Forecasting stock trend direction with support vector machine

Abstract

Financial markets facilitate international trade, are indicative of the future prospects of organizations and economies, and are drivers of economic growth (Hsu, Lessmann, Sung & Johnson, 2016). Hence, the prediction of financial market assets with reference to previously observed data has drawn considerable attention as an active research area (Zhu, Wang, Xu & Li, 2008). The financial market is a non-linear dynamic system that is influenced by many interdependent factors (Abu-Mostafa & Atiya, 1996). Such are macroeconomics, political sentiments, news, general economic conditions as well as the expectations and psychology of active investors (Novak & Veluscek, 2015). As a result of these ambiguous complexities coupled with the large amount of noise in financial market data, modelling stock trends has been regarded as a challenging task (Polimenis & Neokosmidis, 2014). This paper therefore addresses the stock trend prediction problem as a classification task and models it using Support Vector Machine (SVM). It also explores different feature selection algorithms applicable for SVM and finally draw comparisons amongst results generated by other machine learning methods.