(WP 2020-04) Heterogeneity in Individual Expectations, Sentiment, and Constant-Gain Learning

The adaptive learning approach has been fruitfully employed to model the formation of aggregate expectations at the macroeconomic level, as an alternative to rational expectations. This paper uses adaptive learning to understand, instead, the formation of expectations at the micro-level, by focusing on individual expectations and, in particular, trying to account for their heterogeneity. We exploit survey data on output and inflation expectations by individual professional forecasters. We link micro and macro by endowing forecasters with the same information set that they would have as economic agents in a benchmark New Keynesian model. Forecasters are, however, allowed to differ in the constant gain values that they use to update their beliefs. We estimate the best-fitting constant gain for each forecaster. We also extract individual measures of sentiment, defined as the degrees of excess optimism and pessimism that cannot be justified by the near-rational learning model, given the state of the economy and the updated beliefs. Our results highlight the heterogeneity in the gain coefficients adopted by forecasters, which is particularly pronounced at the beginning of the sample. The median values are consistent with those typically estimated using aggregate data, and display some moderate time variation: they occasionally jump to higher values in the 1970-80s, and stabilize in the 1990s and 2000s. Individual sentiment is persistent and heterogeneous. Differences in sentiment, however, don\u27t simply cancel out in the aggregate: the majority of forecasters exhibit excess optimism, or excess pessimism, at the same time