Current Research

One part of my recent work applies Statistical Decision Theory to study how to best allocate scarce experimental resources to screen potential innovations in online experiments (A/B tests). See this and this. This problem connects naturally with a classical literature on the ''value of information'' in decision problems, and also with Empirical Bayes methods (as we explain here).

Another part of my research has focused on the use of econometric models for macroeconomic policy evaluation; in particular, identification, estimation, and inference about impulse response functions in structural vector autoregressions (SVARs). I have studied inference based on external instruments (see this and this); the properties of inference based on local projections; inference in a class of set-identified SVARs (see this and this); and results on how to conduct simultaenous inference about impulse responses.

I have also been studying some tools that have received a lot of attention in the machine learning literature. For example, Variational Inference (a popular algorithm used to conduct approximate Bayesian inference in large-scale, parametric models), Dropout Training (a popular method used in the estimation of parameters of neural networks), the analysis of text data (in particular, trying to understand the identification of the parameters in the popular Latent Dirichlet Allocation model), and Statistical Learning Theory (with the hope of using this framework to provide computationally feasible approximations to identified sets in parametric models).

This past spring I received a Presidential Authority grant from the Russell Sage Foundation to study racial and ethnic disparities in police homicides (joint with Dan O'Flaherty and Rajiv Sethi). In this paper we use an Empirical Bayes approach to construct counterfactuals of police homicides for the nation's ten largest police departments.

Click here to see a full list of my papers. Click here for a detailed research statement.