Developing a Nonlinear Model to Predict Stock Prices in India: An Artificial Neural Networks Approach [Dagger]

(ProQuest: ... denotes formulae omitted.)IntroductionThere have been innumerable techniques developed to predict stock markets as the results have been economically fruitful to its developers. These techniques rely on the quality of information used in different prediction models; however, many uncertain and interrelated factors also affect stock prices and their importance may be difficult to measure numerically. At the outset, stock markets are complicated and not entirely comprehensible. The returns of the stock market are difficult to predict. A vast amount of research has been carried out to analyze the complexity, nonlinearity, nonstationarity and chaotic nature of the stock market in order to come out with a better stock market prediction model.The present study aims at predicting the stock prices of CNX Nifty 500 using assorted independent variables using Artificial Neural Networks (ANNs). Neural Networks (NNs) are computer programs consisting of computing nodes and interconnections between nodes (Yao et al., 1999). They are recognized as effective tools for financial forecasting (Yao and Tan, 2001) and can 'learn' from experience as do humans, cope with nonlinear data, and deal with partially understood application domains, such as stock market behaviors. Moreover, the fundamental stock market indicators, gross domestic product, interest rate, gold prices and exchange rates and technical indicators, including closing prices, opening prices, highest prices and lowest prices, can be incorporated into neural networks to help improve predictive outputs (Yao et al., 1999).Owing to an ability to learn nonlinear mappings between inputs and outputs, ANNs are one of the more popular machine learning methods used for predicting stock market prices (Egeli et al., 2003). One of the major application areas of ANNs is forecasting. Recent research activities in ANNs have shown that ANNs have powerful pattern recognition capabilities (Widrow et al., 1994).Literature ReviewYoungohc and George (1991) demonstrated that the NN approach is capable of learning a function that maps inputs to output and encoding it in magnitudes of the weights in the network connection. They compared the technique with multivariate discriminant analysis approach and indicated that the NN approach can significantly improve the predictability of stock price performance.Trippi and DeSieno (1992) applied a neural network system to model the trading of Standard & Poor's 500 index futures. They found that the NN system model outperforms passive investment in the index. Based on the empirical results, they favor the implementation of NN systems into the mainstream of financial decision making.Yao et al. (1999) did a seminal work on neural network model to relate technical indicators to future trends in the Kuala Lumpur Composite Index (KLCI) of Malaysian stocks. The technical indicators used as inputs for the NN model included moving average, momentum and Relative Strength Index (RSI) ). Their experiment used many neural networks with the training method of back-propagation. However, they did not train the NNs sufficiently nor use fundamental factors for their predictions. As a result, the robustness of their model for prediction involving other time periods was not good due to insufficient training and lack of usage of fundamental factors.Ray et al. (2000) studied the relationship between the real economic variables and the capital market in the Indian context for the period after liberalization. In this paper, the authors apply modern ANNs and VAR to study nonlinear relationship between the variables. The results show that variables like the interest rate, output, money supply, inflation rate and exchange rate have considerable influence on the stock market movement for the period of study. The study verified the fact that macroeconomic variables affect the stock market returns in the post-reform era. …