Heckman-type maximum likelihood estimators of the gravity equation: A Monte Carlo study

We propose a heteroskedastic Heckman model to consistently estimate the gravity equation in the presence of heteroskedasticity and zero trade values. The Heckman-type Maximum Likelihood estimator allows for different error term distributions, non-linear forms of both selection and measure equations, and explicitly estimates the variance process. Monte Carlo simulations show that the proposed Heckman technique outperforms traditional estimators of gravity equation. Unlike what is commonly claimed in the literature, we report significantly lower GDP elasticities and we find that the conditional bilateral trade variance is not likely to be proportional to the mean. The proposed Heckman model could be used for a wide range of other applications.