Extended Viewer

Model Insight - Predicting Risk at Short Horizons - January 2013

Predicting risk at short horizons requires a delicate balance between two effects. On the one hand, it is best to give more weight to recent observations, as these contain the most relevant data; on the other hand, giving too much weight to recent observations can lead to increases in sampling error. In this paper, we study how to optimally balance these two effects through appropriate model calibration. Central to this challenge is the identification of a reliable measure of risk forecasting accuracy. Using such measures, we compare the accuracy of risk forecasts for several volatility estimation techniques. We find that the Volatility Regime Adjustment approach, based on cross-sectional observations, provided the most accurate forecasts of all methods considered.