Forecasting Volatility and Asset Allocation: Are Garch-Type Models Useful?

Abstract

The purpose of this study is to determine whether fund managers would do better in allocating their funds among different asset classes by using the new Generalized Autoregressive Conditional Heteroskedastic (GARCH) models. Recently, a spate of papers has been written claiming that GARCH models can substantially improve volatility forecasts. This paper, using a GARCH framework on its own, and a GARCH framework together with investors' expectations of volatility, shows that these claims are unfounded. The GARCH models perform as badly as the traditional equally weighted historical averages of volatility. The GARCH models combined with investors' expectations of volatility improve the forecasting performance only negligibly. Fund managers should spend their time and money looking elsewhere for improvements on their forecasts of volatility of major asset classes.

Keywords: GARCH, Forecasting, Asset Allocation, Implied Volatility

JEL Classification: G0,C5