Social Learning in Continuous Time: When are Informational Cascades More Likely to be Inefficient?

In an observational learning environment rational agents may mimic the actions of the predecessors even when their own signal suggests the opposite. In case early movers? signals happen to be incorrect society may settle on a common inefficient action, resulting in an inefficient informational cascade. This paper models observational learning in continuous time with endogenous timing of moves. This is the first paper with homogenous access to information that gives an analytical approximation for the probability of inefficient cascades. This permits the analysis of comparative statics results. In contrast to the general impression in the literature, the effect of an increase in signal quality on the likelihood of an inefficient cascade is shown to be non-monotonic. If agents do not have strong priors, an increase in signal quality may lead to a higher probability of inefficient herding. The analysis also suggests that markets with quick response to investment decisions, such as financial markets, are more prone to inefficient collapses.