Filtered Historical Simulation (FHS)
Filters historical returns for changing volatility before resampling shocks.
How it works
Filtered historical simulation estimates conditional volatility, standardises past shocks by their then-current volatility and reuses those shocks at the forecast volatility. This combines adaptive volatility with an empirical innovation distribution. A multivariate implementation also needs a consistent treatment of dependence across risk factors.
Worked example
A past loss of 2% observed on a day with estimated volatility 1% becomes a two-standard-deviation shock. With a current volatility forecast of 1.5%, reusing that standardised shock implies a scenario loss of 3%. The scaled scenarios are then applied to today’s exposures.
Validation and limitations
Check residual stationarity, leverage effects and portfolio repricing after rescaling. A historical sample may omit a relevant crisis or liquidity shock. Validate both calm and turbulent periods and compare with unfiltered historical VaR.