Belief in mean reversion and the law of small numbers

Studies find that people systematically underestimate the likelihood of streaks in a random sequence. In a canonical coin‐tossing environment, this paper shows that the evidence can be explained by a belief in mean reversion. Such beliefs are represented as if the bias of the coin is history‐dependent and “self‐correcting.” In a Bayesian inference setting, a belief in mean reversion ensures that the agent never rules out the true parameter.