We propose a generalization of random coefficients models, in which the regression model is an unknown function of a vector of regressors, each of which is multiplied by an unobserved error. We also investigate a more restrictive model which is additive (or additive with interactions) in unknown...