A Comparison of Long Memory and Regime Switching Models of Exchange Rates
This event is in the past.
Wayne State University
Title: A Comparison of Long Memory and Regime Switching Models of Exchange Rates
Abstract: A random walk process remains the toughest model to beat in exchange rate forecasting. Given that both structural change and fractional integration can easily be confused for a unit root in finite samples, we model exchange rates for ten countries against the U.S. dollar using a Markov switching (MS) autoregressive model and a fractionally integrated (FI) model over a sample spanning the past two decades. We discover that a two-state MS model can describe most of the exchange rates well, with a dominant state featuring low-volatility appreciation and a dominated state featuring high-volatility depreciation. Evaluation of in-sample t reveals superior performance of the MS model over the random walk. Most importantly, we show that even if a process is explosive or strongly mean-reverting in both regimes, it could be confused with a single-state random walk process. Alternatively, long memory evidence is strong in some currencies, yet it does not translate to better in-sample fit.