Estimation in the Linear IV Model with Many IVs
This project compares the finite-sample performance of post-Lasso (Belloni, Chen, Chernozhukov and Hansen, 2012), RJIVE (Hansen and Kozbur, 2014), and RLMIL (Carrasco and Tchuente, 2015) in the linear IV model with many IVs.
Description of the data generating process and implementation of the estimators can be find in
Zhento Shi (2017): "Econometric Estimation with High-Dimensional Moment Equalities", Journal of Econometrics
- Eddie Gao, my research assistant, develops the code under my supervision.
- I design the DGP, and develop the code of REL and BC-REL for comparison, as in the paper.
- We thank G.Tchuente for sharing his RLIML code.
- Belloni, Chen, Chernozhukov and Hansen share the post-Lasso code with their published paper.
The code is written in Matlab.
master_IV.mis the master file for the simulation.
dgpLinearIV.mgenerates the data
zin each simulation.
post_lasso.mimplements the post-lasso (BCCH, 2012).
LassoShooting2.mdoes the pre-selection of instruments.
tsls.mimplements two-stage-least-square estimation.
process_optionsare supportive functions to drive
RJIVE.mimplements RJIVE (Hansen and Kozbur, 2014).
Rliml.mimplements regularized LIML estimation under Tikhonov regularization (Carrasco and Tchuente, 2015).
output_bias_rmse.mcomputes the bias and RMSE for the estimation result.