Heterogeneous Expectations and the Predictive Power of Econometric Models

A recent literature questions the mainstream omniscient rational agent, suggesting that agents act as, and have the same bounded rationality of, econometricians. Heterogeneous expectations may then arise because of the different forecasting models used by individuals, who select disparate predictors according to the peculiar net benefits of each model. Net benefits are assumed to be a function of mean square forecasting errors (MSE). Consequently, as in Carroll’s epidemiological approach, an implicit assumption is that the level of disagreement across agents cannot Granger cause model-based MSE. Instead, survey expectations on GDP growth show that the information flow runs exclusively from heterogeneity to MSE. Moreover, variance decompositions point out that survey expectations entropy and MSE are not contemporaneously correlated, enforcing the detected causal chain. Results are robust to several predictors, nonlinearities, and suggest looking also at other possible causes of disagreement.