Improving Risk Forecasts Through Cross-Sectional Observations

Abstract

Volatility forecasting requires a delicate balance between two opposing effects. On the one hand, we should give more weight to recent observations, because they contain the most relevant data. On the other hand, giving too much weight to recent observations leads to undesirable increases in sampling error. In this article, the authors study how to optimally balance these two effects. Central to this challenge is the identification of a reliable measure of risk forecasting accuracy. We The authors examine several widely used measures, highlighting serious shortcomings in some of the approaches as well as introduceing a new technique for volatility estimation that refines traditional volatility forecasts by incorporating cross-sectional observations. The authors show that our their technique improves the accuracy of risk forecasts. We They argue that our their cross-sectional technique permits placing more weight on recent observations while mitigating the detrimental effects of sampling error.