Abstract
Identification of a coefficient associated with a time-invariant regressor (TIR) often relies on the assumption that the TIR is uncorrelated with the unobserved heterogeneity across panel units. We derive an estimator which avoids the random-effects assumption by employing a proxy for the unobserved heterogeneity thus extending the existing results on proxy variables from the cross-sectional literature.
In addition, we quantify the sensitivity of the estimates to potential violations of the random-effects assumption when no proxy is available. The utility of this approach is illustrated on the problem of implausibly high distance elasticity produced by gravity models of international trade.