On a generalized method of combining predictive distributions for stock market index

Abstract

Forecast combinations have frequently been found in empirical studies to produce better forecasts on average than methods based on the ex ante best individual forecasting model. In this paper, we study a generalized method of aggregation in the form of a nonlinear transformation of a linear mixture model. The major advantage of the nonlinear transformation is an excellent flexibility to calibrate predictive cumulative distributions. This method proves to be particularly useful to accommodate complex volatility in the stock market. As for applications, we study two stock market indices, namely the Vietnamese VN30 index and the Thai SET50 index. The forecasts are in the form of empirical densities estimated by Bayesian inference.