AbstractThere are occasions when researchers are interested in quarterly forecasts of variables that are released at higher frequencies. In these situations it is common for researchers to convert from the higher frequency to the lower frequency by some method, such as averaging or stock‐end, and then to model the low‐frequency data. This paper shows how to improve quarterly forecasts by using within‐quarter variations of monthly data. We compare the one‐step‐ahead and multi‐step‐ahead forecasts for real GNP generated using our approach with those of Fair and Shiller (1990). Our model is extremely simple and, yet, or perhaps because of, produces a lower RMSE than any model in Fair and S