Bayesian Markov Switching Quantile Regression with Unknown Quantile τ: Application to Stock Exchange of Thailand (SET)

Woraphon Yamaka, Pichayakone Rakpho, Songsak Sriboonchittac

Abstract

This paper introduces a Bayesian Markov Switching quantile regression with unknown- quantile model that allows the quantile level to be an estimated parameter. This will enable the model to reflect the real behavior of the data series. In the conventional estimation, the maximum likelihood is employed for switching model. Nevertheless, there are some concerns that the conventional estimation may face the computation difficulties. Thus, we consider a Bayesian estimation as the alternative estimator for this model. The posterior distribution of the model is constructed from the Asymmetric Laplace Distribution and uninformative prior distribution. The Metropolis Hasting is employed as the sampling method for the posterior and the vector of parameters. Both simulation study and real data application are provided. The results confirm the accuracy of the Bayesian estimation in both simulation and real application study.