Importance Sampling Method for Portfolio Risk

Abstract:

Based on the Gaussian Copula model with arbitrary marginal distribution in portfolio’s market risk or credit risk; To improve the efficiency in Monte Carlo simulation with importance sampling, we first transform loss to a function of a high dimensional normal vector, then the Newton’s method and a method based on the large deviation theory are used to estimate the coefficients in measure transformation, and the method of freezing coefficient is also proposed. Numerical experiments show that compared with standard Monte Carlo method, the algorithm proposed in the paper reduce simulation error greatly and therefore improve computational efficiency.