Mixed Horizon Risk Forecasting
Yang Liu, FSU
4/1/10
The scarcity of weekly and monthly data makes the usual
fixed-frequency approach of risk forecasts inadequate
as both calibration of filtered return distribution
and forecast of value-at-risk have to be done on the
same timescale. To overcome this drawback, we introduce
a high-low frequency approach that converts calibration
results of high-frequency data to risk forecasts on the
low-frequency level (i.e. longer time horizon). Having
passed backtests for both weekly and monthly forecasts,
this new approach can reduce dependence on data availability
substantially and/or offer more flexibility in selecting
the most relevant historical data. In addition, we uncover
better sensitivity to
volatility changes compared with the fixed frequency approach.