(with Jeffrey M. Wooldridge)
Abstract: We study Gaussian imputation for distributional difference-in-differences with aggregated outcomes. Parallel changes in conditional means and log variances identify a Gaussian proxy for the untreated counterfactual distribution. We characterize when effective aggregation makes approximation error small enough for variance savings to improve mean squared error, relating this tradeoff to skewness and relative aggregation and sample sizes. We also show why counterfactual gains need not improve treatment-effect precision. Comparisons with Sieve Gaussian, Flexible Link, and flexible standardized DiD assess moment and shape restrictions, supported by joint inference for estimator discrepancies. Simulations demonstrate substantial counterfactual gains, failures under persistent common shocks, and treatment-effect reversals. An application to community health centers and county mortality finds close Gaussian and Flexible Link estimates, while small treated cohorts and limited overlap strain richer specifications.
PDF file: Draft coming soon
STATA package : (GitHub repository) coming soon
Abstract: I develop a framework linking pretrend tests, data-driven preperiod selection, and estimation in difference-in-differences without additional identifying assumptions. Under baseline identification, an additional preperiod identifies the same effect exactly when its pretrend restriction holds. With compatible covariance construction, exact Wald–J equivalence in extended two-way fixed effects enables cohort-wise Wald–BIC selection and joint minimum-distance (MD) estimation. Under stated conditions, Wald–BIC consistently selects valid restrictions; focused selection addresses local bias–variance tradeoffs. Extensions cover doubly robust and nonlinear DiD. Simulations document finite-sample gains and losses; an application to community health centers illustrates smaller standard errors. These findings motivate reporting MD alongside baseline estimates.
PDF file: Download (version: Nov 16, 2026)
STATA package : (GitHub repository)
Abstract: It is well-known that nonlinear generalized method of moments (GMM) estimators often encounter computational challenges when the moment conditions are over-identifying and nonlinear. To enhance computational properties, I propose a novel GMM estimator based on linearized moment conditions approximated around an underlying exactly-identified (or over-identified) parameter estimate. This estimator demonstrates improved computational performance while maintaining first-order asymptotic efficiency, which is particularly beneficial in cases involving a large number of moment conditions. The enhancement arises from (i) the better-behaved curvature of the GMM objective function (e.g. strict local convexity) for estimating the underlying parameter, and (ii) the availability of a closed-form solution for the final estimate. For any given standard moment condition, I prove the existence of such an underlying parameter, and introduce a straightforward algorithm for its identification. The added dimensions in the underlying exactly-identified parameter can be estimated one element at a time, separately. The method has been applied to Ahn, Lee, and Schmidt's (2013) panel data model with multiple time-varying individual effects. This paper provides a STATA command.
PDF file: Download (version: Jan 30, 2025)
STATA ado file: Download
(Studies in Nonlinear Dynamics and Econometrics, 2024) DOI
(Economics Letters, 2020, Volume 192, Article 109184) DOI