A general Akaike-type criterion for model selection in robust regression

Abstract

Akaike's procedure (1970) for selecting a model minimises an estimate of the expected squared error in predicting new, independent observations. This selection criterion was designed for models fitted by least squares. A different model-fitting technique, such as least absolute deviation regression, requires an appropriate model selection procedure. This paper presents a general Akaike-type criterion applicable to a wide variety of loss functions for model fitting. It requires only that the function be convex with a unique minimum, and twice differentiable in expectation. Simulations show that the estimators proposed here well approximate their respective prediction errors.