Contemporary Trends and Challenges in Finance: Proceedings from the 5th Wroclaw International Conference in Finance 3030430774, 9783030430771

Citation preview

Springer Proceedings in Business and Economics

Krzysztof Jajuga Hermann Locarek-Junge Lucjan T. Orlowski Karsten Staehr   Editors

Contemporary Trends and Challenges in Finance Proceedings from the 5th Wroclaw International Conference in Finance

Springer Proceedings in Business and Economics

Springer Proceedings in Business and Economics brings the most current research presented at conferences and workshops to a global readership. The series features volumes (in electronic and print formats) of selected contributions from conferences in all areas of economics, business, management, and finance. In addition to an overall evaluation by the publisher of the topical interest, scientific quality, and timeliness of each volume, each contribution is refereed to standards comparable to those of leading journals, resulting in authoritative contributions to the respective fields. Springer’s production and distribution infrastructure ensures rapid publication and wide circulation of the latest developments in the most compelling and promising areas of research today. The editorial development of volumes may be managed using Springer’s innovative Online Conference Service (OCS), a proven online manuscript management and review system. This system is designed to ensure an efficient timeline for your publication, making Springer Proceedings in Business and Economics the premier series to publish your workshop or conference volume.

More information about this series at http://www.springer.com/series/11960

Krzysztof Jajuga Hermann Locarek-Junge Lucjan T. Orlowski Karsten Staehr •

•

•

Editors

Contemporary Trends and Challenges in Finance Proceedings from the 5th Wroclaw International Conference in Finance

123

Editors Krzysztof Jajuga Department of Financial Investments and Risk Management Wroclaw University of Economics and Business Wrocław, Poland Lucjan T. Orlowski Department of Economics and Finance John F. Welch College of Business Sacred Heart University Fairfield, CT, USA

Hermann Locarek-Junge Faculty of Management and Economics Technische Universität Dresden Dresden, Sachsen, Germany Karsten Staehr Department of Economics and Finance Tallinn University of Technology Tallinn, Estonia

ISSN 2198-7246 ISSN 2198-7254 (electronic) Springer Proceedings in Business and Economics ISBN 978-3-030-43077-1 ISBN 978-3-030-43078-8 (eBook) https://doi.org/10.1007/978-3-030-43078-8 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

This volume presents papers from the 5th Wroclaw International Conference in Finance held at Wroclaw University of Economics on September 24–25, 2019. We have sought to assemble a set of studies addressing a broad spectrum of recent trends and issues in finance, particularly those concerning markets and institutions in Central and Eastern European countries. In the final selection, we accepted 19 of the papers that were presented at the conference. Each of the submissions has been reviewed by at least two anonymous referees, and the authors have subsequently revised their original manuscripts and incorporated the comments and suggestions of the referees. The selection criteria focused on the contribution of the papers to the modern finance literature and the use of advanced analytical techniques. The chapters have been organized along the major fields and themes in finance: Financial Markets, Corporate Finance, Quantitative Methods in Finance, and Financial Institutions. The part on Financial Markets contains eight papers. The paper by Hermann Locarek-Junge and Stefan Albers evaluates market anomaly of higher returns during period when markets are closed compared to returns when markets are open using data of WIG20 index from Warsaw Stock Exchange and the German DAX. Agata Gluzicka in her paper presents the results of empirical research on the occurrence of the effect of the day on the Warsaw Stock Exchange. The paper by Agata Kliber, Barbara Będowska-Sójka, and Tomasz Hinc examines the relation between volatility and liquidity on the cryptocurrency markets. Ewa Majerowska in her paper identifies the factors affecting rates of return of banks, based on 13 banks listed on the Warsaw Stock Exchange. Lesław Markowski examines the cross-sectional relationships between realized returns and systematic risk measures using sub-sectoral indices quoted on Warsaw Stock Exchange. The paper by Sabina Nowak evaluates the accuracy of the popular trade classification rules for three stock exchanges: the Warsaw Stock Exchange, the Prague Stock Exchange, and the Budapest Stock Exchange. Anna Rutkowska-Ziarko in her paper conducts research on the relation between profitability of companies (given through ROA and ROE) and returns on the stock market. The paper by Bogdan Włodarczyk, Alberto Burchi v

vi

Preface

and Marek Szturo presents the impact of the risk of raw material markets on the credit risks of listed companies for two stock markets, Polish and Italian ones. The part on Corporate Finance contains three papers. The paper by Patrizia Gazzola, Stefano Amelino, and Alessandro Figus is aimed to analyze how socially responsible behaviors can be considered as risk management tools. Julia Koralun-Bereźnicka in her paper verifies whether and how corporate capital structure and its determinants vary between the old and new EU member states. The paper by Dorika Jeremiah Mwamtambulo provides the evidence regarding the characteristics possessed by the boards of East Africa community companies and their relationship to the performance of these companies. The part on Quantitative Methods in Finance contains five papers. The paper by Ewa Dziwok and Martin Wirth investigates different approaches to the construction of a term structure of interest rates that are the base in fund transfer pricing mechanism. Marta Małecka in her paper explores properties of the geometric distribution as means of constructing conditional coverage value-at-risk tests. The paper by Paweł Miłobędzki conducts the research on the clustering the stock prices from Warsaw Stock Exchange on certain digits. Grzegorz Tratkowski in his paper analyzes stock index time series by machine learning algorithms to build investment strategies. The paper by Aleksandra Wójcicka-Wójtowicz, Anna Łyczkowska-Hanćkowiak, and Krzysztof Piasecki presents the use of multicriteria decision-making approach for credit risk assessment. The part on Financial Institutions contains three papers. The paper by Magdalena Chmielowiec-Lewczuk presents the findings concerning the use of cost strategies by insurance companies in Poland. Joanna Głód, Lyubov Klapkiv, Anna BiałekJaworska, and Krzysztof Opolski in their paper study the impact of capital requirements set out in the Solvency II Directive on the amount of dividends paid by life insurance companies. The paper by Marta Karaś and Witold Szczepaniak answers the question whether systemic fragility or systemic contagion was responsible for the spreading of the financial crisis in each of the selected CEE countries. We wish to thank the authors for making their studies available for our volume. Their scholarly efforts and research inquiries made this volume possible. We are also indebted to the anonymous referees for providing insightful reviews with many useful comments and suggestions. In spite of our intention to address a wide range of problems pertaining to financial theory, there are issues that still need to be researched. We hope that the studies included in our volume will encourage further research and analyses in modern finance. Wrocław, Poland Dresden, Germany Fairfield, USA Tallinn, Estonia December 2019

Krzysztof Jajuga Hermann Locarek-Junge Lucjan T. Orlowski Karsten Staehr

Contents

Financial Markets The Rhythm of the Night: Some Anomalies in Open and Close Prices of Polish and German Blue-Chip Stocks . . . . . . . . . . . . . . . . . . . . . . . . Hermann Locarek-Junge and Stefan Albers The Effect of the Day and the Risk Diversification on the WSE . . . . . . . Agata Gluzicka

3 21

Volatility and Liquidity in Cryptocurrency Markets—The Causality Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Barbara Będowska-Sójka, Tomasz Hinc, and Agata Kliber

31

Identification of the Factors Affecting the Return Rates of the Banks Listed on the Warsaw Stock Exchange . . . . . . . . . . . . . . . . . . . . . . . . . Ewa Majerowska

45

Conventional and Downside Betas and Higher Co-moments in the Asset Pricing Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lesław Markowski

55

The Accuracy of Trade Classification Rules for the Selected CEE Stock Exchanges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sabina Nowak

65

Profitability Ratios in Risk Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anna Rutkowska-Ziarko

77

Impact of Commodity Market Risk on Listed Companies . . . . . . . . . . . Bogdan Włodarczyk, Alberto Burchi, and Marek Szturo

89

vii

viii

Contents

Corporate Finance The Double Relationship Between Risk Management and CSR in the Italian Healthcare Sector: The Case of the Lombard “Health Protection Agencies” (ATS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Patrizia Gazzola, Stefano Amelio, and Alessandro Figus Are Corporate Financing Policies Different in Old and New EU Member States? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 Julia Koralun-Bereźnicka Board Characteristics and Performance of East Africa Companies . . . . 125 Dorika Jeremiah Mwamtambulo Quantitative Methods in Finance Different Approaches to the Reference Yield Curve Construction— And Their Application into Fund Transfer Pricing Mechanism . . . . . . . 149 Ewa Dziwok and Martin Wirth Geometric Distribution as Means of Increasing Power in Backtesting VaR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Marta Małecka Price Clustering in Stocks from the WIG 20 Index . . . . . . . . . . . . . . . . 169 Paweł Miłobędzki Construction of Investment Strategies for WIG20, DAX and Stoxx600 with Random Forest Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Grzegorz Tratkowski Application of the SAW Method in Credit Risk Assessment . . . . . . . . . 189 Aleksandra Wójcicka-Wójtowicz, Anna Łyczkowska-Hanćkowiak, and Krzysztof Piasecki Financial Institutions Cost-Management Strategies Applied by Insurance Companies in Poland in the Years 2016–2018; Empirical Research . . . . . . . . . . . . . 209 Magdalena Chmielowiec-Lewczuk Dividends of Life Insurance Companies and the Solvency Capital Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Joanna Głód, Lyubov Klapkiv, Anna Białek-Jaworska, and Krzysztof Opolski Fragility or Contagion? Properties of Systemic Risk in the Selected Countries of Central and East-Central Europe . . . . . . . . . . . . . . . . . . . 231 Marta Karaś and Witold Szczepaniak

About the Editors

Krzysztof Jajuga is a professor of finance at Wroclaw University of Economics, Poland. He holds master, doctoral, and habilitation degrees from Wroclaw University of Economics and Business, Poland, title of professor given by President of Poland, honorary doctorate from Cracow University of Economics, and honorary professorship from Warsaw University of Technology. He carries out research within financial markets, risk management, household finance, and multivariate statistics. Hermann Locarek-Junge is a professor of Finance and Financial Services at TU Dresden, Faculty of Management and Economics. He graduated in the field of business and economics, and earned his Ph.D. at University of Augsburg, Germany, and he also studied business informatics and has been appointed as a professor in that field at Essen University. Since then, he has been visiting professor and research fellow at some international institutions and universities. During his academic career, he did research work for several banks. Lucjan T. Orlowski is a professor of economics and finance and a Director for the Doctor of Business Administration (DBA) in finance program at Sacred Heart University, Fairfield, Connecticut. His research interests include monetary economics and stability of financial markets and institutions. He has authored numerous books, chapters in edited volumes, and over 80 articles in scholarly journals. He is a Doctor Honoris Causa recipient from Wroclaw University of Economics. Karsten Staehr is a professor at Tallinn University of Technology, Estonia. He is also employed part-time as a research supervisor at Eesti Pank, the central bank of Estonia. He holds a master’s degree from the Massachusetts Institute of Technology and a master’s degree and a Ph.D. from the University of Copenhagen. He undertakes research and policy analysis within macroeconomics, public economics, monetary economics, and transition economics. He is an associate editor of the Baltic Journal of Economics and on the editorial board of several other journals. ix

Financial Markets

The Rhythm of the Night: Some Anomalies in Open and Close Prices of Polish and German Blue-Chip Stocks Hermann Locarek-Junge and Stefan Albers

Abstract Stock returns on many stock exchanges worldwide are much higher when markets are closed for trading versus when they are open. We evaluate this temporal market anomaly with respect to market betas and volatility for the example of the WIG20 index in the Warsaw stock exchange and the German DAX. We provide an in-depth characterization of stock returns, emphasise the importance to decompose return components and reveal fundamental differences between Polish and German stock markets. Furthermore, we address possible reasons for the higher rates of night returns, e.g. news flow, liquidity and bid-ask spreads.

1 Introduction This study deals with temporal anomalies of stock returns and explore whether they also occur on the Polish and German Exchange. A recent publication of New York Times brought the attention of a broad readership to the fact, that the returns between the close of the market to reopen on the next morning are much higher than the returns that can be earned (on average) comparing the open to the following close prices of the trading period. Recent research has put focus on the effects and the reasons for these different return rates (cf. Basdekidou 2017; Berkman et. al. 2012; Blose and Gondhalekar 2014; Blose et. al. 2018; Branch and Ma 2006; Branch and Ma 2014; Fuertes et. al. 2015; Kelly and Clark 2011; Lachance 2015; Liu and Tse 2017; Lou et al. 2018; Monteiro and Manso 2017). A few studies already dealt with the phenomenon in the nineties, but did not dig deeper (Greene and Watts 1996; Haugen 1999). Usually, in the past stock exchanges had their trading hours from nine to five or even shorter. Different countries close their Financial Exchanges at their own H. Locarek-Junge (B) · S. Albers Technische Universität Dresden (Dresden University of Technology), Dresden, Germany e-mail: [email protected] S. Albers e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_1

3

4

H. Locarek-Junge and S. Albers

holidays and each Stock Exchange has its own working hours or working days (i.e. some observe a Friday/Saturday weekend while most other are closed Saturday and Sunday). Some Stock Exchanges have extended trading hours by electronic networks either before or after the regular trading hours of the listing exchange, e.g. FWB (Germany) for its XETRA system. Recently, facing competition from proprietary trading platforms and dark pools, more and more exchanges decided to extend trading hours. However, such trading tends to be limited in volume compared to regular trading hours when the exchange is open. Even if markets close at one Stock Exchange, trading may be open at another market. Therefore, one could assume that stocks trade around the clock somewhere on the world.1 But this is not only wrong for weekends, but also during the week, because usually most stocks are traded only in their own markets. Only very few stocks, mostly of large companies, have a secondary listing, usually in the U.S. markets. Therefore, we can still separate trading (daytime) and non-trading (overnight, weekend) periods. As the next figure indicates, it is a common misinterpretation between the market participants that the majority of returns are created during the daytime, rather than the overnight period. This would seem reasonable, as more people are active during the normal daytime hours as opposed to the overnight period (Fig. 1). Our analyses confirm, but also partially contradict, previous studies. We examine the characteristics of intraday and overnight returns of Polish and German shares included in the DAX and WIG20 index based on OHLC data. Employing betaranked portfolios, we investigate the risk-return relation of the capital asset pricing model and reveal fundamental differences between Polish and German shares. The remainder of this paper is organized as follows. Section 2 presents methods to separate and explore day and night returns. Section 3 reports results. Section 4 concludes.

2 Methods 2.1 Simple Returns and Log Returns Starting from scratch, we define returns as simple returns or as log returns. The numerical difference between simple and log returns is small as long as the return values are close to zero. However, there can be non-negligible differences if we look at volatility and covariance in a stochastic context (see Dorfleitner 2003). Therefore, the return notion matters. Computing stock market returns using simple returns means dividing the difference of two consecutive stock prices by the original price, or dividing two prices and subtract 1: 1 Options and futures markets also tend to have different trading hours depending on the underlying

assets. The German EUREX has recently offered extended trading hours for select liquid futures to support the increasing demand for trading on a global market place (see: https://www.eurexchange. com/exchange-en/trading/thx).

The Rhythm of the Night: Some Anomalies in Open and Close …

5

Fig. 1 Overnight and Intraday Returns to major stock market indices (Knuteson 2018, p. 2)

rtsimple =

Pt − Pt−1 Pt Pt−1 Pt = − = − 1. Pt−1 Pt−1 Pt−1 Pt−1

(1)

For one-period capital market models like Portfolio Selection or CAPM, assuming a normal distribution of simple returns is standard. More complex models of the capital market like continuous models e.g. for options pricing often use log returns. For modelling the stock market, the consequences are different if one assumes log returns.   Pt = ln Pt − ln Pt−1 . (2) rtln = ln Pt−1 Following this computational scheme, adding subsequent period returns gives the total period log return. Any logarithm will do, but mostly the natural logarithm (and the exponential function as its inverse) is used (see Bamberg and Dorfleitner 2003). In this paper, we use the natural logarithm for computing returns. Using the correct formula to compute the total period return or the terminal price PN from serial single returns yields the same result:

6

H. Locarek-Junge and S. Albers

PN = P0 · er1 +r2 +···+r N = P0 · (1 + r1simple ) · (1 + r2simple ) · · · (1 + r Nsimple ). ln

ln

ln

(3)

2.2 OHLC Records and Day- and Night-Returns Since 1992, stock market databases in the U.S. and other places started to report daily trading for stock market indices and individual stocks, giving so called OHLC records. Using log returns (for the sake of simplicity called r during the rest of the paper), we can compute three different returns. Often, also the trading volume was included. The OHLC means, that for one single day, the opening price (O), the closing price (C) as well as the highest (H) and lowest (L) price in the trading period was given in the record. Although the OHLC records were available, empirical research usually used the prices at two consecutive closing auctions. Therefore, the daily log returns in the literature are computed mostly as log returns of closing prices, i.e. rt = ln Ct − ln Ct−1 . Weekly returns can be computed the same way as the sum of five consecutive daily returns, or directly from the closing prices of e.g. two Wednesdays. Quite often, the availability of the other prices in the OHLC record has been neglected. For example, we could use consecutive opening prices in the above formula, or one could eliminate biases that only contaminate closing prices by computing the daily return from the average or sum of the highest and lowest price of the day, e.g.: rt = ln Mt − ln Mt−1 = ln(L t + Ht ) − ln(L t−1 + Ht−1 )   with Mt = (L t + Ht ) 2; Mt−1 = (L t−1 + Ht−1 ) 2.

(4)

The topic of this paper is different: We split the period return r into two separate parts that yield the period return in total. The first part of the returns is called trading period return r T (sometimes day return or intraday return). It is computed from the opening price Ot and the closing price of the same trading period C t . The other, sometimes called night return, overnight return or better non-trading period return r N is computed from the previous closing price C t−1 and the following opening price Ot (Fig. 2). Hence, using the additive property of log returns, we have:   rt = ln Ot − ln Ct−1 + [ln Ct − ln Ot ] = rt,N + rt,T .

(5)

One of the uses of OHLC records is the estimation of the daily volatility or bid-ask spreads when intraday or tick data are absent.

The Rhythm of the Night: Some Anomalies in Open and Close …

7

Trading period return rT

Fig. 2 OHLC-chart with returns (own work) [based on https://datavizcatalogue. com/methods/OHLC_chart. html]

2.3 Volatility Estimates The simplest and most common type of calculation that benefits from only using prices C t from closing auctions is the standard method of estimating a variance and taking its square root. Note that√the estimated volatility is the sample standard deviation multiplied by the factor N /(N − 1) to take into account that we are sampling the population and taking the deviation from the sample mean and not from µ.    σˆ x =

1

(xi − x) ¯ 2. N − 1 i=1 N

(6)

In 1976, Parkinson created the first advanced volatility estimator using the high and low prices during a trading period by in a university working paper (cf. Parkinson 1976). One drawback of this estimator is that it assumes continuous trading, hence it underestimates the volatility as potential movements when the market is closed are ignored. Some studies have shown this measure to be good for actual empirical data although other measures are more efficient based on simulated data. Later in 1980, the Garman/Klass volatility estimator was created (Garman and Klass 1980). As an extension of the Parkinson estimator, it includes opening and closing prices. In Fig. 3, trading is closed initially, with yesterday’s closing when the price was C 0 . The price sample path is then unobservable (although it is visible on the left in the Garman/Klass original chart) until trading opens and the price is O1 . After that, we assume that the entire sample path is continuously monitored, having a high value of H 1 , a low value of L 1 , and a closing value of C 1 . In their formula below, Z is a

8

H. Locarek-Junge and S. Albers

Fig. 3 Price versus time (original chart taken from Garman and Klass 1980, p.2)

scaling factor (e.g. square root of the number of trading days per year).2

σˆ G K

       Z

Ci 2 Hi 2 = − (2 ln 2 − 1) ln . 0.5 ln n Li Oi

(7)

As overnight jumps are ignored (although illustrated), the measure underestimates volatility. The Rogers-Satchell volatility estimator (Rogers and Satchell 1991; Rogers et. al. 1994) created in the early 1990s is able to measure the volatility for securities with non-zero mean properly. It does not, however, handle jumps (hence it underestimates the volatility).

σˆ R S

  N         F

ln CHii ln OHii + ln CL ii ln OL ii . = N i=1

(8)

Later, Yang and Zhang (2000) modified the Garman-Klass volatility measure in order to enable it to handle overnight jumps. They created the most complex OHLC volatility measure that handles both opening jumps and drift. The result is the sum of the overnight volatility (close-to-open volatility), a weighted average of the RogersSatchell volatility and the open to close volatility. The assumption of continuous prices does mean the measure tends to underestimate the volatility slightly. None of the measures can handle the problem that volatility is invisible in the period between 2 The

R package TTR implements the Garman-Klass formula.

The Rhythm of the Night: Some Anomalies in Open and Close …

9

the point in time when trading stops and trading reopens. Here, we have, according to the “bat-in-the-tunnel” analogy, only latent volatility, in contrast to the realized volatility during trading periods, which we can measure with every single trade executed. Their version 1 is given in Formula (9) and version 2 in Formula (10).

σˆ Y Z 1

where k = and σˆ O2 N

  N   2 2  F

ln COi−1i = + 0.5 ln HL ii + (2 ln 2 − 1) ln N i=1    σˆ Y Z 2 = F σˆ O2 N + k σˆ T2 P + (1 − k)σˆ R2 S .

Ci Oi

2  .

(9) (10)

0.34 N +1 N −1

1.34+

 2  2 N N 1

Oi 1

Ci Oi Ci 2 = − ln ; σˆ T P = − ln ln ln N − 1 i=1 Ci−1 Ci−1 N − 1 i=1 Oi Oi

Meanwhile, many sophisticated econometric models for volatility exist. Exponentially weighted (EW) volatilities do not handle regular volatility driving events such as earnings very well. Previous earnings jumps will have the least weight just before an earnings date and the most weight just after earnings. However, this type finds use in the common Value-at-Risk models. The well known GARCH(p, q) models are based on an autoregressive representation of the conditional variance. The most used heteroscedastic model in financial time series is a GARCH(1,1) (cf. Bera and Higgins 1993). However, various other types of GARCH like EGARCH, FIGARCH, FIEGARCH, FIAPARCH or Mixture Memory GARCH (MMGARCH) were proposed in the literature recently (cf. Walther 2017). In the field of realized volatility modeling and forecasting it is common practice to neglect overnight returns and even to truncate intraday returns right after market open and just before market close to avoid so-called microstructure noise (Fig. 4).

2.4 Bid-Ask Spreads Another application for OHLC records is the estimation of bid-ask spreads, when quote data are not available. A market maker or dealer usually sets bid quotes and ask quotes around the fair price, or “value” being the center of the spread. Buyer initiated trades occur at the ask price, while sellers trade at the bid price. Therefore, the bid-ask spread is essentially the difference between the highest price that a buyer is willing to pay for an asset and the lowest price that a seller is willing to accept to sell it. The bid-ask spread s means trading cost, and consecutive seller and buyer

10

H. Locarek-Junge and S. Albers

Fig. 4 Continuous time volatility during the day (own chart similar to Bennet and Gil, 2012)

initiated trades would make prices fluctuate without new information, and contribute to volatility. Roll (1984) proposed a simple implicit measure of the effective bid-ask spread using daily prices only. He assumes, that during the trading period, seller- and buyerinitiated trades are equally likely, and the bid-ask average fluctuates randomly in an efficient market (Roll 1984, p. 1128). The observed trading price (either at bid or at asked) is different from the efficient price: s Pt = Pte + qt ; qt = ±1. 2

(11)

The covariance is minus the square of one-half the bid-ask spread. Similarly, the variance of price changes is s 2 /2 and the autocorrelation coefficient of prices based on bid-ask bounces is −1/2 (Roll 1984, p. 1129). The HL method by Corwin and Schultz (2012) assumes that the high price (low price) are buyer (seller) initiated: s Ht = Hte + ; 2

s L t = L et − . 2

(12)

However, in order to construct the closed-form estimator they violate Jensen’s inequality. Like the other models, they need ad hoc adjustments for holidays and weekends. The paper of Abdi and Ranaldo (2017), where they estimate bid-ask spreads from daily close, high and low prices, lists some other models (HL, Roll, Gibbs, EffTick, and FHT) that do not rely on quote data. Their result is a simple measure for the proportional spread based on Roll (1984). In line with this model, they assume that trade directions qt (+1 or −1) are independent of the efficient price.

The Rhythm of the Night: Some Anomalies in Open and Close …

11

Abdi and Ranaldo in their paper therefore extend the model of Corwin and Schultz (2012) mentioned above. Assuming that the efficient price follows a continuous path in which ct the daily log close price and ηt is the daily log mid-range of close prices, i.e. the average of daily high and low prices, they get the following estimator.   s 2 = 4E (ct − ηt )(ct − ηt+1 ) .

(13)

Note that they use log prices, so that the difference of log process is the log return, and ηt is the geometric mean. None of the authors of the previous models assumes however, that open and close process on any day might show a bias similar to the high and low prices like in Formula (12). If we take the closing price Ct as the result of selling pressure, because day traders want to unwind their positions at the end of the trading day, and we assume that the open prices Ot reflect a demand asymmetry because day traders want to build up their positions in the morning, then we get similar to Eq. (12) the following setting: s s Ot = Ote + ; Ct = Cte − . 2 2

(14)

Hence, the unusual returns in the non-trading period from the close to the open might stem from the bid-ask differences. To test for this hypothesis, we have to acquire or estimate the spreads for the market, e.g. using a Formula like (11) and then correct the open and close prices like in Formula (14). Fortunately, the average spreads for the WIG20 stocks are listed in the rightmost column of a table on the GPW Main Market web page. The bid-ask spreads weighted by average trading volume are approx. 10.25 basis points, i.e. 0.1025 percent per trade. Institutional traders might have much lower trading costs. If we can eliminate other biases, we might be able to explain the high non-trading-period returns using the reported bid-ask-spreads.

2.5 Data The following analysis are based on OHLC data of both blue-chip indices WIG20 and DAX (price index) and corresponding shares obtained from Thomson Reuters Datastream. Since WIG20 is a price index like most international indices, we use DAX price index instead of the more popular DAX performance index to achieve a greater level of comparability of both stock markets.3 The dataset covers the period from 2 January 2014 to 28 December 2018.

3 We

also performed our analyses based on the corporate action adjusted price index of the WIG20 provided by Datastream and the DAX performance index and the results where similar.

12

H. Locarek-Junge and S. Albers

2.6 Portfolio Construction To provide an in-depth analysis of the relationship between the day and night returns and the systematic risk (in terms of beta), we apply a slightly modified version of the portfolio-approach of Hendershott et al. (2018) and Savor and Wilson (2014). Using daily night returns described in Formula 5, we calculate monthly betas for every single stock using a one year rolling window (e.g. first beta of stock i is based on returns from 2014-01-02 to 2014-12-30). Based on these pre-ranked betas, the stocks of the WIG20 (DAX) are then sorted every month into four (six) equally-weighted portfolios consisting of five stocks.4 By averaging daily stock returns within each portfolio, we obtain daily portfolio returns and calculate the post-ranking beta over the whole sample for each portfolio.

3 Results Over the period considered, the WIG20 had shown a negative (logarithmic) return of −6.55%. After a sharp loss in 2015 of −21.7%, it gained 7.7% in 2015 and 22.9% in 2017 but has fallen below the level of 2014 due to a significant loss of −7.9% in 2018 (Fig. 5). Similarly, the DAX (price index) recorded a decline of −2.21% in the period under review. This overall loss is mainly driven by an outstanding sell-off in 2018 when the German blue chip index lost 22.7% (Fig. 6). Whereas both indices show slightly negative daily returns on average, a different picture emerges when interday returns are separated into a night and day return as can be seen in Table 1.

Fig. 5 Historical close prices of WIG20

4 Alternative

numbers of portfolios where also analysed and led to the same overall results.

The Rhythm of the Night: Some Anomalies in Open and Close …

13

Fig. 6 Historical close prices of DAX (price index)

Table 1 Statistics of night, day and interday returns of WIG20 and DAX Night WIG

DAX

Day

Interday

−0.1007

−0.0396

−0.0583

Median

0.0004

−0.0003

0.0000

Arithmetic mean

0.0001

−0.0001

−0.0001

Maximum

0.0120

0.0543

0.0353

Std. Dev.

0.0046

0.0100

0.0111

Skewness

−10.7868

0.0762

−0.2291

Kurtosis

226.2777

1.1011

1.0721

Minimum

−0.1047

−0.0413

−0.0707

Minimum

Median

0.0004

0.0002

0.0005

Arithmetic mean

0.0004

−0.0004

−0.0001

Maximum

0.0270

0.0446

0.0485

Std. Dev.

0.0072

0.0093

0.0117

Skewness

−3.1996

−0.1537

−0.3565

Kurtosis

46.2686

1.9887

2.3738

As in line with Knuteson (2018), night returns of both stock indices are positive on average. This applies even though maximum losses occurred overnight (2016– 06–23 to 2016–06–24, after BREXIT-votum) and maximum gains occurred during trading hours (2016-06-24 for WIG20 and 2016-11-09 for DAX). A further difference between day- and night returns is the extreme high kurtosis and negative skewness of the latter. Thus night returns are leptokurtic and left-tailed asymmetric, which means that there occur more extreme returns compared to normal distribution. These characteristics are illustrated in Figs. 7 and 8 and consistent with the findings of Riedel and Wagner (2015) who found that overnight return innovations exhibit significant tail risk, while intraday innovations do not.

14

H. Locarek-Junge and S. Albers

Fig. 7 Histograms of night (left) and day (right) returns of WIG20

Fig. 8 Histograms of night (left) and day (right) returns of DAX

The pooling of stocks based on night return betas into ranked portfolios indicates the same results and reveals another phenomenon. The sensitivity of portfolio returns to market movements varies significantly between trading hours and non-tradinghours. Regarding the DAX night returns correspond to the capital asset pricing model and show a positive linkage to market beta, i.e. higher systematic risk is related to higher cost of equity capital. However, day returns exhibit a contrary relation as can be seen in Fig. 9.5 These findings for DAX returns correspond to the analyses of Hendershott et al. (2018) who revealed this phenomenon for U.S. stocks and the EU region. But, surprisingly, our examination of the WIG20 indicates a different characteristic for Polish shares. Whereas day returns show a positive linkage to market beta, night returns now decrease with higher beta as displayed in Fig. 10.

5 See

Tables 2 and 3 in the appendix for exact values and detailed statistics for each portfolio.

The Rhythm of the Night: Some Anomalies in Open and Close …

15

Fig. 9 Day and night returns for beta-ranked portfolios for DAX companies and market security lines approximated using ordinary least square estimate

Fig. 10 Day and night returns for beta-ranked portfolios for WIG20 companies and market security lines approximated using ordinary least square estimate

4 Conclusions In this study, we examine the characteristics of day- and night returns of German and Polish shares included in the DAX and WIG20 index. Using daily OHLC prices from 2014 to 2018 we found that positive share price movements stem mainly from positive night returns while day returns are negative on average. The assumption of normal distribution does not hold for night returns, especially for the Polish shares. They are leptokurtic and left-tailed asymmetric, which means that extreme returns occur more often compared to normal distribution.

16

H. Locarek-Junge and S. Albers

By sorting shares into beta-ranked portfolios, we show that the intuitive riskreturn relation of CAPM exists for German shares only during non-trading hours and is inverted during trading hours. In this regard, we confirm the findings from Hendershott et al. (2018) at national level. But analyses of Polish shares indicate different results and contradict those of Hendershott et al. (2018). Polish shares exhibit an inverse relation, i.e. night returns decrease and day returns increase with higher systematic risk. For both countries risk-return relation of shares becomes much less clear when betas are based purely on close-to-close returns. These findings not only provide an in-depth characterization of stock returns, emphasise the importance to decompose return components and reveal fundamental differences between Polish and German stock markets, but have also direct implications for risk management and investment strategies. Even though the reasons for such different characteristics of day and night returns are not yet explored with certainty, there are different approaches such as jump-risk of information releases and liquidity effects (Cliff et al. 2008), cycles of declines and reversals (Abdi 2018), different characteristics of cash-flow news and discount-rate news (Hendershott et al. 2018), retail investors’ herding (Barber and Odean 2008) and market-maker behaviour (Branch and Ma 2006). In this context, we consider bid-ask spreads, varying liquidity as possible influential factors, especially for Polish shares, and therefore as promising directions for future research. Furthermore, the global trend to enable and use extended-hours trading and the introducing of night trading sessions and their potential impact on day and night returns offer potential for further studies.

Appendix See Tables 2 and 3.

The Rhythm of the Night: Some Anomalies in Open and Close …

17

Table 2 Betas and average returns for beta-ranked portfolios of WIG20 and DAX companies

WIG20

DAX

Portfolio

Beta

Average night return

Average day return

Average interday return

A

0.96

B

1.14

0.000578

−0.000599

−0.000021

0.000611

−0.000888

C

1.33

−0.000277

0.000550

−0.000129

0.000421

D

1.77

0.000232

−0.000256

−0.000024

A

0.53

−0.000173

0.000375

0.000202

B

0.65

0.000075

0.000394

0.000469

C

0.70

0.000026

−0.000243

−0.000217

D

0.76

0.000129

−0.000149

−0.000020

E

0.86

0.000062

−0.000086

−0.000023

F

1.04

0.000298

−0.000351

−0.000054

DAX

WIG20

0.0063

−7.1094

126.7167

−0.1008

−0.0020

0.0004

0.0001

0.0061

−5.4969

0.0032

0.0180

0.0056

−6.0328

95.1922

−0.0803

−0.0020

0.0000

−0.0002

0.0021

0.0198

0.0054

−4.5595

57.8066

Quartile 3

Maximum

Stdev

Skewness

Kurtosis

Minimum

Quartile 1

Median

Arithmetic mean

Quartile 3

Maximum

Stdev

Skewness

Kurtosis

0.0008

Arithmetic mean

81.1008

0.0234

0.0027

0.0227

0.0034

0.0006

0.0010

0.0006

Median

−0.1190

−0.0016

−0.0984

−0.0015

Minimum

Quartile 1

Night_B

Night_A

75.7582

−5.7899

0.0066

0.0208

0.0026

0.0000

0.0005

−0.0016

−0.1049

139.5114

−7.4654

0.0071

0.0229

0.0037

0.0006

0.0008

−0.0022

−0.1377

Night_C

84.5342

−5.8985

0.0068

0.0264

0.0027

0.0001

0.0004

−0.0018

−0.1134

224.4241

−10.6664

0.0087

0.0307

0.0034

0.0002

0.0007

−0.0022

−0.1894

Night_D

2.7524

−0.2310

0.0094

0.0534

0.0055

0.0004

0.0006

−0.0044

−0.0439

1.7775

−0.0079

0.0111

0.0645

2.2845

0.0831

0.0102

0.0578

0.0061

0.0004

0.0004

−0.0050

−0.0395

1.2435

0.0058

0.0119

0.0623

0.0062

−0.0007 −0.0009

−0.0002 −0.0006 0.0063

−0.0084

−0.0458

Day_B

−0.0077

−0.0463

Day_A

2.7397

0.0714

0.0106

0.0655

0.0059

2.6450

0.0323

0.0106

0.0578

0.0056

−0.0001

0.0003

−0.0002

−0.0055

−0.0063

−0.0403

14.0021

1.2943

0.0128

0.1398

0.0070

−0.0003

−0.0006

−0.0077

−0.0455

Day_D

−0.0005

−0.0530

2.0303

0.3857

0.0131

0.0855

0.0079

−0.0001

−0.0006

−0.0083

−0.0452

Day_C

1.6090

−0.3695

0.0104

0.0311

0.0056

0.0002

0.0005

−0.0050

−0.0439

1.3989

−0.3132

0.0121

0.0458

0.0075

0.0000

0.0008

−0.0074

−0.0648

Interday_A

Table 3 Statistics of night, day and interday returns for beta-ranked portfolios of WIG20 and DAX companies

1.5815

−0.3352

0.0114

0.0359

0.0067

0.0005

0.0009

−0.0057

−0.0565

1.0680

−0.2518

0.0127

0.0376

0.0076

−0.0003

−0.0003

−0.0084

−0.0631

Interday_B

2.3634

−0.4719

0.0119

0.0381

0.0064

−0.0002

0.0002

−0.0069

−0.0723

0.5440

0.0690

0.0142

0.0479

0.0093

0.0004

0.0000

−0.0088

−0.0522

Interday_C

1.5390

−0.4353

0.0117

0.0328

0.0065

0.0000

0.0006

−0.0061

−0.0625

1.5355

−0.1162

0.0139

0.0526

0.0085

0.0000

−0.0002

−0.0082

−0.0799

Interday_D

18 H. Locarek-Junge and S. Albers

The Rhythm of the Night: Some Anomalies in Open and Close …

19

References Abdi F (2018) Cycles of declines and reversals following overnight market declines. School Financ Res. https://doi.org/10.2139/ssrn.3287680 Abdi F, Ranaldo A (2017) A simple estimation of bid-ask spreads using daily close, high, and low prices. Rev Financ Stud 30:4437–4480. https://doi.org/10.1093/rfs/hhx084 Aboody D, Even-Tov O, Lehavy R, Trueman B (2018) Overnight returns and firm-specific investor sentiment. J Financ Quant Anal 53:485–505. https://doi.org/10.1017/S0022109017000989 Bamberg G, Dorfleitner G (2002) Is traditional capital market theory consistent with fat-tailed log returns? Z Betriebswirtschaft 72(8):865–873 Barber BM, Odean T (2008) All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors. Rev Financ Stud 21(2):785–818 Barclay MJ, Hendershott T (2003) Price discovery and trading after hours. Rev Financ Stud 16(4):1041–1073. https://doi.org/10.1093/rfs/hhg030 Basdekidou VA (2017) The overnight return temporal market anomaly. Int J Econ Financ 9(3):1–19. https://doi.org/10.5539/ijef.v9n3p1 Bennet C, Gil MA (2012) Measuring historical volatility. In: Santander Global Banking and Markets (eds). http://www.todaysgroep.nl/media/236846/measuring_historic_volatility.pdf Bera AK, Higgins ML (1993) ARCH models: properties, estimation and testing. J Econ Surv 7(4):305–366. https://doi.org/10.1111/j.1467-6419.1993.tb00170.x Berkman H, Koch PD, Tuttle L, Zhang YJ (2012) Paying attention: overnight returns and the hidden cost of buying at the open. J Finan Quant 47:715–741. https://doi.org/10.1017/ S0022109012000270 Blose LE, Gondhalekar V (2014) Overnight gold returns. Appl Econ Lett 21(18):1269–1272. https:// doi.org/10.1080/13504851.2014.922661 Blose LE, Gondhalekar V, Kort A (2018) Overnight versus day returns in gold and gold related assets. J Econ Financ 42:526–549. https://doi.org/10.1007/s12197-017-9403-0 Branch BS, Ma A (2006) The overnight return, one more anomaly. SSRN. https://doi.org/10.2139/ ssrn.937997 Branch BS, Ma A (2012) Overnight return, the invisible hand behind intraday returns? J Appl Financ 22:90–100 Cliff M, Cooper MJ, Gulen H (2008) Return differences between trading and non-trading hours: like night and day. SSRN. https://doi.org/10.2139/ssrn.1004081 Corwin SA, Schultz P (2012) A simple way to estimate bid-ask spreads from daily high and low prices. J Financ 67(2):719–760 Dorfleitner G (2003) Why the return notion matters. Int J Theor Appl Financ 06(01):73–86. https:// doi.org/10.1142/S0219024903001797 Fuertes AM, Kalotychou E, Todorovic N (2015) Daily volume, intraday and overnight returns for volatility prediction: profitability or accuracy? Rev Quant Financ Account 45:251–278. https:// doi.org/10.1007/s11156-014-0436-6 Garman MB, Klass MJ (1980) On the estimation of security price volatilities from historical data. J Bus 53(1):67–78. https://doi.org/10.1086/296072 Greene JT, Watts SG (1996) Price discovery on the NYSE and the NASDAQ: the case of overnight and daytime news releases. Financ Manag 25:19–42. https://doi.org/10.2307/3665900 Haugen RA (1999) Beast on wall street: how stock volatility devours our wealth. Prentice Hall, Upper Saddle River, NJ Hendershott T, Livdan D, Rösch D (2018) Asset pricing: a tale of night and day. SSRN. https://doi. org/10.2139/ssrn.3117663 Kelly MA, Clark SP (2011) Returns in trading versus non-trading hours: the difference is day and night. J Asset Manag 12:132–145 Knuteson B (2018) How to increase global wealth inequality for fun and profit. SSRN. https://doi. org/10.2139/ssrn.3282845

20

H. Locarek-Junge and S. Albers

Lachance, ME (2015) Night trading: lower risk but higher returns? SSRN. https://doi.org/10.2139/ ssrn.2633476 Liu Q, Tse Y (2017) Overnight returns of stock indexes: evidence from ETFs and futures. IREF 48:440–451. https://doi.org/10.1016/j.iref.2017.01.005 Lou D, Polk C, Skouras S (2018) A tug of war: overnight versus intraday expected returns. J Financ Econ 134(1):192–213. https://doi.org/10.1016/j.jfineco.2019.03.011 Monteiro JD, Manso JRP (2017) Are there night and day time effects in US equity index returns? A robust econometric analysis. Int J Econ 11:291–313 Parkinson M (1976) The random walk problem: extreme value method for estimating the variance of the displacement (diffusion constant). Working paper, University of Florida Riedel C, Wagner N (2015) Is risk higher during non-trading periods? The risk trade-off for intraday versus overnight market returns. J Int Financ Mark Inst Money 39:53–64. https://doi.org/10.1016/ j.intfin.2015.05.012 Rogers LCG, Satchell SE (1991) Estimating variance from high, low and closing prices. Ann Appl Probab 1(4):504–512 Roll R (1984) A simple implicit measure of the effective bid-ask spread in an efficient market. J Financ 39(4):1127–1139. https://doi.org/10.1111/j.1540-6261.1984.tb03897.x Rogers LCG, Satchell SE, Yoon Y (1994) Estimating the volatility of stock prices: a comparison of methods that use high and low prices. App Financ Econ 4:241–247. https://doi.org/10.1080/ 758526905 Savor P, Wilson M (2014) Asset pricing: a tale of two days. J Finnc Econ 113(2):171–201 Walther Th (2017) Expected shortfall in the presence of asymmetry and long memory. Pac Account Rev 29(2):132–151 Yang D, Zhang Q (2000) Drift-independent volatility estimation based on high, low, open, and close prices. J Bus 73(3):477–492. https://doi.org/10.1086/209650

The Effect of the Day and the Risk Diversification on the WSE Agata Gluzicka

Abstract The article presents the results of empirical research on the occurrence of the effect of the day on the Warsaw Stock Exchange. The analysis was carried out for the daily rates of return of selected stock exchange indices in the period 2010–2018. In addition, the effect of the day for well-diversified portfolios such as Rao’s Quadratic Entropy Portfolios and the Most Diversified Portfolios, was also analyzed. One of the goals of the research was to establish whether the determined effect of the day is reflected in the level of diversification of the constructed portfolios. On the basis of the conducted research, it was determined that the day effect on the WSE occurs only in specific (annual) periods. This dependence was determined both for the rates of return of stocks as well as for the rates of return of portfolios. The occurrence of the effect of the day for a group of indices does not always translate into similar regularities for rates of return of portfolios or for the level of diversification of investment portfolios.

1 Introduction The effect of the day of the week is one of the most frequently analyzed anomalies in the distribution of rates of return. Research conducted over several decades has shown the existence of certain dependencies characteristic for the rates of return on a given day of the week. One of the first studies concerned the US market, for which it was found that average rates of return on Mondays are significantly lower than average rates of return on the other days of the week. In addition, Monday’s rates of return usually have negative values. However the rates of return on Fridays are usually higher than the rates of return on other days of the week (Cross 1973; French 1980).

A. Gluzicka (B) University of Economics in Katowice, Katowice, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_2

21

22

A. Gluzicka

Similar research in subsequent years was conducted for capital markets in various countries. The effect of Monday was also observed for the markets in Great Britain and Canada. However, for Japan and Australia, negative rates of return were found on Tuesdays (Jaffe and Westerfield 1985). The same conclusions were obtained in the analysis of the Italian and French markets. For the Chinese stock market the negative returns were found on Mondays and Tuesdays (Cai et al. 2006). Research conducted by Chinko and Avci (2009) showed the significant negative Monday returns and positive Thursday and Friday returns on the Istanbul Stock Exchange. The following years brought a number of publications in which it was shown that the described effects are variable over time and sometimes disappear at certain periods. The results of such research were presented for example in the papers of Smirlock and Starks (1986), Gay and Kim (1987), Chang and Kim (1988), Johnston et al. (1991), Rossi (2015). Research for the Polish market has shown, among others, positive average rates of return on Mondays and negative on Tuesdays (Szyszka 1999). Other studies showed positive (significantly different from zero) rates of return on Mondays and Fridays (Kompa and Witowska 2007). This article presents the results of research for the Polish investment market conducted for an example group of stocks that are components of the mWIG40 index in the period 2010–2018. The goal of empirical research was to determine whether the effect of the day occurring for the stock rates of return is reflected in the rates of return of diversified portfolios and the level of diversification of these portfolios. In the presented research, the effect of the day was analyzed in the context of maximum and minimum values of rates of return. The effect of the day was considered for two types of well-diversified portfolios: the Rao’s Quadratic Entropy portfolios and the Most Diversified Portfolios.

2 The Models of Well-Diversified Portfolios Different optimization models can be used to construct the well-diversified portfolios. The relationship between the correlation and the risk of portfolio is essential in determining the level of diversification (Markowitz 1952). The Rao’s Quadratic Entropy (Rao 1982a, b) is an example of measure that takes into account this relation. First, this measure was applied as a measure of diversity, mainly in statistics (generalized analysis of variance) and in ecology (in the research of biodiversity). However, this measure can be successfully used in the portfolio analysis, including also the issue of diversification (Carmichael et al. 2015). The value of the Rao’s Quadratic Entropy (the level of diversification), for portfolio consisting of N components with the shares x i , for i =1, 2, …, N, can be calculated according to the formula:

The Effect of the Day and the Risk Diversification on the WSE N 

RQE = 2

di j xi x j

23

(1)

i, j=1

 N where D = di j i, j=1 is a function of diversity. This function measures the differences between any two components of the portfolio. The diversity function can be defined for example by using the Kronecker delta or the covariance matrix of rates of return. However, the best results we can received when the diversity function is expressed by the correlation matrix (Gluzicka 2018). Then the RQE is defined as (Carmichael et al. 2015): RQE =

N    1 − ρi j xi x j

(2)

i, j=1

 N where ρ = ρi j i, j=1 is the correlation matrix of rates of return. The interpretation of the Rao’s Quadratic Entropy is simple: the higher value of RQE, the higher level of diversification of portfolio. In the construction of the well-diversified portfolio, measure RQE is used as an objective function (criterion) which should be maximized. As a result we received portfolio with minimum concentration of information. These portfolios also maximize the effective number of independent risk factors (Carmichael et al. 2015). The second measure of diversification used in the research was the Diversification Ratio (DR). The authors of this measure made an assumption that the diversification effect is connected with the difference between the risk of portfolio and the weighted sum of standard deviations of rates of return for stocks with the non-zero shares (Cheng and Roulac 2007; Choueifaty and Coignard 2008). Cheng and Roulac (2007) defined the diversification ratio as a quotient of the weighted sum of risk of components and the risk of entire portfolio. Formally this ratio is formulated as: DR =

σa σp

(3)

where σ p the standard deviation of the portfolio, σ a the weighted sum of standard deviations for components of non-zero shares, it is calculated as: σa =

N 

xi σi

(4)

i=1

where x i denotes share of i-th component in portfolio and σ i is standard deviation of i-th component, i = 1, 2, …, N.

24

A. Gluzicka

On the base of the value of DR we can’t state how much risk can be diversified, because values of this ratio are higher than 1. The DR can be used as a criterion for ordering portfolio according to the level of diversification. The higher value of DR means the higher level of diversification. Also, this ratio can be used as an objective function to construct the well-diversified portfolio. In this case as a result by maximizing the value of DR we received the Most Diversified Portfolio (MDP). This portfolio maximizes the distance between two definitions of portfolio volatility: the distance between the weighted sum of volatility of assets of portfolio and the total volatility of portfolio (Cheng and Roulac 2007).

3 The Effect of the Day and the Risk Diversification in 2010–2018 on the WSE The main goal of empirical research was to compare whether in the periods for which the effect of the day was found, a similar relationship can be determined for the parameter of the optimal investment portfolio. All calculations were made on the base of the daily rates of return of selected stocks from WSE. Various groups of stocks in different periods were analyzed, among others groups of components of such indices as: WIG20, mWIG40, sWIG80 as well as groups of randomly selected companies. In all cases the conclusions were similar, so below are sample results for the mWIG40 index (index of medium companies). The considered period was from 1st January 2010 to 31st December 2018. In this period only 29 stocks from the mWIG40 were listed without suspensions. In the first part of the research, for 29 selected stocks, the occurrence of the day effect were analyzed. For each stock, in a given year, the average rates of return for individual days of the week were calculated. Also, the rates of return were calculated for the given day of the week for the entire analyzed period 2010–2018. For all stocks in each considered periods the days of the maximum and minimum rates of return were established. The received results indicated that for the analyzed stocks, it was not possible to clearly determine the occurrence of the day effect for all data (stocks, rates of return). This effect, however can be found for selected companies for individual years. Information about the number of stocks with the highest and lowest rates of return on a given day of the week is given in Tables 1 and 2. In particular years, the effect of the day was assumed if at least 10 stocks achieved the maximum (minimum) rates of return on a given day. With this assumption, the maximum rates of return were established in the following days: 2011—Thursday, 2012—Wednesday, 2013—Monday, 2014—Monday, 2015—Friday, 2010–2018—Monday.

The Effect of the Day and the Risk Diversification on the WSE

25

Table 1 The number of stocks with a maximum rate of return on a given day of the week Monday

Tuesday

Wednesday

Thursday

Friday

2010

9

6

4

3

1

2011

7

6

4

11

1

2012

2

5

10

5

7

2013

10

9

1

2

7

2014

10

9

6

2

2

2015

1

1

2

9

16

2016

9

4

8

5

3

2017

7

7

2

8

5

2018

4

3

8

6

8

2010–2018

15

3

4

4

3

Table 2 The number of stocks with a minimum rate of return on a given day of the week Monday

Tuesday

Wednesday

Thursday

Friday

2010

0

6

7

1

14

2011

3

8

2

4

12

2012

4

10

3

3

9

2013

3

1

14

4

7

2014

1

0

0

10

18

2015

13

6

5

3

2

2016

2

4

0

10

13

2017

4

4

9

5

7

2018

8

9

6

4

2

2010–2018

10

2

5

2

10

For analyzed stocks the minimum rates of return were obtained in the following days: 2010—Friday, 2011—Friday, 2012—Tuesday, 2013—Wednesday, 2014—Friday, 2015—Monday, 2016—Friday, 2010–2018—Monday and Friday. In the second part of the research, for the rates of return received for a given day of the week, two types of the well-diversified portfolios were constructed: the Rao’s Quadratic Entropy Portfolios (RQE) and the Most Diversified Portfolios (MDP).

26

A. Gluzicka

Table 3 Days of the week, in which portfolios had the lowest and highest rate of return RQE portfolios

MDP portfolios

Max

Min

Max

Min

2010

Monday

Friday

Monday

Wednesday

2011

Thursday

Tuesday

Thursday

Monday

2012

Friday

Monday

Friday

Monday

2013

Tuesday

Wednesday

Tuesday

Wednesday

2014

Tuesday

Friday

Tuesday

Friday

2015

Friday

Tuesday

Friday

Monday

2016

Monday

Friday

Monday

Friday

2017

Friday

Wednesday

Monday

Wednesday

2018

Friday

Tuesday

Friday

Tuesday

2010–2018

Monday

Wednesday

Monday

Wednesday

Also this part of the study was conducted for each year separately as well as for the entire analyzed period 2010–2018. For each portfolio, the rate of return and the level of diversification (measured by the objective function) were calculated. Table 3 shows the days of the week when portfolios had the lowest and highest rates of return. The days of the week in which the lowest and highest level of diversification determined by a given measure (RQE, DR) were obtained are presented in Table 4. The comparison of results obtained for both types of diversified portfolios, indicates that in almost every case the analyzed effects occurred in both portfolios on the same day. For minimum rates of return, differences were found only for 2010, 2011 and 2015. The maximum rates of return were in different days only in 2017. However, the values of diversification coefficient differed two times: in 2010 (minimum values) and in 2018 (maximum values). Table 4 Days of the week in which portfolios had the lowest and the highest level of diversification RQE portfolios

MDP portfolios

Max

Min

Max

Min

2010

Tuesday

Friday

Tuesday

Monday

2011

Wednesday

Friday

Wednesday

Friday

2012

Thursday

Friday

Thursday

Friday

2013

Monday

Friday

Monday

Friday

2014

Monday

Thursday

Monday

Thursday

2015

Friday

Tuesday

Friday

Tuesday

2016

Wednesday

Friday

Wednesday

Friday

2017

Monday

Wednesday

Monday

Wednesday

2018

Wednesday

Tuesday

Monday

Tuesday

2010–2018

Wednesday

Friday

Wednesday

Friday

The Effect of the Day and the Risk Diversification on the WSE

27

The next step was the analysis of the level of rates of return for the RQE and MDP portfolios and rates of return of the analyzed stocks. Based on the obtained results, it was found that the occurrence of the effect of a day for stocks in a given year does not necessarily translate into the occurrence of the day effect for portfolio’s rates of return. Day-to-day compliance was obtained for RQE portfolios for the following years: 2010, 2011, 2015 and 2016 (maximum rates of return) and for years 2010, 2013, 2014, 2016 and 2018 (minimum rates of return). Similar results were received for MDP portfolios. The effect of the day coincided in 2010, 2011, 2015 and 2016 (maximum rates of return) and also for 2013, 2014, 2015, 2016 and 2018 (minimum rates of return). For the entire examined period 2010– 2018, the effect of Monday (for maximum rates of return) and effect of Wednesday (for minimum rates of return) was found. This relationship was found for both types of diversified portfolios. Analysis of the level of diversification (RQE, DR) showed that for the maximum rates of return, the maximum level of diversification was obtained on the same day only for 2013, 2014, 2015 and 2018. And for minimum values, compliance was recorded for 2010, 2011, 2016, 2017 and 2018. In addition, the received portfolios were also compared in terms of the level of diversification determined by the number of non-zero shares in the portfolio. Results were presented in the Table 5. However in this case any anomalies were not found. The level of the diversification measured by the number of stocks in portfolio is not connected in any way with the maximum and minimum average rates of return in a given day. Only in 2015, 2017 and 2018 the maximum average rates of return and the maximum level of diversification (the biggest number of stocks in portfolio) were in the same days. Table 5 The number of stocks with non-zero shares in the RQE (MDP) portfolios Monday

Tuesday

Wednesday

Thursday

Friday

2010

14 (14)

15 (16)

16 (15)

18 (18)

15 (15)

2011

16 (16)

11 (11)

18 (18)

12 (12)

12 (12)

2012

18 (18)

18 (18)

17 (17)

17 (17)

15 (15)

2013

23 (24)

17 (17)

13 (14)

17 (17)

15 (15)

2014

14 (14)

17 (18)

15 (15)

17 (17)

15 (15)

2015

12 (12)

16 (16)

16 (16)

16 (16)

20 (20)

2016

18 (18)

20 (20)

19 (19)

16 (16)

14 (14)

2017

20 (20)

20 (19)

20 (20)

19 (19)

20 (20)

2018

18 (18)

14 (14)

17 (17)

18 (18)

18 (18)

2010–2018

23 (23)

25 (25)

25 (25)

26 (26)

19 (19)

28

A. Gluzicka

4 Summary The article presents the results of empirical research on the occurrence of the effect of the day on the WSE. First, it was shown that only in selected periods can be determined the occurrence of such an effect. What’s more, anomalies in the distribution of rates of return may affect different days of the week, but in most cases, the dependencies concerned Mondays or Fridays. The research also showed that the occurrence of high (low) rates of return on a given day in specific periods does not guarantee that rates of return of portfolios constructed for the same data will achieve the highest (lowest) rates of return on the same days. Also, the level of diversification, measured by the Diversification Ratio (DR) or Rao’s Quadratic Entropy (RQE), only in a few cases strictly depends on the regularities of the rates of return. As it was mentioned before, this analysis was also conducted for the other groups of stocks. However the results in all cases were very similar. As an extension of the presented research, it is planned to carry out similar research, in which the effect of the month will be analyzed.

References Cai J, Li Y, Qi Y (2006) The day-of-the-week effect: new evidence from the chinese stock market. Chin Econ 29:71–88 Carmichael B, Boevi Koumon G, Moran K (2015) Unifying portfolio diversification measures using Rao’s quadratic entropy. CIRPEE working paper. Available via: https://www.cirano.qc.ca/files/ publications/2015s-16.pdf. Accessed 30 June 2017 Chang E, Kim C (1988) Day-of-the-week effects and commodity price changes. J Futures Markets 8:229–241 Cheng P, Roulac SE (2007) Measuring the effectiveness of geographical diversification. J Real Estate Portfolio Manag 13:29–44 Chinko M, Avci E (2009) Examining the day of the week effect in Istanbul stock exchange. Int Bus Econ Res J 8:45–50 Choueifaty Y, Coignard Y (2008) Toward maximum diversification. J Portfolio Manag 35:40–51 Cross F (1973) The behavior of stock prices on fridays and mondays. Financ Anal J 29:67–69 French K (1980) Stock returns and the weekend effect. J Financ Econ 8:55–69 Gay G, Kim T (1987) An investigation into seasonality in the futures market. J Futures Market 7:169–181 Gluzicka A (2018) Wybrane metody dywersyfikacji portfeli inwestycyjnych. Wydawnictwo Uniwersytetu Ekonomicznego w Katowicach Jaffe J, Westerfield R (1985) Patterns in Japanese common stock returns: day of the week and turn of the year effects. J Financ Quant Anal 6:261–272 Johnston R, Karacaw W, McConnel JJ (1991) Day-of-the-week effects in financial futures. J Financ Quant Anal 26:23–44 Kompa K, Witowska D (2007) Analiza własno´sci stóp zwrotu akcji wybranych spółek. Rynek Kapitałowy. Skuteczne inwestowanie – Zeszyty Naukowe Uniwersytetu Szczeci´nskiego, vol 6, pp 255–266 Markowitz H (1952) Portfolio selection. J Finance 7(1):77–91 Rao RC (1982a) Diversity: its measurement, decomposition, apportionment and analysis. Ind J Stat 44:1–22

The Effect of the Day and the Risk Diversification on the WSE

29

Rao RC (1982b) Diversity and dissimilarity coefficients: a unified approach. Theor Popul Biol 21:24–43 Rossi M (2015) The efficient market hypothesis and calendar anomalies: a literature review. Int J Manag Financ Acc 7:285–296 Smirlock M, Starks L (1986) Day-of-the-week and intraday effects in stock returns. J Financ Econ 9:197–210 Szyszka A (1999) Efektywno´sc´ rynku a anomalie w rozkładach stop zwrotu w czasie. Nasz Rynek Kapitałowy 12

Volatility and Liquidity in Cryptocurrency Markets—The Causality Approach Barbara B˛edowska-Sójka, Tomasz Hinc, and Agata Kliber

Abstract The dependency between volatility and liquidity is thoroughly examined in the contemporary literature on the financial markets. Especially, on the stock markets, liquidity tends to evaporate when volatility increases. Still, very few papers examine such relationships within the cryptocurrency markets. In this paper, we verify whether the volatility and liquidity of cryptocurrencies are interrelated. Our sample consists of 12 highly capitalized and traded cryptocurrencies. We consider both daily and weekly liquidity measures and thus extend the set of proxies. In order to examine the dependency between cryptocurrencies, the causality approach is employed. We use an asymmetric causality test to separate the influence of growths and declines of volatility to the changes of liquidity direction and the other way around. Overall, the empirical results indicate, inter alia, that high volatility is a Granger cause to high liquidity, which means that high volatility attracts investors and induce higher interest in the new financial instruments.

1 Introduction The first digital currency called Bitcoin was created in 2008 by Nakamoto. Since then, various currencies obtain enormous popularity on the financial markets. In 2018 there were over 1600 currencies traded worldwide. The most striking feature is a lack of central authority that governs and controls cryptocurrencies. The supply of these instruments is limited by the design of a protocol (Bouri et al. 2017b). The increase of the cryptocurrency markets is accompanied by the increase in scientific papers devoted to Bitcoin and other highly-capitalized cryptocurrencies. As the process of creating a cryptocurrency is so different from the one observed for traditional instruments, the organizational and institutional issues well-known B. B˛edowska-Sójka · A. Kliber (B) Pozna´n University of Economics and Business, al. Niepodległo´sci 10, 61-875 Poznan, Poland e-mail: [email protected] T. Hinc OLX Group, ul. Królowej Jadwigi 43, 61-872 Poznan, Poland © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_3

31

32

B. B˛edowska-Sójka et al.

from the financial markets should be considered in the new framework. Researchers have shown an increased interest in cryptocurrency markets: Bouri et al. (2017b), Kliber et al. (2019), Dyhrberg (2016), Klein et al. (2018) examine whether Bitcoin serves as a hedge and safe haven for different financial instruments. Their results vary depending on the sample period considered in the study, country and methods used. Bouri et al. (2017a) examine the relationship between returns and volatility during the 2013 crash on the Bitcoin market. Brauneis and Mestel (2018) test the efficiency of selected cryptocurrencies and link it to liquidity. Further, Kliber (2018) studies price, liquidity and information spillovers on the cryptocurrency markets, while Kliber and Włosik (2019) examine volume and price spillovers among the leading Bitcoin exchanges. In this paper, we focus on the dependency between liquidity and volatility which is crucial in the contemporary financial analysis. Such a relationship has been already quite heavily studied on the traditional markets. For the New York Stock Exchange (NYSE) Hiemstra and Jones (1994) find a bidirectional nonlinear causality between returns and volumes. Also, Brooks (1998) confirms the bidirectional volatilityvolume causality, although concludes that the volatility-volume relation prevails the volume-volatility one. Moreover, Chordia et al. (2005) find that the liquidity of large firms causes volatility of small ones. In a more recent paper, Ong (2015) shows that there is the bidirectional Granger volatility-volume causality where trading volume plays the dominant role. Eventually, Hautsch and Jeleskovic (2008) and Gold et al. (2017) show that in the developed markets liquidity changes may help predict future changes in volatility. Much less attention is devoted to the emerging markets; here results confirm bi-directional causality between the variables, but the causality from liquidity to volatility prevails (B˛edowska-Sójka and Kliber 2019). The Granger causality between volatility and liquidity has not been examined on the cryptocurrency markets yet. This paper fulfills this gap by studying whether there is a causal relationship in the Granger sense between volatility and liquidity. Moreover, we study how increases and decreases of volatility influence (ceteris paribus) the direction of changes of liquidity (and vice versa). We use data on twelve cryptocurrencies that are most intensively traded on the market. Our study shows that there exists bidirectional causality between volatility and liquidity. Most often high volatility leads to high liquidity, which might mean that volatility attracts investors to the cryptocurrency market and improves liquidity. For liquidity-volatility direction: low liquidity tends to cause low volatility. In general, we observe stronger causality from volatility to liquidity than from liquidity to volatility. The implication of the paper can be of interest to both practitioners and scholars. First, information about the direction of causality may help market participants to make better investment decisions. Secondly, our results shed some light on the nature and dependencies in the new segment of the financial market, which is still evolving. Our findings also allow one to make comparisons to the analogous features observed for the traditional financial assets.

Volatility and Liquidity in Cryptocurrency Markets …

33

2 Methods In the first part of the research, we run the analysis on daily volatility and liquidity proxies computed on data of the same frequency. Let us denote by Ht the highest price over a given day, L t the lowest one, Ot the opening price, and Ct the closing one. From the wide class of the liquidity proxies (Fong et al. 2017) we use these that can be calculated in daily frequency on the basis of daily data. In the study the following liquidity proxies are calculated: • Corwin and Schultz Efficient Spread Estimator, CSt , is given by the following formula (Corwin and Schultz 2012): 2 exp(αt ) − 1 1 + exp(αt ) √ √  2 · βt − βt γt αt = − √ √ ; 3−2 2 3−2 2   2  Ht Ht+1 2 + ln ; βt = ln Lt L t+1   max(Ht , Ht+1 ) γt = ln min(L t , L t+1 ) CSt =

(1)

• High-Low Range, HLRt , that is a transformation of the Closing Quoted Spread of Chung and Zhang (2014) and was used in e.g. B˛edowska-Sójka (2019): HLRt =

Ht − L t . 0.5 · (Ht + L t )

(2)

• Illiquidity of Amihud (2002), ILLIQ: ILLIQt =

|rt | . ln(volumet )

(3)

where volumet is a turnover within a given day (Olbry´s 2018). In order to avoid the impact of outlier observations we trim the upper distribution of Amihud illiquidity at 95% percentile (Lesmond 2005). • Volatility over volume, VoV t , as proposed in Fong et al. (2017): ln(Ht /L t ) VoVt = √ . volumet

(4)

• Eventually, we compute Abdi and Ranaldo (2017) spread, SAR, only for weekly data as:

34

B. B˛edowska-Sójka et al.

SARw = max(sw , 0),

(5)

where:  sw = 2

 1 5  (Cd − 0.5 · (Hd + L d )) · (Cd − 0.5 · (Hd+1 + L d+1 )) , (6) d=1 5

and subscript d denotes day of the week. In the original paper of Abdi and Ranaldo (2017) they used the expected value inside the square root. Here we decided to replace it by the unbiased estimator of the expected value, that is the sample mean. For the remaining measures in weekly frequency four prices, H t , L t , Ot and C t , are these, that are observed as the last in a given week, whereas volume is aggregated within a whole week. As far as volatility estimates are considered, daily volatility is proxied by: • squared returns, SQR: SQRt = rt 2

(7)

where rt is a daily logarithmic return on day t, • absolute returns, ABR, calculated as the absolute values of returns, • Garman-Klass volatility estimator (GK) (Garman and Klass 1980):  GKt =

 2  2 Ht Ct 0.5 · log − (2 log(2) − 1) · log Lt Ot

(8)

Additionally, in the case of weekly frequency we consider also volatility proxies that require data of higher frequency, that is realized volatility and bipower variation: • Realized variance (Andersen et al. 2003) RV =

T 

rt2

(9)

t=1

• Bipower variation (Barndorff-Nielsen and Shephard 2004) BV =

T 

| rt−1 || rt |

(10)

t=2

where T is the number of the days within a week. Based on daily data we calculate several liquidity proxies and nonparametric volatility estimates in both daily and weekly frequency and conduct asymmetric causality test of Hatemi-J (2012). We set the following research hypotheses:

Volatility and Liquidity in Cryptocurrency Markets …

• Growth in (volatility); • Growth in (volatility); • Decline in (volatility); • Decline in (volatility).

35

volatility (liquidity) does not Granger-cause growth in liquidity volatility (liquidity) does not Granger-cause decline in liquidity volatility (liquidity) does not Granger-cause growth in liquidity volatility (liquidity) does not Granger-cause decline in liquidity

If both hypotheses, (1) liquidity does not cause volatility and (2) volatility does not cause liquidity, are rejected, we conclude that the relationship is bi-directional.

3 Data Source and Sample Preparation We collected the data on cryptocurrency prices and volumes over different periods (see Table 5 in the Appendix), however, all ending in September 2017. The source of data was CoinIntraday (https://www.coinintraday.com/). This service provided open and close prices and volumes of cryptocurrencies. Tables 5 and 6 in the Appendix present the descriptive statistics of the changes in prices in daily and weekly frequency. The mean returns for daily quotations are not statistically different from zero. Median values are lower than means, which suggests negative skewness in the changes of both daily and monthly prices. The distribution is asymmetric and thus there is no normality in the series. We divide the empirical section into two parts: in the first one, we use daily data, while in the second part we aggregate them into weekly ones. The weekly liquidity and volatility measures are based on higher frequency data (daily in this case) such as realized volatility and bipower variation.

4 Empirical Results Tables 1 and 2 present the results of the asymmetric test of causality of Hatemi-J (2012) computed for daily data. Table 1 refers to volatility-liquidity direction; here we find the evidence that high volatility is the Granger cause for high liquidity. This holds for almost all combinations of estimates and proxies (from 56 to 94% of cases). For other relations the results are ambiguous—for some non-parametric measures and proxies they are stronger, e.g. high volatility induces low liquidity if CS is considered as a liquidity proxy (in 92% of cases). However, it does not hold with regard to any other liquidity proxies. On a similar basis low volatility is a Granger cause for high liquidity, if SQR or ABR are employed as volatility estimates and CS is considered as liquidity proxy.

36

B. B˛edowska-Sójka et al.

Table 1 Causality from volatility to liquidity in daily data Causality from

Causality to ILLIQ (%)

HLR (%)

CS (%)

VoV (%)

All (%)

High volatility, low liquidity ABR

67

25

92

25

52

SQR

25

25

92

17

40

GK

42

33

92

17

46

All

44

28

92

19

High volatility, high liquidity ABR

75

92

83

83

83

SQR

58

92

75

67

73

GK

33

100

92

83

77

All

56

94

83

78

Low volatility, high liquidity ABR

42

67

92

17

54

SQR

25

100

92

42

65

GK

33

42

75

33

46

All

33

69

86

31

Low volatility, low liquidity ABR

33

33

58

25

38

SQR

17

25

58

17

29

GK

50

42

42

25

40

All

33

33

53

22

In the case of liquidity-volatility direction (Table 2) we find one causal relationship, such that low liquidity is Granger cause to low volatility. For the remaining relations, the number of cases in which liquidity is a cause of volatility is lower than 50%. The results differ slightly, depending on the proxy used in the study. When we analyze causality from liquidity in daily data, HLR is most often the source of causality when liquidity is low, but not when liquidity is high. In the case of volatility, causality from ABR proxy is observed when volatility is high. For weekly data (see Tables 3 and 4), the strongest causal relations are observed for high volatility and high liquidity. When RV is a volatility estimate, and HLR—the liquidity proxy, the relations are the highest. Quite strong causal relations are also observed for VoV and ILLIQ as liquidity proxies. Low volatility is not a Granger cause for either high or low liquidity. In the case of the reversed relation, from liquidity to volatility, no causality is observed in general. However, for the particular cases of low volatility to low liquidity, we find that HLR is the Granger cause for all volatility estimates, while SAR is a Granger cause for RV and BV only.

Volatility and Liquidity in Cryptocurrency Markets …

37

Table 2 Causality from liquidity to volatility in daily data Causality from

Causality to ABR (%)

SQR (%)

GK (%)

All (%)

Low liquidity to high volatility ILLIQ

42

8

17

22

HLR

25

17

58

33

CS

42

25

50

39

VoV

17

17

25

19

All

31

17

38

ILLIQ

42

25

25

31

HLR

50

42

33

42

CS

33

42

42

39

VoV

25

33

33

31

All

38

35

33

ILLIQ

50

25

0

25

HLR

33

75

58

56

CS

50

50

33

44

VoV

42

33

25

33

All

44

46

29

High liquidity to high volatility

High liquidity to low volatility

Low liquidity to low volatility ILLIQ

67

67

33

56

100

100

100

100

CS

75

67

83

75

VoV

67

75

100

81

All

77

77

79

HLR

Thus, the proxy HLR seems to be the leader of the causality: it occurs either as a Granger cause or appears as a variable related to some volatility estimates when liquidity is high, while no similar pattern can be found when liquidity is low. In general, if there is any causality, this is rather observed in daily than in weekly data and is restricted to high volatility-high liquidity and low liquidity-low volatility cases.

38

B. B˛edowska-Sójka et al.

Table 3 Causality from volatility to liquidity in weekly data Causality from

Causality to ILLIQ (%)

HLR (%)

CS (%)

VoV (%)

SAR (%)

All (%) 25

High volatility, low liquidity RV

17

17

33

8

50

BV

17

17

42

8

42

25

ABR

25

17

50

17

50

32

SQR

8

8

8

0

33

12

GK

8

25

33

8

58

27

All

15

17

33

8

47

High volatility, high liquidity RV

42

92

8

33

50

45

BV

33

83

0

42

58

43

ABR

50

75

42

67

17

50

SQR

50

75

8

42

25

40

GK

33

92

25

75

25

50

All

42

83

17

52

35

0

33

8

17

17

Low volatility, high liquidity RV BV

15

8

25

33

0

25

18

ABR

25

8

42

8

42

25

SQR

0

25

17

0

50

18

GK

0

8

42

17

33

20

All

7

20

28

8

33

Low volatility, low liquidity RV

8

8

8

0

0

5

BV

8

8

17

0

8

8

17

17

8

8

0

10

ABR SQR

0

8

17

0

0

5

GK

25

17

8

17

0

13

All

12

12

12

5

2

5 Conclusion and Discussion In the paper, we examine the causality between volatility and liquidity for 12 cryptocurrencies. We find that high volatility leads to high liquidity, however, this relationship is more pronounced in daily than in weekly data. High volatility seems to attract investors to the cryptocurrency market, perhaps by offering additional returns, and thus liquidity is improved. Besides, in daily data high volatility is most often a

Volatility and Liquidity in Cryptocurrency Markets …

39

Table 4 Causality from liquidity to volatility in weekly data Causality from

Causality to RV (%)

BV (%)

ABR (%)

SQR (%)

GK (%)

All (%)

Low liquidity to high volatility ILLIQ

17

25

8

0

0

10

HLR

42

33

17

8

8

22

CS

8

8

0

0

8

5

VoV

25

17

0

17

8

13

SAR

33

25

17

17

8

20

All

25

22

8

8

7

25

17

17

17

17

18

HLR

8

17

8

0

33

13

CS

8

17

33

25

8

18

VoV

17

8

8

8

17

12

8

17

8

0

0

7

13

15

15

10

15

High liquidity to high volatility ILLIQ

SAR All

High liquidity to low volatility ILLIQ

25

25

17

8

8

17

HLR

33

42

17

8

8

22

CS

0

0

8

17

0

5

VoV

42

33

8

8

0

18

SAR

25

33

25

25

17

25

All

25

27

15

13

7

Low liquidity to low volatility ILLIQ

25

17

33

17

17

22

HLR

58

33

75

83

100

70

CS

25

25

0

0

17

13

VoV

17

25

25

25

58

30

SAR

67

58

8

17

8

32

All

38

32

28

28

40

Granger cause to low liquidity, but only if Estimated Effective Spread (CS) is used as a proxy. In high volatility time, liquidity evaporates easily.

40

B. B˛edowska-Sójka et al.

For liquidity-volatility direction, low liquidity tends to be a Granger cause to low volatility, but this is more pronounced in daily data than in weekly one. We observe stronger causality from volatility to liquidity than in the opposite direction: both volatility and volume attract investors. Since the results for daily and weekly frequency differ, we might conclude that the trading strategies for investors with various investment horizons are also different. Moreover, we find that there is an asymmetry in volatility-liquidity and liquidity-volatility relationships. When we compare the results for cryptocurrencies with the dependencies observed on financial markets for traditional assets, some differences are observed. First, in the traditional markets, we would expect that low liquidity would contribute to high volatility, while high liquidity, to low volatility (and vice versa) (B˛edowska-Sójka and Kliber 2019). The phenomenon of causality from high volatility to high liquidity is typical for the markets of highly speculative assets. Therefore, we can conclude that cryptocurrencies are still perceived as speculation assets. The results are somewhat contradictory—they differ depending on the proxies used. The reason is, that each proxy measure slightly different aspects of volatility and liquidity (B˛edowska-Sójka 2019; B˛edowska-Sójka and Echaust 2019). However, the relationships observed in prevailing cases can be assumed universal. The results of the study have important implications for both scholars and practitioners. They shed new light on the dependencies which have been already studied on traditional markets and show that the cryptocurrencies behave more like speculative assets, and not stable safe-haven ones, as some researchers suggest. Secondly, the information about the direction of causality may help the practitioners make better investment decisions. Acknowledgements This work was supported by the National Science Centre in Poland under grant no. UMO-2017/25/B/HS4/01546 as well as by the grant of Poznan University of Economics and Business: “The Future of money—cryptocurrencies, local currencies and cashless society”. We would like to thank prof. Bogdan Włodarczyk and other participants of the international conference WROFIN2019 in Wrocław for the fruitful discussion and inspiring comments, as well as the participants of the international conference ICOFEP2019 in Pozna´n. We also acknowledge the constructive feedback on the earlier versions of this paper from the participants at SEFIN seminar (https://sefin.ue.poznan.pl).

Appendix See Tables 5 and 6.

04.2016

06.2013

05.2014

04.2015

09.2016

08.2013

Litecoin

Monero

NEM

NEO

Ripple

06.2017

IOTA

Lisk

08.2015

07.2016

Dash

Ethereum classic

02.2014

Bitcoin cash

Ethereum

04.2103

07.2017

Bitcoin

Sample starts from

1365

378

905

1219

1365

534

102

425

777

1316

61

1365

Count

Table 5 Descriptive statistics of the log-returns of daily prices

0

1

1

0

0

0

0

1

1

1

0

0

Mean

7

13

9

8

6

17

12

11

9

9

15

4

Std

−4 −2 −3 −4 −4 −2

−51 −38 −35 −52 −62

−6

−37 −166

−3 −3

−3

−47 −47

−8

−130

−1

−24

25%

−45

Min

0

0

0

0

0

0

0

−1

0

0

−1

0

50%

2

5

4

4

1

4

7

3

3

3

5

2

75%

103

80

57

57

51

92

37

144

41

127

43

21

Max

Volatility and Liquidity in Cryptocurrency Markets … 41

42

B. B˛edowska-Sójka et al.

Table 6 Descriptive statistics of the log-returns of weekly prices Bitcoin

Count

Mean

25%

50%

75%

Max

195

1

Std 9

Min −27

−4

1

6

35

9

0

38

−45

−18

−12

−1

87

Dash

188

4

22

−89

−8

0

10

104

Ethereum

111

5

21

−43

−7

1

13

81

Ethereum classic

61

4

19

−31

−9

1

12

66

IOTA

14

2

33

−51

−21

−1

15

63

Lisk

76

2

36

−198

−9

−3

11

120

Litecoin

195

0

15

−47

−7

0

5

66

Monero

174

2

23

−59

−11

−2

12

132

NEM

129

5

21

−46

−6

1

15

87

Bitcoin cash

NEO Ripple

54

7

36

−75

−10

1

23

114

195

1

22

−59

−7

−1

6

186

References Abdi F, Ranaldo A (2017) A simple estimation of bid-ask spreads from daily close, high, and low prices. Rev Financ Stud 30:4437–4480 Amihud Y (2002) Illiquidity and stock returns: cross-section and time-series effects. J Financ Mark 5:31–56. https://doi.org/10.1016/S1386-4181(01)00024-6 Andersen TG, Bollerslev T, Diebold FX, Labys P (2003) Modeling and forecasting realized volatility. Econometrica 71:529–626 Barndorff-Nielsen OE, Shephard N (2004) Power and bipower variation with stochastic volatility and jumps. J Financ Econometrics 2:1–48. https://doi.org/10.1093/jjfinec/nbh001 B˛edowska-Sójka B (2019) The dynamics of low-frequency liquidity measures: the developed versus the emerging market. J Financ Stab 42:136–142. https://doi.org/10.1016/j.jfs.2019.05.006 B˛edowska-Sójka B, Echaust K (2019) Commonality in liquidity indices: the emerging European stock markets. Systems 7:1–18. https://doi.org/10.3390/systems7020024 B˛edowska-Sójka B, Kliber A (2019) The causality between liquidity and volatility in the Polish stock market. Financ Res Lett 30:110–115. https://doi.org/10.1016/j.frl.2019.04.008 Bouri E, Azzi G, Dyhrberg AH (2017a) On the return-volatility relationship in the Bitcoin market around the price crash of 2013. Economics 11:1–17. https://doi.org/10.5018/economics-ejournal. ja.2017-2 Bouri E, Molnár P, Azzi G et al (2017b) On the hedge and safe haven properties of Bitcoin: is it really more than a diversifier? Financ Res Lett 20:192–198. https://doi.org/10.1016/j.frl.2016. 09.025 Brauneis A, Mestel R (2018) Price discovery of cryptocurrencies: Bitcoin and beyond. Econ Lett 165:58–61. https://doi.org/10.1016/j.econlet.2018.02.001 Chordia T, Sarkar A, Subrahmanyam A (2005) Joint dynamics of liquidity, returns and volatility across small and large firms. Staff Reports, Federal Reserve Bank, New York Chung KH, Zhang H (2014) A simple approximation of intraday spreads using daily data. J Financ Mark 17:94–120. https://doi.org/10.1016/j.finmar.2013.02.004 Corwin SA, Schultz P (2012) A simple way to estimate bid-ask spreads from daily high and low prices. J Finance 67:719–759. https://doi.org/10.1111/j.1540-6261.2012.01729.x Dyhrberg AH (2016) Bitcoin, gold and the dollar—a GARCH volatility analysis. Financ Res Lett 16:85–92. https://doi.org/10.1016/j.frl.2015.10.008

Volatility and Liquidity in Cryptocurrency Markets …

43

Fong KYL, Holden CW, Trzcinka CA (2017) What are the best liquidity proxies for global research? Rev Financ 21(4):1355–1401. https://doi.org/10.1093/rof/rfx003 Garman M, Klass M (1980) On the estimation of security price volatilities from historical data. J Bus 53:67–78. https://doi.org/10.1086/296072 Gold N, Wang Q, Cao M, Huang H (2017) Liquidity and volatility commonality in the Canadian stock market. Math-In-Ind Case Stud 8(7):1–20 Hatemi-J A (2012) Asymmetric causality tests with an application. Empirical Econ 43:447–456. https://doi.org/10.1007/s00181-011-0484-x Hiemstra C, Jones JD (1994) Testing for linear and nonlinear Granger causality in the stock pricevolume relation. J Financ 49(5):1639–1664 Klein T, Pham Thu H, Walther T (2018) Bitcoin is not the new gold—a comparison of volatility, correlation, and portfolio performance. Int Rev Financ Anal 59:105–116. https://doi.org/10.1016/ j.irfa.2018.07.010 Kliber A (2018) Price, liquidity and information spillover within the cryptocurrency market. The case of bitfinex. Safe Bank 4:62–79. https://doi.org/10.26354/bb.4.4.73.2018 Kliber A, Włosik K (2019) Isolated islands or communicating vessels? Bitcoin price and volume spillovers across cryptocurrency platforms. Financ a úvˇer-Czech J Econ Financ 69:324–347 ´ Kliber A, Marszałek P, Musiałkowska I, Swierczy´ nska K (2019) Bitcoin: safe haven, hedge or diversifier? Perception of bitcoin in the context of a country’s economic situation—a stochastic volatility approach. Phys A Stat Mech Appl 524:246–257. https://doi.org/10.1016/j.physa.2019. 04.145 Lesmond DA (2005) Liquidity of emerging markets. J Financ Econ 77:411–452. https://doi.org/10. 1016/j.jfineco.2004.01.005 Olbry´s J (2018) Testing stability of correlations between liquidity proxies derived from intraday data on the warsaw stock exchange. In: Jajuga K, Locarek-Junge H, Orłowski L (eds) Contemporary trends and challenges in finance. Proceedings from the 3rd Wroclaw international conference in finance. Springer proceedings in business and economics. Springer, Cham, pp 67–79 Ong MA (2015) An information theoretic analysis of stock returns, volatility and trading volumes. Appl Econ 47:3891–3906. https://doi.org/10.1080/00036846.2015.1019040

Identification of the Factors Affecting the Return Rates of the Banks Listed on the Warsaw Stock Exchange Ewa Majerowska

Abstract Banks constitute a specific sector of stock-exchange-listed companies. The main factors affecting their rates of return are financial indicators. The aim of this paper is to identify these factors. Using annual data from 1998 to 2018, an unbalanced panel was created, covering 13 banks listed on the Warsaw Stock Exchange. The model proposed has been estimated for selected financial indicators. Next, factor analysis was used, by the means of which the latent factors were calculated. The results of the research allowed identification of the impact the latent factors have on the rates of return and showed that the main indicators determining the rate-of-return level were the operating profit margin, the return on equity, the equity ratio and the share of the operating expenses in the income on core operations.

1 Introduction One of the stocks sectors—the financial sector—includes a sub-sector of listed companies, that is, the banking sector. The companies belonging to this sector, just as other financial companies, differ greatly from non-financial companies. Their specificity is mainly associated with the method of their financing, i.e. bank deposits. They operate according to specific rules, in order to protect their customers’ best interests. They are under strict control of state authorities (in Poland this function is carried out by the Financial Supervision Committee), which is reflected on their balance sheets, which are created on the basis of relevant laws, including the Accounting Act. The purpose of the paper is to identify the factors that influence the rates of return on the assets of the banks traded on the Warsaw Stock Exchange, over the period from 1998 to 2018. The main research hypothesis states that it is possible to identify those factors and that they are the same for all banks. The analysis targeted almost all banks. The data were extracted from databases of Notoria Servis SA and the website stooq.pl. The models proposed were estimated using different approaches: E. Majerowska (B) Department of Econometrics, Faculty of Management, University of Gda´nsk, Gda´nsk, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_4

45

46

E. Majerowska

the OLS, fixed, the effects and the random effects assumptions. The estimations were performed in the Gretl package, the Statistica and Excel software. Two approaches, used for determination of the factors, are presented. The first one is based on the correlations occurring between the variables. The second one uses the factor analysis that allows identification of the latent factors influencing the return rates of the banks under analysis. The results allowed confirmation of the research hypothesis assumed. The paper is organized in the following way. After the introduction, a short overview and theoretical background of return-rate modelling in the banking sector are presented, followed by description of the data and the methodology proposed for the analysis. The next part contains the estimation results, followed by the conclusions drawn from the research.

2 Theoretical Framework The specificity of the financial sector is expressed in the economy by the fact that these enterprises comprise various types of money-market, capital, insurance and pensionmarket entities, which include banks, investment funds and insurance companies (Mishkin 2002). The banking sector is distinguished as part of the financial sector. It plays a dominant role in the economy. Activity of banks is related to the stability of the finances and the entire economy of a given country (Korenik and Korenik, 2004). In most countries, there is a dual (two-tier) banking system, which includes a central bank and commercial banks. Central banks mainly deal with currency issuance and act as the banks of all other banks. Their activities, involving collection of deposits and loan granting, have significant impact on the level of inflation, economic growth and the unemployment rate. These banks cooperate with international institutions, in terms of financial settlements, and manage public debt (Miedziak 2003). Commercial banks provide financial services to their clients. Due to the specificity of their operation and the influence of banks on the country’s economic situation, banks have been subject to strict legal regulations since the economic crisis of 1929–1932. Economists point out that banks, seeking to maximize their profits, make decisions under the conditions of risk, using other people’s money, hence the necessity of the legal regulations. At the international level, the banking system is regulated by the Basel Committee on Banking Supervision. Its activities result in creation of Capital Agreements, which define the capital adequacy standards, called Basel I, Basel II and Basel III (Małecki 2014). The author further writes that the banking sector in Poland differs from those in other developed countries. This results from the short history, the general conditions of the economy, and from the dominating role they play in the financial sector. Thus, accordingly, the author calls these banks universal. What is more, around two thirds of the banks’ capital is comprised of foreign capital. Taking into account the arguments mentioned above, it has been acknowledged that the factors influencing the financial results of banks should be reflected by the

Identification of the Factors Affecting the Return Rates …

47

share prices of the stock-exchange-listed banks, and thus by their rates of return. At the same time, it should be remembered that, with regard to the banking-sector entities, internal factors other than those identified for non-financial corporations are of importance. Viale (2007) states that identification of the common risk factors in financial companies is important in terms of the understanding of asset pricing in general and the public policy purposes. The issue of the impact legal regulations have on the results of the financial sector has been dealt with, among others, by Mohr and Wagner (2013). The authors investigated whether regulatory governance is positively related to financial soundness of the sector. They modelled regulatory governance and financial stability as latent variables, using a structural equation modeling approach. They examined financial stability as well as the macroeconomic and institutional environment. Using data on 55 countries, the authors confirmed that regulatory governance has beneficial influence on financial stability. Many researchers tried to identify the factors influencing the profitability of banks. Those factors mainly include financial ratios. Analyses were based on data from many developed and developing countries. Ullah (2016) analyzed the banking sector’s profits represented by ROA and ROE. The analyses targeted six Islamic banks and forty six conventional banks in GCC (Gulf Cooperation Council) countries. The author found that only internal factors have strong influence on the profitability of banking sector, while external factors have no influence on the banks’ profitability. The same topic was analyzed by Abugamea (2018). The author examined the impact bank-specific and macroeconomic factors have on the profitability of the banking sector in Palestine, identifying some internal factors that are correlated with both the ROA and the ROE. Profitability of the banks in Pakistan was analyzed by Gul et al. (2011). Kawshala and Panditharathna (2017) examined the impact of bank-specific factors on the profitability of domestic commercial banks in Sri Lanka. Banks in Saudi Arabia and Jordan constituted the subject of interest of Almazari (2014), whose research results indicated that there is significant positive correlation between ROA and such variables as TEA, TIA, and LQR, as well as negative correlation between ROA and such variables as NCA, CDR, CIR and SZE (Saudi Arabia). The author also found that there is significant positive correlation between ROA and such variables as LQR, NCA, TEA and CDR, as well as negative correlation between ROA and CIR, TIA and SZE (Jordan). The analysis was aimed at twenty three banks, over the period of 2005–2011. Regehr and Sengupta (2016) argued that size is not the only factor that affects a bank’s long-run profitability. They stated that profitability depends on two elements: individual banks and the markets which they operate on. The issues of modelling the returns on bank assets was also presented by Castrén et al. (2006). They tried to provide better understanding of the factors that may drive unexpected variability in the prices of the banks’ assets. Yang and Tsatsaronis (2012) proved that banks with higher leverage face higher costs of equity. They concluded that higher capital ratios are associated with lower costs of funding. Chan-Lau et al. (2012) showed that better-capitalized and less-leveraged banks have outperformed their outputs. Flannery and James (1984) argued that interest rate sensitivity is related

48

E. Majerowska

to the stock returns of banks. Gandhi and Lustig (2014) introduced a size factor as a component of bank returns. Influence of the bank size on the growth of systemic risk was studied by Laeven, Ratnovski and Tong (2016). Also Kohler ¨ (2015) showed that banks are more stable and profitable, if they increase their share of non-interest income. In one of recent papers, written by Brunnermeier et al. (2019) confirmation of the positive impact non-interest income has on the bank’s tail risk can be found.

3 Data and Methodology On the last day of the 2018, 15 banks were listed on the Warsaw Stock Exchange. Due to some problems with the data, two of these banks were omitted in the analysis. Ultimately, the analysis targeted 13 banks, over the period from 1998 to 2018. The time series constituted a non-balanced panel of yearly data. Description of the banks is provided in Table 1. The general form of the linear model proposed for identification of the factors is as follows: ri,t = β0 +

K 

βk f i,t + ξi,t

(1)

k=1

Table 1 Description of the banks Name of the bank

Abbr.

Date of first listing

No. of shares issued

Market value (mln PLN)

ALIOR BANK

ALR

12.2012

130,553,991

6717.00

BANK HANDLOWY

BHW

06.1997

130,659,600

6977.22

BANK MILLENNIUM

MIL

08.1992

1,213,116,777

10,857.40

BNP PARIBAS BOS´

BNP

05.2011

147,418,918

10,171.91

BOS

02.1997

92,947,671

711.98

GETIN HOLDING

GTN

05.2001

189,767,342

213.68

GETINOBLE

GNB

01.2012

1,044,553,267

575.55

IDEA BANK

IDA

04.2015

78,401,981

167.78

ING BANK ´ ASKI SL ˛

ING

01.1994

130,100,000

25,707.76

MBANK

MBK

10.1992

42,336,982

17,332.76

PEKAO

PEO

06.1998

262,470,034

27,638.09

PKOBP

PKO

11.2004

1,250,000,000

52,250.00

SANTANDER BANK POLSKA

SPL

06.2001

102,088,305

36,731.37

Identification of the Factors Affecting the Return Rates …

49

where r i,t is the rate of return of ith asset over time t, β are the structural parameters, ξi,t is the error term. Elements fk are the selected factors influencing the rates of return. They can be represented by the financial ratios of latent factors, obtained via exploratory factor analysis (EFA). The model will be estimated by a panel OLS, followed by a fixed effects model and a random effects model. Three tests of panel data estimation were applied for selection of the model: a joint significance test applied to different groups of means, the Breusch-Pagan test and the Hausman test. The null hypothesis of the first test (a joint significance test) assumes that the pooled model is correct, while the alternative suggests the use of the fixed effects model. The null hypothesis of the Breusch-Pagan test assumes that the pooled model should be selected as opposed to the alternative one, to make the random effects model better. The last test—the Hausman test—allows determination of whether the random effects model is appropriate (the null hypothesis), in place of the fixed effects model (the alternative hypothesis). Based on the literature review and using the bank balance sheets, the internal ratios (coefficients) were selected. They can be divided into four groups, listed in Table 2. Taking into account the above-mentioned variables, represented by the financial ratios, three approaches to the model have been proposed. In the first one, all the variables are included in the model. Then, only those are selected, which significantly explain the dependent variable. The second approach is based on the correlations between the variables in each group. Only those variables have been selected for the estimation, which are correlated with the dependent variable and not correlated between the groups. The last Table 2 Description of the variables Group

Name of the variable

Abbr.

Group

Name of the variable

Abbr.

Gr. 1

Operating profit margin

MOP

Gr. 3

Earning assets ratio

EAR

Gross profit margin

MGP

Ratio of liabilities to customers

LR

Net profit margin

MNP

The equity ratio

ER

Net interest margin

MNI

Coverage of loans and advances to customers

CL

Return on equity

ROE

Funding ratio of non-income assets

FR

Return on assets

ROA

IMF liquidity ratio

IMF

The interest rate of interest-earning assets

RIE

Operating costs/net banking

OCN

The interest rate of interest-bearing liabilities

ROL

Operating expenses/income from core operations

OEI

The range of interest

RI

Operating costs/assets

OCA

Productive assets

PA

Gr. 2

Gr. 4

50

E. Majerowska

approach involves factor analysis. This method is used for decomposition of the information in a set of variables. The factor analysis assumes that each of the variables is made up of a linear combination of common factors. They constitute hidden factors that affect the variable and possibly affect other variables (Aczel and Sounderpandian 1999). The analysis allows identification of the latent factors, based on given variables. These factors make up the explanatory variables in the model.

4 Results As already mentioned, the model represented by Eq. 1 has been estimated using three approaches. Table 3 presents the correlations occurring between the variables. Based on those correlations, a set of 10 variables was selected, i.e.: MOP, MNI, ROE, RIE, ROL, ER, FR, OCN, OEI and PA. Subsequently, the best model was selected by testing individual significance of the structural parameters, applying the joint significance test to different groups of means, the Breusch-Pagan test and the Hausman. In the next step, factor analysis was applied. The results of all estimations are presented in Table 4. Finally, five models have been estimated. Model (1) was based on selection of the variables that were significant, out of all the variables selected for the study. Models (2) and (3) have been estimated on the basis of the correlations occurring between the variables. In version (2), the variable ROA was included; in (3), it was replaced by the variable ROE. Models (4) and (5) are based on the factor analysis. The results provided in Table 4 show that, in all cases, the pooled model described the return rates on the assets of the banks traded on the WSE, which is indicated by the values of the statistics of the three diagnostic tests used. All models are poorly fitted to the real data. Furthermore, in all cases, there is no first-order autocorrelation of the random components. The main ratios that influence the rates return are the net interest margin (negative relationship), the return on equity (positive), the interest rate on interest-bearing liabilities (negative), the equity ratio (positive) as well as the ratio of the operating expenses to the income from core operations. In the first model, the operating profit margin, the gross profit margin, the net profit margin, the IMF liquidity ratio and the ratio of the operating cost to the number of assets were statistically significant. Additionally, the Choi-meta tests, i.e. stationarity tests of the variables (unit root tests), such as the Inverse chi-square test, the Inverse normal test and the Logit test, were applied. In 13 cases, the variables were stationary, while in the remaining cases, the first differences of the variables were stationary. The factor analysis showed that four latent factors can be identified. Factor 1 includes such variables as ROL and RI, Factor 2—RIE and PA, Factor 3—MOP, the MGP, MNP, ROA, OCN and OCA, Factor 4 included the following variables: ER, CL, IMF and OEI. In model (4), all Factors have been included, while in model (5), only one—the statistically significant one. The properties of the four selected factors are given in Table 5. The value of the cumulative variance indicates that over 74%

1.000

LR

IMF

FR

CL

ER

0.561

EAR

MNI

LR

1.000

MGP

MNP

0.999

MOP

MGP

MNI

−0.426

−0.458 1.000

0.837

−0.157

−0.234

1.000

CL

1.000

0.317

0.330

0.345

ER

1.000

0.993

0.993

MNP

Table 3 The correlations between the variables

1.000

−0.045

0.968

−0.039 1.000

0.862

−0.203

−0.162

IMF −0.387

RI

−0.239

−0.651

FR

ROL

RIE

ROA

ROE

ROA 1.000

0.830

PA

OCA

OEI

OCN

RIE

1.000

0.2823

OEI

1.000

0.037

0.062

ROL

1.000

−0.023

0.837

OCA

1.000

0.035

0.004

0.019

RI

1.000

0.103

0.119

−0.085

PA

1.000

−0.999

0.006

−0.002

−0.016

Identification of the Factors Affecting the Return Rates … 51

52

E. Majerowska

Table 4 Estimates of the model and the test statistics Variables

Model (1)

Model (2)

Model (3)

Model (4)

Model (5)

0.028

−0.742***

−0.078

−0.227

0.002

MOP

4.884**

1.216***

MGP

−6.992***

MNP

2.122*

const

−1.194***

MNI

−0.496*

ROE

1.105**

1.343***

ROL

−4.859**

−2.211*

LR

0.698**

ER IMF

2.170*

2.040*

1.242***

0.207**

1.261***

OEI

−0.993*

OCA

10.316*

Factor 1

−2.624

Factor 2

−0.005

Factor 3

−0.026

Factor 4

0.069*

0.052* 0.011

Adjust. R-squared

0.252

0.165

0.195

0.001

DW

2.138

2.022

2.046

1.848

1.858

Joint sign. test

0.461

0.639

0.354

1.128

1.012

Breusch-Pagan test

2.132

0.902

3.027

1.462

1.480

Hausman test

5.712

5.597

3.421

4.641

0.246

* statistically significant at the level of 0.1 ** statistically significant at the level of 0.05 *** statistically significant at the level of 0.01

Table 5 Factor characteristics Factor

Eigenvalues

Variance (%)

Cumulative eigenvalues

Cumulative variance (%)

Factor 1

1.967

10.351

1.967

10.351

Factor 2

2.381

12.532

4.348

22.883

Factor 3

5.843

30.751

10.190

53.634

Factor 4

3.872

20.377

14.062

74.011

of the variation is explained by the four factors. Models (4) and (5), estimated based on the factor analysis (Table 4), suggest that only Factor 4 influences the rates of return on assets. It targets four of the variables selected, which means that the equity ratio, coverage of the loans and credits for the customers, as well as the IMF liquidity ratio are positively correlated with the rates

Identification of the Factors Affecting the Return Rates …

53

of return. The last variable—the ratio of the operating expenses to the income from core operations, included in Factor 4, shows negative influence.

5 Conclusions The banking sector, which makes up part of the financial sector within listed companies, constitutes a specific part of the economy. Because of the sources of financing bank operations, the factors affecting the rates of return on the shares of the banks listed on the stock exchange include, to a large extent, the internal financial ratios. These ratios are being observed by investors, which, at the same time, determines their willingness to invest in the assets issued by these entities. This means that the better the values of the ratios, from the investors’ perspective, the greater the interest in the values, which should result in an increase in the share prices. Empirical analysis of the shares of the banks traded on the WSE showed that the higher the profit margin, the higher the rates of return. Similarly, the higher the return on equity, the higher the rate of return. Conversely, the higher the value of the interest rate on interest-bearing liabilities, the lower the return rate. As such, identification of internal factors influencing the level of the return rate was possible. It can therefore be concluded that the research hypothesis has been confirmed. Using factor analysis, it was also possible to find the latent factors that significantly influence the variable under analysis. Nonetheless, it should be remembered that the analysis carried out is burdened by a small number of the banks analyzed. The analysis was based on the share quotations of thirteen banks. In a further step, the ownership structure of these banks and their internal policies should be examined as well. Analysis should also be expanded to include additional factors, determined on the basis of balance sheet items. Inclusion of, for example, credit risk, represented by the share of unregulated loans in the total loans, or other indicators that pertain to risk, such as capital adequacy ratios, should be taken into account.

References Abugamea G (2018) Determinants of banking sector profitability: empirical evidence from Palestine. MPRA paper no. 89772 Aczel A, Sounderpandian J (1999) Complete business statistics. The McGraw-Hill, Toronto Almazari A (2014) Impact of internal factors on bank profitability: comparative study between Saudi Arabia and Jordan. J Appl Finance Bank 4(1):125–140 Brunnermeier MK, Dong GN, Palia D (2019) Banks’ non-interest income and systemic risk. Review of corporate financial studies, forthcoming Castrén O, Fitzpatrick T, Sydow M (2006) What drives EU banks’ stock returns? bank-level evidence using the dynamic dividend-discount model. Working paper series no. 677. European Central Bank

54

E. Majerowska

Chan-Lau JA, Liu EX, Schmittmann JM (2012) Equity returns in the banking sector in the wake of the great recession and the European sovereign debt crisis. IMF working paper MCM WP/12/174 Flannery MJ, James CM (1984) The effect of the interest changes on the common stock returns of financial institutions. J Finance 39(4):1141–1153 Gandhi P, Lustig H (2014) Size anomalies in U.S. Bank stock returns. J Finance 70(2):733–768 Gul S, Irshad F, Zaman K (2011) Factors affecting bank profitability in Pakistan. Rom Econ J 39:61–87 Kawshala H, Panditharathna K (2017) The factors effecting on bank profitability. Int J Sci Res Publ 7(2):212–215 Korenik D, Korenik S (2004) Podstawy finansów. PWN, Warszawa Kohler ¨ M (2015) Which banks are more risky? the impact of business models on bank stability. J Financ Stab 16:195–212 Laeven L, Ratnovski L, Tong H (2016). Bank size, capital, and systemic risk: some international evidence. J Bank Finance 69(1):S25–S34 Małecki W (2014) Regulacje sektora bankowego. In: Stacewicz J. (ed) Polityka gospodarcza jako gra w wyzwania i odpowiedzi rozwojowe, Prace i materiały Instytutu Rozwoju Gospodarczego SGH, 94:153–184 Miedziak S (2003) Bankowo´sc´ i podstawy rynku finansowego—wykłady i c´ wiczenia. Diffin, Warszawa Mishkin FS (2002) Ekonomika pieni˛adza, bankowo´sci i rynków finansowych. PWN, Warszawa Mohr B, Wagner H (2013) A structural approach to financial stability: on the beneficial role of regulatory governance. Available via SSRN. https://ssrn.com/abstract=2241083 or https://dx.doi. org/10.2139/ssrn.2241083 Regehr K, Sengupta R (2016) Has the relationship between bank size and profitability changed? Econ Rev Second Q 2016:49–72 Ullah N (2016) Influencing factors of profitability on the banking industry: a case study of GCC countries. MPRA paper no. 69124 Viale AM (2007) Common risk factors in bank stocks. A dissertation Submitted to the Texas A&M University. Available via. https://core.ac.uk/download/pdf/4272705.pdf Yang J, Tsatsaronis K (2012) Bank stock returns, leverage and the business cycle. BIS Q Rev 2012:45–59

Conventional and Downside Betas and Higher Co-moments in the Asset Pricing Relations Lesław Markowski

Abstract This study examined the cross-sectional relationships between realized returns and systematic risk measures using sub-sectoral indices quoted on Warsaw Stock Exchange. In addition to the classical beta, the aim of the study is also to check the impact of higher order co-moments on the sub-indices pricing. The unconditional risk-return relationships are estimated using classical and downside measures and conditional relations in terms of market condition. The downside risk premiums are significantly positive which means, that the downside risk is priced in Polish stock market. The downside risk measures outperform the classical ones. While the market condition is incorporated as the conditioning variable the risk factors acquire significance. Beta coefficient and co-kurtosis generate positive premiums in the up market and negative in the down market. Using the sub-sector indices returns no significant co-skewness pricing in the up market is found. The research show that measures of systematic risk such as the beta coefficient and higher order co-moments in conventional and downside approach are appropriate risk factors in asset pricing.

1 Introduction The rational behaviour of investors allows to achieve the equilibrium of the capital market, described in the modern theory of finance in the form of a capital asset pricing model (CAPM). It is widely used to assess including the cost of capital, therefore its verification of validity is desirable. Along with the testing of the capital asset pricing model, its criticism appeared in the context of different view of risk. Risk-averse investors differently perceive deviations below and above, e.g. the expected value. Deviations above the threshold are treated as a profit, but below the threshold as a potential loss. The risk-averse attitude creates the new conception of risk called downside risk. This theory takes on special significance in the case of abnormal L. Markowski (B) Department of Finance, Faculty of Economic Sciences, University of Warmia and Mazury, Olsztyn, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_5

55

56

L. Markowsk et al.

returns distributions and their asymmetry. Many studies in developed capital markets indicate that downside risk measures outperform classical ones (Post and Van Vliet 2006). Ang et al. (2006) based on individual companies listed on the NYSE, AMEX and NASDAQ show that investors are rewarded with a market premium for the downside risk, which means that the assets with higher downside beta coefficients reaches higher rates of return on average. Furthermore, the downside CAPM reaches higher level of explanatory power of returns than classical CAPM (Chen et al. 2009; Tsai et al. 2014). In the emerging markets, Estrada (2002, 2007) proposed new measures of the downside risk and showed that these measures described returns much better, especially in the situation with skewed returns distribution. The occurrence of asymmetry in returns distributions prompts to use in asset pricing other systematic risk measures called moments or co-moments such as coskewness or co-kurtosis, both in the conventional and downside framework. Investors usually prefer the positive asymmetry of the market portfolio and expect a premium for positive co-skewness. These coefficients measure the contribution of a given asset to the asymmetry and kurtosis of the market portfolio. The support for a significant valuation of the co-moments can be found among others in the works of Duc and Nguyen (2018) and Neslihanoglu et al. (2017). Other research e.g. Mora-Valencia et al. (2017) do not confirm the significance of co-skewness, making its pricing depend on the choice of the estimation method. Chhapra and Kashif (2019) prove that co-skewness and co-kurtosis are not a significant risk source because they fail to uncover average investments returns in Pakistan Stock Exchange. However, it should be noticed, that co-skewness reflects only a certain aspect of the downside risk, independent on the downside beta coefficient. The application in this case has the downside co-skewness. The study of Alles and Murray (2013) show that this measure is important for the assets pricing as opposed to the downside beta. Such results were confirmed in the Korean capital market as well (Truong and Kim 2018). The crucial issues regarding CAPM testing also apply to the methodology of verifying the relationship between returns and systematic risk. In contrast to the unconditional procedure proposed by Fama and MacBeth (1973), the proposals of Pettengill et al. (1995) suggest that relations between the realized rates of return and the beta coefficients are positively correlated with the sign of the difference between the realized market return and the risk-free rate. The analysis and confirmation of conditional relationships has been supported in many studies (Bilgin and Basti 2014; Jagannathan and Wang 1996). The obtained results, based on conditional regression illuminated often pessimistic conclusions regarding the equilibrium model. The problems with testing of CAPM are reflected in this paper in the form of some research questions and theses. Firstly, the downside risk is priced on the Polish capital market. Secondly, beta coefficient is an appropriate measure of risk and is related to the condition of the stock market. Third, investors are rewarded for bearing a risk associated with the co-moments, co-skewness and co-kurtosis of returns’ distribution. These three statements are the main theoretical premises for empirical investigation in this work. The novel contribution to this study is an application of downside co-skewness and co-kurtosis in cross-sectional regressions.

Conventional and Downside Betas …

57

The aim of the article is the verification of unconditional relationships of realized returns in relations to the systematic risk measures and the relationships conditional on market conditions. The scope of the examined assets includes sub-sector indices, representing a significant part of companies listed on the Warsaw Stock Exchange. The Author in this study show the differences in the asset pricing, considering the conventional and the downside approach.

2 Methodology 2.1 Risk Measures The quantification of the risk-return relation in the context of semi-measures is based on the lower partial moments (Rutkowska-Ziarko et al. 2019). The theory distinguishes many forms of the downside beta coefficients differentiating them from the formula and the threshold rate, which can be a risk-free rate (Hogan and Warren 1974; Bawa and Lindenberg 1977) or mean return (Harlow and Rao 1989). In the latter case, the downside beta coefficient can be defined as follows: βiH R =

E[(Rit − E(Ri ))min(R Mt − E(R M ); 0)] . E[min(R Mt − E(R M ); 0)]2

(1)

where Rit , R Mt denote rates of return for the i-th asset and the rate of return of the market portfolio respectively. In this research, the above coefficient is used to study the relationships between realized returns and the risk in the downside approach. The classical portfolio analysis limits investors’ preferences to the first two moments, fully featured the return distribution. An investor characterized by decreasing, absolute risk aversion, building portfolios, draws attention to the marginal contribution of a given asset to the portfolio asymmetry. Therefore, he will prefer stocks that increase the positive skewness of portfolio return distribution. Considering the marginal contribution of a given asset to the asymmetry or kurtosis of the market portfolio, important risk measures may be co-moments called co-skewness and co-kurtosis which are given as (Galagedera et al. 2003):   E (Rit − E(Ri ))(R Mt − E(R M ))2 γi = E[(R Mt − E(R M ))]3

(2)

  E (Rit − E(Ri ))(R Mt − E(R M ))3 . θi = E[(R Mt − E(R M ))]4

(3)

and

58

L. Markowsk et al.

Investors prefer assets with positive co-skewness because they increase skewness of the market portfolio distribution towards the asymmetry of that portfolio. Therefore, the premium for co-skewness depend on the asymmetry of the market portfolio. In a situation where the market portfolio is negatively skewed investors require a positive premium for risk of co-skewness, whereas when the market portfolio is positively skewed investors will be willing to pay for the contribution of an assets with positive co-skewness to the portfolio. Then investors should expect a negative risk premium. The kurtosis of the return distribution is interpreted as the level of concentration of individual observations around the expected return. In addition, the risk of extreme changes in rates of return resulting in a fat-tail, may occurs in the case of leptokurtic distributions with the high kurtosis values, therefore the kurtosis should be treated as a certain aspect of investment risk in each asset. Thus, a positive risk premium will be expected for the kurtosis. Similarly, to the downside beta coefficients, also higher order co-moments of the return distribution as an asymmetric risk measures can be presented in a downside framework as follows (Galagedera 2009): γid

  E (Rit − E(Ri ))min(R Mt − E(R M )); 0)2 = E[min(R Mt − E(R M ); 0)]3

(4)

θid

  E (Rit − E(Ri ))min(R Mt − E(R M )); 0)3 = . E[min(R Mt − E(R M ); 0)]4

(5)

and

Testing the pricing of the above measures is aimed at comparing them with measures in the conventional approach and examine which measures in the downside approach are more useful from the investor’s point of view.

2.2 Modifications of CAPM. Unconditional Relationships in Conventional and Downside Approaches The study of the relationship between systematic risk measures and the realized returns of sub-sector indices is carried out in accordance with a two-stage regression procedure. In the first stage, based on a part of the total research sample, the beta coefficients of individual indices are estimated: Rit = αi + βi R Mt + ξit (i = 1, . . . , N ; t = 1, . . . , T1 )

(6)

where Rit , R Mt —rates of return for the i-th sector and the rate of return of the market portfolio respectively; αi —constant term; βi —beta coefficient of i-th asset; ξit — random error term of i-th equation; T1 —length of estimation period; N—number of

Conventional and Downside Betas …

59

sector indices. Other measures, i.e. the downside beta coefficient and higher order moments are determined in Part 2.1 and 2.2. In the second stage, cross-sectional regression is used where the dependent variables are realized excess return of sub-sector indices above the risk-free rate, and the independent variables are systematic risk measures for the indices estimated in the first stage of the procedure. Unconditional relations are estimated during the CAPM testing period in each observation interval, separable from the previous estimation period. The cross-sectional model is as follows: 

Rit − R f t = λ0t + λ1t R M i + ηit (i = 1, . . . , N ; t = T1 + 1, . . . , T ) 





HR





(7)

d

where R M i (risk measure) is defined as β i , β i , γ i , γ id , θ i and θ i ; R f t —risk-free rate; λ0t , λ1t —parameters of t-th equation; ηit —random error term of t-th equation; T —the whole period length. The average estimation of λ1 which denotes a market risk premium for beta factor should be significantly positive (Tang and Shum 2003): H0 : E(λ1 ) = 0; H1 : E(λ1 ) > 0





for beta coefficients

(8)

A set of hypotheses regarding the parameter λ1 , reflecting the importance of coskewness in the pricing of indices, depending on the sign of asymmetry of the market portfolio distribution is given: H0 : E(λ1 ) = 0, H0 : E(λ1 ) = 0,

H1 : E(λ1 ) > 0 where As M < 0 for co-skewness H1 : E(λ1 ) < 0 where As M > 0 for co-skewness

(9)

where As M denotes skewness of the market portfolio. The verification of pricing for co-kurtosis associated with the dispersion of returns should be associated with a positive risk premium. That is: H0 : E(λ1 ) = 0; H1 : E(λ1 ) > 0

for co-kurtosis.

(10)

Finally, because an asset or portfolio is not correlated with a market portfolio has the expected rate of return equal to the risk-free rate the constant term λ0 should be insignificantly differ from zero: H0 : E(λ0 ) = 0; H1 : E(λ0 ) = 0.

(11)

The test for the above hypotheses is the one-sided or two-sided critical area t test for one mean value.

60

L. Markowsk et al.

2.3 Modifications of CAPM. Relationships in Different Market Conditions A conditional CAPM equation, due to the sign of the market excess return, in the testable version is given as follows: 



Rit − R f t = δλU0t + (1 − δ)λ0tD + δλU1t R M i + (1 − δ)λ1tD R M i + ηit ,

(12)

where δ is a dichotomous variable used to determine the positive and negative market excess return, then δ = 1 if (R Mt − R f t ) > 0 and δ = 0 if (R Mt − R f t ) < 0; R M i (risk measure) is defined as β i , γ i , θ i ; λU0t , λ0tD , λU1t , λ1tD ,—parameters of t-th equation; ηit —random error term of t-th equation. When the market excess return is lower than zero, then an inverse relationship between the risk measures and realized returns are expected. This approach for average parameters λU1 , λ1D in the case of beta coefficient and co-kurtosis is result in hypotheses as follows: 







        H0 : E λU1 = 0; H1 : E λU1 > 0 and H0 : E λ1D = 0; H1 : E λ1D < 0. (13) Rejection of null hypotheses in both cases indicates the occurrence of systematic, conditional relations between these risk measures and the realized subsector indices returns. The set of hypotheses for co-skewness is conditioned by the asymmetry of the market portfolio and is opposite to this sign.

3 Data A dataset for empirical analyses of the CAPM relationships are a full-time series of daily simply returns of 10 sub-sector indices quoted on the Warsaw Stock Exchange in the period from 3 January 2011 to 28 December 2018 what spans 1996 observations. These subindices are: WIG-banking, WIG-construction, WIG-chemicals, WIG-energy, WIG-IT, WIG-media, WIG-real estate, WIG-oil & gas, WIG-food and WIG-telecom. The WIG index is used as the market portfolio approximation and the proxy for the risk-free rate is the interest rate of The Poland 10Y Government Bonds. According to the adopted methodology, the total sample period is divided into two four-year sub-periods. The first of them (2011–2014) is the period of estimation, while the second one (2015–2018) is the tested of the CAPM period. The up and down-market condition are an observation with positive (R Mt − R f t ) > 0 and negative (R Mt − R f t ) < 0 market excess returns. In this way, we can distinguish the number of the following observations: 521 up market days and 477 down market days.

Conventional and Downside Betas …

61

4 Results Firstly, based on the estimation period, the unconditional relationships of subindices returns and risk measures are examined and the estimations are reported in Tables 1 and 2. The results of the risk premiums associated with risk measures are positive, but only in the downside approach statistically significant at 10%. An exception is the premium for co-skewness, which is significant at the 5% level and it is positive what was expected, taking into account the negative asymmetry of the market returns’ distribution (As M = −0.44). The lack of convincing results of CAPM validity is the outcome of overlapping premium values from periods of positive and negative market excess returns. These results, especially those referring to the downside beta, support the earlier results presented in the study (Ang et al. 2006) or the results Table 1 Estimates of unconditional CAPM relations in the conventional framework Coefficient

Mean

t-Stat

p-value

R2

0.139



Model: Rit − R f t = λ0t + λ1t β i + ηit λ0t

−0.0006

−1.030

0.303

λ1t

0.0009

1.221

0.111 0.201



Model: Rit − R f t = λ0t + λ1t γ i + ηit λ0t

−0.0008

−1.279

λ1t

0.0010

1.686

0.118

0.046**



Model: Rit − R f t = λ0t + λ1t θ i + ηit λ0t

−0.0006

−1.012

0.312

λ1t

0.0008

1.244

0.107

0.137

Notes ** indicates significance at the 5% Source own study

Table 2 Estimates of unconditional CAPM relations in the downside framework t-Stat

p-value

R2

−0.0008

−1.178

0.239

0.135

0.0010

1.361

Coefficient

Mean

Model: Rit − R f t =

HR λ0t + λ1t β i

λ0t λ1t



+ ηit 0.087***

Model: Rit − R f t = λ0t + λ1t γ id + ηit 

λ0t

−0.0008

−1.257

λ1t

0.0010

1.462 

0.209

0.134

0.072***

d

Model: Rit − R f t = λ0t + λ1t θ i + ηit λ0t

−0.0008

−1.274

λ1t

0.0010

1.502

Notes *** indicates significance at the 10% Source own study

0.203 0.067***

0.133

62

L. Markowsk et al.

of multifactor models (Lee et al. 2008) and (Tsai et al. 2014). The importance of downside co-skewness can be found separately in developed and emerging markets, among others in Galagedera (2009). Similar results were also obtained for sectoral subindices using monthly rates of return (Markowski 2019). The next part of the analyses are cross-sectional regressions considering market condition according to the Eq. (12). These analyses provide important practical implication on the relationship between the realised returns and the risk factors and they are presented in Table 3. The results of the conditional relations show that the average estimation of the risk premiums for beta coefficient and co-kurtosis are statistically significant at the level of significance α = 0.01 and positive in the up market and negative in the down market. The results are consistent with expectations and imply the companies with high beta and co-kurtosis in periods with a positive market excess return (with a negative market excess return) achieve higher rates of return (lower rates of return) than companies with relatively lower beta coefficients. In relation to beta coefficient the obtained results are consistent with the research like Nurjannah et al. (2012) and Trzpiot and Kr˛ez˙ ołek (2006). The premium for co-skewness as to the sign is inconsistent with the hypotheses in the up market (As M = 1.1) but in the down market the premium has an appropriate sign (As M = −2.2) and turn out to be statistically significant at the 10%. These Table 3 Estimates of conditional on market condition CAPM relations Mean Model: Rit − R f t =

δλU 0t

Up market δ = 1 Down market δ = 0 Model: Rit − R f t = δλU 0t Up market δ = 1 Down market δ = 0 Model: Rit − R f t = δλU 0t Up market δ = 1 Down market δ = 0

t-Stat 

D + δλU β + (1 − δ)λ0t 1t i λU −0.0038 0t λU 0.0099 1t D λ0t 0.0029 D λ1t −0.0091 D + δλU γ + (1 − δ)λ0t 1t i λU 0.0041 0t λU 0.0008 1t D λ0t −0.0061 D λ1t 0.0011 D + δλU θ + (1 − δ)λ0t 1t i 



p-value

R2



Dβ + η + (1 − δ)λ1t it i

−4.861

0.000*

11.282

0.000*

3.444

0.000*

−9.870

0.000*

0.140 0.139



Dγ + η + (1 − δ)λ1t it i

4.924

0.000*

1.013

0.844

−7.313

0.000*

1.388

0.121 0.115

0.082*** 

Dθ + η + (1 − δ)λ1t i it

λU 0t

−0.0025

−3.333

0.001*

λU 1t

0.0082

10.193

0.000*

D λ0t

0.0016

1.979

D λ1t

−0.0074

−8.895

0.048** 0.000*

Notes *, **, *** indicates significance at the 1%, 5% and 10% respectively Source own study

0.139 0.135

Conventional and Downside Betas …

63

results of conditional relationships give an ambiguous conclusion about co-skewness as a measure of risk in the asset pricing.

5 Conclusions The paper presents research verifying the pricing of sub-sector indices on the Warsaw Stock Exchange in the context of the CAPM model. In this study, the unconditional risk-return relations are estimated using conventional and downside measures and conditional relations in terms of market condition. This research provides some important findings. The unconditional relations reveal that the downside beta and downside co-moments are a more appropriate risk measures than their classical counterparts. The downside risk premiums are significantly positive which means, that the downside risk is priced in Polish stock market. While the market condition is incorporated as the conditioning variable the classical risk factors acquire significance. Beta coefficient and co-kurtosis generate positive premiums in the up market and negative in the down market. Using the subsector indices returns the significance of co-skewness pricing is found only in the down market. It would be desirable to examining of robustness risk-return relationship conditional on market conditional volatility regimes. Conditional relationships confirm the validity of assets pricing in accordance with the CAPM postulates, which is often rejected in many tests of this theory. It follows that measures of systematic risk such as the beta coefficient and higher order comoments in conventional and downside approach are appropriate risk factors and may indicate to including these measures in the asset allocation decision in portfolio investment.

References Alles L, Murray L (2013) Rewards for downside risk in Asian markets. J Bank Finance 37(7):2501. https://doi.org/10.1016/j.jbankfin.2013.02.006 Ang A, Chen J, Xing Y (2006) Downside risk. Rev Financ Stud 19(4):1191–1239. https://doi.org/ 10.1093/rfs/hhj035 Bawa VS, Lindenberg EB (1977) Capital market equilibrium in a mean-lower partial moment framework. J Financ Econ 5(2):189–200. https://doi.org/10.1016/0304-405X(77)90017-4 Bilgin R, Basti E (2014) Further evidence on the validity of CAPM: the Istambul stock exchange application. Inz Ekon-Eng Econ 25(1):5–12. https://doi.org/10.5755/j01.ee.25.1.1847 Chen D, Chen C, Chen J (2009) Downside risk measures and equity returns in the NYSE. Appl Econ 41(8):1055–1070. https://doi.org/10.1080/00036840601019075 Chhapra IU, Kashif M (2019) Higher Co-moments and downside beta in asset pricing. Asian Acad Manag J Account Financ 15(1):129–155. https://doi.org/10.21315/aamjaf2019.15.1.6 Duc THL, Nguyen SP (2018) Higher co-moments and asset pricing on emerging stock markets by quantile regression approach. Bus Econ Horiz 141:132–142. https://doi.org/10.15208/beh. 2018.11

64

L. Markowsk et al.

Estrada J (2002) Systematic risk in emerging markets: the D-CAPM. Emerg Mark Rev 3(4):365– 379. https://doi.org/10.1016/S1566-141(02)00042-0 Estrada J (2007) Mean-semivariance behaviour: downside risk and capital asset pricing. Int Rev Econ Financ 16(2):169–185. https://doi.org/10.1016/j.iref.2005.03.003 Fama E, MacBeth J (1973) Risk, return and equilibrium: empirical tests. J Polit Econ 81(3):607–636. https://doi.org/10.1086/260061 Galagedera Don UA (2009) Economic significance of downside risk in developed and emerging markets. Appl Econ Lett 16(16):1627–1632. https://doi.org/10.1080/13504850701604060 Galagedera Don UA, Henry D, Silvapulle P (2003) Empirical evidence on the conditional relation between higher-order systematic co-moments and security returns. Q J Bus Econ 42(1/2):121– 137. Retrieved from https://www.jstor.org/stable/40473367 Harlow WV, Rao RKS (1989) Asset pricing in a generalized mean-lower partial moment framework: theory and evidence. J Financ Quant Anal 24(3):285–311. https://doi.org/10.2307/2330813 Hogan W, Warren J (1974) Toward the development of an equilibrium capital-market model based on semivariance. J Financ Quant Anal 9(1):1–11. https://doi.org/10.2307/2329964 Jagannathan R, Wang Z (1996) The conditional CAPM and the cross-section of expected returns. J Financ 51(1):3–53. https://doi.org/10.1111/j.1540-6261.1996.tb05201.x Lee CL, Robinson J, Reed R (2008) Downside beta and the cross-sectional determinants of listed property trust returns. J R Estate Portf Manag 14(1):49–62. Retrieved from https://www.jstor. org/stable/24883130 Markowski L (2019) Wycena aktywów kapitałowych w klasycznym i dolnostronnym podej´sciu do ryzyka [Capital assets pricing in classical and downside risk approach] Wiadomo´sci Statystyczne. Pol Stat 64(11):58–75 Mora-Valencia A, Perote J, Arias JET (2017) The return performance of cubic market model: an application to emerging markets. Emerg Mark Financ and Trade 53(10):2233–2241. https://doi. org/10.1080/1540496X.2016.1251902 Neslihanoglu S, Sogiakas V, McColl JH, Lee D (2017) Nonlinearities in the CAPM: evidence from developed and emerging markets. J Forecast 36(8):867–897. https://doi.org/10.1002/for.2389 Nurjannah, Galagedera Don UA, Brooks R (2012) Conditional relation between systematic risk and returns in the conventional and downside frameworks: evidence from the indonesian market. J Emerg Mark Financ 11(3):271–300. https://doi.org/10.1177/0972652712466498 Pettengill GN, Sundaram S, Mathur I (1995) The conditional relation between beta and returns. J Financ Quant Anal 30(1):101–116. https://doi.org/10.2307/2331255 Post T, van Vliet PV (2006) Downside risk and asset pricing. J Bank Finance 30(3):823–849. https:// doi.org/10.1016/j.jbankfin.2005.06.005 Rutkowska-Ziarko A, Markowski L, Pyke C (2019) Accounting beta in the extended version of CAPM. Contemporary Trends and Challenges in Finance. In: Jajuga K, Locarek-Junge H, Orłowski LT, Staehr K (eds) Proceedings from the 4th Wroclaw international conference in Finance, Springer, Cham, pp 45–53 Tang GYN, Shum WC (2003) The conditional relationship between beta and returns: recent evidence from international stock markets. Int Bus Rev 12(1):109–126. https://doi.org/10.1016/S09695931(02)00090-2 Truong TT, Kim J (2018) Sustainability managed against downside risk and the cost of equity: evidence in Korea. Sustainability 10(11):3669. https://doi.org/10.3390/su10113969 Trzpiot G, Kr˛ez˙ ołek D (2006) Statystyczna weryfikacja modelu CAPM na przykładzie polskiego rynku kapitałowego [Statistical verification of the CAPM on the example of the Polish capital market]. Zeszyty Naukowe SGGW w Warszawie Ekonomika i Organizacja Gospodarki ˙ Zywno´ sciowej 60: 341–351. Retrieved from http://eiogz.wne.sggw.pl Tsai H, Chen M, Yang C (2014) A time-varying perspective on the CAPM and downside betas. Int Rev Econ Financ 29:440–454. https://doi.org/10.1016/j.iref.2013.07.006

The Accuracy of Trade Classification Rules for the Selected CEE Stock Exchanges Sabina Nowak

Abstract We assess the accuracy of the popular trade classification rules—the tick (T), the reverse tick (RT), the quote (Q), the Lee and Ready (LR) and the Ellis, Michaely and O’Hara (EMO) rules—for the selected Central and Eastern Europe stock exchanges: the Warsaw Stock Exchange (WSE), the Prague Stock Exchange (PSE) and the Budapest Stock Exchange (BSE). We employ the transaction data on the most liquid stocks included in the large cap indices, namely WIG20, PX and BUX, from 9th of January till 14th of March, 2019, and discover the Q rule to be the most successful. It correctly classifies 100% of trades initiated both by the buyers and the sellers in the case of the BSE and 84.35% (82.78%) in the case of the WSE (the PSE). The results obtained for the LR rule are similarly good (96.32% for the BSE, 84.34% for the WSE and 82.86% for the PSE). The third one is the EMO rule with the rate of success of 83.69% for the BSE, 80.40% for the WSE and 67.86% for the PSE. The T and the RT rules are characterized by a considerably low level of accuracy (ranging from 19.91 to 40.65% for the T rule and from 13.04 to 38.37% for the RT rule). The modifications of the T and the RT rules that take into account the preceding and the following transaction price changes allow to obtain the distinctively higher estimates of accuracy statistics.

1 Introduction Trade classification rules (hereafter referred to as TCR) are intended to indicate the party to a trade who initiates a transaction. It may by either a buyer or a seller. Such indication made directly from the data is nowadays in mostly cases inaccessible, since the majority of public databases including transaction data do not contain information of trade initiators and trade direction. The information which party to a trade is a trade initiator is indispensable to specify the trade indicator models used to investigate the intraday price formation S. Nowak (B) Faculty of Management, Department of Econometrics, University of Gda´nsk, Sopot, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_6

65

66

S. Nowak

(Glosten and Harris 1988; Huang and Stoll 1997; Madhavan 1992; McGroarty et al. 2007; Hagströmer et al. 2016). Moreover, the identification of party to a trade which is responsible for initiating a particular transaction is advantageous to clarify many important issues related to the market microstructure. First of all, it may be used to ascertain the information content of trades. Second, it can help to figure out the magnitude of the order imbalance as well as the proportion of the inventory accumulation made by the liquidity suppliers. Third, it helps to assess the price impact of large in volume transactions as well as the magnitude of effective spread (Ellis et al. 2000). In the literature we can find five commonly used TCR: the tick (T), the reverse tick (RT), the quote (Q), the Lee and Ready (1991) (LR) and the Ellis et al. (2000) (EMO) rules. Their descriptions are presented in Table 1. The crucial point, however, is defining when a particular transaction is being initiated by a byer (seller). We will consequently follow an approach employed in the paper of Miłob˛edzki and Nowak (2018) and classify a particular trade as initiated by a buyer (seller) if its resulting price is equal to the best ask (bid) or higher (lower) than that. It is pertinent to note that another attitudes are to be found in the literature: i.e., an investor is assessed to be a trade initiator if he (she) places either a market order or places his (her) order chronologically last. A trade initiator as also understood as the last party to agree to the trade or the party whose decision causes the trade to occur (Lee and Radhakrishna 2000; Odders-White 2000; Chakrabarty et al. 2012). Table 1 Trade classification rules descriptions Rule

Abbreviation

Definition

Tick

T

Trade is recognized to be initiated by a buyer (seller) if its price is higher (lower) than the nearest different price of a previous transaction

Reverse tick

RT

Trade is recognized to be initiated by a buyer (seller) if its price is higher (lower) than the nearest different price of a following transaction

Quote

Q

Trade is recognized to be initiated by a buyer (seller) if its price is higher (lower) than the mid-point pricea . Trades executed at the mid-point price are not classified

Lee and Ready

LR

Two step procedure: (1) the Q rule applies (2) for trades that are not classified the T rule applies

Ellis, Michaely and O’Hara

EMO

Two step procedure: (1) trade is assessed to be initiated by a buyer (seller) if its price is equal to the best ask (the best bid) (2) if the price is in between the quotes, the T rule applies

Explanations: a Mid-point price is equal to the arithmetic mean of the bid and ask prices

The Accuracy of Trade Classification Rules for the Selected …

67

2 Literature Review The accuracy of the TCR differs depending upon the financial market maturity, the level of trade price, the transaction size and the amount of time that passed by from the previous trade. The findings of the literature review on the TCR accuracy divided for empirical works devoted either to the US financial markets or the other international markets were presented in the paper of Miłob˛edzki and Nowak (2018). Here we would like to mention only the general findings. First, majority of results obtained for the US financial markets (including the NYSE, the NASDAQ and the AMEX) exhibit a considerably high level of success rates of the T, the Q, the LR as well as the EMO rules. Their accuracy varies between 41.3 and 98.9% for the T rule, 37.5 and 84% for the Q rule, 39.1 and 93.4% for the LR rule and, finally, between 72.8 and 81.9% for the EMO rule (compare Miłob˛edzki and Nowak (2018), summary in Table 2). Second, the LR and the EMO rules tend to outperform the T and the Q rules on the US financial markets. However, the differences between the accuracy of all the TCR are rather minor (Ellis et al. 2000; Finucane 2000; Lee and Radhakrishna 2000; Odders-White 2000; Chakrabarty et al. Table 2 Stocks in the research sample

Stock

Index

Industry

Alior

WIG20

Corporate banks

Energa

WIG20

Electric power

KGHM

WIG20

Copper mining

Orange

WIG20

Telecommunication resellers

Orlen

WIG20

Petroleum refining

Central European Media Enterprises Ltd

PX

Broadcasting

Kofola CeskoSlovensko

PX

Non-alcoholic beverages

Komercni Banka

PX

Banks

O2 Czech Republic

PX

Integrated telecommunications services

PFNonwovens

PX

Synthetic fabrics

Magyar Telekom Tavkozlesi

BUX

Integrated telecommunications services

MOL

BUX

Oil and gas refining and marketing

Opus Global

BUX

Industrial conglomerates

OTP Bank

BUX

Banks

Richter Gedeon Vegyeszeti Gyar

BUX

Generic pharmaceuticals

68

S. Nowak

2007, 2012, 2015; Rosenthal 2012). Third, similar findings are stemming from the empirical analyses of the other developed markets stock exchanges, including the Deutsche Börse (Poppe ¨ et al. 2016), the Frankfurt Stock Exchange (Theissen 2001), the Australian Stock Exchange (Aitken and Frino 1996), to mention just a few. Nevertheless, still little is known on the issue of TCR accuracy for emerging markets except the Taiwan Stock Exchange (Lu and Wei 2009), the Borsa Istanbul (Aktas and Kryzanowski 2014), the B3—Brasil Stock Exchange (Perlin et al. 2014) and the Warsaw Stock Exchange (Olbry´s and Mursztyn 2015; Miłob˛edzki and Nowak 2018). Lu and Wei (2009) obtained a noticeably high level of the TCR rates of success ranging from 74.2% (the T rule) to 96.5% (the LR rule). Perlin et al. (2014) achieved the comparable results in reference to the T rule (72%). Aktas and Kryzanowski (2014) outperformed the results of Lu and Wei (2009) receiving the lowest accuracy rate equal to 86.8% (for the EMO rule) and the highest one equal to 96.4% (the LR rule). Results obtained by Miłob˛edzki and Nowak (2018) were even more outstanding—the estimates of accuracy ratios ranged from 95.3% for the EMO rule to 100% for both the Q and the LR rules. Nevertheless, their outcomes for the T and the RT rules were significantly lower (25.6% for the T and 16.7% for the RT, respectively). However, in the literature there is still a lack of similar surveys devoted to the European developing countries. Hence, in this paper we would like to extend the research conducted for European emerging markets, employing the data on 15 stocks listed on the selected Central and Eastern Europe (CEE) stock exchanges: the WSE, the PSE and the BSE.

3 Data We examined 15 stocks included in the blue chip indices listed on the three CEE stock exchanges: 5 stocks from WIG20 index listed on the WSE, 5 stocks from PX index listed on the PSE, and 5 BUX index constituents from the BSE. The stocks were selected on purpose, each stock within particular index was to be a different industry representative. The research sample is described in Table 2. We investigated the transaction data in the period from 9th of January till 14th of March 2019 collected from the Refinitiv (previously Thomson Reuters) Eikon 4 database, application Time & Sales.1 The transaction data are stamped to the nearest millisecond. Each transaction record in the database includes information on the trade price and its volume, together with the best bid and best ask, the size of bid and ask and the trade flag. The latter information enables to recognize the transaction type. However, the trade flags turned out to be different for each stock exchange in the sample. For the WSE, 8 different transaction flags could be recognized, namely: 1 The data are collected from the Refinitiv Eikon 4 database under the partnership agreement between

the University of Gdansk and the Refinitiv company.

The Accuracy of Trade Classification Rules for the Selected …

69

open price, auction trade, regular order (transactions with those flags accounted for 0.03% of all trades in the sample), normal trade, auction trade, regular order (0.03%), auction trade, regular order (2.65%), normal trade, normal price, regular order (0.02%), normal price, regular order (97.14%), cross market trade, cross trade, cross order (0.03%), trading at last, regular order (0.09%) and normal trade, trading at last, regular order (less than 0.01%). I decided to considered only the normal price, regular order transactions which limited the data set to 689,942 trades from the initial of 710,229. In the case of the PSE, only 2 flags were indicated: auction trade (1.33%) and correction of last trade (0.06%). The rest of transactions (98.61%) were not labelled. Eventually, we took into account only those N/A transactions. The size of the final data set was equal to 25,331 trades; we resigned of 356 records. For the BSE transactions only two flags were assigned: auction trade (0.55%) and normal price (99.45%). In this case we analysed only those latter records (87,039 from the initial number equal to 87,517). Eventually, we took into consideration 802,312 trades, including 689,942 from the WSE, 25,331 from the PSE and 87,039 from the BSE.

4 Empirical Results In the beginning, following the definition described in the Introduction, we identified the trades initiated by the buyer (buys), seller (sells) and trades executed in between the quotes (including those closer to best ask/best bid and not identified, i.e. executed at the mid-point price). The results of such identification are given in Table 3. Table 3 Trades included in the final dataset by their type Trade type

WSE

PSE

No. of trades

Fraction

BSE

No. of trades

Fraction

No. of trades

Fraction

Buy—above best ask

354

0.05

110

0.43

576

0.66

Buy—at best ask

400,574

58.06

11,322

44.70

37,417

42.99

Inside—closer to best ask

6191

0.90

1596

6.30

3701

4.25

838

0.12

1324

5.23

7497

8.61

Inside—closer to best bid

6382

0.93

1614

6.37

4625

5.31

Sell—at best bid

273,741

39.68

7146

28.25

322,270

37.08

Sell—above best bid

1862

0.27

2209

8.72

953

1.09

689,942

100.00

25,331

100.00

87,039

100.00

Inside—not identified

All trades

70

S. Nowak

What is interesting is the discrepancy between the WSE and two other stock exchanges. Worth of noting is that only 1.94% of trades from the WSE were executed in between the quotes, whereas such percentage was considerably higher and equal to 17.90% and 18.18% in the case of the PSE and the BSE, respectively. The results obtained with reference to the WSE are similar to those achieved by Miłob˛edzki and Nowak (2018) on the basis of the sample of 20 stocks from WIG20 index in the period of April 24–5 October 2017. In Table 4 we summarize the results of the comparison between the classification success rates for the selected TCR. The obtained findings show that, in the case of the WSE, the Q rule performs the best correctly identifying 84.45% of trades. The second one is the LR rule which correctly identifies 83.34% of trades and misclassifies only 7.88% of the inside the quotes trades, and 16.88% of the trades executed at best ask. The third one is the EMO misclassifying as much as 19.60% of all trades. However, the T and the RT rules work very poorly misclassifying 80.09% and 86.96% of the trades, respectively. Those results are comparable to the previous findings from the WSE, where the estimates of accuracy statistics were equal to 100% both for the Q and the LR rule and 95.34% for the EMO rule (Miłob˛edzki and Nowak 2018). For the PSE, the LR rule outperforms the others and identifies correctly 82.86% of trades. The second best is the Q rule with almost the same success rate (82.78%). The obtained results put the EMO rule in third place—it misclassifies almost one third of all trades (32.14%). The T and the RT rules perform weakly, but more successfully than on the WSE: they misclassify 60.53% and 62.93% of trades, respectively. In the case of the BSE, the LR rule performs the best correctly identifying 100% of trades. The second place belongs to the LR rule which correctly identifies 96.32% of all trades and misclassifies 20.24% of those inside the quotes. The third one is the EMO misclassifying as much as 16.31% of all trades. The T and the RT rules perform better than on the WSE and the PSE, but their success rate is still unsatisfying—they misclassify 59.35% and 61.63% of trades, respectively. Having obtained such a poor results for the T and the RT rules for all stock exchanges, we decided to follow the concept introduced in the paper of Miłob˛edzki and Nowak (2018) that involves modifying those rules by accounting for either the preceding or the following transaction price changes. We investigated performance of 10 modified TCR with sequence lengths from one up to five (T1–T5 and RT1–RT5). The summary of their accuracy rates is described in Table 5. In the case of modification of the T and the RT rules their accuracy increases with the growing number of the preceding (following) price changes being considered— results obtained for the T5 (RT5) rules are higher than those reached for the T1–T4 and the RT1–RT4. The highest success rates are equal to 56.19% (T5) and 34.54% (RT5) in the case of the WSE listed stocks, 65.01% (T5) and 60.24% (RT5) for the PSE and, eventually, 65.92% (T5) and 61.63% (RT5) for the BSE. For all trades the estimates of accuracy rates for the modified T rule are higher in comparison to those obtained for the modified RT rule.

The Accuracy of Trade Classification Rules for the Selected …

71

Table 4 The accuracy of five selected trade classification rules SE

Trade type

WSE

Buy—above ba

T

BSE

Q 20.90

LR

100.00

EMO

100.00

8.47

Buy—at ba

16.71

11.04

73.12

73.12

71.35

Inside—closer to ba

10.56

11.08

100.00

100.00

10.56

Inside—not identified

92.12

68.14

100.00

92.12

92.12

Inside—closer to bb

12.25

12.17

100.00

100.00

12.25

Sell—at bb

24.90

15.92

100.00

100.00

97.37

3.60

2.79

84.26

84.26

5.32

Buy

16.70

11.05

73.14

73.14

71.29

Inside

16.46

15.17

100.00

99.51

16.46

Sell

24.75

15.83

99.89

99.89

96.75

Sell—above bb

PSE

RT 8.47

All trades at the WSE

19.91

13.04

84.35

84.34

80.40

Buy—above ba

15.45

98.18

100.00

100.00

15.45

Buy—at ba

33.40

31.50

69.78

75.23

65.00

Inside—closer to ba

35.96

35.46

77.01

77.01

35.96

Inside—not identified

51.96

27.87

94.71

48.49

64.50

Inside—closer to bb

41.26

32.96

68.90

69.83

30.98

Sell—at bb

42.61

40.90

99.97

99.97

93.14

Sell—above bb

55.27

59.85

100.00

100.00

55.27

Buy

33.23

32.14

70.07

75.47

64.52

Inside

42.52

32.36

79.29

66.12

42.52

Sell

45.60

45.37

99.98

99.98

84.21

All trades at the PSE

39.47

37.07

82.78

82.86

67.86

Buy—above ba

9.20

94.44

100.00

100.00

9.20

Buy—at ba

36.53

36.89

100.00

100.00

92.83

Inside—closer to ba

42.77

33.21

100.00

100.00

42.77

Inside—not identified

57.28

30.97

100.00

57.28

57.28

Inside—closer to bb

48.02

29.10

100.00

100.00

48.02

Sell—at bb

41.35

41.90

100.00

100.00

92.15

Sell—above bb

23.19

66.11

100.00

100.00

23.19

Buy

36.12

37.76

100.00

100.00

91.56

Inside

51.18

30.95

100.00

79.76

51.18

Sell

40.83

42.59

100.00

100.00

90.17

All trades at the BSE

40.65

38.37

100.00

96.32

83.69

Explanations: ba—best ask, bb—best bid

PSE

All trades at the WSE

59.77

Sell—above bb

Sell

61.43

Sell—at bb

47.51

59.25

Inside—closer to bb

45.57

53.53

Inside—not identified

Inside

30.14

Inside—closer to ba

Buy

47.77

50.31

Buy—at ba

20

40.97

32.94

Sell

Buy—above ba

25.52

Inside

Sell—at bb

5.91

41.21

Inside—closer to bb

27.66

23.03

Inside—not identified

Buy

85.08

Inside—closer to ba

Sell—above bb

27.68

20.03

Buy—at ba

11.86

Buy—above ba

WSE

T1

Trade type

SE

T2

Table 5 The accuracy of the modified T and RT rules

66.29

45.79

55.55

63.42

67.17

58.74

17.6

56.08

55.88

20.91

41.88

51.93

33.2

35.27

7.89

52.23

32.36

78.52

27.94

35.28

14.41

T3

69.7

45.66

60.3

63.88

71.49

60.78

11.1

59.02

60.67

21.82

48.23

59.6

39.73

40.69

9.77

59.94

40.08

72.43

34.94

40.71

16.38

T4

71.82

45.94

63.43

64.19

74.18

62.33

8.01

60.84

63.83

21.82

52.84

65.05

46.25

44.66

11.44

65.41

47.92

65.99

41.87

44.68

17.8

T5

73.38

46.01

65.69

64.33

76.17

63.14

6.27

61.65

66.12

21.82

56.19

68.92

52.7

47.56

13.21

69.3

55.78

54.42

49.28

47.58

18.36

58.71

31.72

45.71

64.92

56.79

38.17

14.73

39.29

45.19

100

20.89

25.49

20.36

17.75

4.08

25.64

18.52

48.69

18.43

17.73

33.33

RT1

64.93

31.63

53.31

66.82

64.35

40.46

9.37

41.17

52.85

100

26.09

31.86

23.48

22.22

5.05

32.04

22.39

36.28

22.87

22.2

41.53

RT2

68

31.72

57.67

67.18

68.25

41.51

6.72

42.54

57.26

100

29.77

36.37

25.51

25.38

5.69

36.58

25.02

27.8

25.7

25.36

46.33

RT3 50

69.91

31.67

60.56

67.32

70.71

42.32

5.14

42.92

60.17

100

32.5

39.69

26.87

27.75

6.28

39.92

27

22.67

27.31

27.73

RT4

(continued)

71.22

31.67

62.58

67.45

72.39

42.75

4.23

43.23

62.22

100

34.54

42.15

28.01

29.54

6.87

42.38

28.64

19.33

28.54

29.52

52.82

RT5

72 S. Nowak

56.55

35.64

61.17

57.89

27.7

51.58

47.99

57.03

Inside—closer to ba

Inside—not identified

Inside—closer to bb

Sell—at bb

Sell—above bb

Buy

Inside

Sell 53

52.13

All trades at the BSE

15.45

51.69

Buy—at ba

T1

Buy—above ba

All trades at the PSE

Trade type

Explanations: compare Table 4

BSE

SE

Table 5 (continued) T2

59.06

64.95

45.33

59.62

29.91

65.98

66.77

23.88

61.98

60.26

18.06

57.77

T3

62.48

69.28

43.66

64.36

31.16

70.41

69.75

16.97

65.14

65.05

20.14

61.15

T4

64.58

71.88

42.49

67.39

31.69

73.06

71.22

12.71

66.9

68.12

20.49

63.4

T5

65.92

73.49

41.63

69.43

32.21

74.7

72.22

9.88

67.71

70.17

21.18

65.01

49.92

57.82

27.69

52.28

68.84

57.49

33.45

19.31

37.45

51.59

96.88

48.01

RT1

55.56

65.06

26.13

59.5

70.72

64.89

35.39

13.81

39.5

58.91

97.74

53.72

RT2

58.6

68.87

25.06

63.58

71.25

68.8

36.43

10.38

40.58

63.05

98.26

56.84

RT3

60.42

71.07

24.22

66.2

71.56

71.05

37.06

8.04

40.96

65.7

98.44

58.84

RT4

61.63

72.46

23.81

67.91

71.88

72.48

37.66

6.62

41.31

67.44

98.44

60.24

RT5

The Accuracy of Trade Classification Rules for the Selected … 73

74

S. Nowak

5 Conclusions We checked for the accuracy of five trade classification rules (T, RT, Q, LR, EMO) using the tick data on 15 large cap stocks listed at the selected Central Eastern Europe stock exchanges: the WSE, the PSE and the BSE in two-month period January–March 2019. We investigated 15 stocks comprising 5 stocks from each selected CEE stock exchanges, included in WIG20, PX and BUX indices, and representing different industries. The Q rule turned out to be most accurate: it correctly classifies 100% of trades initiated both by the buyers and the sellers in the case of the BSE and 84.35% (82.78%) in the case of the WSE (the PSE). The second best is the LR rule with similar rates of success: 96.32% for the BSE, 84.34% for the WSE and 82.86% for the PSE. The third one is the EMO rule with the accuracy level of 83.69% for the BSE, 80.40% for the WSE and only 67.86% for the PSE. The T and the RT rules exhibit the noticeably lower levels of accuracy (19.91–40.65% for the T rule, 13.04–38.37% for the RT rule). The proposed modifications of the T and the RT rules ensure a more convincing performance. The accuracy rates of the modified T (RT) rules exceed 65% (60%) in the case of the PSE and the BSE and are substantially lower, equal to 56% (34.4%) for the WSE stocks.

References Aitken M, Frino A (1996) The accuracy of the tick test: evidence form the Australian stock exchange. J Bank Finance 20:1715–1729 Aktas OU, Kryzanowski L (2014) Trade classification accuracy for the BIST. J Int Financ Markets Inst Money 33:259–282 Chakrabarty B, Li B, Nguyen V, Van Ness RA (2007) Trade classification algorithms for electronic communications network trades. J Bank Finance 31(12):3806–3821 Chakrabarty B, Moulton PC, Shkilko A (2012) Short sale, long sale, and the Lee-ready trade classification algorithm revisited. J Financ Markets 15(4):467–491 Chakrabarty B, Pascual R, Shkilko A (2015) Evaluating trade classification algorithms: bulk volume classification versus the tick rule and the Lee-ready algorithm. J Financ Markets 25:52–79 Ellis K, Michaely R, O’Hara M (2000) The accuracy of trade classification rules: evidence from Nasdaq. J Financ Quant Anal 35(4):529–551 Finucane TJ (2000) A direct test of methods for inferring trade direction from intra-day data. J Financ Quant Anal 35(4):553–576 Glosten LR, Harris LE (1988) Estimating the components of the bid-ask spread. J Financ Econ 21:123–142 Hagströmer B, Henricksson R, Nordén LL (2016) Components of the bid-ask spread and variance: a unified approach. J Futures Markets 36(6):545–563 Huang RD, Stoll HR (1997) The components of the bid-ask spread: a general Approach. Rev Financ Stud 10:995–1034 Lee CMC, Radhakrishna B (2000) Inferring investor behavior: evidence from TORQ data. J Financ Markets 3:83–111 Lee CMC, Ready MJ (1991) Inferring trade direction from intraday data. J Finance 46(2):733–746

The Accuracy of Trade Classification Rules for the Selected …

75

Lu YC, Wei YC (2009) Classification of trade direction for an equity market with price limit and order match: evidence from the Taiwan stock market. Invest Manage Financ Innov 6(3):135–147 Madhavan A (1992) Trading mechanisms in securities markets. J Finance 47(2):607–641 McGroarty F, Gwilym O, Thomas S (2007) The components of electronic inter-dealer spot FX bid-ask spreads. J Bus Financ Account 34(9, 10):1635–1650 Miłob˛edzki P, Nowak S (2018) The accuracy of trade classification rules for the Warsaw stock exchange. In: XIIth professor Aleksander Zelia´s international conference on modelling and forecasting of socio-economic phenomena. Conference proceedings. Fundacja Uniwersytetu Ekonomicznego w Krakowie, Kraków, pp 316–325 Odders-White ER (2000) On the occurrence and consequences of inaccurate trade classification. J Financ Markets 3:259–286 Olbry´s J, Mursztyn M (2015) Comparison of selected trade classification algorithms on the Warsaw stock exchange. Adv Comput Sci Res 12:37–52 Perlin M, Brooks C, Dufour A (2014) On the performance of the tick test. Quart Rev Econ Financ 54:42–50 Poppe ¨ T, Moos S, Schiereck D (2016) The sensitivity of VPIN to the choice of trade classification algorithm. J Bank Financ 73:165–181 Rosenthal DWR (2012) Modeling trade direction. J Financ Econ 10(2):390–415 Theissen E (2001) A test of the accuracy of the Lee/Ready trade classification algorithm. J Int Financ Markets Inst Money 11(2):147–165

Profitability Ratios in Risk Analysis Anna Rutkowska-Ziarko

Abstract The aim of the study was to examine the correlation between the accounting profitability of the company and its rate of return on the capital market. In addition, betas and accounting betas were compared. The correlation between total variability and semi-variability of profitability ratios and rates of return was also analysed. The risk of a company was considered in variance and downside approaches. For calculating the downside risk, both the Bawa and Lindenberg formula and the Harlow and Rao formula were used. A positive correlation between the average value of the quarterly profitability ratios (ROA and ROE) and the average quarterly rates of return on the Warsaw Stock Exchange was identified. Similarly, companies with higher volatility and semi-volatility of the profitability ratios were simultaneously characterized by larger fluctuations in rates of return on the stock market. The correlations between market betas and accounting betas were statistically significant only in a downside approach. Accounting profitability had a greater effect on the rates on return and the risk for large and medium companies compared to small ones.

1 Introduction One of the fundamental objectives of a company is profit generation. For listed companies, the financial result is one of the factors affecting the market value of company stocks. An analysis of company profitability is thus very important for both stock market investors and managerial staff. Every business is exposed to risk. This risk could be examined based on its effects, which are manifested by variability in the value of financial indices or by variability in listed company stock prices. Another approach involves an analysis of the effect of a given risk factor on the value of an analysed variable.

A. Rutkowska-Ziarko (B) University of Warmia and Mazury in Olsztyn, Olsztyn, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_7

77

78

A. Rutkowska-Ziarko

The importance of risk management research was highlighted in a recent special issue of the British Accounting Review (Woods et al. 2017) on accounting and risk. An editorial to the journal (Woods et al. 2017) stated that risk, the measurement of risk and the theory of risk, is an under-developed area of research in management and accounting academic journals. In addition, this article gave a wide review of literature and studies on that topic. In the article, measures of systematic and total risk were considered. The emphasis is put on accounting beta and downside accounting beta. Accounting beta is a systematic risk measure, based on accounting information as proposed by Hill and Stone (1980). It reflects how the changes in the profitability in the whole market or sector appeal to the profitability of the company. Toms (2012, 2014) proposed the accounting-based risk (ABR) model as a contrast to CAPM. The ABR model focuses on revenue and cost behaviour and its impact on the accounting rate of return. Nekrasov and Shroff (2009) proposed using a one-factor model with accounting beta when it was not possible to calculate market risk measures. The accounting beta for ROE was used as one of the risk factors for the US capital market. Annual accounting data were used and they emphasize the problem with a short time-series of accounting data that produce some problems with estimation. They emphasized that using quarterly data instead of annual data can improve the precision of estimation. It is important for investors to have access to trustworthy and frequent information on a company’s performance. In Poland, all public companies are obliged to provide quarterly public financial reports. The article is focused on the risk in downside approach. There are many supporters of downside risk, so it is difficult to cite all works in this area. But the work of Konchitchki et al. (2016) should be noted here as being very close to the concept of downside accounting beta. The authors combined the downside approach with calculating risk based on accounting information. They noted that the downside volatility of accounting figures is more important in risk analysis than the upside states. The aim of the study was to examine the correlation between accounting profitability of the company and its rate of return on the capital market. In addition, betas and accounting betas were compared and the correlation between total variability and semi-variability of profitability ratios and rates of return was analysed.

2 Downside Beats, Downside Accounting Betas and Semi-variance In the classical Capital Asset Pricing Model the systematic risk is measured by the beta coefficient (βi ), which is usually calculated as follows (Barucci and Fontana 2017, p. 220):

Profitability Ratios in Risk Analysis

79

βi =

C O Vi M , 2 SM

(1)

where C O Vi M is covariance of the rate of return for stock i and market portfolio 2 is variance of market portfolio rates of return. rates of return, S M The term “accounting beta” originated in the 1970s (e.g. Beaver et al. 1970). It was understood as using any accounting information in risk analysis. In this work and previous studies, the approach proposed by Hill and Stone (1980) was used. They restrict the name “accounting beta” to the regression coefficient where a profitability ratio for a given company is a dependent variable and the broad market index for the ratio is an independent variable. According to Hill and Stone (1980), every beta calculated on the market price (βi , βiB L , βiH R ) may be called market beta. In the same way, beta calculated on accounting information can be described as accounting beta. In this article, accounting beta is analysed based on return on assets and return on equity (β(R O A), β(R O E), βiB L (R O A), βiB L (R O E), βiH R (R O A), βiH R (R O E)). The accounting beta is calculated in a similar way to the market beta. Accounting Beta coefficient for return on assets (βi (R O A)) is calculated as follows: βi (R O A) =

C O Vi M (R O A) , 2 SM (R O A)

(2)

where C O Vi M (R O A) is covariance of the profitability ratio of company i and mar2 ket portfolio ratios (market indices of profitability ratios), S M (R O A)—variance of market profitability ratios. Downside risk measures considered only some deviation on the left side of the distribution. In the article, the downside risk measures based on semi-variance and co-semi-variance of returns are used. Market betas are calculated in concordance with the Bawa and Lindenberg (1977) formula and the Harlow and Rao (1989) formula. The main difference between these two formulae is the target level of the rate of return required by investors. Bawa and Lindenberg took the risk-free rate as the target point, whereas Harlow and Rao used average returns of the market portfolio. In the article, the formulas for downside beta are given in the form written by Galagadera (2007). The downside beta proposed by Bawa and Lindenberg (1977) can be written as follows: βiB L =

E{(Ri − R f ) · min[(R M − R f ), 0]} , E{min[(R M − R f ), 0]2 }

(3)

where R f is the risk-free rate, Ri is the rate of return for stock i, R M is market portfolio rates of return. Using the proper covariance between asset i and market portfolio, the downside beta of the Harlow and Rao formula is given by:

80

A. Rutkowska-Ziarko

βiH R =

E{(Ri − μi ) · min[(R M − μ M ), 0]} , E{min[(R M − μ M ), 0]2 }

(4)

where μi , μ M are the average returns of security i and market portfolio, respectively. Relationships between different downside beta coefficients are presented in the work of, among others, Markowski (2018). The concept of downside beta can be combined with the concept of accounting beta. The result of this is the measure of downside risk based on accounting information called downside accounting beta (Rutkowska-Ziarko and Pyke 2017). When the Bawa and Lindenberg formula is adopted to calculating downside accounting beta, the risk-free rate should be replaced by another target point suitable for a given accounting ratio. It can be the long-term average of the profitability ratio in the sector or market (Rutkowska-Ziarko and Pyke 2017). Additionally, the market index for a given profitability ratio is needed. The basic solution is using the mean of a ratio (Hill and Stone 1980). This mean can be weighted in several different ways. Beaver et al. (1970) proposed using the arithmetic mean or mean scaled by market value. A possible solution is weighting the average value of the ratio in relation to assets, equity or sales. Using the Bawa and Lindenberg formula transformed for downside accounting beta for return on assets (βiB L (R O A)), R f was replaced by the average long-term return of assets for market portfolio (R O A M ): βiB L (R O A) =

E{(R O Ai − R O A M ) · min[(R O A M − R O A M ), 0]} E{min[(R O A M − R O A M ), 0]2 }

.

(5)

where R O Ai and R O A M are return on assets for company i and for the market portfolio. Return on assets for the market portfolio was calculated as follows: R O AM =

N  i=1

M Vi R O Ai ∗  N , i=1 M Vi

(6)

where M Vi is the market value of company i. The average long-term return of assets for market portfolio is taken: R O AM =

T 1  R O A Mt , T t=1

(7)

where R O A Mt is return on assets for market portfolio in time t. In a similar way, the formula of downside beta of Harlow and Rao is adopted to calculate the downside accounting beta βiH R (R O A): βiH R =

E{(R O Ai − R O Ai ) · min[(R O A M − R O A M ), 0]} E{min[(R O A M − R O A M ), 0]2 }

,.

(8)

Profitability Ratios in Risk Analysis

81

where: R O Ai =

T 1  R O Ait , . T t=1

(9)

and R O Ait is return on assets for company i in time t. The downside accounting beta for a given profitability ratios can be calculated in the same way. In this paper, the total risk was measured using symmetric and asymmetric approaches. In the symmetric approach, standard deviation—the classical measure of dispersion was used. In the asymmetric approach semi-deviation was considered. The target point for semi-deviation of rates of return (d S) is the risk-free rate (R f ): dS =



E{min[(Ri − R f ), 0]2 }.

(10)

Semi-deviation of return on assets (d S(R O A)) is calculated as follows:  d S(R O A) =

E{min[(R O Ai − R O A M ), 0]2 }.

(11)

The semi-deviation for a given profitability ratios can be calculated in the same way.

3 Data The data of stocks quoted on the Polish stock market are used in the study. The sample of quarterly returns starting from January 2010 to December 2017 is obtained. In this period, 62 stocks are designated, 10 from the WIG20, 21 from the mWIG40 and 31 from the sWIG80 indices. The WIG20 index is based on the value of a portfolio of 20 of the largest and most liquid companies on the WSE, the mWIG40 index comprises 40 medium-sized companies listed on the WSE and the sWIG80 index comprises 80 smaller companies listed on the WSE. The sample chosen in this way allowed considering the company size using two well-known profitability ratios (ROA and ROE). Quarterly financial reports were used for calculating profitability ratios. The WIG index is used as the market portfolio approximation. The Warsaw Interbank Offer Rate (WIBOR 3M) for three-month investment was used as the risk-free rate. All data are collected from the EMIS database.

82

A. Rutkowska-Ziarko

4 Empirical Results Three samples of data were considered, all companies, large and medium companies (covered by WIG20 and mWIG40 indices) and small companies (covered by sWIG80 index). The critical value of the Pearson coefficient for sample size amount to 62 is 0.2108, 0.2500, 0.3248 for significance level 10%, 5% and 1% level respectively. The critical value of the Pearson coefficient for sample size amounted to 31 is 0.3009, 0.3550, 0.4556 for significance levels of 10%, 5% and 1% respectively. Based on quarterly financial reports, profitability ratios such as return on assets (ROA) and return on equity (ROE) were calculated for every company. For each company, the market betas and accounting betas were determined using two different approaches: the risk measured by standard deviation and downside risk. Downside beta is calculated in two ways, according to the Bawa and Lindenberg (1977) formula and the Harlow and Rao (1989) formula. Risk measures of total variability (semi-variability) based on market prices and accounting profitability ratios were also used. A comparison of risk measures employed in the article is presented in Tables 1 and 2. Table 3 presents the descriptive statistics of analysed profitability and risk measures for all securities. Mean values of beta and downside beta HR coefficients are very similar, taking into account all characteristics and both are higher, on average than the values of downside beta BL. It should be noted that the latter has a wider range and higher dispersion than the former. The sample of securities was characterized by average quarterly ROA and ROE at 0.014 and 0.027, respectively. The profitability of Polish companies is high. In comparing ROA and ROE, it can be seen that equity makes up about 52% of enterprise assets, which is a reasonable level of leverage. The maximum and minimum levels of profitability ratios for firms used in the investigation is realistic, although Floria Table 1 Risk measures based on accounting profitability ratios Variance approach

Downside approach

Measures of sensitivity

Accounting beta for ROA (ROE)

Harlow and Rao downside accounting beta for ROA (ROE) Bawa and Lindenberg downside accounting beta for ROA (ROE)

Total risk

Standard deviation of ROA (ROE)

Semi-deviation of ROA (ROE)

Table 2 Risk measures based on market prices Variance approach

Downside approach

Measures of sensitivity

Beta coefficient

Harlow and Rao downside beta Bawa and Lindenberg downside beta

Total risk

Standard deviation

Semi-deviation

Profitability Ratios in Risk Analysis

83

Table 3 Summary statistics of profitability and risk measures for all stock companies Variable

Mean

Median

Min

Max

Std. Dev.

Skewness

R

0.029

0.029

−0.019

0.096

0.029

0.256

RO A

0.014

0.014

−0.009

0.042

0.010

0.469

RO E

0.028

0.022

−0.030

0.108

0.022

0.793

β

0.878

0.800

−0.011

2.002

0.487

0.409

βHR

0.880

0.878

0.042

1.973

0.492

0.299

β BL

0.779

0.739

−0.611

2.072

0.535

−0.071

β(R O A)

0.291

0.135

−1.733

5.449

1.028

2.375

β(R O E)

0.279

0.075

−1.262

3.891

0.921

1.700

β H R (R O A)

0.261

0.017

−1.850

5.366

1.020

2.346

β H R (R O E)

0.258

0.043

−1.359

3.953

0.904

1.728

β B L (R O A)

0.348

0.276

−2.734

4.637

1.335

0.442

β B L (R O E)

0.318

0.337

−3.135

3.889

1.339

0.272

S

0.171

0.161

0.087

0.311

0.056

0.885

dS

0.103

0.102

0.042

0.206

0.035

0.585

S(R O A)

0.019

0.015

0.004

0.062

0.012

1.708

S(R O E)

0.039

0.033

0.004

0.128

0.029

1.389

d S(R O A)

0.014

0.012

0.000

0.045

0.011

1.281

d S(R O E)

0.031

0.024

0.003

0.111

0.025

1.524

Notes

R, β,

β H R,

β B L , S, d S,

R O A,

R O E, β(R O A), β(R O E), β H R (R O A), β H R (R O E),

β B L , (R O A), β B L (R O E), S(R O A), S(R O E), d S(R O A), d S(R O E), denote average rate of return, beta coefficient, Harlow and Rao downside beta, Bawa and Lindenberg downside beta, standard deviation of rate of return, semi-deviation of rate of return, average value of profitability ratio return on assets, average value of profitability ratio return on equity, accounting beta for ROA, accounting beta for ROE, Harlow and Rao downside accounting beta for ROA, Harlow and Rao downside accounting beta for ROE, Bawa and Lindenberg downside accounting beta for ROA, Bawa and Lindenberg downside accounting beta for ROE, standard deviation of ROA, semi-deviation of ROA, standard deviation of ROE, semi-deviation of ROE, respectively

and Leoni (2017) considered firms with extreme values of profitability ratios (min of ROE equalled −760, max of ROE equalled 767) and one outlier can skew the results. Market betas are much higher compared to their accounting equivalents. Making a comparison, one can see that accounting betas are much more dispersed than market betas. The minimum values of accounting betas are negative and very low, while maximum values are very high. The reason for this is probably using a data set of firms from disparate sectors. The profitability ratios vary between businesses and react in different ways to the economic situation on the market. Only β B L is symmetrically distributed (approximately). Other systematic risk measures are positively skewed. This means that, in some cases, reaction to the market situation is positive and very strong.

84

A. Rutkowska-Ziarko

Table 4 Correlations between market and accounting measures of profitability and risk Variables

Sample of stock companies All

Large and medium

Small

R

RO A

0.4763***

0.5369***

0.3910**

R

RO E

0.5254***

0.5340***

0.5173***

β

β(R O A)

0.0616

0.2776

−0.2324

β

β(R O E)

0.0244

0.1902

−0.2014

βHR

β H R (R O A)

0.0439

0.2085

−0.2242

βHR

β H R (R O E)

0.0206

0.1595

−0.2263

β BL

β B L (R O A)

0.2750**

0.4031**

0.1601

β BL

β B L (R O E)

0.2942**

0.3925**

0.2194

S

S(R O A)

0.1755

0.1238

0.2225

S

S(R O E)

0.2162*

0.1286

0.3076*

dS

d S(R O A)

0.3873***

0.4260**

0.3533*

dS

d S(R O E)

0.4270***

0.4596***

0.4271**

62

31

31

Number of companies

Notes *, **, ***, indicates significance at the 10%, 5% and 1% level respectively

Table 4 presents the correlations between market and accounting measures of profitability and risk. There is a positive and significant correlation between the average value of the profitability ratios (ROA and ROE) and the rates of return on investment in their shares for all samples under study. This is supported by earlier research for Polish construction and food sector (Rutkowska-Ziarko 2015; Rutkowska-Ziarko and Pyke 2018a, b). The strength of the correlation is similar for the sample of large and medium companies and the sample of small companies for return on equity. Looking at return on assets, the strength of correlations is higher for the sample of large and medium companies. In a previous study for the Polish food sector, significant positive correlations between market and accounting betas were found; for downside risk, the correlations were higher compared to the variance approach (Rutkowska-Ziarko and Pyke 2017). For the Polish construction sector, similar results were obtained, but the correlation between market and accounting measures of systematic risk for profitability on equity (ROE) was not statistically significant. In the mentioned studies (RutkowskaZiarko and Pyke 2017, 2018a, b), downside betas were calculated according to Bawa and Lindenberg (1977). In the current study, significant correlations were found between accounting betas and market betas only for the Bawa and Lindenberg formula which does not occur for the sample of small companies. This indicates that accounting betas are more strongly correlated with market beta when similar companies are considered—for example, companies from one sector. Accounting beta gives additional information to market beta on the risk of the company.

Profitability Ratios in Risk Analysis

85

Furthermore, there is a positive correlation between the variability (semivariability) of profitability ratios (ROA and ROE) and standard deviation (semideviation) of rates of return on the Polish stock exchange. Pearson correlations coefficients are significant for downside approach, both for ROA and ROE for every case. In the variance approach, the strength of correlation is lower and not always significant. In earlier works for the Polish food sector, the correlation between variability (semi-variability) for profitability ratios and rates of returns on the capital market indicated that it was stronger for ROA compared to ROE. The correlations were slightly higher in the downside approach (Rutkowska-Ziarko 2015). For the Polish construction sector, similar results were obtained, but the strongest correlation occurred for the semi-deviation of ROE. The correlations for the variance approach for both ROA and ROE were very low (Rutkowska-Ziarko and Pyke 2018a, b). Knowledge of the variability (semi-variability) of a company’s profitability is important for investors and managers. Table 5 compares classical and downside calculating methods for systematic and total measures of risk. It can be seen that there is a close relationship between the variance and semi-variance approach, the correlation coefficients are positive and high and all of them are significant at the 1% level for every kind of measure and every sample. In the study, two basic profitability ratios (ROA and ROE) were studied. It is unclear if measures based on different profitability ratios give different information or not. In Table 6, the correlation between the average level of ROA and ROE is presented. In addition, the correlations between risk measures based on ROA and ROE are compared. Pearson correlation coefficients are positive and high, and all of them are significant at a 1% level for every kind of measure and every sample. Table 5 Correlations between risk measures in the variance and downside approaches Variables

Sample of stock companies All

Large and medium

Small

0.9491***

0.9554***

0.9517***

β

βHR

β

β BL

0.8138***

0.7817***

0.8453***

β(R O A)

β H R (R O A)

0.9552***

0.9739***

0.9119***

β(R O A)

β B L (R O A)

0.7552***

0.7618***

0.7540***

β(R O E)

β H R (R O E)

0.9574***

0.9761***

0.9119***

β(R O E)

β B L (R O E)

0.7542***

0.7724***

0.7375***

S

dS

0.8002***

0.7433***

0.8310***

S(R O A)

d S(R O A)

0.8095***

0.7779***

0.8440***

S(R O E)

d S(R O E)

0.8272***

0.8191***

0.8433***

62

31

31

Number of companies

Notes *, **, ***, indicates significance of Pearson’s correlation coefficient at the 10%, 5% and 1% level respectively

86

A. Rutkowska-Ziarko

Table 6 Correlations between measures of profitability and risk based on ROE and ROA Variables

Sample of stock companies All

Large and medium

Small

RO A

RO E

0.7577***

0.7176***

0.8144***

β(R O A)

β(R O E)

0.9315***

0.9229***

0.9485***

β H R (R O A)

β H R (R O E)

0.9515***

0.9487***

0.9559***

β B L (R O A)

β B L (R O E)

0.8829***

0.8889***

0.8844***

S(R O A)

S(R O E)

0.8439***

0.7572***

0.9014***

d S(R O A)

d S(R O E)

0.9201***

0.9217***

0.9206***

62

31

31

Number of companies

Notes *, **, ***, indicates significance of Pearson’s correlation coefficient at the 10%, 5% and 1% level respectively

5 Conclusions In the study, the Polish stock market is considered. The study sample is divided into two subsamples. The first subsample employs large and medium companies, while the other subsample employs small companies. This allowed the possible variety between stock companies to be observed for different sizes. Especially in the context of the influence of profitability ratios and their changeability on the behaviour on rates of returns on the Polish capital market. The relation between average accounting profitability in the long term of a company and its average rates of returns on Warsaw Stock Exchange is observed. This occurs, both for ROA and ROE. The variability of the profitability ratios has an impact on the variability of returns on the Polish capital market. It takes place as well for semi- variability and it is more significant. Measures of risk based on ROE are strongly and positive correlated with those based on ROA. In earlier articles, a statistically significant correlation between the average level of profitability ratios and the average rates of return for the Polish food and construction sectors was found. Correlations were also observed for the variability of profitability ratios and risk in the mentioned sectors on the Warsaw Stock Exchange (RutkowskaZiarko 2015; Rutkowska-Ziarko and Pyke 2018a, b). Previous studies provided evidence that there is a positive correlation between different kinds of market beta and accounting beta for food and construction companies (Rutkowska-Ziarko and Pyke 2017, 2018a, b). In this study, the phenomenon was observed for the sample of all companies and for the sample of large and medium companies. For the sample of small companies, the correlations were not statistically significant and not always positive. The positive and significant correlations occurred between downside betas and downside accounting betas calculated with the BL formula.

Profitability Ratios in Risk Analysis

87

In conclusion: • Profitability ratios such as ROA and ROE give important information for investors on the Warsaw Stock Exchange. Both of these ratios are useful in risk analysis. • On the Polish capital market, accounting profitability is more influential on the rates on return and risk for large and medium companies compared to small ones. • The strongest correlation was observed between market beta and accounting beta calculated with the BL formula. • Risk measures based on accounting profitability ratios are a valuable complement to measures based on stock prices. When investors and managers consider these measures together, they receive a comprehensive view of a company’s risk.

References Barucci E, Fontana C (2017) Financial markets theory. Equilibrium, efficiency and information. Springer Bawa VS, Lindenberg EB (1977) Capital market equilibrium in a mean-lower partial moment framework. J Financ Econ 5:189–200 Beaver WH, Kettler P, Scholes M (1970) The association between market-determined and accounting determined risk measures. Account Rev (3):654–682 Florio C, Leoni G (2017) Enterprise risk management and firm performance: The Italian case. The British Accounting Review 49:56–74. https://doi.org/10.1016/j.bar.2016.08.003 Galagedera DUA (2007) An alternative perspective on the relationship between downside beta and CAPM beta. Emerg Mark Rev 8(1):4–19. https://doi.org/10.1016/j.ememar.2006.09.010 Harlow WV, Rao RKS (1989) Asset pricing in a generalized mean-lower partial moment framework: theory and evidence. J Financ Quant Anal 24(3):285–311 Hill NC, Stone BK (1980) Accounting betas, systematic operating risk, and financial leverage: a risk-composition approach to the determinants of systematic risk. J Financ Quant Anal 15(3):595– 637 Konchitchki Y, Luo Y, Ma MLZ et al (2016) Accounting-based downside risk, cost of capital, and macroeconomy. Rev Account Study 21:1–36. https://doi.org/10.1007/s11142-015-9338-7 Markowski L (2018) The relationships between beta coefficients in the classical and downside framework: evidence from warsaw stock exchange. In: Jajuga K, Locarek-Junge H, Orlowski LT (eds) Contemporary trends and challenges in Finance. Springer Proceedings in Business and Economics. Springer, Cham, pp 45–53 Nekrasov A, Shroff PK (2009) Fundamentals-based measurement in valuation. Account Rev 84(6):1983–2011 Rutkowska-Ziarko A (2015) Influence of profitability ratio and company size on profitability and investment risk in the capital market. Folia Oecon Stetinesia 15(23):151–161. https://doi.org/10. 1515/foli-2015-0025 Rutkowska-Ziarko A, Pyke C (2017) The development of downside accounting beta as a measure of risk. Econ Bus Rev 45(4):55–65. https://doi.org/10.18559/ebr.2017.4.4 Rutkowska-Ziarko A, Pyke C (2018) Validating downside accounting beta: evidence from the polish construction industry. In: Jajuga K, Locarek-Junge H, Orlowski L (eds) Contemporary trends and challenges in Finance. Springer Proceedings in Business and Economics. Springer, Cham, pp 81–87. https://doi.org/10.1007/978-3-319-76228-9_8 Rutkowska-Ziarko A, Pyke C (2018) Using accounting information in risk analysis, collegium of economic analysis annals, warsaw school of economics. CollIum Econ Anal 49:547–554

88

A. Rutkowska-Ziarko

Toms S (2012) Accounting-based risk measurement: an alternative to capital asset pricing model derived discount factors. Aust Account Rev 22:398–406. https://doi.org/10.1111/j.1835-2561. 2013.00201.x Toms S (2014) Accounting-based risk management and the capital asset pricing model: an empirical comparison. Aust Account Rev 24(2):127–133. https://doi.org/10.1111/j.1835-2561.2013. 00201.x Woods M, Linsley P, Maffei M (2017) Accounting and risk special issue. Br Account Rev 49:1–3

Impact of Commodity Market Risk on Listed Companies Bogdan Włodarczyk, Alberto Burchi, and Marek Szturo

Abstract Commodity market is one of the most important element of the global economy as a global mechanism of valuation and distribution of goods, it has additionally become a kind of barometer of investor attitudes. This applies to companies whose operations are strongly related to the commodity market. The impact of market risk of commodities on the value of a company and its solvency is possible, when there is a correlation between the prices of commodities and share prices of the company. These types of relationships are not yet fully understood. This is due to the complexity of the processes that shapes the relationship between commodity markets and stock market. The purpose of this article is to assess the credit risk of listed companies whose activities are related to commodity markets, which means they are exposed to the market risk of commodities. Two stock markets have been selected for this purpose: the Italian market with a mature and well-established position in the global capital trading system and the stock market in Poland as representatives of developed economy, characterized by different stage of institutional development. In the opinion of the authors, companies with a significant difference in the probability of default for the baseline and crisis scenario are more exposed to the impact of commodity market risk. Keywords Market risk · Credit risk · Commodity market · Stock market · Bank risk JEL Classification Q02 · Q40 · Q41 · G14 · C12

B. Włodarczyk (B) · M. Szturo University of Warmia and Mazury in Olsztyn, Olsztyn, Poland e-mail: [email protected] M. Szturo e-mail: [email protected] A. Burchi University of Perugia, Perugia, Italy e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_8

89

90

B. Włodarczyk et al.

1 Introduction The relationship between the threats of commodity market and the risk of the stock market may be an important transmission channel of specific phenomena affecting the exposure of the bank’s assets to risk (investment and credit). For example, the real interest rate level determines the demand for raw materials, including the size of their inventories (Bekaert et al. 2013). The purpose of this article is to assess the credit risk of listed companies whose activities are related to commodity markets, which means they are exposed to the market risk of commodities. Two stock exchange markets have been selected for this purpose: the Italian market with a mature and well-established position in the global capital trading system and the stock market in Poland as representative of developed economy, characterized by different stages of institutional development. The growth of the global money supply has become the leading determinant of price changes in international trade (Epstein 2005). Financialization presents itself as specific domination of the financial sector in the structure of modern economy (Henderson et al. 2015). Until the end of the last century, financial investors when creating an investment took into the account forms of assets such as: shares, bonds, currencies, real estate (Crawford and Sen 1996). Then, along with the development of modern portfolio theory (Markowitz 1952), alternative financial instruments have been included. A significant role was played by future contracts for commodities which prices initially showed a lack of correlation with changes in the prices of classic financial assets. This resulted in a significant difference in the price changes between commodity prices and the prices of classical financial assets. Until now, the prices of various financial assets showed statistical correlation (Tang and Xiong 2012). The collapse of the stock market in 2001 and publications of the negative correlations between returns of shares and returns of futures contracts on raw materials drew the attention of financial investors to the commodity futures market in order to reduce their portfolio risk. As a result, there was a significant inflow of funds into the financial instruments related to commodity futures contracts. The extreme intensification of this process is defined as the financialisation of commodity markets. The phenomenon in some cases of strong correlation between prices of commodities and prices of classic financial assets became its expression. It is noteworthy that before 2008, the correlation coefficient between commodity and share prices was very low (close to zero) (Bicchetti and Maystre 2012). After the collapse of Lehman Brothers, this correlation was already at a relatively high level. This phenomenon should be associated with a new active strategy of financial investors and a gradual departure from the passive strategy, which was characterized by the so-called index traders. This is reflected in the fact that the share of these traders in the total amount of funds invested in commodity assets dropped from over 80% in 2005 to only about 30% in 2011 (Dudzi´nski 2012). Commodity markets become more vulnerable to sudden and sharp adjustments due to changes in global financial markets. The best example of this can be strong declines in commodity and food prices during the financial crisis. As pointed out, in

Impact of Commodity Market Risk on Listed Companies

91

just a few months (July 2008–February 2009), the total raw material and food prices index—calculated by IMF—decreased by nearly 60%, and oil prices by nearly 70% (Filis et al. 2011). The demand reported by financial investors should be treated as an expression of demand for financial assets (futures contracts on commodity markets), and not as an expression of the actual demand for products in the real sector of the economy (production and trade) (Hamilton and Wu 2015). Financialisation of commodity markets has created—from the theoretical point of view—an anomaly of the world price information function (Jurado et al. 2015). On developed capital markets, a wide range of shares of companies involved in various activities on commodity markets is available. As research in developed markets show, building a portfolio of shares giving satisfactory exposure to commodity risk requires a precise selection of companies for which the correlation between commodity prices and company profits will be the strongest (Tang and Xiong 2012). Investments in raw material companies—in addition to obtaining exposure to the risk of commodity prices—entails exposure to two additional types of risk: specific risk of the company and systemic risk of the stock market. Specific risk is related to the quality of company management, capital structure or even the quality of communication with shareholders (Filis et al. 2011). Commodity price risk means the strength of the relationship between company share prices and commodity prices. Chung (2003) showed the existence of such a strong correlation in the case of gold mining companies. However, systemic risk means that in the case of some commodity companies, the correlation of their prices with the stock market represented by broad stock market indices is stronger than the link with commodity price risk (Kilian and Murphy 2014). Investments in shares of companies operating in strictly defined sectors of commodity market show moderate correlations with relevant commodity price indices (Silvennoinen and Thorp 2013). Considering the above, we state that a commodity-related company is susceptible to the impact of commodity risk in at least two ways: 1. by reducing company value as a response of investors to forecast changes in commodity prices; 2. by increasing of the company debt caused by losses resulting from unfavorable fluctuations in the prices of goods. In our opinion, researches on the issue of the impact of commodity risk on corporate debt are rather small achievement in the literature. In our article, we try to address this problem using an approach based on measuring the credit risk of commodityrelated companies. This is interesting because the results of our research can be part of the analysis of financial institutions’ exposure to commodity risk. Default (credit) risk depends on the proportion between the value of assets and liabilities. The value of cash flows from income assets is significant, which shapes the rate of return on these assets. In relation to liabilities, the value of debt-related cash flows determines the cost of debt, which is significant. The linear correlation between changes in commodity prices and changes in the value of assets and liabilities of companies may indicate the existence of a relationship

92

B. Włodarczyk et al.

between commodity risk markets and default risk. Considering operational activity of the materials sector companies, fundamental impact on the value of their assets and liabilities comes from prices of commodity markets. If capital market is efficient, everything that affects the value of the company’s assets and liabilities is reflected in the share price (Bartram 2005). Hence, Granger causality testing was used for verification of the existence of such dependence among the companies in the surveyed markets.

2 The Scope of Research and Research Methods The study involved 20 companies from the following sectors: energy, mining, chemical, telecommunications and petroleum; listed on two European stock exchanges (Warsaw Stock Exchange 10 companies, Borsa Italiana Milan 10 companies). The selection of companies was based on purposive sampling. Their selection criteria was their relationship with the commodity markets in real terms (supplier/recipient). The selection of companies was also based on the results of our previous research using Granger causality in relation to time series of the values of commodity indices and time series of share prices of listed companies (Włodarczyk 2018). The study period included June 2018. The default (credit) risk analysis of listed companies was conducted using the Merton Model, taking into account the probability of default for both the base and stress scenarios. The baseline scenario referred to normal company’s share price, which was adopted as a 250 periodic straight moving average. The stress scenario took into account the share price, which was the minimum price in the last 52 weeks. Data from the Wall Street Journal portal has been used for these calculations (https:// quotes.wsj.com/PL/XWAR/, accessed on 25/06/2018). The probability of default (Merton Model) is defined as the probability of the asset value falling below the debt threshold at the end of the time horizon T: P D = P{ln(VT ) < ln(L)}     ln(L/VT ) − μ − σ 2 /2 (T − t) P D = N(0,1) √ σ T −t where PD VT L σ μ T t

probability of default, value of the company’s assets in period T, nominal value of company’s debt, company asset’s return volatility (standard deviation), drift parameter, average of the company asset’s returns, debt maturity, current period 0 ≤ t ≤ T,

Impact of Commodity Market Risk on Listed Companies

93

N(0,1) (·) cumulative standard normal distribution function. The debt threshold is the exercise price of the call option held by the shareholders. The option’s underlying security is the company’s asset value. Due to the unobservability of the market value of the company’s assets, the option pricing model was used to estimate these values. For this purpose, it is assumed that the value of the company’s equity is the value of the call option on the company’s assets. The estimation of asset volatility was made using numerical method, assuming that the value of company’s equity reflects all relevant information, including those related to the impact of commodity market price volatility.

3 The Results of the Warsaw Stock Exchange Study In relation to two companies (Konsorcjum Stali, Tauron), the credit risk was at a high level (probability of default PD for up to 10 years oscillated in the range of 50–100% for two scenarios) (Fig. 1). On the opposite side there were three companies (Grupa Azoty, Mangata, PKN Orlen) with very low credit risk (the probability of default PD within 10 years fluctuated in the range of 0% to 1% for two scenarios) (Fig. 1). The remaining 5 companies (Astarta, Boryszew, Grupa Lotos, Serinus, KGHM) could be graded in a gradual manner credit risk, with the probability of default in the range of 1–50% for at least one scenario in the first 10 years (Fig. 1). The companies could also be divided due to the sensitivity of the level of credit risk, taking into account the baseline and stress scenario. This sensitivity was related to the impact of market valuation (share price) on the level of credit risk of a given company. If the curve for the baseline scenario largely coincided with the curve for the stress scenario, the company was characterized by credit risk insensitive to changes in the price of its shares (Grupa Azoty, Tauron, Mangata). The big difference between the curves meant a significant sensitivity of the company’s credit risk level to the stock market valuation of its shares (Astarta, Boryszew, KGHM, Konsorcjum Stali, Grupa Lotos, Serinus, PKN Orlen) (Fig. 1). Companies with increased credit risk sensitivity to their valuation were potentially more exposed to commodity market risk. Considering our earlier Granger causality studies, a group of companies was identified that were linked with at least one commodity index and had significant credit risk sensitivity to market valuation. This group included seven companies: Astarta Holding NV (causal relation exists with 14 commodity indices), Boryszew (5 commodity indices), Grupa Lotos (6 commodity indices), KGHM (17 commodity indices), Konsorcjum Stali (4 commodity indices), PKN Orlen (1), Serinus Energy Inc. (6) (Fig. 1).

94

B. Włodarczyk et al.

Fig. 1 Probability of default estimated by the Merton model for companies (Warsaw Stock Exchange). Note The number in brackets specifies the number of commodity indices for which Granger causality was identified in relation to share prices of given company stocks. Source Own study

Impact of Commodity Market Risk on Listed Companies

95

4 The Results of the Italian Stock Exchange-Borsa Italiana Study In relation to the two studied companies (Enel, Terna), the credit risk was at a high level (the probability of default (PD) in the time frame of up to 10 years fluctuated in the range of 50–100% for both scenarios) (Fig. 2). This is confirmed by the opinions on the two companies issued by rating agencies. The ratings are almost identical and are positioned close to the “non-investment grade” area for all the major agencies (S&P’s, Moody’s, Fitch). Two companies (Prysmian, STMicroelectronics) experienced a very low credit risk (the probability of default (PD) in the time frame of up to 10 years fluctuated in the range of 0–1% for two scenarios) (Fig. 2). These two companies are those that are less close to the energy production sector and are mainly involved in manufactures electric power transmission and telecommunications cables (Prysmian) and production of multinational electronics and semiconductor (STMicroelectronics). The remaining eight companies could be classified in a way that would increase the credit risk, in at least one scenario, in the first 10 years, the probability of default ranged from 1 to 50%. Companies with a high credit risk sensitivity to their value (A2A, Eni, Italgas, Saipem, Snam) were potentially more exposed to the risk of the commodity market (Fig. 2). Group of four companies was identified, in regards to which there was a significant relationship between the risk of the commodities market and their credit risk. They include companies listed on the Italian stock exchange: A2A (4 raw material indices), Enel (5 raw material indices), Saipem (6 raw material indices), Italgas (4 raw material indices). These companies share the involvement in the energy sector, however the identification of commodity indices related to credit risk is not univocal. In other words, the risk of these companies does not depend in a preponderant and exclusive way on the fluctuations of the oil or other energy commodity market. In the group of the above companies (Fig. 2) exposed to credit risk related to commodity markets, the high level of credit risk at the time of measurement (June 2018) was recorded in the following cases: Italgas, Snam and Terna. All these companies are among the top 40 for capitalization and are included in the constituents of the most important index of the Italian market: the FTSE-MIB. But most of all, these companies represented the energy sector and played a significant role within it. Companies such as Prysmian, STMicroelectronics, and Tenaris were characterized by an initial low credit risk and a relatively small number of raw material market indices had an impact on their share prices. This result leads us to assume, on equal terms, a positive relationship between the level of credit risk and the relationship between the risk and the commodities market. That is, we can hypothesize that as credit risk increases, the company increases its sensitivity to fluctuations and the dynamics of the commodities market. However, this aspect requires appropriate and in-depth further analysis. Given the lack of relationship between the price of shares and the dynamics of raw material

96

B. Włodarczyk et al.

Fig. 2 Probability of default estimated by the Merton model for companies (Italian Stock Exchange—Borsa Italiana). Note The number in brackets specifies the number of commodity indices for which Granger causality was identified in relation to share prices of given company stocks. Source Own study

Impact of Commodity Market Risk on Listed Companies

97

market indices, we can state that this companies did not constitute the main channel of transmission of raw material market risks on the Italian exchange during the studied period.

5 Conclusion Presented results allow for the following conclusions: 1. In the opinion of the authors, companies with a significant difference in the probability of default for the baseline and crisis scenario are more exposed to the impact of commodity market risk. In this case, the impact relates to a decrease in their market value, which increases the probability of default of these companies. 2. A comparison of companies from two European stock exchanges leads to the conclusion that more companies which are susceptible to the impact of the commodity market on their credit risk are on the Warsaw Stock Exchange. However, the differences are not great. The Italian stock exchange is characterized by more even distribution of these characteristics among the surveyed companies. 3. Generally, it can be stated that conditions favoring the impact of commodity market risk on the credit risk of listed companies have been observed. These conditions come down to a situation where instability on the commodity market changes into a decrease in the value of a given company related to commodities (sensitivity to stress scenario). Then it causes an increase in the share of debt in the company’s financing structure and an increase in the probability of default. This conclusions are the starting point for broader research into the model approach to this phenomenon and the classification of companies due to its intensity. This is especially important in relation to the deepening of the credit risk analysis from the point of view of financial institutions lending to goods related companies. The complexity of the above relations and the significant share of raw material companies on the studied stock exchanges means that the commodity market will increasingly affect the situation of investors (including banks), especially those involved in financing enterprises dependent on revenue or costs of raw materials.

References Bartram SM (2005) The impact of commodity price risk on firm value—an empirical analysis of corporate commodity price exposures. Multinatl Financ J 9(3/4):161–187. https://doi.org/10. 17578/9-3/4-2 Bekaert G, Hoerova M, Duca ML (2013) Risk, uncertainty and monetary policy. J Monet Econ 60(7):771–788. https://doi.org/10.1016/j.jmoneco.2013.06.003 Bicchetti D, Maystre N (2012) The synchronized and long-lasting structural change on commodity markets: evidence from high frequency data. UNCTAD Discussion Papers, no. 208, October. Retrieved from https://unctad.org/en/PublicationsLibrary/osgdp2012d2_en.pdf

98

B. Włodarczyk et al.

Chung SY (2003) Do financial analysts value corporate hedging strategies? A case of precious metal mining firms. SSRN Electron J. https://doi.org/10.2139/ssrn.492722 Crawford G, Sen B (1996) Derivatives for decision makers: strategic management issues. Wiley, New York Dudzi´nski J (2012) Ewolucja działalno´sci inwestorów finansowych na rynkach towarowych. Acta Universitatis Lodzienis, Folia Oeconomica 273:77–78. Retrieved from https://hdl.handle.net/ 11089/2061 Epstein G (ed) (2005) Financialization and the world economy. Edward Elgar Publishing, Northampton Filis G, Degiannakis S, Floros C (2011) Dynamic correlation between stock market and oil prices: the case of oil-importing and oil-exporting countries. Int Rev Financ Anal 20(3):152–164. https:// doi.org/10.1016/j.irfa.2011.02.014 Hamilton JD, Wu JC (2015) Effects of index-fund investing on commodity futures prices. Int Econ Rev 56(1):187–205. https://doi.org/10.1111/iere.12099 Henderson B, Pearson N, Wang L (2015) New evidence on the financialization of commodity markets. Rev Financ Stud 28(5):1285–1311. https://doi.org/10.1093/rfs/hhu091 Jurado K, Ludvigson SC, Ng S (2015) Measuring uncertainty. Am Econ Rev 105(3):1177–1216. https://doi.org/10.1257/aer.20131193 Kilian L, Murphy DP (2014) The role of inventories and speculative trading in the global market for crude oil. J Appl Econ 29(3):454–478. https://doi.org/10.1002/jae.2322 Markowitz HM (1952) Portfolio selection. J Financ 7:77–91. https://doi.org/10.1111/j.1540-6261. 1952.tb01525.x Silvennoinen A, Thorp S (2013) Financialization, crisis and commodity correlation dynamics. J Int Financ Markets Inst Money 24:42–65. https://doi.org/10.1016/j.intfin.2012.11.007 Tang K, Xiong W (2012) Index investment and the financialization of commodities. Financ Anal J 68(6):54–74. https://doi.org/10.2469/faj.v68.n6.5 Włodarczyk B (2018) Rynek surowców a ryzyko bankowe. Wydawnictwo UWM, Olsztyn

Corporate Finance

The Double Relationship Between Risk Management and CSR in the Italian Healthcare Sector: The Case of the Lombard “Health Protection Agencies” (ATS) Patrizia Gazzola, Stefano Amelio, and Alessandro Figus Abstract The aim of the paper is to analyze how socially responsible behaviors can be considered as risk management tools. In particular, the underlying objective is to highlight the existence of a link between CSR and risk management within the healthcare sector of the Lombardy Region (Italy). The research is divided into two sections and the approach used combines both descriptive analysis and quantitative analysis methods: in the first part will be analyzed the concept of corporate social responsibility and risk management, describing the same concepts in the healthcare sector. In the second part, in order to highlight the CSR-risk management link, the paper analyses the web sites of the 8 health protection agencies of the Lombardy Region, created following the reform of the social and health system in Lombardy (Regional Law 23/2015). In our paper we demonstrate how a double bond (a double relationship) between CSR and RM exists. The first link classifies the CSR as an RM tool. At the same time, the RM can be considered a tool to demonstrate the social responsibility of the institution (or as a tool to prove that an institution is socially responsible).

1 Introduction The concept of corporate social responsibility (CSR) has been extensively studied in the literature (Singh 2016; H˛abek and Wolniak 2016; Arru and Ruggieri 2016; Amelio 2016; Gazzola and Mella 2017; Saka et al. 2018; Dyck et al. 2019), mainly P. Gazzola (B) University of Insubria, Varese, Italy e-mail: [email protected] S. Amelio University of Milano, Bicocca, Italy e-mail: [email protected] A. Figus M. Kozybayev North State Kazakhstan University, Petropavl, Kazakhstan e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_9

101

102

P. Gazzola et al.

with reference to the private sector (profit and non-profit) (Castelló and Lozano 2009; Gazzola et al. 2019a). On the contrary, this issue has not found fertile ground in the healthcare sector although the strong relationship that links healthcare to society is important (Jamali et al. 2008, 2010). In literature, the theme of risk and risk management (RM) in the healthcare sector followed the same evolutionary path of CSR. In fact, it has gained on significance only in the last 10 years (Capocchi et al. 2019). In the healthcare sector, CSR, although little investigated, plays an important role due to the relationship between healthcare and society (Russo 2016): the healthcare receives from the society the mandate to take care of patients. The most important task of the health firms is to provide service to society (Abela 2001). The health sector is “however particular” and different compared to the other economic sectors as the characteristics of the “patient” are different from those of the “client”. This is why a healthcare company cannot use methods and strategies deriving from other sectors (Gazzola et al. 2019b). As Russo (2016) states, CSR literature in the healthcare sector could be divided into three groups, based on the relationship between society and health: • social responsibility and organization; • social responsibility and social impact; • social responsibility and competitiveness. The first group derives from Spencer et al. (1999) opinion for which organization is the key link between the economic-financial, human and social dimension: “a healthcare organization […] is […] a provider organization with an administrative structure consisting of a board of directors, management personnel and professionals, and which supplies […] services to individual patients and groups of patients”. In relation to this first group, the organization could suffer the risk that the three mentioned dimensions (economic-financial, human and social dimension) may not be in equilibrium with each other and this could lead to negative consequences for the survival of the company. The second group comes from the Drucker (1989) thought “their first social responsibility is to do their job” from which it emerges that they must be responsible for their impact, acting as a member of a community. In this case, the risk to be managed consists in the possibility that the community (patients and non-patients) suffers negative externalities due to errors committed by healthcare organizations. In particular, as highlighted below, patients could suffer negative consequences deriving from adverse events (damage or discomfort) attributable to medical treatment (clinical risk). These risks need to be identified and managed by healthcare organizations. CSR is also a tool to generate profit and to get competitive advantages (third group). In this sense, CSR could be considered as a share responsibility to use resources effectively to deliver better health (Galvin 2010), an instrument to promote a more competitive, efficient and accessible healthcare sector. In relation to this aspects, more and more institutions operating in the healthcare sector adopt CSR activities in their work programs (Lubis 2018) such as within the PARMs (annual

The Double Relationship Between Risk Management …

103

risk management plan). In particular, CSR activities influence the hospital reputation among stakeholders and, consequently, the hospital value (Susanto 2009; Ihlen et al. 2011; Doda 2015). For this reason, Lubis (2018) demonstrates that CSR should be considered a strategic tool to be used to avoid such risks. This was confirmed also by Bem et al. (2019). In literature there are many studies on the relationship between CSR and RM (McGuire et al. 1988; Feldman et al. 1997; Orlitzky and Benjamin 2001; Husted 2005a, b; Godfrey et al. 2009; Oikonomou et al. 2012; Salama et al. 2011). In general, these studies show a positive correlation between the performance of CSR activities and the presence of risks within the company. Consequently, if CSR activities reduce corporate risk, CSR activity becomes critical and vital in sectors intrinsically characterized by risk, such as the healthcare sector. Starting from these assumptions, the aim of this paper is to demonstrate how a double relationship between CSR and RM exists. In particular, the considered hypotheses are: H1—CSR is an RM tool. H2—RM is a tool to demonstrate the social responsibility of the institution (or as a tool to prove that an institution is socially responsible). To demonstrate the existence of this double bond, the paper is divided into two sections. In the first part will be analyzed the concept of risk management, describing the same concept in the healthcare sector. In the second part, in order to highlight the CSR-risk management link, the paper analyses the web sites of the 8 health protection agencies of the Lombardy Region, created following the reform of the social and health system in Lombardy (Regional Law 23/2015) which has, among other things, transformed the former ASL (Local Health Authority) into ASST (Agency of health protection).

2 Risk, Risk Management and the Healthcare Sector Business risk can be defined as “a risk inherent in a firm’s operations as a result of external or internal factors that can affect a firm’s profitability” (Jo and Na 2012). There are in particular two types of risk: systematic risk (risk that affects most corporate assets and is also called market risk) and unsystematic risk (which affects a small number of assets and is called firm-specific unique risk) (Ross et al. 2011). There are relations between the concept of risk and that of uncertainty (Knight 2012). In essence, the risk exists because there is no certainty about the outcome of a given event. For these reasons every economic-social system must adopt risk management systems: there is no sustainable social-economic development if the institutions do not assume the risks deriving from their activity and manage them. The RM in particular is defined as the set of coordinated activities, useful for guiding and controlling an organization with reference to the risk (ISO 31000 “Risk management—Principles and guidelines”). Even in finance, risk is an essential factor that

104

P. Gazzola et al.

needs to be governed. Risk management is therefore a financial instrument (Stulz 1996). As Capocchi et al. (2019) remember, RM in the healthcare sector is “a system composed of several processes by which organizations try to estimate, measure, and prevent risk in order to reduce negative impacts on different variables, such as technical and economic aspects”. To reduce risks in healthcare over time, various tools have been used: from insurance coverage (Harrington and Niehaus 2003) to the most recent managerial tools, techniques and methodologies (Messano et al. 2014), including training (Capocchi et al. 2019) and various CSR activities. The characterizing and priority part of the risk profile of the healthcare companies is constituted by the clinical risk dimension (Cagliano et al. 2011; Sale 2005; Kohn and Corrigan 1999), defined as the probability that a patient is the victim of an adverse event (damage or discomfort), attributable, even if involuntarily, to medical treatment lend him during a period of hospitalization and able to cause a prolongation of the period of hospitalization, a worsening of health conditions or death. Consequently, as suggested in the doctrine (Lucas 1997; Reason 2001), to reduce errors, it is necessary to adopt risk management systems, but risk management systems in the broad sense, as in fact the tool of CSR activities in the present study is considered.

3 CSR and RM: Is There a Link? As previously stated, the objective of this paper is to demonstrate the existence of a double bond that correlates CSR and RM. In literature, several authors have implicitly identified this link, although they focused primarily on a unidirectional link (CSR → RM or alternatively RM → CSR) without grasping explicitly the existing two-way relationship between the two concepts (CSR → RM and simultaneously RM → CSR). In other words, the bidirectional link can be explained as: The first link classifies CSR as an RM tool. At the same time, the RM can be considered a tool to demonstrate the social responsibility of the institution (or as a tool to prove that an institution is socially responsible). Story and Price (2006), showing the results of their research, demonstrate that CSR is a tool of RM (first link), indeed “CSR activities were important to responding organizations mainly as a means of improving risk management systems, enhancing the image of organisation as well as for ethical reasons”. Moreover, they later state that classifying the risks, a typical step of the RM, is a CSR activity. Unerman (2008) believes that CSR is a tool to minimize reputation risk (first link) when he states that “a prime motive for corporations to report on issues of social responsibility is a desire to minimize risks to their reputations”. The company’s reputation is indeed “a valuable asset which needs to be protected and developed, and a key aspect of this reputation is stakeholders’ perceptions of the CSR corporation— or, more precisely, perceptions of how well the CSR policies corporation, practices and outcomes meet stakeholders’ social and environmental values and expectations”. Companies generally use social reports as a means of increasing corporate reputation

The Double Relationship Between Risk Management …

105

(Clarke and Gibson-Sweet 1999), especially when “negative incidents occur that expose CSR shortcomings of particular corporations or industries”, of which the healthcare sector is certainly part, being exposed more than other industries to risks of error. Moreover, in the context of CSR and competitiveness (Russo 2016), of fundamental importance is the moral reputation hospital. Husted (2005a, b), conducting a literature review, found a negative correlation between CRS and RM: even in this case evidence of the first bond (first link) emerges but not of the second. In particular, considering CSR as a form of investment, he states that the greater the investment in CSR activities, the lower is the business risk to manage. The higher RM, less CSR, is not verified. Jo and Na (2012) found that “CSR engagement inversely affects firm risk after controlling for various firm characteristics”. Also in this case, there is evidence of the first link (“CSR engagement helps their risk management effort”) but not of the second. Fundamental is what they say later, namely that CSR is a more powerful tool than other forms of risk management insurance. Castellò and Lozano (2009) explicitly emphasize that “In the risk management posture, CSR is seen as a tool to protect reputational value.” In this case, reputational value is considered by the authors as an element of risk to be managed (therefore as an element of RM). CSR activities consequently act as a tool for risk management (first link). As from the conducted literature review emerges, the main topic studied by the authors is attributable to the first link. The second link is traceable (but only minimally) in Story and Price (2006), although the authors simply state that classifying the risks is an activity that falls within the corporate social responsibility and not that through this activity the company proves to be socially responsible. Indirectly, however, they come to say that RM is a way to demonstrate the CSR of a company.

4 Results and Discussion In Italy, The National Health Service (SSN) is a system of structures and services that have the purpose of guaranteeing all citizens, under conditions of equality, universal access to the equitable provision of health services, in implementation of Article 32 of the Constitution. On the basis of 2019 data, “Lombardy Region” confirms its first place in terms of health attractiveness and reliability. In the Lombardy Region, the health system includes eight Health Protection Agencies (ATS) and twenty seven Agency of health protection (ASST), as a result of regional law n. 23 of 11 August 2015, as well as various other types of structures subject to regional socio-health planning. ATS and ASST replaced the ASL (local health authorities) and the AOs (hospital companies). In order to highlight the CSR-risk management link, starting from the assumption that the twenty seven Agency of health protection (ASST) are subjected to the eight Health Protection Agencies (ATS), the paper analyses the web sites of the 8 health protection agencies (ATS) of the Lombardy Region, which are: ATS Bergamo, ATS

106

P. Gazzola et al.

Brescia, ATS Insubria, ATS Val Padana; ATS Milano Città Metropolitana, ATS Brianza, ATS Pavia, ATS Montagna. It was therefore decided to analyze all the ATSs operating in the Lombardy Region, as the twenty seven Agency of health protection (ASST) are subjected to the management and coordination of the ATSs. Therefore the results achieved by concentrating on the eight ATSs can be extended to the twenty seven ASSTs subordinate to them. In Italy, there are 101 healthcare agencies, comparable to the Lombard ATS, so the sample analyzed in the present study represents about 8% of the Italian population. The Lombardy Region represents a virtuous case in Italy: a study conducted on 7 indicators (satisfaction on health services, active mobility, passive mobility, health expenditure, impoverished families due to out-of-pocket health costs, legal costs for litigation disputes and unfavorable judgments, political costs) confirms the top positions in the ranking. As a consequence, worthy of attention and analysis, although the agencies analyzed represent only 8% of national agencies. Moreover this region, inhabited by more than 10 million inhabitants (16.5% of the Italian population, data of 2018) represents the most populous of Italy, and has almost double the number of residents compared to Lazio, which follows it in the ranking. In order to evaluate the level of connection, this part is developed through the following steps: • selection of the companies surveyed (8 ATS of Regione Lombardia) • selection of the documents to be analyzed • analysis of the main result upon the objective of the research. The first step of the analysis consists in the selection of the 8 ATS of the Lombardy Region and in the consequent exploration of the relative websites. Once the sample under study is identified, the second step consists in selection of the documents to be analyzed. Since the study focuses on the analysis of CSR-RM link, the selected documents are mainly of two types: • annual risk management plan • sections dedicated to CSR on the website. The annual risk management plan (PARM) is the business tool to promote and implement initiatives for the operational definition and risk management. The Plan is drawn up annually in line with regional indications and with the 18 Ministerial Recommendations on clinical risk. In PARM, projects and actions are identified, based on clinical risk objectives, which will be developed during the year and the correlated centers of responsibility, resources and mechanisms for monitoring the progress of activities will be defined. The Plan proposal is formulated by the Risk Manager in agreement with the Coordination Group for Risk Management activities and the Claims Evaluation Committee. The results of the second step are summarized in Table 1. As it is possible to understand from the table, all the ATS present the PARM annually, although for an ATS the PARM (with the words “Annual risk management plan”) has not been traced but only a document (dated 2016) entitled “Evaluation document of risks”.

The Double Relationship Between Risk Management …

107

Table 1 CSR-RM No.

ATS

PARM

CSR Social report

Section within the website

1

322—ATS DELL’INSUBRIA

Yes 2018

No

Yes

2

324—ATS DELLA BRIANZA

Yes 2018

No

Yes

3

321—ATS DELLA CITTA’ METROPOLITANA DI MILANO

Yes/no 2016

No

Yes

4

323—ATS DELLA MONTAGNA

Yes 2018

Yes (mandate report)

Yes

5

327—ATS DELLA VAL PADANA

Yes 2018

No

Yes

6

325—ATS DI BERGAMO

Yes 2018

Yes (social report)

Yes

7

326—ATS DI BRESCIA

Yes 2018

No

Yes

8

328—ATS DI PAVIA

Yes 2018

No

Yes

In relation to the CSR, only 2 ATS have published the social report (document not required by the Italian law) but, all the ATS have sections within the website that can be linked to the social responsibility of the agency. In reality this latter result is not surprising given the activity carried out by the ATS, within a sector with a high impact on the human person such as health. A further step consists in reading the PARMs to identify any financial data related to the topic of risk management. The objective is to investigate the existence of a relationship between risk management and economic-financial data: in particular, does risk management allow the improvement of numerical data? From the analysis of the PARMs of all the ATS investigated, it emerges the absence of a section dedicated to financial data (since it is not a mandatory datum according to the recommendations). The two social reports have also been investigated. Also in this case the same conclusions are confirmed: there are no references to emerging economic-financial data following risk management. It is clear that these issues included in the RM of the healthcare companies analyzed, are typical CSR themes. The first link highlighted above therefore clearly emerges, namely that CSR is an RM tool (H1 is verified). These agencies, in fact, make use of the CSR (and expose on their website or within the social report the sustainable activities they carry out) to manage the risk to which they are subjected. Indeed, the analysis of the PARMs shows that these sustainable activities are included in the list of strategies implemented by the ATS (and made explicit in the PARMs), as required by the Ministerial Recommendations. Ultimately, the socially responsible

108

P. Gazzola et al.

activities carried out by the Agencies (CSRs) are present within the PARMs (RM tool). However, the second link also emerges, namely that RM can be considered a tool to demonstrate the social responsibility of the institution (or as a tool to prove that an institution is socially responsible). In 6 cases, in fact, the ATS do not publish social reports (they only have sections within the website that could be compared to the CSR). But despite this, they are socially responsible companies since, by publishing the PARM, they are able to prove that they are socially responsible towards the various stakeholders involved. Moreover, to greater demonstration of this, the 2 ATS that publish the social report, dedicate inside of it some sections to the topic of the risk (H2 is verified). Also in this case it is possible to observe how these agencies use the PARM (RM tool) to prove that they are socially responsible. In fact, in this document, as emerged just above, strategies and projects of a CSR nature are listed. PARM is drawn up in line with regional indications and with Ministerial Recommendations; it represents a document of fundamental importance for risk management but, by virtue of the contents, it is substantiated in a document with which the agency demonstrates to carry out sustainable activities. CSR activities emerges in the PARM, consequently CSR becomes one with risk management; it then flows under the umbrella of the RM: naturally the synergy at this point is two-way. Companies also divide the RM into sections, which is why a risk manager for the environment (CSR) is also established within the organizations.

5 Conclusions Based on the analysis conducted, the study demonstrates the double connection between CSR and RM in the healthcare sector (H1 and H2). This paper presents elements of originality as there are no papers in the literature that jointly consider CSR and RM in the healthcare field under the double perspective of analysis. There is also a gap in this respect in the non-health field. The main limitation consists in the sample investigated: the analysis is limited to the 8 ATS of the Lombardy Region; therefore it does not consider the ASLs and the AOs (hospital companies) of the other Italian Regions. The results achieved can be extended to the 27 ASST of the Lombardy Region, in fact as mentioned, the ASSTs are subordinated and controlled by the ATS. Therefore, the results are related to the entire Lombardy Region. It is expected that they can also be extended to other Italian Regions, although the regional laws are partially different and consequently also the organizational structure that follows.

The Double Relationship Between Risk Management …

109

References Abela AV (2001) Profit and more: catholic social teaching and the purpose of the firm. J Bus Ethics 31(2):107–116. https://doi.org/10.1023/A:1010746913619 Amelio S (2016) The connection between IAS/IFRS and social responsibility. Manag Dyn Knowl Econ 4(1):7–30 Arru B, Ruggieri M (2016) I benefici della Corporate Social Responsibility nella creazione di valore sostenibile: il ruolo delle risorse di competenza e del capitale reputazionale [The benefits of Corporate Social Responsibility in creating sustainable value: the role of resources of competence and reputational capital]. Economia Aziendale Online 7(1):17–41. https://doi.org/10.6092/20385498/7.1.17-41 Bem A, Siedlecki R, Pr˛edkiewicz P, Gazzola P, Ryszawska B, Ucieklak-Je˙z P (2019) Hospitals’ financial health in rural and urban areas in Poland: does it ensure sustainability? Sustainability 11(7):1932 Cagliano AC, Grimaldi S, Rafele C (2011) A systemic methodology for risk management in healthcare sector. Saf Sci 49(5):695–708. https://doi.org/10.1016/j.ssci.2011.01.006 Capocchi A, Orlandini P, Pierotti M, Luzzi L, Minetti L (2019) Risk management in the healthcare sector and the important role of education and training activities: the case of Regione Lombardia. Int J Bus Manag 14(1):180–191 Castelló I, Lozano J (2009) From risk management to citizenship corporate social responsibility: analysis of strategic drivers of change. Corp Governance Int J Bus Soc 9(4):373–385. https://doi. org/10.1108/14720700910984927 Clarke J, Gibson-Sweet M (1999) The use of corporate social disclosures in the management of reputation and legitimacy: a cross sectoral analysis of UK Top 100 Companies. Bus Ethics Eur Rev 8(1):5–13. https://doi.org/10.1111/1467-8608.00120 Doda S (2015) The importance of corporate social responsibility. J Sociol Res 6(1):86–91 Druker PF (1989) The new reality. Harper & Row, New York Dyck A, Lins KV, Roth L, Wagner HF (2019) Do institutional investors drive corporate social responsibility? International evidence. J Financ Econ 131(3):693–714 Feldman J, Soyka P, Ameer P (1997) Does improving a firm’s environmental management system and environmental performance result in a higher stock price? J Investing 6:87–97. https://doi. org/10.3905/joi.1997.87 Galvin RS (2010) Thinking clearly about payment reform. Am J Manag Care 16(2):89–91 Gazzola P, Mella P (2017) Can CSR influence employees satisfaction? Economia Aziendale Online 7(4):331–337. https://doi.org/10.13132/2038-5498/7.4.331-337 Gazzola P, V˘at˘am˘anescu EM, Andrei AG, Marrapodi C (2019a) Users’ motivations to participate in the sharing economy: moving from profits toward sustainable development. Corp Soc Responsib Environ Manag 26(4):741–751 Gazzola P, Grechi D, Ossola P, Pavione E (2019b) Certified Benefit Corporations as a new way to make sustainable business: the Italian example. Corp Soc Responsib Environ Manag 26(6):1435– 1445 Godfrey PC, Merrill C, Hansen J (2009) The relationship between corporate social responsibility and shareholder value: an empirical test of the risk management hypothesis. Strateg Manag J 30(4):425–445. https://doi.org/10.1002/smj.750 H˛abek P, Wolniak R (2016) Assessing the quality of corporate social responsibility reports: the case of reporting practices in selected European Union member states. Qual Quant 50(1):399–420. https://doi.org/10.1007/s11135-014-0155-z Harrington SE, Niehaus G (2003) Capital, corporate income taxes, and catastrophe insurance. J Financ Intermediation 12(4):365–389. https://doi.org/10.1016/j.jfi.2003.07.001

110

P. Gazzola et al.

Husted BW (2005a) Risk management, real options, corporate social responsibility. J Bus Ethics 60(2):175–183 Husted BW (2005b) The relationship between corporate social responsibility and shareholder value: an empirical test of the risk management hypothesis. J Bus Ethics 60(2):175–183. https://doi.org/ 10.1002/smj.750 Ihlen Ø, Bartlett JL, May S (2011) Corporate social responsibility and communication. The handbook of communication and corporate social responsibility, pp 3–22 ISO 31000, Risk management—principles and guidelines Jamali D, Safieddine AM, Rabbath M (2008) Corporate governance and corporate social responsibility synergies and interrelationships. Corp Governance Inter Rev 16(5):443–459. https://doi. org/10.1111/j.1467-8683.2008.00702.x Jamali D, Hallal M, Abdallah H (2010) Corporate governance and corporate social responsibility: evidence from the healthcare sector. Corp Governance Int J Bus Soc 10(5):590–602. https://doi. org/10.1108/14720701011085562 Jo H, Na H (2012) Does CSR reduce firm risk? Evidence from controversial industry sectors. J Bus Ethics 110(4):441–456 Knight FH (2012) Risk, uncertainty and profit. Courier Corporation Kohn LT, Corrigan JM (1999) To err is human: building a safer health system. In: Donaldson MS (ed) Committee on quality of health care in America. Institute of Medicine, Washington, DC Lubis AN (2018) Corporate social responsibility in health sector: a case study in the government hospitals in Medan Indonesia. Verslas: teorija ir praktika 19(1):25–36 Lucas D (1997) The causes of human error. In: Redmill R (ed) Human factors in safety critical systems. Butterworth Heinemann, Oxford, pp 37–65 McGuire JB, Sundgren A, Schneeweis T (1988) Corporate social responsibility and firm financial performance. Acad Manag J 31:854–872. https://doi.org/10.5465/256342 Messano G, De Bono V, Di Folco F, Marsella LT (2014) Past and present of risk management in healthcare Oikonomou I, Brooks C, Pavelin S (2012) The impact of corporate social performance on financial risk and utility: a longitudinal analysis. Financ Manag 41(2):483–515. https://doi.org/10.1111/j. 1755-053X.2012.01190.x Orlitzky M, Benjamin J (2001) Corporate social responsibility and firm risk: a meta-analytic review. Bus Soc 40(4):369–396. https://doi.org/10.1177/000765030104000402 Reason J (2001) Understanding adverse events: the human factor. In: Vincent C (ed) Clinical risk management. Enhancing patient safety. BMJ Books, London, UK Ross S, Westerfield R, Jordan B (2011) Essentials of corporate finance, 7th edn. McGraw-Hill Irwin, Boston Russo F (2016) What is the CSR’s focus in healthcare? J Bus Ethics 134(2):323–334 Saka C, Noda A, Jimichi M (2018) Cultural influence on corporate social responsibility disclosure in East Asia. Int Rev Bus 18:1–28 Salama A, Anderson K, Toms JS (2011) Does community and environmental responsibility affect firm risk: evidence from UK panel data 1994–2006. Bus Ethics Eur Rev 20:192–204. https://doi. org/10.1111/j.1467-8608.2011.01617.x Sale D (2005) Understanding clinical governance and quality assurance. Palgrave Macmillan, New York Singh J (2016) The influence of CSR and ethical self-identity in consumer evaluation of cobrands. J Bus Ethics 138(2):311–326. https://doi.org/10.1007/s10551-015-2594-4 Spencer E, Mills A, Rorty M, Werhane P (1999) Organization ethics for healthcare. Oxford University Press, New York Story D, Price TJ (2006) Corporate social responsibility and risk management. J Corp Citizensh 22((Summer)):39–51 Stulz RM (1996) Rethinking risk management. J Appl Corp Financ 9(3):8–25

The Double Relationship Between Risk Management …

111

Susanto AB (2009) Reputation-driven corporate social responsibility Pendekatan strategic management dalam CSR. Erlangga, Jakarta Unerman J (2008) Strategic reputation risk management and corporate social responsibility reporting. Account Auditing Accountability J 21(3):362–364. https://doi.org/10.1108/ 09513570810863941

Are Corporate Financing Policies Different in Old and New EU Member States? Julia Koralun-Bere´znicka

Abstract The research aim is to verify whether and how corporate capital structure and its determinants vary between the old and new EU member states. The empirical research based on the BACH-ESD database provided by the European Commission covers private firms from 12 countries in the period 2000–2017. Apart from considering the length of EU membership, a number of firm-specific determinants and industry features are captured in order to identify the main differences in corporate financing patterns. The methods employed include analysis of variance as well as panel data modelling performed on the two subgroups of countries. Findings reveal that although corporate financing patterns differ significantly across the compared groups, with companies in old member countries more heavily dependent on debt, the determinants of corporate financing choices are less varied and provide more support for the pecking order theory. The results also indicate that the relevant importance of the country, industry, and firm size effects depends on the length of EU membership.

1 Introduction Although there have been several enlargements of the EU since its foundation, it is still common to distinguish between the so called old and new member states (MS). This is probably the result of the particularly spectacular Eastern enlargement in 2004, which had a significant and symbolic meaning, as it opened the integration of the Western countries with the Eastern and Central Europe by including the largest number of countries and people at a time. Given the now 15 years’ long membership period of the countries accessing in 2004, a question arises whether the aforementioned distinction is still justified. The recently lively discussion on the future of the EU, fuelled by Brexit, makes the problem even more valid. The aim of this paper is to analyse the differences between the old and new MS in terms of corporate financing policies and the importance of factors affecting these decisions. J. Koralun-Bere´znicka (B) University of Gda´nsk, Gda´nsk, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_10

113

114

J. Koralun-Bere´znicka

As indicated in the literature review section, financing dilemmas have been an issue of obvious interest to both business practitioners and academics, who produced a virtually uncountable number of research studies aimed at identifying the factors underlying capital structure decisions. By comparing corporate financing policies and capital structure determinants across the old and new EU member countries, the study contributes to the existing literature in several ways. Firstly, it updates the knowledge on the extent of the gap remaining between the two categories of MS in terms of corporate financing. Secondly, it provides new insights into the capital structure determinants by taking into account the length of EU membership not only as a factor directly affecting capital structure, but also as an indirect factor which may entail the diversified impact of primary factors on financial leverage within each of the two groups of countries. Finally, this study includes private enterprises instead of the more commonly investigated public company data.

2 Literature Review The relevance of the country effect in capital structure has been reported by a number of studies, as summarised e.g. by Venanzi (2017), and is variously attributed to such country-specific features as economic development (Demirgüç-Kunt and Maksimovic 1999), legal environment (La Porta et al. 1997; Joeveer 2006), institutional environment (Rajan and Zingales 1995; Graham and Harvey 2001; Bancel and Mittoo 2004; Brounen et al. 2006), the financial sector development (Fan et al. 2012; Graham et al. 2015), creditors’ protection (Hall et al. 2004), or GDP growth (De Jong et al. 2008). Although the occurrence of country specifics is considered as a state of the art, two approaches to explaining country effect in capital structure can be recognised in corporate finance literature. According to the first one, due to the substantial country-level differences in such areas as corporate taxation, legal system or macroeconomic situation, determinants of corporate debt tend to be verified with reference to a given country. This view is supported by multiple studies, where the significance of capital structure factors is verified predominantly for a single ˇ country solely—(López-Gracia and Sogorb-Mira 2008; Crnigoj and Mramor 2009; Degryse et al. 2012; Sayilgan et al. 2011; Pepur et al. 2016). In the case of an international sample leverage determinants are analysed in a country-by-country manner (Rajan and Zingales 1995; Mokhova and Zinecker 2013). In the other approach cross-country samples are investigated as a whole rather than as individual countries (Demirgüç-Kunt and Maksimovic 1999; Joeveer 2006), without necessarily disregarding the country effect, which in this case is often captured by the inclusion of country dummy variables (Antoniou et al. 2008; Aggarwal and Kyaw 2009; Alves and Ferreira 2011; Acedo-Ramirez and Ruiz-Cabestre 2014). In this study yet another approach is adopted, where the analysed group of countries is divided into two supposedly homogeneous categories according to their EU membership length. The comparisons are then performed across the two categories of the old and new MS. However, following the recent advice by Venanzi (2017), who questions

Are Corporate Financing Policies Different in Old …

115

Table 1 Theoretical predictions on the selected determinants of capital structure Factor

Theory, expected sign of the relation with debt, example supporting studies POT

TOT

Collateral

−

(8), (9)

+

Liquidity

−

(8), (9)

+

(8), (9)

Growth

+

(7), (8), (9),

−

(4), (8), (9)

+

(7), (8), (9)

(3), (4), (8), (9)

Profitability

−

(2), (8), (9)

Working capital

−

(6)

Risk

+

(7), (10)

−

(7), (8), (9)

Non-debt tax shields

+

(1), (5)

−

(7), (8), (9)

Firm size

−

(8), (9)

+

(3), (7), (8), (9)

(1) Bradley et al. (1984), (2) Myers and Majluf (1984), (3) Titman and Wessels (1988), (4) Rajan and Zingales (1995), (5) Graham (2003), (6) Chiou et al. (2006), (7) Frank and Goyal (2009), (8) Hussain and Hussain (2015), (9) Pepur et al. (2016), (10) Sibindi (2016)

any cross-country homogeneity, it seems purposeful to capture the within-category country effect by including country dummy variables in each case. Corporate financing policies are explained by a number of competing theories. The static trade-off theory (TOT) by Modigliani and Miller (1958) and the peckingorder theory (POT) by Myers and Majluf (1984) are the two leading theoretical explanations of debt choices and at the same time considered the most appropriate in the context of SMEs. A summary of the selected, most commonly verified factors affecting capital structure along with the theoretically expected sign of their relation with leverage based on literature review is shown in Table 1. Taking into account the two main theoretical concepts of capital structure as well as empirical findings from previous studies, the following research hypotheses are put: H1: the level of financial leverage is significantly different for old and new MS; H2: the significance and (or) direction of impact of firm-specific variables and industry on debt vary across the two groups of countries; H3: the relative importance of the country, size and industry effect in capital structure depends on the length of EU membership. Verification of these hypotheses, based on reliable European data source, would add to the hitherto literature by further investigation of the direct and indirect impact of the length of EU membership.

3 Data and Methodology This paper investigates the capital structure diversity and its determinants across the old and new EU members (9 and 3 countries, respectively) using the BACH-ESD database provided by the European Commission (de France 2018). The database

116

J. Koralun-Bere´znicka

contains harmonised financial information aggregated by country, industry and size for the following EU countries: Austria, Belgium, Czech Republic, Germany, Denmark, Spain, France, Italy, Luxemburg, Poland, Portugal, and Slovakia. Therefore, the sample is limited only to these MS, which is a trade-off between the data range and its comparability. The analysis includes companies of three size groups: small, medium, and large in the twelve countries and in sixteen industries according to the NACE classification: A, B, C, D, E, F, G, H, I, J, L, N, P, Q, R, S. The time span covers yearly data from 2000 to 2017. Including the four-years’ period before the eastern enlargement was meant to extend the time series and cover the pre-accession period, during which the assumed differences in corporate financing might have also been present. The dependent variable is the median of the debt to assets ratio (D/A). The variables employed as explanatory variables are defined in Table 2. Certainly, the range of variables does not consider all the potential factors affecting capital structure, which are virtually countless. The choice of variables was limited to those most commonly explored in studies from the field. The first stage of the analysis was aimed at verifying whether the differences in capital structure between old and new MS are significantly different. For this purpose, the one-way analysis of variance (ANOVA) was employed with the total Table 2 Construction of explanatory variables

Variable

Definition

Collateral (COL)

Tangible fixed assets/total assets (Q2)

Liquidity (LIQ)

Cash and bank/total assets (Q2)

Growth (GRT)

(Total assetst —total assetst-1 )/total assetst-1

Profitability (PRF)

Net profit or loss for the period/equity (Q2)

Risk (RSK)

(EATt −EATt-1 )/EATt-1

Non-debt tax shields (NTS)

Depreciation on fixed assets/net turnover (Q2)

Working capital (WCR)

Operating working capital/net turnover (Q2)

Sale size (SAL)

Logarithm of net turnover

Industry dummies (D_IND)

A, B, C, D, E, F, G, H, I, J, L, N, P, Q, R, S

EU membership dummy (D_EU)

OLD, NEW

Country dummies (D_CT)

AT, BE, CZa , DE, ES, FR, IT, LU, PL, PT, SKa

Size (D_SIZE)

S, M, L

EAT—earnings after taxes, industry symbols as in NACE, a In case of missing data for Q2 the mean of ratios was used instead

Are Corporate Financing Policies Different in Old …

117

debt ratio as the dependent variable and the length of EU membership (long vs. short) as the grouping factor. Then, to verify the significance of the determinants and the hypothesised effects, panel data regression models were estimated, which can be specified by formula (1) for the whole set of countries: (D/A)icst = α + β1 C O L icst + β2 L I Q icst + β3 G RTicst + β4 P R Ficst + β5 RS K icst + β6 N T Sicst + β7 W C Ricst + β8 S AL icst + β9 D_EUicst + β10 D_S I Z E icst + β11 D_I N Dicst + ξicst ,

(1)

and by (2) for the two groups of old or new MS, estimated separately: (D/A)icst = α + β1 C O L icst + β2 L I Q icst + β3 G RTicst + β4 P R Ficst + β5 RS K icst + β6 N T Sicst + β7 W C Ricst + β8 S AL icst + β9 D_C Ticst + β10 D_S I Z E icst + β11 D_I N Dicst + ξicst , (2) where i refers to the industrial section, c—to country, s—to size group and t—to the time period. As can be seen from the formulas, they only differ by one variable: in the case of model (1) the dummy variable for EU membership was included (D_EU), whereas in the case of model (2), the length of EU membership was replaced by country dummies (D_CT). The model specification is based on the fixed effects model defined by Baltagi (2008). Both models were estimated by OLS with standard errors robust for heteroscedasticity and autocorrelation. In order to evaluate the importance of the effects of industry, size, and country or length of EU membership the Akaike’s information criterion (AIC) was used for models estimated without the given effect.

4 Results The results of ANOVA with the length of the EU membership as the grouping factor performed on the whole sample, i.e. for all countries, industries, size groups and years indicate the statistically significant differences in financing patterns between the old and new EU MS, as shown by the value of the F-statistics equal to 1320.0 (p = 0.000). The basic statistics shown in Table 3 inform about the generally lower level of leverage in new MS compared to old MS. Table 3 Basic statistics of total debt ratio Countries

N

Mean

Std. dev.

Std. error

All

6416

0.649

0.130

0.002

Old MS

5641

0.668

0.115

0.002

New MS

775

0.504

0.140

0.005

118

J. Koralun-Bere´znicka

The differences in debt level proved statistically significant also when the ANOVA was applied separately on each size group of firms, each industrial section and each year, which indicates that the lower reliance on debt of companies in new MS is considerable and persistent. These results to some extent resemble previous findings (Jensen and Uhl 2008) on the existence of SME financing gap in transition countries. The authors attribute this gap to a lack of institutional development, affecting credit availability. The estimation results for the panel regression models specified by Formulas (1) and (2) are shown in Table 4. Model (2) was also estimated for all 12 countries together, although the results are not reported here. In each case the Hausman specification test indicated the appropriateness of the fixed effects model. The comparison of the impact of the firm-specific variables on debt reveals rather only minor differences between the old and new MS in terms of both their significance and direction. As for the sign of the ratios’ impact, only the working capital and growth ratio appear to affect leverage differently between the compared groups. However, the sign of the growth variable should not even be considered due to the insignificance of this variable, as opposed to the working capital, which appears positive but insignificant for old countries, whereas significantly negative for the new ones. Other variables do not differ in terms of direction, although their significance does vary across the two groups. Apart from the aforementioned growth ratio, which proved insignificant in both groups, the only variable of the same significance is the ratio of logarithmized sales. This proxy of firm size proved significantly negative in both groups of countries. The significance of all the other variables is different for old and new MS. Certainly, these result do not indicate that the differences between explanatory variables across old and new MS are significant, which would need to be confirmed by further tests, not employed in this study. The joint significance test for dummy variables of size indicate the relevance of this effect in all four estimations, i.e. in model (1), in model (2) for old MS and for new MS, as well as in model (2) for all countries (not reported in Table 4). Similarly, the country effect proved significant in the same estimations, except for model (2) for new MS. However, the industry effect appeared significant only in the case of old MS. The importance of the effects can also be evaluated with the use of one of the information criterion values, e.g. as in this case the Akaike’s information criterion (AIC), whose values are shown in Table 5. They indicate that the omission of industry dummies in model (1) results in the largest decrease of the model’s explanatory power, as illustrated by the increase of the AIC value. The effect of the length of EU membership comes next in terms of its contribution to the explanatory power of model (1), whereas the size effect appears the weakest. Based on the analysis of AIC values the order of importance of the compared effects is the same for old countries and for all countries in the sample: the country effect is followed by the industry effect and then by the size effect. However, when prioritising the effects for new MS only, these are the industry-specific features which matter most, followed by the size effect and finally by the country specifics. However, the relatively low importance of the country effect in the new MS can be attributed to the small number of countries taken into account.

Are Corporate Financing Policies Different in Old …

119

Table 4 Estimation results of panel regressions for D/A Variable Const.

Model (1) (all countries)

Model (2) (old MS)

Model (2) (new MS)

Estimate

Estimate

Estimate

Std. error

0.780***

Std. error

Std. error

0.604***

0.058

0.064

0.924***

0.150

COL

−0.242***

0.043

−0.030

0.041

−0.478***

0.077

LIQ

−0.447***

0.139

−0.678***

0.181

−0.161

0.215

PRF

0.059

0.073

0.317***

0.064

0.016

0.033

DPR

0.432***

0.165

0.169

0.154

1.480***

0.319

−0.178***

0.064

0.043

0.057

−0.544***

0.116

WCR GRT

0.002

0.004

0.000

0.003

0.003

0.009

RSK

−0.001*

0.001

0.000

0.000

−0.005*

0.003

SAL

−0.001

0.008

−0.024**

0.010

−0.049**

0.023

M

0.031***

0.010

0.015

0.009

0.058***

0.015

L

0.028**

0.013

0.032***

0.011

0.059**

0.025

EU_OLD

0.118***

0.012

0.007

0.025

0.048**

0.022

−0.045***

BE DE

0.031**

ES

−0.072***

FR

0.007

IT LU

0.114***

0.016

−0.140***

0.025

−0.039**

0.016

PL PT SKa B

−0.046*

0.026

C

−0.010

0.025

−0.088*** 0.015

0.025

0.139***

0.039

D

0.023

0.028

0.009

0.027

0.013

0.040

−0.031

0.023

−0.016

0.041

E

0.010

0.025

0.017

0.018

F

0.086***

0.025

0.124***

0.022

0.126***

0.035

G

0.040

0.026

0.097***

0.025

0.173***

0.040

H

0.059**

0.025

0.092***

0.021

0.108**

0.045

I

0.046*

0.026

0.058***

0.021

0.099

0.062

0.026

0.025

0.022

−0.082

0.052

−0.039

J

−0.029

0.042

L

0.031

0.032

0.024

0.027

N

0.072***

0.025

0.104***

0.021

0.093**

0.046

0.052

P

−0.020

0.031

0.024

0.028

−0.165**

0.083

Q

−0.063**

0.026

−0.044*

0.025

−0.028

0.051

R

−0.019

0.031

0.016

0.027

−0.080

0.055 (continued)

120

J. Koralun-Bere´znicka

Table 4 (continued) Variable S

Model (1) (all countries)

Model (2) (old MS)

Model (2) (new MS)

Estimate

Estimate

Std. error

Estimate

Std. error

0.021

−0.034

0.053

Std. error

0.006

0.025

0.012

No. obs.

5438

4520

918

R2

0.458

0.565

0.662

0.456

0.562

0.651

140.9 [0.000]

86.61 [0.000]

60.93 [0.000]

2.783 [0.006]

2.482 [0.014]

3.474 [0.001]

−1.670 [0.096]

1.636 [0.104]

1.790 [0.074]

0.492 [0.624]

Adj.

R2

Hausman test Joint significance size country industry

0.607 [0.544]

Notes Due to the missing data for profitability ratio in DK it was not included in the model; *Significant at the 10% level, **5%, ***1%

Table 5 Values of Akaike’s information criterion for models explaining D/A Countries, (model)

Omitted effect

All (1)

−9847.8

All (2)

−11,851.3

−9151.4

Old MS (2)

−10,897.9

New MS (2)

−1846.5

Nonea

Countryb

Industryc

Sized

EU membershipe

−9072.7

−9759.3

−9151.4

−10,574.4

−11,726.0

−9057.5

−9552.9

−10,818.8

−1826.3

−1550.7

−1776.6

a All effects included (size, industry, country or length of EU membership); b Size and industry effect

included, country or length of EU membership effect omitted; c Size and country or length of EU membership effect included; d Industry and country or length of EU membership effect included; e Industry and size effect included

5 Conclusions The study finds evidence for the persistence of significant differences in financial leverage between the old and new EU member countries. The share of debt in the new MS is considerably lower than in old MS and the differences remain significant across all size groups, industrial sections and throughout the whole period. This conclusion provides support for the research hypothesis H1 and is in line with other studies in the field (Jensen and Uhl 2008; K˛edzior 2012).

Are Corporate Financing Policies Different in Old …

121

As for the comparison of the significance and direction of the firm-specific variables on debt ratio between the old and new MS, it appears that although their impact is far from identical, it is difficult to distinguish any clear differences in terms of the importance of factors affecting leverage. The sign of the relation between debt and the majority of variables was found the same for all but two factors, namely working capital ratio and firm growth. More differences, however, were identified in terms of significance. The most common pattern observed here is the situation, where a given factor has the same direction for old and new MS, but different significance, as in the case of assets tangibility, financial liquidity, profitability, non-debt tax shields, risk and firm size. The occurrence of such differences provides only weak support for hypothesis H2. Consequently, it is difficult to attribute either of the two leading capital structure theories to the analysed groups of countries. Although the relation between some variables and debt implies the adequacy of the static trade-off theory, as in the case of the significantly positive profitability–leverage relation in old MS, generally the results of this study indicate the relevance of the pecking order predictions for both old and new MS. With reference to last hypothesis (H3) concerning the relative importance of the country, industry and size effect in capital structure depending on the length of membership period, it can be concluded that the prioritisation of these effects differs across the old and new MS. Contrary perhaps to the common sense expectations that the old EU member countries should be more homogeneous than the new ones and that therefore the country-specific features should be less pronounced within this group of countries, the country specifics was found to prevail over the industry effect here, whereas the size effect was of the lowest importance. In the other group, i.e. the new MS, it was the industry effect which proved most important in terms of its influence on corporate debt. The industry effect was followed by the size effect, whereas the country specifics proved to be the least relevant. These findings provide support for H3. Summarising, this study highlights mainly the relevance of the country effect in corporate capital structure, whose both direct and indirect effect appears not to be irrelevant even in the supposedly more harmonised old EU members. It also indicates that although the level of indebtedness differs significantly depending on the length of EU membership, the factors explaining corporate financing decision are less varied. The main limitation of the study comes from the compromise between extending data range and maintaining its harmonisation, which results in covering only about 43% of the EU countries. The study could be further extended by covering a larger number of MS, especially newer members, as well as by including different debt measures, which is left for future research.

122

J. Koralun-Bere´znicka

References Acedo-Ramirez MA, Ruiz-Cabestre FJ (2014) Determinants of capital structure: United Kingdom versus continental European countries. J Int Finance Manage Account 25(3):237–270 Aggarwal R, Kyaw NNA (2009) International variations in transparency and capital structure: evidence from European firms. J Int Finance Manage Account 20(1):1–34 Alves PFP, Ferreira MA (2011) Capital structure and law around the world. J Multinat Finance Manage 21:119–150 Antoniou A, Guney Y, Paudyal K (2008) The determinants of capital structure: capital marketoriented versus bank-oriented institutions. J Finance Quant Anal 43(1):59–92 Baltagi BH (2008) Econometric analysis of panel data. Wiley, Chichester Bancel F, Mittoo UR (2004) Cross-country determinants of capital structure choice: a survey of European firms. Financ Manage 33(4):103–132 de France B (2018) Bank for the accounts of companies harmonised—European sectoral references database. https://www.bachesd.banque-france.fr. Accessed 13 Apr 2017 Bradley M, Jarrell G, Kim E (1984) On the existence of an optimal capital structure. J Finance 39(3):857–878 Brounen D, De Jong A, Koedijk K (2006) Capital structure policies in Europe: survey evidence. J Bank Finance 30(9):1409–1442 Chiou J, Cheng L, Wu H (2006) The determinants of working capital management. J Am Acad Bus 10(1):149–155 ˇ Crnigoj M, Mramor D (2009) Determinants of capital structure in emerging European economies: evidence from Slovenian firms. Emerg Markets Finance Trade 45:72–89 Degryse H, de Goeij P, Kappert P (2012) The impact of firm and industry characteristics on small firms’ capital structure. Small Bus Econ 38:431–447 Demirgüç-Kunt A, Maksimovic V (1999) Institutions, financial markets and firm debt maturity. J Finance Econ 54(3):295–336 Fan JPH, Twite GJ, Titman S (2012) An International comparison of capital structure and debt maturity choices. J Finance Quant Anal 47(1):23–56 Frank M, Goyal V (2009) Capital structure decisions: which factors are reliably important? Finance Manage 38(1):1–37 Graham JR (2003) Taxes and corporate finance. https://ssrn.com/abstract=264516 Graham JR, Harvey C (2001) The theory and practice of corporate finance: evidence from the field. J Finance Econ 60(2–3):187–243 Graham JR, Leary MT, Roberts MR (2015) A century of capital structure: the leveraging of corporate America. J Finance Econ 118(3):658–683 Hall GC, Hutchinson PJ, Michaelas N (2004) Determinants of the capital structures of European SMEs. J Bus Finance Account 31(5–6):711–728 Hussain MI, Hussain MA (2015) Determinants of capital structure and testing of theories: a study on the listed manufacturing companies in Bangladesh. Int J Econ Finance 7(4):176–190 Jensen NS, Uhl FT (2008) Capital structure in European SMEs. An analysis of firm- and country specific variables in determining leverage. Aarhus School of Business, University of Aarhus. https://pure.au.dk/portal/files/3244/Capital_Structure_in_European_SMEs.pdf Joeveer K (2006) Sources of capital structure: evidence from transition countries. CERGE-EI working papers wp306. The Center for Economic Research and Graduate Education—Economics Institute, Prague de Jong A, Kabir R, Nguyen T (2008) Capital structure around the world: the roles of firm- and country-specific determinants. J Bank Finance 32(9):1954–1969 K˛edzior M (2012) Capital structure in EU selected countries—micro and macro determinants. Arg Oecon 1(28):69–117 La Porta R, Lopez-de-Silanes F, Shleifer A, Vishny R (1997) Legal determinants of external Finance. J Finance 52(3):1131–1150

Are Corporate Financing Policies Different in Old …

123

López-Gracia J, Sogorb-Mira F (2008) Testing trade-off and pecking order theories financing SMEs. Small Bus Econ 31(2):117–136 Modigliani F, Miller MH (1958) The cost of capital, corporation finance, and the theory of investment. Am Econ Rev 48(3):261–275 Mokhova N, Zinecker M (2013) The determinants of capital structure: the evidence from the European union. Acta Universitatis Universitatis Agriculturae et Silviculturae Mendelianae Brunensis 61(7):2533–2546 Myers SC, Majluf N (1984) Corporate financing and investment decisions when firms have information that investors do not have. J Finance Econ 13(2):187–221 ´ Pepur S, Curak M, Poposki K (2016) Corporate capital structure: the case of large Croatian companies. Econ Res - Ekonomska Istraživanja 29(1):498–514 Rajan RG, Zingales L (1995) What do we know about capital structure? some evidence from international data. J Finance 50(5):1421–1460 Sayilgan G, Karabacak H, Kuçukkocaooglu G (2011) The firm-specific determinants of corporate capital structure: evidence from Turkish panel data. Int J Econ Financ Issues 1:1–11 Sibindi AB (2016) Determinants of capital structure: a literature review. Risk Gov Control Financ Markets Inst 6(4):228–237 Titman S, Wessels R (1988) The determinants of capital structure. J Finance 43(1):1–19 Venanzi D (2017) How country affects the capital structure choice: literature review and criticism. Int J Econ Finance 9(4):1–11

Board Characteristics and Performance of East Africa Companies Dorika Jeremiah Mwamtambulo

Abstract In the management of agency problems, the board of directors is an essential tool for monitoring the activities of the managers. The board of directors has specific characteristics associated with its effectiveness and efficiency, which can be transformed into the general performance of the company. This study aimed at providing evidence regarding the characteristics possessed by the East Africa community companies’ board of directors and their influence on the companies performance. Simultaneous system of equations and annual data from 16 companies for a period between 2003 and 2017 was used in the study. The study observed that the market performance measures to be highly influenced by the changes in the characteristics of the board such as board size, the proportion of women, proportion of independent directors and proportion of foreign directors. On the other hand, the account performance measures were less influenced by the board characteristics, and only ROE was more associated with the changes on board characteristics. During this period, the market was more overwhelmed by new information regarding the board characteristics as compared to the performance of the account measures.

1 Introduction The question as to whether the characteristics of the board or on how well the board is diversified is having a significant influence on the performance of the company has been a point of interest in many studies. These include the research studies of Dwyer et al. (2003), Rose (2007), Amran (2011), Dobbin and Jung (2011), Ujunwa et al. (2012), Abdul Latif et al. (2013), Lückerath-Rovers (2013), Rose et al. (2013), Wellalage and Locke (2013), Darmadi (2013), Abidin et al. (2014), Ilaboya and Obaretin (2015), Molenkamp (2015), Vishwakarma and Kumar (2015), Ararat et al. (2015), Hidayat and Utama (2016), Shehata et al. (2017), Conyon and He (2017), Gordini and Rancati (2017), Abdallah et al. (2018). While the board characteristics D. J. Mwamtambulo (B) Wroclaw University of Economics and Business, Wroclaw, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_11

125

126

D. J. Mwamtambulo

include factors such board size, the duality of the Chief Executive Officer (CEO), the proportion of external directors, diversity on the other hand look on things such as the inclusion of women and foreigners in the board. The reasons as to why the characteristics of the board are of interest can be explained by understanding the company’s agency problems and the management process of the agency problems. Agency problems in a company arise when there is a separation of powers between the owners and management (Cornett et al. 2008; Bodie et al. 2009, 2014; Berk et al. 2012; Berk and DeMarzo 2014; Cottrell 2016). Owners are faced with the problem of information asymmetry, the difference in interest regarding the direction the company ought to take and differences in expectations with the managers (Shapiro 2005). Information asymmetry is a situation where the managers have more information regarding the company than the owners. As not all the information regarding the company is made public by the managers and the owners cannot request for a tailormade one, this problem continues to exist. Owners are forced to find other means of obtaining reliable information regarding the company. This, however, comes with an additional cost to the owners. The difference in interest, on the other hand, is the problem where the manager act on their self-interest rather than the interest of the owners of the company. On the other hand, difference in expectation is just that sometimes the actions of the managers might be not good enough to the expectations of the owners. On dealing with the agency problem, the owners incur what is known as an agency cost. A cost that the owners must pay to ensure that the manager’s interests are aligned with that of theirs. As addressed by Meckling and Jensen (1976), the cost can be in the form of monitoring cost, bonding and residual loss. This includes the act of aligning the managers’ incentives with the interest of the owners, that is aligning their compensations with their performance (Ederer and Manso 2008; Lawler et al. 2012; Masulis and Reza 2015). Researches have shown that several problems arise when management compensations are aligned with performance. An issue of backdating of options (Yermack 1995, 1997; Lie 2005; Burns and Kedia 2006; Heron and Lie 2007, 2009; Adam and Schwartz 2009; Collins et al. 2009; Veld and Wu 2014; Biggerstaff et al. 2015; Devos et al. 2015; Tee and Wiley 2018), underestimation of financial statement in the accounts scandals of Enron and WorldCom (Lyke and Jickling 2002; Kaplan and Kiron 2007; Scharff 2007; Giroux 2008; Weber Willenborg and Zhang 2008; Lal Bhasin 2013). The board of directors is a system of monitoring created by owners to mitigate the agency problem. The board of directors serve the responsibilities of the owners. Some of these responsibilities include the nomination of the management and their compensation settings, discipline and dismissal of managers, decide on significant investments to be undertaken by the company. In a country like the USA, the board of directors have a fiduciary duty to the owners. The significant of the board of directors was raised after the collapse of many companies in Europe and the USA. Many recommendations and regulation were set regarding the board of directors as a result. The Cadbury Commission Recommendation (1992) was a result of the problem in the UK in 1991. Sarbanes-Oxley Act (2002) was passed after the collapse of Enron and WorldCom in the USA, Dodd-Frank Act (2010) was set after the financial crisis.

Board Characteristics and Performance of East Africa Companies

127

They emphasised on the need of the board of directors to be firstly independent. This includes the requirement of the majority of the board members to becoming outside of the company. This is in order to avoid a situation of a ‘captured board’, more so there is a need for the audit and the compensation committee to be made of all independent directors if not the majority. They point out the necessity of nonexistence of duality, where the CEO should not hold both the title of the managing director and the chairman of the board. The need for constant rotation of the external auditor and most significantly the obligation of the external auditors to disclose the percentage of non-audit work performed with a company. Both the recommendation and rules and regulations imply for a well-performing board of directors that eliminate the agency problem. According to Berk and DeMarzo (2014), a well-performing board is an investment with a positive net present value to the company. In light of the above requirements, recent literature has pointed out other desirable characteristics that are possessed by a good function board. They include the size of the board, where a small sized board has proved to be efficient (Yermack 1996; Dalton et al. 1998; Haniffa and Hudaib 2006; Guest 2009; Ujunwa et al. 2012; Zainal et al. 2013; Abidin et al. 2014; Appiadjei et al. 2017). Inclusion of women in the board (Campbell and Vera 2009; Boulouta 2013; Oba and Fodio 2013; Taghizadeh and Saremi 2013; Lückerath-Rovers 2013; Peni 2014; Ararat et al. 2015; Molenkamp 2015; Willows and Van Der Linde 2016; Kim and Starks 2016; Rossi et al. 2017; Shehata et al. 2017; Chan et al. 2017; Conyon and He 2017; Gordini and Rancati 2017; Haque 2017; Sarpong-Danquah et al. 2018; Green and Homroy 2018), participation of foreigners and members from different ethnicity including minority (Marimuthu and Kolandaisamy 2009; Darmadi 2011; Jhunjhunwala and Mishra 2012; Ujunwa 2012; Wellalage and Locke 2013). After their Independence, all member states of the East Africa Community strived to rebuild their economy. This is reflected by growth in their Gross Domestic Product (GDP) over the years and more so in the recent years were more than +5% growth in the GDP had been observed (Mensah 2016). Most of the member states are expected to move to middle-income countries by 2025. The countries have also increased participation of private and foreign individuals through opening their doors to free trade. Moreover, the countries have worked towards increasing women participation as reflected by their high ranking in gender and equality index. Companies operating in this market have experienced a tremendous amount of growth over the years, and so is their freedom to trade in all the East Africa capital markets. Their trading volumes in the markets have significantly contributed to the growth of GDP. For example, by 2016 the Nairobi Stock Exchange contributed to 36% of the GDP while the Dar es Salaam and Kampala Stock Exchange contributed for about 38% and 41% respectively (Citizen 2016; Delloite 2016; EAC 2019). Due to their growth and significant contribution to economic growth, there is always a question regarding their sustainability and the management of the agency problems. Even though these companies have implemented all the requirements needed for a proper and efficient board functioning, nevertheless no evidence has accounted regarding the relationship between these board characteristics to the performance. This study aimed at providing new evidence in this area. The significance of this study is on giving new

128

D. J. Mwamtambulo

literature regarding the management of agency problems by East Africa member community through the board of directors’ characteristics and on how they influence the performance of the companies. Moreover, it provides an outlook for the companies trading in these markets and opportunities to strategies in increasing their performance and thus operating a sustainable business.

2 Literature Review Literature has pointed out the following characteristics of the board to have a significant effect on the performance of the company.

2.1 Board Size Board size is measured by the number of directors in the company’s board. The efficient of the board is measured on its ability to making clear and objective goals, useful strategies and effective decisions that largely transform the company. The overall outcome of this is regulated by the performance of the company. It is argued that small board size is always associated with better performance. This is associated with time used to pass a decision on the board, where it is far shorter in a small-sized board compared to in a large-sized board. This can be seen in the findings of Yermack, (1996), Dalton et al. (1998), Haniffa and Hudaib (2006), Abidin et al. (2014), Guest (2009), Ujunwa et al. (2012), Zainal et al. (2013), Rose et al. (2013) and Berk and DeMarzo (2014), Appiadjei et al. (2017) were large board size negatively affected the company’s performance. Similar findings were observed by Alabebde (2016) in the UK and Johl et al. (2015) in Malaysia. Evidence has also shown that large board is also associated with better performances. By increasing the number of individuals with different backgrounds and experience in its board, the company allows the creation of new ideas and methods that help to transform the company. This was supported by Ahmadi et al. (2018) who observed a positive relationship between the French CAC 40 companies board size to the factor of ROE. In their study, Ahmadi et al. (2018) found no such relationship with ROA. Similarly, Rashid (2018) in Bangladesh observe a positive correlation between the board size and firm performance and that of Vishwakarma and Kumar (2015) for India top IT firms Abdallah et al. (2018), in Jordan, Kyereboah-Coleman and Biekpe (2006) in Ghana, Johl et al. (2013), and Shukeri et al. (2012), Abdullah and Ismail (2013) in Malaysia, Jackling and Johl (2009) in India, Gaur et al. (2015) and in New Zealand, Haque (2017) in the UK. Chen and Al-Najjar (2012) also observed that the size of the board to be influenced by the existence of supervisory board and managerial ownership. While the studies of Yammeesri and Herath (2010), Müller (2014), Sarpong-Danquah et al. (2018) observe no relationship between the performance and the board size. Darmadi

Board Characteristics and Performance of East Africa Companies

129

(2011) found rather than size, the proportion of the young board members positively and significantly influence the performance of Indonesia firms. Hidayat and Utama (2016) observed a U shaped, non-linear relationship between the board size and the performance of Indonesia firms. In this study a positive relationship between the board size and the company performance was examined. The hypothesis tested in the study is stated below. Hypothesis 1 A small board size is associated with a positive performance of the company.

2.2 The Proportion of Independent Directors The board is made of three types of directors. The inside directors, grey directors and outside/independent directors (Berk and DeMarzo 2014). Inside directors are directors that are directly involved with the management and daily operations of the company. They include the chief executive officer, managers and other employees of the company. Grey directors are those who are not directly involved with the company, but they have a significant interest in the firm. They include directors from financial institutions, creditors, debtors and employees of the parents’ company. Outside directors are directors with no interest in the company; hence, they act independently. Both the Cadbury 1992 and Sarbanes-Oxley acts require the board of directors to be made mostly by independent directors for better performance of the company. This can be seen in the studies of Alabebde (2016) for listed Britain companies observed the proportion of outside directors positively and significantly affected the performance of the companies. Müller (2014) also found a positive relationship within the UK FTSE 100 and Weir et al. (2002) in the UK listed companies. Other include that of Heenetigala and Armstrong (2011) in Sri-Lanka, Anderson and Reeb (2004) for USA S&P 500, Jackling and Johl (2009) in India, Liu et al. (2015), Weisbach and Hermalin (2003), Rashid (2009) in Bangladesh, Amran (2011) in Malaysia, Ilaboya and Obaretin (2015) in Nigeria, Vishwakarma and Kumar (2015), in India, Hidayat and Utama (2016) in Indonesia, Sarpong-Danquah et al. (2018), in Ghana, and Ahmadi et al. (2018) in France, have shown that the high number of independent directors in the board plays a significant role in the general performance of the company. On the other hand, Guest (2009) in UK for a period between 1981 and 2002, Taghizadeh and Saremi (2013), Shukeri et al. (2012) in Malaysia, Yammeesri and Herath (2010) in Thailand, Eulerich et al. (2014) in Germany found a negative relationship between the board independence and the performance of the company. In the study of Yammeesri and Herath (2010) and Gaur et al. (2015) rather than the external board, the internal board significantly influence the performance of New Zealand firms. While Bhagat and Black (1998) in the US firm in 1991 Leung et al. (2014), in Hong Kong between 2005 and 2006, Rashid (2018) in Bangladesh, Taghizadeh

130

D. J. Mwamtambulo

and Saremi (2013), Johl et al. (2013) in Malaysia and Terjesen et al. (2016) in 47 countries found no relationship whatsoever between board independent and the performance of the company. But in the study of Rashid (2018) in the same regions observed a positive correlation between the size and the independence of the board. The board independence not only affects performance it also affects risk management as was noted by Ahmad et al. (2015) in Malaysia. This study examined on the positive influence of the large proportion of independent directors in the board as stated in the hypothesis below. Hypothesis 2 Large proportion of external directors is positively associated with the performance of the company.

2.3 Separation of Chairman and CEO Positions On avoiding the problem of a captured board, there is a need to separate the position of the board chairman and CEO. At any given time, the CEO should not hold both the position of the board chairman and that of the managing director. Much of the literature has argued for duality, this includes that of Brickley et al. (1997) in the USA, Haniffa and Hudaib (2006) in Malaysia Jackling and Johl (2009), in India, Peni (2014) in the US Gaur et al. (2015), in New Zealand, Ahmadi et al. (2018), in France and Abdallah et al. (2018) in Jordan have observed evidence to support the duality to be positively related to the performance of the firm. Brickley et al. (1997) argued by separating the position of chairperson from the CEO, the company incur new additional costs such as the cost of monitoring the new board chairman, the cost of information asymmetry between the chairperson and the CEO, the cost to pinpoint the person to be accountable during crisis, the cost of training new CEO in case of a change in management and the new CEO will feel much as an outcast in the board compared if the predecessor was also the chairperson of the board. But in the findings of Duru et al. (2016) in USA, Kyereboah-Coleman and Biekpe (2006) in Ghana, Ehikioya (2009), Ujunwa (2012) in Nigeria supported the need of separation of these position as this was overseen by an increase in the company’s performance. While Shukeri et al. (2012) did not observe any tied up relationship between separating these position to the performance of the firm. But there is still argument on how busy the CEO-Chairman or the members of the board should be. Haniffa and Hudaib (2006) observed a positive relationship on how active the directors are to the performance of the company. Abdul Latif et al. (2013) observed a non-statistical significant positive relationship while Peni (2014) argue against the CEO-Chairman multiple directorships which lead to low performance. In this study, it was anticipated that absence of duality in the board of directors is associated with better performance. This is provided for hypothesis 3 below. Hypothesis 3 Separation of CEO and Chairman Position is positively associated with the general performance of the company.

Board Characteristics and Performance of East Africa Companies

131

2.4 Proportion of Women In many countries, the percentage of women on the board has been low. Over the years, many factors have accounted for this problem. These include that women are not much of aggressive hasslers; they do not want to compete for more significant opportunities available to them. They are also believed to be emotional and meticulous (Julizaerma and Sori 2013). Rawlinson (2018) pointed out the following factors to contribute to lower participation of women in the board: (1) They have little understanding of the complex business issues. (2) They do not possess the required credentials for the position. (3) The board’s environment does not go well with women. (4) Man, prejudice towards women in business. (5) The business world has always been considered a man’s world; it becomes difficult for men to accept a new gender. (6) Women are treated as a minority group with a certain number of required individuals. (7) Moreover, the investor is interested on the performance rather than the composition of the board Treanor (2017) showed Britain’s boardrooms lack the number of female directors after examining the FTSE 100 companies for ten years, the number of influential non-executive women increased from 6% in 2007 to 8% in 2017. Similarly, the participation of women on the board only risen from 11% in 2007 to 27.7% in 2017. The literature on the effect of women participation inboard on firm performance has been mixed. Marinova et al. (2016), Isidro and Sobral (2015), Rose et al. (2013); Shukeri et al. (2012), Yasser (2012), Carter et al. (2010), Miller and Del Carmen Triana (2009), Rose (2007), Farrell and Hersch (2005) found participation of women on the board to had no significant influence on the performance of the company. Kochan et al. obtained similar findings (2003), Erhardt et al. (2003), Rose (2007), Wang and Clift (2009), Matsa and Miller (2011), Jurkus et al. (2011), Ahmad et al. (2015), Vishwakarma and Kumar (2015), Marimuthu and Kolandaisamy (2009), Daunfeldt and Rudholm (2012), Shukeri et al. (2012) and Lückerath-Rovers (2013). At the global level, Noland et al. (2016) observed a total of 91 countries and 21,980 firms and found no relationship between gender and performance while the study of Terjesen et al. (2016) for 47 countries and 3876 firms observed a positive correlation on the increase on the number of women in the board to the performance of the firm. The study of Shrader et al. (1997), Smith et al. (2006), Campbell and Vera (2009), Chan et al. (2017), Oba and Fodio (2013), Corkery and Taylor (2012), Johl et al. (2013), Taghizadeh and Saremi (2013), Dobbin and Jung (2011), Carter et al. (2003), Dwyer et al. (2003), Carter et al. (2007), Gul et al. (2013), Appiadjei et al. (2017), Boulouta (2013), Ararat et al. (2015), Kılıç and Kuzey (2016), Low et al. (2015), Green and Homroy (2018), Conyon and He (2017), Haque (2017), Rossi et al. (2017), Liu et al. (2014), Peni(2014), Martín-Ugedo and Minguez-Vera (2014), Nguyen et al. (2015), Julizaerma and Sori (2013), Taghizadeh and Saremi (2013), LückerathRovers (2013), Molenkamp (2015), Hutchinson et al. (2015), Kim and Starks (2016), Willows and Van Der Linde (2016), Zainal et al. (2013), Appiadjei et al. (2017), Chan et al. (2017), Gordini and Rancati (2017), Shehata et al. (2017). Sarpong-Danquah

132

D. J. Mwamtambulo

et al. (2018) observed that a large proportion of women on top management significantly affect performance. Low et al. (2015) noted that the influence of women on the firm performance dies away in the countries were women participation is part of the culture. Green and Homroy (2018) observed that the number of women in the board increased if the CEO had a daughter leading to the enactment of rules and regulations that are favourable to women working in the company. Conyon and He (2017) emphasised the heterogeneity of female contribution to firm performance. In their study, a significant contribution of women participation in board was observed on already performing companies as compared to those who were underperforming. While Dwyer et al. (2003) noted the female contribution in performance was conditioned to the sound strategies and its organisational culture. Nguyen et al. (2015) emphasise on the need for a specific proportion of women on the board to affect performance. They observed that the performance of the company increases with an increase in the proportion of women up to a rate of 20%. Thereon with 1% increase in the proportion of women, this did not make any significant change in the performance of the company. While in Joecks et al. (2013) a negative relation between gender and performance were observed up to a critical mass point of 30% (exactly three women in the board) were a positive relationship was observed. The critical mass point requirements was also observed by Liu et al. (2014) were companies with three or more female directors in their boards performed better than those with less than three women. In their observations female inside directors had more impact on performance compared to the independent female directors and more so in non-state owned companies than in state owned. According to Bernile et al. (2018), Conyon and He (2017), Bernardi and Guptill (2008), Welch et al. (2008) and Bird and Brush (2018) women showed more ethical awareness than their counterparties, had higher degree of risk aversion and they cared for the longevity over fast growth of company. This was also portrayed by Haque (2017) were companies with a large proportion of women favoured the decision that were environmental friendly and future sustainable. Giannetti and Zhao (2015) also reported that companies with more diversity had more stock returns, flexible strategies but the stocks were general volatile and had many conflicts among its board members. Low et al. (2015) revealed that women use tougher monitoring thus diversity allow better managerial accountability and Puthenpurackal and Upadhyay (2013) observed that the effect on performance of women in the company was through their corporate experience and the company’s environment. Gallego-Álvarez et al. (2010) on their analysis of data of corporation listed in Madrid Stock Market 2004–2006 found Company with higher gender diversity not necessary outperformed their counterparties. In their case, gender diversity in the board made no significant difference in terms of performance. In the work of Adams and Ferreira (2009), Darmadi (2011), Ujunwa et al. (2012), Ujunwa (2012), Abdullah and Ismail (2013), Darmadi (2013), Wellalage and Locke (2013), Joecks et al. (2013), Abdullah (2014), Eulerich et al. (2014), Shehata et al. (2017) a negative relation between an increase in the proportion of women in the board to the performance of the firm Adams and Ferreira (2009) observe that female directors were active and have high chances of joining the monitoring committee. They argued against the

Board Characteristics and Performance of East Africa Companies

133

enforcement of female quotas in the company as it has an adverse reaction to the firm value. Similar, Ahern and Dittmar (2012) in Norway, observed the companies that were forced to increase the number of women in the board, experience value loss in their performance. In the light of the above findings, this study tested the positive relationship between an increase in the number of women in the board to the performance of the company. Hypothesis 4 Large proportion of women on the board is positively associated with the performance of the company.

2.5 The Proportion of Foreign Board Members An increase in the number of foreigners on the board is associated with better performance. Foreigners in the board bring in new ideas, methods, techniques and strategies that have been adopted in their native countries that can be applied into the company (Oxelheim and Randøy 2003). By introducing the new methods, procedures or challenges act into the company as a shock that stimulates an increase in the performance of the company. This can be observed by the finding of Oxelheim and Randøy (2003), Carter et al. (2003), Miller and Del Carmen Triana (2009), Ujunwa et al. (2012), Ujunwa (2012), Zainal et al. (2013), Müller (2014) and Marimuthu and Kolandaisamy (2014). Also, Smith et al. (2006), Shukeri et al. (2012), Ujunwa (2012), Abdullah and Ismail (2013), Wellalage and Locke (2013) observed that introducing different ethnicity in the board positively influenced the performance of the company while Carter DA et al. (2010) noted on no relation whatsoever between ethnicity measured by inclusion of minority in the board similar to the findings of Rose (2007), Darmadi (2011), Jhunjhunwala and Mishra (2012), Rose et al. (2013) and Molenkamp (2015). However Darmadi (2011), and Eulerich et al. (2014) observed a negative relationship between the nationality and the performance of the company. A significant positive relationship between the proportion of foreigners in the boards and the performance of the company has been observed in the literature over the years. Similarly, in this study such relationship was examined as stated in hypothesis 5 below. Hypothesis 5 Large proportion of foreigners on the board is positively associated with the performance of the company.

134

D. J. Mwamtambulo

3 Methodology 3.1 Data Financial and board structure information for the period between 2003 and 2017 from the non-financial listed companies in Tanzania, Kenya and Uganda markets were collected. Both the financial and board structure information was obtained from the companies’ annual reports available in their respective websites. The period between 2003 and 2017 was selected to accommodate the Tanzania and Uganda markets which opened their doors in January and April of 1998 respectively. A total of 15 years will provide a better account of the progress on the board characteristics and performance of the companies after allowing a grazing period of 5 years. The Rwanda stock market, which was launched on 31st January 2011, was dropped from the study due to a short period coverage. The current listing in the East Africa Community capital markets included a total of 73 non-financial firms when cross-listings are excluded. Financial institutions were not included in the data due to different policies regarding their performance accountability. The total number of companies which were listed for the covered period were 23 in Kenya, 5 in Tanzania and 1 in Uganda. Companies that were listed after 2003 were dropped from the analysis. Depending on the availability of data, a total of 16 companies were selected, of which 10 were from Kenya, 5 from Tanzania and 1 from Uganda were used in the study.

3.2 Methodology Performance measure included the market return measured by Tobin q ratio obtained by taking the Market value of the capital plus the book value of debt divided by the book value of assets. The accounting measure for performance included that of Return on Equity (ROE) and Return on Asset (ROA), which will examine the return for each unit of invested equity and asset, respectively. Both the market and accounting returns were used to mitigate the biases of the accounting measures, which are more susceptible to the type of account policy adopted by a company. In examining the relationship between board characteristics, the problem of endogeneity between the factor of performance and gender diversity and size of the board is always of concern (Carter et al. 2007; Lefort and Urzúa 2008; Brick and Chidambaran 2010; Wellalage and Locke 2013; Leung et al. 2014; Low et al. 2015; Nguyen et al. 2015; Kılıç and Kuzey 2016; Marinova et al. 2016; Terjesen et al. 2016; Rashid 2018). Board size and gender can influence the performance of the company and at the same time increase in performance may lead to an increase in the board size and number of women on the board. On this case, it is difficult to determine the causality between these factors. The best approach is using instrumental variables, and the best instruments are the value of the lag of the dependent variable. Adding lags into the model transform the model to dynamic panel models and the best estimator for this type of models is the

Board Characteristics and Performance of East Africa Companies

135

Generalised Methods of Moment (GMM). However, this model has a restriction on the number observations (N) which should be significantly higher than the time (T) variable (i.e. N > T). In this study N = 16, while T = 15, it is a case of Small N and Short T, GMM is not an appropriate estimation measure. Fixed effect models are also not applicable due to small sample bias, by including both the lags and the fixed effects it increases the biasedness of the estimators (Flannery and Watson 2013). Due to the characteristics of the data, the following system of simultaneous equations was solved to estimate the relationship between performance and board characteristics: Yit = β0 + β1 Board Sizeit + β2 Prop.Womenit + β3 Female Ex.Dirit + β4 Prop.Foreignersit + β5 Prop.Ext.Dirit + β6 Separ.Chairit + β7 Firm Sizeit + β8 Leverageit + εit

(1)

Board Sizeit = β0 + β1 Yit−1 + β2 Prop.Womenit + β3 Female Ex.Dirit + β4 Prop.Foreignersit + β5 Prop.Ext.Dirit + β6 Separ.Chairit + β7 Firm Sizeit + β8 Leverageit + εit (2) Prop.Womenit = β0 + β1 Yit−1 + β2 Board Sizeit + β3 Female Ex.Dirit + β4 Prop.Foreignersit + β5 Prop.Ext.Dirit + β6 Separ.Chairit + β7 Firm Sizeit + β8 Leverageit + εit

(3)

where Yit is the Performance Measures, that is Return on Assets (ROA), Return on Equity (ROE) and Tobin Q, BoardSizeit is the total number of Directors on the board. Prop. Womenit is the proportion of women on the board. FemaleEx.Dir it is a dummy variable, it is 1 in the existence of a female executive director and 0 if otherwise. Prop. Foreignersit is the proportion of foreigners on the board. Prop.Ext.Dir it is the proportion of external directors on the board. Separ.Chair it is also a dummy variable, it is one (1) if there is a separation of the position of the Chairman and CEO and zero (0) if otherwise. FirmSizeit is the control variable which is measured by taking the natural logarithm of Total Assets (lnT Ait−1 ) similarly Leverageit which is given by Debt to Equity Ratio and the error term εit : αi + μit accounting for the personal effects and errors of predictions.

4 Findings and Discussions 4.1 Descriptive Statistics Table 1 shows the descriptive statistics results for the data. During the period between 2003 and 2017, an average of 10 directors constituent the board and less than 1% were women, and they were less likely to hold the CEO position. On average of 36%

136

D. J. Mwamtambulo

Table 1 Descriptive statistics results Variable

Obs

Minimum

Maximum

Mean

Standard deviation

Board characteristics Board Size

240

3

26

9.1125

4.0437

Prop. Women

240

0

0.4

0.0869

0.1068

Female. Ex_Dir

240

0

1

0.0167

0.1283

Prop. Foreign Dir

240

0

0.8571

0.3596

0.2250

Prop.External_Dir

240

0

0.6667

0.6388

0.1581

Separ_Chairman

240

0

1

0.9375

0.2426

Critical mass point

240

0

1

0.1125

0.3166

Size

240

20.3275

27.7316

24.8826

1.4770

Leverage

240

−2.2032

38.4459

0.4882

2.5653

Tobin q

240

0.1392

9.2701

1.8277

1.6803

ROE

240

−11.8574

1.1759

0.1388

1.0867

ROA

240

−0.3015

1.2020

0.1778

0.1991

Firm characteristics

Performance Measures

Source Results from data analysis for the period between 2003 and 2017

of the board were foreigners, while more than 60% of the directors were independent directors. In the board, there were high chances the chairman and CEO positions to be occupied by different people. The companies had an average market return of 1.8, ROE and ROA stood at an average of 14% and 18% respectively. Companies were highly geared, and there were substantial variations between the least to the most geared company. The results for the three systems of the equations are presented in Tables 2, 3 and 4. In Table 3, results for ROE are presented where an increase in the proportion of women and foreigners on the board had a positive sign on ROE. The percentage of the external directors was also positive and significant at 5% while that of board size was positive and significant at 10% levels. The number of females holding the CEO position or separation of the position of the chairperson and the CEO had no significant impact on ROE. Moreover, the companies with more than three women on the board had a decline in their performance as reflected by a significant negative relationship with ROE. These results allowed for the rejection of Hypothesis 1 and 3 and not to reject Hypotheses 2, 4 and 5. Hypothesis 1 was rejected as a large size of the board rather than small board size was associated with better performance in the Eat Africa’s companies. No significant relationship between the past ROE and the proportion of women on the board, although a significant relationship between the size and past performance of ROE was observed. Leverage had a negative relationship with performance and no significant relationship with the board size and the proportion of women on the board. The proportion of independent directors decreased as the size of the company

Board Characteristics and Performance of East Africa Companies

137

Table 2 Regression results for ROE ROE Analysis 1 Coefficient

Analysis 2 t-value

Analysis 3

Coefficient

t-value

−31.4893***

−7.55

Intercept

2.5032

0.6

Board Size

0.2145*

1.89

Prop. Women

7.5634***

4.21

−0.9799

−0.4

Female. Ex_Dir

0.0425

0.1

−2.4088

−1.57

Prop. Foreign Dir

1.5353***

5.7

−0.7575

−0.78

Prop. External_ Dir

1.9052**

2.28

−7.0586***

−5.10

Coefficient 0.4257*** −0.0019 0.0348 −0.1004***

t-value 3.43 −1.04 0.84 −3.92

0.0045

0.12

−0.0068

−0.24

Separ_Chairman

−0.6367

−1.23

3.9994***

4.34

Critical mass value

−2.2259***

−5.85

1.3121

1.49

0.3012*

1.66

Size

−0.2303

−1.11

1.6717***

10.55

−0.0158***

−3.07

Leverage

−0.3385***

−0.0642

0.40

−0.0004

−0.21

R square

0.7668

0.4720

0.4671

Adjusted R Square

0.7577

0.4512

0.4438

lag ROE −22.19

0.2139*** 0.0067

11.17 1.38

Source Results from data analysis (Note ***, **, * are for 1%, 5%, 10% significant levels respectively)

increased while the number of women increased the proportion of foreign directors decreased. The chances for the appointment of women in the board increase when there are three (3) or more directors on the board. The size of the company had a strong relationship with both the size and the proportion of women on the board. As the size of the company increased, more directors were introduced on the board, and so is the number of women on the board. ROA was positively influenced by the number of external directors on the board and negatively affected by a number of women on the board and when the CEO powers in the board are reduced. All hypotheses except for Hypothesis 2 were rejected. Both the size and the proportion of women on the board were influenced by the past performance of ROA. Having more than three (3) women on the board negatively affected ROA while positively affected the proportion of women on the board. Size of the company negatively affected the performance and the percentage of women on the board but positively influenced the size of the board. The market return had a positive relationship with the factor of board size, the proportion of women in the board, female CEOs’ and with both the proportion of the foreigner and independent directors in the board. It negatively related to the decrease in the CEO powers on the board and when there were three or more women on the

138

D. J. Mwamtambulo

Table 3 Regression results for ROA ROA Analysis 1

Analysis 2

Coefficient

t-value

Intercept

2.7664

3.73

Board Size

0.1168

7.38

Analysis 3

Coefficient

t-value

−32.0778***

−7.88

2.6583

5.44

−2.1412

−0.89

Female. Ex_Dir

0.2173

2.44

−2.7818*

−1.85

Prop. Foreign Dir

0.5336

7.75

−1.2946

−1.34

Prop. External_Dir

0.5962***

4.74

-6.2125***

-4.51

Separ_Chairman

−0.4986***

−5.46

4.5433***

4.95

Critical mass value

−0.7175***

−7.26

1.6221*

1.86

3.8313***

3.54

−0.1571***

−4.71

1.6373***

10.53

−0.0245

−0.32

Size Leverage

-0.0031

-0.93

0.4042*** −0.0023

Prop. Women

Lag ROA

Coefficient

0.0296 −0.1108***

t-value 3.26 0.72 0.72 −4.23

0.0127

0.33

−0.0017

−0.06

0.2154***

11.29

0.0623**

2.02

−0.0133**

−2.64

0.0002

R square

0.5972

0.4934

0.4721

Adjusted R Square

0.5814

0.4735

0.4490

0.08

Source Results from data analysis (Note ***, **, * are for 1%, 5%, 10% significant levels respectively)

board. Under these results, Hypotheses 1 and 3 were rejected while Hypothesis 2, 4 and 5 were not rejected. Similar hypothesis 1 was rejected because an increase in the size of the board had a significant impact on the performance than a small board of director. Same to the separation of powers of the CEO rather than it is positive, it negatively affected the performance. The market was a bit sceptical regarding the decrease in powers of the CEO. The market performance was significantly influenced by the factor of the size of the company and leverage. While the increase in leverage had a positive impact on the performance of the company, but an increase in the size of the company had a negative impact on the profitability of the company. Increase in the size of the board had a negative impact on the number of independent directors on the board. Similarly, when the number of women on the board increased, it led to a decline in the proportion of foreigners on the board. Both the size and the proportion of women positively related to past market performance.

Board Characteristics and Performance of East Africa Companies

139

Table 4 Regression results for Tobin Q Tobin Q Analysis 1 Coefficient Intercept Board Size

Analysis 2 t-value

48.0343***

7.27

1.3701***

8.72

Analysis 3

Coefficient

t-value

−33.6862***

−8.18

Coefficient 0.2676** −0.0032

10.5602***

3.21

−2.6504

−1.08

Female. Ex_Dir

1.9101**

2.40

−1.9594

−1.3

Prop. Foreign Dir

2.6562***

5.15

−1.1124

−1.16

Prop. External_ Dir

7.8445***

6.38

−6.3897***

−4.64

0.0201

Prop. Women

0.0421 −0.1052***

t-value 2.15 −1.84 1.06 −4.27 0.54

Separ_Chairman

−5.6563***

-6.82

4.0919***

4.51

0.0092

Critical mass value

−3.3279***

−4.96

1.6068

1.84

0.2135***

0.3983***

3.23

0.0142***

−2.4095***

−7.76

1.7209

11.07

0.0091

−1.83

−0.0566

−0.76

−0.0002

0.09

Lag Tobin q Size Leverage

0.0695**

2.22

R square

0.5227

0.4890

0.5068

Adjusted R Square

0.5039

0.4689

0.4852

0.34 11.59 4.52

Source Results from data analysis (Note ***, **, * are for 1%, 5%, 10% significant levels respectively)

4.2 Discussion Mixed results have been observed between the two accounting measures and between the market and accounting measures in relation to the board characteristics. An increase in the board size, the proportion of women and foreigners’ directors in the board positively influence the performance of the company on returns to equity holders (ROE). This was not observed in the management of assets to generate profit (ROA) but rather a decrease in the power of the CEO negatively impact the utilisation of assets to generate income (ROA). An increase in the proportion of independent directors’ positive influence better returns to equity holders and at the same time on the management of the assets and its productivity. A board with three or more women leads a decline in the efficiency on the management of assets and generation of income to equity holders. The market reaction is quite different from the accounting measures. An increase in the size of the board, the proportion of women, the percentage of foreigners, the proportion of independent directors and the appointment of female CEO were positively received in the market. An increase in the proportion of women had a positive impact on the market, but this effect changes to negative when the critical

140

D. J. Mwamtambulo

mass is reached. Increase in the size of the company was also not entirely wellreceived by the investors as observed by the negative relationship between size and the Tobin q, while an increase in leverage was more tolerable.

5 Conclusion This study aimed at examining the relationship between the board characteristics and the performance of the companies in the East Africa markets. A total of 16 nonfinancial listed companies for the period between 2003 and 2017 were analysed. Mixed results were observed between accounting and market measures. Factor of ROE had a positive relationship with board size, the proportion of women, foreigners and external directors. The existence of independent directors had a positive relationship with the ability of the company on the management of assets. In accounting analysis, the board characteristics work toward improving the return to equity holders while it cannot be said the same on the management of assets. The market had a different reaction to the characteristics of the board. They responded positively with an increase in the size of the board, the proportion of women, foreigners and independent directors. They were more intrigued by the existence of a female CEO, but they were less intrigue when the number of female board members exceeded three. More so, with the increase in the size of the company. In changing the board characteristics, the market and the accounting reaction are quite different, and this difference is significant. While doing this study, it was observed that there was a significant drop in the performance of all the listed companies between 2015 and 2017. These drops were substantial and were not explained by the current model, thus raising red flags regarding their future. There is a need for further examination in these markets to understand the factors behind this phenomenon. This is an area of interest for future studies.

References Abdallah A et al (2018) The relationship between board size and CEO duality and firm performance: evidence from Jordan. Int J Account Fin Risk Manag 3(3):16–20. https://doi.org/10.11648/j. ijafrm.20180303.11 Abdul Latif R et al (2013) Multiple directorships, board characteristics and firm performance in Malaysia. Management. https://doi.org/10.5923/j.mm.20130302.07 Abdullah SN (2014) The causes of gender diversity in Malaysian large firms. J Manag Gov. https:// doi.org/10.1007/s10997-013-9279-0 Abdullah SN, Ismail KNIK (2013) Gender, ethnic and age diversity of the boards of large Malaysian firms and performance. J Pengurusan Abidin ZZ, Kamal NM, Jusoff K (2014) Board structure and corporate performance in Malaysia. Int J Econ Fin. https://doi.org/10.5539/ijef.v1n1p150 Adam AM, Schwartz MS (2009) Corporate governance, ethics, and the backdating of stock options. J Bus Ethics. https://doi.org/10.1007/s10551-008-9941-7

Board Characteristics and Performance of East Africa Companies

141

Adams RB, Ferreira D (2009) Women in the boardroom and their impact on governance and performance. J Financ Econ. https://doi.org/10.1016/j.jfineco.2008.10.007 Ahern KR, Dittmar AK (2012) The changing of the boards: the impact on firm valuation of mandated female board representation. Q J Econ. https://doi.org/10.1093/qje/qjr049 Ahmad RAR et al (2015) Board characteristics and risk management and internal control disclosure level: evidence from Malaysia. Procedia Econ Fin. https://doi.org/10.1016/s22125671(15)01147-8 Ahmadi A, Nakaa N, Bouri A (2018) Chief executive officer attributes, board structures, gender diversity and firm performance among French CAC 40 listed firms. Res Int Bus Fin 44(July 2017):218–226. https://doi.org/10.1016/j.ribaf.2017.07.083 Alabede JO (2016) Effect of board diversity on corporate governance structure and operating performance: evidence from the UK listed firms. Asian J Account Gov. https://doi.org/10.17576/ ajag-2016-07-06 Amran NA (2011) Corporate governance mechanisms and firm performance: evidence from Malaysian companies. Int Rev Bus Res Papers Anderson RD (2004) Board composition: balancing family influence in S&P 500 firms. In: European universities from the enlightenment to 1914. https://doi.org/10.1093/acprof:oso/9780198206606. 003.0014 Appiadjei A, Oppong G, Atta FNE (2017) Gender diversity and firm performance. Int J Econ Comm Manag V(10):1–16 Ararat M, Aksu, M. and Tansel Cetin, A. (2015) How board diversity affects firm performance in emerging markets: evidence on channels in controlled firms. Corp Gov Int Rev. https://doi.org/ 10.1111/corg.12103 Berk J, DeMarzo P (2014) Corporate Finance, The Prentice-Hall series in finance. https://doi.org/ 10.1016/S1574-0102(03)01011-2 Berk J, DeMarzo P, Harford J (2012) Fundamentals of corporate finance, 2nd edn. Prentice-Hall, New York. https://doi.org/10.1017/CBO9781107415324.004 Bernardi RA, Guptill ST (2008) Social desirability response bias, gender, and factors influencing organizational commitment: an international study. J Bus Ethics. https://doi.org/10.1007/s10551007-9548-4 Bernile G, Bhagwat V, Yonker S (2018) Board diversity, firm risk, and corporate policies. J Financ Econ. https://doi.org/10.1016/j.jfineco.2017.12.009 Bhagat S, Black BS (1998) The non-correlation between board independence and long-term firm performance. SSRN. https://doi.org/10.2139/ssrn.133808 Biggerstaff L, Cicero DC, Puckett A (2015) Suspect CEOs, unethical culture, and corporate misbehavior. J Financ Econ. https://doi.org/10.1016/j.jfineco.2014.12.001 Bodie Z, Kane A, Marcus AJ (2009) Essentials of investments, 9th edn. MacGraw-Hill. https://doi. org/10.1017/CBO9781107415324.004 Bodie Z, Kane A, Marcus AJ (2014) Investments, 10th ed. McGraw Hill Education, New York Boulouta I (2013) Hidden connections: the link between board gender diversity and corporate social performance. J Bus Ethics. https://doi.org/10.1007/s10551-012-1293-7 Brick IE, Chidambaran NK (2010) Board meetings, committee structure, and firm value. J Corp Finan. https://doi.org/10.1016/j.jcorpfin.2010.06.003 Brickley JA, Coles JL, Jarrell G (1997) Leadership structure: separating the CEO and chairman of the board. J Corp Fin 3:189–220. https://doi.org/10.1016/s0929-1199(96)00013-2 Brush C, Bird B (2018) A gendered perspective on organizational creation. Entrepreneurship Theory Pract. https://doi.org/10.1177/104225870202600303 Burns N, Kedia S (2006) The impact of performance-based compensation on misreporting. J Financ Econ.https://doi.org/10.1016/j.jfineco.2004.12.003 Campbell K, Vera AM (2009) Female board appointments and firm valuation: short and long-term effects. J Manag Gov. https://doi.org/10.1007/s10997-009-9092-y Carter D et al (2007) The diversity of corporate board committees and firm financial performance. SSRN. https://doi.org/10.2139/ssrn.972763

142

D. J. Mwamtambulo

Carter DA et al (2010) The gender and ethnic diversity of US boards and board committees and firm financial performance. Corp Gov Int Rev. https://doi.org/10.1111/j.1467-8683.2010.00809.x Carter DA, Simkins BJ, Simpson WG (2003) Corporate governance, board diversity, and firm value. Financ Rev. https://doi.org/10.1111/1540-6288.00034 Chan S-G, Yap L-KI, Zainudin R (2017) ‘Gender diversity and firms’ financial performance in Malaysia. Asian Acad Manag J Account Fin. https://doi.org/10.21315/aamjaf2017.13.1.2 Chen CH, Al-Najjar B (2012) The determinants of board size and independence: evidence from China. Int Bus Rev. https://doi.org/10.1016/j.ibusrev.2011.09.008 Citizen (2016) Tanzania : stock market investors up by 132 percent. The Citizen, pp 2–3. Available at: https://allafrica.com/stories/201610280163.html Collins DW, Gong G, Li H (2009) Corporate governance and backdating of executive stock options. Contemp Account Res. https://doi.org/10.1506/car.26.2.4 Conyon MJ, He L (2017) Firm Performance and Boardroom Gender Diversity: A Quantile Regression Approach. Journal of Business Research. https://doi.org/10.1016/j.jbusres.2017. 02.006 Corkery J, Taylor M (2012) The gender gap: a quota for women on the board. Corp Gov eJournal. Cornett MM, Marcus AJ, Tehranian H (2008) Corporate governance and pay-for-performance: the impact of earnings management. J Financ Econ. https://doi.org/10.1016/j.jfineco.2007.03.003 Cottrell P (2016) Corporate finance. SSRN. https://doi.org/10.2139/ssrn.2779908 Dalton, Dan R, Daily CM, Johnson JL, Ellstrand AE (1998). Meta-analytical reviews of board composition, leadership structure, and financial performance. Strateg Manag J 19:269–290 Darmadi S (2011) Board diversity and firm performance: the Indonesian evidence. Corp Ownership Control. https://doi.org/10.22495/cocv8i2c4p4 Darmadi S (2013) Do women in top management affect firm performance? Evidence from Indonesia. Corp Gov (Bingley). https://doi.org/10.1108/CG-12-2010-0096 Daunfeldt S-O, Rudholm N (2012) Does gender diversity in the boardroom improve firm performance? HUI working papers no. 60 Delloite (2016) Uganda economic outlook 2016 the story behind the numbers. Available at: https:// www2.deloitte.com/content/dam/Deloitte/ug/Documents/tax/EconomicOutlook2016UG.pdf Devos E, Elliott WB, Warr RS (2015) CEO opportunism?: option grants and stock trades around stock splits. J Account Econ. https://doi.org/10.1016/j.jacceco.2015.02.004 Dobbin F, Jung J (2011) Board diversity and corporate performance: filling in the gaps: corporate board gender diversity and stock performance: the competence gap or institutional investor bias? North Carolina Law Rev Duru A, Iyengar RJ, Zampelli EM (2016) The dynamic relationship between CEO duality and firm performance: the Moderating role of board independence. J Bus Res. https://doi.org/10.1016/j. jbusres.2016.04.001 Dwyer S, Richard OC, Chadwick K (2003) Gender diversity in management and firm performance: the influence of growth orientation and organizational culture. J Bus Res. https://doi.org/10.1016/ S0148-2963(01)00329-0 EAC (2019) Capital markets—an overview. Available at: https://www.eac.int/financial/capitalmarkets Ederer F, Manso G (2008) Is pay-for-performance detrimental to innovation? SSRN. https://doi. org/10.2139/ssrn.1270384 Ehikioya BI (2009) Corporate governance structure and firm performance in developing economies: evidence from Nigeria. Corp Gov Int J Bus Soc. https://doi.org/10.1108/14720700910964307 Erhardt NL, Werbel JD, Shrader CB (2003) Board of director diversity and firm financial performance. Corp Gov. https://doi.org/10.1111/1467-8683.00011 Eulerich M, Velte P, Van Uum C (2014) The impact of management board diversity on corporate performance—an empirical analysis for the German two-tier system. Probl Perspect Manag. https://doi.org/10.11113/jt.v71.3854 Fan JPH et al (2016) Does board structure in banks affect their performance? J Financ Econ.https:// doi.org/10.1016/j.jbankfin.2012.12.016

Board Characteristics and Performance of East Africa Companies

143

Flannery MJ, Watson K (2013) Estimating dynamic panel models in corporate Finance. J Corp Fin 19:1–19. https://doi.org/10.1016/j.jcorpfin.2012.09.004 Gallego-Álvarez I, García-Sánchez IM, Rodríguez-Dominguez L (2010) The influence of gender diversity on corporate performance. Revista de Contabilidad. https://doi.org/10.1016/S11384891(10)70012-1 Gaur SS, Bathula H, Singh D (2015) Ownership concentration, board characteristics and firm performance: a contingency framework. Manag Decis 53(5):911–931. https://doi.org/10.1108/ MD-08-2014-0519 Giannetti M, Zhao M (2015) Board ancestral diversity and firm performance volatility. SSRN. https://doi.org/10.2139/ssrn.2700058 Giroux G (2008) What went wrong? Accounting fraud and lessons from the recent scandals. Soc Res Gordini N, Rancati E (2017) Gender diversity in the Italian boardroom and firm financial performance. Manag Res Rev. https://doi.org/10.1108/MRR-02-2016-0039 Green CP, Homroy S (2018) Female directors, board committees and firm performance. Eur Econ Rev. https://doi.org/10.1016/j.euroecorev.2017.12.003 Guest PM (2009) The impact of board size on firm performance: evidence from the UK. Eur J Fin. https://doi.org/10.1080/13518470802466121 Gul FA, Hutchinson M, Lai KMY (2013) Gender-diverse boards and properties of analyst earnings forecasts. Account Horiz. https://doi.org/10.2308/acch-50486 Haniffa R, Hudaib M (2006) Corporate governance structure and performance of Malaysian listed companies. J Bus Fin Account. https://doi.org/10.1111/j.1468-5957.2006.00594.x Haque F (2017) The effects of board characteristics and sustainable compensation policy on carbon performance of UK firms. Br Account Rev. https://doi.org/10.1016/j.bar.2017.01.001 Heenetigala K, Armstrong AF (2011) The impact of corporate governance on firm performance in an unstable economic and political environment: evidence from Sri Lanka. SSRN. https://doi. org/10.2139/ssrn.1971927 Heron RA, Lie E (2007) Does backdating explain the stock price pattern around executive stock option grants? J Financ Econ. https://doi.org/10.1016/j.jfineco.2005.12.003 Heron RA, Lie E (2009) What fraction of stock option grants to top executives have been backdated or manipulated? Manag Sci. https://doi.org/10.1287/mnsc.1080.0958 Hersch PL, Farrell KA (2005) Additions to corporate boards: the effect of gender. J Corp Fin. https:// doi.org/10.1016/j.jcorpfin.2003.12.001 Hidayat AA, Utama S (2016) Board characteristics and firm performance: evidence from Indonesia. Int Res J Bus Stud 8(3):137–154 Hutchinson M, Mack J, Plastow K (2015) Who selects the “right” directors? An examination of the association between board selection, gender diversity and outcomes. Account Fin. https://doi. org/10.1111/acfi.12082 Ilaboya OJ, Obaretin O (2015) Board characteristics and firm performance: evidence from Nigerian quoted companies. Acad J Interdiscip Stud 4(1):283–290. https://doi.org/10.5901/ajis.2015. v4n1p283 Isidro H, Sobral M (2015) The effects of women on corporate boards on firm value, financial performance, and ethical and social compliance. J Bus Ethics. https://doi.org/10.1007/s10551014-2302-9 Jackling B, Johl S (2009) Board structure and firm performance: evidence from India’s top companies. Corp Gov Int Rev. https://doi.org/10.1111/j.1467-8683.2009.00760.x Jhunjhunwala S, Mishra RK (2012) ‘Board diversity and corporate performance : the Indian evidence. J Corp Gov Joecks J, Pull K, Vetter K (2013) Gender diversity in the boardroom and firm performance: what exactly constitutes a “critical mass?” J Bus Ethics. https://doi.org/10.1007/s10551-012-1553-6 Johl SK, Kaur S, Cooper BJ (2013) Board characteristics and firm performance: evidence from Malaysian public listed firms. J Econ Bus Manag. https://doi.org/10.7763/joebm.2015.v3.187

144

D. J. Mwamtambulo

Johl SK, Kaur S, Cooper BJ (2015) Board characteristics and firm performance: evidence from Malaysian public listed firms. J Econ Bus Manag 3 (2):239–243 Julizaerma MK, Sori ZM (2013) Gender diversity in the boardroom and firm performance of Malaysian public listed companies. Procedia—Soc Behav Sci. https://doi.org/10.1016/j.sbspro. 2012.11.374 Jurkus AF, Park JC, Woodard LS (2011) Women in top management and agency costs. J Bus Res. https://doi.org/10.1016/j.jbusres.2009.12.010 Kaplan R, Kiron D (2007) Accounting fraud at WorldCom. Business Kim D, Starks LT (2016) Gender diversity on corporate boards: do women contribute unique skills? Am Econ Rev. https://doi.org/10.1257/aer.p20161032 Kılıç M, Kuzey C (2016) The effect of board gender diversity on firm performance: evidence from Turkey. Gender Manag. https://doi.org/10.1108/GM-10-2015-0088 Kochan T et al (2003) The effects of diversity on business performance: report of the diversity research network. Hum Resour Manag. https://doi.org/10.1002/hrm.10061 Kyereboah-Coleman A, Biekpe N (2006) The relationship between board size, board composition, CEO duality and firm performance: experience from Ghana. Corp Ownership Control Lal Bhasin M (2013) Corporate accounting fraud: a case study of Satyam Computers Limited. Open J Account. https://doi.org/10.4236/ojacct.2013.22006 Lawler EE, Benson GS, McDermott M (2012) What makes performance appraisals effective? Compensation Benefits Rev. https://doi.org/10.1177/0886368712462331 Lefort F, Urzúa F (2008) Board independence, firm performance and ownership concentration: evidence from Chile. J Bus Res. https://doi.org/10.1016/j.jbusres.2007.06.036 Leung S, Richardson G, Jaggi B (2014) Corporate board and board committee independence, firm performance, and family ownership concentration: an analysis based on Hong Kong firms. J Contemp Account Econ. https://doi.org/10.1016/j.jcae.2013.11.002 Lie E (2005) On the timing of CEO stock option awards. Manag Sci. https://doi.org/10.1287/mnsc. 1050.0365 Liu Y et al (2015) Board independence and firm performance in China. J Corp Fin. https://doi.org/ 10.1016/j.jcorpfin.2014.12.004 Liu Y, Wei Z, Xie F (2014) Do women directors improve firm performance in China? J Corp Fin. https://doi.org/10.1016/j.jcorpfin.2013.11.016 Low DCM, Roberts H, Whiting RH (2015) Board gender diversity and firm performance: empirical evidence from Hong Kong, South Korea, Malaysia and Singapore. Pac-Basin Fin J. https://doi. org/10.1016/j.pacfin.2015.02.008 Lückerath-Rovers M (2013) Women on boards and firm performance. J Manag Gov. https://doi. org/10.1007/s10997-011-9186-1 Lyke B, Jickling M (2002) WorldCom: the accounting scandal. The Library of Congress Marimuthu M, Kolandaisamy I (2009) Can demographic diversity in top management team contribute for greater financial performance? An empirical discussion. J Int Soc Res Marimuthu M, Kolandaisamy I (2014) Ethnic and gender diversity in boards of directors and their relevance to financial performance of Malaysian companies. J Sustain Dev. https://doi.org/10. 5539/jsd.v2n3p139 Marinova J, Plantenga J, Remery C (2016) Gender diversity and firm performance: evidence from Dutch and Danish boardrooms. Int J Hum Resour Manag. https://doi.org/10.1080/09585192. 2015.1079229 Martín-Ugedo JF, Minguez-Vera A (2014) Firm performance and women on the board: evidence from Spanish small and medium-sized enterprises. Feminist Econ. https://doi.org/10.1080/ 13545701.2014.895404 Masulis RW, Reza SW (2015) Agency problems of corporate philanthropy. Rev Financ Stud. https:// doi.org/10.1093/rfs/hhu082 Matsa DA, Miller AR (2011) Chipping away at the glass ceiling: gender spillovers in corporate leadership. Am Econ Rev. https://doi.org/10.1257/aer.101.3.635

Board Characteristics and Performance of East Africa Companies

145

Meckling WH, Jensen MC (1976) Managerial behaviour, agency costs and ownership structure. J Financ Econ. https://doi.org/10.1016/0304-405X(76)90026-X Mensah C (2016) Africa’s economy; trends, challenges, prospects and potentials. Available at: https://ssrn.com/abstract=2863660 Miller T, Del Carmen Triana M (2009) Demographic diversity in the boardroom: mediators of the board diversity-firm performance relationship. J Manag Stud. https://doi.org/10.1111/j.14676486.2009.00839.x Molenkamp J (2015) Board diversity and firm performance : the Dutch evidence. In: 5th IBA bachelor thesis conference Müller V-O (2014) The impact of board composition on the financial performance of FTSE100 constituents. Procedia—Soc Behav Sci. https://doi.org/10.1016/j.sbspro.2013.12.573 Nguyen T, Locke S, Reddy K (2015) Does boardroom gender diversity matter? Evidence from a transitional economy. Int Rev Econ Fin. https://doi.org/10.1016/j.iref.2014.11.022 Noland M, Moran T, Kotschwar BR (2016) Is gender diversity profitable? Evidence from a global survey. SSRN. https://doi.org/10.2139/ssrn.2729348 Oba VC, Fodio MI (2013) Boards’ gender mix as a predictor of financial performance in Nigeria: an empirical study. Int J Econ Fin. https://doi.org/10.5539/ijef.v5n2p170 Oxelheim L, Randøy T (2003) The impact of foreign board membership on firm value. J Bank Fin. https://doi.org/10.1016/S0378-4266(02)00395-3 Peni E (2014) CEO and chairperson characteristics and firm performance. J Manag Gov. https:// doi.org/10.1007/s10997-012-9224-7 Puthenpurackal J, Upadhyay A (2013) Board gender diversity and firm performance: the impact of information environment., twenty years after Cadbury, ten years after Sarbanes-Oxley: challenges of corporate governance Rashid A (2009) Board composition, board leadership structure and firm performance: evidence from Bangladesh. In: Proceedings of the 2009 AFAANZ conference Rashid A (2018) Board independence and firm performance: evidence from Bangladesh. Future Bus J 4. https://doi.org/10.1016/j.fbj.2017.11.003 Rawlinson K (2018) FTSE firms’ excuses for lack of women in boardrooms’ pitiful and patronising. The Guardian, pp 2018–2020 Rose C (2007) Does Female Board Representation Influence Firm Performance? The Danish evidence. Corporate Governance. https://doi.org/10.1111/j.1467-8683.2007.00570.x Rose C, Munch-Madsen P, Funch M (2013) Does board diversity really matter? Gender does not, but Citizenship does. Int J Bus Sci Appl Manag Rossi F, Hu C, Foley M (2017) Women in the boardroom and corporate decisions of Italian listed companies: does the “critical mass” matter? Manag Decis. https://doi.org/10.1108/MD-01-20170029 Sarpong-Danquah B et al (2018) Corporate governance and firm performance: an empirical analysis of manufacturing listed firms in Ghana. Account Fin Res. https://doi.org/10.5430/afr.v7n3p111 Scharff MM (2007) Understanding WorldCom’s accounting fraud: did groupthink play a role? J Leadersh Organ Stud. https://doi.org/10.1177/107179190501100309 Shapiro SP (2005) Agency theory. Ann Rev Sociol. https://doi.org/10.1146/annurev.soc.31.041304. 122159 Shehata N, Salhin A, El-Helaly M (2017) Board diversity and firm performance: evidence from the U.K. SMEs. Appl Econ. https://doi.org/10.1080/00036846.2017.1293796 Shrader CB, Blackburn V, Iles P (1997) Women in management and firm financial performance: an exploratory study. J Manag Issues. https://doi.org/10.2307/40604152 Shukeri SN, Shin OW, Shaari MS (2012) Does board of director’s characteristics affect firm performance? Evidence from Malaysian public listed companies. Int Bus Res. https://doi.org/10.5539/ ibr.v5n9p120 Smith N, Smith V, Verner M (2006) Do women in top management affect firm performance? A panel study of 2,500 Danish firms. Int J Prod Perform Manag. https://doi.org/10.1108/ 17410400610702160

146

D. J. Mwamtambulo

Taghizadeh M, Saremi S (2013) Board of directors and firms performance: evidence from Malaysian public listed firm. IPEDR. https://doi.org/10.7763/IPEDR Tee K, Wiley M (2018) Backdating of executive stock options: Comparing financial and nonfinancial industries. J Financ Crime. https://doi.org/10.1108/JFC-01-2017-0001 Terjesen S, Couto EB, Francisco PM (2016) Does the presence of independent and female directors impact firm performance? A multi-country study of board diversity. J Manag Gov. https://doi. org/10.1007/s10997-014-9307-8 Treanor J (2017) Number of senior women in Britain’s boardrooms unchanged in 10 years. The Guardian, pp 2017–2019 Ujunwa A (2012) Board characteristics and the financial performance of Nigerian quoted firms. Corp Gov (Bingley). https://doi.org/10.1108/14720701211275587 Ujunwa A, Nwakoby I, Ugbam CO (2012) Corporate board Diversity and firm performance: evidence from Nigeria. Corp Ownership Control Veld C, Wu BHT (2014) What drives executive stock option backdating? J Bus Fin Account. https:// doi.org/10.1111/jbfa.12077 Vishwakarma R, Kumar A (2015) Does corporate governance increases firm performance of IT industry ? An empirical analysis. BVIMSR’s J Manag Res Wang Y, Clift B (2009) Is there a “business case” for board diversity? Pac Account Rev. https://doi. org/10.1108/01140580911002044 Weber J, Willenborg M, Zhang J (2008) Does auditor reputation matter? The case of KPMG Germany and ComROAD AG. J Account Res. https://doi.org/10.1111/j.1475-679X.2008.00298.x Weir C, Laing D, Mcknight PJ (2002) Internal and external governance mechanisms: their impact on the performance of large UK public companies. J Bus Fin Account. https://doi.org/10.1111/ 1468-5957.00444 Weisbach MS, Hermalin BE (2003) Boards of directors as an endogenously determined institution: a survey of the economic literature. SSRN Electron J Welch CL, Welch DE, Hewerdine L (2008) Gender and export behaviour: evidence from womenowned enterprises. J Bus Ethics. https://doi.org/10.1007/s10551-007-9652-5 Wellalage NH, Locke S (2013) Corporate Governance, board diversity and firm financial performance: new evidence from Sri Lanka. Int J Bus Gov Ethics. https://doi.org/10.1504/ijbge.2013. 054416 Willows G, Van Der Linde M (2016) Women representation on boards: a South African perspective. Meditari Accountancy Res. https://doi.org/10.1108/MEDAR-01-2016-0001 Yammeesri J, Herath SK (2010) Board characteristics and corporate value: evidence from Thailand. Corp Gov. https://doi.org/10.1108/14720701011051910 Yasser QR (2012) Affects of female directors on firms performance in Pakistan. Mod Econ. https:// doi.org/10.4236/me.2012.37104 Yermack D (1995) Do corporations award CEO stock options effectively? J Financ Econ. https:// doi.org/10.1016/0304-405X(95)00829-4 Yermack D (1996) Higher market valuation of companies with a small board of directors. J Financ Econ. https://doi.org/10.1016/0304-405X(95)00844-5 Yermack D (1997) Good timing: CEO stock option awards and company news announcements. J Fin. https://doi.org/10.1111/j.1540-6261.1997.tb04809.x Zainal D et al (2013) Corporate board diversity in Malaysia: a longitudinal analysis of gender and nationality diversity. Int J Acad Res Account Fin Manag Sci 3(1):136–148

Quantitative Methods in Finance

Different Approaches to the Reference Yield Curve Construction—And Their Application into Fund Transfer Pricing Mechanism Ewa Dziwok and Martin Wirth

Abstract The paper investigates different approaches to the construction of a term structure of interest rates—reference rates that are the base for a Fund Transfer Pricing mechanism (FTP). While many positions in the literature focus on FTP mechanism as a part of asset liability management (ALM) process without any closer look at term structure construction, we identify features that let measure the behavior of the yield curve and detect the consequences of the model’s choice. The results show that the arbitrarily chosen model of the reference yield could have significant consequences for risk management process of a financial institution. The study provides a twofold contribution to the literature describing FTP mechanism. First it introduces more complex approach to the reference rate modeling inside FTP mechanism and shows the differences between considered models. Moreover it focuses on the construction of the reference curve itself and shows two different approaches covering a parsimonious model as well as Smith-Wilson one.

1 Introduction Fund Transfer Pricing (FTP) mechanism has been an essential part of bank management for almost 40 years. Its beginnings go to the 1970s, when the era of stable interest rates came to the end. The FTP mechanism was employed into banks as a respond to an increasing volatility of interest rates of interest rates to help to manage risk that was connected with it (van Deventer et al. 2005). In spite of the fact that during the period of low interest rates the FTP mechanism wasn’t treated as essential, the consequences of the last global financial crisis showed that might play the fundamental role in liquidity management. Shortly after, the E. Dziwok (B) University of Economics in Katowice, Katowice, Poland e-mail: [email protected] M. Wirth University of Applied Sciences BFI Vienna, Vienna, Austria e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_12

149

150

E. Dziwok and M. Wirth

regulators established it as a regulatory requirement. What is more, the supervisors expect from banks to demonstrate how their FTP structure is related to the best practices of liquidity management (BSBC 2008; BSBC 2013; IIF 2007). Lack of precise manual from regulators and bank’s autonomy in establishing liquidity measurement and management process caused a lot of functional inefficiencies in FTP structure (EBA 2010; Grant 2011). What is more, the yields or term structure segments are affected by the existence of a liquidity premium components that are difficult to extract and measure (Fleckenstein et al. 2014; d’Amico et al. 2014). Institutions should have an adequate internal transfer pricing mechanism based on reference rate delivered from the market in a form of the yield curve. The role of FTP has been widely described lately by Wyle and Tsaig (2011), Elliot and Lindblom (2015) who emphasized its importance in business unit profitability measurement, interest rate risk and funding liquidity risk management. The study provides a twofold contribution to FTP mechanism. First it introduces more complex approach to the reference rate modeling inside FTP mechanism and then it shows the differences between considered models. The data taken into account come from Polish and Euro money market and cover the period between 2006 and 2018 and the results let point out the differences in goodness of fit depending of the model and—in consequence—have an effect on the fund transfer pricing mechanism. The paper is organized as follows. Section 2 impact of a yield curve construction on FTP process. Section 3 reports descriptive results from different yield curve models and Sect. 4 concludes.

2 An Impact of a Yield Curve Construction on FTP Process The idea of fund transfer pricing mechanism FTP could be generally understood as an internal tool that takes sources from a business unit that accumulates funds and moves them to the business unit offering funds. The base of the mechanism is created on the reference rate which is often market determined in a form of a fixing. The first step is a construction of a reference curve (or yield curve) through interest rate term structure model. Among plenty methods of a yield curve construction, two main groups of models could be taken into account: parametric ones introduced by Nelson and Siegel (1987) and extended by Svensson (1994) as well as Smith-Wilson (2000) model. Both types give a lot of possibilities for further analysis and forecasting. The term structure of interest rates gives the relationship between the yield of the investment with the same credit quality but different term to maturity. Generally a term structure is typically built with a set of liquid and common assets; the problem arises in a case of non-liquid market with a small number of data. One of solutions is to analyze several types of models and then to choose this one which let achieve the best approximation. The inter-relation between discount factor, spot rate and forward one (in continuous time) could be illustrated as below:

Different Approaches to the Reference Yield Curve Construction …

P(τ ) = δ(τ ) = e− i (τ ) ·τ = e−

τ 0

151

f (m)dm

(1)

where P(τ ) δ(τ ) i(τ ) f (τ ) τ

price of a bond discount factor (δτ) spot rate forward rate term to maturity.

To fit the curve it is necessary to choose an interpolation method, (a form of the theoretical function) which let receive discount factors δ(τ ) for all maturities (between zero and infinity).

2.1 Parsimonious Models The utilization of a parametric model (Nelson-Siegel) let calculate forward rates directly. It guarantees achievement of different shapes of theoretical term structure thorough the estimation of four parameters δ(τ ) = δ(τ | β0 , β1 , β2 , v):  τ  −τ · e v f (τ ) = β0 + β1 + β2 v i(τ ) = β0 + (β1 + β2 )

1−e τ v1

− vτ

1

(2a)

− β2 · e

− vτ

1

(2b)

where f (τ ) [β0 , β1 , β2 , v] β0 β1 β2 υ1

instantaneous forward rate vector of parameters describing the curve parameter which shows a limit in infinity, β0 > 0 parameter which shows a limit in infinity, β0 + β1 ≥ 0 parameter which shows a strength of curvature parameter which shows a place of curvature, υ1 > 0.

The utilization of the second parametric model (Svensson) let calculate more sophisticated shapes of the yield curve thorough the estimation of six parameters δ(τ ) = δ(τ | β0 , β1 , β2 , β3 , ν1 , ν2 ):   τ τ −τ −τ · e v1 + β3 · e v2 f (τ ) = β0 + β1 + β2 v1 v2   τ − −τ 1 − e v2 1 − e v1 − vτ − vτ − β2 · e 1 + β3 −e 2 i(τ ) = β0 + (β1 + β2 ) τ τ v1

v2

(3a)

(3b)

152

E. Dziwok and M. Wirth

where β3 parameter which shows a strength of curvature υ2 parameter which shows a place of curvature, υ2 > 0. When a set of discount factors (for all cash

flows) are calculated from forward rates, a vector of theoretical prices P = Pl l=1,...,k can be described as a product of a cash flow matrix C and a vector of discount factors (in a functional form): P

 = C · δ(τ1 ) δ(τ2 ) . . . δ(τk ) T

(4)

A set of parameters is estimated by minimizing mean square errors between market prices and theoretical ones (taken from the fitted curve):

k l=1



Pl − Pl k

2 → min

(5)

where Pl − Pl a price error of l-th asset k - number of bonds. The goodness of fit comparison (for prices and yields respectively) is possible by the calculation of errors through time. A low mean value proves the flexibility of the model and shows its ability to fit the data quite accurately.

2.2 Smith-Wilson Model The Smith-Wilson procedure was initially devoted to a yield curve construction for insurance industry. Using the Smith-Wilson model the yield curve is built as a result of an exponential spline optimization problem to create the optimal discount curve, and finally a smooth term structure of interest rates. The Smith-Wilson procedure creates the discount function as linear combinations of kernel functions Kj . The unknown parameters ζj needed to calculate the kernel functions are solutions to a linear system based on the symmetric functions W (τ, τ j ) plus the discounted ultimate forward rate. The input parameters are: • the Ultimate Forward Rate (UFR), continuously compounded • α—controls the speed of extrapolation to the UFR:

P(τ ) = e−U F R·τ +

N  j=1

  ζ j · W τ, τ j τ ≥ 0

(6a)

Different Approaches to the Reference Yield Curve Construction … 



W τ, τ j = e

−U F R·(τ +τ j )

·

⎧ ⎨ ⎩





α · min τ, τ j −

153

 ⎫ e−α·max(τ,τ j ) · eα·min(τ,τ j ) − e−α·min(τ,τ j ) ⎬ ⎭

2

(6b) where N number of zero coupon bonds τ j , j = 1, 2, …, N the maturities of all cash payments ζ j , j = 1, 2, …, N parameters to fit the actual yield curve.

3 Data and Results For the research IRS market rates were taken into account from euro (EUR) and the Polish market (PLN) since the beginning of 2006–2018. They were represented by two panels of data with maturities up to 10 years.

3.1 Parametric Model Considering the way of MSE error calculation (5) and following Nelson-Siegel and Svensson parametric model two sets of instantaneous rates can be found both for EUR and PLN (Dziwok 2019). To achieve these results two macros were written in VBA code that helped to receive theoretical prices for each of analyzed days. As a result, two vectors of MSE were calculated. Taking into account the similarity of the results only the plots of errors for Nelson-Siegel is shown here (Figs. 1 and 2). The methods let analyze the sensitivity of the model to disturbances in the market. The plots of MSE errors (differences between theoretical and market prices) for chosen methods let analyze the sensitivity of the model to disturbances in the market. 0,00025

EUR 0,00020 0,00015 0,00010 0,00005 0,00000

Fig. 1 MSE errors between market and theoretical prices for EUR. Source Own calculations

154

E. Dziwok and M. Wirth

0,00025

PLN 0,00020 0,00015 0,00010 0,00005 0,00000

Fig. 2 MSE errors between market and theoretical prices for PLN. Source Own calculations

From the beginning of financial crisis the volatility of assets’ rates had become very high which caused problems with the data fitting. The highest value of errors was observed during financial crises and accelerated in 2009–2010 period.

3.2 Smith-Wilson Model The basic risk free term structure is based on zero coupon swap yields. Forward rates beyond 10 years equal the unconditional ultimate forward rate (UFR). Extrapolation and intrapolation between zero yields uses the Smith-Wilson method while extrapolation towards UFR starts at the last liquid swap maturity (the entry point). The evolution of the two yield curves using Smith-Wilson model and linear interpolation was computed and the mean squared error between the two yield curves was presented (Figs. 3 and 4).

Fig. 3 Mean squared error between Smith-Wilson and linear interpolation for EUR. Source Own calculations

Different Approaches to the Reference Yield Curve Construction …

155

Fig. 4 Mean squared error between Smith-Wilson and linear interpolation for PLN. Source Own calculations

An important conclusion following from the analysis above is the fact that the parametric models can be used to determine the FTP reference curve. The selection problem presented here (how to find the best method of the reference yield construction by adopting a comparison of errors) shows that the best results were achieved by an implementation of the MSE price methodology (through a minimizing of the sum of squared errors of market and theoretical prices). An implementation of the Smith-Wilson method provided an exact fit to the swap rates. The main disadvantage of the exact fit is that the Smith-Wilson yield curve is subject to more fluctuations relative to the smoothed cubic spline. The smoothed cubic spline on the other hand does not provide an exact fit to the market rates—smoothing have a little practical impact on the ability to match quotes. Unfortunately, the Smith-Wilson method may result in unrealistic yield curves if the estimation sample reflects an untypical yield curve. For most swap markets, the problem will be minimal, but there may be big difficulties. While for the EUR yield curve the Smith-Wilson interpolation resulted in a plausible interpolated yield curves on all dates, the interpolation for PLN resulted in unrealistic yield curves for some dates. Figure 5 shows the interpolated yield curves for PLN for January 27th, 2015 where the steep inversion at the short end of the curve plus the flat structure for longer maturities resulted in an oscillating interpolation. Such an oscillating yield curve implicitly results in implausible forward rates.

156

E. Dziwok and M. Wirth

Fig. 5 Linear interpolation and Smith-Wilson interpolation for January 27th, 2015 for PLN. Source Own calculations

4 Summary The results show that the arbitrarily chosen model of the reference yield has significant consequences for liquidity risk management process of a financial institution. The chosen measure—Mean Squared Error methodology—let compare selected models and assess their applicability into FTP mechanism (Table 1). The analyzed data show the differences between considered models and emphasize the pros and cons of both types. An important conclusion coming from the analysis above is the fact that the parametric models have better characteristics for the FTP reference curve construction. One way to overcome the problem that SmithWilson interpolation may result in very implausible yield curves would be to use Table 1 A comparison of yield curve models Type of model

Pros

Nelson-Siegel

Sufficient flexibility to capture most popular shapes of yield curves; four parameters easy to estimate and interpret

Svensson

Enables to create more complicated yield curves with more than one local maximum or minimum

Smith-Wilson

Provides a perfect fit to liquid market data

Type of model

Cons

Nelson-Siegel

Estimation problems (nonlinear), cannot handle all yield curve shapes, forward rates are always positive

Svensson

No significant improvement of the estimates when compared with the Nelson-Siegel model

Smith-Wilson

Choice of alpha (the speed of convergence to the UFR) is chosen subjectively, the discount factor, may become negative, interpolation may result in oscillating and implausible yield curves

Different Approaches to the Reference Yield Curve Construction …

157

smoothed cubic spline interpolation. This would not result in a yield curve that provides a perfect fit to market data but the interpolations would be more plausible. Linear interpolation would be another approach resulting in a plausible interpolation with a perfect fit to market data, but the resulting yield curve contains kinks and thus would no longer be differentiable.

References BSBC (2008) Basel Committee on banking supervision principles for sound liquidity risk management and supervision, September, BIS. https://www.bis.org BCBS (2013) Basel Committee on banking supervision: Basel III: the liquidity coverage ratio and liquidity risk monitoring tools. https://www.bis.org D’Amico et al (2014) Tips from tips: the informational content of treasury inflation-protected security prices. FEDS Working Paper 2010-19 Dziwok E (2019) The role of a reference yield fitting technique in the fund transfer pricing mechanism. In: Jajuga K, Locarek-Junge H, Orłowski L, Staehr K (eds) Contemporary trends and challenges in finance. Springer proceedings in business and economics. Springer, Cham EBA (2010) European Banking Authority guidelines on liquidity cost benefit allocation (CP 36), October. https://www.eba.europa.eu EIOPA (2016) Consultation paper on the methodology to derive the UFR and its implementation. https://eiopa.europa.eu/ Elliot V, Lindblom T (2015) Funds transfer pricing in banks: Implications of Basel III. In: Elliot V (ed) Essays on performance management systems, regulation and change in Swedish banks. BAS Publisher, Gothenburg Fleckenstein M, Longstaff FA, Lustig H (2014) The tips-treasury bond puzzle. J Fin 69(5):2151– 2197 Grant J (2011) Liquidity transfer pricing: a guide to better practice. Australian Prudential Regulation Authority Working Paper, Sydney IIF (2007) Institute of International Finance. Principles of liquidity risk management. Nelson CR, Siegel AF (1987) Parsimonious modeling of yield curves. J Bus 60:473–489 Smith A, Wilson T (2000) Fitting yield curves with long term constraints. Bacon & Woodrow Research Notes Svensson LEO (1994) Estimating and interpreting forward interest rates: Sweden 1992–1994. NBER Working Paper Series #4871 Van Deventer DR, Imai K, Mesler M (2005) Advanced financial risk management: tools and techniques for integrated credit risk and Interest rate risk management. Wiley, Singapore Wyle RJ, Tsaig Y (2011) Implementing high value funds transfer pricing systems. Technical report (september), Moody’s analytics. https://www.moodysanalytics.com/-/media/whitepaper/ 2011/11-01-09-implementing-high-value-fund-transferpricing-systems.pdf

Geometric Distribution as Means of Increasing Power in Backtesting VaR Marta Małecka

Abstract We explore properties of the geometric distribution as means of constructing conditional coverage VaR tests. We study properties of these tests using asymptotic convergence of the test statistics. In this way, we replace Monte Carlo simulated distributions. We provide a unified framework that allows for effective comparison of various procedures. To achieve comparability we modify test statistics and adapt them to the conditional coverage hypothesis. We show that two tests that indirectly use properties of the geometric distribution—the test based on the General Method of Moments and the test based on the Gini coefficient—may be conveniently implemented with the use of known theoretical distributions. We argue that replacing Monte Carlo simulations with these distributions does not pose the risk of overrejecting correct risk models. We also demonstrate their efficiency at detecting incorrect models. We include practical guidelines about significance level and sample size that ensure accurate and efficient testing.

1 Introduction The contemporary literature on extreme risk management often focuses on methods of testing Value-at-Risk (VaR). The reason for the importance of this topic is that the concept of VaR, well known in economics since 1994 (Morgan 1994), has been deemed fundamental for testing risk models by the global banking supervision (Basel Committee on Banking Supervision 1996, 2016, 2017). The VaR testing framework is commonly based on the violation process that indicates whether the actual returns go below the estimated VaR. Ideally, this process should satisfy the conditional coverage condition, which refers to the overall number of VaR violations and their independence. According to the mainstream contemporary literature, to improve the power of an early conditional coverage Christoffersen’s VaR test (1998), VaR violations are transformed into a duration process. Then the M. Małecka (B) Department of Statistical Methods, University of Łód´z, Łód´z, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_13

159

160

M. Małecka

concept of a duration is used for testing, which opens the possibility to utilize a wide range of methods related to the geometric distribution. The idea of the duration-based conditional coverage VaR test was first proposed by Christoffersen and Pelletier (2004). However, instead of the geometric distribution, they used the exponential distribution, being its continuous analogue. Their idea was continued by Haas (2005), who showed the superiority of discrete distributions over their continuous counterparts. Following this line, Berkowitz et al. (2011) proposed a discrete geometric test, which was generalized into the geometric-VaR test by Pelletier and Wei (2016). The geometric-distribution-based VaR backtesing methods were also extended by Candelon et al. (2011), who developed a General Method of Moments (GMM) VaR test, Ziggel et al. (2014), who proposed to improve power by including squared durations and Kramer and Wied (2015), who employed the Gini coefficient. The duration-based VaR tests have the main advantage of increasing the power of VaR backtesting. However, their real-life applicability is limited by the number of shortcomings. First, due to unavailability of test statistic distributions, practical implementation needs to be conducted through the Monte Carlo test technique. This requires time-consuming simulations. Second, the duration-based tests suffer from the rarity of observations. Third, statistical properties of the duration-based VaR tests cannot be compared in a straightforward manner. This is because some of them suffer from size distortions, being the consequence of using asymptotic distributions, while others rely on Monte Carlo simulated distributions. What is more, previous power studies use various data generating processes, with the consequence that the results cannot be compared. All these problems impede practical implementation of new VaR testing techniques, potentially more powerful than the early Christoffersen’s test. We seek to bridge the gap between propositions of testing VaR developed by the academia and expectations from financial market practitioners. To achieve this we reject any approaches based on Monte Carlo simulations and rely only on readilyavailable distributions. Our aim is to compare statistical properties of new, geometric distribution based methods proposed to test conditional coverage of VaR failures. For reference, we also provide their comparison to the early but influential conditional coverage VaR test, which uses the properties of the Markov chain. Our contribution to the existing literature is twofold. First, we present properties of the VaR tests implemented with the use of known distributions, which is aimed at bringing them closer to practical application. We, therefore, conduct the geometric and the geometricVaR tests relying on the asymptotic distribution of the likelihood ratio, instead of the originally proposed Monte Carlo simulated distributions. In the Gini test, we replace the simulated distribution with the asymptotic distribution of the Gini coefficient. Second, we provide a unifying framework that allows for a comprehensive comparison of the tests with respect to fundamental statistical properties—the size and the power. The unified framework includes a test hypothesis formulated in terms of the conditional coverage property, the size study based on asymptotic distributions and a power study conducted in a realistic setting, adjusted for size distortions. In order to have the same conditional coverage hypothesis tested by all examined statistics,

Geometric Distribution as Means of Increasing …

161

we modify the Gini test, so that it includes a parameter relating to the unconditional VaR coverage. The paper is structured as follows. In the second section, we present the recent geometric-distribution-related VaR tests as the alternative to the basic Markov-chain procedure. In the third section, we provide details of the Monte Carlo study of test properties and discuss the results. The final section summarizes and concludes.

2 Geometric Distribution Based Methods of Testing VaR The hypotheses in the VaR testing framework refer to the properties of the violation process, defined as It = 1{ Rt 0 and yi = 0 otherwise. Digit Z becomes the ending digit whenever the propensity to cluster is positive. In such circumstance the probability of stock price to cluster reads (Greene 2003, p. 680)  Pr(yi = 1|X i , Hi ) = 

 Xi β , exp(Hi γ )

where (·) is the c.d.f. for the standard normal distribution. Thus if variable Wi j is included in both X i and Hi the marginal effects that are of the ultimate interest yield   β j − (X i β)γ j Xi β ∂Pr(yi = 1|X i , Hi ) . =φ ∂ Wi j exp(Hi γ ) exp(Hi γ ) I estimate them in Stata using hetprobit followed by mfx command evaluated at the medians of independent variables. 3 The

price for any stock crossed the threshold except for JSW for which 63,278 observations are removed accordingly. 4 In case of the greater tick size the ending digits are 00, 05, …, 95.

172

P. Miłob˛edzki

3 Empirical Results The percentage of transactions executed at a price with the ending digit Z for stocks of the smaller minimum tick size is stacked in Table 1. The figures indicate that in the period in question the stock prices cluster on 0 and 5. The ending 0 appears in 18.53% of all transaction prices while the ending 5—in 11.77% of them. The percentage of all other ending digits is almost alike and falls in between 8.41 and 9.41%. The strongest tendency for price clustering exhibit Alior and JSW for which the percentage of ending 0 and 5 exceed the whole smaller minimum tick size stock average at 30.30% by approximately 9%. Any tendency for price clustering demonstrates PGNiG with the frequency of both digits sum equal to 20.79%. The figures for stocks of the larger minimum tick size document the presence of price clustering on the ending 00 alone with the frequency of 11.76% (see Table 2). The remaining double ending digits are almost equally distributed across all stocks with the frequency in between 4.03 and 6.71%. The strongest tendency for price clustering demonstrates LPP for which the appropriate frequency amounts to 34.65%. On the opposite side are KGHM and Orlen with the frequency of the ending 00 less than 10%. The results from the probit analysis are stacked in Tables 3 and 4. The estimates of the W1 test statistic ensure that the spread, S, and the log of trading volume, lnT V , do impact the propensity for price to cluster for all stocks. The estimates of the W2 test statistic support the decision to use a heteroskedastic probit. The estimates of the t(γ1 ) and t(γ2 ) test statistics indicate it is the trading volume that more often drives the variance of error terms than the spread does. The probability for stock price to cluster differs across the stocks of PLN 0.01 and PLN 0.05 minimum tick size being—on average—much higher for the first. The median probability in the first group of stocks (0.2876—PKO BP) is about 2.05 times as large as the median probability in the second group (0.1406—BZWBK). Nevertheless, when they are adjusted in the way to take into account the ending digit probabilities for ‘no clustering’ case, the difference in median probabilities becomes rather a pretty small.5 As the marginal effects are concerned the probability of stock price to cluster increases with the trading volume—as expected—but its increase across the stocks in the response to 1% increase in the trading volume is very much diverse ranging from 0.0076 (Energa) to 0.0279 (Alior) for the stocks of PLN 0.01 minimum tick size and from 0.0044 (Orlen) to 0.0603 (LPP) for the stocks of PLN 0.05 minimum tick size. The response to one tick increase in the spread is mixed in the sign, however. It amounts to 0.0330 for PGNiG and to 0.0257 for PGE while the effect for many other

5 In the case of ‘no clustering’ the probability of ending digit(s) Z (XZ) occurrence for the stocks of

PLN 0.01 (0.05) minimum tick size equals to 0.1 (0.05). In such circumstance the median probability for stocks from the first group is 0.2876/0.1 = 2.876 as large as that for the ‘no clustering’ case while the median probability for stocks belonging to the second group is 0.1406/0.05 = 2.812 as large.

193,553

131,474

166,759

139,909

228,874

203,636

269,089

282,932

124,624

279,752

161,599

380,782

410,703

2,973,686

Asseco

Polsat

Enea

Energa

Eurocash

JSW

Lotos

Orange

PGE

PGNiG

PKO BP

PZU

All

N

Alior

Stock

18.53

20.09

17.47

11.11

13.26

12.56

19.61

27.36

20.41

14.83

14.73

17.21

19.73

26.57

0

Digit 8.97

8.18

8.72

8.89

8.96

9.44

8.24

7.79

8.92

9.72

8.72

7.79

8.64

10.85

1

8.49

7.94

8.97

9.66

8.85

8.84

8.65

7.38

8.22

9.22

8.45

8.37

8.71

7.79

2

8.41

7.90

8.81

9.68

9.33

10.07

7.80

6.67

8.49

9.22

8.94

8.56

8.46

6.92

3

8.65

8.53

8.74

10.26

9.22

10.89

8.37

6.75

8.25

9.72

9.04

8.96

7.92

7.19

4

11.77

12.92

11.96

9.68

11.47

10.79

12.10

12.13

11.70

11.28

11.65

10.53

11.68

12.66

5

Table 1 Percentage of ending digit Z for the stocks of PLN 0.01 minimum tick size (Z %)

8.51

8.09

8.68

9.83

9.59

9.30

8.30

7.32

8.27

8.44

9.30

8.80

8.80

7.16

6

8.48

8.16

8.72

9.94

9.76

9.59

8.26

7.05

8.27

8.53

9.65

8.92

8.03

6.41

7 7.57

9.90

9.51

8.37

9.03

8.87

9.00

9.92

10.34

10.13

8.60

7.54

8.37

10.14

8

8.77 8.57

9.41

9.72

9.01

9.07

10.00

9.11

9.40

8.84

8.58

10.38

10.56

10.23

9

30.30

33.01

29.43

20.79

24.73

23.35

31.71

39.49

32.11

26.11

26.38

27.74

31.41

39.23

0+5

Price Clustering in Stocks from the WIG 20 Index 173

1,627,165

N

133,040

147,660

493,241

78,978

92,706

377,300

304,240

1,627,165

Stock

BZWBK

CCC

KGHM

LPP

MBank

Orlen

PeKaO

All

92,706

MBank

All

78,978

LPP

377,300

493,241

KGHM

304,240

147,660

CCC

PeKaO

133,040

BZWBK

Orlen

N

Stock

6.71

7.48

6.69

5.94

2.88

6.74

7.30

7.09

50

Digits

11.76

10.15

7.71

16.85

34.65

9.45

15.18

14.52

00

Digits

4.13

3.64

4.59

4.25

2.38

4.23

4.27

4.43

55

4.57

4.29

4.36

5.51

6.62

4.24

4.57

5.11

05

4.81

4.52

5.39

4.09

2.42

5.36

4.35

4.28

60

4.90

4.85

4.73

4.93

3.83

5.44

4.29

4.76

10

4.03

3.64

4.78

3.76

2.27

4.23

3.76

3.52

65

4.29

4.41

4.39

3.86

3.04

4.59

3.84

4.19

15

4.60

4.29

5.00

4.14

2.63

5.12

4.18

4.19

70

4.90

5.70

4.94

4.26

2.46

5.18

4.42

4.32

20

Table 2 Percentage of ending digits X Z for the stocks of PLN 0.05 minimum tick size (Z %) 25

4.24

4.08

4.71

3.97

3.03

4.25

4.06

4.27

75

4.32

5.00

4.60

3.75

2.78

4.15

4.00

4.26

30

4.94

4.95

5.06

4.75

3.32

5.21

4.77

4.80

80

4.64

4.99

4.91

4.26

2.28

4.87

4.51

4.07

35

4.33

4.27

4.64

4.08

3.71

4.49

4.03

3.90

85

4.11

4.33

4.49

3.82

2.45

4.12

3.69

4.11

40

5.07

5.15

5.08

4.58

4.97

5.33

4.84

4.61

90

4.70

5.37

5.19

3.64

2.41

4.70

4.38

4.28

45

4.80

4.39

4.31

5.89

9.59

4.25

5.20

5.07

95

4.15

4.48

4.43

3.68

2.28

4.05

4.36

4.23

174 P. Miłob˛edzki

Price Clustering in Stocks from the WIG 20 Index

175

Table 3 Results from the heteroskedastic probit for the stocks of PLN 0.01 minimum tick size Stock

Marginal effecta

Statistic W1

W2

t (γ1 )

t (γ2 )

S

lnT V

y

987.62*

168.07*

1.69

−12.95*

0.0477*

0.0279*

0.3874

Asseco

1253.75*

399.16*

4.28*

−19.02*

0.0264

0.0207*

0.3085

Polsat

2453.08*

462.38*

0.95

−21.48*

−0.0895*

0.0187*

0.2709

Enea

1655.53*

988.97*

8.05

13.89*

0.8198*

0.0132*

0.2525

Energa

3112.22*

373.45*

0.06

−18.91*

1.4337*

0.0076*

0.2442

Eurocash

1563.57*

718.36*

−0.48

−26.80*

0.0084

0.0257*

0.3119

JSW

568.12*

287.85*

0.19

−16.45*

0.0401*

0.0182*

0.3872

Lotos

3303.98*

668.83*

−0.08

−25.83*

0.1085*

0.0215*

0.3109

935.76*

334.90*

5.75*

8.08*

1.8071*

0.0081*

0.2244

Alior

Orange

3088.48*

125.73*

4.19*

−10.85*

2.5744*

0.0121*

0.2305

PGNiG

291.57*

66.41*

6.09*

−5.63*

3.3003*

0.0088*

0.2019

PKO BP

3051.53*

328.75*

1.52

−17.49*

0.2543*

0.0161*

0.2876

PZU

1862.68*

902.68*

6.16*

−28.61*

0.8529*

0.0171*

0.3138

PGE

W1 (W2 )—Wald test statistic under H0 : β1 = β2 = 0 (H0 : γ 1 = γ2 = 0) distributed as χ 2 (2) t (γi ) statistic under H0 : γi = 0 distributed as N (0, 1); a evaluated at variable medians y = Pr(Z = 0 ∨ 5); *—significant at the 5% significance level

Table 4 Results from the heteroskedastic probit for the stocks of PLN 0.05 minimum tick size Marginal effecta

Stock

Statistic t (γ2 )

S

lnT V

y

BZWBK

3445.21*

460.85*

0.56

−18.52*

−0.0213*

0.0225*

0.1406

CCC

6133.48*

1142.69*

−1.56

−32.71*

−0.0137*

0.0171*

0.1434

KGHM

4362.21*

631.63*

5.92*

−24.62*

0.2237*

0.0087*

0.0928

LPP

1964.01*

432.21*

7.84*

−13.69*

−0.0044*

0.0603*

0.3475

MBank

3922.72*

925.94*

8.98*

−27.23*

−0.0152*

0.0235*

0.1572

Orlen

3088.89*

680.20*

8.11*

−23.78*

0.1629*

0.0044*

0.0677

PeKaO

5446.01*

580.88*

−23.99*

0.0865*

0.0098*

0.0979

W1

W2

t (γ1 )

−1.08

W1 (W2 )—Wald test statistic under H0 : β1 = β2 = 0 (H0 : γ 1 = γ2 = 0) distributed as χ 2 (2) t (γi ) statistic under H0 : γi = 0 distributed as N (0, 1); a evaluated at variable medians y = Pr(Z = 00); *—significant at the 5% significance level

stocks (Alior, Polsat, Eurocash, JSW, Lotos, PKO BP, BZWBK, CCC, LPP, MBank, Orlen, PeKaO) —although statistically significant—is negligible in the magnitude.

176

P. Miłob˛edzki

4 Conclusion I use a high quality transaction data on the stocks included in the WSE index WIG 20 from May to September 2017 and find that their prices tend to cluster on certain final digits as 0, 5 and 00. Having employed the probit analysis I reveal that the tendency for stock prices to cluster generally increases with increases in the traded volumes but not in the spreads. The response to one tick increase in the spread across the stocks is mixed in the sign and for the most of them—although statistically significant— is negligible in the magnitude As the traded volumes and the spreads at the WSE exhibit either the day of the week or the hour of the day effect or both (Miłob˛edzki and Nowak 2018b), a detailed analysis of price clustering within the trading day is left for a further research.

References Anderson H, Dunstan A, Marshall BR (2015) Cultural stock price clustering in the Chinese equity market. Chin Econ 48:449–467 ap Gwilym O, Meng L (2010) Size clustering in the FTSE100 index futures market. J Futures Markets 30:432–443 Baig A, Blau BM, Sabah N (2019) Price clustering and sentiment in bitcoin. Finance Res Lett 29:111–116 Bharati R, Crain SJ, Kaminski V (2012) Clustering in crude oil prices and the target pricing zone hypothesis. Energy Econ 34:1115–1123 Ball CA, Torus WN, Tschoegl AE (1985) The degree of price resolution: The case of the gold market. J Futures Markets 5:29–43 Brown P, Chua A, Mitchell J (2002) The influence of cultural factors on price clustering: evidence from Asia-Pacific stock markets. Pacific-Basin Finance J 10(3):307–332 Brown P, Mitchell J (2008) Culture and stock price clustering: evidence from The Peoples’ Republic of China. Pacific-Basin Finance J 16:95–120 Cai BM, Cai CX, Keasey K (2007) Influence of cultural factors on price clustering and price resistance in China’s stock markets. Account Finance 47:623–641 Chaudhry SM, Bajoori E, Nandeibam S (2019) Clustered pricing in the corporate loan market: theory and empirical evidence. J Econ Behav Organ 157:275–296 Chen T (2014) Price clustering and price barriers: international evidence. Nang Yan Bus J 3(1):1–16 Chiao C, Wang ZM (2009) Price clustering: evidence using comprehensive limit-order data. Financ Rev 44:1–29 Das S, Kadapakkam PR (2018) Machine over mind? Stock price clustering in the era of algorithmic trading. North Am J Econ and Finance (in press) Greene WH (2003) Econometric analysis. Prentice Hall, Upper Saddle River Grossman SJ, Miller MH, Cone KR, Fischel DR, Ross DJ (1997) Clustering and competition in asset markets. J Law and Econ 40:23–60 Harris L (1991) Stock price clustering and discreteness. Rev Finance Stud 4(3):389–415 He Y, Wu C (2006) Is stock price rounded for economic reasons in the Chinese markets? Global Finance J 17(1):119–135 Hu B, Jiang C, McInish T, Zhou H (2017) Price clustering on the Shanghai Stock Exchange. Appl Econ 49(28):2766–2778 Huang RD, Stoll HR (2001) Tick size, bid-ask spreads, and market structure. J Finan Quant Anal 36(4):503–522

Price Clustering in Stocks from the WIG 20 Index

177

Ikenberry DL, Weston JP (2007) Clustering in US stock prices after decimalisation. Eur Finan Manag 14(1):30–54 Kleidon AW, Willig RD (1998) Why do Christie and Schultz infer collusion from their data? Available at SSRN: https://ssrn.com/abstract=6680 Lien D, Hung PH, Hung IC (2019) Order price clustering, size clustering, and stock price movements: evidence from the Taiwan stock exchange. J Empirical Finance 52:149–177 Miłob˛edzki P, Nowak S (2018a) The accuracy of trade classification rules for the Warsaw Stock ´ Exchange. In: Papie˙z M, Smiech S (eds) The 12th Professor Aleksander Zelia´s international conference on modelling and forecasting of socio-economic phenomena proceedings. Foundation of the Cracow University of Economics, Kraków, pp 316–325 Miłob˛edzki P, Nowak S (2018b) Intraday trading patterns on the Warsaw Stock Exchange. In: Jajuga K, Orłowski LT, Staehr K (eds) Contemporary trends and challenges in finance. Springer Proceedings in Business and Economics. Springer International Publishing, Cham, pp 55–66 Mishra AK, Tripathy T (2018) Price and trade size clustering: evidence from the national stock exchange of India. Q Rev Econ Finance 68:63–67 Narayan PK, Narayan S, Popp S, D’Rosario M (2011) Share price clustering in Mexico. Int Rev Financ Anal 20(2):113–119 Niederhoffer V (1965) Clustering in stock prices. Oper Res 13:258–265 Niederhoffer V (1966) A new look at clustering in stock prices. J Bus 39:309–313 Ohta V (2006) An analysis of intraday patterns in price clustering on the Tokyo Stock Exchange. J Bank Finance 30:1023–1039 Osborne MFM (1962) Periodic structure in the Brownian motion of stock prices. Oper Res 10(3):345–379 Palao F, Pardo A (2012) Assessing price clustering in European carbon markets. Appl Energy 92:51–56 Sonnemans J (2006) Price clustering and natural resistance points in the Dutch stock market: a natural experiment. Eur Econ Rev 50:1937–1950 Sopranzetti BJ, Datar V (2002) Price clustering in foreign exchange spot markets. J Finan Markets 5:411–417 Uchwała (2017) The Warsaw Stock Exchange detailed exchange trading rules in UTP system adopted by Resolution No. 1038/2012 of the WSE Management Board dated 17 October 2012, including amendments adopted by the end of 2017. https://www.gpw.pl/pub/files/PDF/regulacje/ SZOG_pl.pdf. Accessed 20 July 2019 Urquhart A (2017) Price clustering in bitcoin. Econ Lett 159:145–214

Construction of Investment Strategies for WIG20, DAX and Stoxx600 with Random Forest Algorithm Grzegorz Tratkowski

Abstract Machine learning provides powerful tools for data analysis, especially in regression and classification problems what may be used in creation of investment strategies. This paper present an efficient way of utilization of one of the machine learning algorithms on examples of stock indices: Stoxx600, WIG20 and DAX. This work concentrates on time series analysis of stock indices with Random Forest algorithm to create investment strategies based on future probabilities of declines and upswings. Taking into account some macroeconomic characteristics, technical indicators and consensus estimates, the models are trained to provide a buy signal if the output probability is above a specific threshold and sell signal in case of the opposite situation. The examination of the strategies efficiency indicates the differences in determinants among chosen stock indices.

1 Introduction This paper investigates the problem of the efficient usage of Random Forest algorithm directed for the creation and verification of investment strategies. The word efficient in the previous statement means that the algorithm is used in such a manner that the maximum amount of information is derived from the dataset as in many published papers it is not properly conducted. The area of interest of using complex algorithms for construction of investment strategies has been widely covered, mainly in forecasting equity indices, however in most papers the Machine Learning algorithms have not been correctly used thus the research projects are not comparable. The papers can be differentiated in many ways. Firstly, the inputs taken into the models are usually only closing prices (Halliday 2004; Hansson 2017), but sometimes some additional variables are used like some technical indicators (Kara et al. 2011) or financial indicators (Cao et al. 2011). Secondly, the evaluation metrics differ among G. Tratkowski (B) Wrocław University of Economics and Business, Wrocław, Poland e-mail: [email protected]

© Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_15

179

180

G. Tratkowski

the authors: Janeski and Kalajdziski (2011) used Mean Squared Error and Mean Absolute Error, Atsalakis and Valavanis (2009) used HIT ratio and Root Mean Square Error. Thirdly, the strategies or forecasts are created for specific markets and periods: Moghaddam et al. (2016) analyzed a 99-day period for NASDAQ and Qiu et al. (2016) 237-month period for Nikkei 225. Other differences, which can be outlined, are algorithms’ selection, data regularization method, regression or classification method of forecasting, training–testing data split or methods used for comparison. In this work, the author suggests another approach in tackling the problem of using the Machine Learning algorithm called Random Forest for the creation of investment strategies. The additional purpose of this research is to highlight good practices and describe all the steps needed to use ML algorithms properly. For presentation the three indices were chosen: Polish WIG20 which comprises of the largest twenty companies listed on the Warsaw Stock Exchange, DAX which contains thirty largest German stocks listed on the Frankfurt Stock Exchange and Stoxx Europe 600 including six hundred companies from seventeen European countries. The whole dataset included the period from January 2004 to July 2019. The Random Forest was chosen as a representative of the Machine Learning algorithm. The remaining part of this research paper is organized into 4 sections. Section 2 the general idea of investment strategy and some characteristics of Random Forest algorithm. Section 3 describes the steps needed for correct data preparation in order to run is through the algorithm. Section 4 summarizes the methodology of training the algorithm on the dataset. Section 5 concerns the evaluation of the strategies, discussion and conclusion.

2 Investment Strategy and Random Forest An investment strategy for portfolio manager is a set of rules determining in which kind of financial instruments to invest and in what time (Basile et al. 2016). The investment strategy requires identification of the investment’s purpose and its limitations. The set of rules should allow the investor to recognize a signal either to buy an asset or sell it in appropriate time. In this paper the signals would be based on probabilities of next index upswing derived from Random Forest algorithm. In reference to the probability of the index’s increase is greater than determined threshold, it is a buy signal. If it is below this specific level, a sell signal is derived. Other example of a strategy based on probability can use two thresholds: for sell, buy and “do nothing” signals. When the strategy’s framework is determined, the investor should test it on historical data to verify its past performance. This process is known as a backtesting. The simplest way to evaluate the strategy is put its performance in comparison to the benchmark. A good benchmark for strategy based on the stock index or only one asset is the index or the asset itself, in other words ‘buy and hold’ strategy. In the following research the described approach was introduced and the strategy was evaluated, based on the assumptions of the beating the index.

Construction of Investment Strategies for WIG20, DAX …

181

Random Forest belongs to the ensemble classification and regression algorithms of Machine Learning (Breiman 2001). The name comes from the fact that the structural elements of the algorithm are decision trees in a mean of predictors. Each of the trees is constructed with some portion of randomness, so finally all trees should be differentiated after the training time. The results from the predictive process of Random Forest algorithm are derived as a majority voting of the trees in classification problems and as a mean prediction in regression problems. Because of the random initiation of decision trees, the algorithm is designed to handle large datasets. Thanks to the all above characteristics, Random Forest might be suitable to analyze complex economic environments and might be considered as helpful in the creation of investment strategies by predicting assets price changes. One of the main advantages of this algorithm is that Random Forest could solve both classification and regression problem by the use of categorical and continuous input variables. Thanks to the structure of multiple decision trees it can automatically handle some missing values in dataset, as the trees are build randomly by a limited number of variables. A similar situation refers to outliers, the outstanding datapoints are only taken into consideration by some limited amount of trees, so it minimizes their impact or eliminates it completely. Random Forest is considered as a very stable algorithm in case of changes in the inputs. If a new datapoint is introduced in the dataset, the structure of the algorithm would not vary much as the new point may affect very limited number of trees and not all of them. The main and the most important disadvantage of Random Forest is its complexity. It requires the a lot of decision trees and combines their results, so it demands much more computational power than single decision tree. However the interpretation of the whole algorithm is more complicated than standard models, but still it is simpler to track the decision process than in other Machine Learning algorithms like neural networks, which are sometimes considered as black boxes.

3 Data Preparation For the purpose of this research, the monthly data from January 2004 to July 2019 for WIG20, DAX and Stoxx600 was in the form of 187 observations for each index. The data included some financial indicators for each index like price, dividend yield, P/E ratio, market estimates on 12 and 18 months forward financials and growths, actual financials on EPS, EBIT, Sales, etc. In total, 51 variables for each index were used. Besides, some macroeconomic indicators were used as a common part for the indices: commodities prices, inflation, interest rates, bond yields, unemployment, exchange rates, and other economic indicators; 52 macroeconomic variables in total. All the variables used are listed in Tables 2 and 3 in Appendix. When the input variables are not in the same range, for example, the oil price is expressed in dollars and bond yields in percentages, the data needs normalization to keep similar variances among variables (Specht 1991). The normalization can make

182

G. Tratkowski

the training process more efficient in time and results (Jayalakshmi and Santhakumaran 2011). There are several types of normalization, but usually, the Min-Max method (i.e. Li and Xiong 2005) or Z-Score method (Pan et al. 2016) are used. When it comes to the time series analysis and Machine Learning usage the normalization process should be done on a rolling basis, which means that the statistics taken into consideration in normalization methods should only be based on historical data. For example, in the Z-Score method expressed in Eq. 1, every next observation would take other average and standard deviation than the previous one: 

x =

(xi − μi ) . σi

(1)

The rolling basis can be achieved in two ways. The first one is to roll a fixed window over time so that the average (μi ) and standard deviation (σi ) would always be calculated with a specific number of values. The second approach is to always capture the beginning of the dataset and expand the interval over new entries. In this paper, the Z-Score approach with a fixed 5 years rolling window was adopted, but additionally, the absolute or percentage changes (dependings on the datatype) were kept as input variables. The normalization of the observations finished the data preparation step. All 41 index and 62 macroeconomic variables were transformed into changes and 5 years rolling z-scores yielded 206 variables in total for each index.

4 Training the Algorithm The Random Forest algorithm is a combination of different decision trees randomly selected for predictions, which can be used for solving classification and regression problems (Breiman 2001). The training is based on a binary classification problem where 1 stands for next month upswing and 0 for a decrease or no change in index value over a month. This approach in training the algorithm is called supervised learning and requires an additional supervised binary series so that the input variables are mapped to 1 if the change of the index over the next month is positive and to 0 otherwise. In most of the research papers on using the Machine Learning to forecast a time series or create a strategy, the dataset is split on training and testing subsets, for example, 70% to 30% respectively. However, this approach loses the ability of the algorithm to learn over time. Hart (1994) suggested a time series cross-validation for models describing a one step forward prediction. Similarly to the rolling approach in Z-Score, the time series cross-validation concerns a window rolling over the dataset. In this paper, the rolling period of 5 years is used and it means that the model is fitted on the first 60 observations and the forecast is performed only on the 61st observation. After that training, the rolling window moves forward and includes observations from the second to the 61st and forecasts only next month’s market move. This approach catches the latest available data and limits the need for splitting

Construction of Investment Strategies for WIG20, DAX …

183

the dataset on training and testing subsets. Only the first sixty observations can be treated as training data. The optimization of the hyperparameters should always be a part of the Machine Learning training process. Hyperparameters are these which should be set before the training begins and actual model parameters are derived from the learning process. There are several methods of finding a good set of hyperparameters: manual search, grid search or random search (Bergstra and Bengio 2012). In the case of the Random Forest, the examples of hyperparameters are the number of trees, the maximum depth of a tree, and the number of variables in each split inside a tree. There is no specific rule determining which set of hyperparameters would be optimal for a certain dataset and that is why the current methods are based on some type of searching for the best combination. In this paper the values of hyperparameters were derived using the grid search method, which fits the model with one specification, measures the accuracy of the model and changes for other specification. The next step in optimizing the training process is to perform the correct variables selection. This operation may reduce the chances of overfitting and increases the effectiveness of the model. Random Forest algorithm enables to get relative variables’ importance, which is called permutation importance (Genuer et al. 2010). The process involves fitting a model, permutation of the input variable, calculation of the error on the output, taking the average of errors for each variable and creation of relative ranking. Calculation of variables importance provides a tool for variable selection process called Recursive Feature Elimination. In every step, the importance of all used input variables is determined and the measure of the model’s effectiveness is calculated. Then the least important variable is removed the whole process is repeated. The input variables subset is chosen as the one having the best evaluation. If there are two or more subsets with the same score, the one with the lowest number of variables is chosen. The last important feature that needs to be determined before training the algorithm is to choose the effectiveness measure. When it comes to regression problems, a lot of metrics exist like Mean Absolute Error or Root Mean Square Error. However, for classification problems, the measures are limited to the output of a confusion matrix, which describes the amounts of predicting true positives, true negatives and false negatives, false positives. One measure derived from this matrix is accuracy (HIT ratio), which describes the percentage of true positives and true negatives to all predictions. Even though this metric seems to be suitable for binary classification for investment strategy, there are two problems connected with it. The first issue is the possible misinterpretation of the accuracy ratio without taking into consideration the imbalance of the data—the accuracy of 75% would be a score below expectations if there are 80% chances of a correctly and randomly guessing a value. Another issue with this metric is that from the investment strategy perspective, it would be more important to correctly forecast the index move when the move itself would be larger, as small moves in the index would have a less significant influence on the strategy’s performance. Despite these facts, the accuracy metric was chosen, but finding a better evaluation score suitable for investment strategies might be a subject of further research.

184

G. Tratkowski

Since the rolling approach is used, all steps are repeated every time the rolling window moves forward. It means that all hyperparameters and variables used in a model in a specific point in time may be different, thus not a single model is trained for the whole dataset, but instead for several ones. Once all the above is determined, the models forecast the probability of the next market move. The probability is determined as the average of each tree’s forecast, so the output is a time series of predicted probabilities for each analyzed index. The whole described above process was performed using Python language with PyCharm serving as an environment.

5 Results and Conclusions The results of the predictions for each index are presented in Fig. 1. The last necessary step to derive an investment strategy is determining the buy and sell signals out of the probability. Assuming that both long and short positions are available on each of the indices, an example of a strategy that can be obtained from probability is the sell signal if the probability is below a certain threshold, and buy signal if it is above. The threshold level was determined in 3 years period by maximizing the performance within that interval. The results of the adaptation of this strategy for each index are presented on as red dashed lines in each panel in Fig. 1 and the performances of the indices and strategies are presented in Table 1. The obtained results are satisfying for WIG20 and Stoxx 600 indices as the performance of the strategies are above the indices. Applying the same methodology for these three time-series ended with different outcomes. The DAX index next month move proved to be the hardest to predict by this algorithm, as the strategy based on the forecasted probability was not able to outperform the market. Both Table 1 and Fig. 1 indicates that in cases of WIG20 and Stoxx600 the strategies resulted in significantly better returns than passively investing in the index. For WIG20 the strategy outperformed the market by 3.9% annually over the ten years period. Strategy derived on Stoxx600 managed to achieve 2.2% of return above the annualized performance of the index. In case of DAX the strategy underperformed the index by 1.9%. The hit ratios of strategies and indices are not significantly different from each other. That may indicate that such a metric does not include difference in power of price movements, so having a very similar hit ratios but the quite differentiated returns means that the strategies were able to correctly guess larger moves in case of WIG20 and Stoxx600. A worse result for DAX strategy might be caused by the fact that DAX is a total return index whereas the Stoxx600 and WIG20 are price indices. Total return means that the value of the index incorporates dividends and it might imply its abnormal movements in comparison to other market variables. However this issue requires a further research. The main purpose of this paper was to show the methodology of creating an investment strategy with the usage of the Machine Learning algorithm. The results show

Construction of Investment Strategies for WIG20, DAX …

185

1.0

4500 4000

0.8

3500 3000

0.6

2500

0.4

2000 0.2

1500 1000 2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

0.0

1.0

550 500

0.8

450 400

0.6

350 0.4

300 250

0.2

200 150 2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

0.0

15000

1.0

13000

0.8

11000

0.6

9000 0.4

7000

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

0.0 2006

3000 2005

0.2

2004

5000

Fig. 1 Price index (black line) since Feb’04, predicted probability (grey bars) since Mar’09 and strategy long and short performance (red dashed line) for WIG20 (top panel), Stoxx 600 (middle panel) and DAX (bottom panel)

186

G. Tratkowski

Table 1 Indices and strategies’ performances Index

Total return (%)

4.9

Hit ratio (%)

Index return Strategy return

141.5

8.8

49.6

Stoxx600

Index return

123.1

7.9

58.7

Strategy return

175.4

10.1

58.4

Index return

217.1

11.6

57.9

Strategy return

165.6

9.7

56.8

DAX

65.9

Annualized return (%)

WIG20

47.6

that this approach is not universally effective for different indices, but presents some evidence for its effectiveness in investment strategies. Further research is needed in the investigation on the usefulness of this methodology for other classes of assets, like commodities or fixed-income assets. Additionally, finding a better evaluation metrics suitable for investment strategies may enhance the forecasting power of this approach.

Appendix See Tables 2 and 3.

Table 2 List of specific variables for each index Dividend yield

12m fwd YOY growth EBIT

3m % in 12m fwd CPS

Total return index

12m fwd YOY growth sales

3m % in 12m fwd DPS

Price

12m trl BPS

3m % in 12m fwd EBIT

Price high

12m trl CPS

3m % in 12m fwd sales

Price low

12m trl DPS

3m % in 18m fwd BPS

Historical EPS growth

12m trl EBIT

3m % in 18m fwd CPS

Long term average growth

12m trl sales

3m % in 18m fwd DPS

Market value

18m fwd BPS

3m % in 18m fwd EBIT

Opening price

18m fwd CPS

3m % in 18m fwd sales

P/E ratio

18m fwd DPS

6m % in 12m fwd BPS

12m fwd BPS

18m fwd EBIT

6m % in 12m fwd CPS

12m fwd CPS

18m fwd P/B

6m % in 12m fwd DPS

12m fwd DPS

18m fwd P/C

6m % in 12m fwd EBIT

12m fwd EBIT

18m fwd P/D

6m % in 12m fwd Sales

12m fwd P/B

18m fwd sales

6m % in 18m fwd BPS

12m fwd P/C

18m fwd YOY growth BPS

6m % in 18m fwd CPS (continued)

Construction of Investment Strategies for WIG20, DAX …

187

Table 2 (continued) 12m fwd P/D

18m fwd YOY growth CPS

6m % in 18m fwd DPS

12m fwd Sales

18m fwd YOY growth DPS

6m % in 18m fwd EBIT

12m fwd YOY growth BPS

18m fwd YOY growth EBIT

6m % in 18 m fwd sales

12m fwd YOY growth CPS

18m fwd YOY growth Sales

12m fwd YOY growth DPS

3m % in 12m fwd BPS

Table 3 List of common variables Austria—unemployment rate

Oil brent $

Copper price

PLN/USD

EU OECD CLI

Polish 10Y yield

EU OECD CLI normalized

Polish inflation

EU OECD CLI trend restored

Polish interest rate

EUR TWI

Polish interest rate. 1

Euro CPI

Polish OECD CLI amplitude adjusted

Euro export prices

Polish OECD CLI normalized

Euro interbank offer rate adjusted

S&P500

Euro money M3

SHANGHAI SE index

Euro retail sales

Sweden PMI

G7 OECD CLI amplitude adjusted

Synthetic EUR TWI

German 10Y yield

UK benchmark 10Y price index

German CPI total

US benchmark 10Y price index

German HICP

US OECD CLI amplitude adjusted

German interest rate

US OECD CLI normalized

German OECD CLI amplitude adjusted

US OECD CLI trend restored

German OECD CLI normalized

US treasury yield

German OECD CLI trend restored

USD TWI

German unemployment rate

USD/CNY

Germany benchmark 10Y price index

USD/EUR

References Atsalakis GS, Valavanis KP (2009) Forecasting stock market short-term trends using a neuro-fuzzy based methodology. Expert Syst Appl 36(7):10696–10707 Basile I et al (2016) Asset management and institutional investors. Springer International Publishing, Switzerland Bergstra J, Bengio Y (2012) Random search for hyper-parameter optimization. J Mach Learn Res 13(Feb):281–305 Breiman L (2001) Random forests. Mach Learn 45(1):5–32 Cao Q et al (2011) The three-factor model and artificial neural networks: predicting stock price movement in China. Ann Oper Res 185(1):25–44

188

G. Tratkowski

Genuer R et al (2010) Variable selection using random forests. Pattern Recogn Lett 31(14):2225– 2236 Halliday R (2004) Equity trend prediction with neural networks. Res Lett Inf Math Sci 6:15–29 Hansson M (2017) On stock return prediction with LSTM networks. Lund University, Lund Hart JD (1994) Automated kernel smoothing of dependent data by using time series cross-validation. J Roy Stat Soc 56(3):529–542 Janeski M, Kalajdziski S (2011) Neural network model for forecasting Balkan stock exchanges. In: International conference on intelligent computing. Springer, Berlin, Heidelberg, pp 17–24 Jayalakshmi T, Santhakumaran A (2011) Statistical normalization and back propagation for classification. Int J Comput Theory Eng 3(1):1793–8201 Kara Y et al (2011) Predicting direction of stock price index movement using artificial neural networks and support vector machines: the sample of the Istanbul Stock Exchange. Expert Syst Appl 38(5):5311–5319 Li RJ, Xiong ZB (2005) Forecasting stock market with fuzzy neural networks. In: 2005 International conference on machine learning and cybernetics, vol 6. IEEE, pp 3475–3479 Moghaddam AH et al (2016) Stock market index prediction using artificial neural network. J Econ Finan Adm Sci 21(41):89–93 Pan J et al (2016) The impact of data normalization on stock market prediction: using SVM and technical indicators. In: International conference on soft computing in data science. Springer, Singapore, pp 72–88 Qiu M et al (2016) Application of artificial neural network for the prediction of stock market returns: the case of the Japanese stock market. Chaos Solitons Fractals 85:1–7 Specht DF (1991) A general regression neural network. IEEE Trans Neural Networks 2(6):568–576

Application of the SAW Method in Credit Risk Assessment Aleksandra Wójcicka-Wójtowicz, Anna Łyczkowska-Han´ckowiak, and Krzysztof Piasecki

Abstract Credit risk assessment usually is a complex process, which consists of many successive steps and numerous criteria. Selection of good customers and rejection of potentially bad ones is vital as it directly and significantly affects the quality of bank’s credit portfolio. Also, ordering the decision alternatives is an important part of the whole decision-making analysis which takes place before making a final decision. The importance and complexity of the problem on one hand call for strictly analytical methods, however, on the other, also for a method which enables intuitive decision-making, imprecision and inaccurate linguistic ranks based on experts’ personal experience. The paper presents the utility of Simple Additive Weighting method in case of a credit risk assessment. The presented illustrative example bases on experts’ knowledge and their perception and evaluation of various linguistic, frequently imprecise criteria. Therefore, the order scale is described by trapezoidal oriented fuzzy numbers.

1 Introduction The process of a credit risk evaluation is a complex one and usually consists of at least three main steps. Firstly, the potential borrower must go through a scrutinized analysis of their financial statements (usually covering at least three successive years). Secondly, it is classified into a rating class and the recommendation is given. At that point, from a risk management perspective, transparency throughout the process is crucial and it allows the bank to see an accurate and comparable picture. However, this A. Wójcicka-Wójtowicz (B) · K. Piasecki Pozna´n University of Economics and Business, Pozna´n, Poland e-mail: [email protected] K. Piasecki e-mail: [email protected] A. Łyczkowska-Han´ckowiak WSB in Pozna´n, Pozna´n, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_16

189

190

A. Wójcicka-Wójtowicz et al.

is also the point in this process, at which the measurable and quantitative process stops as finally, the decision is taken at the meeting of a credit assessment committee. That is why such decision making is unstable and changeable. The mentioned Committee consists of a number of experts—banks’ employees (higher level managers). Their main role is to make a final decision based on their experience but also their ability to look at the future projections of the borrower’s business and not just their past performance. Any underwriting agreements, financial projections and the health of the borrower’s industry are all very important, as they will be leading indicators of potential volatility in loan payments, however, assessed features can frequently be conflicting or excluding one another. The final decision can be consistent with the recommendation or it can reject the recommendation. The potential borrowers may be evaluated by means of a rating scale assessed by Oriented Fuzzy Numbers (OFN) (Wójcicka-Wójtowicz and Piasecki 2019). Therefore, the main subject of our paper is the problem of an evaluation of the potential debtor with the use of OFNs. We take into account a multiple attributes decision-making approach based on Simple Additive Weighting (SAW) method. SAW method is applied to deal with qualitative dimensions. The SAW process considers the professional experience of each expert and the total scores for alternative features. The utilization of the considered method provides experts with a support system allowing them to make a decision in a shorter time. For these reasons, we apply Oriented Fuzzy SAW (OFSAW) method (Piasecki and Roszkowska 2018) for solving the problem of assessing the standing of the potential debtor. The proposed system integrates fuzzy set theory and the SAW method to evaluate the available alternatives. SAW method was further developed in (Piasecki et al. 2019b). To simplify the perception of this paper, some elements of the above-mentioned article, are repeated in that research. The paper is organized as follows. Section 2 presents the overview of credit risk assessment methods. Section 3 outlines the mathematical formulation of oriented fuzzy numbers (OFNs). The general linguistic approach to the borrowers’ evaluation is presented in Sect. 4. The OF-SAW method is described in Sect. 5. Section 6 presents the numerical example, which illustrates the procedure of the proposed OFSAW method and contributes into the understanding of the process of borrowers’ evaluation. Finally, Sect. 7 concludes the article, summarizes the main findings of this research and proposes some future research directions.

2 Credit Risk Assessment Methods—Overview The Basel Capital Accord within the Internal Rating Based Approach (IRB) allows banks to use their own (or chosen) rating models for the estimation of probabilities of default (PD). Also, banks can use various approaches to classify the borrowers or to create their ranking and, as a result, attribute a specific rating. The methods, which are most widely used in credit risk assessment and the evaluation of borrowers,

Application of the SAW Method in Credit Risk Assessment

191

usually belong to a group of parametric methods such as linear discriminant analysis (e.g. Altman 1968; Karels and Prakash 1987), regression analysis (West 2000), credit scoring (Bonilla et al. 2000; Wallis 2001) and non-parametric methods such as decision trees (Kabari and Nwachukwu 2013), neural networks (Haykin 2011; West 2000; Pacelli and Azzollini 2011), expert systems or support vector machines (Wang and Ma 2012), machine learning (Davis et al. 1992) etc. All these methods have their limitations. Amongst those limitations we can enlist: misclassification and indirect discrimination, variations from market to market, problems with accommodating changes, assuming a specific normality or homoscedasticity that are often violated in real world or model selection on trial and error process (Samuel and Asogbon 2016). There are also many credit risk mainstream probability of default (PD) models divided into a group of standard and reduced models. In this approach it is the PD level that is eventually translated into a credit rating. In the group of standard models, we can find models of Merton (1974), Geske (1979), Hull et al. (2004), Black-Cox (1976), or more recent models of Chen-Hu-Pan (2006) and others. The reduced mod˝ els were researched by, for instance, Jarrow-Turnbull (1995), Madan-Unal (1998), Lando (1998), Duffie-Singleton (1999) and others. More on those models can be found in Saunders and Allen (2010). All the above-mentioned models represent the foundation of most of further developed models. Under BIS 1998 banks had to develop their Value-at-Risk methodology. In the scope of this approach that originated from standard and reduced models (so-called “new approach”) we can find well-known models of rating agencies, consulting firms and other financial institutions, such as MKMV (originally proposed by Merton), CreditMetrics (introduced by RiskMetrics Group) and Credit Suisse Financial Product’s CreditRisk + or McKinsey’s CreditPortfolioView, etc. These models, and their modifications, are constantly used by banks. However, it must be stressed that they are mostly successful in the USA due to calibration data used for their creation (more in Gordy 2000; Lopez and Saidenberg 2000). However, there is frequently a problem with applying those methods (no available data for non-quoted company on their market asset value) or with their ultimate efficiency due to their assumptions or strictly quantitative character of data which can lack the overall big picture. Moreover, it is important to remember that in the field of credit risk assessment research, we can also include, not just the bank’s debtors but also other creditors (e.g. trading companies) that play a role of creditor in trade transactions with a delayed payment (trade credit). Using that tool, one party is always threatened by credit risk. Therefore, for them intuitive systems of debtors’ assessment such as OFSAW are really essential. Another approach is the statistical enterprise trade credit risk assessment model for evaluation of trade risk of small and micro enterprises (Kanapickiene and Spicas 2019).

192

A. Wójcicka-Wójtowicz et al.

Within the area of credit risk assessment and debtors’ ratings there are also many modifications and extensions.1 They appear as a result of the shortcomings of existing models. The significance of experts’ knowledge and experience, as well as other qualitative factors in credit risk assessment and debtors’ classification, are recognised as increasingly influential and helpful in decision-making process. In (Samuel and Asogbon 2016) neural network and fuzzy logic systems for credit risk evaluation was developed and their performances were evaluated based on prediction accuracy metric. The conclusion was that despite comparable results, the fuzzy inference system could be easily understood by any user, however, the decisions made by the neural network system is not easily understood by the user, and in this case the user has no choice than to accept the output given by the neural network as the most appropriate output without any explicit reasoning. Also, in (Dadios and Solis 2012) a hybrid fuzzy logic and neural network algorithm (HFNN) to solve credit risk management problem is tested. It is shown that HFNN model can solve credit risk management problem and is capable of self-learning similar to the traditional neural network. It can also generate the rules behind the discrimination of each account subjected to it and in this manner, it behaves much like a traditional fuzzy logic system.

3 Oriented Fuzzy Numbers—Basic Facts Objects of any considerations may be given as elements of a predefined space X. The basic tool for an imprecise classification of these elements is the notion of fuzzy sets introduced by Zadeh (1965). Any fuzzy set A is unambiguously determined by means of its membership function μ A ∈ [0, 1]X , as follows A = {(x, μ A (x)); x ∈ X}.

(1)

From the point-view of multi-valued logic (Łukasiewicz 1922/23), the value μ A (x) is interpreted as the truth value of the sentence “x ∈ A”. By the symbol F(X) we denote the family of all fuzzy sets in the space X. Dubois and Prade (1979) have introduced fuzzy numbers (FNs) as such a fuzzy subset in the real line which may be interpreted as imprecise approximation of a real number. The ordered FNs were intuitively introduced by Kosi´nski et al. (2002) as an extension of the FNs concept. Ordered FNs usefulness follows from the fact that it is interpreted as FNs with additional information about the location of the approximated number. Currently, ordered FNs defined by Kosi´nski are often called Kosi´nski’s numbers (Prokopowicz and Pedrycz 2015; Prokopowicz 2016; Piasecki 2019). The competent discussion on the current state of knowledge on Kosi´nski’s numbers is presented in (Prokopowicz et al. 2017). A significant drawback of Kosi´nski’s theory is 1 A review on financial risk assessment (including credit and bankruptcy risks) can be found in Chen

et al. (2016).

Application of the SAW Method in Credit Risk Assessment

193

that there exist such Kosi´nski’s numbers which, in fact, are not FNs (Kosi´nski 2006). For this reason, the Kosi´nski’s theory was revised by Piasecki (2018). If an ordered FN is determined with use of the revised definition, then it is called Oriented FN (OFN). The OFN definition fully corresponds to the intuitive Kosi´nski’s definition of ordered FNs. In this paper, we restrict our considerations to the case of Trapezoidal OFNs (TrOFN) defined as fuzzy subsets in the space R of all real numbers in the following way. Definition 1 (Piasecki 2018) For any monotonic sequence (a, b, c, d) ⊂ R, TrOFN ↔ −→ ← → T r (a, b, c, d) = T is the pair of the orientation a, d = (a, d) and a fuzzy subset T ∈ F(R) determined explicitly by its membership functions μT ∈ [0, 1]R as follows ⎧ ⎪ 0, x ∈ / [min{a, d}, max{a, d}], ⎪ ⎪ ⎨ x−a , x ∈ [min{a, b}, max{a, b}[, μT (x) = μT r (x|a, b, c, d) = b−a (2) ⎪ 1, x ∈ [min{b, c}, max{b, c}], ⎪ ⎪ ⎩ x−d , x ∈ ]min{c, d}, max{c, d}]. c−d The symbol KT r denotes the space of all TrOFNs. Any TrOFN describes an imprecise number with additional information about the location of the approximated number. This information is given as orientation of OFN. If a < d then TrOFN ← → −→ T r (a, b, c, d) has the positive orientation a, d. For any z ∈ [b, c], the positively ← → oriented TrOFN T r (a, b, c, d) is a formal model of linguistic variable “about or ← → slightly above z”. If a > d, then OFN T r (a, b, c, d) has the negative orientation −→ ← → a, d. For any z ∈ [c, b], the negatively oriented TrOFN T r (a, b, c, d) is a formal model of linguistic variable “about or slightly below z”. Understanding the phrases “about or slightly above z” and “about or slightly below z“depends on the applied ← → pragmatics of the natural language. If a = d, then TrOFN T r (a, a, a, a) = a describes un-oriented real number a ∈ R. Kosi´nski has introduced the arithmetic operators of dot product  for TrOFNs in a following way: ← → ← → β  T r (a, b, c, d) = T r (β · a, β · b, β · c, β · d).

(3)

In Piasecki (2018), the sum  for TrOFNs is determined as follows ← → ← → T r (a, b, c, d) T r ( p − a, q − b, r − c, s − d) ← → T r (min{ p, q}, q, r, max{r, s}) (q < r ) ∨ (q = r ∧ p ≤ s) = ← . → T r (max{ p, q}, q, r, min{r, s}) (q > r ) ∨ (q = r ∧ p > s)

(4)

194

A. Wójcicka-Wójtowicz et al.

  ↔ ↔ Let us consider the pair K, L ∈ K2T r represented by the pair (μ K , μ L ) ∈

2 [0, 1]R of their membership functions. On the space KT r , we introduce the relation ↔

↔

K. G E.L, which reads: ↔

↔

“T r O F N K is gr eater than or equal to T r O F N L.”

(5)

This relation is a fuzzy preorder  G E ∈ F K2T r defined by its member2 ship function νG E ∈ [0, 1]KT r (Piasecki 2019; Piasecki  et al.  2019c). From the ↔

↔

point of view of the multivalued logic, the value νG E K, L

is considered as a

truth-value of the sentence (5). In (Piasecki 2019), it is shown that for any pair (T r (a, b, c, d), T r (e, f, g, h)) ∈ K2T r we have ⎧ ⎪ 0 < α − γ, ⎨ 0, ← → ← → α−γ , νG E T r (a, b, c, d), T r (e, f, g, h) = α+δ−β−γ α − γ ≤ 0 < β − δ, . ⎪ ⎩ 1, β −δ ≤0

(6)

where α = max{a, d},

(7)

β = max{b, c},

(8)

γ = min{e, h},

(9)

δ = min{ f, g}.

(10)

Therefore, for any pair (T r(a, b, c, d), [[e]]) ∈ KT r × R ⊂ K2T r we get ← → νG E T r (a, b, c, d), [[e]] ⎧ ⎪ max{a, d} < e, ⎨ 0, {a,d}−e max = max{a,d}−max{b,c} , max{a, d} ≥ e > max{b, c}, ⎪ ⎩ 1, 0 ≤ c − f.

(11)

Application of the SAW Method in Credit Risk Assessment

195

4 Linguistic Approach—Order Scales Credit granting is a decision which is vitiated by credit risk understood as the possibility that credit will be defaulted. Credit lenders tend to minimize this risk. For this reason, they evaluate borrowers in terms of many criteria. Any borrower attributes can be evaluated by means of numerical values. By its very nature of things, each such assessment is an imprecise information. Therefore, in dealing with such a situation with imprecise information, the use of linguistic assessments, instead of numerical values, may be more useful. Following (Herrera and Herrera-Viedma 2000), we can say that an application of imprecise linguistic assessments for decision analysis is very beneficial because it introduces a more flexible framework which allows us to represent the information in a more direct and adequate way when we are unable to express it precisely. This way, the burden of quantifying a qualitative concept is eliminated. In the first step of any linguistic approach, we should determine the imprecision granularity, i.e., the cardinality of the linguistic term set used for showing the information. The imprecision granularity indicates the capacity of distinction that may be expressed. The knowledge value is increasing with the increase in granularity. The typical values of cardinality used in the linguistic models are odd ones, usually between 3 and 13. It is worth to note that the idea of granular computing goes from Zadeh (1997) who wrote “fuzzy information granulation underlies the remarkable human ability to make rational decisions in an environment of imprecision, partial knowledge, partial certainty and partial truth.” Also, Yao (2004) pointed out that “the consideration of granularity is motivated by the practical needs for simplification, clarity, low cost, approximation …”. For review variety of application linguistic models in decision-making see for example (Herrera and Herrera-Viedma 2000). In general (Herrera and Herrera-Viedma 2000), any linguistic value is characterized by means of a label with semantic value. The label is an expression belonging to a given linguistic term set. Finally, a mechanism of generating the linguistic descriptors is provided. In our model of credit risk assessment, all linguistic assessments are linked with Tentative Order Scale (TOS) given as a sequence T O S = {Bad, Average, Good} = {C, B, A} = {V1 , V2 , V3 }.

(12)

Any element of TOS is called a reference point. Now, we focus on the problem of TOS enlargement of intermediate values. For this purpose, we use the following orientation phrases: • • • • •

“much below” described by the symbol “−−”, “below” described by the symbol “−”, “around” described by the symbol “∼”, “above” described by the symbol “+”, “much above” described by the symbol “++”.

196 Table 1 Order scales

A. Wójcicka-Wójtowicz et al. TOS

EOS

Semantic meaning

C−−

Much below bad

C−

Below bad

C~

Around bad

C

Bad C+

Above bad

C++

Much above bad

B−−

Much below average

B−−

Below average

B~

Around average

B

Average B+

Above average

B++

Much above average

A−−

Much below good

A−

Below good

A~

Around good

A

Good A+

Above good

A++

Much above good

NOS

← → T r 1, 1, 43 , 14 ← → T r 45 , 1, 34 , 24

← → T r 24 , 1, 1, 46 ← → T r (1, 1, 1, 1) ← → T r 43 , 1, 54 , 64 ← → T r 1, 1, 45 , 47 ← → T r 2, 2, 47 , 54

← → T r 49 , 2, 74 , 64

← → T r 46 , 2, 2, 10 4 ← → T r (2, 2, 2, 2)

← → T r 47 , 2, 94 , 10 4

← → T r 2, 2, 49 , 11 4

← → 9 T r 3, 3, 11 4 , 4

← → 13 11 10 T r 4 , 3, 4 , 4

← → 14 T r 10 4 , 3, 3, 4 ← → T r (3, 3, 3, 3)

← → 13 14 T r 11 4 , 3, 4 , 4 ← → 15 T r 3, 3, 13 4 , 4

Any order label is determined as a composition of reference point and orientation phrases. The set of all order labels is called Extended Order Scale (EOS). In Table 1, we can see TOS and EOS proposed for credit risk assessment. In information sciences, natural language word is considered as a linguistic variable defined as a fuzzy subset in the predefined space X. Then, these linguistic variables may be transformed with the use of fuzzy set theory (Zadeh 1975a, b, c). From decision making point view, the linguistic variable transformation methodologies are reviewed in (Herrera et al. 2009; Herrera and Herrera-Viedma 2000; Martınez et al. 2010). Let us assume that each reference point V j is represented by the number j ∈ N. On the other side, the semantic meaning of any orientation phrase is imprecise. For this reason, any order label may be considered as imprecise approximation of its reference point. Thus, each order label from applied EOS should be represented in the real line R by FN (Chen and Hwang 1992). For convenience of future calculations, we can always restrict this representation to representation by trapezoidal FN. Moreover, we observe that orientation phrases determine the orientation of FN representing approximated reference point. Therefore, any order label can be represented by TrOFN. This approach is more faithful than representation of order labels

Application of the SAW Method in Credit Risk Assessment

197

by trapezoidal FN. On the other hand, an omission of information about order labels’ orientation causes unbelievable assessment of borrowers (Piasecki et al. 2019b). For these reasons, all order labels will be represented by TrOFNs. The family of all TrOFNs representing considered EOS will be called Numerical Order Scale (NOS). In our credit risk assessment task, we use NOS introduced in (Wójcicka-Wójtowicz and Piasecki 2019). All applied order scales are presented in Table 1.

5 Simple Additive Weighting Method—Overview Let us take into account the problem of borrower’s evaluation. The evaluation template distinguishes all borrower’s attributes which are evaluated. Any borrower may be evaluated by means of a scoring function which takes into account experts’ preferences with respect to all evaluation criteria and their relative importance. The process of determining evaluation template is an important part of credit risk analysis, as well as constructing a scoring function, which is realized in the pre-evaluation phase. Because borrowers are often characterized by several contradictory criteria, the multi-criteria techniques are useful for building borrower-scoring function. In (Hwang and Yoon 1981; Mardani et al. 2015) it is shown that the most popular techniques used for multi-criteria evaluation is the Simple Additive Weighting (SAW) method (Churchman and Ackoff 1954). The SAW method is a scoring method based on the concept of a weighted average of criterion ratings. The SAW method is also called Simple Multi Attribute Rating Technique (Edwards 1977). In a general case of criterion ratings expressed by FNs, the fuzzy SAW method was introduced by Chou and Chang (2008). In the considered task of a credit risk evaluation, the individual criterion ratings are expressed by TrOFNs. For this reason, we need the SAW method linked with TrOFNs. Such SAW method should be equipped with scoring function determined on the space KnT r = K × K × · · · × K. Roszkowska and Kacprzak (2016) tentatively introduce the Oriented Fuzzy SAW (OF-SAW) method with criterion ratings expressed trapezoidal Kosi´nski’s numbers. In (Piasecki and Roszkowska 2018; Piasecki et al. 2019a) OF-SAW method is modified in a way that it is compatible with the revised theory of ordered FNs (Piasecki 2018). In this case, criterion ratings are given as TrOFNs. Below, we adapt the OF-SAW method (Piasecki et al. 2019a) to the needs of assessing a single borrower. We intend to evaluate a borrower characterized by attributes record A ∈ A where A is an anticipated set of potential borrowers. For this case OF-SAW method can be described by the following procedure: Step 1: Define a multi-criteria evaluation problem by criteria set D = {C1 , C2 , . . . , Cn }. Step 2: Determine the weight vector n w = (w1 , w2 , . . . , wn ) ∈ R+ . 0

(12)

198

A. Wójcicka-Wójtowicz et al.

where w1 + w2 + . . . + wn = 1.

(13)

and w j is the weight of the criterion C j denoting the importance of this criterion in considered evaluation problem Step 3: For each evaluation C j ( j = 1, 2, . . . , n), determine its scope Y j . Step 4: Determine the evaluation template Y = Y1 × Y2 × . . . . × Yn ⊃ A.

(14)

Step 5: Define the NOS O ⊂ Ktr . Step 6: Define the evaluation function X : Y × D → O ⊂ Ktr in such way that the value X A, C j ∈ O is equal to evaluation of attributes record A from the point-view of the criterion C j ( j = 1, 2, . . . , n). ←−→ Step 7: Determine the scoring function S AW : Y → KT r given for any A ∈ Y by the identity ←−→ S AW (A) = (w1  X (A, C1 ))(w2  X (A, C2 ))  · · · (wn  X (A, Cn )).

(15)

For a given evaluation template Y, any classical scoring method of credit risk assessment can be presented as a pair ( f, L) (Mays 2001; Anderson 2007) where: • f : Y → R is a given scoring function, • L ∈ R is a predetermined level of acceptance of a credit/loan application. Let us consider a credit application of a borrower characterized by attributes record A ∈ A. If the following condition is fulfilled f (A) ≥ L ,

(16)

then the application is acceptable (Mays 2001; Anderson 2007). In this section to assess the creditworthiness we suggest the use of a scoring ←−→ function S AW : Y → KT r . Therefore, we suggest extending the inequality (16) into a following form ←−→ S AW (A). G E.[[L]],

(17)

The fulfilment of the above inequality is tantamount to a sentence: Cr edit application based on attributes r ecor d A is acceptable.

(18)

Application of the SAW Method in Credit Risk Assessment

199

←−→ Then the value νG E S AW (A), [[L]] is truth-value of the sentence (18). For ←−→ this reason, we interpret the value νG E S AW (A), [[L]] as a degree in which the considered credit application is acceptable. Therefore, the value ←−→ accept(A, L) = νG E S AW (A), [[L]] ,

(19)

will be called the acceptance degree (acceptance level). This value can be a significant premise for the credit committee to take a final decision to grant the funding.

6 Numerical Example—Case Study This paper has applied 16 criteria that are qualitative and positive for selecting a good potential debtor amongst the analysed ones and ranking them. The introduced method is used in a case study. The data was collected from two experts in the banking field. We will call these experts George and John. Both experts are the active members of a credit assessment committee with a long business experience in that field.2 The research was conducted in the following steps: Step 1. Preparation of an appropriate assessment form (template). The most important part of that research stage was to establish the qualitative criteria, basing on experts’ business experience in credit risk assessment. Eventually, the experts settled on 16 criteria which, apart from the quantitative analysis of financial ratios, influence the final decision. The chosen criteria are presented in Tables 2 and 3. Step 2. Incorporate weights of the criteria. 1 In the research the weighs are equal and amount to 16 due to the number of considered criteria. In this way, we got evaluation template. which was applied.

Step 3. Set the acceptance level. Before the beginning of the evaluation process, the experts were informed that the acceptance level is represented by “a middle point between reference points ‘Average’ and ‘Good’”. The experts evaluated credit application provided by ‘Enterprise A’ and ‘Enterprise B’. Step 4. Experts fill in the form. They expressed their individual, professional opinion on the specific criterion in relation to an analysed enterprise by attributing the criterion to a single rank of EOS. 2 The

personal data of experts and any data concerning the Bank as well as any business and decision-making actions involved in the process, are subject to confidentiality.

200

A. Wójcicka-Wójtowicz et al.

Table 2 Evaluation of the ‘Enterprise A” attributes No.

Criterion

Evaluations George

← → T r 34 , 1, 45 , 46

← → T r 46 , 2, 2, 10 4 ← → T r (2, 2, 2, 2) ← → 15 T r 3, 3, 13 4 , 4

10 4



1

Prospects of business

C+

2

Quality of suppliers

B~

3

Quality of customers

B

4

Clean criminal record of the board members

A++

5

Clean criminal record of a chairperson

A++

← → T r 3, 3,

13 15 4 , 4

6

Experience of the board members

A++

← → T r 3, 3,

13 15 4 , 4

7

Experience of a chairperson

A++

← → T r 3, 3,

13 15 4 , 4

8

Operations range—Poland

A++

← → T r 3, 3,

13 15 4 , 4

9

Operations range—abroad

C~

← → T r 24 , 1, 1, 64

10

Risk of the market

C++

11

Risk of the trade

B+

12

Risk of the suppliers

A−

13

Risk of the customers

A−

14

Diversification of products

B

← → T r 1, 1, 45 , 47

← → T r 47 , 2, 49 , 10 4

← → 11 10 T r 13 4 , 3, 4 , 4

← → 11 10 T r 13 4 , 3, 4 , 4 ← → T r (2, 2, 2, 2)

15

Diversification of sales markets

B

← → T r (2, 2, 2, 2)

C++

← → T r 1, 1, 45 , 47

16

Diversification of supply market

B

← → T r (2, 2, 2, 2)

B++

← → T r 2, 2, 49 ,



B+

John ← → T r 47 , 2, 49 ,

B++

← → 9 T r 3, 3, 11 4 , 4

← → 9 T r 3, 3, 11 4 , 4

← → T r 2, 2, 49 , 11 4

B++

← → T r 2, 2, 49 ,

11 4

B++

← → T r 2, 2, 49 ,

11 4

B++

← → T r 2, 2, 49 ,

11 4

A−

← → T r 13 4 , 3,

B−

← → T r 49 , 2, 47 , 46

B

← → T r (2, 2, 2, 2)

A−− A−−

B B B C++





11 10 4 , 4



← → T r (2, 2, 2, 2) ← → T r (2, 2, 2, 2) ← → T r (2, 2, 2, 2) ← → T r 1, 1, 45 , 47

11 4



Step 5. Transform the experts’ evaluations into NOS. The evaluations given by each expert were transformed into NOS. Evaluations of ‘Enterprise A’ attributes and ‘Enterprise B’ attributes are presented respectively in Tables 2 and 3. All obtained results show that the experts differ in their perception of the importance of the qualitative features when assessing the same entity. The results in bold shown in Tables 2 and 3 present the dominant variants. It is clear that each individual expert perceives the criteria differently. Therefore, as a final assessment, we ought to calculate mean SAW value representing common opinion of both experts. Basing on the discussed form, filled by the experts, the calculations were performed for each enterprise. Then, for each expert we determine the value of a scoring function SAW (15). For each evaluated enterprise, taking into account the values

Application of the SAW Method in Credit Risk Assessment

201

Table 3 Evaluation of the ‘Enterprise B” attributes No.

Criterion

Evaluations George

1

Prospects of business

B

2

Quality of suppliers

B++

3

Quality of customers

B++

4

Clean criminal record of the board members

A+

← → T r (2, 2, 2, 2)

← → T r 2, 2, 94 , 11 4

← → T r 2, 2, 94 , 11 4

← → 13 14 T r 11 4 , 3, 4 , 4

5

Clean criminal record of a chairperson

A+

← → T r 11 4 , 3,

13 14 4 , 4

6

Experience of the board members

A+

← → T r 11 4 , 3,

13 14 4 , 4

7

Experience of a chairperson

A+

← → T r 11 4 , 3,

13 14 4 , 4

8

Operations range—Poland

A+

← → T r 11 4 , 3,

13 14 4 , 4

9

Operations range—abroad

C+

← → T r 34 , 1, 45 , 46

10

Risk of the market

C

11

Risk of the trade

C~

12

Risk of the suppliers

C++

13

Risk of the customers

C+

14

Diversification of products

B++

← → T r (1, 1, 1, 1)

← → T r 24 , 1, 1, 46 ← → T r 1, 1, 45 , 47 ← → T r 34 , 1, 45 , 46

← → T r 2, 2, 94 , 11 4

15

Diversification of sales markets

A–

← → T r 3, 3,

11 9 4 , 4

16

Diversification of supply market

A–

← → T r 3, 3,

11 9 4 , 4

John A~ B+ B+ A++







A++

← → T r 3, 3,

13 15 4 , 4

A++

← → T r 3, 3,

13 15 4 , 4

A++

← → T r 3, 3,

13 15 4 , 4

B−

← → T r 49 , 2, 47 , 46

B−

← → T r 49 , 2, 47 , 46

A

← → T r (3, 3, 3, 3) ← → T r (3, 3, 3, 3) ← → T r (3, 3, 3, 3)

A

← → T r (3, 3, 3, 3)

C++

← → T r 1, 1, 45 , 74

C++

← → T r 1, 1, 45 , 74

C++

← → T r 1, 1, 45 , 74

A A





← → 14 T r 10 4 , 3, 3, 4

← → 7 T r 4 , 2, 49 , 10 4

← → T r 47 , 2, 49 , 10 4 ← → 15 T r 3, 3, 13 4 , 4

of the scoring functions designated by all experts, we calculate their average value denoted by the term Mean SAW. The value Mean SAW represents the opinion given by experts’ team. All these values are presented in Table 4. Table 4 Values of scoring function SAW Enterprise A B

Evaluations SAW by George ← → 72 75 83 T r 70 32 , 32 , 32 , 32 ← → 136 145 158 T r 127 64 , 64 , 64 , 64

SAW by John ← → 132 136 T r 132 64 , 64 , 64 , ← → T r 150 64 ,

146 64



152 159 175 64 , 64 , 64



Mean SAW ← → 138 143 156 T r 137 128 , 128 , 128 , 64

← → 288 304 333 T r 267 128 , 128 , 128 , 128

202

A. Wójcicka-Wójtowicz et al.

Table 5 Acceptance degree

Acceptance degree Enterprise

By George

By John

From Mean SAW

A

0.375

0.000

0.000

B

0.000

0.9375

0.6111

In the next step, for each assessed value of SAW scoring function the acceptance degree values are calculated (19). Due to the fact that the acceptance level is represented by “a middle point between reference points ‘Average’ and ‘Good’”, we assume that the acceptance level is given as L=

5 1 · (2 + 3) = , 2 2

(20)

Here, we utilise the relationship (11). All values acquired in this manner are presented in Table 5. The values presented in Table 5 allow us to formulate the following findings • • • • • •

credit application of Enterprise A is weakly accepted by George; credit application of Enterprise A is not accepted by John; credit application of Enterprise A is not accepted by experts’ team; credit application of Enterprise B is not accepted by George; credit application of Enterprise B is strongly accepted by John; credit application of Enterprise B is accepted at medium level by experts’ team.

The final decision of granting the credit (loan) is up to the credit committee. The committee can take into consideration the opinions presented above.

7 Conclusions Credit risk assessment usually is a complex process which consists of many successive steps and numerous criteria. The distinction between a good and a bad customer and following rejection of those in the latter group is vital for a bank as it directly and significantly affects the quality of bank’s credit portfolio. Also, ordering the decision alternatives is an important part of the whole decision-making analysis which takes place before making a final decision. The nature of the problem enables intuitive decision-making, imprecision and inaccurate linguistic ranks based on experts’ personal experience. The calculations, conducted in a numerical example presented in the paper, show the utility of SAW method in case of a credit risk assessment and the order scale is described by oriented fuzzy numbers (OFN). The estimation of the acceptance level should be a subject of further research.

Application of the SAW Method in Credit Risk Assessment

203

References Altman E (1968) Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. J Financ 23:589–609 Anderson R (2007) The credit scoring toolkit. Oxford Press, Oxford Black F, Cox JC (1976) Valuing corporate securities: some effect of bond indenture provisions. J Financ 31(2):351–367 (Papers and proceedings of the thirty-fourth annual meeting of the American Finance Association Dallas, texas december 28–30, 1975 (May 1976)) Bonilla M, Olmeda I, Puertas-An R (2000) Application of genetic algorithms in credit scoring. WASL J Chen SJ, Hwang CL (1992) Fuzzy multiple attribute decision making: methods and applications. Springer Verlag, Berlin. https://doi.org/10.1007/978-3-642-46768-4 Chen RR, Hu SY, Pan GG (2006) Default prediction of various structural models. www.defaultrisk. com/pp_score_63.htm Chen N, Ribeiro B, Chen A (2016) Financial credit risk assessment: a recent review. Artif Intell Rev 45(1). https://doi.org/10.1007/s10462-015-9434-x Chou S, Chang Y (2008) A decision support system for supplier selection based on a strategy-aligned fuzzy SMART approach. Expert Syst Appl 34(4):2241–2253 Churchman CW, Ackoff RL (1954) An approximate measure of value. J Oper Res Soc Am 2(1) Dadios EP, Solis J (2012) Fuzzy-neuro model for intelligent credit risk management. Intell Inf Manag 4:251–260. https://doi.org/10.4236/iim.2012.425036 Davis RH, Edelman DB, Gammerman AJ (1992) Machine learning algorithms for credit-card applications. IMA J Math Appl Bus Ind 4:43–51 Dubois D, Prade H (1979) Fuzzy real algebra: some results. Fuzzy Sets Syst 2:327–348. https:// doi.org/10.1016/0165-0114(79)90005-8 Duffie D, Singleton KJ (1999) Modeling term structures of defaultable bonds. Rev. Financ. Stud. 12(4):687–720 Edwards W (1977) How to use multi-attribute utility measurement for social decision making. IEEE Trans Syst Man Cybern 7(5):326–340. https://doi.org/10.1109/TSMC.1977.4309720 Geske R (1979) The valuation of compound options. J. Financ. Econ. 7:63–81 Gordy MB (2000) A comparative anatomy of credit risk models. J Bank Finance 24:119–149 Haykin S (2011) Neural networks and learning machines, 3rd edn. PHI Learning Private Limited, New Dehli-110001 Herrera F, Herrera-Viedma E (2000) Linguistic decision analysis: steps for solving decision problems under linguistic information. Fuzzy Sets Syst 115:67–82. https://doi.org/10.1016/S01650114(99)00024-X Herrera F, Alonso S, Chiclana F, Herrera-Viedma E (2009) Computing with words in decision making: foundations, trends and prospects. Fuzzy Optim Decis Mak 8:337–364. https://doi.org/ 10.1007/s10700-009-9065 Hull JC, Nelken I, White A (2004) Merton’s model, credit risk, and volatility skews. J Credit Risk 1(1):3–28 Hwang CL, Yoon K (1981) Multiple attribute decision making: methods and applications. SpringerVerlag, Berlin. https://doi.org/10.1007/978-3-642-48318-9 Jarrow RA, Turnbull SM (1995) Pricing derivatives on financial securities subject to credit risk. J. Financ. 50(1):53–85 Kabari LG, Nwachukwu EO (2013) Credit risk evaluating system using decision tree—neuro based model. Int J Eng Res & Technol (IJERT) 2(6) Kanapickiene R, Spicas R (2019) Credit risk assessment model for small and micro-enterprises: the case of Lithuania. Risks 7(2):67 Karels G, Prakash A (1987) Multivariate normality and forecasting of business bankruptcy. J Bus Financ Account 14(4):573–593 Kosi´nski W (2006) On fuzzy number calculus. Int J Appl Math Comput Sci 16(1):51–57

204

A. Wójcicka-Wójtowicz et al.

´ ezak D (2002) Drawback of fuzzy arithmetics—new intuitions and Kosi´nski W, Prokopowicz P, Sl˛ propositions. In: Cholewa T, Moczulski W (eds) Methods of artificial intelligence Burczy´nski. Gliwice, Poland, pp 231–237 Lando D (1998) On Cox processes and credit risky securities. Rev. Deriv. Res. 2(2–3):99–120 Lopez JA, Saidenberg MR (2000) Evaluating credit risk models. J Bank Finance 24:151–165 Łukasiewicz J (1922/23) Interpretacja liczbowa teorii zda´n, Ruch Filozoficzny (trans: ‘A numerical interpretation of the theory of propositions’). In: Borkowski L (eds) Jan Łukasiewicz—selected works. North-Holland, Amsterdam, Polish Scientific Publishers, Warszawa, Poland, vol 7, pp 92–93 (1970) Madan DB, Unal H (1998) Pricing the risk of default. Rev. Deriv. Res. 2(2):121–160 Mardani A, Jusoh A, Zavadskas EK (2015) Fuzzy multiple criteria decision-making techniques and applications—two decades review from 1994 to 2014. Expert Syst Appl 42:4126–4148. https:// doi.org/10.1016/j.eswa.2015.01.003 Martınez L, Ruan D, Herrera F (2010) Computing with words in decision support systems: an overview on models and applications. Int J Comput Intell Syst 3(4):382–395. https://doi.org/10. 1080/18756891.2010.9727709https://doi.org/10.2991/ijcis.2010.3.4.1 Mays E (ed) (2001) Handbook of credit scoring. Glenlake Publishing, Chicago Merton R (1974) On the pricing of corporate debt: the risk structure of interest rates. J Financ 29:449–470 Pacelli V, Azzollini M (2011) An artificial neural network approach for credit risk management. J Intell Learn Syst Appl 103–112. https://doi.org/10.4236/jilsa.2011.32012 Piasecki K (2018) Revision of the Kosi´nski’s theory of ordered fuzzy numbers. Axioms 7(1). https:// doi.org/10.3390/axioms7010016 Piasecki K (2019) Relation “greater than or equal to” between ordered fuzzy numbers. Appl Syst Innov 2(3):26. https://doi.org/10.3390/asi2030026 Piasecki K, Roszkowska E (2018) On application of ordered fuzzy numbers in ranking linguistically evaluated negotiation offers. Adv Fuzzy Syst 1:1–12. https://doi.org/10.1155/2018/1569860 Piasecki K, Roszkowska E, Łyczkowska-Han´ckowiak A (2019a). Simple additive weighting method equipped with fuzzy ranking off evaluated alternatives. Symmetry 11(4). https://doi.org/10.3390/ sym11040482 Piasecki K, Roszkowska E, Łyczkowska-Han´ckowiak A (2019b). Impact of the orientation of the ordered fuzzy assessment on the simple additive weighted method. Symmetry 11(9). https://doi. org/10.3390/sym11091104 Piasecki K, Łyczkowska-Han´ckowiak A, Wójcicka-Wojtowicz A (2019c) The relation “greater than or equal to” for trapezoidal ordered fuzzy numbers. In: 37th international conference on mathematical methods in economics 2019, Conference Proceedings, Houda M, Remeš R (eds) ˇ ˇ University of South Bohemia in Ceské Budˇejovice, Faculty of Economics, Ceské Budˇejovice, Czech Republic, pp 61–66 Prokopowicz P (2016) The directed inference for the Kosinski’s fuzzy number model. In: Abraham A, Wegrzyn-Wolska K, Hassanien A, Snasel V, Alimi A (eds) Proceedings of the second international afro-european conference for industrial advancement AECIA 2015. Advances in intelligent systems and computing, vol 427. Springer, Cham https://doi.org/10.1007/978-3-31929504-6_46 Prokopowicz P, Pedrycz W (2015) The directed compatibility between ordered fuzzy numbers— a base tool for a direction sensitive fuzzy information processing. Artif Intell Soft Comput 119(9):249–259. https://doi.org/10.1007/978-3-319-19324-3_23 Prokopowicz P, Czerniak J, Mikołajewski D, Apiecionek Ł, Slezak D (2017) Theory and applications of ordered fuzzy number. Tribute to Professor Witold Kosi´nski; Stud Fuzziness Soft Comput 356 (Springer, Berlin, Germany) Roszkowska E, Kacprzak D (2016) The fuzzy SAW and fuzzy TOPSIS procedures based on ordered fuzzy numbers. Inf Sci 369:564–584. https://doi.org/10.1016/j.ins.2016.07.044 Samuel OW, Asogbon MG (2016) Comparative analysis of neural network and fuzzy logic techniques in credit risk evaluation. Int J Intell Inf Technol 12(1)

Application of the SAW Method in Credit Risk Assessment

205

Saunders A, Allen L (2010) Credit risk management. In and out of the financial crisis: new approaches to value at risk and other paradigms, 3 edn. Wiley Wallis LP (2001) Credit scoring. Business credit magazine Wang G, Ma J (2012) A hybrid ensemble approach for enterprise credit risk assessment based on support vector machine. Expert Syst Appl 39(5):5325–5331 West D (2000) Neural network credit scoring models. Comput Oper Res 27:1131–1152 Wójcicka-Wójtowicz A, Piasecki K (2019) A scale of credit risk evaluations assessed by ordered fuzzy numbers. SSRN E-J, Dx. https://doi.org/10.2139/ssrn.3459822 Yao YY (2004) Granular computing. Comput Sci 31:1–5 Zadeh LA (1965) Fuzzy sets. Inf Control 8:338–353. https://doi.org/10.1016/S00199958(65)90241-X Zadeh LA (1975a) The concept of a linguistic variable and its application to approximate reasoning. Part I. Information linguistic variable. Expert Syst Appl 36(2):3483–3488 Zadeh LA (1975b) The concept of a linguistic variable and its application to approximate reasoning. Part II. Inf Sci 8(4):301–357. https://doi.org/10.1016/0020-0255(75)90046-8 Zadeh LA (1975c) The concept of a linguistic variable and its application to approximate reasoning. Part III. Inf Sci 9(1):43–80. https://doi.org/10.1016/0020-0255(75)90017-1 Zadeh LA (1997) Toward a theory of fuzzy information granulation and its centrality in human reasoning and fuzzy logic. Fuzzy Sets Syst 90:111–127. https://doi.org/10.1016/S01650114(97)00077-8

Financial Institutions

Cost-Management Strategies Applied by Insurance Companies in Poland in the Years 2016–2018; Empirical Research Magdalena Chmielowiec-Lewczuk

Abstract The purpose of this paper is to present the findings concerning the use of cost strategies by insurance companies which run business activity in Poland. The work includes a theoretical outline of the issues presented here, in which a definition of the term “cost strategy” will be proposed, that has been formulated on the basis of analysis of relevant literature. The present paper will also contain a discussion of cost strategies that can be implemented in the insurance company and present and elaborate more extensively on the findings regarding the use of the cost strategies by insurance companies in Poland in the years 2016–2018. Summing up the findings we can formulate the following conclusions: insurance companies in Poland pursue some cost strategies or other—for the whole of the insurance companies in the last year of the study, i.e. 2018 they made up ¼ of all the insurers (14 entities altogether in all the groups); the group of insurance companies in which information on pursuing cost strategies appeared most frequently are life insurance companies, operating as joint-stock companies; and the most popular cost strategies pursued by insurance companies in Poland are strategies aiming at cost reduction (for example in 2018 12, out of 14 business entities altogether, provided information indicating the use of a cost strategy whose objective was to reduce costs).

1 Introduction Elaboration and implementation of adequate strategies constitute a basis for management of each and every business entity. Insurance companies, whose activity is based on selling financial products remain vulnerable to changing market conditions, which are, in turn, shaped both by customer expectations and new legal regulations. Because of their activity and conditions in which these entities operate they must

M. Chmielowiec-Lewczuk (B) Wrocław University of Economics and Business, Wrocław, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_17

209

210

M. Chmielowiec-Lewczuk

pay special attention to efficiency in achieving the targets set with respect to financial management. One of the basic conditions for achieving them is preparing and conducting appropriate cost-related operations. The purpose of this paper is to present the findings concerning the use of cost strategies by insurance companies which run business activity in Poland. The work includes a theoretical outline of the issues presented here, in which a definition of the term “cost strategy” will be proposed, that has been formulated on the basis of analysis of relevant literature. The present paper will also contain a discussion of cost strategies that can be implemented in the insurance company. Further on in this paper, I will present and elaborate more extensively on the findings regarding the use of the cost strategies by insurance companies in Poland in the years 2016–2018. The research was carried out on the basis of a detailed analysis of data contained in reports released by insurance companies on their solvency and financial condition. The obligation to draw up such reports arose due to introduction of the project Solvency II. Owing to these reports, it is possible to obtain partial information on strategies of insurance companies which is usually classed as commercially sensitive and access to it is seriously restricted. While analyzing the data from the report, special attention was paid to any information on cost-related operations. Each time it was checked whether a given insurance company followed a definite cost strategy or whether it actually undertook any cost-related activities. If information either on strategies or activities was not available, some other information regarding costs was taken into account (i.e., their structure, or most important items). It should be noted, that the reason behind these reports was to obtain data from insurance companies indicted by the supervisory authority and concerning their financial performance and factors shaping their financial condition. In other words the role of costs is of vital importance.

2 Cost Strategies Implementable by Insurance Companies The term ‘cost strategy’ is hard to come by in the literature. That is the reason why any attempt at defining it must begin with specifying of the very notion “strategy”. The definition of strategy which stems from the area connected with strategic management has been shaping over the last several decades and still is not quite free from ambiguity. In the literature one can find various approaches explaining the meaning of the term “strategy”. According to some authors strategy relies on defining long-term objectives which correspond to general direction of operations and such allocation of means that will guarantee their performance. Other authors in turn are of the opinion that the term strategy refers to targets set by the company and connected with its mission and aims, or define strategy as a choice model of strategic behavior which is determined by the following factors: budget behavior, strategic adaptation and strategic discontinuity rejecting previous experience

Cost-Management Strategies Applied by Insurance Companies …

211

(A. D. Chandler, P. Wright, Ch. D. Pringle, M. J. Kroll, H. I. Ansoff, K. R. Andrews, H. Mintzberg, M. Moszkowicz—based on Nowodzi´nski 2013; Janasz et al. 2010). As for the term “cost strategy” as such, it invariably appears when it comes to presenting M. E. Porter’s five forces model, where one of the strategies he elaborated is the cost leadership strategy. Its principal aim is to reduce costs to such an extent that someone following the strategy becomes the cheapest producer on the market (Porter 1999, p. 50). The cost strategy, as understood by Porter has little to do with the one that is the object of study of the present work. Cost strategy is not the same thing as the cost leadership strategy presented by Porter. The latter should only be viewed as one of many cost strategies that can be implemented by the insurance company. The cost strategy in the insurance company is a phrase that has a wider denotation, and cost reduction may be but one of its aims. It should be noted that in the considerations and findings presented here, the cost strategy is treated as an element of cost management in the insurance company, and depends on the choice and setting out the course of action related to cost areas which are conducive to target attainment in matters of financial management of the insurance company (Chmielowiec-Lewczuk 2018, p. 122). Issues related to cost strategy of the insurance company are relatively unknown and rarely encountered both in the world and domestic literature. It is possible, however, to find publications presenting articles that deal with development of strategies in the insurance company or papers dealing with questions of cost management. There are actually three texts deserving mention. In the year 2005 an article was published in which findings of the study carried out by a team of twelve researchers headed by R. Massey were presented. The subject of the project was development and application of theoretical models of strategy for non-life insurance companies. For that purpose the researchers used seven strategy models such as: competitive advantage model, M. E. Porter’s five forces model, value chain model PESTL, SWOT, product life cycle, blue ocean theory and innovation value theory. Even though none of the strategies was directly referred to as a cost strategy, many of them relate to the problem of costs. Another interesting publication dealing with the problem of costs in the insurance company appeared in the year 2014. Its author, S. McGregor described the implementation process for modification of the hitherto existing cost accounting system used by AXA Equitable Life Insurance Company in the United States of America. McGregor presents an analysis of issues related to costs, managers’ opinions, actuaries’ expectations and organizational arrangements. It also discusses the entire process of changes to the cost management system. The third item is an article which was published in 2010, and whose authors are P. Francis and S. Butler. They focus their attention on reduction of costs borne by the insurer accompanying indemnities and other payments due under the system adopted. The authors point out the role played by this cost item which, in their opinion, is the crucial factor in cost management in the insurance company. They propose to apply one of the four models of insurance claims management as a solution enabling the insurer to supervise relevant operations as well as use a matrix based on claims segmentation, which allows for cost reduction in the insurance company.

212

M. Chmielowiec-Lewczuk

The insurance company must coordinate its cost strategy with financial management targets it has set. It can opt for one of the following strategies (ChmielowiecLewczuk 2018, p. 126): (1) (2) (3) (4) (5) (6) (7)

strategy subordinated to cost reduction strategy subordinated to profitability strategy subordinated to price policy strategy subordinated to insurance risk strategy subordinated to financial liquidity strategy subordinated to value creation in the insurance company strategy subordinated to capital cost minimization.

Characteristics of the cost strategies implementable in the insurance company is presented in Table 1. A cost strategy can be implemented in the insurance company both as aid in achieving just one goal of financial management as well as in attaining several. One cannot, therefore, ascribe each of the enumerated strategy to only one goal. It should also be added that the insurance company does not have to follow just one cost strategy, it can pursue a few strategies at a time. In order to determine in practice, the best cost strategy for a particular insurance company, a two-step model may be used (see Chmielowiec-Lewczuk 2018, pp. 188– 190). The first stage involves establishing how important realization of particular objectives of finance management over a given period is (it can be assumed that the period in question is one year) attaching weight to them (this operation is carried out by managers responsible for finance management or implementation of a given cost strategy in the insurance company). The next step is to determine a cost strategy facilitating finance management in the insurance company relying on the assumption that its task is to guarantee such a realization of objectives of finance management as the insurance company had assumed earlier. Therefore, cost strategy should be subordinated to objectives of finance management.

3 Research Findings Concerning Cost Strategies Applied in Insurance Companies Poland in the 2016–2018 Insurance companies under examination were divided into four groups: 1. insurance companies of type 1 (life insurance companies) doing business as joint-stock companies (25 economic entities), 2. insurance companies of type 1 (life insurance companies) doing business in the form of mutual insurance companies (2 economic entities), 3. insurance companies of type 2 (non-life insurance companies) doing business as joint-stock companies (24 economic entities),

Cost-Management Strategies Applied by Insurance Companies …

213

Table 1 The cost strategies implementable in the insurance company Cost strategy

Crucial objective

Instruments applied

Main problem areas

Subordinated to cost reduction

Target cost level

Cost budgeting, activity based costing, responsibility accounting

Acquisition costs, claims management expenses, administrative costs (traditional approach) Operating cost (approach based on business operations); internal unit costs (responsibility accounting)

Subordinated to profitability

Target profitability level

Product life cycle costing, Kaizen costing, client cost accounting

Profitability of products, clients, distribution channels, technical and reinsurance operations, investment activity

Subordinated to price policy

Target rate of gross written premium

Target cost accounting, budgeting, costs of damages and benefits control

Acquisition costs, claims settlement costs, damages and benefits

Subordinated to insurance risk

Target level of covered risk (allowing for capital level and outward reinsurance)

Budgeting and costs of damages and benefits control coordinated with a given solvency level

Damages and benefits, solvency, own resources, capitals, methods of insurance risk evaluation

Subordinated to financial liquidity

Timing of payment of damages, benefits and other expenses with the date of deposit liquidation

Cost budgeting coordinated with budgeting of cash flows

Relation: costs–expenses (defining the relation between cost bearing and cash flow generating), cash flows

Subordinated to value creation of the insurance company

Maximization of value of the insurance company for owners

Value creation chain

Value creation areas connected with costs

Subordinated to cost capital maximization

Minimum weighted average cost of capital

Capital budgeting, debt ratio (financial leverage)

Capital structure, capital costs

Source Chmielowiec-Lewczuk (2018, pp. 127–128)

214

M. Chmielowiec-Lewczuk

4. insurance companies of type 2 (non-life insurance companies) operating in the form of mutual insurance companies (9 economic entities). Such grouping of insurance companies is dictated by the necessity to take account of differences in: legal form—financial management in joint-stock companies serves different purposes from those in the case of mutual insurance companies; the same holds true for their business activity (life insurances differ from property insurances particularly where saving aspect of the product is concerned). Research findings for the first group are presented in Fig. 1. It should be explained that: • the heading ‘no information on costs’ denoted the situation where the report included only data on the volume of the costs incurred with no additional description or interpretation of the cost level, • ‘providing significant information on costs’ denoted the situation where the report included explanations as to the volume of the costs incurred (e.g., when it was given that acquisition costs had the greatest share in the overall expenses or when a significant rise in administrative expenses took place, etc.), • ‘taking significant steps in relation to costs’ denoted the situation in which the report gave data on application of specific operations, methods of calculation, allocation or cost grouping (e.g., when it was given that activity-based costing or internal cost allocation methods were used in cost calculations),

Life Insurance Companies (Joint-Stock Companies)

41%

38% 25%

35% 27%

25%

22%

18% 13% 2016

35%

14% 2017

9% 2018

No information on costs Providing significant information on costs Taking significant steps in relation to costs Application of cost strategies

Fig. 1 Application of cost strategies in life insurance companies operating as joint-stock companies. Source Own elaboration based on the analysis of information found in reports released by insurance companies (see: the Web Index)

Cost-Management Strategies Applied by Insurance Companies …

215

• ‘application of cost strategies’ denoted the situation in which it was specifically underscored in the report that the company undertook adequate steps related to costs, which were subordinated to previously assigned goals or strategy, or when the volume of costs is a direct result of their application (i.e. it was announced that saving schemes had been adopted or that reduction in certain operations was foreseen with a view to cost curbing). In Fig. 1, we see that in the reports released by insurance companies belonging to group one an approach dominates in which important information on costs is given, which means that entities from this group add to the data presented explanations vis-à-vis the volume of incurred costs. There are also a considerable number of companies, which restrict themselves to putting in their reports just cost data without any additional description whatsoever. It can also be observed, however, that in this group of insurance companies there was an increase in the number of entities providing clear information on the basis of which we can learn that they take action related to costs resulting from implementation of concrete strategies and are subordinated to certain objectives. In the analyzed period, there was an increase in insurance companies which applied cost strategies, and at the same time, there occurred a decrease in entities undertaking only the most fundamental steps related to costs (e.g. use internal calculations but unsubordinated to any definite strategy). This is a positive development for it attests to their maturity in matters of management. On the other hand, it may also reflect this growing awareness of the need to make such information public in order to create a positive image of the insurance company. The second group of insurance companies embraces entities operating in Type 1 in the form of mutual insurance companies. This group includes only 2 economic entities in Poland, and in the year 2018 when the research was conducted, only in one of them steps were implemented which are ascribable to a concrete strategy, and which can be classified as a cost reduction strategy, since it said in the report that, “the decrease in costs occurred as a result of changes in the structure of employment and better process management”. The findings for Group 3 of insurance companies, which are represented by jointstock companies type 2, were presented in Fig. 2. Among joint-stock companies of type 2 the situation looks a bit different from that in the analogous group of type 1. Although similar results were obtained in respect of the most popular method of presenting information about costs, i.e., the most common situation was where insurance companies provide important information on costs, i.e., apart from presenting data on the volume of costs, their reports also include additional explanations or short characteristics of the costs. What is, however, different in this group, is a clearly decreasing number of entities which point out definite strategies implemented as part of cost management. This situation does not result from the fact that insurance companies resign from cost strategies to make way for important steps to curb costs, for these are also not very popular with this group of insurers. The results presented in Fig. 2 may therefore be the result of the following factors:

216

M. Chmielowiec-Lewczuk

Non - life Insurance Companies (Joint-Stock Companies)

55% 48%

48% 30%

5%

13% 2016

26%

22%

18%

18%

4% 2017

9% 2018

No information on costs Providing significant information on costs Taking significant steps in relation to costs Application of cost strategies Fig. 2 Application of cost strategies in non-life insurance companies operating as joint-stock companies. Source Own elaboration based on the analysis of information found in reports released by insurance companies (see: the Web Index)

• the insurers in this group have a stable cost situation and do not have to use any cost strategies, • they either do not need or do not want to publicly disclose information on the cost strategies they apply and restrict themselves only to explanations as to the volume and character of their costs. The last group is comprised of insurance companies, which carry out their business activity in the form of mutual insurance companies of type 2. This group is represented by a slightly bigger number of entities that the analogous group of type 1. Type 2 which embraces property insurance companies is represented in Poland by nine business entities. The results for this group of insurers are shown in Fig. 3. Insurance companies operating in the form of mutual insurance companies of type 2 are characterized by different objectives from those registered as joint-stock companies. That is why, based on the results obtained, we can observe that a relatively large group is formed by entities which provide only cost data, as well as by those which complement the data with explanatory information. In the study there were, however, no such companies in which we could observe, on the basis of the data provided in the reports, any significant operations related to costs. Also the findings indicating the use of cost strategies are slightly unexpected. Although the number of companies using them is on the decrease, the fact that mutual insurance companies, whose business activities are not primarily aimed at generating profit, implement

Cost-Management Strategies Applied by Insurance Companies …

217

Non - life Insurance Companies (Mutual Insurance Companies)

56%

50%

44% 38%

33%

33%

22% 0%

0%

2016

13%

2017

0%

11%

2018

No informaƟon on costs Providing significant informaƟon on costs Taking significant steps in relaƟon to costs ApplicaƟon of cost strategies

Fig. 3 Application of cost strategies in non-life insurance companies operating as mutual insurance companies. Source Own elaboration based on the analysis of information found in reports released by insurance companies (see: the Web Index)

them at all is a positive development. Of course, due to the fact that the last group is relatively small, cost strategies are used only by individual entities.

4 Summary and Conclusions The use of cost strategies was observed in each of the groups under examination. To the most commonly implemented belong strategies aimed at cost reduction or profitability. Some others appeared occasionally, e.g., strategy subordinated to insurance risk—1 entity in 2016. There appeared a situation where it was impossible to identify the type of strategy, for it said in the report, that “operations related to costs are established within the company” (type 2, joint-stock companies, 2017). Summing up the findings we can formulate the following conclusions: • insurance companies in Poland pursue some cost strategies or other—for the whole of the insurance companies in the last year of the study, i.e. 2018 they made up ¼ of all the insurers (14 entities altogether in all the groups), • the group of insurance companies in which information on pursuing cost strategies appeared most frequently are insurance companies of type 1, operating as jointstock companies,

218

M. Chmielowiec-Lewczuk

• the most popular cost strategies pursued by insurance companies in Poland are strategies aiming at cost reduction (for example in 2018 12, out of 14 business entities altogether, provided information indicating the use of a cost strategy whose objective was to reduce costs). It should be remembered, however, that information regarding the use of cost strategies by insurance companies in Poland is obtained from reports on solvency and financial condition, to preparing and publishing of which insurers have been obliged since 2016. On one hand, these reports constitute an unusually interesting source of information on insurance companies, but on the other, serve to create a certain image of the company as an institution of public trust. For this reason, during the examination it was possible to observe very different approaches towards presenting of data. Some of the companies under examination confined themselves to an absolute minimum of information, which they are obliged to provide, and some of the entities, quite the opposite, indicated very clearly their financial objectives, implemented strategies and measurable achievements in the area of finances. It should also be added, that if the report by an insurance company did not include information indicating the use of any cost strategy, it does not automatically mean that the entity did not pursue one or did not implement any measures in the area of costs. The study was carried out on the basis of information that is in the public domain, as no other data is available for presentation in this article. Insurance companies should always keep their costs under control, but these actions should be designed to be compatible with financial management objectives. Implementation of cost strategies should be a conscious action in the insurance company and adapted to the current market conditions.

References Badal V et al (2005) The application of strategic models to non-life insurance markets. GIRO working party paper. https://www.actuaries.org.uk/documents/application-strategic-models-nonlife-insurance-markets. Access 10.01.2018 Chmielowiec-Lewczuk M (2018) Modelowanie strategii kosztowej w zakładzie ubezpiecze´n a współczesne uwarunkowania rynku. Modeling of cost strategy in insurance companies versus contemporary market conditions. Publishing House of Wroclaw University of Economics, Wrocław Publisher Francis P, Butler S (2010) Cutting the cost of insurance claims. Taking control of the process. https:// www.strategyand.pwc.com/reports/cutting-cost-insurance-claims-taking. 11.01.2018 Janasz K et al (2010) Zarz˛adzanie strategiczne: koncepcje, metody, strategie [Strategic management: concepts, methods, strategies]. Publishing House Difin, Warsaw McGregor S (2014). Product costs: application in an insurance company. IMA Educ Case J 7(3) Nowodzi´nski P (2013) Zarz˛adzanie strategiczne współczesnym przedsi˛ebiorstwem. Otoczenia a strategia [Strategic management of a contemporary company. Environment versus strategy]. Cz˛estochowa University of Technology Porter ME (1999) Competition strategy. Methods of analysis of sectors and competitors. Polish Economic Publishers, Warsaw

Cost-Management Strategies Applied by Insurance Companies …

219

Reports on solvency and financial condition published by insurance companies in Poland for years 2016–2018 (as per the Web Index, access June 2019): https://www.aegon.pl/Strona-glowna/Ofirmie/informacje_o_firmie/Aegon-Towarzystwo-Ubezpieczen-na-zycie-Spolka-Akcyjna/; https://www.allianz.pl/dla-ciebie/szkody-i-obsluga/dokumenty/archiwum-sprawozdaniafinansowe/; https://www.aviva.pl/o-naszej-firmie/dane-o-spolkach/raporty-roczne/; https://axa. pl/o-nas/axa-w-polsce/spolki/#panel-5312-1; https://santander.aviva.pl/informacje-o-santanderaviva-ubezpieczenia/sprawozdanie-o-wyplacalnosci-i-kondycji-finansowej.html; https://www. cardif.pl/pl/pid3533/sprawozdania-wyplacalnosci-kondycji-finansowej.html; https://www. compensa.pl/compensa/; https://www.concordiaubezpieczenia.pl/biuro-prasowe/sprawozdaniasfcr,335.html; https://www.ergohestia.pl/o-ergo-hestia/raporty/; https://tueuropa.pl/raportyroczne.htm; https://www.generali.pl/o-generali/raporty-roczne.html; https://interpolska.pl/onas/; https://www.metlife.pl/o_metlife/ogloszenia/; https://www.nn.pl/dla-ciebie/notowania-iwyniki-finansowe/raporty-finansowe.html; https://www.openlife.pl/o-nas/raporty-publiczne/; https://pkoubezpieczenia.pl/o-nas/pko-zycie-tu#raporty; https://www.ubezpieczeniapocztowe. pl/o-nas#dane-o-towarzystwach; https://www.unum.pl/dane-finansowe/; https://www.pzu. pl/_fileserver/item/1516392; https://saltus.pl/o-nas/zycie/; https://www.signal-iduna.pl/onas/sprawozdania-finansowe; https://www.uniqa.pl/o-nas/o-firmie; https://viennalife.pl/files/ wyniki-finansowe/2018-Sprawozdanie-na-temat-wyplacalnosci-i-kondycji-finansowej.pdf; https://www.warta.pl/informacje-finansowe; https://www.macif.com.pl/onas/; https://rejentlife. com.pl/ujawnienia-publiczne/; https://pkoubezpieczenia.pl/o-nas/pko-tu#raporty; https://das. pl/das/towarzystwo-ubezpieczen/wyniki-finansowe-i-publikacje/; https://www.eulerhermes. com/pl_PL/obsluga-klienta/pliki-do-pobrania.html; https://www.interrisk.pl/interrisk/wynikifinansowe/; https://www.kuke.com.pl/o-kuke/sprawozdanie-wyplacalnosc-ii/; https://www. link4.pl/o-link4/dane-firmy-link4; https://www.tuirpartner.pl/node/54; https://www.tuzdrowie. pl/Onas_raporty; https://www.ubezpieczeniapocztowe.pl/o-nas#dane-o-towarzystwach; https:// saltus.pl/o-nas/tuw/; https://tuw-cuprum.pl/o-firmie; https://tuwmedicum.pl/o-firmie/; https:// polskigaztuw.pl/sprawozdania-o-wyplacalnosci-i-kondycji-finansowej-sfcr/; https://www. tuwpzuw.pl/o-nas.html; https://www.tuw.pl/PL/ujawnienie_publiczne.html; https://www.tuz.pl/ ujawnienia-publiczne

Dividends of Life Insurance Companies and the Solvency Capital Requirements Joanna Głód, Lyubov Klapkiv, Anna Białek-Jaworska, and Krzysztof Opolski

Abstract The aim of the article is to discuss the impact of capital requirements set out in the Solvency II Directive on the amount of dividends paid by life insurance companies in Poland. The study was conducted with the use of an estimator of random effects of panel data of the entire market consists of 24 insurance companies. The data was taken from Solvency and Financial Condition Reports for the total period of Solvency II Directive obligatory use (2016–2018). Life insurance companies are required to apply a number of legal regulations that determine the amount of dividend paid. However, simultaneously they are exposed to numerous risks which are not covered by capital requirements. Our results indicate that the amount of dividend is strongly correlated with the capital requirements. Capital requirements determine the level of profits earned in life insurance companies. There is a negative correlation between the capital retention (outflow) and the financial position of the life insurance company.

1 Introduction The study aims to analyse the relationship between the capital requirement and the dividends paid in life insurance companies operating on the Polish market as an example of a strongly regulated sector. As was admitted by Dickens (2009), dividends policy on regulated and nonregulated industries is different. Regulation influences the direction of dividend policy in such regulated markets as banking, insurance, real estate investment, telecommunication, etc. (Rozeff 1982). Insurance companies are subject to statutory reporting requirements that must hold their financial liquidity and solvency. It has made more sense after the global financial crises when the risks invoked high losses. To prevent insolvency in the insurance market J. Głód · A. Białek-Jaworska · K. Opolski Faculty of Economic Sciences, University of Warsaw, Warsaw, Poland L. Klapkiv (B) Maria Curie-Skłodowska University, Lublin, Poland e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_18

221

222

J. Głód et al.

the Solvency II was implemented. The new requirements create a higher demand for capital, which must represent the level of risk (risk-based capital). As was shown in the banking sector, solvency capital requirements can influence the dividend policy. Higher requirements force banks to restrict dividend payments; more stringent capital requirements are likely to force banks to retain profits to meet regulatory capital requirements (Ashraf et al. 2016). The insurance sector is similar to banking in the context of capital requirements, so it can be suggested that the solvency capital requirements also reduce the value of the dividend. There is no research that study directly this phenomenon on the insurance market. But the problem is important because insurers looking for their optimal dividend policy and authority influence on this policy. The other suggestion that enforces the research interest to a selected problem says that the dividend increases by European insurers seem to signal falling future corporate earnings (Basse 2019). The paper verifies the hypotheses that dividend pay-out is strongly correlated with the Solvency Capital Requirements, its sub-modules and the Minimum Capital Requirements. Our results contribute to the literature by providing new evidence on a character of linkage between the new one factor (capital requirements) and the dividend policy of life insurers. Based on the positive character of such a relation between capital requirements and dividends, it can be stated that in the case of the Polish market, dividends paid by life insurers were not limited by the new capital requirements.

2 Dividends Payments of Life Insurers The question of the dividends payment and dividends policy in insurance companies were analysed in different contexts: taxation policy influence (Chen 1990), the relation between stock prices and dividends per share (Lee and Forbes 1980), firm’s capacity ratio (Lee and Forbes 1982). Dividends were studied in the context of asymmetrical information between owners and financial markets. Due to Basse et al. (2011), in the condition of asymmetrically distributed information between investor and manager, dividend smoothing is a relevant phenomenon in the case of the Italian insurance market. It means that the market makes a negative interpretation of dividends cutting. Akhigbe et al. (1993) showed that life’s insurers share price response to dividend increases less that industries companies due to low level of capital maintained. It must be admitted that the majority of articles deal with the problem of optimal dividend policy and its ruin. De Finetti (1957), and Dong and Yin (2012) showed that certain barrier strategy maximizes expected discounted dividend payments and proposed the value of dividends paid out to shareholders as a criterion of optimal performance. It also has been determined by Gerber and Shiu (2005) that due to the aim of maximizing the payout to shareholders during the period of the company’s activity, the optimal is the barrier strategy, i.e. all inflowing surpluses are paid as dividends until the level of capital is below the barrier. But De Finetti (1957) argued

Dividends of Life Insurance Companies and the Solvency …

223

that such a strategy can invoke the destruction of the company. Based on the analysis of European and Germany insurers after the global crisis, Reddemann et al. (2010) showed that cutting dividends and thereby improving financial strength and complying with regulatory standards, companies do not necessarily have to fear major negative consequences due to investors assuming this measure to be an indubitable sign for future problems. Also, Basse (2019) confirmed that higher dividend pay-out could weaken the financial strength and (as a result) the ability of financial services firms to take more risks. Niehaus (2018) attempted to analyse the flow of capital based on life insurance companies. He verified the hypothesis of a negative relationship between the results of an insurance company and the dividend received (paid). This correlation means a situation in which the insurer achieved good results and thus led to the payment of a profit (in the form of a dividend). Niehaus (2018) noted that the sensitivity of dividends to performance increases significantly during the financial crisis. Nevertheless, a negative relationship between the financial results of an insurance company and capital flows may seem counterintuitive. The visible discrepancy is easily solved by the institutional context: in the insurance sector (and especially in the life insurance sector) the deterioration of the company’s capital, resulting from unfavourable financial results, may be associated with a reduced demand for the company’s products and a negative impact on its value. Consequently, for insurers, it may cause an increase in the risk of losing liquidity, the effects of which may spread to the entire financial market in the long run. Powell et al. (2008) focused on verifying the relationship between the dividend and the results of the insurance company (belonging to the business group). They demonstrated a positive link between capital transfers and the performance of the insurance company. They argued that the additional financing was provided to the companies in the group with the most favourable investment opportunities. At the same time, they indicated that affiliates with financial problems do not receive such aid.

3 Research Design In the study, the data of all life insurance companies operating in Poland obtained from the all available Solvency Financial Capital Reports (SFCR) published on the website were used. The manually collected database contains information for the years 2016–2018, i.e. the entire term of the Solvency II Directive and includes data from all Polish life insurance enterprises (a total of 24 life insurers). The total number of observations analysed in our research is limited by the total number of life insurers in Poland and the length of obligatory use of the Solvency II rules. The study was carried out on a sample of panel data, for maximum possible period of 3 years (2016–2018). The choice of the most effective estimator to estimate the dependence between individual variables in the case of panel data is determined by the results obtained

224

J. Głód et al.

during the Hausman test. In this test, the zero hypothesis assumes that the individual effects are independent of the explanatory variables, resulting in a result that the applied estimator is not encumbered. Where it is not discarded, the estimator for a model with random effects is considered more effective. Otherwise, i.e. when a zero hypothesis is discarded, it is assumed to be a more effective model estimator with fixed effects (similarly, it is considered to be an unencumbered estimator in such a situation). The model described below examines how the capital requirements calculated by companies (SCR together with individual modules and MCR), together with the amount of premium written and the level of own funds influence the amount of dividend paid. Our research allows to verify the research hypotheses formulated. The study covers the panel data of 24 life insurance companies for the years 2016–2018 published in SFCR. Table 1 presents variables used in the model. The data analysed in the model come from the reports of all 24 insurance companies operating in Poland, from the period of 3 years. However, for some variables it is visible that this figure is reduced by one observation. This is due to the fact that some insurance companies did not calculate the capital requirement for specific risks (e.g. market risk) due to lack of such an activity (Table 2). The factors which influence the above disproportion are, among others: dividend policy, the size of the business, financial results achieved, legal regulations. Analysis of capital requirements, both their main “representatives”—scr and mcr, as well as submodules—scr_market, scr_default, scr_life, scr_health, scr_operat, Table 1 Definition of the variables

Variables

Definition of variables

Explained variable lndiv

Ln of amount of dividend paid by insurance companies

Regressors lnpremium

Ln of gross written premium

lneoaol

Ln of excess of assets over liabilities

lnownfunds

Ln of own capital of insurer

lnmcr

Ln of minimum capital requirements

lnscr

Ln of solvency capital requirements

lnscr_market

Ln of capital requirements for market risk

lnscr_default

Ln of capital requirement for counterparty default risk

lnscr_life

Ln of capital requirements for life underwriting risk

lnscr_health

Ln of capital requirements for health risk

lnscr_operat

Ln of capital requirements for operational risk

Dividends of Life Insurance Companies and the Solvency …

225

Table 2 Descriptive statistics of the variables Variable

N

Std. err.

Min

Max

lndiv

72

Mean 6.5106

5.2856

0.0000

14.1930

lnpremium

72

12.8731

1.5138

9.3638

15.9630

lneoaol

72

12.6392

1.5832

9.8817

16.4448

lnownfunds

72

12.5974

1.5558

9.9142

16.4358

lnmcr

72

10.7851

1.0981

9.6629

14.0197

lnscr

72

11.6269

1.5037

8.4295

15.0591

lnscr_market

71

10.4054

2.0769

3.6636

14.7071

lnscr_default

71

8.0083

2.0188

1.6094

12.3590

lnscr_life

71

11.2180

1.7170

6.9660

14.7065

lnscr_health

72

6.4913

4.8428

0.0000

13.1197

lnscr_operat

71

9.3102

1.4079

5.9375

12.5960

Bold indicates explained variable

allows to state that their sizes in the tested sample behave in an analogous way. However, in their case, discrepancies between individual values, apart from the abovementioned factors, may be the result of not conducting business in some areas of risk. This can be observed for the requirement corresponding to the measurement of insurance risk for health insurance (the minimum value for this variable is: PLN 0 thousand). The highest ratios are achieved by the requirements responsible for market and life underwriting risk. The main objective of the study is to examine the impact of capital requirements set under the Solvency II Directive on the amount of paid dividends in life insurance companies. Therefore, in the study we estimate the model described in the following equation: lndivit = β0 + β1lnpr emium + β2 lneaol + β3lnown f unds + β4 lnmcr + β5lnscr + β6lnscr _mar ket + β7lnscr _de f ault + β8lnscr _li f e + β9lnscr _health + β10 lnscr _operat + uit + εit However, due to the strong correlation between the estimated variables (presented in Table 3), equation was divided into nine components (X) in order to obtain meaningful estimates. Each of them was estimated separately, and the results of the estimates are presented in Table 4. The form of the individual equations is as follows: lndivit = β0 + β1 X + β2 lnscr _health + uit + εit

0.61*

0.59*

0.56*

0.56*

0.44*

0.35*

0.64*

0.27*

0.58*

2. lnpremium

3. lneoaol

4. lnownfunds

5. lnmcr

6. lnscr

7. lnscr_market

8. lnscr_default

9. lnscr_life

10. lnscr_health

11. lnscr_operat

0.94*

0.15*

−0.08*

0.94*

0.94*

0.74*

0.93*

0.97*

0.93*

0.99*

1

3

0.89*

0.82*

0.90*

0.92*

0.80*

0.89*

0.89*

1

2

*significant at the level of 5% (95% confidence level)

1

0.50*

1. lndiv

1

Table 3 Correlation matrix of the variables

0.94*

0.15*

0.94*

0.73*

0.92*

0.97*

0.94*

1

4

0.85*

0.28*

0.89*

0.61*

0.83*

0.93*

1

5

0.93*

0.12*

0.98*

0.70*

0.93*

1

6

0.89*

0.02*

0.88*

0.77*

1

7

0.75*

−0.18*

0.65*

1

8

0.92*

0.11*

1

9

0.05*

1

10

1

11

226 J. Głód et al.

38.48%

R-sq between

45.01%

22.4900*** 42.90%

20.7500*** 36.44%

15.5400***

5.9900*

(8.3696)

−21.8941***

(0.1752)

0.2590##

(0.7936)

2.4778***

lndiv

40.10%

17.8600***

5.6800*

(6.3976)

−17.6795***

(0.1654)

0.3335**

(0.5486)

1.8943***

lndiv

***Significant at the level of 1%, **Significant at the level of 5%, *Significant at the level of 10%, ## Significant at the level of 15%

14.3700***

Test Walda

7.3400**

(6.6651)

(6.4692)

(6.9142)

7.5800**

−21.0644***

−21.6134***

16.0415**

7.1400**

(0.1648)

(0.1625)

(0.1671)

0.3039*

(0.5317)

2.0323***

lndiv

0.2951*

(0.5146)

2.0736***

lndiv

0.4352***

(0.5200)

1.5324***

lndiv

Test Hausmana

_cons

lnscr_health

lnscr_operat

lnscr_life

lnscr_default

lnscr_market

lnscr

lnmcr

lnownfunds

lneoaol

lnprzypis

Table 4 Results

27.98%

11.1100***

5.7800*

(4.3512)

−5.9012##

(0.1774)

0.4164**

(0.3938)

0.9314**

lndiv

20.48%

7.4900**

8.1500**

(3.4501)

−0.8477##

(0.1854)

0.4691**

(0.3600)

0.5367##

lndiv

50.60%

26.7100***

6.9800**

(5.0484)

−17.9981***

(0.1532)

0.3193**

(0.4449)

1.9975***

lndiv

44.95%

16.4500***

9.1700**

(5.3280)

−12.8459**

(0.1610)

0.3706**

(0.5584)

1.8192***

lndiv

Dividends of Life Insurance Companies and the Solvency … 227

228

J. Głód et al.

The control variable in models was the scr_health variable which measures the capital requirement for underwriting risk in health insurance. This requirement represents the third largest volume of life insurance activity in terms of resources devoted to it. Based on the Table 3, the largest correlation with the dividend is observed for: surplus of assets over liabilities (lneoaol), level of own funds held (lnownfunds), capital requirement for life insurance risk (lnscr_life) and capital requirement for operational risk (lnscr_operat) variables. The weakest correlation is for the lnscr_health, used as a control variable in the estimated models. A strong correlation (92% on average) is between the financial characteristics: premium written (lnpremium), surplus of assets over liabilities (lneoaol), own funds (lnownfunds) and the capital requirements: lnscr and lnmcr. Our results (Table 4) do not give a basis to reject the hypotheses that dividend pay-outs are correlated with the Solvency Capital Requirements, its sub-modules and the Minimum Capital Requirements. There is a positive relationship between the dividend paid out and the financial standing of life insurance companies (premium written, excess of assets over liabilities and own funds). Correlation between all the variables estimated in the models is high, which may indicate a strong relationship, which in turn may be caused by: existing legal regulations, as well as internal regulations (applied directly in life insurance companies, or also imposed by “recipients” of capital transfers). According to the theoretical assumptions described above, the estimator used to verify the hypothesis is a random effects model. The test by which the most effective form of the model was determined is the Hausman test. The results obtained in this diagnosis did not allow to reject the zero hypothesis on the effectiveness of the model used, i.e. the model assuming the existence of individual, impossible to estimate random effects (at the significance levels: 5 or 10%). The above conclusions are also confirmed by the results obtained by estimating Wald’s statistics, on the basis of which an alternative hypothesis was adopted, indicating that a better estimator for the analyzed dependencies is a random effects model. The preliminary analysis of the estimation allows to state that all of the analysed variables positively influence the amount of dividends paid by life insurance companies. The change in some of them (by 1%) results in a more than double increase in the amount of funds paid out as a share in the generated profit. The greatest influence is observed for the lnmcr, lneoaol, lnownfunds variables. An increase in the minimum capital requirement of 1% results in an increase in dividend payment of 2.48%. An increase in the value of the excess of assets over liabilities by 1% results in an increase in the dividend payment by 2.07%. An increase in the volume of own funds held, not burdened with any liabilities, contributes to an increase in the amount of capital transferred (in the form of dividends) by 2.03%. To conclude the life insurers within the analyzed risks follow a cautious investment policy. They apply the prudent investor principle. The volatility of market risk and insolvency risk does not affect the dividends.

Dividends of Life Insurance Companies and the Solvency …

229

4 Conclusions The result of the empirical analysis showed a positive relation between all selected variables and dividend. But the most significant linkage is between dividends and Minimum Capital Requirements, the surplus of assets over liabilities and the volume of own funds held, not burdened with any liabilities. This linkage can be explained by a few reasons: when insurers need to allocate higher capital, it has two sources: internal or external. The internal source is a profit part of which can be divided between shareholders (owners) as dividends. If managers decide to fit capital requirements using profit, it will deal with the reduction of dividends. But dividend cuts or omissions could be interpreted as negative signals by external stakeholders (Basse 2019). From the other side, managers can increase the dividend to enlist external sources of capital. The higher dividends can be a signal for stakeholders which confirms a good financial performance of the company and its potential growth. In the case of the Polish insurance market, there is no clear evidence of the dividend smoothing as a phenomenon as like as in Italian insurance market (Basse et al. 2011). With the raising of the required capital (MCR) it is also increases dividend pay-out, but at the same time companies have a good financial performance. So, Basse’s (2019) thesis about weak performance of financial company in the result of high dividend pay-out can be discussed. It is also possible to agree with Gerber and Shiu (2005) in the context of payout barrier and it was confirmed by a positive correlation between dividends and surplus of assets over liabilities, the volume of own funds held, not burdened with any liabilities and gross written premiums. It means that companies in Poland having a good financial performance don’t reduce dividends as well as Niehaus (2018) showed. But one of the main factors of dividend policy among Polish insurers is supervisor recommendation about dividend pay-outs. These recommendations are published each year and contain the algorithm for dividend calculation. The amount of dividend is closely relating to the capital requirements as a percentage criterium. That explains why it has to be taken into consideration as a kind of restriction for dividend policy. Our results contribute to the literature by providing an evidence on a linkage between the capital requirements and the dividend policy of life insurers. Based on the positive character of such a relation, we state that in the case of the Polish market, dividends paid by life insurers were not limited by the new capital requirements. Acknowledgements The article was prepared during the project financed under the Bekker Programme by the Polish National Agency for Academic Exchange in 2019, Contract No. PPN/BEK/2018/1/00426/U/00001 (research scholarship at the Queensland University).

230

J. Głód et al.

References Akhigbe A, Borde S, Madura J (1993) Dividend policy and signaling by insurance companies. J Risk Insur 60(3):413–428. https://doi.org/10.2307/253036 Ashraf BN, Bibi B, Zheng C (2016) How to regulate bank dividends? Is capital regulation an answer? Econ Model 57:281–293. https://doi.org/10.1016/j.econmod.2016.05.005 Basse T (2019) The impact of the financial crisis on the dividend policy of the European insurance industry: additional empirical evidence. Z Gesamte Versicherungswissenschaft 108(1):3–17. https://doi.org/10.1007/s12297-019-00429-w Basse T, Reddemann S, Riegler J (2011) Dividend policy and the global financial crisis: empirical evidence from the Italian insurance industry. ZVersWiss 100(1):131–140. https://doi.org/10.1007/ s12297-010-0092-4 Chen CY (1990) Risk-preferences and tax-induced dividend clienteles: evidence from the insurance industry. J Risk Insur 57(2):199–219 Dickens R (2009) In: Baker HK (ed) Dividend policy in regulated industries in dividends and dividend policy. Wiley, Hoboken, pp 1–19 Dong H, Yin C (2012) Complete monotonicity of the probability of ruin and de Finetti’s dividend problem. J Syst Sci Complex 25(1):178–185 Finetti B (1957) Su un’impostazione alternativa della teoria collettiva del rischio. In: Proceedings of transactions of xv international congress of actuaries, NY, pp 433–443 Gerber HU, Shiu ES (2005) On optimal dividend strategies in the compound Poisson model. North Am Actuarial J 10(2):76–93. https://doi.org/10.1080/10920277.2006.10596249 Lee C, Forbes S (1980) Dividend policy, equity value, and cost of capital estimates for the property and liability insurance industry. J Risk Insur 47(2):205–222 Lee C, Forbes S (1982) Income measures, ownership, capacity ratios and the dividend decision of the non-life insurance industry: some empirical evidence. J Risk Insur 59(2):269–289 Niehaus G (2018) Managing capital via internal market transactions: the case of life insurance. J Risk Insur 85(1):69–106 Powell LS, Sommer DW, Eckles DL (2008) The role of internal capital markets in financial intermediaries: evidence from insurers groups. J Risk Insur 75(2):439–461 Reddemann S, Basse T, von der Schulenburg JM (2010) On the impact of the financial crisis on the dividend policy of the European insurance industry. Geneva Pap Risk Insur Issues Pract 35(1):53–62 Rozeff MS (1982) Growth, beta and agency costs as determinants of dividend payout ratios. J Financ Res 3:249–259. https://doi.org/10.1111/j.1475-6803.1982.tb00299.x

Fragility or Contagion? Properties of Systemic Risk in the Selected Countries of Central and East-Central Europe Marta Kara´s and Witold Szczepaniak

Abstract The paper quantifies two aspects of systemic risk in selected countries of Central and Eastern Europe over the period of 2006–2018: fragility and contagion. The paper adds to existing literature in three ways. Firstly, we present the results of calculation of two well-known measures (SRISK and CoVaR) for countries for which these measures were not calculated before due to technical problems with application, which we overcome by using our own proposals of modification. Secondly, by proposing our own methods of proxying we are able to include a significantly bigger number of systemically important financial institutions in the measurement of systemic risk than any other scientific paper quantifying this risk for the analyzed region. Thirdly, we propose an analysis of the two aspects of systemic risk (fragility or contagion) in the periods of the global financial crisis, European debt crisis and recently, during the economic stagnation period. The conclusions from the paper may serve as an important insight into how systemic risk changes with changing economic and financial markets’ condition and how it varies between different Central and East-Central European countries. Keywords Systemic risk · Global financial crisis · Emerging financial systems · Systemic fragility · Contagion JEL Classification G01 · G21 · C58 · E44 · E58

1 Introduction Research on systemic risk has progressed significantly since the global financial crisis. Works by Benoit et al. (2017) but also Bisias et al. (2012) and Hattori et al. M. Kara´s · W. Szczepaniak (B) Department of Financial Investments and Risk Management, Wroclaw University of Economics and Business, Wroclaw, Poland e-mail: [email protected] M. Kara´s e-mail: [email protected] © Springer Nature Switzerland AG 2020 K. Jajuga et al. (eds.), Contemporary Trends and Challenges in Finance, Springer Proceedings in Business and Economics, https://doi.org/10.1007/978-3-030-43078-8_19

231

232

M. Kara´s and W. Szczepaniak

(2014) provide the overview of an vast number various measures. The analysis of the theoretical and empirical findings points to an observation that systemic risk materializes in three areas: in sudden changes of systemic liquidity, in fragility of financial institutions that builds-up over time and in contagion effects that intensify in systemic events. The aim of this paper is to quantify two aspects of systemic risk in the selected countries of Central and East-Central Europe1 over the period of 2006–2018: fragility and contagion. The paper adds to existing literature in three ways. Firstly, we present the results of calculation of two well-known measures (SRISK and CoVaR) for countries for which these measures were not calculated before due to application problems. We overcome them by using our own proposals of modification. Secondly, by proposing our own methods of proxying, to the best of our knowledge, we are able to include a significantly bigger number of systemically important financial institutions in the measurement of systemic risk than any other scientific paper quantifying this risk for the Central and East-Central European region. Thirdly, and perhaps most importantly, we propose an analysis illustrating which of the two aspects of systemic risk (fragility or contagion) was the dominating factor in the periods of the global financial crisis, European debt crisis and recently, during the economic stagnation period. The conclusions from the paper may serve as an important insight into how systemic risk changes with changing economic and financial markets’ condition and how it varies between different the Central and East-Central European countries. The layout of the paper is as follows. Firstly we briefly define the two notions important for the research: the financial system and systemic risk, outlining the literature findings that support the distinction of fragility and contagion effects in this risk. Then we present the selected systemic risk measures, outlining our propositions of modification and the technical detail behind the computations. The empirical section reports the results of systemic risk quantification, including the analysis of fragility and contagion effects. The paper concludes with policy recommendations.

1 The

study follows the geographic definitions of Central and Eastern Europe and those based on cultural proximity (see e.g. Encyclopedia Britannica, The World Factbook or German Standing Committee on Geographical Names). Please note that this definition differs from that of e.g. OECD. This choice allows for comparison of systemic risk in the geographical region, attaining the economic divergence at the same time. Of special interest to us is the ability to present the contrast in systemic risk developments between the countries with most developed financial systems (Germany), those which recently entered the group of economically developed (e.g. Poland) and those which are still in the emerging phase of economic development (e.g. Romania). We also consider Baltic countries due to their specificity in terms of the impact of financial systems of Sweden and Norway combined with financial system proximity to the other analyzed states. Such a mix allows us to present a more comprehensive picture of systemic risk in the geographical region of Central and East-Central European region. Some countries such as e.g. Albania, Belarus, Serbia, Slovenia and others had to be excluded from the study due to the shallowness/illiquidity of their stock markets rendering necessary data nonexistent.

Fragility or Contagion? Properties of Systemic Risk …

233

2 Financial System and Systemic Risk—Definitions A wider literature analysis allows to systematize at least two main areas for systemic risk triggers materialization: • fragility of financial institutions, theoretically elaborated on e.g. by Boissay et al. (2016), Adrian and Shin (2014), Martin et al. (2014), Freixas and Rochet (2013), Acharya and Yorulmazer (2008); the empirical studies of e.g. Benoit et al. (2016), Adrian and Shin (2014), Gabrieli and Georg (2014), Acharya and Merrouche (2013), Freixas and Rochet (2013); supported by fragility-based systemic risk measures of Brownlees and Engle (2017), Blei and Ergashev (2014), Jakubik and Slaˇcík (2013) or even the early measure by Nelson and Perli (2007); • risk spill-over, notionally studied by e.g. Biais et al. (2016), Acemoglu et al. (2015), Koeppl et al. (2012); empirically studied by Markose et al. (2012), Iyer and Puri (2012), Iyer and Peydró (2011) and others; measured with, inter alia, the proposals of Adrian and Brunnermeier (2016), Diebold and Yilmaz (2014), Hollo et al. (2012), but also De Nicolo and Lucchetta (2011). In systemic risk analysis the financial system is most commonly defined as a set of Systemically Important Financial Institutions (SIFIs)—the “institutions whose distress or disorderly failure, because of their size, complexity and systemic interconnectedness, would cause significant disruption to the wider financial system and economic activity” (FSB 2011, p. 1). From this follows that the financial systems of the analyzed countries may be modelled as collectives of locally systemically important financial institutions (GSIFIs and OSIIs), quantifying systemic risk as reaction of these collectives to low probability systemic triggers.

3 Selected Risk Measures and the Estimation Methods The empirical study utilises two well-established quantile-based systemic risk measures, one based on the concept of Value at Risk (VaR) and one on Expected Shortfall (ES). Based on our previous studies (see Kara´s and Szczepaniak 2017; Jajuga et al. 2017), in the empirical part of this study we include: Adrian and Brunnermeier’s (2016) Conditional VaR (CoVaR) (with the modifications of Kara´s and Szczepaniak 2017, 2019) and (Brownlees and Engle’s 2017) SRISK. These measures have been successfully applied to measure systemic risk around the global financial crisis for advanced economies, including especially USA and the biggest Western Europe countries, but also for the CEE countries (see the literature cited above). Moreover, these measures have been proven to be complementary. The measures based on Marginal Expected Shortfall (MES) answer the question: which institutions are most fragile; whereas CoVaR indicates which financial institution contributes most to systemic risk. Adrian and Brunnermeier (2016) themselves call

234

M. Kara´s and W. Szczepaniak

CoVaR complementary to high-frequency marginal expected shortfall-based measures (in: extended version of the working paper—Adrian and Brunnermeier 2011, p. 5), such as SRISK. At the same time, Benoit et al. (2014, p. 17) empirically show very low concordance between SRISK and Delta CoVaR—only 9.9%—while the similarities between the other measures studied by the authors are at approximately 50%. Kuziak and Piontek (2018, p. 155) on the other hand, confirm for Poland that Delta CoVaR is not driven by fragility of financial institutions, demonstrating that the CoVaR measure is a risk spill-over measure.

3.1 Fragility Measure—SRISK The SRISK measure draws from the Expected Shortfall (ES) concept: N  E S Mt (u) = E t−1 (R Mt |R Mt < u) = wit E t−1 (Rit |R Mt < u). i=1

Marginal Expected Shortfall (MES), indicating the extreme contribution os the system s to systemic risk is a partial derivative of ES: M E Sit (u) =

∂ E S Mt (u) = E t−1 (Rit |R Mt < u), ∂wit

Assuming a crisis—an equity decline conditional on the system’s equity falling below the assumed (marginal) threshold C in the next six months, one may define the Long Run Expected Shortfall (LRMESS) as:   L R M E Si,t (C) = 1 + exp γ · M E Si,t (C) . SRISK indicates the shortage of equity (increased quasi-leverage—[D it ; Wit ]) expected in the event of a system-wide crisis and it draws from the above described LRMES: S R I S K it = max[0; k(Dit + (1 − L R M E Sit )Wit ) − (1 − L R M E Sit )Wit ] Individual LSIFIs’ SRISKS are aggregated to obtain the final System-level SRISK.

3.2 Risk Spill Over Measure—Delta CoVaR Conditional Value at Risk of the system, measures systemic value at risk on condition that a given analysed financial institution materializes its marginal risk:   P R Mt ≤ CoV a Rit| |C(Rit ) = α.

Fragility or Contagion? Properties of Systemic Risk …

235

In this and previous studies (see e.g. Kara´s and Szczepaniak 2017), we use Delta CoVaR—the measure derived from CoVaR, which stands for the difference between the system’s VaR if the analysed SIFI is at risk financially and the system’s VaR if the financial condition of this institution is normal (median): q

q

q

q

CoV a Rit = (CoV a Rit |Rit = V a Rit ) − (CoV a Rit |Rit = V a Rit0.5 ) We obtain system-wide Delta CoVaR by aggregating individual Delta CoVaRs of all analysed Local Systemically Important Financial Institutions.

3.3 Estimation For estimating the quantile-based systemic risk measures, we adopt a twodimensional process, where Rt is a vector (Rst , Rit ) and H t is a conditional variance–covariance matrix:  Rt = Ht υt , The matrix takes the form of:  Ht =

 σit σst ρit σst2 , σit σst ρit σit2

with a conditional standard deviation σ for the system’s (s) rate of return and the institution’s (i) rate of return, as well as the conditional correlation ρ it . At the  same time, the i.i.d. υ t vector (εit , εst ) is such that E(υ t ) equals zero and E(υt υ t ) = I2 is a two by two units matrix (see Benoit et al. 2014). Conditional volatility σ is estimated using GJR-GARCH model, while the conditional correlation of the financial institution and the financial system (ρ it ) is estimated with the GJR-GARCH DCC model. The conditional expected value for each LSIFI is based on: V a Rit = σit Fi −1 (q) q

For the LSIFIs’ contribution to the systemic conditional VaR, we apply the equation:   q q CoV a Rit = γˆ V a Rit − V a Rit0.5 , where: γˆ = equation:

ρˆi,t σˆ s,t σˆ i,t

. The estimation of the marginal expected shortfall uses the

      q  2 ˆ E t−1 εi,t |εs,t < κ , M E Si,t V a Rs,t = σˆ i,t ρˆi,t Eˆ t−1 εs,t |εs,t < κ + σˆ i,t 1 − ρˆi,t

236

M. Kara´s and W. Szczepaniak

where:   Eˆ t−1 εs,t |εs,t < κ =

 κ−εs,t  εs,t t=1 K h   κ−ε T s,t t=1 K h

T

and 



Eˆ t−1 εi,t |εs,t < κ = for κ =

q

V a Rs,t σs,t

, K (x) =

function—and h = T (2017):

−1 5



x h

−∞ k(u)du

 κ−εs,t  εi,t t=1 K h   κ−ε T s,t t=1 K h

T

,

for k(u)—the normal distribution density

. LRMESS is determined following Brownlees and Engle

  L R M E Si,t (C)  1 + exp 18 · M E Si,t (C) . In the selection of the financial institutions for the study we follow the regulators who identified global and other systemically important institutions (GSIFI and OSII) for the region analysed. Table 1 presents the list and systemic characteristics of these biggest and/or most globally interconnected SIFIs. One of the used modifications consists in proxying (see Kara´s and Szczepaniak 2019) which allows us to expand the sample of included banks by about 50% in comparison to other contemporary studies of the selected geographical region.

4 Empirical Results and Short Discussion The following section presents the empirical results of the calculations of fragility and contagion for each of the ten analyzed countries. We used mainly Thomson Reuters DataStream for collecting the raw data, as well Microsoft Excel and several Matlab tool-packs for calculations. The results were normalized using the min–max normalization procedure to enable comparisons of the two aspects of systemic risk. We also run quantile regressions for the comparison of SRISK and Delta CoVaR for each country to confirm that these two aspects of systemic risk vary significantly in the tails of their distribution (showing that they describe different aspects of systemic risk). Figures 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10, as well as Table 2 and the Appendix, present the results. From the graphs presented above, as well as the descriptive statistics provided in Appendix, one may ascertain that systemic risk in all analyzed countries increased significantly in the period of the global financial crisis (2008/2009) and (less prominently) during the European public debt crisis (2012/13). Of particular interest is the observation that for all the analyzed countries except the Baltic states and Poland,

Fragility or Contagion? Properties of Systemic Risk …

237

Table 1 Local SIFIs for the analysed region

Bulgaria

Czechia

Bank

Avg. SIS

Type

Ultimate EU parent (owns or controls)

UniCredit Bulbank A.D.

1880

OSII

UniCredit S.p.A.

United Bulgarian Bank A.D.

1120

OSII

KBC Group NV

First Investment Bank A.D.

1100

OSII

x

DSK Bank EAD

1040

OSII

OTP Bank Nyrt.

Societe Generale Expressbank A.D.

720

OSII

Société Générale S.A.

Raiffeisenbank (Bulgaria) EAD

670

OSII

Raiffeisen Bank International A.G.

Eurobank Bulgaria A.D.

630

OSII

Eurobank Ergasias S.A.

Central Cooperative Bank A.D.

520

OSII

x

Piraeus Bank Bulgaria A.D. ˇ Ceskoslovenská obchodní banka, a.s.

310

OSII

Piraeus Bank S.A.

2100

OSII

KBC Group NV

Komerˇcní banka, a.s. ˇ Ceská spoˇritelna, a.s.

1475

OSII

Société Générale S.A.

1435

OSII

Erste Group Bank A.G.

UniCredit Bank CZ and SK, a.s.

1100

OSII

UniCredit S.p.A.

500

OSII

Raiffeisen-Landesbanken-Holding GmbH

Swedbank AS

3190

OSII

Swedbank A.B.

AS SEB Pank

1930

OSII

Skandinaviska Enskilda Banken A.B.

350

OSII

Luminor Group A.B.a

OTP Bank Nyrt.

3095

OSII

x

UniCredit Bank Hungary Zrt.

960

OSII

UniCredit S.p.A.

Kereskedelmi és Hitelbank Zrt.

830

OSII

KBC Group NV

ERSTE BANK HUNGARY Zrt.

655

OSII

Erste Group Bank A.G.

Raiffesenbank, a.s. Estonia

AS Luminor Bank Hungary

(continued)

238

M. Kara´s and W. Szczepaniak

Table 1 (continued) Bank

Latvia

Lithuania

Poland

Romania

Avg. SIS

Type

Ultimate EU parent (owns or controls)

Raiffeisen Bank Zrt.

600

OSII

Raiffeisen Bank International A.G.

CIB Bank Zrt.

420

OSII

Intesa San Paolo S.p.A.

Swedbank A.S.

425

OSII

Swedbank A.B.

SEB banka A.S.

420

OSII

Skandinaviska Enskilda Banken A.B.

DNB banka A.S.

425

OSII

Luminor Group A.B.a

SEB bankas A.B.

4280

OSII

Skandinaviska Enskilda Banken A.B.

Luminor Bank A.B.

2050

OSII

Luminor Group A.B.a

Swedbank A.B.

1895

OSII

Swedbank A.B. (itself; Sweden)

Šiauli˛u bankas A.B.

640

OSII

x

PKO BP S.A.

1580

OSII

x

Bank Polska Kasa Opieki S.A.

1050

OSII

x

Bank Zachodni WBK S.A. ´ aski ING Bank Sl˛ S.A.

960

OSII

Banco Santander

950

OSII

ING Bank N.V.

mBank S.A.

930

OSII

Commerzbank A.G.

Millennium Bank S.A.

424

OSII

Banco Comercial Portugues

Bank Handlowy w Warszawie S.A.

440

OSII

x

Deutsche Bank Polska S.A.

400

OSII

Deutsche Bank A.G.

Banca Transilvania S.A.

1620

OSII

x

UniCredit Bank S.A.

1525

OSII

UniCredit S.p.A.

Banca Comercial˘a Român˘a S.A.

1390

OSII

Erste Group Bank A.G.

BRD - Groupe Societe Generale S.A.

1165

OSII

Société Générale S.A.

Raiffeisen Bank S.A.

1000

OSII

Raiffeisen Bank International A.G. (continued)

Fragility or Contagion? Properties of Systemic Risk …

239

Table 1 (continued) Bank

Slovakia

Germany

Avg. SIS

Type

Ultimate EU parent (owns or controls)

Alpha Bank România S.A.

445

OSII

Alpha Bank

OTP Bank Romania S.A.

305

OSII

OTP Bank Nyrt.

Garanti Bank S.A.

300

OSII

Turkiye Garanti Bankasi AS

Všeobecná Úverová Banka A.S.

2070

OSII

Intesa San Paolo S.p.A.

Slovenská Sporiteˇlˇna A.S.

1800

OSII

ERSTE Group Bank A.G.

Tatra Banka A.S.

1390

OSII

Raiffeisen-Landesbanken-Holding GmbH

ˇ Ceskoslovenská Obchodná Banka A.S.

1205

OSII

KBC Group N.V.

Deutsche Bank A.G.

2765

GSIFI

x

Commerzbank A.G.

830

OSII

x

Unicredit Bank A.G.

470

OSII

UniCredit Group

ING DiBa A.G.

145

OSII

ING Bank N.V.

Source Own study, raw data accessed from EBA’s database (2019) Avg SIS Average Systemic Risk Score, OSII Other Systemically Important Institution, GSIFI Global Systemically Important Financial Institution (EBA 2019) a DNB and Nordea before 2018 Fig. 1 Germany. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

1 0.9

Germany Fragility vs. Contagion SRISK deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

240 Fig. 2 Poland. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

M. Kara´s and W. Szczepaniak Poland Fragility vs. Contagion 1 SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Fig. 3 Czechia. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Czechia Fragility vs. Contagion 1 0.9

SRISK deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

fragility quantified from 2007 to 2019 increased. This shows that the global financial crisis and the following strains of the financial system left a permanent mark in the region risk-wise. Meanwhile, in term of the risk spill-over potential, all countries show patterns similar to this of Germany, pointing to the necessity of the transnational monitoring of systemic risk. Interestingly Baltic financial systems are most prone to contagion, confirming the negative impact of the business models of the foreign parent banking institutions. Finally it is worth noticing that systemic risk generally follows the same patterns regardless of the level of development of the economy of the financial system in each of the countries.

Fragility or Contagion? Properties of Systemic Risk … Fig. 4 Slovakia. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

241

Romania Fragility vs. Contagion 1 SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Fig. 5 Hungary. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

1

Hungary Fragility vs. Contagion SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Our final exercise in this study is quantile regression whose purpose is to check in a statistically prudent manner whether the differences in the fragility-based measure (SRISK) and contagion-based measure (Delta CoVaR) are significant. Thus we aim to that both contagion and fragility drive systemic risk in the analyzed region, but with varying effect over time. We choose to investigate three quantiles of distribution: left and right tail, as well as the middle area. This corresponds to three states of systemic risk materialization: overage state of the system (q = 0.5), periods of especially low systemic risk (q = 0.95) and periods of the biggest systemic distress (q = 0.05). In order to confirm significance of the R2 statistic we run 3 types of tests, as data is characterized by atypical and varying shapes of distribution (see Appendix). As can

242 Fig. 6 Romania. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

M. Kara´s and W. Szczepaniak Romania Fragility vs. Contagion 1 SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Fig. 7 Bulgaria. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

Bulgaria Fragility vs. Contagion 1 SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

be seen in Table 2, the results of the tests vary between the quantiles. As expected, in the quantile regarding the distress period the result is sensitive to the type of test. All in all however, the results confirm that the differences between fragility and contagion are significant, these two drivers of systemic risk are different, they do coexist in all the analyzed periods and they change with the financial system condition—they become most different in distress (in the negative tail of the distribution where systemic risk materializes). This allows us to turn to a more in-depth analysis of these two aspects of systemic risk. Structural properties of the region play a key role in its systemic risk. In Central and East-Central Europe foreign owners control over 90% of the total banking assets

Fragility or Contagion? Properties of Systemic Risk … Fig. 8 Lithuania. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

1

243

Lithuania Fragility vs. Contagion SRISK

0.9

deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Fig. 9 Latvia. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

1 0.9

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Lativia Fragility vs. Contagion SRISK deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

in the region (Radulescu et al. 2018, pp. 7–8). A great majority of these banks are local SIFIs. This poses a macroprudential challenge. In fact the studied countries used a wide variety of macroprudential tools even before the global financial crisis (Dumicic 2018, p. 10), unfruitfully. In her analysis of eleven CEE’s prudential systems, Dumicic (2018) points out, that most of the subsidiaries of big Western European banks were able to circumvent regulations via their close ties to mother institutions. This was also possible due to the existing internal markets identified in an empirical study of De Haas and van Lelyveld (2010). An example is how Romanian and Bulgarian foreign-owned banks (especially Greek ones) deleveraged sharply, affecting credit provision to the real economy (Radulescu et al. 2018, pp. 3–4), while local regulators were not able to counteract

244 Fig. 10 Estonia. Source Own study; raw data accessed from Thomson Reuters DataStream, EBA’s database and annual reports of the studied institutions

M. Kara´s and W. Szczepaniak

1 0.9

Estonia Fragility vs. Contagion SRISK deltaCoVaR

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

this process. In Lithuania, Latvia and Estonia, not only do the foreign-own bank subsidiaries adopt the business models of their parent—even when this is not optimal for their local environment, but they are also strongly dependent on all the decisions of their parents (Joˇcien˙e 2015, p. 51). Barkauskaite et al. (2018) find this to be an important macroprudential factor, pointing to inadequacies of the existing methods used by the regulators. The above described properties suggest that fragility of the region may play a crucial role in systemic risk, while the links to the Western European financial sector might make the analyzed systems prone to contagion. The choice to study fragility and contagion in the selected financial systems was dictated by these unique properties. Fragility relates to a state in which the immunity of a given financial institution to a systemic trigger is significantly lower. Accumulation of fragility seems a key aspect of financial stability, as it directly determines the immunity of a given system to a systemic trigger (Benoit et al. 2017). For the financial system, fragility may be created either by a set of fragile systemically important institutions or the structural properties of the system itself, such as the level of common exposures which results from the decisions made by financial system participants. Thus it may come from concentration risks, especially tail risk and from various kinds of overexposure to debt and default. In the analyzed region it comes from all these characteristics, specifically due to the ownership structure of the banking sector and the predominant business models present there. These very same characteristics may create a risk spill-over potential. Contagion may be defined as “the probability that the instability (…) will spread to other parts of the financial system with negative effects, leading to a system-wide crisis” (Smaga 2014, p. 11). It most often results from the interdependence between the participants of the financial system. Otherwise, contagion may be perceived as a transmission mechanism which is unexplainable by economic fundamentals. It translates the micro-scale risks into a systemic trigger.

Wald

0.087