Adrien Verdelhan
MIT Sloan
April 2013
Optimal Sovereign Debt and Default Workhorse Model Benchmark Empirical Spreads
Sovereign Risk Premia Portfolios of EMBI Excess Returns Cross-sectional Asset Pricing Model
Literature
if government defaults:
it is excluded from international capital markets for a stochastic number of periods, faces a direct output cost YtB ,def .
Endowment: borrowers
Arellano (2008): Yti = e zt Aguiar and Gopinath (2007): Two components: Yti = e zt it
i, a transitory component zt
i i
zti it gti
g ,i
= z (1 z ) + z zti 1 + = Gti it 1 ,
z ,i t g ,i t .
and
z ,i
= 0.
Investors
Lenders are:
risk neutral, behave competitively, supply any quantity of funds demanded at a price Qt ,
Lenders maximize expected prots QB + 1 1+r B , where denotes the expected default probability (zero recovery in case of default).
Recursive Equilibrium
State variables:
x = [y i , s ] , Markovian with conditional density f (x , x ). B is the quantity of debt coming to maturity.
Value of the option to default or stay in the contract is: v o (B , x ) = max {v c (B , B , x ), v d (x )} Value of default is: v d (x ) = u (y def ) +
x
250
200
150
100
50
0 1992
1994
1996
2004
2006
2008
3000
ECU
2500
EMBI G Spread (basis points)
2000
BEL
1500
VEN ARG PAK UKR
1000
DOM KAZ
500
BUL RUS CHIL CHI MAL POL SOU MEX COL HUN PAN BRA TUN PER EGY
EL VIE
LEB
BBB
BB S&P Ratings
CCC
CC
Then expected excess returns should be zero: Not true in the data!
We can express the bond price as: Pt = Et (Xt +1 ) + covt (Mt +1 , Xt +1 ), Rtf
For a given identical default probability, Mexican sovereign bonds are more risky than Thai bonds for the US investor.
Study:
sovereign debt issued by emerging and developing countries in US dollars.
Findings
Large cross section of average holding period excess returns between countries with low and high probability of default and low and high connection to the US:
Use Standard and Poors credit ratings to measure default probability; Use bond market betas to measure connection to the US.
A model with optimal sovereign borrowing and default and external habit preferences replicates qualitatively these results.
Pure Peso explanations of excess returns unlikely in simulated data; Risk premia link countries: countries default more frequently when investors risk aversion is high.
Key Mechanism
Investors risk aversion aects debt quantities and prices:
Take the example of countries with business cycles positively correlated to the US These countries are riskier (i.e tend to default in bad times) higher excess returns They are more likely to default when investors risk aversion is high
Every period, tradeo between paying back debt and borrowing again vs default If investors risk aversion is high, risk premia are high, and so are borrowing costs Borrowing again is less attractive, and thus defaults are more likely.
r e ,m log total excess return on MSCI US equity index Betas are computed on 200-day rolling windows iEMBI ,t .
Data from Standard and Poors. Start from S&Ps credit ratings for foreign currency denominated long-term debt. Convert letter grade credit ratings into numerical index (1 to 23, a higher number means a higher probability of default).
EMBI Portfolios
Countries are ranked on two dimensions:
their default probabilities, their bond market betas.
At the end of each month t , sort all countries into 2 groups on the basis of their iEMBI ,t (information up to date t ). Within each group, sort countries into 3 portfolios on the basis of their default probabilities at the end of each month t .
,j Compute the log excess return rte+ 1 for each portfolio j = 1, 2, ..., 6 by averaging: 1 ,j ,i rte+ rte+ 1 = 1. Nj i Pj
7.15 1.52
15.22 1.47
0.11 0.05
0.16 0.09
0.30 0.08
0.33 0.09
Annualized monthly excess returns. Monthly Data. Higher S&Ps credit ratings is higher default probability. Sample period is 1/1995 5/2011.
Spreads between high and low default probability countries: 470 bp on average.
Spreads between high and low bond market beta countries: 570 bp on average.
where M is the stochastic discount factor of US investors. M is linear in the pricing factors f : Mt +1 = 1 b (ft +1 ), where b is the vector of factor loadings. Candidate risk factors:
returns on US stock market; returns on US corporate bond market: USBBB .
No Arbitrage Restrictions
No arbitrage implies:
e Risk Factor = E [RRisk Factor ]
6
14
Actual Mean Excess Return (in %)
6
14
Actual Mean Excess Return (in %)
12 10
12 10
4
8 6 41 2
4
8 6 4 2
12
10
15
10
15
Risk Prices
R2 84.38
RMSE 1.31
54.08 60.38
Mean
s? Signicant s?
Portfolio 1 2 3 4 5 6 All j 0 (%) jUS BBB 0.76 [0.09] 0.85 [0.12] 0.87 [0.13] 1.06 [0.16] 1.31 [0.20] 1.84 [0.38] R 2 (%) 38.93 32.14 21.31 29.44 27.07 28.93 5.79 44.69 2 ( ) p value
0.17 [0.12] 0.20 [0.19] 0.02 [0.23] 0.03 [0.25] 0.10 [0.30] 0.05 [0.50]
zt is the CBOE volatility index VIX and summarizes all the information set of the investor.
The market prices of risk associated with the USBBB factor increase in bad times, when the implied US stock market volatility is high.
bUSBBB
R2 90.48
RMSE 3.39
2.52 3.90
Mean
Macro Implications?
Cost of debt, borrowing and default choice are all endogenous. We need a model of optimal sovereign borrowing and default:
N-1 small open economies (borrowers), 1 large developed economy (US, lender), Framework a la Eaton and Gersovitz (1981), Aguiar and Gopinath (2006) and Arellano (2008), But:
Lenders are risk-averse: external habit preferences as in Campbell and Cochrane (1999). The borrowers endowment is composed of a transitory component and a time-varying mean. One source of heterogeneity among borrowers: countries dier in their correlation with respect to the US business cycle.
Maximum spreads between high and low-correlation countries: 2c r e 2 2 1.5% 13% 80bp .
Matching the data with CRRA implies a very high risk-aversion coecient and implausible risk-free rates.
External habit implies that risk aversion is time-varying and higher in bad times.
Endowment: lenders
Emerging countries only dier according to their conditional correlation to the developed economy: E(
zi
) = i .
Investors
Lenders are:
risk averse, behave competitively, supply any quantity of funds demanded at a price Qt , have external habit preferences over their consumption endowment described by: (Ct Ht )1 1 Et ( ) t . 1 t =0 Ht is the external habit level and corresponds to a time-varying subsistence level or social externality.
If lenders were risk neutral, there would be no role for covariances in sovereign bond prices.
Model Parameters
Variable Lenders Risk-aversion Mean consumption growth (%) Standard deviation of consumption growth (%) Persistence of the surplus consumption ratio Mean risk-free rate (%) Borrowers Endowment Permanent: Persistence Permanent: Standard deviation (%) Permanent: Mean (%) Temporary: Persistence Temporary: Standard deviation (%) Temporary: Mean (%) Preferences Risk-aversion Discount factor Direct default cost (%) Probability of re-entry (%) g rf 2.00 1.89 1.50 0.87 1.00 Notation Value
g g g z z z
0.20 4.00 2.31 0.90 2.00 Var (z )/2 2.00 0.80 4.00 10.00
Results
A model with both large sovereign debt levels and large spreads:
time-varying mean growth rates incentives to borrow (e.g., good times ahead, but want to consume now) large debt But, sometimes, ... unexpected bad news (e.g., negative temporary shocks) defaults large spreads
Defaults and bond prices depend not only on the borrowers, but also on the lenders economic conditions
Default Frontiers
Default Policy Sets (Defaults occur in the area below the frontier)
Low mean growth rate (g), high riskaversion (low s) Low mean growth rate (g), low riskaversion (high s) High mean growth rate (g), high riskaversion (low s) High mean growth rate (g), low riskaversion (high s)
0.06
0.04
Endowment shock z
0.02
0.02
0.04
0.06 0.5 0.45 0.4 0.35 0.3 0.25 Debt level B today 0.2 0.15 0.1
Impact of riskaversion 1
low z low z, no r.p. high z high z, no r.p.
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
Bond price Q
0.5
Bond price Q
0.3 0.2 Debt level B tomorrow 0.1
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0 0.4
0 0.4
0.1
Simulated Data
Generate simulated data for 36 small open economies. The only source of heterogeneity across countries is the correlation of their endowment process with the US endowment.
Correlations range from -0.5 to 0.5.
Simulation Results
Reasonable consumption and output dynamics, and large debt/GDP ratios A cross-section of sovereign bond excess returns...
smaller than in the data (180 bp vs 500 bp)
reminder: only one-quarter bonds, no term premium
Implications
Monetary unions:
More synchronized business cycles higher sovereign risk premia.
Conclusion
Empirically, countries with higher bond betas oer higher returns.
Timing of expected sovereign payos matters. Low payos in bad times = risky sovereign bonds.
Models of optimal borrowing and defaults with risk neutral investors cannot account for our empirical ndings. Investors risk aversion aect optimal debt quantities and prices.
Habit preferences lead to a cross-section of sovereign bond excess returns. Riskier countries are more likely to default when investors risk aversion is high.