Treating measurement error in Tobin's q

We compare the ability of three measurement error remedies to deliver unbiased estimates of coefficients in investment regressions. We examine high-order moment estimators, dynamic panel estimators, and simple instrumental variables estimators that use lagged mismeasured regressors as instruments. We show that recent investigations of this question are largely uninformative. We find that all estimators can perform well under correct specification, all can be biased under misspecification, and misspecification is easiest to detect in the case of high-order moment estimators. We develop and demonstrate a minimum distance technique that extends the high-order moment estimators to be used on unbalanced panel data