Estimating nonlinear effects of management styles in the US equity market using a classifier neural network

The authors use a classifier neural network to predict stock factor returns. Training files with four months of fundamental factor data classify stocks as "buy" "hold" or "sell" based on the next month\´s return. Out-of-sample testing indicates that this technique is effective. The classifier\´s probability predictions for each state are used to estimate expected return and variance for the following month