This study explores the out-of-sample forecasting ability of exchange rate models using real-time macroeconomic data with bilateral exchange rates and short-term interest rates from the U.K. and Western Offshoot countries, that is, Australia, Canada, New Zealand, and the U.S. We employ the Taylor rule with monetary policy inertia to forecast the out-of-sample exchange rate changes across these economies using various methods, such as ordinary least squares, non-linear least squares, and generalized method of moments. Our findings show that the exchange rate forecasting performance of the relative Taylor rule model is mostly superior to that of a naïve random walk process even after utilizing real-time macroeconomic data. By comparing them to the results obtained from ex post revised data, we confirm the desirability of real-time data for estimating the model. Our results also present the invalidity of the assumptions concerning the homogeneous coefficient and symmetric reaction of real exchange rates. Furthermore, we split the sample at the point where some structural breaks occur through a policy interest rate’s deviation from its forecast obtained from the original Taylor rule model and estimate the exchange rate model in each subsample. The superiority of the model persists in terms of its predictability during relatively moderate deviation periods for each of the currency pairs. Using forward-looking inflation/output gaps, including stock prices, and employing alternative econometric approaches, we also corroborate the desirable forecasting abilities for several model specifications.