The authors propose a new benchmark to evaluate forecasts of averaged series, such as the monthly real price of oil. The new benchmark is based on the last high-frequency observation of the underlying series and allows forecasters to test for predictability. The authors also warn that forecast comparisons with the conventional benchmark can introduce spurious predictability. In an application to the real price of crude oil, the authors find that the new benchmark overturns the existing evidence for oil-price predictability: the real price of oil is more difficult to predict and behaves more similar to the prices of financial assets than implied by the academic literature. The authors’ results also highlight that incorporating information from high-frequency observations into forecasting models can yield large gains in forecast-accuracy. Such gains are likely to occur in any setting where forecasters work with averaged data and the underlying series are very persistent.