Back-Testing: Traffic Lights Principle

Back-testing is used in the VaR model.

To check the model reliability, we find the number of days (out of 250 working days by default), when actual losses in portfolio value exceeded forecast values. On average, real loss exceeds VaR value in one case out of hundred, or 2.5 times per 250 working days. These cases are named exceptions, and the loss in such a case is named extreme loss.

The so-called "traffic-lights principle" is used to classify models by the level of their reliability. Depending on the number of exceptions, the model is considered to belong to one of the zones: green (if 0 to 4 outliers occurred), yellow (5 to 9 outliers) or red (10 or more outliers).