Website Paper
Fields
Econometrics
Job market paper
Non-parametric Identification of Dynamic Treatment Effects: a Graphical Approach to Causality
This paper proposes a unified framework to study different types of dynamic treatment effects (DTE), extending the dynamic potential response (DPR) framework in Torgovitsky (2019). The DPR framework is used to study the causal effect of the immediate preceding past on the present in his paper. In this paper, I first show that DTE, such as the causal effect of the distant past on the present, can be non-parametrically identified in a DPR framework which excludes the distant past from directly entering the function of the present potential response. I then develop an algorithm leveraging a novel graphical representation of the DPR model. This graph-based algorithm provides a computationally tractable means of scaling the DPR framework beyond binary variables to multi-valued variables, circumventing the curse of dimensionality inherent in dynamic treatment effect studies. An application to the 2008 Survey of Income and Program Participation (SIPP) sheds further light on the long-term scarring effect of unemployment.