Dynamic Local Average Treatment Effects in Time Series

This paper studies identification, estimation, and inference of dynamic LATEs in IV models. We show that compliers can be identified individually in time series; this allows both improved interpretability of IV estimates and a direct test of the exclusion restriction. We document widespread weak identification in the literature, but also strong identification in relevant subsamples. We propose a method (based on dynamic programming) to select the most strongly identified subsample and improve estimation and inference, along with new robust procedures. Application to monetary policy.Profilo del ricercatore: Ricercatore in possesso di dottorato in Economia e/o Finanza e con esperienza su metodi di serie storiche per dati economici