Testing the rationality of expectations using aggregate data

In this paper it is argued that tests of rationality of expections based on aggregate data, specifically when the aggregate expectations series come from business survey data, are not conclusive. In fact it is shown that even when individual agents have rational expectations, aggregate expectations series and aggregate prediction errors should not pass the traditional rationality tests. This can account for the observed persistent correlation in deviations of real GNP from trend in a Lucas-supply function without the need of a lagged output variable. But it also suggests that the macroeconomic implications of rational expectations will be weakened.