Measurement errors in investment equations

We use Monte Carlo simulations and real data to assess the performance of methods dealing with measurement error in investment equations. Our experiments show that fixed effects, error heteroscedasticity, and data skewness severely affect the performance and reliability of methods found in the literature. Estimators that use higher-order moments return biased coefficients for (both) mismeasured and perfectly measured regressors. These estimators are also very inefficient. Instrumental-variable-type estimators are more robust and efficient, although they require restrictive assumptions. We estimate empirical investment models using alternative methods. Real-world investment data contain firm-fixed effects and heteroscedasticity, causing high-order moments estimators to deliver coefficients that are unstable and not economically meaningful. Instrumental variables methods yield estimates that are robust and conform to theoretical priors. Our analysis provides guidance for dealing with measurement errors under circumstances researchers are likely to find in practice.