Directed Attention and Non-Parametric Learning

This paper examines the implications of learning for the effects of ambiguity aversion. The key result is that since agents naturally choose to learn about the sources of uncertainty that reduce utility the most, information acquisition attenuates the most severe effects of ambiguity aversion. The specific setting we study is the canonical consumption/savings problem. Agents endogenously learn most about income dynamics at the very lowest frequencies. While ambiguity aversion typically implies in this setting excessive extrapolation of income shocks, that effect is eliminated here. Furthermore, deviations of consumption from the full-information benchmark are largest at high frequencies, so the model naturally generates overreaction of consumption to predictable short-run income variation.