Know these first
- L6.1.09: Copulas
Math Atoms
By the end you will be able to
- 1Build a two-level sector model that keeps the average correlation and changes its structure, and compare the whole book's tail with one factor.
- 2Simulate defaults with Gaussian and t copulas, and say why ν is a modelling choice and when a grouped t copula is needed.
- 3Report a simulated VaR and ES with their uncertainty, and read a convergence table.
- 4Explain importance sampling by a factor shift and its likelihood ratio, and measure the variance it saves.
Why it matters
One factor drives every number so far: the twelve firms, three mortgage books and five consumer books all rise and fall with the same economy. Give the sectors their own factors and the book's 99.9% loss falls from £134.7m to £115.6m. Keep one factor but swap the Gaussian copula for a t with ν = 10, and it rises to £281.2m. Same PDs, same LGDs, same exposures. Only simulation can answer these questions for a real book, and only a careful reading of the answers stops them being misused.