Mixture sutoregressive models for Thai stock market data

Abstract

The autoregressive (AR) model is one of the most widely used time series forecasting models. The standard AR model was established using the normal distribution, which is violated in some datasets, notably financial data. Therefore, alternative distributions are proposed in the literature, such as the concept of mixture distributions. This concept is also applied to time series modeling in the family of mixture autoregressive models that combine different autoregressive components. Specifically, we consider both the univariate mixture autoregressive model and the multivariate mixture autoregressive model based on the normal and t distributions. In this study, we construct the EM algorithm to estimate parameters and investigate the performance of this method compared with the MLE. The analysis focuses on top stocks from two different sectors in the market, namely energy and utility and electronic components, with each sector comprising three stocks. The fitted models are compared with the family of mixture autoregressive models by using AIC, HQIC, BIC, and MSE of predictions. The results indicate that the EM algorithm is preferred for Thai stock market data.