A common justification for boundedly rational expectations is that agents receive partial feedback about the equilibrium distribution. I formalize this idea in the context of the "Bayesian network" representation of boundedly rational expectations, presented in Spiegler (2015). According to this representation, the decision maker forms his beliefs as if he Öts a subjective causal model - captured by a directed acyclic graph (DAG) over the set of variables - to the objective distribution. When the causal model is misspecified, the belief systematically distorts the objective distributionís correlation structure. I show that when the DAG is perfect, the representation is the outcome of extrapolating from limited feedback - in the form of a long spreadsheet with randomly missing values - using a variant on a familiar imputation technique. When the DAG is imperfect, this foundation breaks down.

## Date:

Sun, 29/11/2015 - 16:00 to 17:00

## Location:

Elath Hall, 2nd floor, Feldman Building, Edmond Safra Campus

imputation.pdf | 254 KB |