage products. All other market data used in the analysis is obtained from Thomson Reuters DataScope Tick History archive through the Securities Industry Research Centre of Asia Pacific (SIRCA) and Thomson Reuters Datastream.17 I calculate weekly realized volatility as the sum of squared returns of 5-min intraday mid-quotes as obtained from Thomson Reuters DataScope Tick History.18 To compute absolute weekly returns, I use the 15

16 17 18

The distinction between investing and speculating is not straightforward and can depend on various factors, such as the traded asset, the intended holding period, and the leverage ratio. Generally the amount of risk associated with a trade can be used to differentiate between investing and speculating. Accordingly, trades that involve high risks can be classified as speculating, whereas trades that involve comparatively lower risks can be classified as investing. Further, the motivation typically differs in the sense that speculators aim for abnormally high returns by entering bets that are either right or wrong, whereas investors aim for ‘satisfactory’ returns and assume moderate risks. Given these differences, leverage products and investment products seem well suited to proxy for speculating and investing, respectively. That is, trades in products that have the respective stock as underlying. I thank SIRCA for providing access to the Thomson Reuters DataScope Tick History archive. This approach to measuring realized volatility is commonly used in the literature since intraday returns have been shown to contain valuable information about volatility estimates at the daily/weekly level. For details, on how to capture realized

80

5.3. Data

Table 5.2: Descriptive Statistics of Trading in Structured Products This table reports the descriptive statistics for all structured products in the sample. All statistics are reported separately for investment products and leverage products. I report the number of products and distinguish between ’calls’ (products that benefit from rising security prices) and ’puts’ (products that benefit from falling security prices). Trading volume and number of trades in the sample period are reported by trade direction (’buy’ or ’sell’) and product type (’call’ or ’put’). Mean and median trade sizes are also reported. Further, I report the number of submitted orders, the number of executed orders, marketable orders (all orders executed within 10 seconds after submission), limit orders (all orders not executed within 10 seconds after submission) and stop orders (all orders that contain a stop buy or stop sell feature). Investment Products

Leverage Products

Products Calls Puts

60,623 444

64,107 23,646

Volume (in Mil. EUR) Calls Puts Buys Sells

20,815 63 11,829 9,049

4,741 706 2,797 2,650

Trades Calls Puts Buys Sells Mean Size (EUR) Median Size (EUR)

504,494 2,674 313,543 193,625 42,649 10,704

932,428 126,936 592,326 467,038 5,110 1,810

Orders Submitted Executed Marketable Limit Stop

642,031 558,344 411,470 230,561 16,033

1,635,386 1,126,316 498,890 1,136,496 150,497

Chapter 5. Retail Investor Information Demand

81

opening price on the first trading day of the week and the closing price of the last trading day of the week. I obtain weekly market capitalization data from Thomson Reuters Datastream, as the market value of all equity types at market close on the last trading day of the previous week. Table 5.3 gives an overview of all variable definitions used in the empirical analysis of this chapter.

5.4 5.4.1

Empirical Results Time Series Properties of Information Demand and Supply

Table 5.1 reports descriptive statistics of the information demand and information supply variables. Information demand exhibits positive skewness for the majority of stocks as well as the index, which implies that the mass of the distribution is concentrated on the left, with relatively large variations between observations. Further, it displays positive excess kurtosis for most stocks and the index, indicating a peaked distribution. The dispersion of information demand fluctuates substantially between sample stocks, with coefficients of variation ranging from 34.14 for Deutsche Bank AG and 244.74 for Wacker Chemie AG. In comparison, the index displays relatively low dispersion, with a coefficient of variation of 46.59, indicating a rather stable information demand throughout the sample period. To assess the time series properties, I perform Augmented Dickey-Fuller (ADF) tests on the logarithmically transformed information demand time series.19 The results, reported in Table 5.4, indicate that information demand is stationary around a deterministic trend. Since the reported search volume is already normalized by Google to account for increasing internet adoption, this indicates that the information demand of internet users re-

19

volatility, see, for example, Andersen et al. (2001), Andersen et al. (2003), Andersen et al. (2005) and Andersen et al. (2011). The purpose of the logarithmical transformation is to stabilize the variance of the time series. For details, see Luetkepohl and Xu (2012). For robustness, I also perform Phillips Perron (PP) tests, which yield similar results. See Dickey and Fuller (1979) and Phillips and Perron (1986) for details on the methodology.

82

5.4. Empirical Results

Table 5.3: Variable Definitions This table reports the definition and the computation methodology for all variables used for the empirical analysis in Chapter 5. Name Information Information Demand (ID)

Information Supply (IS)

Retail Investor Trading Volume

Trades

Order Aggressiveness Order Uncertainty Imbalance (IMB)

Market Data Realized Volatility (RV)

Absolute Return

Abnormal Return Market Cap

Definition Weekly logarithmically transformed search volume, as obtained from Google Trends based on the search queries as specified in Table 5.1; Detrended and seasonality adjusted. Computed for all stocks (ID) as well as the index (ID Index ) Weekly logarithmically transformed number of news articles, in which the respective company is mentioned. Obtained from the news aggregation database LexisNexis. Detrended and seasonality adjusted. Computed for all stocks (IS ) as well as the index (IS Index ). Included are 30 leading German news sources (online and print). News providers include e.g. ’Handelsblatt’, ’Financial Times Deutschland’, ’Spiegel Online’, ’Die Zeit’, ’FAZ’, ’Boersen Zeitung’ (the full list is available upon request). As search query, I use the full company name including the legal form. For the German stock market index DAX, I use ’dax’ as search query. Duplicates are not included. Weekly logarithmically transformed aggregated trading volume of all executed orders in products that have the respective stock/index as underlying. Computed separately for investment products and leverage products. Weekly logarithmically transformed aggregated number of trades in products that have the respective stock/index as underlying. Computed separately for investment products and leverage products. The weekly ratio of marketable orders (all orders executed within 10 seconds after submission) to all submitted orders. The weekly ratio of orders that include either a stop buy or a stop sell option to all submitted orders. Measures whether individual investors are predominately long or short positioned. Calculated separately for investment products and leverage products. Computed as volume-based measure (IMB Volume - see Equation 5.2) and trade-based measure (IMB Trades - see Equation 5.3). I calculate weekly realized volatility for all stocks (RV ) and the index (RV Index ) as the sum of squared returns of 5-min intraday mid-quotes as obtained from Thomson Reuters Tick History (SIRCA). Since the weekly 5-minute intervals vary between weeks (e.g. due to public holidays), I multiply this weekly measure with the ratio of total yearly 5-min intervals to the number of intervals in the respective week and then take the square root to obtain a yearly measure. Absolute return is computed as the return between the opening price on the first trading day of the week and the closing price of the last trading day of the week. I compute absolute returns for each stock (ABS Ret) and the index (ABS Ret Index ). Computed for each stock as the return in excess of the German stock market index DAX. Logarithmically transformed consolidated market value of all equity types at market close on the last trading day of the previous week, as obtained from Thomson Reuters Datastream. Denoted in Million EUR.

Chapter 5. Retail Investor Information Demand

83

garding investing-related information has grown more strongly than overall search engine usage. I detrend information demand by taking the residuals from a linear regression of information demand on time. Next, I attempt to shed some light on the question whether detrended information demand exhibits deterministic seasonal variations. Information demand might exhibit such patterns because of factors such as public holidays (less demand for information) or quarterly earnings announcements (more demand for information) that are clustered around the same time periods each year and are likely to have an important impact on retail investor behavior. Given the data limitation of a relatively short sample period, conclusively testing for seasonality effects is difficult. To nonetheless evaluate seasonal variations in information demand, I run an F-test for equality of means across months. That is, I average my weekly observations during each sample month and then test whether the 36 monthly data points are equal during the sample period. The results, also reported in Table 5.4, indicate that most of my sample stocks exhibit monthly variation – which implies that seasonal patterns are present. To account for them, I regress the detrended information demand of each stock on weekly dummy variables. The dummy loadings are then subtracted from the detrended information demand variable.20 Nonetheless, F-tests show no significant differences in means for the adjusted time series, which indicates that my data meets the required statistical properties for further analysis. Also, Jarque-Bera normality tests indicate that the adjusted time series are normally distributed. In all further analysis, I use the logarithmically transformed time series (referred to as ‘information demand’). In line with previous studies (Thompson et al. (1987), Vlastakis and Markellos (2012)), I find information supply to vary considerably across stocks. Information supply also exhibits excess kurtosis and positive skewness for the majority of stocks. To evaluate the time series properties, I proceed analogously to the information demand variable. The results 20

Again, it is important to note that the adopted demeaning procedure is limited in the sense that it can only approximately account for seasonality patterns. Due to the limited length of my data set, more sophisticated procedures to deseasonalize the time-series could not be applied. Also see Vlastakis and Markellos (2012), who face similar data limitations and resort to a corresponding approach.

84

5.4. Empirical Results

Table 5.4: Time Series Properties of Information Demand and Supply This table reports the time series properties of the raw (logarithmically transformed original time series) and the adjusted time series of information demand and information supply. ’Normality’ reports Jarque-Berra normality test statistics of the raw time series. ’Stationarity’ reports Augmented Dickey-Fuller (ADF) test statistics for the raw time series and the detrended time series. Detrended time series are computed as the residuals of a linear regression against a weekly time trend. ’Seasonality’ reports the results of F-tests for equality of means across months. To obtain the deseasonalized time series, I regress the detrended time series on weekly dummy variables. The dummy loadings from these regressions are then substracted from the detrended time series, to obtain deseasonalized time series. */**/*** indicate statistical significance at the 10%/5% and 1% level, respectively.

Siemens AG Volkswagen AG SAP AG E.ON SE BASF SE Daimler AG Dt. Telekom AG Bayer AG Allianz SE Deutsche Bank AG BMW AG RWE AG EADS N.V. Dt. Post AG ThyssenKrupp AG MAN SE Commerzbank AG Infineon Tech. AG Wacker Chemie AG Dt. Lufthansa AG AIXTRON SE Sky Dt. AG SolarWorld AG Q-Cells SE Nordex SE Stocks Pooled DAX

Information Demand Stationarity Seasonality Raw Detrend. Detrend. Deseas.

Normality Raw

Deseas.

3.73 9.90*** 4.30* 5.65** 2.96 3.52 10.76*** 4.33* 5.79** 8.18*** 3.15 4.50* 1.10 4.29* 2.38 4.84** 1.23 7.52*** 4.82** 5.45** 3.74 2.09 5.81** 2.14 1.60

7.44*** 15.05*** 9.76*** 10.88*** 14.10*** 8.09*** 16.30*** 4.69** 9.22*** 11.19*** 6.10** 14.22*** 4.21* 9.53*** 10.41*** 8.00*** 3.13 17.13*** 9.00*** 6.60*** 5.90** 4.28* 6.32*** 3.68 4.28*

1.48 1.32 3.41*** 3.39*** 3.15*** 1.64* 3.85*** 2.37*** 1.76* 1.60 4.35*** 1.01 0.72 1.39 2.84*** 3.88*** 0.59 2.23** 2.62*** 4.14*** 3.83*** 1.12 0.77 0.90 2.66***

0.01 0.15 0.09 0.27 0.03 0.05 0.06 0.04 0.03 0.07 0.18 0.09 0.09 0.06 0.14 0.07 0.04 0.13 0.05 0.21 0.09 0.08 0.05 0.18 0.12

506.08*** 416.38*** 9.83*** 586.40*** 106.53*** 51.16*** 879.83*** 100.18*** 43.44*** 33.17*** 2923.10*** 63.69*** 15.46*** 10.16*** 14.42*** 2.29 48.46*** 50.55*** 161.51*** 77.48*** 5.25* 21.76*** 7108.86*** 15.65*** 1136.85***

0.37 181.50*** 1.25 6.17** 0.04 8.40** 27.07*** 16.89*** 227.81*** 0.06 42.99*** 116.24*** 12.64*** 0.36 1.71 3.50 13.28*** 0.64 3.36 89.66*** 13.19*** 14.90*** 7.40** 1.32 6.10**

4.55* 6.00**

8.78*** 7.24***

2.28** 3.05***

0.10 0.04

575.54*** 307.53***

31.87*** 21.60***

continued on the next page. . .

Chapter 5. Retail Investor Information Demand

85

Table 5.4 - continued Information Supply Stationarity Seasonality Raw Detrend. Detrend. Deseas.

Normality Raw

Deseas.

Siemens AG Volkswagen AG SAP AG E.ON SE BASF SE Daimler AG Dt. Telekom AG Bayer AG Allianz SE Deutsche Bank AG BMW AG RWE AG EADS N.V. Dt. Post AG ThyssenKrupp AG MAN SE Commerzbank AG Infineon Tech. AG Wacker Chemie AG Dt. Lufthansa AG AIXTRON SE Sky Dt. AG SolarWorld AG Q-Cells SE Nordex SE

21.38*** 13.22*** 16.78*** 18.95*** 21.95*** 9.95*** 20.83*** 19.55*** 20.08*** 15.67*** 41.56*** 10.66*** 19.08*** 36.10*** 22.54*** 20.07*** 8.94*** 27.40*** 20.11*** 18.52*** 27.56*** 13.85*** 22.29*** 24.16*** 29.15***

21.86*** 14.74*** 25.64*** 18.94*** 27.66*** 15.96*** 25.45*** 26.15*** 32.09*** 26.96*** 48.23*** 14.28*** 23.40*** 36.08*** 29.69*** 21.71*** 17.03*** 28.44*** 28.03*** 20.85*** 27.26*** 19.57*** 27.39*** 32.69*** 32.08***

1.62* 1.36 1.30 2.81*** 1.20 3.87*** 1.85* 1.54 2.98*** 3.89*** 2.00** 2.43*** 2.87*** 1.31 2.70*** 1.60 4.55*** 2.43*** 1.16 1.64* 3.52*** 1.39 2.93*** 2.53*** 1.57

0.15 0.20 0.23 0.00 0.29 0.00 0.05 0.12 0.00 0.00 0.03 0.01 0.00 0.22 0.00 0.11 0.00 0.01 0.32 0.09 0.00 0.18 0.00 0.01 0.11

32.54*** 27.87*** 254.15*** 602.99*** 183.84*** 17.33*** 249.09*** 153.01*** 19.67*** 16.92*** 99.80*** 257.93*** 258.74*** 266.27*** 411.51*** 642.76*** 7.31** 150.41*** 509.06*** 225.49*** 291.57*** 872.93*** 357.26*** 192.91*** 1784.79***

0.06 1.98 0.20 41.13*** 0.90 2.35 0.44 1.11 0.87 0.89 0.57 2.94 10.76*** 1.63 0.50 0.95 1.46 0.51 3.55 5.30* 0.36 1.47 2.17 3.07 3.86

Stocks Pooled DAX

20.81*** 5.43**

25.69*** 8.87***

2.28** 3.31***

0.09 0.07

315.45*** 5.85*

3.56 8.78**

86

5.4. Empirical Results

indicate that information supply is also stationary around a deterministic trend. This finding is contradictory to information supply proxies in previous literature. For example, Vlastakis and Markellos (2012) use data from Thomson Reuters News Scope Database and find no deterministic trend. An explanation for these differing findings is that the increasing internet adoption has lead to accelerated news reporting in freely available sources as used in this study, whereas the number of news of professional financial data providers such as Thomson Reuters has remained more constant. I adopt the same methodology as for information demand to detrend and deseasonalize information supply. F-tests for equality of means find no evidence for seasonality in the adjusted time series. Further, Jarque-Bera normality tests provide evidence that the majority of adjusted information supply time series is normal-distributed. In all further analysis, I use the detrended and deseasonalized information supply time series (referred to as ‘information supply’).

5.4.2

Correlation and Causality

Next, I evaluate the causal relationship between information demand and information supply. Potential interlinkages between the two variables are not straightforward and likely have undergone substantial changes with the universal internet adoption of today’s information society. Different arguments can be put forward as to the causal relationship between the two variables: First, information supply might predict future information demand. The underlying argument is that market participants first consume investing-related information in news media (information supply), which then triggers a demand for more information. Alternatively, the causal relationship might go from information demand to information supply. This would be the case, if information is already spreading among market participants, before it reaches official news sources. For example, rumors about firm-specific developments might cause market participants to gather more detailed information online, before this information is available to news providers and thus only subsequently incorporated in information supply. Further, it also seems plausible that information demand and information supply interact dynamically.

Chapter 5. Retail Investor Information Demand

87

For example, if rumors spreading on the internet are picked up by the media, this is likely to stir interest among a greater number of people, which in turn again can lead to increased information demand. First, the results of a simple correlation analysis between information demand and information supply imply a positive contemporaneous relationship between the two variables for the majority of stocks (see Table 5.5). To study the causal relationship between information demand and information supply, I perform Granger-causality tests separately for each stock and the index.21 Specifically, I determine the most suitable lead-lag framework based on the Aikaike Information Criterion (AIC).22 The results are also reported in Table 5.5. At the 5% level of statistical significance, I find a bi-directional relationship for 3 sample stocks. For 10 (5) stocks, I find information demand (supply) to Granger-cause information supply (demand). On the index-level, I also find information demand to Grangercause information supply. These results do not allow for unambiguous conclusions, as to the direction of the relationship. However, they show a tendency of information demand to Granger-cause information supply. This implies that information discovery often takes place via alternative information channels (e.g. discussion boards, blogs, social media), before it is incorporated in news media. This finding is in line with a growing body of literature that finds valuable information content in alternative online sources.23 Finally, the results indicate that information demand and information supply capture something different. For example, the average information demand for Aixtron SE is the second highest in the sample. At the same time, this stock is only rarely covered by the news media. Such differences between information demand and information supply seem in line with economic intuition. For example, certain firm-specific topics might be highly newsworthy to the media (and thus be incorporated in information supply) and at the same time only hardly be interesting for retail investors (and thus only have little impact on information demand). Given these findings, 21 22 23

See Granger (1969) for details on the methodology. For robustness, I also employ the Schwarz Information Criterion (SIC), which yields similar results. See Akaike (1974) and Schwarz (1978), respectively. See, for example, Antweiler and Frank (2004) and Bollen et al. (2011).

88

5.4. Empirical Results

Table 5.5: Correlations and Granger Causalities of Information Demand and Supply This table reports Pearson correlation coefficients between information demand (ID) and information supply (IS). Further, it shows the results of GrangerCausality tests between the two variables. Aikaike’s information Criterion (AIC) is used to determine the most suitable lead-lag model for each security. F-tests are then employed to determine whether lagged values of information demand (information supply) have predictive power on information supply (information demand). */**/*** indicate statistical significance at the 10%/5% and 1% level, respectively. Correlation Coef.

IS → ID F-Test

ID → IS F-Test

Siemens AG Volkswagen AG SAP AG E.ON SE BASF SE Daimler AG Deutsche Telekom AG Bayer AG Allianz SE Deutsche Bank AG BMW AG RWE AG EADS N.V. Deutsche Post AG ThyssenKrupp AG MAN SE Commerzbank AG Infineon Technologies AG Wacker Chemie AG Deutsche Lufthansa AG AIXTRON SE Sky Deutschland AG SolarWorld AG Q-Cells SE Nordex SE

-0.07 0.20** 0.19** 0.38*** 0.22*** 0.20** 0.17** 0.26*** -0.03 0.03 0.06 0.18** 0.12 -0.04 0.10 0.18** -0.36*** 0.19** 0.07 0.18** 0.12 0.48*** 0.24*** 0.30*** 0.27***

3.27* 4.02*** 0.06 3.53* 1.77 0.18 0.73 2.84* 6.68** 1.92 0.36 2.81* 0.04 0.11 4.69** 12.3*** 0.72 3.44** 6.68*** 1.75 5.28*** 3.46** 0.57 2.57* 2.05

0.37 3.71* 3.95*** 7.97*** 1.11 2.99** 4.60** 4.96*** 0.18 6.40*** 0.53 2.54* 0.18 3.07* 2.43* 4.71** 9.89*** 1.18 3.4* 1.99 10.35*** 3.23** 4.36** 5.02*** 4.34**

Stocks Pooled Index

0.15* -0.16**

2.87* 1.64

3.74** 5.20***

Chapter 5. Retail Investor Information Demand

89

investigating the financial market impact of information demand poses an important research question. In the following, I focus on the link between information demand and trading behavior of retail investors in investment products and leverage products.

5.4.3

Trading Activity

The first question is, whether information demand has an impact on trading activity and whether it affects investment activity (trading in investment products) differently than speculative trading activity (trading in leverage products). To investigate this question empirically, I run regressions separately for investment products and leverage products. I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects.24 For robustness, I run the regression twice for each product category, once with trading volume as dependent variable and once with number of trades as dependent variable.25 The regression model is specified as follows:

Measurei,t = α + β1 IDi,t + β2 lag(ID)i,t + β3 ISi,t + β4 lag(IS)i,t +β5 ID Indext + β6 lag(ID Index)t + β7 IS Indext (5.1)  +β8 lag(IS Index)t + βj Controlsj,i,t + i,t , j

where ID (IS ) denotes firm-specific information demand (supply) and ID Index (IS Index ) denotes market-wide information demand (supply). As controls, I use firm-specific and market-wide realized volatility, firmspecific and market-wide absolute return and firm-specific market capitalization. The results are reported in Table 5.6. 24

25

The main benefit of this methodology is that it allows to account for correlations in the regression residuals across two dimensions. Here, the residuals might be correlated across sample firms (firm effect) and across time (week effect). Hence, clustering for those two dimensions according to Thompson (2011), allows me to obtain robust standard errors. For details on the methodology, see Petersen (2009) and Thompson (2011). Trading volume and number of trades are stationary during the sample period.

90

5.4. Empirical Results

I find a significant positive contemporaneous relationship between firmspecific information demand and trading activity in speculative products. Further, high firm-specific information demand in a given week predicts high trading volume in leverage products in the following week. In contrast, I find a negative link (contemporaneously and predictive) between marketwide information demand and firm-specific trading activity in leverage products. This implies that firm-specific and market-wide information demand are differently motivated. Whereas firm-specific information demand is likely to be associated with gathering further information as basis for investment decisions, market-wide information demand might be driven by uncertainty regarding current market developments. This explanation seems particularly plausible, given the ongoing sovereign debt crisis during the sample period: Information demand of retail investors in the stock market index is likely driven by interest in the market reaction to political and economic developments. Since the economic and political situation in the sample period is largely dominated by a high level of uncertainty, it seems plausible that market-wide information demand has a negative effect on trading activity. For investment products, I do not find a significant relationship between firm-specific information demand and trading activity. In contrast, information supply has a significant positive impact on trading activity. This indicates that long-term investing is associated with a comparably lower information demand than short-term speculating. An explanation for these differing results could be that long-term investors are less interested in current firm-specific short-term developments and thus display lower information demand. Alternatively, it can be argued that investment decisions are comparably less information-based than speculative trading decisions. Lastly, it could be the case that long-term investments tend to be based on advise from financial institutions, whereas speculative decisions are taken more independently.26 In line with speculative trading activity, 26

This reasoning is based on the legal duty for financial advisors to inform customers about the risks involved with financial products. The Lehman bankruptcy was followed by damage suits by many retail investors, claiming that banks had neglected their legal duty to inform about risks associated with structured products. Hence, it can be assumed that banks have adopted a conservative stance and refrain from actively promoting risky leverage products. For details on the legal duty

0.04 0.02 0.29 0.28 -0.23 -0.61 0.13 -0.07 0.36 0.93 0.95 3.04 0.96 4.69

0.44

ID lag(ID) IS lag(IS) ID Index lag(ID Index) IS Index lag(IS Index) RV RV Index ABS Ret ABS Ret Index Market Cap Constant

R2

(0.40) (0.25) (2.32)** (2.36)** (-0.73) (-3.68)*** (0.66) (-0.45) (0.43) (0.94) (1.72)* (3.42)*** (8.27)*** (4.28)*** 0.34

0.02 0.03 0.21 0.21 0.03 -0.66 0.18 0.06 0.85 0.07 1.12 1.72 0.64 -2.35

(0.35) (0.60) (6.10)*** (5.99)*** (0.18) (-4.08)*** (1.31) (0.43) (4.96)*** (0.18) (1.55) (1.38) (38.56)*** (-12.11)***

Investment Products Volume Trades Coef. t-stat. Coef. t-stat.

0.35

0.15 0.11 0.14 0.07 -0.46 -0.53 0.46 0.32 1.70 -0.23 2.69 0.87 0.69 6.00

(3.66)*** (2.60)*** (3.98)*** (2.14)** (-2.33)** (-3.32)*** (3.39)*** (2.37)** (9.33)*** (-0.62) (3.57)*** (0.70) (39.03)*** (29.24)***

0.32

0.13 0.08 0.12 0.08 -0.43 -0.53 0.35 0.26 1.76 0.09 2.60 1.20 0.54 -1.06

(3.55)*** (2.36)** (4.27)*** (2.81)*** (-2.55)** (-3.82)*** (3.00)*** (2.36)** (10.56)*** (0.29) (3.60)*** (1.11) (37.68)*** (-6.15)***

Leverage Products Volume Trades Coef. t-stat. Coef. t-stat.

This table reports regression results of weekly information demand on the trading activity of retail investors in investment and leverage products with individual stocks as underlying. I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects. I analyze both, effects of firm-specific and market-wide information demand on trading activity. As controls, I use information supply, firmspecific and market-wide realized volatility, stock-specific and market-wide absolute return and market capitalization. For robustness, the regression is run twice, once with weekly trading volume as proxy for trading activity and once with the number of trades as proxy. ∗/ ∗ ∗/ ∗ ∗∗ denotes significance on the 10%/5% and 1% level, respectively. T-statistics are reported in parentheses.

Table 5.6: Information Demand and Trading Activity

Chapter 5. Retail Investor Information Demand 91

92

5.4. Empirical Results

market-wide information demand negatively predicts trading activity in investment products.

5.4.4

Order Submission Strategies

In this section, I analyze whether a relationship exists between information demand and aggregated order submission behavior of retail investors. As described in Section 4.1, retail investors trading in the EUWAX segment can generally choose to submit limit orders or market orders. This choice allows me to investigate whether an increased information demand influences retail investors’ desire for immediate order executions. I measure weekly demand for immediate order execution (order aggressiveness) as the ratio of marketable orders (that is, orders executed within 10 seconds after submission) to all submitted orders. In addition, investors can submit stop orders. For buy orders, this means that an order is executed at the next best price when the prespecified stop buy price is touched. For example, this order type can be beneficial if an investor does not want to watch the stock market actively, but wants to ‘jump on the bandwagon’ based on the expectation that prices will further rise, once they move beyond a certain price. For sell orders, the order is executed at the next best price after the pre-specified stop loss threshold is touched. This order type can be useful for investors who want to automatically exit the market to limit losses, based on the expectation that prices will further drop, once they fall below a certain price. Generally, stop orders are useful for investors who are uncertain regarding the direction of security prices. Put differently, an investor who is highly confident that the market will move in a certain direction has no need to use stop orders. Thus, a higher (lower) ratio of stop orders to submitted orders can be interpreted as an increase (decrease) in uncertainty to inform and a discussion of Lehman damage suits in Germany, see Buck-Heck (2011). Further, leverage products tend to be traded by more experienced retail investors (before being allowed to trade these products, retail investors are legally required to state that they have considerable trading experience), who tend to take independent investment decisions.

Chapter 5. Retail Investor Information Demand

93

among investors. I measure this type of uncertainty as the weekly ratio of stop orders to all submitted orders (order uncertainty). Previous literature on order submission strategies of investors predominantly argues that informed investors trade more aggressively and tend to use market orders.27 Hence, I expect a positive relationship between information demand and order aggressiveness since retail investors who actively obtain information on a financial topic are likely to be better informed than others. Alternatively, it can be argued that retail investors who have engaged in information gathering at least tend to be more confident that they are well-informed and trade more aggressively as a result. Due to the higher execution risk for leverage products (small fluctuations of the underlying can result in substantial price fluctuations of leverage products), I particularly expect a positive relationship between information demand and order aggressiveness for leverage products. I run the regression model as outlined in Section 5.4.3, but use the two order type variables order aggressiveness and order uncertainty as dependent variables. Further, I add an additional control variable for the number of trades. Again, I perform the regressions separately for investment products and leverage products. Table 5.7 shows the results. I find a positive contemporaneous relationship between stock-specific information demand and order aggressiveness in leverage products. I attribute this to a decrease in information uncertainty: The information demand of retail investors and the associated information gathering leads to a higher level of information about the respective company. As a result, information uncertainty is reduced and investors have incentives to trade more aggressively. Although the contemporaneous relationship is also positive for investment products, it is economically and statistically insignificant. Further, market-wide information demand has a strong significant and economic positive relationship with the order aggressiveness of leverage products. This also applies to investment products (only contemporaneously). I attribute this effect of market-wide information demand to 27

See, for example, Angel (1994) and Harris (1998), who argue that informed traders tend to use market orders, whereas limit orders are used by liquidity traders. However, as argued by Kaniel and Liu (2006), it should also be noted that informed investors might prefer limit orders.

0.00 0.00 -0.01 -0.02 0.08 -0.01 0.03 0.07 -0.09 -0.36 0.00 -0.37 0.00 0.02 0.67

0.14

ID lag(ID) IS lag(IS) ID Index lag(ID Index) IS Index lag(IS Index) RV RV Index ABS Ret ABS Ret Index Market Cap Trades Constant

R2

(0.71) (0.16) (-2.75)*** (-3.60)*** (3.66)*** (-0.36) (1.71)* (3.75)*** (-4.28)*** (-7.85)*** (-0.05) (-2.30)** (-0.76) (7.75)*** (25.33)*** 0.33

0.00 0.00 0.01 0.00 0.03 -0.03 -0.01 0.01 0.03 -0.04 0.09 0.01 0.00 -0.03 0.14

(-1.44) (0.18) (3.01)*** (0.36) (3.25)*** (-4.10)*** (-1.44) (1.65)* (4.00)*** (-2.36)** (2.39)** (0.24) (2.24)** (-11.69)*** (13.06)***

Investment Products Order Aggressiveness Order Uncertainty Coef. t-stat. Coef. t-stat.

0.15

0.01 0.00 0.00 0.00 0.11 0.04 -0.02 0.03 0.06 -0.32 0.05 -0.23 -0.01 0.00 0.44

(1.77)* (-0.71) (-1.36) (-2.22)** (8.76)*** (3.64)*** (-2.59)*** (2.86)*** (3.61)*** (-12.57)*** (0.91) (-2.46)** (-4.75)*** (1.17) (27.80)***

0.12

0.00 0.00 -0.01 0.00 0.04 -0.02 0.00 0.01 -0.03 -0.09 0.04 0.05 0.00 -0.01 0.15

(-0.12) (-0.10) (-3.52)*** (-2.25)** (5.35)*** (-3.61)*** (-0.11) (1.41) (-3.78)*** (-6.26)*** (1.63) (0.98) (3.52)*** (-12.77)*** (15.77)***

Leverage Products Order Aggressiveness Order Uncertainty Coef. t-stat. Coef. t-stat.

This table reports regression results of weekly information demand on order aggressiveness and order uncertainty of retail investor trading in investment and leverage products with individual stocks as underlying. I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects. I analyze both, effects of firm-specific and market-wide information demand on order types. As controls, I use information supply, firm-specific and market-wide realized volatility, stock-specific and market-wide absolute return, market capitalization and number of trades. ∗/ ∗ ∗/ ∗ ∗∗ denotes significance on the 10%/5% and 1% level, respectively. T-statistics are reported in parentheses.

Table 5.7: Information Demand and Order Submission Strategies

94 5.4. Empirical Results

Chapter 5. Retail Investor Information Demand

95

an increased economic uncertainty among investors and hence, an increased desire to enter or exit the market immediately. I find no impact of stock-specific information demand on order uncertainty. However, contemporaneous market-wide information demand is associated with an economically and statistically significant increase in order uncertainty for both, investment products and leverage products. This supports prior reasoning that market-wide information demand reflects stock market interest attributable to economic uncertainty, which in turn induces retail investors to take safety precautions in the form of stop orders. In sum, I provide evidence for a positive relationship between information demand and both, order aggressiveness and order uncertainty. This relationship particularly applies to leverage products. Further, the more pronounced impact of market-wide information demand on order submission again indicates that market-wide information demand influences retail investor behavior more thoroughly than stock-specific information demand.

5.4.5

Positioning

The information demand variable per se does not reveal qualitative information about the underlying drivers that cause retail investors to display an increased interest. For example, I do not know if the information demand towards a topic is fueled by negative or positive events associated with this topic. Similarly, information supply only quantifies the number of news, but not the informational content (e.g. positive or negative). This section investigates, if regardless of the lacking qualitative informational content of the variable, a relationship between information demand and systematic trading patterns of retail investors exists. My data set allows for a more differentiated analysis than previous studies since structured products enable investors to benefit from rising as well as falling prices. This allows for a more complete picture than the analysis of trading in ordinary equity since investors cannot only express ‘short interest’ by selling equities they own, but deliberately enter short positions. The possibilities for betting on falling prices are particularly pronounced for leverage

96

5.4. Empirical Results

products. Theoretically, I do not expect a significant relationship between neither information demand nor information supply on the positioning of retail investors in leverage products. However, for investment products, the arguments of Barber and Odean (2008) are more valid and it is more likely to find a positive relationship between information demand and long positioning. To determine, on which side of the market retail investors are in a given security, I introduce the following imbalance measure that contains both, the option type of the products and the trade direction.28 I measure weekly imbalances separately for investment products and leverage products. For robustness, I use both, trade-based and volume-based imbalance measures in all analysis. Specifically, the imbalance measures are defined as follows:  IM B V olumei,t =

i,k,t

(vi,k,t × di,k,t × oi,k,t )  , i,k,t vi,k,t

 IM B T radesi,t =

i,k,t

(di,k,t × oi,k,t )  ,

(5.2)

(5.3)

i,k,t

where vi,k,t denotes transaction volume of executed order k with security i as underlying (trade price × size), di,k,t denotes the trade direction of order k (buy: di,k,t = 1, sell: di,k,t = −1), and oi,k,t denotes the option type of the traded product (call: oi,k,t = 1, put: oi,k,t = −1 ). The values of the resulting weekly imbalance measures are always between -1 (100% short positioning / closing of long positions) and 1 (100% long positioning / closing of short positions). I run the regression model as outlined in Section 5.4.3 and use the imbalance measures as dependent variables. Again, I perform the regressions separately for investment products and leverage products. Table 5.8 reports the results. I find no relationship between information demand and positioning in leverage products. For investment products, information demand positively predicts trading imbalances. An explanation is that, whereas lever28

This measure follows similar methodology as the EUWAX sentiment index, which is based on leverage products with DAX as underlying. For details, see Burghardt (2010).

0.01 0.04 -0.02 -0.04 -0.37 0.04 0.10 -0.02 -0.01 -0.30 -0.69 -0.94 0.01 0.17 0.08

ID lag(ID) IS lag(IS) ID Index lag(ID Index) IS Index lag(IS Index) RV RV Index ABS Ret ABS Ret Index Market Cap Constant

R2

(0.38) (2.57)** (-1.70)* (-2.76)*** (-5.15)*** (0.65) (1.94)* (-0.39) (-0.12) (-2.18)** (-2.81)*** (-1.94)* (1.37) (2.16)** 0.13

0.01 0.03 -0.01 -0.02 -0.42 0.14 0.08 -0.10 0.07 -0.64 -0.89 -0.67 0.03 0.07

(1.03) (1.75)* (-1.06) (-1.45) (-6.99)*** (2.77)*** (1.70)* (-2.44)** (1.14) (-5.60)*** (-3.55)*** (-1.67)* (5.02)*** (0.96)

Investment Products IMB Volume IMB Trades Coef. t-stat. Coef. t-stat.

0.03

0.00 0.01 0.02 0.01 -0.12 0.00 -0.04 0.04 0.21 0.30 -0.67 -0.65 -0.01 0.01

(-0.02) (0.73) (1.95)* (1.73)* (-2.47)** (0.09) (-1.09) (0.99) (4.62)*** (3.36)*** (-4.06)*** (-2.14)** (-1.98)** (0.23)

0.03

0.01 0.01 0.02 0.01 -0.10 0.01 -0.01 0.05 0.19 0.20 -0.62 -0.39 0.01 -0.08

(0.76) (0.86) (2.60)*** (2.08)** (-2.65)*** (0.19) (-0.26) (1.88)* (5.57)*** (2.79)*** (-4.93)*** (-1.57) (1.41) (-1.68)*

Leverage Products IMB Volume IMB Trades Coef. t-stat. Coef. t-stat.

This table reports regression results of weekly information demand on retail investor positioning as measured by weekly order imbalances in investment and leverage products with individual stocks as underlying. I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects. I analyze both, effects of firm-specific and market-wide information demand on imbalances. As controls, I use information supply, firm-specific and market-wide realized volatility, stock-specific and market-wide absolute return and market capitalization. ∗/ ∗ ∗/ ∗ ∗∗ denotes significance on the 10%/5% and 1% level, respectively. T-statistics are reported in parentheses.

Table 5.8: Information Demand and Positioning

Chapter 5. Retail Investor Information Demand 97

98

5.4. Empirical Results

age products are commonly used to ‘go short’, this is not the case for investment products.29 Hence, the arguments of Barber and Odean (2008) that increased interest in a company leads to net buying activity of retail investors apply to investment products. I find market-wide information demand to be negatively associated with trading imbalances for both leverage and investment products. This is consistent with my previous reasoning that market-wide information demand is most pronounced when ‘bad things’ happen. Accordingly, investors seem induced to sell products or enter short positions in weeks with high market-wide information demand.

5.4.6

Return Predictability

Da et al. (2011) and Bank et al. (2011) document a positive return pressure of information demand in the short-term that is subsequently reversed. However, they also find that this positive price pressure is driven by the smaller capitalized stocks in their sample and does not apply to blue chips. The results of the positioning analysis in Section 5.4.5 indicate that retail investor trading in leverage products does not contribute to an upward pressure on security prices. However, there is some evidence that retail investors tend to enter long positions in investment products in response to increased information demand, which might contribute to an upward price pressure.30 By analyzing abnormal stock returns, I can also shed light on the question whether an upward price pressure of information demand exists and direct equity trading differs from trading in structured products. To analyze the impact of information demand on returns, I adopt similar methodology as Da et al. (2011) and regress future weekly abnormal stock returns at different horizons on information demand, information supply and various control variables. I compute abnormal returns for each stock as the return 29

30

E.g. in my sample there are only 400 products (reverse bonus/discount certificates) with stocks as underlying that benefit from falling prices as opposed to 23,646 in leverage products. Trading in structured products has no direct impact on the prices of the respective underlyings. However, banks that have issued structured products, are likely to hedge their exposure, which then affects security prices.

Chapter 5. Retail Investor Information Demand

99

in excess of the return of the German stock market index DAX.31 Again, I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects. I run the following regression model and use abnormal returns as dependent variable: Measurei,t = α + β1 IDi,t + β2 ISi,t + β3 ID Indext + β4 IS Indext  + βj Controlsj,i,t + i,t , (5.4) j

where ID (IS ) denotes firm-specific information demand (supply). As controls, I use firm-specific realized volatility, absolute return and market capitalization. The regression results are reported in Table 5.9. I do not find stock-specific information demand in a given week to contribute to an upward return pressure in any of the subsequent 4 weeks. Since the sample consists of large-capitalized stocks, this finding is in line with Da et al. (2011), who only find evidence of price pressure caused by information demand for small-capitalized stocks. In contrast, information supply predicts negative abnormal returns throughout the following 4 weeks.32 This finding corresponds to Fang and Peress (2009), who find a persistent negative link between media coverage and returns. A potential explanation for this negative relationship is a negative news bias, which means that firm-specific information supply is dominated by bad news.33 If bad news are then only gradually incorporated in asset prices after becoming available, the result is negative abnormal returns.

5.5

Conclusion

The intertemporal analysis reveals that causality tends to run from information demand to information supply on the stock-level as well as the 31

32 33

The alternative would be to derive abnormal returns from the market model or the Fama-French three factor model. For robustness, I repeat the analysis based on abnormal returns from a market model and obtain corresponding results. However, the R2 values of all regression specifications imply that my regression models can only explain a very low fraction of the variation in abnormal returns. See, for example, Green et al. (2012), who document a bad news bias in the business press.

0.000 -0.003 -0.003 0.033 0.001 -0.013

0.004

ID IS RV ABS Ret Market Cap Constant

R2

(0.21) (-2.49)** (-0.36) (0.87) (1.41) (-1.32)

Week 1 Coef. t-stat.

0.008

0.000 -0.003 0.001 -0.075 0.001 -0.006

(0.15) (-2.68)*** (0.14) (-1.95)* (0.88) (-0.57)

Week 2 Coef. t-stat.

0.006

0.000 -0.004 -0.005 0.027 0.001 -0.011

(0.29) (-3.31)*** (-0.75) (0.89) (1.42) (-1.25)

Week 3 Coef. t-stat.

0.005

0.000 -0.003 0.000 0.003 0.002 -0.015

-(0.06) (-2.50)** (-0.02) (0.07) (1.73)* (-1.65)*

Week 4 Coef. t-stat.

This table reports the results of regressing future weekly abnormal stock returns at different horizons on information demand, information supply and the control variables realized volatility, absolute returns and market capitalization. Abnormal returns for each stock are computed as the return in exccess of the German stock market index DAX. I use robust clustered standard errors according to Thompson (2011), which allow me to simultaneously cluster for firm and week effects. ∗/ ∗ ∗/ ∗ ∗∗ denotes significance on the 10%/5% and 1% level, respectively. T-statistics are reported in parentheses.

Table 5.9: Information Demand and Returns

100 5.5. Conclusion

Chapter 5. Retail Investor Information Demand

101

market-level. Hence, information discovery often takes place via alternative information channels before it is incorporated in traditional news media. I find a significant positive effect of firm-specific information demand on contemporaneous and future speculative trading activity. No such relationship exists for long-term oriented investing activity. This implies that investing activity either relies less on information gathering online or that investing activity is comparably less information-driven. Market-wide information demand, however, leads to subsequently lower trading activity in speculative products but at the same time also to more aggressive trading by retail investors. This particularly applies to speculative trading. Further, high market-wide information demand is associated with higher order uncertainty, which supports the reasoning that market-wide information demand is driven by uncertainty of retail investors about overall market developments. I find no effect of firm-specific information demand on retail investor positioning in leverage products. However, market-wide information demand is associated with short interest expressed by retail investors. I argue that this behavior stems from increased uncertainty with regards to the overall market development. I do not find retail investor information demand to exert an upward price pressure on equity prices. Generally, firm-specific and market-wide information demand capture something different. Whereas firm-specific information demand is driven by gathering and using information with regards to trading activities, market-wide information demand seems to be driven by interest in the financial market impact of economic or political developments. Further, the impact of market-wide information demand on trading behavior indicates that retail investor information demand reflects uncertainty regarding the overall market development. In all analyses, the effect of market-wide information demand is economically and statistically more pronounced than the effect of firm-specific information demand. This supports the theoretical reasoning of Peng and Xiong (2006) that market-wide information demand often plays a more prominent role in the financial decision making process of retail investors than firm-specific information demand: The speculating as well as the investing behavior in structured products is more influenced by overall market developments than by company fundamentals. As a result,

102

5.5. Conclusion

the informational efficiency of the market for structured products seems limited.34 A limitation of my analysis is that information demand time series are only available at weekly frequency. Since information and financial markets evolve rapidly, a more detailed analysis of the dynamic interlinkages between the variables would require data at higher frequency. Currently, no such data is available at daily or intra-daily frequency over extended periods. Finally, the fact that I focus on the aggregated trading behavior has the advantage that I can draw conclusions for the general population of retail investors trading structured products. On the other hand, this also somewhat poses restrictions on the scope of the empirical analysis. I do not have information about the specific investor behind the trades in the sample and thus cannot assess the specific motivation behind the observed trading behavior. Since the general population of retail investors in my sample is far from being a homogeneous group, more specific and detailed conclusions would require a more detailed data set.

34

Incorporating market-wide information in investment decisions does not automatically hint at a low informational efficiency. However, investors who want to speculate on overall market fluctuations would do well to trade products with DAX (or other indices) as underlying, to reduce the idiosyncratic risks associated with speculating on individual stocks. In contrast, when trading structured products with individual stocks as underlying, stock-specific information should play a greater role.

103

Chapter 6

Media Sentiment and Leveraged Trading 6.1

Introduction

The impact of information arrival on investor behavior has likely undergone drastic changes in recent years. The rapid technological developments in the information and communication industry have greatly influenced the way how new information is disseminated to traders, how information is processed and how it is translated into trading decisions. Nowadays, even retail investors have access to real-time media feeds that are delivered to their mobile devices. Previous work, such as Riordan et al. (2013) and Gross-Klussmann and Hautsch (2011), has shown that professional traders, as well as the growing number of algorithmic traders indeed rapidly act on news. However, the impact of intraday newswire messages on the trading behavior of retail investors is still unclear. To address this question, I analyze how retail investors trade with leverage products around Dow Jones Newswire messages. Specifically, I analyze how the trading intensity changes around news and how retail investors adjust their order submission strategies. Leverage products are particularly well-suited for this purpose. First, the speed of information processing and the timely reac-

104

6.1. Introduction

tion to new information is highly pronounced for leveraged positions since market movements triggered by new information have an amplified impact on price movements. In addition, leveraged products are typically used for short-term speculation on current market developments. Thus, new pieces of information are likely to play an important role for trading decisions in these products. Second, leverage products are typically traded by experienced investors, who trade actively. Hence, these investors are likely to closely watch current market developments and incorporate new information in their trading decisions. In this chapter, I investigate various aspects of information arrival and trading in leverage products. I cluster the newswire messages in my sample by two distinct categories: The sentiment of news (positive, neutral, negative), as measured by the textual tonality contained in news messages, as well as the type of news (normal, mandatory, analyst), as classified by Dow Jones. This allows me to study potential asymmetric reactions on the trading behavior of retail investors, depending on the sentiment of news and the type of news. Based on this classification, I analyze two aspects of intraday retail investor trading in leverage products: The trading intensity and the order submission strategy around public information arrival. I find that retail investors trade more intensively around news. They enter positions to speculate on the outcome of news, ahead of information arrival and also engage in feedback trading after news releases. However, this only applies to newswires that carry a positive or a neutral sentiment. No such effect is found for negative news, which implies that retail investors are intimidated by negative news and refrain from trading more actively around them. Further, I find that after positive news, retail investors increasingly trade in the direction of the news sentiment (increase in long positions), whereas after negative news they trade more in the opposite direction (increase in long positions). My analysis of order submission strategies indicates that retail investors aggressively trade on news articles with a negative sentiment. After negative analyst news, I obtain opposing results: Whereas retail investors trade more aggressively in calls, they trade less aggressively in puts. Ahead of positive mandatory news, retail investors have an increased desire to take

Chapter 6. Media Sentiment and Leveraged Trading

105

safety precautions by using stop orders intensively. I argue that this is due to an increased uncertainty regarding the outcome of scheduled mandatory news. In sum, I can conclude that retail investor behavior is strongly affected by public information arrival. However, depending on the sentiment as well as on the type of news, retail investors adjust their trading behavior in leverage products differently. The remainder of this chapter is structured as follows: Section 6.2 provides an overview of previous research on retail investors and information arrival. I describe the news data and the trading data in Section 6.3. In Section 6.4, I report the empirical results of my analysis on information arrival and trading behavior in leverage products. Section 6.5 concludes.

6.2

Literature Review

Generally, two strands of literature are related to the empirical analysis in this chapter: (1) The role of media sentiment in financial markets and (2) the trading behavior of investors around news. The Role of Media Sentiment in Financial Markets A growing number of studies attempts to quantify the textual content of financial news, to analyze how their sentiment affects financial markets. The broad research question behind all these studies is to provide insights into the (intertemporal) relationship between public information arrival and the stock market. Tetlock (2007) analyzes the interaction between the influential Wall Street Journal column ‘Abreast of the Market’ and the U.S. stock market on a daily basis. He semantically quantifies the textual content of the columns and finds that high negativity exerts a downward pressure on stock returns that is subsequently reversed. Also, trading volume is found to increase after columns with abnormally high or low levels of negativity. Tetlock et al. (2008) build on the analysis of Tetlock (2007) and investigate whether the textual content in Wall Street Journal and Dow Jones News Service news comprises valuable information to predict the stock market. Their findings suggest that the linguistic content in news media indeed

106

6.2. Literature Review

contains valuable information about fundamentals. Further, they show that particularly negative sentiment does have a high predictive capability regarding earnings announcements and stock returns. Finally, Loughran and McDonald (2011) analyze 10-K filings and confirm that the textual content of financial information can contribute to a better understanding of stock returns. Given the growing importance of social media sources in financial news dissemination, various studies also focus on the link between social media sentiment and financial markets. Antweiler and Frank (2004) study the relationship between the informational content of stock message boards onYahoo!Finance and the U.S. stock market. They find that firms with high posting volume exhibit increased trading activity and higher return volatility. Bollen et al. (2011) find that investor sentiment in Twitter can predict Dow Jones Industrial Average (DJIA) price movements. A positive mood about the DJIA leads to rising prices, whereas a negative mood predicts falling prices. The finding that social media sentiment has an effect on financial markets already hints at a link between public information arrival and retail investor trading since the social media activity on financial topics is likely to be driven by retail investors to a large extent. Investor Behavior around News Previous literature that specifically addresses the news trading behavior of retail investors includes Kaniel et al. (2012), who analyze retail investor trading around earnings announcements. They find that retail investor trading activity increases around earnings announcements and provide some evidence for skillful trading.1 More broadly, Kelley and Tetlock (2013) use all news articles from Wall Street Journal and Dow Jones News Service on U.S. stocks from 2003 to 2007 and analyze whether retail investors can correctly predict the tone of these news, as well as abnormal returns. They find that daily order imbalances of individual investors can predict positive abnormal stock returns at the monthly horizon, with no evidence of return reversal. They find partial evidence that retail investors can correctly anticipate the tone of news, which is the case for aggressive orders by retail investors. Further, Tetlock (2011) finds for the U.S. that stock markets react to stale news, but the price impact is partially reversed in subsequent weeks. This effect 1

For details, see Section 2.5 of this thesis.

Chapter 6. Media Sentiment and Leveraged Trading

107

is attributed to retail investors overreacting to stale information. Barber and Odean (2008) analyze the effect of attention on retail investor trading and use news coverage as measured by Dow Jones News Services as one of their proxies. They find that retail investors are attracted by news and are net buyers of stocks on days with news coverage for the respective stocks. Finally, there is a trend in the literature to investigate the market reaction to news at a high-frequency time horizon. This seems adequate, given that in today’s information society information often becomes available in real-time and can thus easily be incorporated into trading decisions at high speed. This is all the more the case, since many news provider have started to provide machine-readable news feeds that can be processed and transformed into trading decisions by algorithmic traders within milliseconds.2 Gross-Klussmann and Hautsch (2011) analyze high-frequency market reactions to stock-specific intraday news. They use data from the Reuters NewsScope Sentiment Engine, that is already pre-processed in the sense that news messages are tagged with sentiment based on an automated linguistic analysis performed by Reuters. Their findings indicate that intraday volatility and trading volume are influenced by stock-specific news messages. Finally, Riordan et al. (2013) study the impact of news on trading activity, price discovery and liquidity. Their results show increased trading activity around news, irrespective of their sentiment. Further, they find that liquidity increases around positive and neutral news, but decreases around negative news. They argue that negative news contain the most valuable information and induce the strongest market reactions.

6.3 6.3.1

Data News Data

My news data set consists of stock-specific Dow Jones Newswires (DJNS) messages that cover constituents of the German blue-chip index DAX dur2

See, for example the following commercial services: Thomson Reuters News Analytics, Dow Jones Elementized News Feed, Deutsche Boerse AlphaFlash.

108

6.3. Data

ing the sample period from April 2010 to November 2012.3 Each news is tagged with all firms that are covered in the text. However, in many cases news are tagged with a company although the article mostly deals with related companies or topics (e.g. competitors). Hence, to ensure that news are actually directed at the company in question, I restrict my analysis to those news that have the respective company name in their headline. Further, each news contains information about the type of news. This allows me to classify each news message into the following categories: normal, mandatory, or analyst news. Normal news are regular financial news articles (e.g. articles published in the Wall Street Journal), mandatory news are mandatory filings that are legally required by listed companies (e.g. directors’ dealings), and analyst news contain the assessments of sell-side analysts. To exemplarily illustrate the information contained in the DJNS data, Table 6.1 provides a selection of individual news on the sample firm Adidas AG. For my empirical analysis, I only consider news that arrive during continuous trading. Further, I delete all stock-specific news that arrive within one hour after the previous message. The reason for doing this is that very often a brief message that contains the key facts is immediately released when a news story becomes available and then later followed by a more detailed version of the original message. Since, I am interested in novel news, the follow-up messages are useless for my analysis. Also, I delete all duplicates from the data set, which are messages on the same company with corresponding timestamps. Overall, this leaves me with 6,909 messages. Figure 6.1 reports the number of intraday news (2 month moving average) across the sample period. In a next step, these news messages are enriched with sentiment based on a semantic analysis of the textual content. Specifically, I adopt an automated dictionary-based approach, in which sentiment is determined based on a dictionary that classifies individual words into positive and negatives ones depending on their connotation. Such an approach has been adopted successfully in the financial context, for example, by Antweiler and Frank (2004), Tetlock (2007) and Tetlock et al. (2008). The main benefits of automated textual analysis, in comparison to manual analysis, are that they (1) allow to classify a large number of news and (2) the results are independent of any person3

I thank Interactive Data for providing access to the data.

Timestamp

2012/06/2909:55:14

2011/03/0210:47:31

2010/11/04/11:31:53

Company

Adidas AG

Adidas AG

Adidas AG

Mandatory

Analyst

Normal

Type

PRESS RELEASE: Adidas: Nine Months 2010 Results

MARKET TALK: A Good Opportunity To Buy Adidas

Adidas Raises 2011 Sales Guidance Despite Profit Slump In 4Q

Headline

Currency-neutral Group sales up 10% in Q3. Net income more than doubles in first nine months. adidas Group refines 2010 financial outlook. Earnings per share to increase between 10% and 15% in 2011....

Adidas (ADS.XE) drops 1.1% to EUR55.83 following weak fiscal 4Q results from its US competitor Nike (NKE), but this provides a good opportunity to buy adidas shares, Silvia Quandt Research analyst Mark Josefson says. Adidas is winning Nikes market share, he adds...

German sportswear and equipment maker Adidas AG (ADS.XE) Wednesday cautiously raised its 2011 sales guidance as net profit for the final quarter of 2010 slumped 64% due to higher marketing costs. All our brands scored with consumers in an improving worldwide economy in the fourth quarter, Adidas Chief Executive Herbert Hainer said in a statement. Adidas now expects 2011 sales to rise by a mid- to high-singledigit percentage range, after previously guiding for a mid-single-digit rise....

Textual Content

This table shows three novel intraday Dow Jones Newswires messages for the sample firm Adidas AG.

Table 6.1: Illustration of Dow Jones Newswires Data

Chapter 6. Media Sentiment and Leveraged Trading 109

110

6.3. Data

400

Number of News

350 300 250 200 150 100 50 0 Apr 10

Aug 10

Dec 10

Apr 11

Aug 11

Dec 11

Apr 12

Aug 12

Figure 6.1: Number of Intraday News This figure plots the monthly number of intraday news (2month moving average) during the sample period from April 2010 to November 2012.

ally induced biases that are likely to exist in manual analysis. However, the quality of this approach to semantic analysis heavily depends on the ability of the underlying dictionary to effectively capture the connotation in the respective context. A dictionary that has often been used in various domains to analyze textual sentiment is the Harvard-IV-4 dictionary, originally developed in the psychological and sociological context. However, Loughran and McDonald (2011) convincingly question the use of the Harvard dictionary in the financial context and argue that particular words often have different meanings in the financial context than in other areas of life. For example, the word ‘liability’ does not carry a negative meaning in the financial context, whereas in everyday use it is negatively afflicted. Thus, it cannot be used to accurately capture the meaning of words in the financial context. To overcome this shortcoming, Loughran and McDonald (2011) develop an alternative dictionary that is specifically directed at capturing the sentiment of words in the financial domain. Following their reasoning, I employ the Loughran and McDonald (2011) dictionary and measure the sentiment of each Dow Jones Newswire message as a ratio of

Chapter 6. Media Sentiment and Leveraged Trading

111

positive to negative words.4 Specifically, my news sentiment measure is defined as follows:   i,k posi,k − i,k negi,k , (6.1) N ews Sentimenti,k =  i,k (posi,k + negi,k ) where posi,k (negi,k ) denotes positive (negative) words in news message k on stock i according to the Loughran and McDonald (2011) word list. The resulting measure provides an indication of the stock-specific tonality of each news message in my sample.5 It can take values ranging from -1 (only negative words) to +1 (only positive words). Table 6.2 reports descriptives statistics of the news data, for each firm in the sample. Figure 6.2 provides an overview of the sentiment distribution of news messages across all sample stocks. A large number of news are entirely 1,600 1,400

Number of News

1,200 1,000

Analyst

800

Mandatory

600

Normal

400 200 0

Sentiment of News

Figure 6.2: Sentiment Distribution of Intraday News This figure plots the frequency of intraday news according to the raw news sentiment based on the Loughran and McDonald (2011) word list. Further, it illustrates the occurence of different news types by sentiment classes. 4 5

The full dictionary is available for download at http://www3.nd.edu/ mcdonald/Word Lists.html. It is important to note, that such a simple approach is unable to capture the precisely ‘correct’ sentiment of each word and news document. However, I believe that for the purpose of my analysis, it works efficiently by accurately providing a broad indication regarding the overall direction of sentiment (e.g. positive or negative) for the news in my sample.

Total Mean Median Max Min Standard Deviation

Siemens AG Deutsche Bank AG Volkswagen AG SAP AG Daimler AG E.ON SE Deutsche Lufthansa AG BMW AG RWE AG Deutsche Boerse AG Deutsche Telekom AG Bayer AG Commerzbank AG BASF SE Allianz SE ThyssenKrupp AG MAN SE Muenchener Rueck AG HeidelbergCement AG Merck KGaA Infineon Technologies AG Fresenius Medical Care AG Adidas AG Henkel AG Metro AG Linde AG Deutsche Post AG Fresenius SE Beiersdorf AG K+S AG 749,880 24,996 17,218 71,940 5,564 17,946

71,940 32,002 52,168 53,070 45,438 39,452 5,564 36,077 23,354 9,360 41,249 44,140 8,569 51,679 39,385 11,732 11,445 19,948 7,519 15,712 6,799 14,945 10,794 18,646 12,051 19,309 15,790 11,661 11,718 8,361

MV

6,909 230 188 684 40 158

684 624 433 371 368 343 339 324 317 276 269 257 238 217 196 179 169 133 132 130 120 116 115 101 100 99 94 73 52 40 5,219 174 145 523 33 124

523 412 337 102 310 310 306 269 290 229 225 192 182 162 144 153 146 109 113 90 92 45 100 84 71 33 73 44 37 36

Type of News All Normal

1,399 47 23 263 1 59

159 195 95 263 54 22 22 44 18 33 34 47 39 44 43 15 17 15 14 23 19 60 7 10 14 57 11 20 4 1

Mandatory

291 10 10 18 1 4

2 17 1 6 4 11 11 11 9 14 10 18 17 11 9 11 6 9 5 17 9 11 8 7 15 9 10 9 11 3

Analyst

2,343 78 54 273 12 71

263 190 184 273 160 66 64 160 46 77 61 117 61 82 66 30 37 23 24 48 40 59 32 27 23 45 32 26 15 12 1,459 49 39 169 13 36

169 112 108 36 102 56 71 65 69 40 52 41 26 65 31 27 51 15 58 30 27 13 49 18 25 37 19 15 18 14

Tone of News Sent>0 Sent=0

3,107 104 88 322 14 76

252 322 141 62 106 221 204 99 202 159 156 99 151 70 99 122 81 95 50 52 53 44 34 56 52 17 43 32 19 14

Sent0’, ‘Sent=0’ and ‘Sent 0

9

10 11 12 13 14 15 16 17 18 19 Trading Hour

(b) News Sentiment

Figure 6.3: News Characteristics by Trading Hour The two graphs plot the number of news during the respective trading hours and show the distribution within each trading hour, according to the sentiment of news and the type of news.

6.3.2

Trading Data

The second data source used in my analysis, is trading data in leverage products from the EUWAX trading segment of Boerse Stuttgart.6 Specifically, I consider all trades in warrants and knock-out products in the data set. 6

For details on the data set, see Section 4.1 of this thesis.

6.4. Empirical Results

1.600

1.600

1.400

1.400

1.200 1.000

Analyst Mandatory Normal

800 600 400 200 0

Number of News

Number of News

114

1.200 1.000

Sent < 0 Sent = 0 Sent > 0

800 600 400 200

Monday

Tuesday Wednesday Thursday

Friday

Trading Day

(a) News Types

0

Monday

Tuesday Wednesday Thursday

Friday

Trading Day

(b) News Sentiment

Figure 6.4: News Characteristics by Weekday The two graphs plot the number of news during the respective trading days and show the distribution within each trading day, according to the sentiment of news and the type of news.

This sample also covers the period from April 2010 to November 2012. As underlyings, I focus on the 30 DAX constituents, for which I have obtained Dow Jones Newswire messages.7 Table 6.3 reports descriptive statistics for each underlying in my sample. The number of leverage products in my sample is 114,622. With a total number of 80,871, more than 70% of the leverage products in my sample are calls, whereas only close to 30% are puts. The total trading volume during the sample period is EUR 5.097 billion. The largest share of trading volume is in calls, which account for EUR 4,411 billion or 87% of the trading volume.

6.4 6.4.1

Empirical Results Trading Intensity around News

In this section, I evaluate whether retail investors trade more intensively around news. More specifically, I test whether trading activity reacts differently according to the sentiment of news and the type of news. To do this, I classify the news in my sample into positive, neutral and negative news. For each firm, I sort all messages into terziles according to their sentiment. The messages in the lowest terzile are classified as negative, the ones in the 7

That is, trades in leverage products which have the respective stocks as underlying.

Total Mean Median Max Min Standard Deviation

Deutsche Bank AG BMW AG Beiersdorf AG Deutsche Post AG Deutsche Telekom AG Fresenius SE Fresenius Medical Care AG Deutsche Boerse AG MAN SE HeidelbergCement AG Henkel AG Infineon Technologies AG Linde AG Merck KGaA RWE AG Daimler AG SAP AG Siemens AG Metro AG ThyssenKrupp AG Volkswagen AG Commerzbank AG Deutsche Lufthansa AG Allianz SE Muenchener Rueck AG Adidas AG BASF SE Bayer AG E.ON SE K+S AG

Underlying

80,871 2,696 2,326 6,359 910 1,476

6359 4349 5256 1925 4485 2184 2707 3521 2424 1272 2416 2271 5015 3666 5471 2696 2155 1215 2380 1499 1283 3555 1654 1016 1283 1750 2077 1200 910 2877 33,751 1,125 889 2,818 334 662

2818 1906 2385 659 1972 1115 916 2082 1179 475 1220 826 2031 1553 1740 985 862 440 826 647 508 1604 761 481 493 737 955 334 411 830

No. Products Calls Puts

849,291 28,310 16,371 116,806 4,168 27507

116,806 52,601 77,998 9,418 55,226 14,314 41,947 27,905 22,559 8,995 18,149 20,110 69,635 52,417 69,048 22,971 14,593 6,304 14,033 8,995 6,095 40,977 6,740 4,168 7,339 7,870 14,130 4,995 6,455 26,498

Trades Calls

121,116 4,037 2,658 13,218 752 3623

13,218 5,478 10,938 1,639 7,035 3,611 2,498 8,884 3,175 1,365 3,865 1,807 12,348 4,688 9,843 2,715 2,640 817 2,727 1,602 1,240 6,709 1,584 1,225 1,154 1,670 2,676 752 981 2,232

Puts

4,411 147 91 579 21 151

578.9 289.6 547.1 51.5 285.0 86.2 148.5 189.6 93.3 33.7 124.4 88.1 241.9 282.0 403.0 97.3 85.2 30.7 95.1 36.2 27.7 229.4 20.7 20.6 51.5 38.0 88.4 26.2 21.0 100.9 686 23 14 75 3 21

66.3 43.6 58.6 6.6 44.9 19.2 12.7 45.5 25.7 6.5 40.3 8.3 75.1 26.2 50.3 10.3 11.7 2.7 16.9 6.6 5.5 31.6 9.8 4.2 14.5 6.2 18.1 2.9 2.8 12.6

Vol. (Mil. EUR) Calls Puts

417,126 13,904 8,457 54,091 2,456 13012

54,091 25,038 38,592 4,961 27,261 7,714 17,859 15,777 10,666 4,451 8,971 9,093 36,003 23,571 32,619 11,721 7,661 3,262 7,150 4,693 3,081 21,668 3,644 2,456 3,929 4,402 7,943 2,658 3,294 12,897 995,485 33,183 22,893 126,561 7,353 29130

126,561 60,061 77,474 13,956 61,388 21,008 46,025 37,795 28,908 10,227 26,269 24,883 75,462 58,603 85,706 24,778 18,252 9,160 20,154 10,593 9,346 44,691 10,445 7,353 11,611 11,311 20,906 7,653 8,119 26,787

Order Types Market Limit

128,768 4,292 3,134 14,673 1,161 3514

14,673 7,535 12,715 2,114 6,701 2,588 5,335 5,166 3,567 1,705 1,804 3,476 7,314 7,069 9,923 4,138 3,363 1,275 1,986 1,860 1,217 7,221 1,450 1,349 1,180 2,164 2,904 1,161 1,301 4,514

Stop

This table reports descriptives statistics for the trading data. For each underlying, I report the total number of leverage products (warrants and knock-out products), the number of trades and the trading volume separately for ’calls’ and ’puts’. Further, for each underlying, I also report the total number of marketable orders (’market’), limit orders (’limit’) and stop (stop limit and stop buy) orders (’stop’). Summary statistics are provided across all underlyings.

Table 6.3: Descriptive Statistics of Trading in Leverage Products

Chapter 6. Media Sentiment and Leveraged Trading 115

116

6.4. Empirical Results

middle terzile as neutral and the ones in the highest terzile as positive. I believe that this sorting methodology is superior to taking the actual sentiment measures (that is, news with a raw negative (positive) sentiment score are classified as negative (positive) news), since the prevalent overall perception on a company strongly influences how news are perceived by market participants. For example, for a firm that currently experiences financial distress, a news message with a slightly positive tone is likely to be perceived as an extremely positive message. In contrast, during a period of success, a news message that is only slightly positive might be perceived as negative news. In addition, the raw sentiment scores strongly rely on the underlying dictionary and the number of positive and negative word classifications contained in them.8 By using a relative measure, where the classification depends on the sentiment relative to all other news messages on the respective company, I can avoid that my results are driven by the specific design of the underlying word list.9 Since I know for each news item whether it is positive, neutral, or negative, I can analyze various aspects of retail investor trading: Can retail investors correctly anticipate the tone of news? Do they engage in news-feedback trading? Does their trading behavior differ, with respect to the sentiment of news? Further, I can differentiate between different types of news (normal, mandatory, analyst). This allows me to study which news types have the most pronounced impact on retail investor behavior and whether they behave differently before and after the respective news types. To study whether retail investor trading activity differs between positive, negative and neutral news messages, I adopt similar methodology as in Riordan et al. (2013) and proceed as follows: I build intervals I1 , I2 , and I3 for different periods before and after news:

8

9

Specifically, the Loughran and McDonald (2011) dictionary contains 354 positive words and 2329 negative words. Due to this imbalance between positive and negative news, raw sentiment scores are likely to be negatively biased. For robustness, I repeat all following analysis based on the following classification according to the raw sentiment scores: Sentiment ≥ 0.1 = positive, Sentiment ≤ -0.1 = negative, Sentiment >-0.1