This study examines exchange rate predictability based on different types of monetary policy rules in Taiwan. The out-of-sample exchange rate predictive accuracy is compared based on the Taylor rule fundamentals to a naïve random walk model. We find that both short-horizon and long-horizon out-of-sample exchange rate predictive power outperform the random walk process in many cases. The stronger evidence relates to the Taylor rule models with interest rate smoothing. The strongest evidence comes from the specifications which involve higher-order interest rate smoothing in the trade-weighted Taiwan Dollar rate. The findings are confirmed by the contemporaneous Taylor rules, the homogeneous coefficients, and the examinations of the nonlinear least squares approaches.