Job Market Paper
"High Dimensional GMM Estimation and Inference Using the Variant of the Conservative Lasso" (draft coming soon)
Presentation: Midwest Econometrics Group 2022 Conference
I consider estimation and inference in high dimensional GMM models. I propose an estimator using the two-step variant of the conservative Lasso without the sub-gaussian and two stage sparsity assumptions. In my model, I allow the number of instruments and structural parameters to be greater than the sample size. I first derive an oracle inequality for my two-step GMM estimator. Then I show that the upper bound in my oracle inequality is not worse than that given by Caner and Kock (2019). The strong oracle property of my estimator is also showed which is not established in previous high dimensional GMM estimation studies. Afterwards, I provide the desparsified version of my estimator and show that it leads to uniformly valid inference even with the conditionally heteroskedastic errors. The results of simulations show that my estimator has a good performance especially in the situation of the high dimensional instruments and parameters in terms of mean squared error, test size and power.
Work in Progress
"High Dimensional Correlated Random Effects Models With Structured Sparsity Estimator"