A comparison of linear regression and neural network methods for predicting excess returns on large stocks

摘要: Recent studies have shown that there is predictable variation in returns of financial assets over time. We investigate whether the predictive power economic and variables employed above can be enhanced if statistical method linear regression replaced by feedforward neural networks with backpropagation error. A shortcoming too many free parameters allow network to fit training data arbitrarily closely resulting an "overfitted" network. Overfitted poor generalization capabilities. explore two methods attempt overcome this reducing complexity The results our experiments confirm network, while making better predictions for within-sample data, makes out-of-sample data. explored paper, clearly help improve forecasts. show one cannot say forecasts are conditionally efficient respect any degree confidence. However, some