FRM Exam Part II · Empirical Properties of Correlation: How Do Correlations Behave in the Real World?

Correlation Breakdown and Risk Model Implications

Updated 11 October 2026 · Fact-checked

Correlation breakdown is when correlations between assets change sharply in stress, usually rising toward 1 in crashes, so diversification fails. Models that assume constant correlation, such as basic parametric VaR, then understate risk. You solve questions by comparing normal and stressed correlations, recomputing portfolio risk, and judging the model's bias.

Understand Implications for Risk Models and Correlation Breakdown

Correlation measures how two assets move together. Risk models need it to combine individual risks into portfolio risk. The simplest models use one fixed correlation matrix, estimated from past data and applied to all future periods.

Empirical studies show this is wrong. Correlations are not constant. They move over time, tend to mean-revert toward a long-run level, and are higher when markets are volatile or falling. In equities, correlations typically rise in down markets more than in up markets. This is asymmetric behaviour.

Correlation breakdown is the term for the failure of historical correlations in a crisis. Assets that looked diversifying in calm periods fall together. Diversification benefit shrinks exactly when you need it. Portfolio volatility and VaR then exceed what the calm-period matrix predicted.

Care with wording: breakdown means the estimated relationship stops holding, not always that correlation goes to +1. Part of the rise in measured correlation is a statistical effect of higher volatility, but the practical lesson is the same. A constant-correlation model is biased toward understating tail risk.

Risk managers respond in several ways. They stress correlations directly by setting them to higher values, such as 1, in scenarios. They use stressed-period correlation estimates, dynamic models such as EWMA or GARCH-type correlations, and regime-based approaches. They also treat correlation as a source of model risk and report VaR sensitivity to correlation inputs.

Key formulas to remember

Two-asset portfolio variance: σp² = w1²σ1² + w2²σ2² + 2·w1·w2·ρ·σ1·σ2; Portfolio risk rises with ρ. Weights w1 and w2 are portfolio proportions.
Parametric VaR (normal, zero mean): VaR = z × σp × portfolio value; z = 1.645 at 95% and 2.326 at 99% (one-tailed). Higher ρ raises σp and so VaR.
Upper bound on portfolio risk: With ρ = 1: σp = w1σ1 + w2σ2; The no-diversification case. Long-only positions give the maximum risk at ρ = 1.
Diversification benefit: Benefit = (undiversified VaR) − (diversified VaR); Undiversified VaR is the sum of standalone VaRs. Benefit shrinks as ρ rises.

How to solve Implications for Risk Models and Correlation Breakdown questions

Use this method for numerical and conceptual questions on correlation in stress and risk models.

1Identify what the question gives: volatilities, weights or exposures, and the base and stressed correlations.
2Decide whether the position is long-long, or a hedge (offsetting positions). Hedges lose protection when correlation falls; long-only diversified positions lose when correlation rises.
3Compute portfolio variance with the two-asset formula, once with base ρ and once with stressed ρ.
4Take the square root to get σp, then multiply by z and portfolio value if VaR is asked.
5Compare: the increase in VaR is the loss of diversification benefit.
6Interpret: state that a constant-correlation model calibrated to calm data understates tail risk.
7Match any remedy to the problem: stressed correlations, dynamic models, or scenario analysis.

Quickest way: Bound and direction check

When to use it: When options are numerically spread and you have under two minutes.

Compute the two bounds: ρ = 1 gives the sum of weighted volatilities, ρ = 0 gives the square root of the sum of squares.
If the stressed ρ is 1, the answer is the upper bound directly.
Eliminate any option above the ρ = 1 bound for long-only portfolios.
For conceptual items, pick the answer saying correlations rise in stress and constant-correlation models understate risk.

Common mistakes in Implications for Risk Models and Correlation Breakdown

Saying correlations always go to 1 in a crisis.
Textbook shorthand is remembered as a rule.
Fix: Say correlations tend to rise, especially in down markets. Some pairs, such as safe-haven assets, can move the other way.
Treating a hedge like a diversified long portfolio.
Students link higher correlation with higher risk without checking the position signs.
Fix: Check signs first. For a long-short hedge, lower or unstable correlation weakens the hedge.
Forgetting the square root when converting variance to volatility.
Time pressure when computing σp².
Fix: Always write σp = √variance before applying z.
Calling a stressed-correlation VaR a probability statement.
Students keep the 95% or 99% label.
Fix: Stressed scenarios are conditional what-if exercises. They show loss size, not a confidence level.
Thinking longer data history fixes breakdown.
More data seems safer.
Fix: A long calm sample dilutes crisis behaviour. Use stressed-period estimates or dynamic correlation models.

Worked examples

Example 1

A portfolio holds ₹50 crore in Asset A (volatility 20% annual) and ₹50 crore in Asset B (volatility 30%). Base correlation is 0.20. In stress it is 0.90. Compute one-year 99% parametric VaR in both cases (z = 2.326), to the nearest ₹ crore.

Show the solution

Weights are 0.5 each. Terms: w1²σ1² = 0.25×0.04 = 0.0100; w2²σ2² = 0.25×0.09 = 0.0225.
Cross term coefficient: 2×0.5×0.5×σ1σ2 = 0.5×0.06 = 0.03.
Base: variance = 0.0100 + 0.0225 + 0.03×0.20 = 0.0385. σp = √0.0385 = 0.1962.
Base VaR = 2.326 × 0.1962 × 100 crore = ₹45.6 crore, about ₹46 crore.
Stress: variance = 0.0325 + 0.03×0.90 = 0.0595. σp = √0.0595 = 0.2439.
Stress VaR = 2.326 × 0.2439 × 100 = ₹56.7 crore, about ₹57 crore.

Answer: Base VaR is about ₹46 crore; stressed VaR is about ₹57 crore. Higher correlation raises VaR by roughly ₹11 crore.

Example 2

A risk model uses a constant correlation estimated from a calm five-year sample. During a crisis, daily losses repeatedly exceed the 99% VaR. Which explanation and response is most appropriate? (A) Correlations fell, so add more diversification. (B) Correlations rose in stress, so the model understated risk; add stressed correlation scenarios. (C) Volatility fell, so reduce the confidence level. (D) The sample was too short, so only extend the calm window.

Show the solution

Exceptions clustered in a crisis point to understated portfolio risk.
Empirical evidence says correlations rise in stressed, falling markets, cutting diversification.
A constant matrix from calm data misses this, so VaR is too low.
Option A has the direction wrong. C contradicts a crisis. D keeps the calm bias.
The fix is stressed correlations, stressed-period calibration or dynamic models.

Answer: (B)

Exam tips

Always check whether positions are long-only or hedged before deciding if higher correlation is bad.
Memorise the direction: correlations rise in volatile, down markets, and constant-correlation VaR understates tail risk.
In numerical items, compute the ρ = 1 bound first. It often identifies the answer fast.
Know the remedies by name: stressed correlations, dynamic models, scenario analysis, and reporting sensitivity to correlation inputs.
Read wording like 'always' with suspicion; GARP prefers 'tend to'.

Practice questions from Empirical Properties of Correlation: How Do Correlations Behave in the Real World?

Implications for Risk Models and Correlation Breakdown: frequently asked questions

What is correlation breakdown in stress?

It is when historical correlations stop describing how assets move in a crisis. Typically correlations among risky assets rise, so diversification fails when it is needed most. Models using calm-period correlations then understate losses.

Why does the constant correlation assumption fail in VaR?

Empirically, correlations vary over time, mean-revert and rise in volatile, falling markets. A single fixed matrix cannot capture this. VaR built on it usually understates tail risk.

How do you stress test correlations?

Replace the base correlations with higher or stressed values, including 1 in an extreme case, and recompute portfolio loss. You can also use correlations from a past crisis period. Compare results with the base case to see the lost diversification benefit.

Does a higher correlation always raise portfolio risk?

For long-only portfolios of risky assets, yes. For hedged positions with offsetting exposures, the effect depends on the signs, and a hedge can fail when correlation falls or becomes unstable.