A1382
Title: Challenges of cross-validation in post-double-Lasso: A Monte Carlo study
Authors: Adrian Drexel - University of Regensburg (Germany) [presenting]
Abstract: Monte Carlo simulations evaluate the performance of post-double-Lasso, revealing that using cross-validated $\lambda$ (CV-$\lambda$) in Lasso can be disadvantageous compared to the X-independent $\lambda$, particularly in small samples. Additionally, in settings characterized by approximate sparsity, even post-Lasso with CV-$\lambda$ occasionally outperforms post-double-Lasso with CV-$\lambda$. These results highlight the importance of the penalty choice in high-dimensional econometric models.
