Stock Market Forecasting

Abstract

Forecasting for anything is very difficult and it has been in the domain of newlinelinear statistics from long. Linear models have the advantage that they can be newlineeasily understood and analyzed in great detail and they are easy to explain and newlineimplement. However, they may be totally inappropriate if the underlying newlinesystem is nonlinear as is the case with most of the natural real world systems newlineand stock market is one of them. Work has been done in this area by newlineresearchers using statistical and non-statistical methods with acceptable newlineprediction accuracy but I thought of developing an optimum model which is newlinenot only accurate but has less risk also. newlineThis research aims at developing an optimum model using Artificial newlineNeural Network to predict the price of a stock to a reasonable extent of newlineaccuracy. In this research, Multilayer Feedforward Neural Network with newlinebackpropagation was used to develop a model for predicting the price of a newlinestock. Backpropagation is a neural network learning algorithm. The model is newlineoptimized by using the derived parameter method of optimization. newlineEighty experiments on companies of five different sectors viz. IT, Cement, newlineAutomobile, Petroleum and Steel were performed using four Backpropagation newlineAlgorithms, namely, Gradient Descent with Momentum Backpropagation newlineTraining Algorithm, Resilient Backpropagation Algorithm, Conjugate newlineGradient Backpropagation Algoirthm with Fletcher-Reeves Updates, newlineConjugate Gradient Backpropagation with Polak-Ribiere Updates Algoirthm. newlineiii newlineThese eighty experiments were performed by varying network parameters newlineand results obtained were recorded and displayed in tabular and graphical newlineform. The results obtained were analysed and inferences were drawn using newlinederived parameter method of optimization. Optimization can be done by newlineusing the maximum value of a derived parameter dp where newlinedp = (Product of dependent variables to be maximized or 1) / (Product of newlinedependent variables to be minimized or 1). newlineAfter analyzing the results, it was observed that the optimized model using newlinethe followin