FRM Exam Part II · Financial Correlation Modeling - Bottom-Up Approaches

Default Correlation and Bottom-Up Credit Models Explained

Updated 11 October 2026 · Fact-checked

Default correlation measures how likely obligors are to default together. Bottom-up models build portfolio loss from each obligor's default probability, exposure, recovery and a dependence link. In the one-factor Gaussian copula, a common factor drives joint defaults. Joint probability = p1 × p2 + ρ × σ1 × σ2, where σ = √(p(1 − p)).

Understand Default Correlation and Bottom-Up Credit Models

A bottom-up model starts with single obligors. For each one you estimate a probability of default, an exposure and a recovery. Then you link them with a dependence structure and add up the losses to get the portfolio loss distribution. A top-down model skips this and models the portfolio loss or an index directly, without tracking each name.

Default correlation is the correlation between the default indicators of two obligors. A default indicator is 1 if the obligor defaults in the horizon and 0 if not. It is a Pearson correlation of two Bernoulli variables. Default correlations are usually small, often a few percent, but they matter a lot. They fatten the tail of the portfolio loss distribution. Higher correlation means more years with very few defaults and a few years with many.

The structural approach (Merton style) says an obligor defaults when its asset value falls below a threshold tied to its debt. If the asset values of two firms are correlated, their defaults are correlated. The one-factor Gaussian copula makes this practical. Each obligor's standardised asset return is Xi = √ρ × M + √(1 − ρ) × Zi, where M is a common systematic factor and Zi is an idiosyncratic factor. Both are independent standard normals. Obligor i defaults if Xi < N⁻¹(PDi).

Given the value of M, defaults are independent. This is the key trick. The conditional default probability is N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]. A bad market (low M) raises it. Integrating over M gives the unconditional distribution. For a large homogeneous portfolio this leads to the Vasicek loss distribution.

Asset correlation ρ in the copula is not the same as default correlation. Default correlation is lower than asset correlation for the same pair, and it depends on the PDs and the horizon. Do not mix them up in the exam.

Key formulas to remember

Joint default probability: P(A and B default) = pA × pB + ρD × √[pA(1 − pA)] × √[pB(1 − pB)]; ρD is the default correlation. If ρD = 0, defaults are independent.
Default correlation: ρD = [P(A and B) − pA × pB] ÷ {√[pA(1 − pA)] × √[pB(1 − pB)]}; Correlation of two default indicators over the same horizon.
One-factor Gaussian copula: Xi = √ρ × M + √(1 − ρ) × Zi; M and Zi are independent standard normals. ρ is the asset correlation.
Default threshold: Obligor i defaults if Xi < N⁻¹(PDi); N⁻¹ is the inverse standard normal CDF.
Conditional PD: PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]; Defaults are independent once M is fixed. A low M means stress.
Portfolio expected loss: EL = Σ (PDi × LGDi × EADi); Expected loss does not depend on correlation. Unexpected loss does.
Conditional default probability given another default: P(B | A) = P(A and B) ÷ P(A); Useful for contagion-style questions.

How to solve Default Correlation and Bottom-Up Credit Models questions

Use this method for any question on default correlation, copula joint defaults or bottom-up loss.

1Identify what is asked: joint probability, default correlation, conditional PD, or portfolio loss.
2List the inputs: PDs, horizon, LGD, EAD, and whether the given correlation is a default correlation or an asset correlation.
3If the correlation is a default correlation, use the Bernoulli formula with σ = √[p(1 − p)] for each name.
4If the correlation is an asset correlation, convert PDs to thresholds with N⁻¹, then use the one-factor structure or the conditional PD formula.
5Check the horizon. All PDs and correlations must refer to the same period.
6Compute the result and apply LGD and EAD only after the default probability is settled.
7Sanity check: the joint probability must be at most the smaller PD, and must exceed the independent product when correlation is positive.
8Interpret: say what higher correlation does to tail loss, unexpected loss and economic capital.

Quickest way: Fast joint default calculation

When to use it: When you are given two PDs and a default correlation and asked for the joint default probability or a conditional probability.

Compute the independent product pA × pB.
Compute σA = √[pA(1 − pA)] and σB likewise.
Add ρD × σA × σB to the product.
For a conditional probability, divide the joint by pA.
Eliminate options that violate the bounds: joint ≤ min(pA, pB) and joint > pA × pB if ρD > 0.

Common mistakes in Default Correlation and Bottom-Up Credit Models

Treating asset correlation as default correlation.
Both are called correlation and both appear in the same chapter.
Fix: Check the wording. Asset correlation goes into the copula. Default correlation goes into the Bernoulli joint formula. Default correlation is usually lower.
Using σ = p(1 − p) without the square root.
Students confuse the variance of a Bernoulli variable with its standard deviation.
Fix: Variance is p(1 − p). Standard deviation is √[p(1 − p)]. The formula needs standard deviations.
Thinking higher correlation raises expected loss.
Students link correlation with more risk in general.
Fix: Expected loss is a sum of individual terms and ignores correlation. Correlation widens the distribution and raises unexpected loss and tail loss.
Forgetting that defaults are independent only conditional on M.
The copula is introduced as a dependence tool, so the conditional independence step gets skipped.
Fix: Fix M first, treat defaults as independent, then integrate over M.
Mixing horizons, such as a one-year correlation with a five-year PD.
Data in the question arrives in different forms.
Fix: Put every input on the same horizon before computing.
Confusing bottom-up with top-down.
Both produce a portfolio loss number.
Fix: Bottom-up models name each obligor and aggregate. Top-down models work on the aggregate loss or an index without obligor detail.

Worked examples

Example 1

Two obligors have one-year default probabilities of 4% and 9%. Their default correlation is 0.10. What is the joint default probability?

Show the solution

Independent product = 0.04 × 0.09 = 0.0036.
σA = √(0.04 × 0.96) = √0.0384 = 0.19596.
σB = √(0.09 × 0.91) = √0.0819 = 0.28618.
Correlation term = 0.10 × 0.19596 × 0.28618 = 0.005608.
Joint = 0.0036 + 0.005608 = 0.009208.

Answer: About 0.92%, compared with 0.36% if defaults were independent.

Example 2

In a one-factor Gaussian copula, a firm has PD = 2.5% and asset correlation ρ = 0.25 (so √ρ = 0.5). Given a systematic factor M = −2, what is the conditional default probability? Use N⁻¹(0.025) = −1.96.

Show the solution

√(1 − ρ) = √0.75 = 0.8660.
Numerator = −1.96 − (0.5 × −2) = −1.96 + 1.00 = −0.96.
Divide: −0.96 ÷ 0.8660 = −1.1085.
N(−1.1085) ≈ 0.134.

Answer: Conditional PD is about 13.4%, far above the 2.5% unconditional PD, showing how a stressed market raises defaults.

Exam tips

Read whether the question gives asset correlation or default correlation before touching a formula.
Use the bounds check to eliminate options: joint probability cannot exceed the smaller PD.
Expect conceptual questions: why correlation changes unexpected loss but not expected loss, and why the copula is a bottom-up tool.
Know the contrast between bottom-up and top-down models, including that bottom-up needs more data and is slower.
Remember the Gaussian copula criticism: it has no tail dependence and often uses a single stable correlation, which understates stress.

Practice questions from Financial Correlation Modeling - Bottom-Up Approaches

Default Correlation and Bottom-Up Credit Models: frequently asked questions

What is the difference between bottom-up and top-down correlation models?

Bottom-up models build portfolio loss from individual obligors, each with its own PD, exposure and recovery, linked by a dependence structure. Top-down models describe aggregate portfolio or index loss directly. Bottom-up gives detail but needs more inputs.

How do I calculate joint default probability?

Multiply the two PDs, then add default correlation times the two Bernoulli standard deviations. Each standard deviation is √[p(1 − p)]. The result must not exceed the smaller PD.

What does the one-factor Gaussian copula assume?

Each obligor's asset return depends on one common factor and one idiosyncratic factor, both standard normal. An obligor defaults when its return falls below N⁻¹(PD). Given the common factor, defaults are independent.

Is default correlation the same as asset correlation?

No. Asset correlation drives the copula. Default correlation is the correlation of default indicators and is usually much smaller. It also depends on PDs and the horizon.