A Comparison of FIML and Robust Estimates of a Nonlinear Macroeconomic Model

The prediction accuracy of six estimators of econometric models are compared. Two of rthe estimators are ordinary least squares (OLS) and full-information maximum likelihood. (FML). The other four estimators are robust estimators in the sense that they give less weight to large residuals. One of the four estimators is approximately equivalent to the least-absolute-residual (LAR) estimator, one is a combination of OLS for small residuals and LAR for large residuals, one is an estimator proposed by John W. Tukey, and one is a combination of FIML and LAR. All of the estimators account for the first-order serial correlation of the error terms. The main conclusion is that robust estimators appear quite promising for the estimation of econometric models. Of the robust estimators considered in the paper, the one based on minimizing the sum of the absolute values of the residuals performed the best. The FIML estimator and the combination of the FIML and LAR estimators also appear promising.