Another Look at the Identification of Dynamic Discrete Decision Processes

This paper presents an econometric approach to estimate the behavioral effects of counterfactual policy experiments in the context of dynamic decision models where the current utility function and the distribution of unobservables are nonparametrically specified. Previous studies have shown that the identification of the current utility function in dynamic decision models requires of stronger assumptions than in static decision models. We show in this paper that knowledge of the current utility function (or of a 'normalized' utility function) is not necessary to identify counterfactual choice probabilities in dynamic models. To identify these counterfactuals we need the probability distribution of the unobservables and the difference between the present value of choosing always the same alternative and the present value of deviating one period from this strategy. We show that both functions are identified from the factual choice probabilities under similar conditions as in static decision models. Based on this result we propose a nonparametric procedure to estimate the behavioral effects of counterfactual experiments in dynamic decision models. We apply this method to evaluate the effects of an investment subsidy program in the context of a model of machine replacement.