Optimization of Predicted Portfolio Using Various Autoregressive Neural Networks

This work presents a neural networks approach for stock returns and uses mean-variance model for the selection of predicted portfolio thus formed. Four types of different neural network models have been used and their outputs have been compared at various regression orders. A new type of predictor called autoregressive moving reference neural network predictor has been used in all the four neural network models. In this predictor the differences between the values of the series of returns and a determined past value are the regression variables. To evaluate the performance of the predictor, various error measures have been used, taking the average of these error measures, the overall performance of the predictor has been tested. Experiments with real data from National stock exchange of India (NSE) were employed to examine the accuracy of this method.