Short Term Firm-Specific Stock Forecasting with BDI Framework

Abstract

In today’s information age, a comprehensive stock trading decision support system which aids a stock investor in decision making without relying on random guesses and reading financial news from various sources is the need of the hour. This paper investigates the predictive power of technical, sentiment and stock market analysis coupled with various machine learning and classification tools in predicting stock trends over the short term for a specific company. Large dataset stretching over a duration of ten years has been used to train, test and validate our system. The efficacy of supervised non-shallow and prototyping learning architectures are illustrated by comparison of results obtained through myriad optimization, classification and clustering algorithms. The results obtained from our system reveals a significant improvement over the efficient market hypothesis for specific companies and thus strongly challenges it. Technical parameters and algorithms used have shown a significant impact on the predictive power of the system. The predictive accuracy obtained is as high as 70–75% using linear vector quantization. It has been found that sentiment analysis has strong correlation with the future market trends. The proposed system provides a comprehensive decision support system which aids in decision making for stock trading. We also present a novel application of the BDI framework to systematically apply the learning and prediction phases.