Pooling Forecasts In Linear Rational Expectations Models

Estimating linear rational expectations models requires replacing the expectations of future, endogenous variables either with forecasts from a fully solved model, or with the instrumented actual values, or with forecast survey data. Extending the methods of McCallum (1976) and Gottfries and Persson (1988), I show how to pool these methods and also use actual, future values of these variables to improve statistical efficiency. The method is illustrated with an application using SPF survey data in the US Phillips curve, where the output gap plays a significant role but lagged inflation plays none.