Forecasting period-average exchange rates requires using high-frequency data to efficiently construct forecasts and to test the accuracy of these forecasts against the traditional random walk hypothesis. To achieve this, we construct the first real-time dataset of daily effective exchange rates for all available countries, both nominal and real. The real-time vintages account for the typical delay in the publication of trade weights and inflation. Our findings indicate that forecasts constructed with daily data can significantly improve accuracy, up to 40 per cent compared to using monthly averages. We also find that unlike bilateral exchange rates, daily effective exchange rates exhibit properties distinct from random walk processes. When applying efficient estimation and testing methods made possible for the first time by the daily data, we find new evidence of real-time predictability for effective exchange rates in up to fifty per cent of countries.