Gambling behavior in multiple-choice betting games

Abstract In a multistage gambling process, a gambler is required to apportion his capital over the available alternatives, and is paid according to the outcome of the game. His problem is to choose a betting policy that is optimal with respect to some criterion. Examinations of logarithmic and power utility functions yield proportional stationary Markov betting policies which are generalized to cases where the vector of probabilities of outcomes is unknown. The sensitivity of the model to departures from optimality is investigated. Twenty-nine subjects participated individually in a multiple-choice multistage gambling experiment. While neither the logarithmic nor the power model was strictly supported, the former was favored by the data. Both the current amount of capital and the pattern of previous outcomes, irrelevant factors in both models, were found to affect subjects' betting decisions.