Performance of MS-GARCH models : Bayesian MCMC-based estimation

In this chapter, both Maximum likelihood estimation (MLE) and Bayesian MCMC estimation methods are used to test their parameters estimation power while estimating a Markov-Switching generalized autoregressive conditional heteroscedasticity (MS-GARCH) model. The monthly exchange rates of BRICS countries for the period from 1997 to 2017 were used for this empirical analysis. MS(2)-GARCH (1,1) is estimated using both the MLE and Bayesian MCMC. For both methods of estimation, the models were found to be adequate and can be used for further analysis. Prior estimation for the MS (2)-GARCH (1,1), various nonlinearity and nonstationarity tests were estimated with the aim of testing the presence of nonlinearity and the result revealed that the exchange rates were nonlinear in nature. For the comparative analysis, the models with Bayesian MCMC estimates outperformed the one with MLE estimates using error matrices. Furthermore, Diebold-Mariano test was used to assess the predictive accuracy of the models and results confirmed that models with Bayesian MCMC performed better.