VaR Backtesting Quant Interview Guide
VaR backtesting quant interview guide covering VaR forecasts, exceptions, coverage, clustering, model limits, and examples.
Candidates validating risk forecasts and exception counts.
VaR backtesting checks exception frequency
If one-day 95 percent VaR is calibrated, losses should exceed the VaR threshold roughly five percent of the time over suitable samples.
Exceptions can cluster
Even if the average exception rate looks acceptable, clustered exceptions may reveal regime changes, volatility misspecification, or dependence errors.
Concrete example
A model with ten exceptions in two hundred days may pass a rough coverage check, but five consecutive exceptions deserve investigation.
Backtests do not prove safety
VaR backtesting is historical evidence. It does not guarantee future tail behavior, especially when market structure or positions change.
Common mistakes
Candidates often count exceptions without discussing sample size, confidence level, clustering, or whether the VaR horizon matches the portfolio.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.