Toward Better Risk Forecasts

Abstract

Volatility forecasts crucial to many financial applications usually assume implicitly that the frequency of the data should match the forecast horizon; portfolio managers typically rely on risk models estimated using monthly data to produce monthly volatility forecasts, for example. For longer–term forecasts, this practice has two drawbacks: Volatility estimates can be based on stale data; and return events occurring within long sampling intervals are obscured, confounding estimation. Monthly volatility risk measures constructed using higher–frequency data seem to be more robust than those using low–frequency data. Microstructure effects can explain the differences in estimates.