Dynamic Choices with Social Interactions

We introduce the study of dynamic, forward-looking equilibrium choices in environments characterized by social interactions. Agents’ preferences capture inter-temporal links - such as, e.g., habits and addictions - as well as attitudes to conform to a social reference group. We characterise equilibrium behaviour as a system of linear non-stationary Markovian policy rules, for each individual and in each time period. We then derive conditions for the identification of the parameters of the dynamic social interaction model with panel data. We finally illustrate the empirical implications of the model by bringing it to data in the context of adolescents’ smoking behaviour, where addiction effects are arguably a first-order concern. Indeed, we find strong evidence for the dynamic effects we emphasize in the paper. In our empirical implementation, peer effect estimates for a misspecified static model typically used in the extant literature are approximately six times larger than those of a dynamic specification.