Have Instrumental Variables Brought Us Closer to the Truth
调查255篇金融顶刊论文发现，工具变量估计值在约80%的研究中大于未处理内生性的估计值，平均为后者的9倍，并解释了这种“大得不合理”的现象，为注重识别的研究者提出最佳实践建议。
A survey of 255 papers that rely on the instrumental variable (IV) approach for identifying causal effects published in the âBig Threeâ finance journals reveals that IV estimates are larger than their corresponding uninstrumented estimates in about 80% of the studies, regardless of whether the potential endogeneity is expected to create a positive or negative bias based on economic reasoning. The magnitude of the IV estimates is, on average, nine times of that of the uninstrumented estimates even when economic insights do not suggest a downward bias of the latter. This study provides several explanations to the âimplausibly largeâ IV estimates in finance research, and proposes best practices for identification-conscientious researchers. Received January 20, 2017; editorial decision April 7, 2017 by Editor Gregor Matvos