Skip to content
All apps

Default correlation surface in 3D

About this app

Explore how default correlation depends on probability of default and asset correlation using a 3D surface.

Default correlation

The default correlation ρ of two loans is the correlation of their default indicators: ρ = (p_ij − pᵢ · pⱼ) / √(pᵢ(1 − pᵢ) · pⱼ(1 − pⱼ)), where p_ij is the joint default probability. Portfolio models derive p_ij in different ways. In CreditMetrics it comes from a Gaussian copula with asset correlation ρₐ: p_ij = Φ₂(Φ⁻¹(pᵢ), Φ⁻¹(pⱼ); ρₐ). In CreditRisk+ a shared background factor with coefficient of variation Ω fluctuates, and approximately p_ij ≈ pᵢ · pⱼ · (1 + Ω²).

What the app shows

The app draws ρ as a 3D surface over the default probabilities PDᵢ and PDⱼ, each from 0.5 % to 15 %. Depending on the model, one slider controls Ω (CreditRisk+) or ρₐ (CreditMetrics). A chosen PD pair is marked on the surface, and the stats panel gives the default correlation from both models and the difference between them. With “Overlay both surfaces” they are drawn on top of each other.

Try it

Rotate the view and pick a PD pair. Turn on the overlay and compare the shapes: in CreditRisk+ ρ grows with the PDs, while in CreditMetrics the shape also depends on the asset correlation. Change Ω and ρₐ and look for regions where one model gives a much higher default correlation than the other, even though the PDs are identical.

Frequently asked questions

How do you calculate the default correlation between two loans?
Default correlation is the correlation of two 0/1 default variables: ρ = (p_ij − pᵢpⱼ) / √(pᵢ(1 − pᵢ)pⱼ(1 − pⱼ)). Besides the individual PDs you need the joint default probability p_ij. Since it is hard to observe directly, it is usually derived from a model.
Why do CreditMetrics and CreditRisk+ give different default correlations?
CreditMetrics derives joint defaults from correlated asset returns via a Gaussian copula. CreditRisk+ creates correlation through a fluctuating background factor that moves default rates together. The two approaches depend on the PDs in different ways, so they give different correlations even for identical PDs.
Why is default correlation usually much lower than asset correlation?
Defaults are rare events in the tail of the distribution. Even when two asset returns are clearly correlated, both rarely fall below their thresholds at the same time. The correlation of the default indicators is therefore much lower for small PDs.
What does the 3D default correlation app show?
It shows default correlation as a rotatable surface over two PDs, for CreditRisk+, CreditMetrics or both overlaid. You set Ω, the asset correlation and a PD pair and read off each model’s correlation and the difference.

Subject: Credit risk

More from Credit risk

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps