The predictability of Asian exchange rates: evidence from Kalman filter and ARCH estimations

Abstract

This study provides evidence on long-horizon predictability of exchange rates. We employ the Kalman filter and ARCH-family models to account for time-varying parameters and conditional variances. At long forecasting horizons (6–12 months), we find the Kalman filter and/or ARCH models generally outperform the naive random-walk model for the Singapore dollar and the Japanese yen, and outperform the OLS and AR(1) forecasts. At the 12-month horizon, our estimation produces significantly lower forecast errors than previous models. The Singapore dollar has the lowest short-run forecast error, while the Malaysian ringgit has the lowest long-run forecast error. The Japanese yen is least predictable, with the largest forecast errors.