A proposal to evaluate stock performance forecasts when empirical distributions are not normally stationary

Abstract

This paper deals with the main problems related to predictability of asset returns when data series are not normally stationary distributed. The statistical analysis includes several normality tests on returns series of Banamex-30 stocks first, and then an application of mixture of probability distributions and stochastic processes to series, which are not normal stationary. As a means to avoid the normality assumption when forecasting asset returns, we introduce a second-order Markov model.