Home > Topics > Financial Analytics > Backtesting Risk Models

Backtesting Risk Models

You built a VaR model. It says "You won't lose more than ₹1 Lakh with 99% confidence". How do you know if the model is right? Backtesting.

What is Backtesting?

Backtesting involves comparing the Model's Forecasts (VaR) against Actual Returns (P&L) over a historical period.

The Process

  1. Take the last 250 days.
  2. For each day, check: Did the Actual Loss exceed the VaR Calculation?
  3. Exception: If Loss > VaR, it's called an "Exception" or "Breach".

Evaluating Results (Kupiec POV Test)

  • If you calculate 95% VaR, you expect exceptions 5% of the time.
    • In 250 days, you expect 250 * 0.05 = 12.5 exceptions.
  • Green Zone: If you get 10-15 exceptions, the model is Good.
  • Red Zone: If you get 40 exceptions, the model is Underestimating Risk.
  • Yellow Zone: If you get 0 exceptions, the model is Overestimating Risk (holding too much capital).
Note

Reality Check: Banks are penalized by regulators if their models fail backtesting (Red Zone). They are forced to hold more capital (Multiplier effect), which hurts profitability.

Loading quiz…