Dynamic choice models

In this chapter, we explore dynamic choice problems - where choices made in the present are both dependent on those in the past and will also affect those in the future. We start by considering the dynamic choice problem from two different perspectives: that of the decision maker and that of the analyst. In doing so we identify three different mechanisms that influence dynamic behaviour. Existing dynamic models in the literature typically focus on only one of two of these mechanisms: forward-looking models that try to account for forward planning, or habitual behaviour and learning models that describe how we learn from previous choices. Instead, we propose a general parametric dynamic choice model based on first principles that unifies these two approaches under a single framework. We then extend the framework to account for: (i) the Markov assumption in the anticipation of future explanatory variables and habitual behaviour and learning; (ii) agent effects in the utility function and the anticipation of future explanatory variables; and (iii) variance inflation in the error terms as the prediction interval increases. Finally, we use the general model to show how different types of dynamic choice models in the literature can be derived through simple assumptions on the model parameters.