Structured Risk Assessment and Value-at-Risk

Abstract

An important question for corporate finance officers is whether risk management systems, such as Value at Risk (VaR), currently are producing accurate results. In contrast to previous research on assessing the accuracy of risk systems or VaR, which has focused on backtesting a large sample of historical observations, we provide tools for real-time assessment, using a time window that varies adaptively with the data. The objective is to quickly signal if the estimation process is systematically biased, subject to a specified rate of false detections. For example, if the volatility is systematically underestimated by 25 percent our procedure detects this in an average of 25 observations. Previous techniques have often backtested thousands of observations. We also discuss the trade-off between increasing detection power at the risk of detecting meaningless errors and suggest a parameter to specify the balance desired for a specific application.