Testing Theories of Gambler’s Fallacy and Hot-hand Fallacy : Probability Bias vs. Randomized Choice

This paper proposes a new method to identify gambler's fallacy (GF) and hot-hand fallacy (HHF). In the classical method, subjects are classified as exhibiting GF/HHF even they are just randomizing/indifferent between options. In our method, subjects play the game twice and we vary the payoff between different options, and their probability beliefs are elicited. We experimentally test three competing theories of GF and HHF: probability bias, randomized choice, and bounded rationality. We find that the classical method substantially overestimates the proportion of subjects exhibiting GF and HHF due to wrongly classifying randomized and mistaken choices as exhibiting these biases. About 7% (18%) of subjects exhibit true GF (HHF), while the classical method suggests about 51% (48%). We show that probability bias, randomized choice, and bounded rationality accounts for 31%, 41%, and 20% of behavioral GF, while the shares are 40%, 26%, and 18% for behavioral HHF