A New Bayesian Unit Root Test in Nonlinear Heteroskedastic Models

Abstract

This study proposes a Bayesian testing procedure to detect the presence of local non-stationarity in double Markov switching autoregressive GARCH models. To implement a test, a new posterior odds analysis is proposed that uses an adaptive Markov Chain Monte Carlo scheme, which includes a new extension to solve the likelihood identification problem of the autoregressive coefficient with a unit root. Our simulation study demonstrates that the proposed Bayesian test performs properly. Further, our empirical study confirms that the proposed method successfully detects the local non-stationarity in the double Markov switching autoregressive GARCH model.