Illiquidity or Credit Deterioration: A Study of Liquidity in the US Corporate Bond Market during Financial Crisis Nils Friewald WU Vienna Rainer Jankowitsch WU Vienna Marti Subrahmanyam New York University Italian Treasury Tuesday, June 28th 2011 nyustern_logo.jpg (JPEG Image, 182x161 pixels) http://www.nyucareerstudy.org/nyustern_logo.jpg
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 2/28 Liquidity is an important price factor The financial crisis has shown that credit and liquidity risk are key determinants of asset pricing. It is important to understand their (relative) effects and how they change during periods of crisis. It is also relevant to ask if there are interactions between these important factors. The most affected financial markets were over-the-counter markets, which makes research challenging. The US corporate bond market is an ideal laboratory for testing as detailed transaction data (since 2004) are available.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 3/28 Dramatic increase of average US corporate bond yield spread Spread in % 0 2 4 6 8 10 GM/Ford Crisis Normal Period Sub prime Crisis Jul 2005 Jul 2006 Jul 2007 Jul 2008
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 4/28 What are we doing in the paper? We employ a wide range of liquidity proxies (bond characteristics, trading activity variables and liquidity measures) to explain yield spread (changes) while controlling for credit risk. We examine three different regimes in our sample period which allows as to compare liquidity effects during two periods of crisis (GM/Ford crisis, sub-prime crisis) with a more normal period in between. We analyze investment vs. speculative grade bonds to provide evidence whether liquidity is priced differently in these sub-segments. We use panel regressions and Fama-MacBeth regressions to analyze liquidity effects.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 5/28 Relevant papers on liquidity Impact of liquidity on asset prices Amihud and Mendelson (JFE, 1986) liquidity priced Amihud, Mendelson and Pedersen (FTF, 2006) overview Evidence for corporate bond markets Longstaff, Mithal and Neis (JOF, 2005) reduced-form models Huang and Huang (WP, 2003) structural models Nashikkar, Subrahmanyam and Mahanti (forthcoming JFQA) reduced-form models with bond-level liquidity Bond characteristics and trading activity Fisher (JPE, 1959) first paper on liquidity effects in bonds Elton, Gruber, Agrawal and Mann (JOF, 2001) explain part of the bond yield spread with credit and other factors Edwards, Harris and Piwowar (JOF, 2007) analysis of bond liquidity
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 6/28 Relevant papers on liquidity Liquidity measures Roll (JOF, 1984) Roll measure Amihud (JFM, 2002) Amihud measure Chen, Lesmond and Wei (JOF, 2007) LOT measure Mahanti, Nashikkar, Subrahmanyam, Chacko and Mallik (JFE, 2008) latent liquidity Jankowitsch, Nashikkar and Subrahmanyam (JBF, 2011) price dispersion measure Liquidity studies covering the financial crisis Bao, Pan and Wang (JOF, 2011) focus on Roll measure Dick-Nielsen, Feldhütter and Lando (forthcoming JFE) various proxies
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 7/28 The three hypotheses that we test Hypothesis 1 Liquidity is an important price factor in the US corporate bond market. Amihud and Mendelson (1986) show that investors demand a premium for holding illiquid assets where there is a clientele effect. Duffie et al. (2007) find that liquidity premia are driven by transaction costs due to search frictions, inventory holding costs and bargaining power.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 8/28 The three hypotheses that we test Hypothesis 2 Liquidity effects are more important in periods of financial distress. Duffie et al. (2007) demonstrate that in periods of crisis, liquidity is more important because inventory holding and search costs are higher, and asymmetric information becomes more relevant. Archarya et al. (2009) point out that banks face more stringent capital requirements when holding illiquid assets and access to liquidity is difficult. Sadka (2010) finds that during crises investors may have shorter horizons, e.g. to meet VaR requirements and margin calls. Bao et al. (2011) and Dick-Nielsen et al. (2010) also show that liquidity effects are more important during the sub-prime crisis.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 9/28 The three hypotheses that we test Hypothesis 3 Liquidity effects are more important for bonds with high credit risk. Based on a regime switching model Archarya et al. (2009) show that liquidity is substantially different between investment and speculative grade bonds. Chen et al. (2007) find evidence that in periods of crisis, flight-to-quality effects are expected which result in lower price reactions for investment grade bonds.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 10/28 Data sources Four different data sources Transaction data from TRACE Consensus market valuations from Markit Credit ratings from Standard & Poor s Bond characteristics, swap and Treasury data from Bloomberg Merged data sample Period from Oct 1, 2004 to Dec 31, 2008 3,261 firms 23,703 bonds 691,016 bond-weeks 23.5 mln trades
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 11/28 Set of proxies to capture liquidity Bond characteristics Amount issued Coupon Age Maturity Liquidity measures Amihud measure Price dispersion measure Roll measure Zero-return measure Trading activity variables Number of trades Trade volume Trading interval... expected effect on liquidity
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 12/28 Liquidity measures based on transaction data 09:00 11:00 13:00 15:00 17:00 94 96 98 100 Washington Mutual Inc CUSIP 939322AE3 (Jan 15, 2008) Trade Time Price
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 13/28 Liquidity measures based on transaction data Omnicom Group CUSIP 681919AT3 (Jan 15, 2008) Price 94 96 98 100 09:00 11:00 13:00 15:00 17:00 Trade Time
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 14/28 Liquidity measures based on transaction data Amihud measure Amihud t = 1 N t N t j=1 r j v j, r j... return based on traded prices v j... traded volume N t... number of observations
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 15/28 Liquidity measures based on transaction data Roll measure Roll t = 2 Cov( p j, p j 1 ). Price dispersion measure Price dispersion t = 1 N t Nt j=1 v (p j m t ) 2 v j, j j=1 p j... traded price v j... traded volume m t... market-wide valuation N t... number of observations
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 16/28 Descriptive statistics Q 0.05 Q 0.50 Q 0.95 Mean SD Yield Spread (%) 0.52 1.92 7.67 2.87 2.95 Rating 1.30 7.03 15.46 8.00 4.14 Bond Amount Issued (bln) 0.00 0.20 1.25 0.32 0.50 Characteristics Coupon (%) 3.03 5.97 9.13 5.98 1.87 Maturity (yr) 0.45 5.20 24.87 7.62 7.63 Age (yr) 0.47 2.77 10.36 3.80 3.61 Trading Activity Volume (mln) 0.02 0.39 5.23 1.35 2.53 Variables Trades 1.45 2.46 8.44 3.47 4.50 Trading Interval (dy) 1.50 4.48 7.80 4.46 2.18 Liquidity Amihud (bp per mln) 0.68 38.33 260.68 78.38 137.21 Measures Price Dispersion (bp) 1.69 33.64 106.84 41.53 35.56 Roll (bp) 24.48 155.98 420.86 185.12 144.69 Zero-Return (%) 0.00 0.01 0.16 0.03 0.08
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 17/28 Liquidity effects in corporate bond yield spreads Dependent variable: Yield Spread; panel regression (1) (2) (3) (4) Eco.Sig. (bp) * Intercept 0.073 0.073 0.072 0.072 Lagged Yld.Spr. 0.285 0.283 0.282 0.280 Volume 0.020 0.011 1.8 Trades 0.007 0.005 1.5 Trading Interval 0.007 0.007 2.5 Amihud 0.050 0.048 6.1 Price Dispersion 0.074 0.070 3.4 Roll 0.051 0.051 3.1 Zero-Return 0.077 0.070 0.8 Rating Dummies Yes Yes Yes Yes R 2 0.074 0.077 0.084 0.086 Observations 691,016 691,016 691,016 691,016 *SD of yield spread changes is 75.6 bp
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 18/28 Liquidity effects in corporate bond yield spreads Among all liquidity proxies, the Amihud and the price dispersion measure are most important in terms of their t-statistics and economic significance. Among the trading activity variables, the volume and trading interval are of particular importance. Liquidity measures are more relevant than trading activity variables in terms of relative improvement in R 2. A one SD move of all proxies in the direction of greater illiquidity increases the yield spread by 19.2 bp (SD of spread change is 75.6 bp.) Liquidity effects explain about 14% of the explained market-wide corporate yield spread variation. Hence, liquidity is an important price factor driving yield spread changes.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 19/28 Liquidity effects in periods of financial distress Descriptive statistics GM/Ford Normal Sub-prime Yield Spread (%) 2.34 1.88 5.00 Rating 8.82 8.38 7.63 Traded Bonds (thd) 5.23 5.92 5.19 Market-Wide Trades (thd) 20.43 20.71 22.77 Market-Wide Volume (bln) 7.65 8.06 6.99 Amount Issued (bln) 0.43 0.45 0.54 Coupon (%) 6.26 6.24 6.23 Maturity (yr) 7.57 7.75 8.31 Age (yr) 3.91 4.36 4.76 Volume (mln) 1.51 1.44 1.53 Trades 4.48 4.06 5.33 Trading Interval (dy) 3.31 3.38 3.37 Amihud (bp per mln) 66.48 53.21 89.20 Price Dispersion (bp) 46.36 39.75 70.02 Roll (bp) 164.28 142.82 209.77 Zero Return (%) 0.02 0.02 0.03
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 20/28 Liquidity effects in periods of financial distress Dependent variable: Yield Spread; panel regression... Liquidity Proxies Liquidity Proxies Liquidity Proxies GM/Ford Dummy Sub-prime Dummy Volume 0.014 0.003 0.014 Trades 0.003 0.002 0.004 Trading Interval 0.003 0.004 0.007 Amihud 0.033 0.007 0.027 Price Dispersion 0.042 0.005 0.053 Roll Measure 0.008 0.003 0.078 Zero-Return Markit 0.050 0.024 0.082 Rating Dummies Yes Observations 691,016 R 2 0.101
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 21/28 Liquidity effects in periods of financial distress Dependent variable: Yield Spread; Fama-MacBeth regression GM/Ford Crisis Normal Period Sub-prime Crisis Intercept 1.848 1.444 4.441 Amount Issued 0.254 0.182 0.325 Coupon 0.157 0.114 0.351 Maturity 0.011 0.018 0.060 Age 0.005 0.004 0.043 Volume 0.001 0.011 0.043 Trades 0.045 0.032 0.034 Trading Interval 0.007 0.003 0.006 Amihud 0.086 0.072 0.170 Price Dispersion 0.350 0.274 0.452 Roll 0.072 0.081 0.113 Zero-Return 0.237 0.039 0.612 Rating Dummies Yes Yes Yes R 2 0.591 0.602 0.497 Observations 3,815 3,845 3,187
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 22/28 Liquidity effects in periods of financial distress Trading during periods of crisis was focused on fewer bonds, with a larger number of smaller size trades. We observe a flight-to-quality during the sub-prime crisis, which we do not for the GM/Ford crisis. Liquidity is far more important in times of crisis, particularly during the sub-prime crisis. The economic significance of the liquidity measures more than doubles during the sub-prime crisis. Among the liquidity measures, the Amihud and price dispersion measure are the most promising proxies.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 23/28 Interaction effects between liquidity and credit ratings Spread in % 0 2 4 6 8 10 Investment Grade Speculative Grade Jan 2005 Jan 2006 Jan 2007 Jan 2008
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 24/28 Interaction effects between liquidity and credit ratings Dependent variable: Yield Spread; panel regression Intercept 0.076 Lagged Yld.Spr. 0.303 Lagged Yld.Spr. Spec. Grade Dummy 0.038 Liquidity Proxies Liquidity Proxies Spec. Grade Dummy Volume 0.004 0.012 Trades 0.006 0.002 Trading Interval 0.007 0.010 Amihud 0.046 0.025 Price Dispersion 0.080 0.032 Roll 0.060 0.001 Zero-Return 0.083 0.009 Spec. Grade Dummy Rating Dummies 4.881 Yes R 2 0.095 Observations 637,814
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 25/28 Interaction effects between liquidity and credit ratings In general, trading is focused on the investment grade segment. Higher trading activity in the GM/Ford crisis for speculative grade bonds shuffling of bonds due to clientele preferences. Lower number of trades and bonds are observed in the speculative grade segment in the sub-prime crisis flight-to-quality. The regression analysis shows that bonds with higher credit risk are less liquid and react more strongly to liquidity changes. A one SD move in the direction of greater illiquidity increases the yield spread by 13.8 bp for investment grade bonds vs. 37.6 bp for speculative grade bonds. We find a particularly strong reaction of speculative bonds in the sub-prime crisis.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 26/28 Conclusion Liquidity is an important risk factor for corporate bond pricing. Liquidity effects explain about 14% of the explained market-wide corporate yield spread variation. During periods of crisis the economic impact of the liquidity measures increases significantly (more than doubles in the sub-prime crisis.) More pronounced liquidity effects are seen in the speculative grade segment, particularly in the sub-prime crisis. Results are relevant for pricing, risk management, and regulatory policy.
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 27/28 Important events in the US corporate bond market Mar 16: GM issues profit warning May 5: Downgrade of GM to BB and Ford to BB+ Oct 8: Delphi defaults Jan 23: Ford announces 30,000 layoffs Jul 17: At leat 90% loss of two Bear Stearns hedge funds specialized in sub-prime debt Aug 7: American Home Mortgage defaults Sep 15: Lehman Brothers defaults Sep 25: Washington Mutual defaults Apr 05 Jul 05 Oct 05 Jan 06 GM/Ford Crisis Normal Period Mar 05 Feb 06 Jul 07 Jul 07 Oct 08 Sub-Prime Crisis Jan 09 Jan 09
Motivation Literature Hypothesis Data Proxies Results Conclusion Appendix 28/28 Time-series and cross-sectional regression models Time-series regression model (Panel) (Yield Spread) i,t = α 0 + α 1 (Yield Spread) i,t 1 + β (Rating Dummies) i,t + γ (Trading Activity Variables) i,t + λ (Liquidity Measures) i,t + ɛ i,t Cross-sectional regression model (Fama-MacBeth) Yield Spread i = α 0 + α 1 Rating Dummies i + β Bond Characteristics i + γ Trading Activity Variables i + λ Liquidity Measures i + ɛ i