Group-Specific Heterogeneity in Short Binary Outcome Panels (2025) [Job Market Paper]  [GitHub]  [Google Drive]

• Abstract: When the number of time periods is small, identification of binary outcome panel data models that allow for heterogeneity is inherently difficult. This paper presents identification and estimation results for such models in the presence of latent group-specific heterogeneity. We assume that each unit belongs to a time-invariant latent group, so the joint distribution of the binary outcomes conditional on the covariates is a finite mixture model. Under mild conditions, the group-specific conditional outcome models are fully nonparametrically identified, with identification being possible with as few as two time periods. We also provide conditions under which group-specific average marginal effects are identified. In addition, we develop a novel semiparametric estimator for the setup of interest and study its asymptotic properties. Simulations indicate that the estimator performs well in finite samples. We illustrate the estimation procedure and how to interpret the latent groups empirically in the context of homeownership in Germany.

On Missing Covariates in Linear Panel Data Models with Fixed Effects (2025)

• Abstract: This paper presents a single-imputation-based method to address missing covariate values in a linear panel data model with fixed effects. We map the estimation problem into a generalized method of moments (GMM) framework and propose an estimator that improves on the standard fixed effects estimator on the unbalanced panel and the imputed data set in terms of efficiency. Allowing for flexible parametric imputation models, we follow Windmeijer (2005) and provide an estimator for the asymptotic variance that corrects for higher order terms due to the estimation of the GMM weight matrix. Additionally, we present two bootstrap procedures that eliminate the need for reimputing the data set or estimating potentially unavailable cross-derivatives. While we focus on parametric imputation models, we present regularity conditions that allow for nonparametric imputation models and derive the influence function in this setup. In simulations, confidence intervals based on the proposed imputation-based estimator are up to 21% smaller than the ones based on the unbalanced panel fixed effects estimator, all while achieving the nominal coverage rate.

Missing Data in Asset Pricing Panels (with Joachim Freyberger, Andreas Neuhierl, and Michael Weber): The Review of Financial Studies, Volume 38, Issue 3, March 2025, Pages 760–802.

• Abstract: We propose a simple and computationally attractive method to deal with missing data in in cross-sectional asset pricing using conditional mean imputations and weighted least squares, cast in a generalized method of moments (GMM) framework. This method allows us to use all observations with observed returns; it results in valid inference; and it can be applied in nonlinear and high-dimensional settings. In simulations, we find it performs almost as well as the efficient but computationally costly GMM estimator. We apply our procedure to a large panel of return predictors and find that it leads to improved out-of-sample predictability. 

A Simple and Fast Bootstrap for Multi-Step Semiparametric GMM Estimators (with Joachim Freyberger)