identifying search frictions and selling pressures

Transcription

1 selling pressures Copenhagen Business School Nykredit Symposium October 26, 2009

2 Motivation Amount outstanding end 2008: US Treasury bonds $6,082bn, US corporate bonds $6,205bn. Average daily trading volume in 2008: US Treasury bonds $553.1bn, US corporate bonds $14.3bn The OTC corporate bond market is a search-and-bargain market Large dispersion of prices on a given point in time. Understanding the price formation in the cross-section (investor sophistication and bond maturity) and over time gives insights into the impact of illiquidity on bond prices: Important in risk management, portfolio decisions, portfolio liquidation. Corporate bonds studied here, but results applicable to most OTC markets

3 Key results 1. I model how corporate bond prices are affected by search occasional, and test key predictions by structurally estimating the model. 2. Selling pressure: Theory: under selling pressure institutional prices decrease more than retail prices Empirical: Two periods of selling pressure identified, the downgrade of Ford/GM in 2005 and the current crisis 3. The impact of search costs and occasional selling pressure on yields can explain the credit spread puzzle

4 Example 1: Ford 7.45% 07/16/2031 transaction price November 21, 2005 (normal market) transaction price April 5, 2005 (more sellers than buyers) small large small large

5 Example 2: Citigroup Inc 7.25% 10/01/2010 Prices for Citigroup Inc % 10/01/2010 on March 11 and customer buy interdealer costumer sell price :00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 March 11 March 12

6 Example 2: Citigroup Inc 7.25% 10/01/ Prices for Citigroup Inc % 10/01/2010 on March 11 and customer buy, $0 99K interdealer, $0 99K costumer sell, $0 99K customer buy, $ K interdealer, $ K costumer sell, $ K customer buy, $1,000K costumer sell, $1,000K price :00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 March 11 March 12

7 Related papers Related papers are (among others) Search models: Stigler (1961,JPE), Garbade and Silber (1976,JPE), Duffie, Garleanu, and Pedersen (2005,Econometrica), Duffie, Garleanu, and Pedersen (2007,RFS), Vayanos and Wang (2007), Vayanos and Weill (2008,JF), Weill (2008) Corporate bond transaction costs: Hong and Warga (2000), Schultz (2001), Chakravarty and Sarkar (2003), Bessembinder, Maxwell, and Venkaraman (2006,JFE), Edwards, Harris, and Piwowar (2007, JF), Goldstein, Hotchkiss, and Sirri (2007,RFS)

8 Related papers Related papers are (among others) Credit spread puzzle: Huang and Huang (2003,WP), Longstaff, Mithal, and Neis (2005,JF), Blanco, Brennan, and Marsh (2005,JF), Hackbarth, Miao, and Morellec (2006,JFE), Chen, Collin-Dufresne, and Goldstein (2009,RFS), Chen (2009;JF), Bhamra, Kuehn, and Strebulaev (2009,RFS), Ericsson and Renault (2006,JF) Empirical evidence on search: Duffie and Ashcraft (2007), Jostikasthira (2007,WP), Newman and Rierson (2003). Empirical evidence on selling pressure:acharya, Schaefer, and Zhang (2008,WP)

9 Setup Two types of (risk-neutral) agents trade a corporate bond in the economy: investors and dealers An investor is high (no cost) or low (holding cost δ) Interpret holding cost as a liquidity shock Investors switch low to high with intensity λ u and high to low with intensity λ d Investors trade through dealers An investor i finds a dealer with intensity ρ i, i =1,..., N Dealers maximize profits and have no inventory (immediately unload in the interdealer market) The bond has finite maturity T, defaultable with intensity λ D, and pays coupons If there is excess demand for bonds, firms stand ready to issue Setup Prices Illustration Implications

10 Prices If sellers are not constrained, bid and ask prices are A i = ΔVh i z + M(1 z) B i = ΔVl i z + M(1 z) where ΔVl i is the reservation price of seller, ΔVh i is buyer s reservation price, M the interdealer price, and z the bargaining power of the dealer. The interdealer price is between seller s and buyer s reservation price [Let ΔV h = minδv i i h and ΔV l = maxδv i i l then M = qδv h +(1 q)δv l ] Setup Prices Illustration Implications

11 Model search search selling investor dealer immediately dealer buying investor Setup Prices Illustration Implications bargain price B bargain price A bargained prices depend (among others) on outside options of investor (ease of finding counterparties) and number of sellers vs buyers

12 The model is different from that in Duffie, Garleanu, and Pedersen (2005,2007) in two ways 1. Investors trade through a financial intermediary, not directly with each other 2. The traded asset has finite maturity Setup Prices Illustration Implications

13 Bid-ask spreads in equilibrium (Cor. 2.1) In equilibrium, the bid-ask spread for investor i with search intensity ρ i is given as A i B i zδ = λ u + λ d + ρ i (1 z)+r +1/T + λ D Smaller for short maturities Smaller for institutional investors Setup Prices Illustration Implications

14 Prices after a liquidity shock (Th. 2.3) Assume a fraction of s of high investors are hit by a liquidity shock and become low Assume segmented market for institutions and retail investors Prices decrease after a liquidity shock Bid-ask spreads are not affected by a liquidity shock Prices of institutional investors decrease more than prices of retail investors (ask, bid, midprice) Setup Prices Illustration Implications

15 TRACE data Transactions data for almost all U.S. corporate bonds from TRACE Straight coupon bullet bonds October 1, 2004-June 30, 2009 More than 10 million trades

16 Roundtrip costs TRACE does not have buy/sell indicators (only after November 2008). What can we do to extract buys and sells? of roundtrip costs is based on unique roundtrip trades (URT) For a given bond on a given day, if there are exactly 2 or 3 trades for a given volume, and they occur within 15 minutes, they are part of a URT Ask is the largest price and bid is the smallest price in a URT 973,600 URTs in the sample

17 methodology I fit the model to demeaned prices because the goal is to extract information from the cross-sectional variation in prices Any price for a given bond on a given day is demeaned with the average of all prices for this bond on this day Six levels of investor sophistication sorted according to trade size: $0 10K, $10K 50K, $50K 100K, $100K 500K, $500K 1, 000K, and $1, 000K I assume that 1. r = 0.05 (riskfree rate) 2. λ d =0.02 (default intensity) 3. λ u = 1 ( up intensity) 4. λ d =0.1 ( down intensity) 5. δ = 7 (holding costs) 6. z = 0.9 (dealer bargaining power)

18 ML estimation I assume that errors between fitted and actual demeaned prices are normally distributed with mean 0 and variance w tb =min(1, T tb ) σ 2 where T is the maturity of a bond b and day t ɛ i tb (Θ) = Pi tb ˆP i tb (Θ) N(0, w tbσ 2 ) Parameters are estimated by a ML procedure: max Θ L(Θ,σ Y )= T N b 2N tb t=1 b=1 i=1 1 2πσ 2 exp( ɛi tb (Θ)2 2σ 2 ) Equilibrium parameter estimates based on the period October July 2007

19 Estimated parameters Search parameters: ρ 1 ρ 2 ρ 3 ρ 4 ρ 5 ρ 6 69 (0.3) 108 (0.4) 142 (1.6) 208 (4.1) 378 (38.2) 474 (44.3) It takes smallest retail investors almost 4 business days to complete a trade (learn how to trade, keep-up-to-date about information) Large institutional investors complete a trade in half a day Model fit Selling pressure Credit spread puzzle

20 How are actual roundtrip costs fitted? Matches downward slope of bid/ask spreads in trade size 0-5K 6-10K 11-20K 21-50K K fitted (in bps) 73.2 (0.0) actual (in bps) 87.1 (0.4) fitted (in bps) 28.1 (0.0) actual (in bps) 34.9 (0.3) 56.9 (0.0) 68.6 (0.2) 50.7 (0.0) 67.0 (0.2) 48.1 (0.0) 59.3 (0.2) 35.1 (0.0) 49.0 (0.3) K K K >1000K 25.3 (0.0) 22.3 (0.2) 14.2 (0.0) 14.2 (0.2) Matches increase of bid/ask spreads in maturity: fitted (in bps) 26.3 (0.1) actual (in bps) 23.4 (0.6) 0-2m 2m-4m 4-6m 6m-1y 1-3y 3-5y 5-30y (0.1) (0.1) (0.1) (0.0) (0.0) (0.0) 25.1 (0.4) 27.4 (0.4) 32.5 (0.2) 45.7 (0.1) 55.8 (0.2) 76.6 (0.2) 12.8 (0.0) 7.5 (0.1) Model fit Selling pressure Credit spread puzzle

21 Time-variation in selling pressure To estimate periods of selling pressure I do the following. For each month in the sample keep all estimated parameters fixed Estimate the size of liquidity shock 0 s 1 (fraction of high investors shocked) that best fit that month s prices. This is again done by ML as in the initial parameter estimation Remember: Retail vs institutional prices identify shocks Model fit Selling pressure Credit spread puzzle

22 Liquidity shocks Bear Sterns take-over Lehman Brothers default selling pressure credit spreads peak credit crunch (aug. 07) Model fit Selling pressure Credit spread puzzle GM/Ford downgrade

23 Liquidity shocks around GM/Ford downgrade Ford GM Other bonds selling pressure Model fit Selling pressure Credit spread puzzle 10 0 oct04 dec04 feb05 apr05 jun05 aug05 oct05 dec05

24 Impact of search costs and selling pressure on average yields To find the effect of search costs and occasional selling pressure on yields across maturity: I define the search premium as the midyield paid by an average corporate bond investor minus the (mid)yield of an investor with ρ = ( corporate bond market vs Treasury market) For the same average investor I define the selling pressure premium as the midyield on trades occuring under a liquidity shock of 19% (sample average) minus the midyield in absence of a shock Model fit Selling pressure Credit spread puzzle

25 Maturity effect of search and selling on yields premium due to occasional selling pressure premium due to search costs basis points Model fit Selling pressure Credit spread puzzle bond maturity

26 Liquidity premia pre and post credit crunch liquidity premium, October 04-July 07 liquidity premium, August 07-June 09 basis points Model fit Selling pressure Credit spread puzzle bond maturity

27 Conclusion In this paper I model corporate bond prices in a model with search frictions, and find that under selling pressure institutional prices decrease more than retail prices show how to structurally estimate a search model examine the extent of selling pressure during the downgrade of GM/Ford and the subprime crisis show that search frictions explain the credit spread puzzle Conclusion Future research

28 Open questions This paper raises a number of questions: This paper examines aggregate liquidity shocks in the corporate bond market. What about the cross-section? The municipal bond market is even more illiquid than the corporate bond market. How is the nature of liquidity and liquidity shocks in this market? Even in very liquid markets search frictions matter (Duffie and Ashcraft (2007)). Can we understand the Treasury market better through a model with search buying pressure? Conclusion Future research