Robust Estimation and Inference for Doubly High-Dimensional Instrumental Variables Models

Endogeneity and heavy-tailed errors pose substantial challenges for high-dimensional

statistical inference. This talk considers linear instrumental variables

models in which both the number of endogenous regressors and the number

of instruments may exceed the sample size. A double bias correction accounts

for regularization and first-stage estimation. We derive a Bahaduj representation

with explicit remainder bounds, supporting confidence intervals for individual

coefficients and Wald tests for general linear hypotheses. We establish

quantitative coverage guarantees and characterize the Wald statistic's

asymptotic distributions under the null and local alternatives. Simulations

assess finite-sample performance, and a mouse obesity data analysis illustrates

the proposed methods' application to exploratory gene-expression analysis.

赵得霖，福州大学数学与统计学院副教授。2020年本科毕业于厦门大学经济学院统计学专业，2025年博士毕业于中国人民大学统计与大数据研究院，师从朱利平教授。主要研究方向包括独立性检验、高维统计推断、稳健统计和高维正则化方法等。

相关研究成果已发表于《Journal of Machine Learning Research》《Statistica Sinica》《Statistics and Computing》等期刊。