Empirical estimation of multilayer perceptron for stock market indexes

Abstract

The return on investment of stock market index is used to estimate the effectiveness of an investment in different savings schemes. To calculate Return on Investment, profit of an investment is divided by the cost of investment. The purpose of the paper is to perform empirical evaluation of various multilayer perceptron neural networks that are used for obtaining high quality prediction for Return on Investment based on stock market indexes. Many researchers have already implemented different methods to forecast stock prices, but accuracy of the stock prices are a major concern. The multilayer perceptron feed forward neural network model is implemented and compared against multilayer perceptron back propagation neural network models on various stock market indexes. The estimated values are checked against the original values of next business day to measure the actual accuracy. The uniqueness of the research is to achieve maximum accuracy in the Indian stock market indexes. The comparative analysis is done with the help of data set NSEindia historical data for Indian share market. Based on the comparative analysis, the multilayer perceptron feed forward neural network performs better prediction with higher accuracy than multilayer perceptron back propagation. A number of variations have been found by this comparative experiment to analyze the future values of the stock prices. With the experimental comparison, the multilayer perceptron feed forward neural network is able to forecast quality decision on return on investment on stock indexes with average accuracy rate as 95 % which is higher than back propagation neural network. So the results obtained by the multilayer perceptron feed forward neural networks are more satisfactory when compared to multilayer perceptron back propagation neural network.