Generating Random Optimising Choices

This paper provides an efficient way to generate a set of random choices on a set of budgets which satisfy the Generalised Axiom of Revealed Preferences (GARP), that is, they are consistent with utility maximisation. The choices are drawn from an approximate uniform distribution on the admissible region on each budget which ensures consistency with GARP, based on a Markovian Monte Carlo algorithm due to Smith (1984). This procedure can be used to extend Bronars‘ (1987) method as it can be used to approximate the power of tests for conditions for which GARP is a necessary but not sufficient condition (e.g., homotheticity, separability, risk aversion, etc.). For example, it allows to approximate the probability that a set of random choices which happens to satisfy GARP is also consistent with homotheticity. The approach can also be applied to production analysis and nonparametric tests of cost minimisation.