Evaluating Value-at-Risk Models with Desk-Level Data

Abstract

We present new evidence on disaggregated profit and loss (P/L) and Value-at-Risk (VaR) forecasts obtained from a large international commercial bank. Our dataset includes daily P/L generated by four separate business lines within the bank. All four business lines are involved in securities trading and each is observed daily for a period of at least two years. Given this unique dataset, we provide an integrated, unifying framework for assessing the accuracy of VaR forecasts. We use a comprehensive Monte Carlo study to assess which of these many tests have the best finite-sample size and power properties. Our desk-level data set provides importance guidance for choosing realistic P/L generating processes in the Monte Carlo comparison of the various tests. The Caviar test of Engle and Manganelli (2004) performs best overall but duration-based tests also perform well in many cases.

Keywords: Risk Management, Backtesting, Volatility, Disclosure

JEL Classification: G21, G32