Every multivariate distribution simultaneously encodes two things: how each component behaves individually (the marginals) and how they move together (the dependence). These two aspects are fundamentally separable, and Sklar's theorem is the theorem that proves it. The copula is the pure dependence object: it captures the joint distribution of probability-transformed rvs , each uniform on , stripping away every marginal feature. A model builder can then mix any marginals (Student-, lognormal, whatever EVT suggests) with any copula (Gaussian, , Clayton, Gumbel), producing a fully specified multivariate model.
This separation power is what the credit crisis exposed. Li (2000) fixed the Gaussian copula and calibrated it to default correlations — but left the marginal default probabilities governed by their own model. The combined model had zero tail dependence between default events regardless of the calibrated correlation. Copula theory makes this failure precise.
Learning objectives. By the end of this note you should be able to:
- state the definition of a copula and verify the three axioms;
- state and apply Sklar's theorem: both the existence/uniqueness direction and the construction direction;
- extract the copula from a given bivariate joint distribution;
- simulate from a meta-distribution (arbitrary marginals + chosen copula);
- state the three fundamental copulas (independence, comonotonicity, countermonotonicity) and their financial interpretations.
In this lesson
-
- Definition and basic properties
-
- Sklar's theorem
-
- Fréchet bounds and fundamental copulas
-
- Implicit copulas from known distributions
-
- Meta-distributions and simulation
Before you start
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Quantitative Risk Management syllabus.