My current research mainly focuses on causal inference, missing data, and machine learning
My current research mainly focuses on causal inference, missing data, and machine learning
Working Papers:
Efficient Instrumental Variable Estimation under Sequential Missingness (Job Market Paper) Link
(Presented at Texas Econometrics Camp 2025)
(Scheduled to present at SEA Annual Meeting 2026, ASSA Annual Meeting Poster Session 2026)
Abstract: Missing data poses a challenge for instrumental variable estimation when endogenous variables and instruments are incompletely observed across survey modules, linked records, or multiple waves. I introduce a sequential missing-at-random condition that allows instrument availability to depend on observed endogenous variables and permits dependence among instrument observation indicators. This relaxes the common restriction that instrument availability depends only on the always observed variables. Under this condition, I establish identification and develop a two-step inverse probability weighted GMM estimator that combines available moments across observation patterns using sequential inverse probability weights. I establish consistency and asymptotic normality and derive feasible weighting matrices that minimize asymptotic variance within the proposed estimator class. Monte Carlo simulations illustrate the consequences of misspecified observation models and the benefits of flexible estimation. An application to family size and schooling using an extended sample from the Indonesia Family Life Survey finds statistically significant evidence that larger sibling counts reduce children’s schooling under the maintained identifying assumptions. The resulting framework generalizes IPW-GMM estimation under weaker restrictions on data availability.
Heterogeneous Treatment Effect in Panel Data with Machine Learning (Draft available upon request)
(Presented at Texas Econometrics Camp 2026)
Abstract: Average treatment effects can conceal differences that matter for policy, yet identifying these differences is challenging in the presence of unobserved individual heterogeneity. I propose a framework for estimating sorted group average treatment effects (GATES) and conducting classification analysis (CLAN) in high-dimensional panel data with binary treatment and unit fixed effects. Building on existing panel methods, I develop weighted estimators to measure treatment effects across groups formed by machine learning predictions and characterize the most and least affected populations. I study conditions for valid inference that accommodate estimated treatment probabilities and within-unit dependence. An application to maternal smoking and birth weight examines how the effects vary across mothers. The framework connects panel data identification strategies with interpretable measures of treatment effect heterogeneity.
Work in Progress:
Variable Selection Model with Machine Learning Imputation for Incomplete Big Data
Publications:
Jiqing Zhao, Da Lu, Yucheng Zhou. "TOPSIS-based Investment Risk Assessment of Countries along the Belt and Road". Journal of Hubei University of Technology. 2018, 33(2): 33-35.