Forecaster (Mis-)Behavior

Abstract

We document two stylized facts in expectational data. First, professional forecasters overrevise their macroeconomic expectations. Second, such overrevisions mask evidence of both over- and underreactions to public signals. We show that the first fact is inconsistent with standard models of noisy rational expectations, but consistent with behavioral and strategic models. The second fact, in contrast, presents a puzzle for existing theories. We propose an extension of noisy rational expectations that allows forecasters to be overconfident in their information. We show that this feature when combined with the endogeneity of public signals leads to over- and undereactions consistent with the data.

Myopia and Anchoring

“…As noted earlier, rational expectations requires that the coefficient of this regression be zero. Bordalo et al (2020) and Broer and Kohlhas (2019) argue that this coefficient is negative in the data, supporting the presence of overconfidence. Our own take is that the evidence is inconclusive: the relevant coefficient switches signs across variables…”

“…At the same time, we explain why the evidence on the underreaction of average forecasts provided in CG is more "reliable" for our purposes than the conflicting evidence on the overreaction of individual forecasts provided in Bordalo et al (2020) and Broer and Kohlhas (2019). In an extension that adds a behavioral element as in those papers (a form of overconfidence), we can vary the theory's implications about individual forecasts without varying the structural relation between average forecasts and aggregate outcomes.…”

“…But even if we take for granted those papers' preposition of systematic bias in beliefs, this does not necessarily upset either our theoretical results or our proposed quantitative strategy. We illustrate this point by augmenting our model with the same kind of overconfidence as Broer and Kohlhas (2019): whereas the actual level of noise is σ, agents perceive it to be σ ˆ , for some σ ˆ < σ. (For completeness, underconfidence, or σ ˆ > σ , is also allowed.)…”

“…Suppose, now, that the analyst knows all parameters except σ ˆ and σ and wishes to quantify the equilibrium effects of the friction under consideration (as we do, for example, in Section V). Suppose further that the analyst combines the CG coefficient with the individual-level counterpart estimated in Bordalo et al (2020) and Broer and Kohlhas (2019). Then, the CG coefficient alone allows the identification of σ ˆ and the quantification of its effect on the actual dynamics.…”

Monetary Policy with Opinionated Markets

“…Our empirical analysis of interest rate disagreements is related to a literature that uses survey data to document belief distortions about macroeconomic outcomes. Much of the recent literature focuses on whether agents over-or underreact to data (e.g., Coibion and Gorodnichenko (2015); Bordalo et al (2020); Broer and Kohlhas (2018); Angeletos, Huo and Sastry (2020); Ma et al (2020)). In contrast, we focus on the relationship between disagreements on different macroeconomic variables (see also Andrade et al (2016); Giacoletti, Laursen and Singleton (2021); Bauer and Chernov (2021)).…”