Weak Identification: Causes, Consequences, and Solutions.
Location
Public Policy : 367
Date & Time
February 18, 2015, 4:00 pm – 5:00 pm
Description
Summary:
In contexts where the data contains only limited information about model parameters, commonly-used econometric techniques may break down, resulting in tests with distorted size and biased estimates. The best-studied example of this phenomenon is the weak instruments problem, which arises when instrumental variables are only weakly correlated with the endogenous regressors of interest, but such issues also arise in a wide range of other contexts. In this talk, I review the recent literature on weak identification, discussing why the usual econometric techniques may break down, how such problems may be detected, and recent advances in identification-robust inference.