Recovery Cyclicality and PD-LGD Correlation
Defaults don't just happen more often in recessions — recoveries get worse at the same time, because a downturn that pushes more borrowers into default is also flooding the market with distressed collateral to sell.
Prerequisites: The Credit Triangle: Spread, Hazard Rate and LGD, Probability of Default and Loss Given Default
A simple credit model treats probability of default and recovery rate as two unrelated dials — the chance a borrower defaults, and separately, how much you'd get back if it did. Historical data says that's wrong: the two move together, and in the worst possible direction. Recessions raise default rates and depress recoveries at the same time, because the same economic weakness that pushes borrowers over the edge also collapses the value of what's left to sell.
PD-LGD correlation is the empirical tendency for loss given default to rise precisely when probability of default rises — in recessions, more borrowers default and each default recovers less, because distressed sellers of collateral (real estate, equipment, inventory) all hit the market at once, driving prices down exactly when supply from defaults is highest.
Why the two move together
In a normal year, a defaulted company's assets — plant, inventory, receivables — get sold into a market with reasonably healthy demand, supporting decent recovery values. In a recession, many companies default simultaneously, all trying to sell similar assets into a market where buyers are scarce and credit for buyers is tight. The result is a downturn LGD effect: recovery rates on defaults occurring during recessions are systematically lower than the long-run average recovery rate used in normal-times credit models.
Worked example
A bank's long-run average assumption is a 10-year cumulative default rate of 8% for a loan portfolio, with an average recovery rate of 45% (LGD of 55%), giving an expected loss estimate of of exposure.
During an actual recession year within that 10-year window, defaults spike to double the average annual rate, and — because it is specifically a downturn — the observed recovery rate on those defaults falls to 25% (LGD of 75%) rather than the long-run 45%. Using the stressed, correlated inputs for that single bad year: expected loss becomes — nearly three times the long-run average estimate, driven by both inputs moving against the lender simultaneously, not just one.
What this means in practice
Basel-style regulatory capital frameworks require banks to use a downturn LGD, not a long-run average LGD, when calculating regulatory capital specifically to capture this correlation — using average-times LGD assumptions in stress scenarios systematically understates loss in exactly the scenario capital is meant to protect against.
Modeling default probability and loss given default as independent inputs is a common simplification that understates tail risk. The two are correlated in exactly the scenarios that matter most — systemic downturns — which is why stress-testing frameworks insist on jointly stressed, not independently stressed, PD and LGD assumptions.
Related concepts
Practice in interviews
Further reading
- Altman, Resti and Sironi, 'Default Recovery Rates in Credit Risk Modelling: A Review'