Forecasting Stock Prices using Neural Networks on a Beowulf Cluster

R.M. Rahman, R.L. Thulasiram, and P. Thulasiraman (Canada)


Neural Networks, Parallel Algorithms, Financial Forecast ing, Performance Evaluation.


In this paper, we focus on the problem of stock price fore casting in a parallel environment using Neural Network (NN). Currently, with traditional algorithms training a neu ral network (NN) for stock price forecasting consumes pro hibitively large amount of time before it can be effectively utilized. We develop a parallel algorithm for a Backprop agation NN and implement it on a Beowulf cluster using Message Passing Interface (MPI). By establishing strong correlation between the output and the target values we show that the training and testing errors are reduced signifi cantly, which ensures the accuracy of the predicted results. We have shown that the training time is reduced signifi cantly with multiprocessors, although the performance of the algorithm with 8 processors is only 50%. However, as a first attempt (to the authors’ knowledge) to study stock price forecasting using NN in a parallel environment, the current results are quite encouraging to pursue this research work further.

Important Links:

Go Back