Uniform Inference in Linear Error-in-Variables Models: Divide-and-Conquer

Abstract

It is customary to estimate error-in-variables models using higher-order moments of observables. This moments-based estimator is consistent only when the coefficient of the latent regressor is assumed to be non-zero. We develop a new estimator based on the divide-and-conquer principle that is consistent for any value of the coefficient of the latent regressor. In an application on the relation between investment, (mismeasured) Tobin's $q$ and cash flow, we find time periods in which the effect of Tobin's $q$ is not statistically different from zero. The implausibly large higher-order moment estimates in these periods disappear when using the proposed estimator.