Discussion

Following up on Ricardo Reis’s discussion, Martin Eichenbaum asked a question on disagreement in models with heterogeneous beliefs, which sparked a general discussion on the topic. In practice, disagreement between agents seems to be very persistent. He asked whether the authors’ model and the modification suggested by the discussant were consistent with this observation.Alp Simsek secondedEichenbaum’s comment, pointing out that most models with dispersed information, including the authors’, have the implication that if an individualwere to elicit other agents’ expectations, shewould account for this new information. In otherwords, if average beliefs were part of the current information set, individuals would update their own beliefs in response. This does not seem to be supported by the data, he claimed. Simsek’s own empirical analysis of the Blue Chip Financial Forecasts in Ricardo Caballero and Alp Simsek (“Monetary Policy with OpinionatedMarkets” [Working Paper no. 27313, NBER, Cambridge, MA, 2020]) finds evidence in support of confident disagreement: past individual forecasts are abetter predictor for future individual forecasts than past consensus forecasts. This fact seems to indicate that forecasters have dogmatic beliefs and do not consider the other agents to have useful informationwhen forming expectations for the future. It is important to take this into account for understanding how agents respond to new information, he claimed. The authors answered to the remarks on the role of disagreement among agents arguing that it does not have material consequences for the implications of their theory. There are two arguments that support this conclusion. First, the actual level of information precision is irrelevant for shaping aggregate outcomes. It is the perceived level of precision