Volatility dependent smooth transitions and abrupt switches: Why they are needed for better forecasting the FX rates

Abstract

Exchange rate prediction is a problematic area. There still does not exist a structural exchange rate model that can consistently predict future exchange rates better than a driftless random walk model. This paper draws attention to two important points regarding this problem. First of all, many structural exchange rate models inherently depend on the uncovered interest parity (UIP) rule. However, empirical evidence almost universally rejects UIP. Therefore, this paper firstly questions whether it would be possible to improve the predictions of UIP by converting it into a nonlinear form since the workhorse UIP specification has traditionally been linear. Secondly, this paper also discusses that a nonlinear transformation is indeed a necessity given that exchange rates typically follow meandering time paths. Inspired by a Bank for International Settlements (BIS) report, UIP model is estimated by volatility-dependent Threshold autoregression (TAR) and Smooth transition regression (STR) specifications using a dataset on two popular carry trade currencies against the US dollar and the Japanese yen. Estimations clearly favor TAR and STR over a linear specification. Although random walk model remains as the champion, results are still indicative of the usefulness of a volatility-dependent regime switching framework for improving the prediction performances of various other structural models that are dependent on UIP.