A Bayesian Confidence Interval for Value-at-Risk

This study assesses the accuracy of the value-at-risk estimate (VaR). On the basis of posterior distributions of the unknown population parameters, we develop a confidence interval for VaR that reflects the genuine information available about the portfolios for which the VaR is calculated. This approach is more accurate than that in Dowd (2000) as it avoids explaining the behaviour of the population parameters on the basis of distributions of sample parameters. We find that the accuracy of both the confidence interval and the VaR estimate depend more dramatically on the sample size than what Dowd’s results suggest. In addition, we not only find that the impact of the confidence level and the holding period at which the VaR is predicated are negligible compared to that of the sample size (as in Dowd), but also that the confidence interval is far from being symmetric.