Risk Analytics
Backtest a VaR model
Does five days of loss above a 99% VaR threshold prove the model failed?
An exception occurs when realized loss exceeds the forecast threshold. Compare frequency and clustering over a defined window.
Expected exceptions = number of observations × (1 − confidence level)
250 × (1 − 0.99) = 2.5 expected exceptions. Five observed is 5 ÷ 250 = 2%.
Over 250 days, a 99% model expects 2.5 exceptions. Five observed exceptions imply an observed rate of 2%, which merits investigation.
250 × (1 − 0.99) = 2.5 expected exceptions. Five observed is 5 ÷ 250 = 2%.
What is the expected number of exceptions over 250 days at 95%?