A nonlinear model for the conditional expectations of asset returns

Abstract

Several financial theories suggest a factor structure for the conditional expectations of asset returns. The number of factors can often be used to discriminate among competing theories. In this paper we generalize the approach of Costa et al. (1992) employing a neural network architecture to model the conditional expectation of asset returns. The identification problem of such models is discussed, and likelihood inference developed. An application illustrates the approach.