Volatility Forecasts by Clustering: Applications for VAR Estimation

Abstract

It is well known that volatility is time-varying and clustered. However, few studies have explored the information content of volatility clustering and its implications for investors’ risk aversion. This information is particularly important in turbulent periods, such as financial crisis. We present a volatility cluster partition model to forecast volatility and apply it to risk management. We find that our model substantially outperforms the GARCH model and improves financial risk management using the value-at-risk metric.

Keywords: Volatility forecasts, Fisher's optimal dissection, value-at-risk