Mixture distribution-based forecasting using stochastic volatility models

Abstract

Many traditional econometric methods forecast the conditional distribution of asset returns by a point prediction of volatility. Alternatively, forecasts of this distribution may be generated from a mixture of distributions. This paper proposes a method by which information extracted from the estimation of a standard stochastic volatility model (using non-linear filtering) can be used to generate mixture distribution forecasts. In general, it is found that forecasts based on mixture distributions are superior to those simply using point predictions of volatility. In terms of mixture distribution forecasts, the method proposed in this paper is found to be superior to a number of competing approaches.