Econ Papers Daily / 期刊论文
工具变量模型的识别与维度稳健检验
An identification-and dimensionality-robust test for instrumental variables models
摘要
Using modifications of Lindeberg’s interpolation technique, I propose a new identification-robust test for the structural parameter in a heteroskedastic instrumental variables model. My analysis accommodates both scenarios where the number of instruments is much larger than the sample size and scenarios with few instruments, thus making the test broadly applicable. The proposed test statistic has a limiting chi-squared distribution so long as an auxiliary parameter can be consistently estimated. This is possible with standard methods when the number of instruments is small or using machine learning methods when the intruments are high-dimensional. To improve power, a simple combination with the sup-score statistic is proposed. The flexibility of my asymptotic approach allows for higher quality first-stage estimators to be used in the construction of the test statistic which provides noticeable power improvements over existing tests, both in Monte Carlo simulations calibrated to empirical data and in an application to the returns to education. Motivated by empirical practice, I also point out that first-stage F-statistics calculated on LASSO-selected variables may be misleading indicators of identification strength.
中文摘要
通过对林德伯格插值技术的改进，我提出了一种用于异方差工具变量模型中结构参数的新识别稳健检验。我的分析既适用于工具数量远大于样本量的情形，也适用于工具数量较少的情形，从而使该检验具有广泛的适用性。只要辅助参数能够一致估计，所提出的检验统计量便具有极限卡方分布。当工具数量较少时，可使用标准方法实现这一点；当工具为高维时，可使用机器学习方法实现。为提高功效，提出了与sup-score统计量的一种简单组合。我的渐近方法的灵活性允许在构造检验统计量时使用更高质量的第一阶段估计量，这在基于经验数据校准的蒙特卡洛模拟中以及在教育回报率的应用中，均相较现有检验提供了显著的功效改进。受实证实践启发，我还指出，基于LASSO所选变量计算的第一阶段F统计量可能是识别强度的误导性指标。