Mixed-Copula VaR for Portfolio Risk Evaluation

Abstract

This paper proposes a novel Mixed-copula VaR (MCV) model for financial portfolio risk management and a novel investment strategy based on it. VaR (Value at Risk) is a traditional risk metric in computational finance to measure how much a set of investments might lose in a disadvantageous situation. Previous VaR models assume that the yield rates follow a single distribution (e.g. normal distribution) for simplicity, which is far from reality. In order to improve the adaptivity and the extendability of the VaR method, this paper constructs an MCV model with several families of distributions and designs a fast EM algorithm to compute the mixing weights. It further leads to a strategy for portfolio investment. Experiments by Monte Carlo simulation verify the intention of MCV. Besides, experiments on two real-world financial data sets indicate that MCV measures portfolio risk more accurately and adaptively, and delivers superior investing performance.