Authors : M.V. Subha, Arul Sulochana and Thirupparkadal Nambi
Abstract: The performance of predictive models depends more on predictor sets than the efficiency of the predictive algorithm. This study investigates the role of global cues as input predictors in improving the predictability of the models. It also attempts to study the efficacy of various machine learning algorithms such as support vector machines, case-based reasoning, decision trees and artificial neural network in forecasting the stock indices. It focuses on studying predictability of the widely-followed Indian stock market indices BSE SENSEX and CNX Nifty using the above-mentioned four machine learning algorithms with two separate set of predictors, a set of commonly used technical indicators and another set of daily global cues such as gold price, crude oil price, exchange rate of strong currencies, LIBOR and close price of major global stock market indices. With its lowest forecasting error values, SVM outperforms other predictive models in terms of all key performance metrics. Among the predictor sets, global cues show a higher level of predictive accuracy.
M.V. Subha, Arul Sulochana and Thirupparkadal Nambi, 2016. Improving Stock Index Predictability of Machine Learning Algorithms with Global Cues. Asian Journal of Information Technology, 15: 329-337.