SOCIAL LEARNING WITH A HIDDEN ACTION

Abstract. We analyze a model of observational learning in which of the two actions taken by agents only one is observable, say, the positive action (investment, henceforth). The negative action (no investment) is hidden and can only be inferred by the lack of invest-ments through time. We investigate the possible consequences of this lack of observability. We find our model has qualitatively the same asymptotic properties as the standard herding model in which both actions are observable (benchmark, henceforth). Namely, there is incomplete learning in the limit and possible settling on the wrong action if and only if private beliefs are bounded. How-ever and perhaps surprisingly, there is always a higher chance of investment than in the benchmark whether the state is good or bad. More likely investment, although welfare improving in the good state, is “overinvestment ” nonetheless as it always leads to lower ex ante welfare than in the benchmark