Essays in Pseudo-Bayesian Learning and Behavioral Macroeconomic Theory

Abstract

The Kalman filter equations make clear that a realistic agent tracking a distant economic variable period by period (i.e. retaining beliefs but not past data) can neither fully consider nor intuitively update the exceedingly large number of conditional second moments required to continually revise her forecast with statistical accuracy. This dissertation leverages basic results in the behavioral literature to put forth a parameterized specification for behavioral belief evolution that yields sharp testable predictions while remaining consistent with current behavioral working knowledge. I ultimately find that predictable inertia in the mental updating of covariances between variables in a stochastic process can have significant effects on first order beliefs, or the forecasts made by economic agents. In particular, the tendency to pay less and less attention to the interaction between variables the farther apart they exist in a process causes agents to inflate their sensitivity to incoming data, resulting in excess volatility in the agent's forecast of future variables. In addition, variables irrelevant to a rational prediction can, under these behavioral conditions, obtain a non-zero response magnitude, which, while dissipating with distance from a rationally relevant realization, can significantly affect the agent's forecast if highly correlated with the target variable.