Identification of Dynamic Treatment Effects by Instrumental Variables

This paper considers the problem of the identification of causal effects using instrumental variables. We are interested in the effects of some treatments on certain outcomes. First, we consider that a participation in a treatment or a program is only possible one time but we have the choice between more than one program. Under a monotonicity condition and an exclusion restriction, pair-wise Local Average Treatment Effects are identifiable. Second, we consider the case where only one program is available but more than one participation is possible, leading to a comparison of sequences of participations (or sequences of programs). In this framework a problem of endogeneity appears: the outcome after one period, affected by the participation in this period, can influence the participation in the next period. Under different versions of the monotonicity condition and the exclusion restriction, identification of the causal effects of sequences of programs are investigated. The introduction of a second period implies a loss of identification for some effects of interest even without any endogeneity problem.