Gevorg Khandamiryan

We develop a covariate-assisted approach to partially identified parameters that are solutions to an under-identified system of linear equations with known coefficients. Examples include bounds on treatment effects, models of unemployment with state dependence, choice-theoretic models of IV, and random utility models. The boundary (i.e., support function) of the proposed identified set is represented as an average of intersections of regression functions, aggregated over the covariate distribution. We show that the boundary is a regular parameter, propose asymptotic theory, and demonstrate using an empirical application to Jobs First.

This paper extends the Generalized Mundlak Estimator (GME) framework of Arkhangelsky and Imbens (2024) to settings with network interference. In a fixed effect setup group-level unconfoundedness is tackled by balancing statistics, which can include group-level averages of regressors, treatments, and their functions, such that it is sufficient to eliminate differences between groups. When units are connected in a logistic network formation model, network's degree sequence is a sufficient statistic for the unobserved heterogeneity, and the joint distribution of covariates and treatments belongs to an exponential family. Group unconfoundedness can be tackled by using this statistic to determine the group membership of each unit and construct balancing scores according to those clusters. A doubly robust AIPW estimator that uses these balancing statistics is consistent and has nice asymptotic properties, as illustrated in a simulation study.