Quantifying Liquidity and Default Risks of Corporate Bonds over the Business Cycle Hui Chen1
Rui Cui2 1
Zhiguo He3
MIT Sloan and NBER 2
3 4
Konstantin Milbradt4
Chicago Booth
Chicago Booth and NBER
Northwestern Kellogg and NBER
October 2015
Outline
Introduction Model Calibration Decomposition Conclusion
Motivation Default risk only accounts for part of corporate bond spreads œ œ
Longstaff, Mithal, Neis (2005): Aaa/Aa 50%; Baa 70% Structural models with macroeconomic risks mostly target default component of credit spreads: Chen, Collin-Dufresne, Goldstein (2009); Bhamra, Kuehn, Strebulaev (2010); Chen (2010)
This paper: structural model to explain total credit spread œ œ
Introducing time-varying search frictions and macro risks Disciplined by matching default probability, credit- & BA-spread moments cross-sectionally and over the business cycle
Previous literature usually posits additive decomposition of credit spreads into one liquidity and one default component This paper: Model-based decomposition of credit-spreads with interactions between default and liquidity components œ œ
framework for policy evaluation recognition of "credit loss" vs "trading loss"
Literature Overview Default-liquidity interactions: He and Milbradt (2014) Our paper: œ œ
Introduction of macroeconomic risks OTC search market with non-constant holding costs
Structural credit model with macroeconomic risks: œ
Chen et al. (2009); Bhamra et al. (2010); Chen (2010),...
Liquidity frictions in OTC search market: œ
Duffie, Garleanu, Pedersen (2005),...
Empirical liquidity & yield estimations: œ
Edwards, Harris, Piwowar (2007), Bao, Pan, Wang (2009),...
Model Setup
Structural credit model in Leland tradition with OTC bond market Aggregate shocks to parameters: 2-state Markov chain: normal state (G) and recession (B) œ œ
Shocks to price of risk & real production process Jumps in secondary market liquidity frictions
Liquidity constrained investors face holding costs: Holding costs related to uncollateralized borrowing & haircuts
Baseline Model: Leland ’94b
Reissue
Firm: dy!Μdt#ΣdZ
Default at yb : D!Αvb
D: c dt
Maturity m
Idiosyncratic Liquidity Shocks
Reissue
Firm: dy!Μdt#ΣdZ
DH : c dt
Maturity m
Ξ
Liq. Shock
Maturity
Default at yb
DL : !c%hc"dt
Secondary Market for Bond Trading
Reissue
Resale
DH : c dt
A!DH
Firm: dy!Μs dt#Σs dZ
Maturity m
Ξs
Liq. Shock
Interdealer Market
Maturity
Default at yb s : DH !ΑsH vb s DL !ΑsL vb s
DL : !c&hcs "dt
s
Λ Intermediation
B!DL #Β!DH &DL "
Haircuts & Holding Costs Liquidity shocks proxy for significant need for cash Uncollateralized financing at rate r + ¬
Collateralized financing via bond at rate r Bond haircut h (depends on riskiness of collateral) œ œ œ
° ¢ Marginal (flow) benefit of bond as collateral: ¬(1 ° h)P y Marginal (flow) benefit from immediate sale of bond: ¬B(y) Holding cost for illiquid bond: relative to cash
Haircuts & Holding Costs Liquidity shocks proxy for significant need for cash Uncollateralized financing at rate r + ¬
Collateralized financing via bond at rate r Bond haircut h (depends on riskiness of collateral) œ œ œ
° ¢ Marginal (flow) benefit of bond as collateral: ¬(1 ° h)P y Marginal (flow) benefit from immediate sale of bond: ¬B(y) Holding cost for illiquid bond: relative to cash
Where does the haircut come from? This paper: Assumed haircut function h (P) (decreasing in P) that delivers linear holding costs hc (P) = ¬ [N ° P] Chen-He-Milbradt 2016: Microfound h (P) via VaR constraint œ
Utilizing (endogenous) inverse relationship between bond volatility and price
Degrees of freedom
Reissue
Resale
DH : c dt
A!DH
Firm: dy!Μs dt#Σs dZ
Maturity m
Ξs
Liq. Shock
Interdealer Market
Maturity
Default at yb s : DH !ΑsH vb s DL !ΑsL vb s
DL : !c&hcs "dt
Λs Intermediation
B!DL #Β!DH &DL "
Note: Only one set of parameters to explain the ratings cross-section.
Calibration: Secondary Bond Market Bond Price Data: TRACE (2005-12) & FISD (1994-2004) Parameters for cash-flow process and SDF from literature When modeling search frictions, 3 "free" parameters (N, ¬G , ¬B ) Target "Investment Grade" BA spreads in both states and accross ratings (6 moments)
Calibration: Secondary Bond Market Bond Price Data: TRACE (2005-12) & FISD (1994-2004) Parameters for cash-flow process and SDF from literature When modeling search frictions, 3 "free" parameters (N, ¬G , ¬B ) Target "Investment Grade" BA spreads in both states and accross ratings (6 moments)
Model Parameters for Search Friction Symbol
Interpretation
G
ª Ø ∏ N ¬
Liquid shock int. Investor bargaining Intermediation int. Holding cost Holding cost
0.7
B
Justification
1.0
Bond Turnover Feldhutter 2012 Anecdotal
0.05 50
20 115 0.06 0.11
Default prob & credit spreads (10 year bonds) Maturity = 10 years Aaa/Aa
A
Baa
Ba
Panel A. Default probability (%) data model
2.1 1.6
3.4 3.9
7.0 7.9
19.0 15.9
Default prob & credit spreads (10 year bonds) Maturity = 10 years Aaa/Aa
A
Baa
Ba
Panel A. Default probability (%) data model
2.1 1.6
3.4 3.9
7.0 7.9
19.0 15.9
Panel B. Credit spreads (bps) State G data model
61.2 86.0
90.2 122
150 182
303 301
State B data model
106 136
159 185
262 261
454 404
Comparative Statics Credit Spread Maturity = 10 years Aaa/Aa
A
Baa
Ba
Credit spreads (bps) State G model hc = 0 hcs = csts
86.0 32.5 95.3
122 57.5 126
182 103 176
301 200 278
261 143 245
404 248 359
State B model hc = 0 hcs = csts
136 59.5 146
185 90.7 185
Default probability matching deteriorates only slightly
Bid-Ask Spread
State G data model hcs = csts
Superior 40 39 45
Investment 50 50 50
State B Junk 70 61 49
Superior 77 111 123
Investment 125 125 125
Model counterpart is a bond with time to maturity of 8 years (mean time-to-maturity of frequently traded bonds)
Junk 218 186 138
Structural Decomposition Standard View: CDS spread measures “Default component” Bond-CDS spread = Credit spread - CDS spread measures “Non-default component” Our view: Default component: 1. pure default from illiquidity free model (different default policy) 2. residual of liquidity-driven default
Liquidity component: 3. pure liquidity from risk-free bonds with same search frictions 4. residual default-driven liquidity
State G (Repo-Treasury spread ¢ =15 bps) �� ������ ���
��� ������ ���
���� �������
� ����� ��� ���/�� ����� ���
���-������ ������� ���� ��� �������-������ ���
State B (Repo-Treasury spread ¢ =40 bps) �� ���� ���
��� ������ ��� � ����� ��� ���/�� ����� ���
���� �������
���-������ ������� ���� ���
�������-������ ���
Changes in Spreads: G ! B �� ����� ���
���� ������� ��� ����� ��� ���-������ ������� � ���� ���
���� ���
���/�� ����� ��� �������-������ ���
Structural decomposition in time series Panel A. Baa-rating
Panel B. Baa-rating 120
150
pure liquidity def-driven liquidity
basis points
100 80
100
60 pure default liq-driven default
50
40 20
0
95Q1
00Q1
05Q1
10Q1
0
95Q1
basis points
Panel C. B-rating 250
400
200
300
150 pure default liq-driven default
10Q1
pure liquidity def-driven liquidity
100
100 0
05Q1
Panel D. B-rating
500
200
00Q1
50
95Q1
00Q1
05Q1
10Q1
0
95Q1
00Q1
05Q1
10Q1
Comp. Statics: More frequent liquidity shocks Panel A. Baa-rating
Panel B. Baa-rating 120
150
pure liquidity def-driven liquidity
basis points
100 80
100
60 50
pure default liq-driven default
40 20
0
95Q1
00Q1
05Q1
10Q1
0
95Q1
basis points
Panel C. B-rating 250
400
200
300
150 pure default liq-driven default
10Q1
pure liquidity def-driven liquidity
100
100 0
05Q1
Panel D. B-rating
500
200
00Q1
50
95Q1
00Q1
05Q1
10Q1
0
95Q1
00Q1
05Q1
10Q1
Liquidity Provision
Debt holding values Injecting Liquidity cxv Liquidity Improves
Alleviating Rollover Losses
Equity holders delay default
Here: Injecting liquidity means improving dealer contact intensity ∏s and reducing the uncollaterialized borrowing premium, lowering holding costs hcs
Liquidity Provision Policy evaluation: What are the effects of “injecting liquidity” on lowering “borrowing cost” (setting state B liquidity parameters same as in G)? Pure liquidity channel is dominant for higher rated bonds; default-driven liquidity channel important for lower rated bonds.
Credit Spread Aaa/Aa G Aaa/Aa B Ba G Ba B
Contributions to Change (%)
w/o policy
w policy
pure LIQ
LIQ!DEF
DEF!LIQ
71.1 96.0 286 364
41.9 44.4 234 262
83 83 46 42
5 3 12 9
12 14 42 49
Conclusion Tractable structural model that embeds aggregate liquidity effects in a capital structure model Ability to explain total credit spread, i.e., credit risk premium and liquidity premium, cross-sectionally (across rating classes) and over the business cycle (across aggregate states) Holding costs motivated by cost of collateralized vs uncollateralized borrowing (haircuts) Counterfactual analysis reveals sizable benefits of injecting liquidity in bad times through default-liquidity interaction terms.
Jensen’s Inequality: Heterogeneity in the Data Rating class defined by the empirical distribution of market leverages as given by distribution dPrating for each quarter Normal A
Normal Baa
Normal Ba
Recession Aaa/Aa
Recession A
Recession Baa
Recession Ba
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0 8 6 4 2 0
Density
2
4
6
8
Normal Aaa/Aa
Market Leverage
Jensen’s Inequality: Implementation David (2008), Bhamra et al. (2010): map each firm-quarter to its model counterpart, then aggregate within each rating class œ
Match empirical leverage distribution for each rating class
D Each market leverage (ML = D+E ) implies an initial cash-flow state y0 , which then gives model-implied credit-spread, liquidity measure and default probability
Average credit spread for each rating class: csrating =
Z1 0
° ¢ cs y (ML) dPrating
Given rating, we explain average credit spread across firms, not credit spread of an average firm because strong non-linearities
Calibration: Fundamental and Aggregate Shocks X ° ∑(s ,s ) ¢ d§t (s ,s ) = °r (st ) dt ° ¥ (st ) dZtm + e t° t ° 1 dMt t° t §t 0 st 6=st
Model Parameters Symbol
Interpretation
r ¢ ≥P e∑ µP ¥ æm æf m !
Risk-free rate T liq premium Transition Density Jump Risk Premium Cash Flow Growth Risk Price Systematic Vol Idiosyncratic Vol Maturity Intensity Primary issuance cost
G
B
0.05 15 bps 40 bps 0.10 0.50 2 0.50 0.045 0.015 0.17 0.22 0.10 0.11 0.25 0.2 0.01
Justification Data Repo-Treasury spread Chen 2010 Chen 2010 Chen 2010 Chen 2010 Equity vol Baa def. prob Data/ Chen et al 2012 Chen 2010
Calibration: Post-Default Bond Market For counterfactual analysis, we need bond recovery w/o post-default illiquidity. So need ultimate recovery Moody’s Default and Recovery Database covering 1987-2012 Risk adjust: discounting these return with a public market benchmark (SP500) over the same horizon, known as Public Market Equivalent (PME)
Table: Mean Annualized Net PME on Defaulted Bonds
Default Time Non-Recession Recession Full Sample
# Def. Bonds 512 130 642
Net PME 0.3126 0.5537 0.3613
Emergence (Yrs) 1.37 1.31 1.35
ˆ ƈ G = 87.96%, ƈ B = 64.68%. Ultimate recovery rate Æ:
Model-implied Bond-CDS Spread Assume CDS contracts perfectly liquid (doubtful empirically; Bongaerts et al 2011: CDS seller earns liquidity premium) CDS contract with maturity T requires flow payment f s.t. ∑Zmin{ø,T } ∏ £ § Q °rø Q °rt E 1{ø∑T } e LGD (s) = E e f · dt 0
œ œ
ø is the time the firm defaults loss-given-default LGD (s) 2 [0, 1]: face-value p (wlog p = 1) minus recovery value right at default at state s
Bond-CDS spread: Bond credit spread minus CDS spread 0 1 1°e°yT ° ¢ ° ¢ y A y ° r ° f = y ° c @1 ° 1°EQ [e°r(ø^T ) ] r
œ
In Leland ’94 world, for small r and y, (1) par-bonds have zero spread, (2) discount bonds have positive spread, (3) premium bonds have negative spread
Model-implied Bond-CDS Spread (10 year bonds) Sample: 2005-2012, firms with CDS, 5- and 10-year bonds Treasury spread 15bps G, 40 bps B (netted out) Bond-CDS spreads (bps) 10 years Aaa/Aa
A
Baa
Ba
State G Credit Spread Bond-CDS spread
71 48
107 53
167 61
286 61
State B Credit Spread Bond-CDS spread
96 69
145 79
221 92
364 107