Corporate Defaults and Large Macroeconomic Shocks Mathias Drehmann Bank of England Andrew Patton London School of Economics and Bank of England Steffen Sorensen Bank of England The presentation expresses the views and analysis of the author and should not be thought to represent those of the Bank of England or the Monetary Policy Committee members. Mathias Drehmann: Bank of England, SRAD, mathias.drehmann@bankofengland.co.uk 1
Corporate Defaults and Large Macroeconomic Shocks Do non-linearities matter for credit risk? Is the impact proportional to the shock size? Is the impact symmetric in the sing of the shock? Is the impact independent of initial conditions? Does estimation uncertainty matter for credit risk? In some settings upper confidence intervals might matter Why is this interesting? Stress testing forecast of severe but plausible shock Capital setting Pricing 2
Why Should Non-linearities Matter? Potential Problem: Standard models are estimated in (log-) linear form If unknown data generating process (DGP) is truly linear no problem If unknown DGP is not linear and interest lies in forecasting small perturbations around the equilibrium no problem But: Stress testing looks at large shocks and we don t know whether DGP is truly linear problem? 3
Estimation of the Non-linear VAR Standard VAR First order Taylor series approximation Impulse response functions iterate one period model forward We follow Jorda (2005) Estimate first, second and third order approximations (assume cross products = 0) Piece-wise regressions for each horizon by OLS Error bands based on covariance matrix of parameter estimates at each horizon Data for macroeconomic VAR GDP growth, inflation (PPI), nominal interest rate Quarterly data from 1992Q4-2004Q3 4
Macroeconomic Risks and Aggregate Defaults Aggregate liquidation rates (LQR) Use logit transformed already non-linear! Include lagged liquidation rates and squares and cubes of macro variables Simulate error bands based on covariance matrix of parameters and residual variance Related literature BoE work on liquidation rates eg. Benito, et al (2001) Hoggarth, Sorensen and Zicchino (2005) 5
Impulse response functions 1 and 3 std. positive/negative shocks to GDP and interest rates Shocks based on Cholesky decomposition Base case: variables are held at sample average Impulse responses are plotted relative to base case Plot confidence intervals for cubic model 6
Impulse Response Functions for LQR Small shocks (+) Large shocks (+) Int. rate GDP Impact is not proportionally to shock size Considerable estimation uncertainty 7
Impulse Response Functions for LQR (II) Positive shocks (3std) Negative shocks (3std) Int. rate GDP Impact not symmetric across sign of shock Interest rates key risk driver 8
Firm specific PDs Macroeconomic Risks and Firm Specific Defaults Estimate firm specific quarterly probit model already non-linear! Derive quarterly series from annual accounts data (over 30,000 UK companies ) and liquidation rates from 1991-2004 Include firm specific variables (the interest cover, the current ratio, the debt to asset ratio, the number of employees, the profit margin and industry dummies) Include squares and cubes of macro variables Simulate error bands based on covariance matrix of parameters Related literature Bunn and Redwood (2003) Beaver (1966), Altman (1968), Wilson (1997a,b), Shumway (2001), Duffie et al (2005), Campell et al (2005) 9
Impulse Response Functions for PDs Small Shocks (+) Large Shocks (+) 1.4 +1 sd shock to GDP 1.4 +3 sd shock to GDP GDP 1.0 0.6 1.0 0.6 Int. rate 0.2 1.50 +1 sd shock to interest rate 1.25 1.00 0.2 1.50 +3 sd shock to interest rate 1.25 1.00 0.75 0.75 Cubic model Lower bound cubic model Upper bound cubic model Linear model Impulse responses more precisely estimated Linear model may overestimate corporate credit risk 10
Impulse Response Functions for PDs (II) 1.4 Positive Shocks (3std) +3 sd shock to GDP Negative Shocks (3std) 1.25-3 sd shock to GDP GDP Int. rate 1.0 0.6 0.2 1.25 1.00 1.50 +3 sd shock to interest rate 0.75 1.00 0.75 0.50 1.1-3 sd shock to interest rates 1.0 0.9 0.8 0.7 Cubic model Lower bound cubic model Upper bound cubic model Linear model Impact is not symmetric 11
Impulse Responses and Initial Values 2003 conditions (+3std) 2% interest rate (+3std) Int. rate 1.4 +3 std shock to interest rate - relative to 2003 conditions 1.3 1.2 1.1 1.0 0.9 1.4 +3 std shock to interest rate - relative to interest rate at 2% 1.3 1.2 1.1 1.0 0.9 Initial conditions matter Large interest rate shocks have larger impact on corporate defaults when interest rates are low 12
Distribution of PDs +3std to interest rates (2003, lag 2) 0.40 +3 SD shock to interest rate, Lag 2 0.35 0.30 Max PD distribution 97.5 critical value PD distribution Base 0.25 0.20 0.15 0.10 0.05 0 9 10 11 12 Estimation uncertainty matters 13
Next steps Robustness checks: Dependence on sample period Structural break? Explore the level effect Focus on company specific PDs Check robustness of firm specific variables Derive full 1 year ahead loss distributions Explore pricing and capital setting implications 14
Conclusion Do non-linearities matter for credit risk? Is the impact proportional to the shock size? Is the impact symmetric in the sing of the shock? Is the impact independent of initial conditions? Does estimation uncertainty matter for credit risk? NO NO NO YES stress testing / capital setting Linear models may overestimate corporate credit risk Pricing 15