FRM Exam Part II · Portfolio Credit Risk

Default Correlation and Copula Models for FRM Part II

Q: What is the difference between default correlation and asset correlation?

Default correlation is the correlation between two binary default indicators over a horizon. Asset correlation is the correlation between the underlying asset values or latent variables that drive default in a structural model. Default correlation is usually far smaller for low PDs.

Q: How do I calculate joint default probability?

Multiply the two PDs, then add default correlation times the square root of the product of each PD(1 − PD). If you have asset correlation instead, use a bivariate normal with thresholds N⁻¹(PD) for each name.

Q: Why is the Gaussian copula criticised?

It has no tail dependence, so it tends to understate how often many names default together in a crisis. It also relies on a single, stable correlation input, which tends to rise under stress.

Q: What is a copula in simple terms?

A copula is a function that joins separate marginal distributions into one joint distribution. It carries only the dependence structure. In credit it links individual default probabilities or default times.

Updated 11 October 2026 · Fact-checked

Default correlation measures how much two borrowers' defaults move together. Joint default probability = PD_A × PD_B + ρ × √[PD_A(1 − PD_A) × PD_B(1 − PD_B)]. A Gaussian copula links each borrower's marginal default probability to a joint distribution using a correlation parameter, usually asset correlation.

Understand Default Correlation and Copula Models

A portfolio loses money in a bad way when many borrowers default together. Each borrower's probability of default (PD) tells you about one name. It says nothing about joint behaviour. You need a dependence measure.

Default correlation is the Pearson correlation between two default indicators. Each indicator is 1 if the borrower defaults over the horizon and 0 if not. It is usually small. For PDs of 1% to 2% it is often a few percent, even when the underlying firms are strongly linked. Reason: the indicators are binary and rare, so the maximum achievable correlation is well below 1 unless the PDs are equal.

Asset correlation is different. It is the correlation between the firms' asset values (or the latent variables driving default) in a structural model such as Merton. A default happens when the asset value falls below a threshold. Asset correlation is usually much larger than the default correlation that results from it. Do not swap them in an answer.

A copula separates a joint distribution into two parts: the marginal distributions and a dependence structure. In credit, you fix each name's marginal default time or default probability from CDS spreads or ratings. The copula then ties them together. In the Gaussian copula, you map each marginal to a standard normal variable and give those variables a joint normal distribution with correlation ρ. A name defaults when its normal variable falls below N⁻¹(PD).

The Gaussian copula is simple and needs few parameters, which is why it spread through CDO pricing. Its weakness is no tail dependence: extreme joint defaults become unlikely as you go further into the tail, whatever ρ you choose. In a crisis, defaults cluster more than it predicts. Alternatives such as the Student-t copula add tail dependence, and Clayton copulas add lower-tail dependence. Correlation is also not stable. It rises in stress, so a single calibrated ρ can understate risk.

Key formulas to remember

Default correlation: ρ_D = [P(A∩B) − PD_A × PD_B] ÷ √[PD_A(1 − PD_A) × PD_B(1 − PD_B)]; P(A∩B) is the joint default probability over the same horizon. Result lies between −1 and 1 but is capped by the PDs.
Joint default probability: P(A∩B) = PD_A × PD_B + ρ_D × √[PD_A(1 − PD_A) × PD_B(1 − PD_B)]; With ρ_D = 0 it reduces to independence: PD_A × PD_B.
Gaussian copula joint default: P(A∩B) = M₂(N⁻¹(PD_A), N⁻¹(PD_B); ρ); M₂ is the bivariate standard normal CDF; ρ is the asset (latent) correlation, not default correlation.
Default threshold: Default if Z ≤ N⁻¹(PD); Z is the standard normal latent variable for the name.
Conditional PD, one-factor Gaussian copula: PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]; M is the common factor, standard normal. A low M (bad economy) raises conditional PD.
Latent variable, one-factor model: Z_i = √ρ × M + √(1 − ρ) × ε_i; M and ε_i are independent standard normals. Pairwise correlation between Z_i and Z_j is ρ.
Bounds on default correlation: Maximum occurs when P(A∩B) = min(PD_A, PD_B); Equal PDs allow a maximum of 1; unequal PDs cap it below 1.

How to solve Default Correlation and Copula Models questions

Use this order for any default correlation or copula question. It keeps the inputs and the correlation type straight.

1Identify what is given: PDs, horizon, and which correlation is quoted (default correlation or asset correlation).
2If default correlation is given, use the joint default formula directly with the two PDs.
3If asset correlation is given, treat it as the Gaussian copula parameter. Convert each PD to a threshold with N⁻¹(PD).
4For a one-factor model, set the common factor M and compute conditional PD using the formula, or use the bivariate normal for the unconditional joint probability.
5Check the answer: joint probability must not exceed the smaller PD, and with positive correlation it should exceed PD_A × PD_B.
6Compute any derived quantity asked for, such as conditional probability P(B | A) = P(A∩B) ÷ PD_A, or portfolio loss.
7Interpret in words: positive correlation fattens the loss tail, and a Gaussian copula may still understate tail clustering.

Quickest way: Joint default from default correlation in under a minute

When to use it: Use when the question gives two PDs and a default correlation and asks for joint default probability or conditional default.

Compute the independent part: PD_A × PD_B.
Compute the spread term: √[PD_A(1 − PD_A) × PD_B(1 − PD_B)].
Multiply the spread term by ρ_D and add to the independent part.
For conditional PD of B given A, divide the joint probability by PD_A.
Eliminate options that exceed min(PD_A, PD_B) or sit below PD_A × PD_B when ρ is positive.

Common mistakes in Default Correlation and Copula Models

Using asset correlation in the default correlation formula.
Both are called correlation and often appear in the same question.
Fix: Check the wording. Asset or latent correlation goes into the copula. Default correlation goes into the covariance-style formula with PDs.
Assuming default correlation equals asset correlation.
Students expect the correlation to carry through unchanged.
Fix: Remember that default correlation is much smaller for low PDs. Asset correlation of 30% might give a default correlation of only a few percent.
Forgetting the (1 − PD) terms in the square root.
Students recall the form of a standard deviation but not the Bernoulli variance.
Fix: Each default indicator has variance PD(1 − PD). Write it out before taking the root.
Claiming the Gaussian copula has tail dependence.
A high ρ seems to imply joint extremes.
Fix: The Gaussian copula has zero tail dependence for ρ < 1. The Student-t copula has positive tail dependence.
Getting a joint probability larger than one of the PDs.
Arithmetic slips in the spread term or the wrong sign.
Fix: Sanity check: joint probability ≤ min(PD_A, PD_B). If it breaks, recompute.
Treating a copula as a model of marginal default risk.
The copula is presented as the whole credit model.
Fix: The copula only supplies dependence. Marginal PDs come from ratings, structural models or CDS-implied hazard rates.

Worked examples

Example 1

Two borrowers have one-year PDs of 2% and 5%. The default correlation is 0.10. Find the joint default probability and the probability that B defaults given that A defaults.

Show the solution

Independent part: 0.02 × 0.05 = 0.001.
Variance terms: 0.02 × 0.98 = 0.0196; 0.05 × 0.95 = 0.0475.
Product = 0.0196 × 0.0475 = 0.000931. Square root ≈ 0.030512.
Spread term: 0.10 × 0.030512 = 0.0030512.
Joint probability = 0.001 + 0.0030512 = 0.0040512, about 0.405%.
Check: 0.405% is below min(2%, 5%) = 2%, and above 0.1%. Good.
Conditional: P(B | A) = 0.0040512 ÷ 0.02 = 0.2026, about 20.3%.

Answer: Joint default probability ≈ 0.405%; P(B defaults | A defaults) ≈ 20.3%, versus 5% unconditionally.

Example 2

In a one-factor Gaussian copula, a loan has PD = 1% and asset correlation ρ = 0.20. Using N⁻¹(0.01) = −2.326, find the conditional PD if the common factor M = −2. Use N(−2.0) ≈ 0.0228... as a guide where helpful: N(−1.43) ≈ 0.0764.

Show the solution

Formula: PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)].
√0.20 = 0.4472; √0.80 = 0.8944.
Numerator: −2.326 − (0.4472 × −2) = −2.326 + 0.8944 = −1.4316.
Divide: −1.4316 ÷ 0.8944 = −1.600.
N(−1.60) ≈ 0.0548.
So conditional PD ≈ 5.5%, up from 1% unconditionally.

Answer: Conditional PD ≈ 5.5% when the common factor is two standard deviations below its mean.

Exam tips

Read which correlation the question gives before choosing a formula. This is the most common trap.
Memorise the one-factor conditional PD formula. Questions often supply N⁻¹ and N values, so you only need to plug in.
For conceptual options, link Gaussian copula to no tail dependence and Student-t copula to tail dependence.
Use the bound check: joint probability cannot exceed the smaller PD.
Expect interpretation items: why correlation rising in stress breaks diversification and CDO tranche assumptions.

Practice questions from Portfolio Credit Risk

Default Correlation and Copula Models: frequently asked questions

What is the difference between default correlation and asset correlation?

Default correlation is the correlation between two binary default indicators over a horizon. Asset correlation is the correlation between the underlying asset values or latent variables that drive default in a structural model. Default correlation is usually far smaller for low PDs.

How do I calculate joint default probability?

Multiply the two PDs, then add default correlation times the square root of the product of each PD(1 − PD). If you have asset correlation instead, use a bivariate normal with thresholds N⁻¹(PD) for each name.

Why is the Gaussian copula criticised?

It has no tail dependence, so it tends to understate how often many names default together in a crisis. It also relies on a single, stable correlation input, which tends to rise under stress.

What is a copula in simple terms?

A copula is a function that joins separate marginal distributions into one joint distribution. It carries only the dependence structure. In credit it links individual default probabilities or default times.