Multiple Regime Models and Exchange Rate Forecasting

Abstract

We extend the basic random walk Markov-Switching model in two ways and evaluate the out-of-sample forecasting performance on the Japanese yen during 1995-2004. First, we estimate both a two-and also a three-regime Markov switching models. Second, we add four exogenous variables as suggested in the monetary theory. According to the modified DieboldMariano forecast equivalence test, the result shows that our modified models, a three-regime random walk model and a two-regime monetary model, outperform a simple random walk for the yen. However, the interpretation of coefficients in the two-regime monetary model is unclear and the exchange-rate disconnect puzzle still remains a subject for further investigation