Charta is on Google Play: practise on the go. Get the app
Charta

Outline the usual mistakes made when stress testing CCR.

FRM® Part II · Credit Risk Measurement and Management · The Evolution of Stress Testing Counterparty Exposures · FRM2.CR.21.h

The idea

Frequent mistakes in stress testing CCR come from not treating it as both credit and market risk. Stressing counterparty PDs while holding exposure at today's level, or shocking market factors while holding credit quality constant, understates the joint effect. Treating the 99th percentile PFE as a stress test is another mistake. It is a statistical measure calibrated to normal conditions, not a scenario with stressed correlations and spreads. Others are assuming exposure and default are independent, and so missing wrong-way risk. Treating collateral as fully effective ignores the margin period of risk, haircuts and disputes. Relying on mild or purely historical scenarios is a further error. So is treating CVA hedges as perfect when basis, jump-to-default and exposure changes exist, or counting the DVA gain as an offset. Mixing risk-neutral spread-implied PDs with real-world PDs when estimating default losses is another. A sound test uses one coherent scenario across the whole book.

A practice question

From the question bank, with the reason each wrong option is wrong.

A bank stresses CCR by tripling each counterparty's PD while holding exposures at today's mark-to-market. What is the main flaw?

  1. PDs should never be stressed in a CCR test, only LGDsWhy not: PD is a key driver of loss and should be stressed along with the market factors.
  2. It double counts because exposure already contains the default probabilityWhy not: Exposure is the MTM-based claim and does not contain the PD.
  3. It ignores that the market moves in the scenario would change exposures, so the joint credit and market effect is missed - correct
  4. Exposure should be set to notional under stressWhy not: Notional overstates the claim; exposure is the positive MTM, which should be revalued under the scenario.

CCR is both credit and market risk. A test that shocks PD but freezes exposure, or the reverse, understates the combined loss.

A harder one

A bank's CCR stress test assumes exposure and counterparty default are independent. For which portfolio would this assumption most understate the stress loss?

  1. Protection bought through CDS from a bank headquartered in the same sovereign as the reference entity - correct
  2. Interest rate swaps with a diversified set of counterparties across many unrelated sectorsWhy not: Confuses diversifying across counterparties with dependence between exposure and default of one counterparty.
  3. Trades that are collateralised daily with cashWhy not: Confuses small exposure from daily cash margining with absence of dependence; the independence assumption matters less when exposure is small.
  4. Short-dated trades that mature within three monthsWhy not: Confuses short tenor with lack of correlation; a short tenor limits time but not dependence.

The protection is worth most when the sovereign defaults, which is also when the selling bank is most likely to fail: a clear wrong-way link that independence ignores.

The hardest tier

A risk manager computes default losses in a stress scenario by backing PDs out of stressed CDS spreads and using them as real-world default probabilities. What is the problem?

  1. CDS spreads cannot be stressed in a scenario, only observed in the market, so default losses must come from historical default ratesWhy not: Spreads can be shocked in scenarios, which is standard in CVA stress tests.
  2. Spread-implied PDs are too low in stress because spreads lag actual defaultsWhy not: Spreads normally move ahead of defaults, and risk-neutral PDs usually exceed real-world PDs.
  3. Spread-implied PDs are risk-neutral and include risk premia and liquidity components, so they overstate real-world defaults and suit CVA P&L stresses more than default loss stresses - correct
  4. Spreads include LGD, so PDs cannot be extracted from themWhy not: The credit triangle gives PD from spread given an LGD assumption.

Risk-neutral probabilities from spreads are consistent with pricing CVA, but they are not forecasts of default. Mixing the two measures gives incoherent stress results.

In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.

Start free →

More in The Evolution of Stress Testing Counterparty Exposures