FRM Exam Part II · Correlation Basics: Definitions, Applications, and Terminology

Empirical Correlation Properties and Terminology for FRM Part II

Updated 11 October 2026 · Fact-checked

Empirical correlation properties are patterns seen in real data. Correlations are not constant: they tend to mean revert, cluster in high or low periods, and rise in market stress. Correlation breakdown means a past relationship fails when you need it. A correlation matrix lists all pairwise correlations. Know each term and its risk implication.

Understand Empirical Correlation Properties and Terminology

Correlation measures how two variables move together, on a scale from -1 to +1. Covariance is the same idea but in raw units, so it has no fixed range. Correlation is covariance scaled by both standard deviations, which makes it comparable across pairs.

In real markets, correlation is not a fixed number. It changes through time. Four patterns come up again and again in FRM Part II. First, mean reversion: correlation drifts back toward a long-run average after moving away from it. Second, correlation clustering: high-correlation periods tend to be followed by high-correlation periods, and low by low, much like volatility clustering. Third, correlations rise in crises: in stressed markets, many assets fall together, so diversification weakens just when you need it. Fourth, correlation tends to be higher when markets fall than when they rise, and in recessions more than in expansions.

Correlation breakdown is the term for a correlation relationship that changes sharply from its historical level, so models built on past data misstate risk. In practice the usual worry is correlations jumping up in stress. It can also mean a hedge relationship that stops working.

A correlation matrix holds the pairwise correlations of n assets. It is square and symmetric, the diagonal is all 1s, and every off-diagonal entry lies between -1 and +1. A valid matrix for risk work must also be positive semi-definite. Risk models use it with volatilities to compute portfolio risk.

Why it matters: a VaR model that uses calm-period correlations will understate loss in a crisis. That is the core risk message the exam wants from you.

Key formulas to remember

Correlation: ρ(X,Y) = Cov(X,Y) ÷ (σX × σY); Unit-free and always between -1 and +1. Defined only when both standard deviations are non-zero.
Covariance from correlation: Cov(X,Y) = ρ × σX × σY; Covariance has units of X times units of Y and no fixed bound.
Two-asset portfolio variance: σp² = w1²σ1² + w2²σ2² + 2 w1 w2 ρ σ1 σ2; Higher ρ gives higher portfolio risk, all else equal. Use it to show the effect of rising correlation.
Correlation matrix properties: Diagonal = 1; ρij = ρji; -1 ≤ ρij ≤ 1; matrix positive semi-definite; Entries chosen one by one can produce an invalid matrix.
Mean reversion (concept): Correlation tends to move toward its long-run mean over time; A tendency, not a guarantee. Speed varies by asset class and period.

How to solve Empirical Correlation Properties and Terminology questions

Use this method for any question on empirical correlation behaviour or terminology.

1Identify the term being tested: mean reversion, clustering, crisis increase, breakdown, covariance versus correlation, or the matrix.
2Recall the plain definition and the direction of the effect.
3If numbers are given, compute covariance or correlation from the formula, or portfolio variance if needed.
4Check the answer is in range: correlation between -1 and +1, variance not negative.
5State the risk consequence: lower diversification, understated VaR, or a failed hedge.
6Eliminate options that claim correlation is constant, always positive in crisis for every pair, or equal to covariance.
7Pick the option that matches both the definition and the stated market condition.

Quickest way: Direction and definition check

When to use it: Conceptual MCQs with no calculation, when time is short.

Crisis or downturn in the stem: expect correlations to be higher and diversification weaker.
Correlation far from its average: expect it to revert toward the mean.
High correlation last period: expect clustering, so high again is more likely.
Any option saying correlation is stable over time is usually wrong.
Covariance has units and no bound. Correlation is unit-free and bounded.

Common mistakes in Empirical Correlation Properties and Terminology

Treating correlation as a constant input.
Textbook portfolio formulas use a single ρ.
Fix: Remember ρ is an estimate that varies with time and market state.
Confusing mean reversion with correlation breakdown.
Both involve correlation changing.
Fix: Mean reversion is a gradual pull toward the average. Breakdown is a sharp departure from the historical relationship.
Confusing correlation with covariance.
Both measure co-movement and share a formula.
Fix: Covariance carries units and is unbounded. Dividing by σX σY gives correlation.
Saying crisis correlations rise for every pair.
Over-generalising the headline rule.
Fix: State it as a general tendency, mostly among risky assets. Some hedges and safe assets can behave differently.
Confusing correlation clustering with volatility clustering.
The names are similar.
Fix: Clustering of correlation means high or low correlation persists. It is a separate property, though similar in spirit.
Assuming any symmetric matrix of values between -1 and 1 is a valid correlation matrix.
Checking only the entries.
Fix: It must also have 1s on the diagonal and be positive semi-definite.

Worked examples

Example 1

Two assets have volatilities of 20% and 30% and a covariance of 0.03. Find the correlation. Then, if correlation rises to 0.8 with volatilities unchanged, find the new covariance.

Show the solution

Correlation = 0.03 ÷ (0.20 × 0.30) = 0.03 ÷ 0.06 = 0.5.
New covariance = 0.8 × 0.20 × 0.30 = 0.8 × 0.06 = 0.048.

Answer: Correlation is 0.5. At ρ = 0.8 the covariance is 0.048.

Example 2

A portfolio holds two assets, 50% each, both with volatility 10%. Compare portfolio volatility at correlation 0.2 in calm markets and 0.9 in a crisis.

Show the solution

Variance = 0.25×0.01 + 0.25×0.01 + 2×0.5×0.5×ρ×0.01 = 0.005 + 0.005ρ.
Calm: 0.005 + 0.005×0.2 = 0.006. Volatility = √0.006 ≈ 7.75%.
Crisis: 0.005 + 0.005×0.9 = 0.0095. Volatility = √0.0095 ≈ 9.75%.
Risk rises by about 2 percentage points with no change in holdings.

Answer: Volatility is about 7.75% in calm markets and about 9.75% in the crisis. Using calm correlation would understate risk.

Exam tips

Link every correlation property to its risk effect: diversification falls and VaR is understated.
Watch wording: gradual drift to the average is mean reversion, a sudden failure is breakdown.
For numeric questions, check the correlation is within -1 to +1 before answering.
Use portfolio variance with a higher ρ to show the effect of stress, and recompute carefully.

Practice questions from Correlation Basics: Definitions, Applications, and Terminology

Empirical Correlation Properties and Terminology: frequently asked questions

Why do correlations increase in a crisis?

Investors sell many assets at once because of losses, margin calls or a flight to safety. Common shocks then dominate individual factors, so assets move together. This reduces diversification when it is needed most.

What is mean reversion of correlation?

It is the tendency of correlation to move back toward its long-run average after rising or falling away from it. It is a tendency, and the speed differs across assets and periods.

How do I interpret a correlation matrix in risk management?

Each entry shows the linear co-movement of a pair. Values near +1 mean little diversification, values near zero mean little or no linear co-movement (the assets are uncorrelated), and negative values mean offsetting moves. Zero correlation does not by itself imply independence, because assets can be uncorrelated yet dependent, for example through a nonlinear relationship or tail dependence. Combine the matrix with volatilities to get portfolio risk.

What is the difference between correlation and covariance?

Covariance measures joint movement in the units of the two variables and has no fixed range. Correlation divides covariance by the product of the standard deviations, giving a unit-free value from -1 to +1.