Artificial Neural Networks - Empirical Analysis of Predictive Accuracy in the Indian Stock Market

Abstract

Stock market movement is driven by numerous factors, both at national and international levels, and because of the multiplicative effect of these factors, the market movement has been majorly random and very less predictable. A number of research studies have been undertaken in the past to model the stock market movement. Research analysts are continuously charting data and conducting fundamental analyses to identify stocks so as to design multi-bagger portfolio’s which can outperform the benchmark index. Any model, which can predict the stock market movement would be helpful to investors to reduce their risk exposure, increase hedging effectiveness and maximize returns. In the context of behavior modeling, it would be noteworthy to understand the complex structure of the human brain, which learns by experience. The learning is facilitated by an arrangement of neurons, which store the information about a particular event and its outcome. These neurons get trained in due course of time by encountering the events and recording the respective outcome, and by predicting the outcome when any similar event is encountered again. A similar logic could be applied to model the stock market movement by designing a neural network and developing a structure of neurons using the historic data, which can be used to predict the market movement. This research is an application of the concept of neural networking, wherein historical data is analyzed using the software MATLAB and designing an optimal network with the minimum root-mean-squared-error, using the NARX (Non-linear Autoregressive Exogenous) model. The network so obtained was applied on the current data, and an attempt was made to predict the closing price of the script under consideration for the next day. It was found that using the model, an accuracy rate of 80% was achieved in predicting the closing price of the script for the next day.