Consequences and Detection of Misspecified Nonlinear Regression Models

Abstract Under general conditions given here, the least squares estimator for the parameters of a misspecified nonlinear model converge strongly to the parameters of a (weighted) least squares approximation to the true model. With additional conditions, the least squares estimator is asymptotically normal. A new, specification-robust estimator of the covariance matrix is obtained, which simplifies to the usual estimator when the model is correct up to an independent additive error. The properties of the approximation and the covariance estimator are exploited to yield new tests for model misspecification. These results are applied to two examples in economics.