Multivariate Stable Distributions and Value at Risk: The Case of the Asian Currency Crisis

Abstract

This paper explores practical applications of multivariate stable distributions to value at risk modeling during the Asian currency crisis. We fit multivariate stable distributions to daily foreign exchange rate data 1996 through 1998 for six Asian currencies using a rolling estimation procedure and backtest daily marginal and conditional probabilities under 95% and 99% value at risk nulls. We also examine gains in value at risk accuracy from using multivariate stable distributions relative to univariate benchmarks such as generalized autoregressive conditional heteroskedasticity or univariate stable models. We find multivariate stable distributions overstate the probability of extreme losses.