High-Dimensional Methods in Microeconometrics

在这个项目中，我提出了在微观计量经济学中实施高维方法的几个进展。该项目的目的是为对这些方法感兴趣的应用微观计量经济学家提供有用的理论和模拟结果。具体地说，在以前工作的基础上，我计划在几个关键方向上推进用于多个结果校正的稀疏组套索的实现：首先，我计划检查先前建议的调优参数选择方法，以了解它们在数据驱动的调优参数选择中的用处。我可以证明交叉验证会导致关于真实信号结构的误导性结果，因此我们需要寻找其他方法。其次，我打算研究一条不同的路径：根据政策目标和/或经济理论选择调整参数。第三，我想比较其他多重测试程序和最近（高维）的多重测试修正与稀疏组套索方法在一个精心设计和经验动机模拟研究中的进展，以找出这些方法中哪一个是最适合的经验研究设置。

In this project I propose several advancements for the implementation of high-dimensional methods in microeconometrics. The aim of this project is to be derive useful theoretical-, as well as simulation results for applied micro-econometricians interested in these methods. Specifically, building on previous work, I plan to advance the implementation of the sparse-group lasso for multiple outcome correction in several crucial directions: First, I plan to examine previously suggested methods for tuning parameter selection for their usefulness in data driven-tuning parameter selection. I can show that cross-validation leads to misleading results concerning the structure of the true signal and we therefore need to look for other methods. Second, I plan to examine a different path: choosing the tuning parameter based on policy objectives and/or economic theory. Third, I want to compare other multiple testing procedures and recent (high-dimensional) advances in multiple testing corrections with the sparse-group lasso method in a well-designed and empirically motivated simulation study to find out which of these methods is the most suitable in which empirical research set-up.