European Sovereign Bond Liquidity and Central Bank Interventions [1 ed.] 9783896447425, 9783896737427

This work investigates liquidity in European sovereign bond markets. Liquidity in financial markets is often neglected w

124 94 4MB

German Pages 158 [161] Year 2018

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

European Sovereign Bond Liquidity and Central Bank Interventions [1 ed.]
 9783896447425, 9783896737427

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

Studienreihe der Stiftung Kreditwirtschaft an der Universität Hohenheim

Gerold Willershausen

European Sovereign Bond Liquidity and Central Bank Interventions

Verlag Wissenschaft & Praxis

European Sovereign Bond Liquidity and Central Bank Interventions

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

Band 56

Gerold Willershausen

European Sovereign Bond Liquidity and Central Bank Interventions

Verlag Wissenschaft & Praxis

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

D100 ISBN 978-3-89673-742-7 © Verlag Wissenschaft & Praxis Dr. Brauner GmbH 2018 D-75447 Sternenfels, Nußbaumweg 6 Tel. +49 7045 930093 Fax +49 7045 930094 [email protected] www.verlagwp.de

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

5

Preface Since the global financial crisis the European Central Bank and other central banks have amended their operational frameworks and introduced new unconventional policies. These measures included direct purchases of securities such as government bonds. A growing strand of financial research investigates the liquidity in financial markets, which have shown high degrees of stress. The implications that can be derived from research on markets of European sovereign bonds are particularly valuable, as these markets play an important role in the European financial system. This study contributes to the existing literature by investigating the liquidity of bond markets and analyzing the impact of unconventional policies. The empirical analysis finds positive returns from a liquidity supplying investment strategy in European bond market. The strategy performs particularly well during stress periods, when the funding becomes more costly and the inventory capacity of financial intermediaries declines. Moreover, monetary policies that improve funding abilities of financial intermediaries and involve direct asset purchases can have a positive e↵ect on market liquidity. Nevertheless, when considering emergency measures their short-term e↵ectives should be balanced against the potential of distorting the incentives of market participants. Hohenheim, March 2018 Prof. Dr. Hans-Peter Burghof

7

Contents List of Figures

9

List of Tables

10

List of Abbreviations

12

1 Introduction 15 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.2 Research Outline . . . . . . . . . . . . . . . . . . . . . . . . 17 1.3 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . 19 2 Market Liquidity 2.1 Market Liquidity and Securities Prices 2.1.1 Trading and Liquidity . . . . . 2.1.2 Measuring Liquidity . . . . . . 2.1.3 Pricing and Liquidity Risk . . . 2.2 Market Participants . . . . . . . . . . 2.2.1 Institutional Investors . . . . . 2.2.2 Individual Investors . . . . . . 2.3 Liquidity Crises . . . . . . . . . . . . . 2.3.1 Time-Varying Liquidity . . . . 2.3.2 Slow-Moving Capital . . . . . . 2.3.3 Central Bank Interventions . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

21 21 21 24 27 29 29 31 32 32 35 36

3 The European Government Bond Market 41 3.1 Market Characteristics . . . . . . . . . . . . . . . . . . . . . 41 3.2 Market Structure . . . . . . . . . . . . . . . . . . . . . . . . 49

8

Contents

4 Short-Term Reversals 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Literature Review . . . . . . . . . . . . . . . . . . . . . 4.3 Data and Methodology . . . . . . . . . . . . . . . . . . . 4.3.1 Sample Data . . . . . . . . . . . . . . . . . . . . 4.3.2 Portfolio Construction . . . . . . . . . . . . . . . 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Return Reversals and Portfolio Characteristics . 4.4.2 Liquidity-driven Reversals . . . . . . . . . . . . . 4.4.3 Reversals under Financial Stress . . . . . . . . . 4.4.4 Robustness: Evidence from Electronic Platforms 4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

53 53 57 59 59 61 64 64 68 73 79 86

5 Liquidity and Unconventional Monetary Policy 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . 5.2 Literature Review and Hypothesis Development . 5.2.1 Monetary Policy and Liquidity . . . . . . 5.2.2 Unconventional Policies . . . . . . . . . . 5.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Bond Market Data . . . . . . . . . . . . . 5.3.2 Variables and Summary Statistics . . . . 5.4 Single Market Analysis . . . . . . . . . . . . . . . 5.4.1 Causalities . . . . . . . . . . . . . . . . . 5.4.2 Volatility, CDS and Return . . . . . . . . 5.4.3 Unconventional Policies across Markets . 5.5 Spillovers and Flights . . . . . . . . . . . . . . . 5.5.1 VAR Approach . . . . . . . . . . . . . . . 5.5.2 VAR Results . . . . . . . . . . . . . . . . 5.6 Robustness . . . . . . . . . . . . . . . . . . . . . 5.6.1 Maturity E↵ects . . . . . . . . . . . . . . 5.6.2 Quoted Depth . . . . . . . . . . . . . . . 5.7 Conclusion . . . . . . . . . . . . . . . . . . . . .

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

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

87 87 91 91 92 95 95 96 101 102 104 110 116 116 119 129 129 130 134

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

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

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

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

6 Conclusion 137 6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Bibliography

141

9

List of Figures 3.1 General Government Debt Securities . . . . . . . . . . . . . 3.2 Holdings of Government Debt . . . . . . . . . . . . . . . . .

43 46

5.1 Rolling window regressions . . . . . . . . . . . . . . . . . . 5.2 Accumulated response of bond liquidity to endogenous variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Accumulated response of bond liquidity to endogenous variables post LTROs . . . . . . . . . . . . . . . . . . . . . . . 5.4 Regressions by maturity group . . . . . . . . . . . . . . . . 5.5 Order book depth by maturity . . . . . . . . . . . . . . . . 5.6 Regressions on depth by maturity group . . . . . . . . . . .

115 124 128 131 132 133

11

List of Tables 4.1 Summary statistics by rating group . . . . . . . . . . . . . . 4.2 Characteristics and portfolio returns of single sorted portfolios 4.3 Liquidity-weighted portfolio returns and cross-sectional regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Liquidity regression . . . . . . . . . . . . . . . . . . . . . . . 4.5 Contrarian alphas by financial stress condition . . . . . . . 4.6 Threshold regression . . . . . . . . . . . . . . . . . . . . . . 4.7 Cross-sectional regressions by state . . . . . . . . . . . . . . 4.8 Summary statistics of matched bond sample . . . . . . . . . 4.9 Liquidity-weighted portfolio on electronic platform . . . . . 4.10 Liquidity regression on electronic platform . . . . . . . . . . 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9

Bonds by issuing country and maturity group . . . . . Summary statistics by issuing country . . . . . . . . . Correlation and Granger causality by issuing country . Regressions on bid-ask spreads by issuing country . . . Structural break tests . . . . . . . . . . . . . . . . . . Summary statistics by liquidity group . . . . . . . . . Granger causality by liquidity group . . . . . . . . . . Variance decomposition of bond liquidity . . . . . . . Variance decomposition of bond liquidity post LTROs

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

62 65 70 72 76 78 80 82 83 85 97 100 103 107 111 118 120 125 127

13

List of Abbreviations AIC APP AT bps BE BIS BoE BoJ CAPM CDS CSD DE ECB EMU EONIA ES EUR EURIBOR FED FI FR FRFA GDP IE IT KfW

Akaike Information Criterion Asset Purchase Program Austria basis points Belgium Bank of International Settlement Bank of England Bank of Japan Capital Asset Pricing Model Credit Default Swap Central Securities Depository Germany European Central Bank European Monetary Union Euro Overnight Index Average Spain Euro Euro Interbank O↵ered Rate Federal Reserve Finland France Fixed-Rate Full Allotment Gross Domestic Product Ireland Italy Kreditanstalt f¨ur Wiederaufbau

14 LLR Lender of Last Resort LTCM Long Term Capital Management LTRO Long Term Refinancing Operation MiFID Markets in Financial Instruments Directive MiFIR Markets in Financial Instruments Regulation MMLR Market Maker of Last Resort MRO Main Refinancing Operations MTS Mercato Telematico dei Titoli di Stato NL The Netherlands NYSE New York Stock Exchange OIS Overnight Index Swap OMT Outright Monetary Transactions OTC Over-the-Counter PT Portugal PTF Principal Trading Firm QE Quantitative Easing SI Slovenia SMP Securities Market Program TARGET Trans-European Automated Real-time Gross settlement Express Transfer system TRACE Trade Reporting and Compliance Engine tr trillion VAR Vector Autoregression VIX US Volatility Index of the Standard&Poor’s 500 Index VSTOXX European Volatility Index of the EURO STOXX 50 Index

15

Chapter 1

Introduction 1.1

Motivation

Not long ago, European government bond markets used to be of interest only for specialized portfolio managers or central bankers. Recently, these markets have gained the attention of policy makers and the broader public. For instance, during the European debt crisis, politicians criticized speculation for driving up bond yields and worsening funding conditions of liquidity-constrained sovereigns. In particular, the surge in the spread of peripheral countries against Germany as a benchmark has been closely followed by the public. A development that some may argue is not based on fundamental evaluation, given its magnitude and speed. Did the narrow corridor of bond prices before the financial turmoil reflect something else? A potential candidate is the faulty assumption of investors that the European Monetary Union (EMU) would convert into a transfer union. Often neglected in this debate are frictions in bond markets, which begin to matter under severe circumstances. Indeed, illiquidity in financial markets, which influences the costs of transacting, among other things, has surged unevenly and strongly during this episode. Against this backdrop, the European Central Bank (ECB) directly intervened in these markets under newly developed policy programs. The

16

1.1. Motivation

securities market program (SMP) was the first instance of direct purchases of government debt. To date, its successor the Outright Monetary Transactions (OMT) has not been implemented. Recently, the Asset Purchase Program (APP) followed these initiatives, and it is expected to be in place until the end of 2017. Previous studies on unconventional policies of major central banks are concerned with the influence on asset prices. For instance, existing evidence supports the conclusion that asset purchases by the Federal Reserve (EFD) have successfully reduced bond yields.1 Nevertheless, the communicated goal for purchasing sovereign bonds under the SMP/OMT program is to address the dysfunctionality in bond markets.2 These e↵orts are due to the fact that European bonds are heterogeneous with respect to their credit risk and illiquidity, which is much in contrast to US treasury markets. Hence, the liquidity in some markets has appeared to be severely impaired, which may result in a deviation in observable prices from the underlying evaluation. To explain the strong di↵erences in market conditions during times with low trading costs and abundant liquidity and periods of market stress, a topical body of research investigates the supply by liquidity providers. Further, studies analyze the demand for liquidity by investors. Most of the existing evidence is provided for equity markets, which are organized differently from bond markets.3 For example, in bond markets the prevailing type of investor is institutions, such as insurance companies or pensions funds. Moreover, the market structure is more opaque than for equities, particularly for voice-trading in over-the-counter (OTC) markets. Further, pan-European electronic platforms market it possible to trade the bonds of many issuers in a single venue; however, pre- and post-trade arrangement are far from perfectly harmonized. In the near future, the supply side of bond markets will likely be subject to noticeable changes. The regulation under MiFID II is targeted to increase transparency in this market, which favors the trend of increased trading on electronic venues. In this context, 1 2 3

See Krishnamurthy and Vissing-Jorgensen (2011) and D’Amico and King (2013). The official press release refers to liquidity and depth in government bond markets. https://www.ecb.europa.eu/press/pr/date/2010/html/pr100510.en.html A growing number of studies provide evidence on US corporate bond markets. See Schestag et al. (2016) for a study on measuring bond liquidity. For studies on the pricing of bond liquidity see Bao et al. (2011) and Acharya et al. (2013).

Chapter 1. Introduction

17

new agents with automated trading strategies may become more important with the chance of lower transaction costs for investors. This thesis aims to provide a better understanding of European bond markets by contributing to the related issues of market liquidity and central bank interventions.

1.2

Research Outline

In the first part of the empirical analysis, I investigate the supply of liquidity in bond markets. For institutional investors with high trading needs, a decline in liquidity supply may result in large costs of trade execution. Moreover, evaporating liquidity supply can cause extreme price movements or market glitches. For instance, on October 15, 2014 10-year US treasury bonds experienced strong intraday price movements, which were accompanied by deteriorated liquidity conditions (Joint Sta↵ Report, 2015). Previous empirical research mostly investigates liquidity supply in equity markets (Hendershott and Seasholes, 2014; Nagel, 2012). For treasury bonds, Green (2004) finds negative returns from liquidity supply around news. In European bond markets, primary dealers largely contribute to the liquidity in secondary markets. Recently, anecdotal evidence indicates that primary dealers have stopped their liquidity supplying service in some bond markets due to increasing costs and regulatory requirements.4 I contribute to the discussion with the following research question: Research Question 1: What are the returns from supplying liquidity in European bond markets? To investigate the issue, I construct a simple short-term contrarian trading strategy to proxy the trading behavior of liquidity suppliers. Trading in the opposite direction of the current price movement is profitable if prices are negatively correlated over time. If fewer liquidity suppliers are in the market, prices will deviate further from the fundamental value, and the return reversal strategy becomes more profitable. I empirically investigate whether short-term reversals are associated with illiquidity in bond 4

At the beginning of 2016 several primary dealers are reported to end their activities in European markets. For instance, see Reuters on January 21, 2016:“Squeeze bank dealers quit European government bond markets.”.

18

1.2. Research Outline

markets. To assess this argument, cross-sectional and time series tests are conducted. Further, I study weather returns from liquidity supply are higher during financial turmoil and funding constraints. Next, the influence of unconventional policies by the ECB is investigated. Previous evidence confirms that monetary policy is related to the liquidity and the pricing premium of illiquidity in equity markets (Jensen and Moorman, 2010; Goyenko and Ukhov, 2009). These findings indicate that an expansive monetary policy eases the financing conditions of financial intermediaries, which results in an increase in market liquidity. The previous findings in the literature suggest that unconventional interventions with an expansive e↵ect on central bank money have a broad e↵ect on bond markets by supporting the supply side. On the other hand, purchasing assets directly without addressing the funding conditions may have a specific e↵ect on the targeted market. Changes in the demands for liquidity, such as temporary selling pressure or correlated trading, may induce the central bank to intervene via asset purchases and de facto become a market maker. The following research question contributes to this issue:

Research Question 2: Do unconventional policy interventions a↵ect liquidity in bond markets? I investigate bond liquidity in a sample of high-frequency data of European intra-dealer markets. The data set makes it possible to calculate precise estimates of alternative liquidity measures in bond markets with di↵erent degrees of credit risk and liquidity. I analyze liquidity in bond markets by linear time series regressions and test for structural stability. Shifts in the parameters can provide indications of a change in liquidity conditions. For instance, Pelizzon et al. (2016) investigate the Italian market and find a structural break in the association of liquidity and credit risk. In addition, I analyze the interdependence of liquidity between markets and follow the approach of Chordia et al. (2005). These authors estimate a vector autoregressive system (VAR) to analyze liquidity linkages between bond and equity markets.

Chapter 1. Introduction

1.3

19

Structure of the Thesis

The remainder of the thesis is structured as follows. Chapter 2 starts with an overview of the related literature on market liquidity, its measurement and pricing. Further, the mechanisms of a liquidity crisis and the interventions of the central banks are discussed. Next, an overview of European government bond markets and a review of the current literature and developments are provided in Chapter 3. Chapter 4 investigates short-term reversals in governments bonds.5 First, I describe the relevant literature, data set and portfolio formation strategy. Then, the portfolio returns are investigated with a particular focus on their association to market liquidity. Further, I analyze return reversals in periods of financial turmoil. For robustness, evidence on an alternative data set is presented. Chapter 5 analyzes the determinants of bond liquidity and the impact unconventional interventions by the ECB.6 The literature discussion allows establishing testable hypotheses. Following the description of the data and variables, I investigate bond liquidity in di↵erent markets. In the second part, the interdependencies across markets are analyzed. For reasons of robustness, maturity e↵ects and order book depth are investigated. Chapter 6 summarizes the main evidence and suggests promising areas of future research.

5

6

This chapter is based on the working paper “Short-Term Reversals in Sovereign Bond” with Hans-Peter Burghof. A previous version of the paper was presented at the 14th INFINITI Conference on International Finance (Dublin, 2016) and the 33rd French Finance Association Conference (Liege, 2016). This chapter is based on the working paper “Unconventional Monetary Policy and Time-Varying Liquidity in Government Bond Markets” with Hans-Peter Burghof. A previous version of the paper was presented at the 65th Annual Meeting of the Midwest Finance Association (Atlanta, 2016).

21

Chapter 2

Market Liquidity “Liquidity [. . . ] is very much a fair weather friend: it is there when you don’t need it, absent when you urgently need it.”(Buiter, 2008).

2.1 2.1.1

Market Liquidity and Securities Prices Trading and Liquidity

The recent financial crisis is a reminder that markets are not without frictions. This observation contrasts with standard financial theory, which assumes that investors can trade without transaction costs and that the prices of securities with identical cash flows should be equal under the law of one price. Garleanu and Pedersen (2011) document that the prices of corporate bonds and their respective credit default swaps (CDS), which are both subject to the same economic risk, are not perfectly equal. The conflict with the law of one price was particular pronounced during the crisis period of 2008-2009. Another example is the failure of the covered interest rate parity, which states that interest rates for the same lender or borrower should be identical across countries if currency risk is eliminated (Co↵ey et al., 2009; Pinnington et al., 2016). Analyzing the trading

22

2.1. Market Liquidity and Securities Prices

behavior of individuals within the structure of securities markets appears to be the starting point for understanding markets frictions. Thus, literature on the broad field of market microstructure incorporates a large body of theoretical and empirical research that is concerned with illiquidity of financial markets and the related issue of price discovery. There exist di↵erent strands of research that are grounded on two intuitive findings: Market participants do not have the same set of information and not every market participant can simultaneously be present in the market. Asymmetric information The first strand of research stresses information asymmetries in financial markets. Bagehot (1971) describes traders with new information that is not reflected in current prices, liquidity motivated traders and traders who falsely believe to be in the first category. The market maker sets the bid-ask spread to account for losses when he trades with an informed trader and gains from transactions with the other types. Glosten and Milgrom (1985) formalize the adverse selection problem and find that higher information asymmetry results in larger bid-asks spreads. Dealers post the bid and ask price, which reflect their valuation of a security contingent on the type of order that they receive. This price setting behavior reflects the idea that the order flow is informative, given that the presence of informed traders changes the likelihood of receiving a buy or a sell order. For instance, a large number of buy orders is a signal for a high valuation, as information traders and liquidity-motivated traders want to buy, resulting in a larger spread between bid and ask prices. A source of the information advantage of some trades may be their ability to transform public information into a private valuation of a company. Kim and Verrecchia (1994) show that a group of market participants can achieve an information advantage by processing the information following the release of earnings news. Thus, the arrival of new public information increases information asymmetries in the market, which results in lower market liquidity. Krinsky and Lee (1996) provide empirical evidence of the influence of quarterly earnings announcements on the information asymmetry in equity markets. The authors document an increase in the adverse selection component of bid-ask spreads prior to and after the release of earnings news. The increase in adverse selection is consistent with the explana-

Chapter 2. Market Liquidity

23

tion that informed investors have the ability to process public information into private valuations. Green (2004) investigates the influence of macroeconomic news on treasury bonds and confirms that adverse selection increases around news releases. Riordan et al. (2013) classify news by sentiment and find di↵erences in the information content on the Toronto Stock Exchange. Negative news appears to be more informative, resulting in worsened liquidity and higher adverse selection costs upon arrival. Risk aversion An alternative line of research discusses the inventory risk of liquidity suppliers and the consequence for liquidity and price movements. Ho and Stoll (1981) show that a risk-averse dealer sets bid and ask prices to regulate the incoming order flow for maintaining his optimal inventory position. The risk aversion of the dealer and the volatility of returns determine bid and ask prices. Grossman and Miller (1988) suggest a more general model in which the demand to execute immediate orders is met by suppliers of immediacy, who o↵er market-making services. Price risk influences the decision of investors to wait another period for trade execution or to trade immediately and transfer the risk to the liquidity provider. The compensation for this service is reflected in the price concession for immediate execution. Consequently, the prices between two periods are negatively correlated. On an aggregate level, a low number of liquidity providers will result in lower market liquidity and stronger negative serial correlation of prices. Recent studies have investigated the empirical implication of risk-averse liquidity suppliers with data on specialist inventory positions and market makers on the New York Stock Exchange (NYSE). Hendershott and Seasholes (2014) show that future stock returns can be predicted with inventory positions. The authors show that intermediaries trade in the opposite direction of contemporaneous returns, and future price reversal is associated with the mean reversion of liquidity providers’ inventory. Hendershott and Menkveld (2014) document temporary deviations from the efficient price, which are charged by risk-averse intermediaries following a transaction. The authors use a state space model to disentangle the permanent price impact due to information asymmetries from transitory price pressure and find the latter has an average size of less than 50 basis points (bps) with an average half-life of 0.92 days. Specialist inventory positions

24

2.1. Market Liquidity and Securities Prices

are positively correlated with future price changes, and the half-lives of the end-of-day inventory ranges from 0.54 days for large stocks to 2.11 days for small stocks. Chapter 4 provides a more detailed discussion on the returns from liquidity supply in government bond markets. Search costs Duffie et al. (2005) show that transaction prices are influenced by the bargaining power of an investor and his ability to search for alternative trading opportunities. The authors consider a market structure similar to OTC markets, in which transactions are negotiated bilaterally. In contrast to previous explanations, no information asymmetries or risk aversion of market makers are assumed. If search frictions are low, then bargaining power of market makers is less important, which results in lower bid-ask spreads. In support of this argument, the empirical evidence on transaction costs in dealer markets finds that large trades by institutional investors are subject to smaller transaction costs than small-sized trades (Harris and Piwowar, 2006; Schultz, 2001). In the market for US corporate bonds, the introduction of TRACE, a system for the mandatory price reporting of OTC transactions, allows studying the e↵ect of increased transparency on transaction costs. Researchers document a subsequent decline in transaction costs of corporate bonds by approximately 5 bps (Edwards et al., 2007). Further, Goldstein et al. (2007) show that the e↵ect of higher transparency due to the publication of trade data on bond liquidity decreases with the size of a transaction. The authors’ conservative estimate of the decline in transaction costs for small trades is 22 bps for a sample of BBB rated bonds. Thus, the literature supports the notion that higher transparency increases the bargain power of investors in opaque markets.

2.1.2

Measuring Liquidity

Demsetz (1968) was one of the first authors to link market liquidity to the spread between bid and ask prices. A simple model explains that suppliers of liquidity are compensated for their service of o↵ering immediate trade execution. There are two equilibrium prices on financial markets: the bid price for immediately selling a stock and the ask price for which a security can be immediately purchased. Thus, the nominal di↵erence indicates the price for immediacy. A larger share of research follows this idea and

Chapter 2. Market Liquidity

25

takes the transaction cost approach for measuring (il-)liquidity in financial markets. However, the concept of liquidity has multiple dimensions, such as the quantity available for selling or purchasing and the time needed for completing a transaction. I summarize di↵erent attempts to measure liquidity below. The potential costs for a small transaction are measured by the bidask spread, which is arguably the most common liquidity proxy used in the literature.1 Similarly, the e↵ective spread indicates the costs of actual transactions. The di↵erence between a transaction price and the mid-price, which is the average of the quoted bid and ask price, indicates the e↵ective costs of a transaction. Glosten (1987) decomposes the e↵ective spread into an adverse selection component and the profit earned by liquidity suppliers. The first component is the price impact, which is measured by the change in mid-prices at the time of execution and after a short time lag. The realized spread, which is the di↵erence between the e↵ective spread and price impact, is interpreted as the profit from a hypothetical liquidity supplier, who reverses the trade later in the day. A di↵erent dimension is the volume available for buying or selling, which can be measured by the depth of the order book in limit order markets. Depth can be measured as the average between quoted quantities at di↵erent levels of the bid and ask side of the order book. Measuring liquidity with quote and trade data may not be possible, if a market is not as transparent as modern electronic limit order markets. Moreover, estimating liquidity on high-frequency observations may be too expensive or the data may not be available for long-term observation periods. Alternatively, there is a larger amount of research on estimating liquidity proxies on low-frequency observations. For instance, Roll (1984) suggests a spread estimator by taking the covariance of subsequent price changes, which reflects the price movement within the bid-ask spread (bidask bounce). Amihud (2002) suggests the simple ratio of the absolute return to the aggregate trading volume, which is positively associated with microstructure estimates of the price impact and fixed trading costs. An approach for quantifying the performance of low-frequency measures is pro1

For other early studies, see Amihud and Mendelson (1980), Tinic (1972). For additional examples, see Amihud et al. (2013).

26

2.1. Market Liquidity and Securities Prices

vided by Goyenko and Ukhov (2009). These authors compare commonly used liquidity measures in equity markets to estimates of transaction costs on high-frequency microstructure data. The findings indicate that the measure of Amihud (2002) shows a significant correlation with the price impact. More recently, Corwin and Schultz (2012) calculate a spread estimator on daily high and low prices. The authors decompose the di↵erence between high and low prices into price volatility and the bid-ask spread. Their spread measure appears to show higher correlation with the underlying bid-ask spread than alternative low-frequency spread measures, such as the Roll (1984) estimator. Fewer studies are concerned with measuring liquidity in bond markets and account for di↵erences in the market structure of fixed income securities. Fleming (2003) finds that the bid-ask spread is an accurate measure of liquidity in US treasury markets, whereas the volume and frequency of transactions are poor proxies of transaction costs. Further, Bao et al. (2011) suggest that the co-variance of the transaction price return captures the illiquidity of US corporate bonds. Jankowitsch et al. (2011) take a di↵erent approach, measuring bond liquidity by the dispersions of transaction prices, which are caused due to search friction and inventory costs in OTC markets. The authors implement the price dispersion measure on bond data of US corporate bond markets and show that the proxy is strongly correlated with standard liquidity measures. A recent study by Schestag et al. (2016) investigates the performance of a larger number of low-frequency liquidity proxies in US corporate bonds. The authors find that measures on price data show reasonable correlation with transaction costs, whereas the high-low spread of Corwin and Schultz (2012) is among the best performing proxies according to time series and cross-sectional tests. Measuring liquidity is an ongoing task in the literature. New developments in trading, such as the emergence of high-frequency traders in equity markets and the increase in institutional ownership requires reassessing the current understanding and methods of quantifying market liquidity (Barardehi et al., 2015). One should be aware that any conclusions on the static liquidity measures discussed above are very specific with regard to the market structure or sampling frequency (Schestag et al., 2016; Goyenko

Chapter 2. Market Liquidity

27

and Ukhov, 2009). An emerging strand of research is likely to be concerned with a dynamic interpretation of liquidity, such as its resilience. The resilience of liquidity can be connected to the quotation of Buiter (2008) at the very beginning of this chapter. Similar to an opportunistic friend, liquidity may swiftly disappear under unfortunate conditions. Examples are the increasing number of flash crashes or market glitches in bond and equity markets, which combine sharp price movement with a surge in illiquidity. Currently there is no consensus in the literature on measuring the resilience.2 Thus, counting the number of outliers of price data may provide an initial indication of the issue.

2.1.3

Pricing and Liquidity Risk

Why should investors care about market liquidity? To answer this question, I summarize several established findings in the financial literature. The seminal paper by Amihud and Mendelson (1986) presents the link between the liquidity of an asset and its market price. Trading costs lower the returns from investing, and investors require compensation with a return premium. The premium lowers the current price, and illiquid securities will trade at a discount. A second finding is that long-term investors, who trade less frequently, hold assets with higher illiquidity because they are less sensitive to trading costs. The clientele e↵ect explains that the return of securities increases concavely with illiquidity. The change in the trading regime on a stock exchange comes close to a natural experiment on the influence of liquidity on asset prices. Jain (2005) finds that a change to a modern trading regime increases the liquidity of transferred stocks and results in higher prices and a lower equity risk premium. Because the event has no relevance for the securities’ cash flows, the price change may be attributed to the increase in liquidity. In US treasury markets empirical evidence supports the link between liquidity and asset prices. Amihud and Mendelson (1991) find that on-the-run bills, which are more frequently traded and consequently have higher liquidity, trade at lower yields than maturity matched treasury notes with identical cash flows and credit risk. 2

For an approach to quantifying liquidity resilience, see Alan et al. (2015).

28

2.1. Market Liquidity and Securities Prices

A recent study by Darbha and Dufour (2015) provide evidence that liquidity explains the cross-sectional yield di↵erences in a sample of European sovereign bonds. Further, a body of research focuses on the implication of shifts in aggregate market liquidity on securities prices. This strand of research is motivated by the observation that aggregate market liquidity changes over time. The global financial crisis stands as a recent example of this timevarying behavior. In addition, Chordia et al. (2000) provide empirical evidence of the co-movement (commonality) of individual stock liquidity with aggregate market liquidity. The result remains robust after controlling for the influence of microstructure variables on a stock level. Against this backdrop, Amihud (2002) shows that unexpected shocks in aggregate illiquidity result in lower stock prices. The author contributes this result to the hypothesis that an unexpected change provides information on the future state of market liquidity, which influences the return required by investors. The association with stock prices appears to be the strongest for the least liquid stock with small market capitalization. Liquidity risk can be understood as the sensitivity of individual asset return to shocks in aggregate market liquidity. Acharya and Pedersen (2005) develop the liquidity adjusted CAPM to model the pricing of systematic market and liquidity risk. An investor prices the gross return risk and transaction costs risk, which a↵ects the net return of his investment. The authors find that the return premium for liquidity risk is 1.1% per year, in addition to 3.5% for the level of liquidity. Thus, investors demand economically significant compensation for holding illiquid stocks with high liquidity risk in their portfolios. Pastor and Stambaugh (2003) confirm that stocks with a greater exposure to liquidity shocks have higher expected returns. The results show that liquidity risk in equity markets is not explained by the common Fama and French (1993) risk factors. Complementing the evidence on US stocks, global liquidity risk is priced in equity markets across developed countries (Liang and Wei, 2012; Lee, 2011). Further, evidence exists on the return premium for liquidity risk in fixed income markets. For US corporate bonds Lin et al. (2011) find an annual return di↵erence between bonds with high and low liquidity risk of 4%, and Alquist (2010)

Chapter 2. Market Liquidity

29

estimates a liquidity premium of 2.8 % per year on an historical sample of sovereign bonds. In summary, this section has discussed the current literature on the economics behind illiquidity in financial markets, the issue of measuring and pricing liquidity. A comprehensive literature review on the pricing of liquidity is provided by Amihud et al. (2013). The authors draw several implications for investors, issuers of securities, policy makers and regulators. For instance, the authors take a clear stance against the taxation of financial transactions, which is currently under debate in many European countries. The negative e↵ect of transaction costs on asset prices suggests that the introduction of a transaction tax increases the costs of capital and hampers corporate investments. A detailed review of research on professional and retail investors is provided in the following section.

2.2

Market Participants

2.2.1

Institutional Investors

The goal of this section is to highlight the relevance of market liquidity from the perspective of institutional investors. First, I discuss decision making on the level of individual professional investors. The established link between liquidity and asset prices raises the question of whether professional investors are able to predict future changes in aggregate liquidity. Next, market participants in market microstructure models are often divided in takers and suppliers of liquidity (see Section 2.1.1). A body of research investigates whether institutional investors act as liquidity suppliers on an aggregate level. Do professional investors incorporate market liquidity into their decision making process? A properly adjusted portfolio structure is a major advantage in states of low market liquidity, because fund performance is linked to capital in and outflows. A recent strand of literature on institutional investing provides empirical evidence of the ability to time market liquidity. Cao et al. (2013) document that equity-orientated hedge funds indeed have liquidity-timing skills. The authors propose a timing model on

30

2.2. Market Participants

an individual fund level, which allows the market beta to change upon the expected liquidity state in the next period. Further, ranking funds by their liquidity-timing ability indicates that the highest timing ability results in significant outperformance in an out-of-sample period from 3 to 12 month. Additional research on debt-oriented hedge funds supports that managers have the ability to time liquidity (Li et al., 2016). A recent extension to the timing literature by Bazgour et al. (2016) investigates whether the timing skills of fund managers extend to well-known style factors, such as momentum, size and growth. The authors find that equity funds adjust the exposure of their portfolios to the size factor in accordance to future market liquidity. This behavior can be explained by the lower liquidity of small stocks. When aggregate market liquidity improves, less liquid stocks will experience a relative outperformance. Accordingly, it appears likely that funds adjust not only their cash levels, but also the exposure of their equity investments to liquidity risk. A related question is whether professional investors demand or supply liquidity in financial markets. Evidence on the ability of fund manager to protect their holdings from evaporating liquidity, may give an indication that professional investors are not present in the market under severe condition. Thus, liquidity timing may exceed a negative externality on the entire market. A topical strand of research investigates if institutional investors are suppliers of liquidity or show an opportunistic trading behavior. Kruttli et al. (2015) find that hedge funds supply liquidity in financial markets. The authors use return autocorrelation as proxy for illiquidity of funds’ portfolios, which follows the idea of Getmansky et al. (2004). As hedge funds with illiquid securities may extrapolate prices if no current information exists, less liquid portfolios will show a positive return autocorrelation. The authors provide empirical evidence that aggregate hedge funds liquidity predicts a large number of asset returns, such as international equities, corporate bonds and currencies. Jylh¨a et al. (2014) take an alternative approach, analyzing the correlation of fund level returns with a liquidity supplying strategy in equity markets. On average more funds appear to be liquidity suppliers than takers. However, the authors observe a shift during stress periods as the share of funds which are liquidity demanders strongly increases. Funds face certain constrain in their ability

Chapter 2. Market Liquidity

31

to exploit trading opportunities, such as investors withdrawals, which are typically mitigated by lock-in periods. Aragon et al. (2014) document that funds with high lock-in requirements are indeed suppliers of liquidity in financial markets.

2.2.2

Individual Investors

In contrast to professional investors, it does not seem likely that individual investors have sophisticated timing skills. Much in contrast, the discussion on retail investing is mainly concerned with behavioral biases (Odean, 1998; Barber and Odean, 2000). Nevertheless, a well-established finding is that individual investors are contrarian investors that trade in the opposite direction of price movements (Goetzmann and Massa, 2002). Similar to the liquidity providers that are described in the microstructure literature (Section 2.1), this observation could indicate that individual investors accommodate the demand for liquidity by institutional investors. Kaniel et al. (2008) find supporting evidence on this argument from individual investors trading on the NYSE. The aggregate net position of individual investors over one week is able to predict future stock returns. The predictive ability is robust when the past returns and trading volume are included, which supports the robustness of the finding that retail investors provide immediacy. Further, Kaniel et al. (2012) and Chen et al. (2014) confirm that individual investors supply liquidity before earnings announcements. The evidence on US and Taiwanese equity markets shows that the trading imbalances of individual investors successfully predict future returns. Thus, the literature shows that liquidity provision may be equally performed by retail investors. This conclusion adds new evidence to the empirical evidence presented in the previous section and theoretical models in market microstructure literature, which focus on professional investors and designated market makers as providers of immediacy. In contrast to the perception of retail investors as noise traders, Kaniel et al. (2012) show that aggregate information on individual investors trading around earnings announcement news is informative and generates positive abnormal returns upon the news’ release. Intuitively, the results reveal that individual investor trading is particularly informative for small stocks,

32

2.3. Liquidity Crises

which are less well covered by analysts and financial media. A recent study by Wang and Zhang (2015) confirms that individual traders add to price discovery and liquidity in US stock markets. The authors contribute the positive association between stock liquidity and individual investors trading with a decline in the information asymmetry in these stocks. Additional evidence is provided by Fong et al. (2014), who show that the endogenous choice of the trading broker can identify investor sophistication. Covering a sample period of 12 years, the authors find a self-selection bias in the choice of a retail broker. Trades via low-cost brokers are less informative and less successful in generating trading profits compared to trades via high-cost brokers with a wider range of services.

2.3 2.3.1

Liquidity Crises Time-Varying Liquidity

The significance of market liquidity may be easily overlooked during normal times but becomes even more apparent during crisis episodes. In a crisis event, aggregate market liquidity declines sharply, leading to an increase in liquidity risk and a corresponding decline in asset prices. Liquidity crises are often combined with financial crises but may also occur due to single shocks in the financial system with far reaching e↵ects. The global financial crisis surrounding the bankruptcy of Lehman brother is an example of the link between financial stress and market liquidity. Nagel (2012) finds that the supply of liquidity in US equites drastically declined during this period. Further, Pedersen (2009) document the association of liquidity risk and securities prices at the start of the crisis. This author finds that the trading pressure of proprietary traders or hedge funds that employed quantitative trading strategies led to a temporary but economically significant price drop in August 2007. The mechanisms of a liquidity crisis are analyzed by Brunnermeier and Pedersen (2009). A key finding is the interdependency between funding ability and market liquidity. The authors formalize the underlying dynamics of a crisis by self-enforcing liquidity spirals, which are triggered by

Chapter 2. Market Liquidity

33

funding shocks. Funding constraint financial intermediaries can provide less liquidity in financial markets, because they are less willing to accumulate larger trading positions. Lower market liquidity leads to lower prices and higher volatility, which further decreases the funding ability of intermediaries, for instance, via higher margin requirements. The spiral e↵ects also predict that the margins of assets with less liquidity risk should increase only modestly compared to high risk assets. Accordingly, the link between market and funding liquidity also predicts a high preference of investors for liquid assets with low liquidity risk, which is also referred to as flight-to-liquidity. Empirical evidence in favor of the link from funding liquidity to market liquidity is presented by Comerton-Forde et al. (2010). The inventory level and revenues of individual specialists on the NYSE explain the future levels of market liquidity. The authors note that these findings should not be unique to the NYSE, because liquidity suppliers in alternative market environments are likely to face funding constraints as well. An additional example is the bankruptcy of Lehman Brothers, which resulted in an exogenous shock to the funding liquidity of the bank’s clients (Aragon and Strahan, 2012). The counterparty risk that funds face when making transactions with the prime broker, such as Lehman Brothers, is even exacerbated by the then common practice of relending stocks (rehypothecation). Following the bankruptcy of Lehman Brothers the market liquidity of equities that were held by its clients deteriorated significantly. Boudt et al. (2013) investigate the reverse direction of causality from market liquidity to funding liquidity. The endogeneity between both types of liquidity is solved via an instrument variable estimation. For high levels of financial stress, market liquidity has a negative impact on funding conditions, which supports the idea of self-enforcing liquidity spirals. Episodes of flight-to-liquidity also find great support in the empirical literature. In a study on European bond markets, Beber et al. (2009) document that investors care about liquidity and credit risk, whereas during stress periods, the role of liquidity becomes more pronounced. Further, the decisions of investors to move assets in or out of the bond market are significantly explained by liquidity and not by credit risk. Goyenko and Sarkissian (2010) measure episodes of flight-to-liquidity by the liquidity of

34

2.3. Liquidity Crises

short-term US treasury debt, give that treasuries have a safe-haven status due to their low credit risk and illiquidity.3 The results show that the bond liquidity predicts liquidity and prices of global equity markets. A di↵erent approach to measuring investors’ preference for assets with low liquidity risk is to investigate the spread of assets with di↵erent liquidity but identical cash flows and credit risk. Recent studies utilize the spread between bonds issued by the Kreditanstalt f¨ur Wiederaufbau (KfW), a government guaranteed agency, and German government bonds. Santis (2014) suggests that the KfW spread is an indicator for flights in European markets. Further, Schwarz (2016) measure liquidity by the KfW spread to explain European sovereign spreads and money market rates. The results show that the spread has higher explanatory power than credit risk in both markets. The empirical literature on the co-movement of liquidity of individual securities provides an alternative explanation compared to the supply-side view proposed by the funding liquidity channel. To explain liquidity commonality, Karolyi et al. (2012) analyze the developed and emerging markets of 40 countries over a 15-year period. The authors find that commonality is also influenced by factors that relate to the demand for liquidity, such as correlated trading decisions, capital inflows and investors sentiment. Moreover, the results show that, for the financial markets of developed and developing countries, demand-side factors are more pronounced than supply-side explanations for the variation in commonality over time. Koch et al. (2016) find that the stock ownership of mutual funds exceeds a strong influence on liquidity co-movement. For instance, funds that face withdrawals from investors are required to sell stocks simultaneously, causing a stronger commonality in stocks with a larger ownership of mutual funds. In summary, the interaction of funding and market liquidity finds large support in the theoretical and empirical literature during the recent financial crisis. Nevertheless, the demand for liquidity by institutional investors, which is driven by sentiment and forced selling, results in high co-movement in liquidity. In order to gain a better understating of liquid3

The authors note that this approach does not market it possible to di↵erentiate between flight-to-quality and flight-to-liquidity, which is discussed by Beber et al. (2009).

Chapter 2. Market Liquidity

35

ity crises and the variation in liquidity over time, more research is required to explore both supply and demand explanations. Although most empirical evidence is presented on equity markets, investigating bond markets and other market structures may provide additional insights into this debate. The following sections of this chapter are concerned with securities prices under a lack of arbitrate capital and address the debate of central bank interventions in a liquidity crisis.

2.3.2

Slow-Moving Capital

In this section, I discuss the implication of a lack of arbitrage capital for financial markets and securities prices. If new capital is slow to arrive to accommodate sudden demand or supply shocks, then the resulting illiquidity will lead to a strong price deviation from the fundamental value. The subsequent price reversal may be delayed until new capital arrives and arbitrageurs seek to exploit the profitable trading opportunity. In the following, I highlight the clientele e↵ects and temporary capital constraints as potential explanations for slow-moving capital.4 The preferred habitat view suggests that the base of potential investors is segmented across securities with di↵erent characteristics. For instance, a group of investors may prefer to hold securities with payo↵s that best suite their investment needs. In addition investors may face investment constraints due to regulations. An example of clientele e↵ects in treasury bonds is provided by Greenwood and Vayanos (2010). The authors identify a high demand of pension funds for long-run UK treasury bonds in exchange for equities following the Pension Act in 2004. As a result of this demand shock the yields of some inflation-linked bonds dropped to extremely low levels of approximately 0.5% in 2006 compared to the historical average of about 3%. Further evidence of clientele e↵ects in US corporate bond markets is presented by Da and Gao (2009). The authors find that on average bond prices drop by approximately 200 bps following a rating downgrade to junk status. Over the course of six month after the downgrade the price recovers and the initial price drop is reduced by half. 4

In addition, Duffie (2010) describes the inattention of investors and formalizes the impact on asset price dynamics.

36

2.3. Liquidity Crises

The authors explain the price pressure and slow recovery by the selling pressure of insurance companies or money market funds, which are subject to a regulatory imposed cap on junk bonds of 20% and 5%, respectively. The persistent price pressure in equity markets is observable for stocks that are removed from an index. Duffie (2010) notes that index tracking funds are likely to induce high selling pressure on delete stocks, which may results in price reversals if other investors are slowly arriving in the markets. Researchers find slow price reversals of about 60 days following index deletions (Chen et al., 2004). Pedersen et al. (2007) note that professional investors face investment constraints in down markets due to capital outflows. The capital constraints of professional investors are particularly pronounced around redemption dates. The convertible bond market experienced high selling pressure in 2005 due to convertible arbitrage hedge funds, which faced withdrawals by institutional investors due to poor performance in the previous year. The authors calculate the theoretical value of convertible bonds based on the underlying stock price, the terms structure and estimates of volatility and credit risk to find that bonds traded at a discount of 2.5% in mid-2005. At the end of 2005, diversified hedge funds exploited the profitable arbitrage opportunity, resulting in the price discovery recovery of convertible bonds. Additional examples of temporary capital constraints of investors are discussed by Duffie (2010). For instance, the increase in the insurance premium of re-insurance companies following catastrophic events takes months or even years to recover to the normal level.

2.3.3

Central Bank Interventions

This section discusses the role of central banks or policy makers in a liquidity crisis. Functioning financial markets play a vital part in an economy, as they provide financing for the real sector and price information. Central bank interventions should be directed at preventing economic costs that may arise due to a temporary distortion in the funding ability or pricing mechanism as outlined in the two previous sections. An additional argument in favor of interventions is the decline in asset prices, which is observable in previous crises. Historical episodes, such as the stock market

Chapter 2. Market Liquidity

37

crash in 1987, document the association between financial crises and liquidity crises (Amihud et al., 1990). The coincidence of financial and liquidity crises raises the possibility of high economic costs. This argument finds support in macroeconomic research that outlines the influence of financial shocks on real economic activity (Bernanke et al., 2007). The decline in asset prices deteriorates the balance sheet of financial intermediaries and potential borrowers, which increase the costs of credit and hampers economic growth (Bernanke and Gertler, 1995; Bernanke and Blinder, 1992). Stylized facts The stock market crash in 1987 is an example of the efforts of the FED to secure financial stability. Under the former Chairman Greenspan, the FED provided emergency liquidity via its discount window and open market operations (Garcia, 1989). In addition, banks were motivated to extend lending to Wall Street firms. The e↵ect of these e↵orts is observable by the boost in excess reverses held by financial institutions in the reserve periods after the crash. In addition, the FED decreased policy rates by about 80 basis points (Neely et al., 2004). The Long Term Capital Management (LTCM) crisis is another example that shows how the FED intervened to safeguard financial stability. The investment strategy of LTCM, a systemically important hedge funds, was to profit from convergence in price di↵erences in related securities, such as convertible bonds or European government debt. The convergence arbitrage created strong exposure to liquidity risk, which was intensified by extreme leverage ratios of 25:1 (Jorion, 2000). Following the announcement by Russia to restructure its debt in 1998, the surge in volatility and liquidity risk caused severe losses in equity capital. The FED organized a bailout of the fund’s private creditors and public money was not involved. However, the FED decreased rates three times by 25 bps to ease market stress and support financial stability (Sauer, 2007). During the recent global crisis in 2007-2009 or the European debt crisis in 2010-2011 central banks showed a response similar to those in the past. The significance of the crisis response is reflected in the balance sheet of major central banks. From 2007 to 2012, the total asset to GDP ratio of the ECB, FED and BoE increased from little more than 5% to approximately 20% (Pisani-Ferry and Wol↵, 2012). Thus, the sheer size of the balance sheet expansion indicates their e↵orts to sustain financial stability

38

2.3. Liquidity Crises

by means of unprecedented measures. A di↵erence from previous crises is that central banks operate close to the zero lower bound and need to rely on unconventional operations. The goals of the quantitative easing programs by the FED and BoE are to support asset prices, in order to revive private credit markets. On the other hand, the asset purchases of the ECB under the SMP and the OMT program are directed to sustain the monetary transmission mechanism by means of supporting liquidity in the European bond markets (De Grauwe, 2013). More recently, the ECB has introduced quantitative easing under the asset purchase program (APP), which involves purchases of private and public debt securities (Demertzis and Wol↵, 2016). In addition, unconventional policies also include an extension of lending facilities or a widening of the collateral base. These measures are directed to improve the financing conditions of the financial sector. Yields and liquidity Since the global financial crisis, a body of research that investigates the e↵ectiveness of unconventional policies has emerged. To address this goal, most researchers are concerned with the impact on the yield curve and other asset prices. Krishnamurthy and Vissing-Jorgensen (2011) apply an event study approach to investigate the influence of the FED’s first and second quantitative easing programs on market rates. Their results show that both programs are e↵ective in reducing rates on US treasuries, mortgage backed securities (MBS) and corporate bonds. Fratzscher et al. (2015) document that unconventional policies in the US have a far-reaching e↵ect on international capital flows and asset prices. For instance, following the second QE program, investors rebalanced their portfolios towards larger holdings of risky assets, which resulted in capital inflows in emerging market equities. Rogers et al. (2014) contribute to this literature and show that the policy announcements of the four major central banks (FED, BoE, ECB, BoJ) influence yields and other securities prices. The early evidence on the FED interventions motivated studies to investigate the bond price e↵ect of the ECB’s unconventional policies. Evidence on the yield e↵ect of the securities market program (SMP) is provided by Ghysels et al. (2014), De Pooter et al. (2015) and Eser and Schwaab (2013). Additional evidence supports an e↵ect of the OMT program (Altavilla et al., 2014).

Chapter 2. Market Liquidity

39

Fewer studies have addressed the impact of the SMP/OMT on frictions in bond markets. The official announcement of the SMP stated that it aims to “ensure depth and liquidity in those market segments which are dysfunctional” (ECB, May 2010). The OMT replaced the SMP, and details were provided on September 6, 2012. De Pooter et al. (2015) find that the impact of weekly purchases under the SMP program reduce the illiquidity premium of purchased bonds. The authors measure the illiquidity premium by taking the spread between bond yields and the respective CDS spread, which holds under the assumption that CDS liquidity is constant over the observation periods. Given that the SMP resulted in a decline in the liquidity premium, it appears reasonable to expect that the liquidity in bond markets has increased as well. Pelizzon et al. (2016) show that the association of bond liquidity, as measured by bid-ask spreads, with credit risk changed strongly around the introduction of the three-year LTRO program in Italian bond markets. The LTRO program allowed financial firms access to central bank money similar to standard facilities with an extended maturity of three years. The authors explain their results by an improvement in funding conditions and find that the non-linear association between CDS and market liquidity is eliminated. Further, the results show that, in contrast to De Pooter et al. (2015), the SMP program does not seem to alter the influence of credit risk on market liquidity. In summary, experiences from past liquidity crises and the potential economic costs provide a strong argument for interventions by central banks in a liquidity crisis. Research mostly confirms an e↵ect of unconventional policies on government bond yields and other asset prices. A di↵erent strand of research has addressed market liquidity in bond markets, which is in line with the communicated goal of the ECB to improve impaired liquidity conditions. To contribute to a better understanding of the impact of these policies, I investigate market liquidity in a large sample of European bonds during the time of the intervention (Chapter 5). The next chapter gives an overview on the characteristics and structure of European bond markets.

41

Chapter 3

The European Government Bond Market 3.1

Market Characteristics

Reinhart and Rogo↵ (2009) document a history of government lending and default of European countries that dates back to the period of 1300 to 1800. For instance, France defaulted eight times on its external debt during this episode. In today’s economy, government bond markets do not solely serve the purpose of financing sovereign expenditures; they also create benchmark yields. Benchmark securities are required for the pricing of other assets, hedging and collateralized lending.1 Thus, liquid markets with low transactions are paramount, given the importance of benchmark yields in modern finance. This argument is supported by the examples of Singapore and Hong Kong, which issued bonds in excess of their financing needs to establish active and liquid secondary markets (McCauley and Remolona, 2000). Supply The Maastricht Criteria set the theoretical upper bound for the supply of European debt securities. Nevertheless, 17 of the 28 EU member 1

For a discussion on benchmark securities see Dunne et al. (2007) or Yuan (2005).

42

3.1. Market Characteristics

states currently have debt levels above the 60% threshold. In comparison to other public debt markets, the European market has gained in economic importance over recent years and is among the three largest public debt markets in the world. In 2014, the total amount of outstanding debt issued by general governments that is denominated in EUR amounts to 7.3 trillion (tr). In comparison, debt issued by the Japanese government and the US treasury equals 6.8 tr EUR and 12.8 tr EUR, respectively. The amount of Euro-denominated debt almost doubled since 2001, when the value of outstanding securities equaled approximately 3.7 tr EUR. Part of the increase is attributed to the changing composition of Euro area countries. Moreover, this development can be explained by the growth in GDP and the appreciation of the Euro, which results in a higher valuation of the debt outstanding. Finally, 75% of public debt was financed via debt securities in 2000 which increased to 77% in 2014. Other financial instruments are loans and currencies.2 Figure 3.1 shows that the composition of general government debt securities issued by the EMU-12 countries, which have been members of the EMU right from the start, has changed markedly from 2005 to 2014. In the pre-crisis period, the majority of debt outstanding was rated AAA, with a total value amount of 2.8 tr EUR. In 2014, this amount declined to 1.7 tr EUR. In the same year, AA rated debt securities amounted to 2.9 tr EUR and 2.7 tr EUR of securities outstanding were rated BBB. The deterioration in credit rating is particularly noticeable in 2012, which is the result of the European sovereign debt crisis, which began in 2010. The share of debt securities with long- and short-term maturities stayed somewhat constant over the last 10 years. In 2014 (2005), about 0.53 tr EUR (0.4 tr EUR) in short-term securities were outstanding, and 6.8 tr EUR (4 tr EUR) were long-term securities. In particular, institutional investors with a long-term liabilities structure, such as pension funds, demand long-term debt securities. Most long-term issues are fixed-rate bonds with a share of approximately 90%.3

2 3

The government debt data are collected by Datastream. Data sources are the BIS, Eurostat and ECB. The data sources are the ECB and the Bank of International Settlement (BIS).

Chapter 3. The European Government Bond Market

43

Figure 3.1 General Government Debt Securities The figure displays the outstanding general government debt securities of the EMU-12 countries grouped by credit rating. The observation periods spans from 2005 to 2014. The source of securities data is the Bank for International Settlement (BIS). The data of securities and rating are collected by Datastream and Bloomberg, respectively.

44

3.1. Market Characteristics

Demand The majority of government bonds are held by large institutional investors, such as banks or insurance companies. Retail investors, on the other hand, typically have a minor contribution to the investor base of sovereign debt. For instance, in Germany, the share of retail investors has markedly declined and investment products for retail investors, such as the German Federal Saving Note (“Schatzbrief”), was terminated in 2013. The reasons may be the high administrative and distributional costs. Thus, it remains to be seen whether the advantages of a broader investor base go beyond the calculated short-term distributional costs (Krause-Junk, 2012). Figure 3.2 presents the share of government debt holdings of the domestic banking sector and of foreign entities; the size of each bubble gives an indication of the total amount of debt outstanding.4 The changes in the composition of debt holdings from 2005 to 2014 are rather heterogeneous across countries. In the top two graphs, countries with a small amount of total debt outstanding, such as Greece, the Netherlands, Finland and Ireland, have a share of foreign holdings above 70%. With the exception of Finland, these countries increased their share of domestic holdings and moved closer to Germany and France, with a share of approximately 60%. Additionally, Italy and Spain have decreased their share of foreign holdings to about 40%. Simultaneously, the banks’ holdings for Germany and Belgium, which were the largest in the sample in 2005, have decreased, whereas the holdings for Italy and Ireland have strongly increased. The opposing developments result in little change in the aggregate share of foreign and financial holdings. At a first glance, the developments presented in Figure 3.2 do not allow for the conclusion of an increase in home bias across European countries. Andritzky (2013) documents that the non-domestic holdings in Euro area countries of about 50% are particularly high in comparison to other developed economies. In particular, however, peripheral countries with high credit risk have decreased their share of foreign holdings. Moreover, on an aggregate European level, this share is markedly lower, given that a quarter of non-residential investors are located within the Euro area. Acharya et al. (2014) find that sovereigns and financial credit risks have become 4

The data is from the sovereign bond holding data base of Merler et al. (2012).

Chapter 3. The European Government Bond Market

2005

2008

45

46

3.1. Market Characteristics 2011

2014

Figure 3.2 Holdings of Government Debt The figure displays the sovereign bond holdings of the banking sector (y-axis) and nonresidents (x-axis). The size of a bubble indicates the total amount of debt securities outstanding. The data is provided by national authorities and compiled by Merler et al. (2012).

Chapter 3. The European Government Bond Market

47

increasingly interdependent since the global crisis, which is explained by the large holdings of sovereign debt in the banking system among other factors. Their results indicate that banks’ credit risk is explained by the sovereign credit risk of the domestic government as well as the risk of foreign sovereigns weighted by the respective cross-country exposure. A question recently addressed in the literature is whether and how the composition of the investor base is linked to bond yields. Recent studies on debt markets in developed countries find that the share of non-domestic investors is negatively correlated with bond yields (Arslanalp and Poghosyan, 2014; Andritzky, 2013). Andritzky (2013) finds that a larger share of nondomestic investors is positively correlated with price volatility because the direction of the fund flows of these investors may change more quickly. This author shows that funds from non-domestic investors and domestic institutional investors follow sovereign yields (pull e↵ect) and not vice versa. Thus, the endogeneity between investor flows and bond yields must be considered to estimate a causal relationship. Addressing these issues, Beltran et al. (2013) confirm that foreign fund flows result in lower US treasury yields. Jaramillo and Zhang (2013) document that a large share of domestic and foreign central banks holdings is associated with lower yields in developed and emerging countries. More generally, the authors find that the holdings of other types of investors other than banks appear to results in lower yields.

Taken together, the adverse feedback between sovereign debt and the European banking system may be solved temporarily by increasing the holdings of central banks. Thus, the recent asset purchase program (APP) by the ECB may have the e↵ect of relaxing this link. However, in the long run governments should benefit by increasing the appetite of foreign investors and diversify their investor base. Finally, individual and nonfinancial investors may also be willing to invest and should not be neglected by domestic debt agencies. Integration Market integration may be defined by the substitutability of securities. In perfectly integrated markets idiosyncratic factors play an insignificant role and consequently no gains from diversification may be

48

3.1. Market Characteristics

achieved by investing in di↵erent markets. Thus, European bonds di↵er markedly with respect to the credit of the issuers and are not perfect substitutes. Against this backdrop, illiquidity also hinders integration. Common findings in this field of research are that EMU members are more integrated than outsiders and the markets are far from perfectly integrated. For instance, Christiansen (2014) measures the degree of integration with yearly factor regressions on bond returns. The author finds that the markets of EMU countries are more integrated than those of non-EMU members and the level of credit risk influences the degree of integration. Abad et al. (2010) find that, for EMU markets, the Eurozone risk is of higher importance than global risk factors. The widening of yield spreads since the crisis indicates a stronger relevance of country-specific factors, such as default risk and liquidity. Accordingly, some authors conclude that markets became increasing fragmented during the crisis period (Christiansen, 2014). The underlying reasons for the higher integration of EMU members in contrast to outsiders go beyond the elimination of currency risk. Pagano and von Thadden (2004) name the elimination of market frictions as the main source of EMU bond market integration. Following the introduction of the single currency, financial intermediaries helped homogenize secondary markets. For instance, pan-European trading platforms allow to trade bonds of di↵erent issuers in a single venue and increases competition among liquidity providers. Following the introduction of electronic trading platforms, Schulz and Wol↵ (2009) find a strong increase in the integration of EMU countries in the year 2000. In addition, post-trade arrangements also influence the costs of trade execution. In particular, the large number of central securities depositaries (CSDs) indicates that the post-trade procedure is still complex and fragmented across European markets (ECB, 2007). In order to achieve a harmonization in the settlement services, cooperation among individual CSDs is facilitated under the TARGET2-Securities system, which is bears similarities to the TARGET2 payment system. In summary, the discussion reveals that bond market integration has been facilitated by the common currency and the synchronization of trading venues. In addition to the benefits from integrated markets, a drawback may be the stronger contagion of financial shocks or spillovers as a conse-

Chapter 3. The European Government Bond Market

49

quence of the higher interdependence. In the following section, I provide a more detailed discussion on the structure of European government bond markets.

3.2

Market Structure

The European bond market has developed to be one of the world’s largest fixed income markets. Since the introduction of the single currency, several e↵orts have been made to harmonize the bond issuance in the primary market and the trading in secondary markets (Pagano and von Thadden, 2004). In addition to trading bonds directly in the cash market, trading in bond futures markets contributes the price finding mechanism (Upper and Werner, 2007). The trading volume in the German bund futures markets is even higher than that in the underlying cash market. The low capital requirement at the time of the transaction, which is limited to the margin requirements, makes futures trading more convenient for hedging and speculation. Further, the market for credit derivatives also contributes to price discovery in bond markets (Palladini and Portes, 2011). Finally, sovereign debt securities are an essential part in collateralized lending activity between financial institutions and for the lending activities of the central bank (H¨ordahl and King, 2008). Thus, government debt occupies a central role in the European financial system, which is becoming increasingly integrated. Most bond markets have established a primary dealers system, which gives registered primary dealer access to primary markets. European primary dealers consist of international investment banks, large systemically important financial institutions and regional and local banks. The benefit of a primary dealer system is to improve market liquidity and quality, give that registered financial firms are required to actively provide trading opportunities.5 Dunn et al. (2006) note that the appeal of the primary dealer business may stem from the income and fees of additional services which are linked to the status of being a primary dealer rather than the market5

For a comprehensive overview across European bond markets and the privileges and requirements of primary dealers, see the Primary Dealer Hand Book (2015).

50

3.2. Market Structure

making operations themselves. The market structure in the secondary market is divided into the intra-dealer segment, which allows dealers to net their positions, and the dealer-to-customer segment. In the intra-dealer market, a larger part of trading occurs on electronic platforms, such as MTS or Eurex (Dunn et al., 2006). At the same time, trades take place on OTC markets, which are more opaque and allow for higher flexibility. The largest electronic intra-dealer trading platform is provided by MTS, which accounts for approximately 70 % of the trading volume on electronic platforms (Persaud, 2006). On domestic MTS platforms, most of the trading occur in debt securities issued by central governments and sovereign entities. EuroMTS is a pan-European platform for benchmark bonds. Benchmark bonds are required to fulfill certain criteria, such as being recently issued (seasonality) and an outstanding value of at least 5 million. However, benchmark bonds trade on the domestic platforms and EuroMTS (Caporale and Girardi, 2011). The market participants on the electronic platforms are primary dealers, with marketmaking obligations, and dealers, which are price takers. Primary dealers submit quoted proposals, which are collected in a limit order book via time and price priority rules (Dufour et al., 2004). Although markets are transparent and the price and volume are disclosed following a transaction, no information about the traders’ identities is disclosed. Moreover, the clearing process is typically preformed via a central counterparty that further protects the participants’ identities. The structure can be described as a hybrid between a dealer market and a limit order market. Dealer-to-costumer markets are more fragmented, and price quotes are provided via electronic systems (request-for-quote), such as Trade Web or BondVision, and via direct communication in OTC markets. OTC markets are less transparent, and data from individual quotes and trades are typically not available. Thus, the higher fragmentation and lower transparency make these markets preferable for customers who want to trade larger volumes (Viswanathan and Wang, 2002). In a recent study, Hendershott and Madhavan (2015) compare the transactions conducted via voice trading (OTC) and an electronic auction platform in US corporate bonds and investigate the endogenous choice between trading arrangements. Their findings show that easier transactions in more liquid instruments occur on

Chapter 3. The European Government Bond Market

51

the electronic multi-dealer venue. Thus, the results are consistent with the view that OTC markets are more preferable for trades with a higher chance of information leakage, although the trading costs are higher and bonds are typically more illiquid. In summary, the co-existence of di↵erent market environments may better suited for the trading needs of market participants rather than a single alternative. The regulations under MiFID II, the implemented of which has been delayed until 2018, will push for higher transparency in the part of the fixed income market that is considered to be the most liquid. The definition of liquid instruments is among the most heavily debated aspects of the regulation, give that transparency in rarely traded securities may discourage suppliers of immediacy to take inventory positions.6 Further, it still remains to be seen which role automated trading strategies will play in European fixed income markets following the regulation. Anecdotal evidence suggests that, at the moment, principal trading firms (PTFs) and high-frequency traders only contribute a small share of the trading volume in electronic secondary markets.7 Understanding the influence of agents with automated trading strategies on market quality and liquidity is important, as they may increase the chances of strong price movements as experienced with US treasury markets in October 2014 (Bech et al., 2016). Under the assumption that measures are introduced to prevent adverse outcomes, these new players may develop innovative solutions and compete with traditional liquidity suppliers, which may lower trading costs for investors (Callaghan, 2016). In Chapter 4, I investigate the supply side of market liquidity in European bond markets during the period 2006-2014. The empirical analysis can add to the understanding of changes in liquidity over time.

6

7

Critical issues with MiFID II from the perspective of practitioners were expressed in the Bond Market Contract Group (BMCG) on October 13, 2015. See: https://www.ecb.europa.eu/paym/groups/bmcg/html/index.en.html. The nexus between transparency and liquidity in European markets is discussed by Dunn et al. (2006). The assessment of the participants in the ECB’s Bond Market Contract Groups is summarized in the discussion on April 2016.

53

Chapter 4

Short-Term Reversals 4.1

Introduction

The supply of liquidity plays an important role in financial markets. In the European sovereign debt crisis the inability and/or unwillingness of private agents to provide liquidity in bond markets required an intervention of the ECB in order to maintain market functionality.1 In particular, market participants with high trading demands, such as mutual funds or insurance companies, may incur a significant amount of costs from trade execution, which are contingent on the current state of liquidity supply. Suppliers of market liquidity are required to “lean against the wind,” and take up trading positions opposite to the current price movements. Thus, liquidity suppliers may be challenged by fluctuations in aggregate demand for liquidity and correlations of trading decisions. Recently, liquidity and its resilience of government bond markets have attracted the attention of regulators and market participants. For instance, the coincidence of extreme price movement and evaporating liquidity on a 1

The ECB announced the securities market program on May 10, 2010 to address the depth and liquidity in dysfunctional market segments. https://www.ecb.europa.eu/press/pr/date/2010/html/pr100510.en.html. A detailed analysis follows in the next chapter.

54

4.1. Introduction

single trading day was observed on October 15, 2014 in the US treasury market (Joint Sta↵ Report, 2015) and on May 7, 2015 in the German bund market (Riordan and Schrimpf, 2015). Yet, there are only few attempts in the literature to investigate the supply side of liquidity in government bonds markets. However, fixed income markets have fundamentally di↵erent features than equity markets, for which reason the existing evidence on stocks may not be conclusive (Hendershott and Seasholes, 2014; Nagel, 2012). Accordingly, several questions remain unanswered. Does liquidity supply in government bond markets yield positive returns? For instance, Green (2004) cannot find positive compensation from providing immediacy in treasury markets. In European bond markets, no related evidence exists to my knowledge. Recent anecdotal evidence in European markets suggests that returns may indeed be negative in recent times.2 A second question addressed in this chapter is if the returns from liquidity supply increase under financial stress. Evidence on equity markets appears to provide conflicting results. For instance, Nagel (2012) find strong returns from liquidity supply, as the compensation for carrying inventory risk increases when funding is more difficult to obtain and only few providers of immediacy are active in the markets. Khandani and Lo (2011) show that systematic sell-o↵s and liquidation under severe stress can cause temporary losses from providing liquidity (“Unwind Hypothesis”). This chapter investigates the liquidity supply in European government bond markets during 2006 to 2014. A simple contrarian trading strategy is constructed to replicate the trading behavior of liquidity suppliers. Bonds with high past returns have likely experienced strong buying pressure, rendering providers of immediacy to hold short positions. Liquidity suppliers will have positive inventory (or long position) in bonds with negative past returns, which have experienced aggregate selling pressure. The short-term reversal strategy approximates the trading of liquidity suppliers, when the aggregate order flow is correlated with past price changes. 2

At the beginning of 2016, several financial institutions reduced their liquidityproviding activities in European bond markets. On January 21, 2016 Reuters reports “Squeeze bank dealers quit European government bond markets.” Further, the e↵ect of higher regulations and costs for the market-making sector on liquidity and market stability were discussed at the 10th AFME European Government Bond Conference. Bloomberg reports on November 11, 2015 “Primary Dealers’ Future Questioned in Europe as Rules Sting.”

Chapter 4. Short-Term Reversals

55

The seminal paper by Lehmann (1990) documents positive return of short-term contrarian investing in US equity markets. There exists a sizable strand of literature that investigates related strategies.3 Di↵erent strands of research name either behavioral biases or imperfect market liquidity as potential causes for negative serial correlation in returns. In this chapter I provide evidence on the latter explanation for government bonds. Previous research studies have investigated liquidity-supplying strategies on equity markets (Hendershott and Seasholes, 2014; Avramov et al., 2006). Regarding the US corporate bonds markets, Bao et al. (2011) measure illiquidity by the transaction price auto-covariance and explore its implication on bond prices. Market microstructure literature finds that transaction costs in financial markets are caused by information asymmetries (Glosten and Milgrom, 1985), search costs (Duffie et al., 2005) and risk aversion of market makers (Grossman and Miller, 1988; Ho and Stoll, 1983). The inventory channel posits that market makers demand compensation for holding risky assets in their inventory, which causes temporary deviations from the fundamental value (Grossman and Miller, 1988). The short-term mispricing is resolved when inventory positions are reversed. Accordingly, prices move in the opposite direction of accumulated inventory positions, which finds support in empirical research (Hendershott and Menkveld, 2014). The liquiditybased explanation leads to distinct testable features of short-term contrarian profits, which are investigated in this chapter. In the cross-section of bonds which are infrequently traded and have high illiquidity returns show stronger negative return autocorrelation, as reverting inventory positions may be more time-consuming than for highly liquid bonds. Aggregate market liquidity in equity and bond markets is found to fluctuate over time (Chordia et al., 2005). Thus, during times when aggregate market liquidity is low return reversals increase. I provide evidence of positive and significant returns from short-term reversal investing in European government bonds. The sample includes low-risk treasury bonds issued by six European sovereigns. The strategy invests in bonds with extremely low performance and sells the best 3

Other early studies on contrarian investing are by Jegadeesh (1990) and Werner F. M. De Bondt (1987).

56

4.1. Introduction

performing bonds. In contrast, bonds with less extreme absolute returns show no significant return predictability. Further, I show that the relative strength strategy as suggested by Avramov et al. (2006), which accords larger weights on bonds with lower liquidity, results in higher returns compared to equally-weighted portfolios. The results are consistent with the explanation that illiquid bonds experience stronger deviations from the efficient price than liquid bonds do, which enhances portfolio profits once the prices reverse. Further, aggregate bond market liquidity explains the time variations in returns from liquidity supply. Individual bond liquidity is proxied by the nominal bid-ask spread or the high-low spread estimator of Corwin and Schultz (2012), whereas market liquidity is the equally-weighted average across all bonds. Previous research finds that aggregate liquidity changes over time and that liquidity of individual securities co-moves with a market-wide aggregate (Chordia et al., 2000; Karolyi et al., 2012). When the level of market liquidity declines, individual bond liquidity is likely to follow. In support of this idea, the coefficient of market illiquidity is significant and positive, which indicates that returns from liquidity supply increase if market liquidity decreases. In addition, the result of the time series analysis on portfolio returns shows that long-short returns cannot be explained by common Fama and French (1993) fixed income risk factors. However, the goal of this paper is not to investigate whether this strategy could be successfully exploited by outside investors. This may not be possible due to high portfolio turnover and transaction costs. Taken together, these findings are strongly consistent with the inventory channel, and complement the evidence of liquidity-driven reversals in stock markets (Avramov et al., 2006; Hendershott and Seasholes, 2014). Recent theoretical models on market liquidity find a non-linear behavior in episodes of financial stress. Brunnermeier and Pedersen (2009) suggest a two-stage equilibrium wherein during normal times, liquidity shocks are bu↵ered by financial intermediaries, while during funding constraints, sello↵ and destabilizing margins may result in a bad equilibrium. The sample period covers two crises episodes, and therefore is well-suited to test the prediction of theoretical literature. The global financial crisis from 2007 to 2009 and the European sovereign debt crisis from 2010 to 2011 have

Chapter 4. Short-Term Reversals

57

dramatically increased the stress level in the financial system and lowered the funding ability of financial institutions. The empirical approach is to estimate a threshold regression, which allows for independent coefficients in multiple regimes. Regime switches are endogenously identified on a pre-defined threshold variable (Hansen, 2000). I suggest the implied volatility index of European equities (VSTOXX) or the spread of unsecured interbank loans and overnight interest swaps (EURIBOR-OIS) in order to identify potential regime changes. During the recent crises, the interest spread indicates episodes of flight-toquality. The volatility index is often referred to as a common stress indicator. I find strong empirical evidence on threshold e↵ects for both variables. Consistent with the theory on liquidity spirals, the results reveal that returns from liquidity supply in the stress state, which are not explained by common risk factors, are highly significant and larger in magnitude than they are in the normal state. Moreover, the increase in returns does not merely reflect higher volatility of bond prices. Supporting evidence shows that the negative correlation between past and future returns intensifies in the stress states, which is consistent with the inventory channel. These findings strengthen the argument that in government bond markets, shortterm price deviations from the underlying value change according to the level of financial stress.

4.2

Literature Review

Contrarian investing is a simple trading strategy, which buys past losers and sells past winners (Werner F. M. De Bondt, 1987). A large body of research revolves round this investment idea, which tries to exploit negative serial correlation of equity markets at di↵erent time horizons.4 The overreaction of investors o↵ers a potential explanation for the apparent conflict with weak-form market efficiency (Jegadeesh, 1990; Lehmann, 1990). A different strand of research finds that short-term reversals evolve as markets 4

Research on contrarian investments finds short-term reversals of a few weeks or long-term reversals of more than a year, for instance Conrad et al. (1997), Werner F. M. De Bondt (1987), and Avramov et al. (2006).

58

4.2. Literature Review

are not perfectly liquid. Thus, temporary deviations from the efficient price can be caused by non-informational trading (Campbell et al., 1993). Large transactions may cause price pressures, which are even greater for lessliquid assets.5 Temporary price deviations from the fundamental value are costs for liquidity takers, which are not captured by static bid-ask spreads (Hendershott and Menkveld, 2014). Thus, for institutional investors with large trading demand price pressures may have a sizeable impact on the realized transaction costs. Empirical evidence on liquidity-based explanations for short-term reversal profits is provided by Avramov et al. (2006). Sorting stocks by liquidity and turnover enhances the subsequent reversals, whereas reversals are particularly strong for stocks that experience the lowest past return. Additional evidence in favor of liquidity-driven reversals is that stocks with high liquidity and low turnover show significantly lower autocorrelation than those with low liquidity and high turnover. These findings support price pressure as an explanation for return reversals. Kaul and Nimalendran (1990) find that return reversals are mostly driven by the bid-ask bounce and vanish when the return is calculated on the true price instead of the observable price. The results by Nagel (2012) show that reversal returns on mid-prices are highly positive. Avramov et al. (2006) note that after accounting for transaction costs, reversal returns lose profitability. Lee et al. (2003) investigate contrarian investing in Australian markets and do not find positive returns after transaction costs. Additional research finds that short-term contrarian investing can be profitable. For instance, De Groot et al. (2012) limit their sample to stocks with low transaction costs and reduce portfolio turnover to show that returns survive transactions costs. Despite the conflicting evidence on the profitability of short-term contrarian investing, market makers may generate a positive payo↵ from providing immediacy for uninformed investors. In addition, liquidity providers may generate returns from di↵erent high-margin activities, which are linked to market-making

5

Campbell et al. (1993) note that stocks have short-term downward sloping demand curves, which allow for short-term deviation from its fundamental value due to non-informed trading.

Chapter 4. Short-Term Reversals

59

activities. Therefore, the nominal profits of liquidity provision do not account for additional benefits of maintaining sound customer relations. Returns from liquidity provision are not constant over time and are influenced by the aggregate liquidity supply and level of market liquidity. Nagel (2012) finds that the liquidity supply in equity markets evaporates during periods of stress. Time variations in return reversals can be predicted by the VIX index, and the compensation of liquidity supply increases during stress periods.6 Khandani and Lo (2011) show that a market-making strategy experienced significant losses in the second half of 2007, when quantitative hedge funds lost record amounts following the beginning of the US subprime crisis. The authors explain losses of the high-frequency long-short strategy in August 2007 as the result of forced liquidations, and the decline in liquidity-providing activity. There exists a growing strand of research on European bond market liquidity. Several authors investigate the pricing of liquidity (Darbha and Dufour, 2015; Bai et al., 2012; Beber et al., 2009). For instance, Beber et al. (2009) finds that the influence of liquidity on bond yields is subject to changes over time and appears highly relevant for investors under stressful market conditions. Pelizzon et al. (2016) document the influence of credit risk on liquidity in Italian markets. Bai et al. (2012) find liquidity linkages between single bond markets and aggregate bond liquidity. This chapter adds to previous research by providing a better understanding on the variation of bond market liquidity over time.

4.3 4.3.1

Data and Methodology Sample Data

Bond market data is retrieved from Thomson Reuters Tick History.7 The sample period spans from January 2006 to December 2014. I include fixed 6

7

The authors note that theoretical models of Brunnermeier and Pedersen (2009) and Adrian and Shin (2010) predict the same higher returns of liquidity supply in high VIX environments. The database is provided by the DALAHO at the University of Hohenheim.

60

4.3. Data and Methodology

rate treasury bonds issued by Austria, Belgium, Finland, France, Germany, and the Netherlands. The main sample of bond market data is from OTC markets. The following observations are collected at daily frequency: covering end-of-day, bid, ask, high, and low prices.8 Bonds with a time to maturity of less than 3 years and more than 15 years are excluded. Further, bonds with infrequent observations are deleted each month.9 A second set of bond data is provided by EuroMTS, an intra-European platform for benchmark bonds. In order to qualify for trading on the centralized European market, bonds need to be issued recently and fulfill additional listing requirements (Caporale and Girardi, 2011). Due to data constraints, the time series is shorter than the main sample (see Section 4.4.4). Intraday observations cover bid and ask quotes, the number of quotes in the bid and ask side, and the daily trading volume. Summary statistics are presented in Table 4.1. Bond returns are calculated on mid-prices. Bonds are grouped by rating of the issuers. The first group requires bonds to have an AAA rating over the entire sample period including bonds issued by Germany, Finland, and the Netherlands. The sub-AAA group hosts bonds issued by Austria, Belgium, and France. Daily return averages are 0.69 bps for the AAA group and 0.64 bps for the sub-AAA group. The liquidity of each bond is calculated by the nominal bid-ask spread. The average spread of the (sub-)AAA group is 0.10 (0.18) EUR. An alternative proxy of liquidity is the high-low spread estimator of Corwin and Schultz (2012), which decomposes the distance of daily high and low prices in spread and volatility.10 The average daily values of both groups are around 0.002. The average daily return on mid-prices on the electronic platform is 0.66 bps and 0.32 bps for the AAA and sub-AAA group, respectively. The bid-ask spread on the electronic platform is mea8

9

10

Sources of OTC data listed for bonds issued by: Finland (Barclays, BNP Paribas, Citigroup, Deutsche Bank, and Nordea); Germany (Bayerische LB, J.P. Morgan, Commerzbank, DekaBank, DZ Bank, CSFB, NATIXIS, and LBBW); and the Netherlands (ABN AMRO, BNP Paribas, Citigroup, CSFB, Deutsche MG, ING Bank, Morgan GTY, NIB, Rabobank International, and SNS Bank). The prices of bonds that are infrequently traded and have less than 75 percent of monthly observations are deleted. Return and liquidity proxies are winsorized at a 99.5 0.5 level. A recent study by Schestag et al. (2016) documents that the high-low spread is among the best performing low-frequency measures of transaction cost proxies on US corporate bonds.

Chapter 4. Short-Term Reversals

61

sured as the time-weighted average on 15-minute intervals. An additional proxy is the absolute di↵erence between the number of bid and ask quotes on the top level of the order book, which are recorded over the entire trading day, and then scaled by the total number of quotes. Further, the log of the Amihud (2002) measure is calculated as the ration of return in absolute terms and trading volume. Finally, note that bonds issued by peripheral countries are not included due to the illiquidity in these markets during the sovereign debt crisis. Accordingly, the bond prices of countries that experienced high sovereign risk, such as Italy, Spain, and Portugal, may be distorted over the observation period. For instance, see studies by Mitchell et al. (2007) or Da and Gao (2009), which document price distortions due to a temporary lack of capital in stock and corporate bond markets.

4.3.2

Portfolio Construction

The return of long-short portfolios is based on selling winner bonds and buying loser bonds. The inventory position of liquidity suppliers most likely comprises bonds with high absolute price changes. Bonds in the inventory of liquidity providers are expected to show strong return reversals, which are caused by temporary deviation from the efficient price. In contrast, bonds that did not experience strong price changes should experience smaller reversals, if any. In the main part of this chapter, the sample is split by default risk in order to minimize the factor exposure of long-short portfolios. Thus, bonds are grouped by issuer credit rating. For brevity, the focus will be on bonds with medium maturity. I arbitrarily choose bonds within the range of 4 to 8.5 years to maturity, which results in a total of 95 (85) bonds in the (sub)AAA-rated group. In the final step, bonds are sorted in four equally sized portfolios based on their past performance. Then two contrarian long-short portfolios are constructed, which invest in the extreme and intermediate return portfolios.

62

4.3. Data and Methodology

Table 4.1 Summary statistics by rating group The table reports the mean and standard deviation of bonds grouped by credit rating. AAA-rated bonds included those issued by Finland, Germany, and the Netherlands; while sub-AAA rated bonds are issued by Austria, Belgium, and France. Return is calculated on daily mid-prices. Bid-ask is the nominal bid-ask spread and the highlow is the spread estimator of Corwin and Schultz (2012). Bid-ask t denotes the timeweighted bid-ask spread and is based on intraday quotes at the top level of the electronic limit order book of EuroMTS. Imbalance is the absolute di↵erence between daily bidand ask-quotes scaled by the total number of quotes. The Amihud measure is the log ratio of return and trading volume in absolute terms. The observation period for OTC markets spans from January 1, 2006 to December 31, 2014. The observation period for EuroMTS is from January 1, 2006 to December 31, 2011.

OTC

AAA group Return (daily) Maturity Coupon Bid-ask High-low spread Time-weighted bid-ask Quote imblance Amihud Sub-AAA group Return (daily) Maturity Coupon Bid-ask High-low spread Time-weighted bid-ask Quote imblance Amihud

Electronic Platform

mean

std

mean

std

0.69 6.72 3.65 0.10 0.002

29.36 2.78 1.42 0.12 0.008

0.66 6.25 3.95

29.34 2.20 0.74

0.06 0.17 -15.4

0.07 0.16 1.4

0.32 6.9 4.04

34.10 2.6 0.60

0.14 0.17 -15.2

0.17 0.17 1.5

0.64 7.5 4.39 0.18 0.002

31.61 3.1 1.47 0.21 0.005

Chapter 4. Short-Term Reversals

63

Bonds are sorted into return portfolios based on the performance over the past five days.11 The holding period following portfolio formation is guided by the duration of reversals, which depends on the time needed by liquidity providers to balance inventory positions. Empirical literature on stock markets suggests di↵erent durations of return reversals.12 To my knowledge, evidence on bond markets does not exist up to this point. The duration of mean reversion of market makers inventory positions may provide guidance on the expected time of price reversals. However, bond markets are organized as dealer markets, and similar data for equity markets cannot be collected. Thus, I explore return reversals over di↵erent holding periods. The portfolio structure is subject to time-variations, which arises naturally as portfolios are constructed on past returns. For instance, a surge in default risk over the formation period would cause low-risk bonds to gain in value in comparison to riskier bonds. Consequently, the contrarian portfolio will be long in high-risk bonds and short in low-risk ones. Similar considerations apply to shifts in the term structure.13 As will be shown in the next section, common factors are of minor importance in explaining the short-term reversal strategy. This result is due to the construction of reversal portfolios within a rather homogenous sub-set of government bonds.14

11 12 13

14

I test for di↵erent formation periods up 15 days. The five-day formation period appears to result in the best-performing portfolios. See Section 2.1 for evidence on equity markets. In general, the turnover in fixed income cash markets is likely to be lower than that in equity markets. In addition to standard risk factors, other time-varying patterns exist, such as industry momentum, which strongly enhance reversal return if considered as described by Da et al. (2011). Research in equity markets suggests sorting stocks on risk-adjusted returns, yet the strategy would still be exposed to risk factors as described in Nagel (2012).

64

4.4. Results

4.4 4.4.1

Results Return Reversals and Portfolio Characteristics

This section investigates the characteristics and holding period returns of single sorted bond portfolios. Following the approach described in the previous section, four portfolios are created for each default risk group. The lowest performing bonds are in portfolio 1, and portfolio 4 includes bonds with the best past performance. Panel A in Table 4.2 shows that portfolio 1 in the (sub-)AAA group has the lowest formation period return of -18.5 bps (-22.5 bps) and the average return in portfolio 4 equals 24.6 bps (27.3 bps). Bonds in the extreme portfolios (1 and 4) appear less liquid than those in the middle portfolios (2 and 3). This pattern is observable for the bid-ask and high-low spread and is more pronounced for the AAA group. This observation is consistent with the finding that less liquid securities are usually more volatile. The relationship may be bi-directional, as price movements increase the risk of liquidity suppliers, leading to higher price movements, etc.15 The average time to maturity of past winner bonds in portfolio 4 appears higher compared to the other portfolios during the observation period. Panel B in Table 4.2 presents the equally-weighted post-formation returns and the Newey-West t-statistics of portfolios 1 to 4. Returns are in basis points. The results show that portfolio returns monotonically increase from portfolio 4 to 1. More precisely, bonds in portfolio 1, which have the lowest past performance, show significant positive returns over the next 10 days and outperform portfolios 2 and 3. In the (sub-)AAA-rated group, the returns are 2.4 bps (1.9 bps) on the next day and increase to 8.4 bps (9.1 bps) over a 10-day holding period. Past winner bonds show negative and significant average return of -1.18 bps over a holding period of one day in the AAA group. For a holding period of five days and longer, the returns are positive but insignificant. On average, bonds with the highest past returns show the lowest future returns. 15

For instance, see the liquidity on equity and bond markets as discussed by Goyenko and Ukhov (2009)

Chapter 4. Short-Term Reversals

65

Table 4.2 Characteristics and portfolio returns of single sorted portfolios Panel A presents portfolio characteristics of single sorted bond portfolios. The mean of formation periods return (t-5;t-1), bid-ask spread, high-low spread, and time to maturity (ttm) is reported. Bid-ask is the nominal bid-ask spread and the high-low spread is the spread estimator of Corwin and Schultz (2012). The standard deviation is denoted in parentheses. Panel B presents the holding period return of return sorted portfolios over 1, 5, and 10 days following the portfolio formation. Panel C presents the raw returns and alpha coefficients of zero costs long-short portfolios. Alphas are calculated on a two-factor regression on Fama and French (1993) TERM and DEF factors. The TERM factor is constructed on the 10-year and 3-month German treasury yields. The DEF factor is calculated by the iBoxx AAA Eurozone sovereign index and the iBoxx BBB Eurozone sub-sovereign bond index. The Newey-West t-statistics are denoted in parentheses. The observation period spans from January 1, 2006 to December 31, 2014.

Panle A: Characteristics portfolio 1 2 3 4

AAA group sub-AAA group (t-5;t-1)Bid/Ask High/low ttm (t-5;t-1)Bid/Ask High/low ttm -18.52 (5.67) -3.20 (3.49) 8.21 (3.46) 24.61 (6.45)

0.0875 (0.060) 0.0774 (0.049) 0.0787 (0.047) 0.0868 (0.060)

0.0024 (0.004) 0.0018 (0.003) 0.0019 (0.003) 0.0025 (0.005)

5.98 (1.17) 5.81 (0.67) 5.9 (0.63) 6.25 (1.21)

-22.52 (2.12) -5.09 (8.97) 9.28 (8.52) 27.30 (3.32)

0.1602 (0.114) 0.1457 (0.097) 0.1508 (0.108) 0.1599 (0.120)

0.0017 (0.002) 0.0016 (0.002) 0.0016 (0.002) 0.0017 (0.002)

6.05 (1.06) 5.96 (0.61) 6.06 (0.56) 6.3 (1.04)

Panel B: Holding period return holding period

1

1 day

2.39 (4.51) 5.63 (2.87) 8.42 (2.87)

5 days 10 days

AAA group 2 3 0.81 (1.62) 3.2 (1.67) 6.37 (2.24)

0.72 (1.44) 3.16 (1.65) 6.28 (2.17)

4

1

-1.18 (2.16) 1.55 (0.77) 5.3 (1.77)

1.87 (3.23) 5.89 (2.42) 9.06 (2.01)

sub-AAA group 2 3 0.88 (1.58) 3.21 (1.39) 5.46 (1.26)

0.64 (1.19) 2.18 (0.94) 4.70 (1.06)

4 -1.06 (1.77) 0.42 (0.17) 3.62 (0.77)

66

4.4. Results Panel C: Long-short return

holding period 1 day 5 days 10 days

AAA group raw alpha 1-4 2-3 1-4 2-3 3.57 (10.20) 4.08 (4.86) 3.12 (2.60)

0.14 (0.86) 0.09 (0.22) 0.04 (0.06)

6.47 (7.19) 4.53 (2.77) 4.09 (1.58)

sub-AAA group raw alpha 1-4 2-3 1-4 2-3

1.3 2.93 0.23 4.62 0.105 (4.04) (7.42) (1.35) (4.99) (2.20) -0.47 5.47 1.02 -0.48 -1.57 (-0.63) (4.45) (2.09) (-0.28) (-1.91) -1.42 5.44 0.76 -3.63 -2.02 (-1.25) (2.91) (1.07) (-1.48) (-1.87)

Next, two contrarian portfolios are created, which are long in loser and short in winner bonds. Panel C reports the results of an extreme portfolio, which is long in portfolio 1 and short in portfolio 4; and a intermediate portfolio, which is long in portfolio 2 and short in portfolio 3. In addition to raw returns, the alphas from a linear regression on Fama-French factors are reported. The standard Fama and French (1993) factors are calculated to control for the term structure (TERM) and default risk (DEF). The TERM factor is the yield spread between 10-year German sovereign bonds and 3-month German treasury bills. The yield spread between the iBoxx Euro sub-sovereign BBB index and the iBoxx AAA Euro sovereign bond index mimics the default factor.16 The following regression equation is estimated: Rt,i = ↵i +

i,D DEFt

+

i,T T ERMt

+ ✏i,t

(4.1)

where Rt,i represents the long-short return of portfolio i on day t. The variable i,D denotes the loading on the default factor, and i,T represents the loading on the term structure factor. ↵i is the regression constant, which is interpreted as the risk-adjusted return.17 16

17

Darbha and Dufour (2015) investigate whether market liquidity is priced in portfolios of European sovereign bonds. The authors use a similar approach in constructing common risk factors for European fixed income markets. For robustness, I control for potential weekly and monthly patterns. Contrarian returns are regressed on a set of dummy variables. The weekly e↵ects are controlled

Chapter 4. Short-Term Reversals

67

The results show that the extreme portfolio produces significant and positive returns over a holding period of one day (3.57 bps) and for up to 10 days (5.44 bps). However, the raw returns of the intermediate long-short portfolio are not larger than 1 bps and are statistically insignificant for most of the tabulated holding periods. The highest alpha in the intermediate portfolio is 1.3 bps for a holding period of 1 day. For the extreme portfolios, alphas of both groups are as high as 6.5 bps for one day and decline for a 10-day holding period to 4.1 bps. Thus, the results support the thesis that bonds that have experienced strong absolute price changes over the formation period show stronger subsequent price reversals. Accordingly, the success of the reversal strategy may be explained by a lower liquidity of bonds in the extreme portfolios, which will be closely investigated in the following section. Finally, the average duration of the reversals is less than five days. In each rating group, alphas decline for holding periods greater than one day. The positive and significant reversal returns represent the profits of market makers, such as a primary dealer, who provide immediate trade execution for liquidity-motivated investors. Returns are calculated on midprice returns and do not include revenue from the bid-ask spread. Nevertheless, the implementation of a reversal strategy may not be limited to the market-making sector. Transaction costs need to be considered in order to assess the profitability from the perspective of outside investors. The average bid-ask spreads are larger than the raw returns or alphas (Panels B and C). For instance, in the liquid AAA-rated group, the relative bidask spread of around 8 bps is larger than the alpha and raw return of 6.5 bps and 3.6 bps, respectively. Clearly, more research is required to provide a definite answer on the profitability of contrarian investing. The following parts of this chapter are concerned with investigating the economic explanation of this return pattern.

for by dummies from Tuesdays to Fridays. Monthly dummies are included from March to November. To account for potential January e↵ects or end-of-the-year e↵ects, a dummy is included for each week in January and December. There is no evidence of strong weekly or monthly patterns. The complete results are not reported and are available upon request.

68

4.4.2

4.4. Results

Liquidity-driven Reversals

The goal of this section is to analyze if reversals are related to bond illiquidity. First, I present evidence on the relative strength strategy to test if less liquid bonds show stronger reversals (Avramov et al., 2006). This approach calculates portfolio weights by the illiquidity of individual bonds. If bonds have lower liquidity, the initial price move following a large transaction and the subsequent reversals should be stronger than that of liquid bonds. Consequently, the profits of the relative strength strategy should yield higher returns than the equally-weighted returns. The portfolio weights of each individual portfolio (1 to 4) i on day t is given by: F Liqi,t 1 wit = PNp F i=1 Liqi,t

(4.2) 1

F where Liqi,t 1 represents the liquidity measure of bond i over the formation period, and Np is the number of bonds in each of the four return sorted portfolios. Liquidity is either measured by the quoted bid-ask spread or the high-low spread.

Panel A in Table 4.3 presents the average of long-short returns and Newey-West t-statistics of di↵erently weighted portfolios. If I calculate portfolio weights by the bid-ask (high-low) spread, the largest average return over the observation period is 4.9 bps (4.9 bps) for the AAA-rated group. In the last three columns, the daily di↵erence with equally-weighted portfolios and within both liquidity proxies are shown. The daily outperformance of the relative strength strategy is more than 1 bps and statistically significant for both rating sorted groups. Next, I control for the influence of the common fixed income risk factors (TERM and DEF). Panel B reports the alpha coefficients of a two-factor regression. Panel C shows the results of a second specification, where the factor regression is augmented by bond market returns. Hamed et al. (2010) find lower liquidity and higher returns from liquidity supply in down markets. The market return is calculated as the equally-weighted return of all bonds in the sample over a period of four weeks.

Chapter 4. Short-Term Reversals

69

Panels B and C show positive and significant regression alphas for liquidity-weighted portfolios. For (sub-)AAA-rated bonds, the alpha coefficients of the reversal strategy increases to 8.5 bps (7 bps). Across both specifications and sub-groups, the liquidity-weighted alphas are higher than that for the equally-weighted strategy. Adding market return as an additional control leaves the results in Panel C almost unchanged. In a final step, I estimate Fama-MacBeth type cross-sectional regressions. The individual bond returns are explained by cross-sectional regressions on a daily frequency. The following equation is estimated: Ri,t = ↵t +

F 1,t Reti,t 1

+

F 2,t Reti,t 1 DLiq

+ Xt + ✏i,t

(4.3)

where Ri,t stands for the holding period return of bond i at time t. RetFi,t 1 are the returns over the formation period from day t-5 to t-1, which is included as an explanatory variable. In order to test if bonds with lower liquidity show stronger reversals, I add an interaction dummy with formation period returns, DLiq , which equals one if the liquidity of a bond is below the median or the 75th quantile.18 As additional controls, a dummy for the country of issuance and a variable for the maturity of the bond are included. Panel D presents the averaged coefficient and Newey-West t-statistics which are denoted in parentheses. The results are estimated on the main sample, which includes bonds with medium time to maturity, and the complete sample with a broader range of maturities. The results show that past returns predict future returns. The average of the time series coefficients is -0.16 (-0.12) in the medium maturity (complete) sample and the t-statistic is -18.22 (-16.97). Further, in both samples the results show an even stronger reversal pattern for less liquid bonds. In the medium maturity sample, the interaction dummy is -0.01 for the 75th bid-ask spread quantile with a t-statistic of -3.45. The results are similar for the larger sample. However, the magnitude of the negative interaction dummy is less than 10% of the unconditional coefficient. Neverthless, the presented 18

I define the 50th or 75th quantiles by comparing bond liquidity within the respective rating and maturity group.

70

4.4. Results

Table 4.3 Liquidity-weighted portfolio returns and cross-sectional regressions Panel A shows the returns of contrarian long-short portfolios with equal and dynamic liquidity weights. The formation period equals five days. Liquidity weights are based on bid-ask spreads and high-low spreads (Corwin and Schultz, 2012) over the formation. The di↵erence between the equally-weighted returns and bid-ask weights (delta 2-3), high-low spread weights (1-3), and between high-low and bid-ask spreads weights (12) is shown. Panel B presents the alpha of a two-factor model on Fama and French (1993) TERM and DEF factors. The TERM factor is constructed on the 10-year and 3-month German treasury yields. The DEF factor is calculated by the iBoxx AAA Eurozone sovereign index and the iBoxx BBB Eurozone sub-sovereign bond index. Panel C presents the alphas on a three-factor model on TERM, DEF, and bond market returns. Panel D presents the average coefficients of daily cross-sectional regressions. Bond return is regressed on a constant (const), the return over the formation period, and an interaction dummy of bond illiquidity (Dliq) and formation return. The dummy equals one if a bond’s illiquidity proxy is in the top 50th or 75th percentile, and zero otherwise. Additional controls are the maturity of a bond and a dummy for the country of issuance. Regressions are estimated on the main sample (medium maturity) and a large sample with maturities from 3 to 15 years. The Newey-West t-statistics are denoted in parentheses. The observation period spans from January 1, 2006 to December 31, 2014.

Panel A: Liquidity-weighted portfolio return

AAA Sub-AAA

High-Low

Bid-Ask

Equal

delta 1-3

delta 2-3

delta 1-2

4.94 (8.95) 3.41 (8.13)

4.85 (10.56) 4.02 (9.48)

3.53 (9.85) 2.93 (7.41)

1.42 (4.38) 0.48 (4.20)

1.33 (6.54) 1.09 (7.77)

0.09 (0.30) -0.61 (-4.36)

Panel B: Two factor model

AAA Sub-AAA

Panel C: Three factor model

High-Low

Bid-Ask

Equal

High-Low

Bid-Ask

Equal

8.47 (5.93) 5.54 (5.76)

6.80 (6.38) 7.05 (6.70)

6.49 (7.22) 4.62 (4.99)

8.45 (5.87) 5.58 (5.74)

6.79 (6.23) 7.11 (6.71)

6.48 (7.09) 4.66 (4.97)

Chapter 4. Short-Term Reversals

71

Panel D: Cross-sectional regression on bond return medium maturity const return Dliq -0.45 (-0.67) q50 Bid-ask -0.49 (-0.74) q75 Bid-ask -0.47 (-0.72) q50 High-low -0.63 (-0.93) q75 High-low -0.42 (-0.63)

-0.16 (-18.22) -0.15 (-17.63) -0.15 (-16.90) -0.15 (-15.50) -0.15 (-17.16)

-0.008 (-2.88) -0.01 (-3.45) -0.005 (-1.71) -0.007 (-1.78)

all maturities const return Dliq -0.32 (-0.63) -0.38 (-0.75) -0.28 (-0.55) -0.41 (-0.80) -0.34 (-0.66)

-0.12 (-16.97) -0.12 (-16.82) -0.12 (-17.09) -0.12 (-16.80) -0.12 (-16.78)

-0.012 (-4.68) -0.012 (-3.97) -0.005 (-2.03) -0.011 (-3.62)

results are highly suggestive that less liquid bonds experience stronger reversals. Market liquidity is subject to variations over time. The compensation to liquidity provision should be high if the market liquidity is low. Therefore, I expect a positive relation between returns and illiquidity. In order to test the predictions of the liquidity channel, I estimated a linear time series regression for the equally-weighted reversal portfolio.19 The following equation is estimated:

Rt = const +

M 3 Liqt 1

+ X + ✏t

(4.4)

where Rt represents the long-short return on day t. The variable LiqtM 1 stands for changes in market liquidity over the formation period, which is measured by the equally-weighted average of each bond’s liquidity esti19

I explain the return of the equally-weighted portfolio, in order to avoid an endogeneity problem for liquidity-weighted returns.

72

4.4. Results

Table 4.4 Liquidity regression The table presents results of linear regressions on the return of equally-weighted longshort contrarian portfolios. The explanatory variables is the di↵erenced market liquidity over the formation period (mktliq), which is measured by the equally-weighted average of each individual bonds liquidity estimates. Additional controls are the Fama and French (1993) TERM and DEF factors, bond market return (mkt) and the Vstoxx index. The DEF factor is calculated by the iBoxx AAA Eurozone sovereign index and the iBoxx BBB Eurozone sub-sovereign bond index. The Newey-West t-statistics are denoted in parentheses. The observation period spans from January 1, 2006 to December 31, 2014.

AAA group #1 mktliq TERM DEF mkt vstoxx const

bid-ask spread #2 #3

#4

209 (3.19)

209 212 202 (3.21) (3.24) (3.11) -1.25 -1.28 -0.98 (-3.03) (-3.13) (-2.46) 0.21 0.27 -0.71 (0.66) (0.83) (-2.59) -33.68 (-0.82) 0.21 (3.10) 3.53 6.46 6.45 2.75 (9.19) (7.18) (7.05) (1.89)

#1

high-low spread #2 #3

#4

1048 (1.42)

1020 1033 1016 (1.40) (1.42) (1.41) -1.25 -1.28 -0.98 (-3.02) (-3.12) (-2.43) 0.20 0.26 -0.73 (0.65) (0.81) (-2.66) -30.88 (-0.76) 0.21 (3.15) 3.53 6.46 6.46 2.68 (9.55) (7.19) (7.08) (1.85)

sub-AAA group #1 mktliq TERM DEF mkt vstoxx const

66.5 (1.33)

bid-ask spread #2 #3

#4

67.1 66.3 65.1 (1.32) (1.29) (1.29) -0.59 -0.60 -0.46 (-1.28) (-1.31) (-0.98) -0.03 -0.02 -0.51 (-0.09) (-0.05) (-1.11) -9.31 (-0.18) 0.11 (1.91) 2.93 4.63 4.65 2.73 (7.50) (4.98) (4.96) (2.02)

#1 2542 (2.11)

high-low spread #2 #3

#4

2522 2521 2548 (1.98) (1.98) (2.02) -0.58 -0.59 -0.44 (-1.26) (-1.28) (-0.95) -0.04 -0.03 -0.52 (-0.10) (-0.07) (-1.11) -7.88 (-0.16) 0.11 (1.93) 2.93 4.60 4.63 2.67 (7.48) (4.99) (4.97) (1.97)

Chapter 4. Short-Term Reversals

73

mates.20 X stands for additional control variables, such as TERM, DEF factors, bond market returns and the VSTOXX index.21 The results in Table 4.4 indicate that illiquid market conditions are associated with higher returns from short-term reversals. Market illiquidity has a significant and positive impact on future returns from liquidity supply, if measured by the bid-ask spread for AAA-rated bonds. Once the additional control variables are included over specifications 2 to 4, the results show only little variation. The coefficient equals 202 with t-statistics of 3.11. A one standard deviation increase in market liquidity over the formation period, results in a 1.2 bps (202 ⇥ 0.0059) increase in reversals returns, all else being equal. Liquidity, as proxied by the high-low spread, has a significant and positive e↵ect on the long-short returns in the subAAA group. The coefficient is 2548 with a t-statistic of 2.02. The daily returns increase by about 1 bps for a one standard deviation increase in market liquidity (2548 ⇥ 0.0004). In summary, the results are consistent with the explanation that prices will deviate further from their fundamental values, if markets are less liquid, resulting in higher returns from providing liquidity. At the same time, exploiting this anomaly via the proposed investment strategy may become more difficult, as transaction costs will increase in an illiquid market environment. The same conclusions hold for liquidity in the cross-section of bonds. Reversals profits are higher for illiquid bonds and the transaction costs for the implementation are higher as well.

4.4.3

Reversals under Financial Stress

This section investigates whether the returns from the short-term reversal strategy are state dependent. The search for a sample split is motivated by 20 21

Market liquidity is di↵erenced, as it is persistent and not stationary during episodes of financial stress. In addition to the OLS regressions presented in the body of the chapter, a quantile regression is preformed to analyze the parameters in the outer quantiles of the return distribution. The results are supportive of the conclusion that illiquidity is positively associated with return reversals. The parameters for market liquidity are positive and increase for large positive return quantiles. Full results are available upon request.

74

4.4. Results

theoretical models on the link between liquidity and the ability of financial intermediaries to fund trading positions, which may result in a two-stage equilibrium as experienced in the recent crisis (Brunnermeier and Pedersen, 2009; Gromb and Vayanos, 2010; Boudt et al., 2013). Previous research finds that the pricing of liquidity risk may depend on the state of financial stress (Acharya et al., 2013; Watanabe and Watanabe, 2008). Nagel (2012) analyzes US equities and successfully predicts future profits from liquidity supply using a volatility index (VIX). Thus, I contribute to the existing research by investing in the link between financial stress and liquidity supply in bond markets. The theoretical model of Brunnermeier and Pedersen (2009) posits that the funding ability of financial intermediaries, such as dealers, hedge funds, or investment banks, and market liquidity is interconnected. During periods of financial stress, for instance, after a negative wealth shock to the balance sheet of financial intermediaries, their funding ability is diminished. When reduced funding ability is substituted with liquidation of assets, sell-o↵s may occur leading to a reduction in market liquidity and depression of asset prices. The resulting decline in asset values reduces the funding liquidity via higher margins, which leaves the market trapped in a self-enforcing liquidity spiral. However, in normal times, liquidity shocks or decline in asset prices can be bu↵ered by financial intermediaries and no destabilizing spiral e↵ects are triggered. Related research discusses intermediary capital and liquidity (Gromb and Vayanos, 2010). However, under particular market circumstances the link between financial stress and returns from providing immediacy may not be straightforward. For instance, the forced liquidation of assets during financial turmoil is described by Khandani and Lo (2011) during the quant crisis in 2007. The authors find that forced sellings may result in a continuation of the common risk factors. Moreover, returns from contrarian investing may turn negative over short holding periods if the momentum factor becomes large enough. It appears reasonable that the e↵ect is symmetric for larger selling or buying pressure. Following the literature discussion in Section 4.2, flights in highly rated government bond markets during periods of market stress may negatively a↵ect reversal returns.

Chapter 4. Short-Term Reversals

75

The goal is to estimate a threshold regression (Hansen, 2000) for longshort returns in order to provide empirical evidence to the proposed question. The underlying idea is to split the sample with the help of an indicator variable, and allow for distinct coefficients and intercepts in each sub-sample. The sample split is not deterministic, but rather is endogenously identified.22 In order to identify the di↵erent states, an economically meaningful threshold variable is required. I link portfolio returns to different indicators of financial stress. The VSTOXX is the implied volatility index of European equities and serves as a general indicator of uncertainty and fear in financial markets. The spread between the overnight interest swap (OIS) and three-month interbank lending rates (EURIBOR) provides information on the funding conditions of large financial institutions. This variable can also be interpreted as a trust indicator in the banking systems, as it is influenced by credit and liquidity risk. First, I investigate long-short returns at di↵erent levels of the VSTOXX and the EURIBOR-OIS spread. Thus, the trading days are independently sorted into high and low stress days. Then, the alphas of the two-factor model are calculated (specification (4.1)). Panel A of Table 4.5 shows that the holding period returns are strictly increasing in the VSTOXX. For instance, in the sub-AAA group, the alpha on days with the lowest stress level is 2.1 bps, which increases to 12.7 bps on high-stress days. Results for the AAA group are similar. If the observation period is divided into days with high and low values of the interbank spread, a similar pattern emerges (Panel B). Alphas from the long-short strategy are monotonically increasing in the EURIBOR-OIS spread for both rating groups. In Panel C, I jointly sort on the VSTOXX and the EURIBOR-OIS spread. For the AAA-rated group, I find that on days with low values of the VSTOXX and EURIBOR-OIS spread, the alpha is 0.87 bps and is statistically insignificant with a t-value of 0.54. The alpha is increasing in financial stress up to 15.2 bps with a t-value of 4.14 and on days with high VSTOXX and EURIBOR-OIS spread. The same pattern holds for alphas 22

The methodology bears similarities to structural break tests, however, the break is not identified by time but rather by the value of a pre-defined threshold variable. I also test for deterministic sub-samples, for example, the global financial crisis and sovereign debt crisis. However, I do not find distinctions in risk-adjusted profits from return reversals, which supports the approach to search for threshold e↵ects.

76

4.4. Results

Table 4.5 Contrarian alphas by financial stress condition The table presents alpha coefficients of equally-weighted long-short contrarian portfolios. Alphas are calculated on a two-factor regression on Fama and French (1993) TERM and DEF factors. The TERM factor is constructed on the 10-year and 3-month German treasury yields. The DEF factor is calculated by the iBoxx AAA Eurozone sovereign index and the iBoxx BBB Eurozone sub-sovereign bond index. The observation period is separated by the VSTOXX index and the EURIBOR-OIS spread in Panels A and B, respectively. In Panel C, days are sorted by the VSTOXX and EURIBOR-OIS spread. The Newey-West t-statistics are denoted in parentheses. The observation period spans from January 1, 2006 to December 31, 2014.

Panel A: VSTOXX

1 2 3 4

Panel B: EURIBOR-OIS

AAA group

sub-AAA group

4.77 (6.06) 7.02 (5.90) 10.09 (5.95) 12.81 (3.60)

2.09 (0.81) 4.26 (2.58) 8.51 (3.07) 12.67 (5.95)

1 2 3 4

AAA group

sub-AAA group

2.01 (1.58) 4.02 (4.84) 8.64 (3.46) 12.25 (9.10)

0.96 (0.81) 2.03 (2.58) 5.70 (3.07) 9.84 (5.95)

Panle C: VSTOXX and EURIBOR-OIS

OIS 1 2 3

1 0.87 (0.54) 4.08 (2.46) 5.26 (4.57)

AAA group VSTOXX 2 0.67 (0.31) 3.55 (1.80) 11.01 (7.52)

3 14.14 (2.89) 6.99 (1.86) 15.21 (4.14)

OIS 1 2 3

1 0.69 (0.54) 3.07 (3.06) 1.00 (0.85)

sub-AAA group VSTOXX 2 3 -1.10 (-0.32) 3.00 (1.64) 6.38 (4.52)

10.24 (2.90) 8.58 (2.19) 16.38 (3.07)

Chapter 4. Short-Term Reversals

77

of the sub-AAA group. For example, on days with high EURIBOR-OIS spread the alpha (t-value) increases with the VSTOXX from 1 bps (0.85) to 16.38 bps (3.07). I develop the initial evidence and investigate a potential sample split in returns from liquidity provision. The threshold regression model identifies the stress regime ( = 2), if the threshold variable qt j at the beginning of the formation period (t j) is above the threshold value . Otherwise, observations are sorted in the normal regime ( = 1). The statistical significance of a sample break is tested by the heteroscedasticity-robust Lagrange multiplier test. Hansen (1996) describes the construction of bootstrapped p-values. Equally-weighted long-short reversal returns are explained by a model of the following form: Rt = ↵  +

 D DEFt

+

 T T ERMt

+ ✏t

(4.5)

where  indicates the regime. TERM and DEF are common risk factors as introduced in Section 4.4.1. Table 4.6 presents LM-test statistics on regime change. P-values are calculated on 1,000 bootstrap replications. The test against no threshold is easily rejected at a 1 percent level for both threshold variables. The VSTOXX index identifies the stress regime for values greater than 18.87 (21.59) in the (sub-)AAA-rated group. The three-month EURIBOR-OIS splits the sample at 0.215 and 0.256 in the AAA and sub-AAA groups, respectively. Table 4.6 shows the regression results for a sample split in the two regimes. The threshold regression on the VSTOXX index reports a negative and significant coefficient of the TERM factor for both regimes in the AAA group. Loadings on the default factors are not statistically significant. The results are similar if the sample split is identified by the EURIBOR-OIS spread. For the sub-AAA group, factor loadings are mostly insignificant. Only in the stress regime, according to the EURIBOR-OIS spread, do the portfolio returns load negative on the TERM factor. Most notable is the di↵erence in the regression constant, which is positive and statistically significant in regimes 1 and 2. I find that the alpha in the stress state, compared to the normal state, is approximately twice the size in the AAA group and thrice the size in the sub-AAA group. These ob-

78

4.4. Results

Table 4.6 Threshold regression The table presents the regression results of a two-regime threshold regression as described in Hansen (2000). The dependent variable is the equally-weighted long-short contrarian portfolio return. Threshold variables are the spread of three-month EURIBOR rates over OIS and the VSTOXX index. The regression includes a constant (alpha) and the two Fama and French (1993) factors for term structure (TERM) and default risk (DEF). The TERM factor is constructed on the 10-year and 3-month German treasury yields. DEF is calculated by the iBoxx AAA Eurozone sovereign index and the iBoxx BBB Eurozone sub-sovereign bond index. The Newey-West t-statistics are denoted in parentheses. Below, the heteroscedasticity-robust LM test on the threshold e↵ect is reported. The P-values are calculated on 1,000 bootstrap replications. The observation period spans from January 1, 2006 to December 31, 2014.

alpha TERM DEF

Threshold LM-Test P-value

AAA group

Sub-AAA group

VSTOXX EURIBOR-OIS =1 =2 =1 =2

VSTOXX EURIBOR-OIS =1 =2 =1 =2

4.26 (5.07) -1.51 (-3.39) 0.19 (0.30)

10.87 3.96 8.85 (8.41) (3.99) (7.62) -1.54 -1.14 -0.86 (-3.47) (-2.38) (-1.96) -0.48 -0.08 -0.44 (-1.51) (-0.16) (-1.36)

18.87 70.24