Correlation Risk and Copulas

In 2007 a portfolio of senior CDO tranches was priced under Gaussian copula correlations near 0.30. By autumn 2008 those correlations approached 1.0 across nearly every reference name. The same model, the same portfolio, gave a VaR of single-digit millions before the breakdown and triple-digit millions after. The exam tests whether you understand correlation as a number that moves under stress, not a fixed parameter.

Correlation risk shows up across every desk:

• Trading book: A long-short portfolio's VaR depends on the correlation between the legs. Falling correlation can weaken the offset between long and short legs; rising correlation generally increases risk for...
• Credit book: Default correlation drives CDO and basket-credit pricing. Senior tranches are short correlation; equity tranches are long.
• Counterparty risk: Wrong-way risk is correlation between counterparty default and portfolio exposure. Right-way risk is the reverse.
• Multi-asset options: Quantos, basket options, best-of/worst-of structures all have correlation embedded in their prices.

Correlation is not a stable parameter. It moves with regime, sector concentration, leverage, and liquidity conditions. Risk management treats it as a stochastic variable, not a constant.

Common mistakes

• Treating correlation as a stable parameter. Correlations move with regime. Calm-period dependence may not describe stress. Use the specified scenario or supporting evidence to stress correlation and inspect signed exposures; there is no universal replacement value.
• Confusing asymptotic and finite-tail dependence. A nondegenerate Gaussian copula has zero asymptotic tail dependence but can still generate substantial joint losses at finite thresholds. A t or Clayton copula may better represent persistent tail dependence; observed joint losses alone do not uniquely select a copula.
• Confusing correlation with dependence. Correlation captures linear dependence; copulas capture full dependence including tail behavior. Two distributions can have correlation 0 but still be highly dependent (e.g., $Y = X^2$ has zero correlation but perfect dependence). Trap: a question reports $\rho = 0$ and asks if the assets are independent: not necessarily; check the copula.

Bottom line

• Financial correlation risk is the chance asset correlations move adversely; stress can weaken diversification or hedges, and signed exposures determine the adverse correlation move.
• Empirical correlation properties: correlations are regime-dependent and mean-reverting; the reading reports Johnson SB fits for equity/default-probability correlations and GEV for bonds.
• Copula functions separate marginal distributions from joint dependence; Sklar's theorem decomposes any joint distribution into marginals plus a copula.
• Tail dependence by family: nondegenerate Gaussian copula has zero asymptotic tail dependence, finite-df Student-t is symmetric, Clayton is lower-tail, Gumbel is upper-tail.

Exam shortcut

Distinguish zero asymptotic tail dependence from zero joint loss probability. Compare copulas using the specified marginals, calibration and tail behavior. Choose stress correlations from the scenario or evidence, not a universal 0.7-0.9 rule. Signed exposures determine which correlation move is adverse.

The full lesson (about 3,477 words, 23 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.

Learning objectives

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18

Browse all free FRM Part II lessons or jump into free FRM Part II practice questions.