ABSTRACT
The limitations of linear forms are apparent, motivating fully nonparametric approaches to modeling functional form. Such so-called nonparametric curve fitting has an enormous literature by now but is not considered at all here. Rather, an alternative strategy is to add nonparametric bits to familiar parametric models possessing some sort of linear regression structure. That is, we enrich the class of standard parametric models for a given setting by wandering nonparametrically near (in some sense) the standard class. As a result, parts of the modeling are captured parametrically, e.g., at the least the linear structure while other aspects are captured nonparametrically. The entire enterprise falls within what is referred to as semiparametric regression modeling.
