Financial Modeling using ANN Technologies : Result Analysis with Different Network Architectures and Parameters

The future is always uncertain and uncertainty leads to risk. The elaborate exercise of forecasting and making predictions is a course of action to which organizations take recourse in order to minimize this uncertainty and the risks arising therefrom. Business decisions, and that too financial business decisions, depend heavily on future predictions. Financial forecasting is a management technique that refers to the estimation of information for the future based on the availability of past financial conditions. Neural networks have been a popular choice among researchers when it comes to modeling financial forecasting. However, the factors like learning rate, momentum, and architectural configuration affect a lot if not selected properly. The selection of an effective architecture and parameter combination improves the accuracy and acceptability of the results by many folds. In this paper, an analysis was made to measure the impact of various combinations of neural architecture and parameters through the application of artificial neural networks (ANN) as a forecasting tool using Zaitun Statistical Package. It was observed that the architecture with 12 neurons in hidden layer and learning rate of 0.03 produced the minimum error.