Counterfactual Estimation in Semiparametric Discrete-Choice Models

We show how to construct bounds on counterfactual choice probabilities in semiparametric discrete-choice models. Our procedure is based on cyclic monotonicity, a convex-analytic property of the random utility discrete-choice model. These bounds are useful for typical counterfactual exercises in aggregate discrete-choice demand models. In our semiparametric approach, we do not specify the parametric distribution for the utility shocks, thus accommodating a wide variety of substitution patterns among alternatives. Computation of the counterfactual bounds is a tractable linear programming problem. We illustrate our approach in a series of Monte Carlo simulations and an empirical application using scanner data