The Law of Small Numbers in Multivariate Sequences
Dr. Matthew Levy
Lecturer in Economics
Department of Economics
London School of Economics
This paper shows that, under modest consistency requirements placed on a decision-maker’s beliefs, distortion of subjective beliefs over the conditional distributions of outcomes cannot be global over all sequences. Specifically, while it is possible for a decision-maker to commit the Gambler’s Fallacy (or Hot-hand Fallacy) over multiple independent sequences, it is always possible to construct a data-generating process consisting only of i.i.d. processes with at least one sequence over which the decision-maker is unbiased in all histories of arbitrary length. This further implies that if the decision-maker does distinguish between histories of i.i.d. sequences, then they cannot commit the Gambler’s Fallacy at all. A simple experiment demonstrates that subjects indeed do not treat all i.i.d. sequences identically, and highlights the importance of framing in creating and maintaining belief in the Gambler’s Fallacy.