News Classification using Support Vector Machine to Model and Forecast Volatility

Volatility modelling and forecasting in the financial market is significant in risk management, monetary policy making, security valuation and portfolio creation. Standard volatility models use historical asset price returns to model and predict volatility. The purpose of this study is to add an exogenous variable to the standard volatility model. The exogenous variables used in this research are the news sentiments from Safaricom news articles extracted from Business daily, a Kenyan news publisher that consistently publishes business news. These news sentiments are the counts of positive and negative articles. Safaricom was chosen due to its huge market capitalization compared to other stocks in Kenya and it also has enough news data points for analysis. The Safaricom news articles were classified into either positive or negative using Support Vector Machine. The volatility model that incorporates news sentiments was formulated and its modelling and forecasting capabilities was compared to some standard volatility models. The empirical results indicate that the news sentiments augmented GARCH model performed best in forecasting volatility compared to standard GARCH and E-GARCH models.