Stock trading decisions using ensemble-based forecasting models: a study of the Indian stock market

In this paper, a two-phase ensemble framework comprising of various non-classical decomposition models, namely, Empirical Mode Decomposition, Ensemble Empirical Mode Decomposition and Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), and machine learning models, namely, Artificial Neural Network and Support Vector Regression (SVR), is proposed for predicting the stock prices. In the first phase, historical stock prices are decomposed to a set of subseries. In the second phase, each subseries is forecasted using machine learning algorithms. Lastly, forecasts of individual subseries are added to obtain the final forecasts. The proposed framework is tested on constituents of Nifty index for a period of 8 years ranging from 2008 to 2015. Performance of the models were analysed using root mean square error. Further, the results were validated statistically using Wilcoxon Signed Rank Test and Friedman Test. CEEMDAN-SVR model outperformed the remaining models. In addition, trading rules were illustrated to determine the optimal timing for buying/selling the stocks. Trading rules based on ensemble models yielded higher return on investment compared to traditional Buy-and-Hold strategy.