後雙重選擇估計式的有限樣本表現

Abstract

在各個研究中，測量因果關係是非常重要的。為了維持解釋變數的外生性，研究者可能需要考慮高維度的控制變數。在此架構下，傳統所使用的最小平方估計法並不適用。為了應對這些問題，Belloni等人（2014）提出了後雙重選擇（Post-Double Selection，PDS）方法，此方法在計量文獻中受到了的重視。雖然Belloni等人(2014)已經證明了PDS方法具有漸近常態性，但在實證使用上，研究者們需要理解PDS在有限樣本中的表現。本文探討在PDS分析中，不同的統計學習方法進行雙重選擇所得出的有限樣本性質，並且藉此重新檢驗技術指標在實證中的顯著性。

Measuring causal effects is crucial in various research. Understanding the impact of an explanatory variable on an outcome variable is essential for evaluating the effectiveness of policy changes or interventions. However, to maintain the exogeneity of explanatory variables, researchers may need to consider high-dimensional control variables. In this framework, the traditional least squares method is inapplicable. To address this problem, Belloni et al. (2014) proposed the post-double selection (PDS) method. This method has received a lot of attention in econometrics. Although Belloni et al. (2014) have proved that the PDS estimator is asymptotically normal under suitable conditions, it is still important to evaluate how the PDS method behaves in finite samples. This study explores the finite-sample performance of the PDS estimator under different choices of statistical learning methods for the double selection. I also apply the PDS method to assess the significance of technical indicators in explaining stock returns.