FINANCIAL NEWS INTERPRETATION FOR STOCK MARKET IMPACT PREDICTION

Abstract

Financial news of companies is an important source of information in the stock market and relevant information could heavily influence investors decision-making and hence their trading activities. As a result, it is essential for traders to identify impactful news that could cause large idiosyncratic price movements and adjust their trading strategies accordingly. Although there have been various studies on the e ect of nancial news on the stock returns, many look at the aggregate e ect of news on the market index or the e ect of news on the unadjusted returns for all stocks. This study takes into account that di erent characteristics of stocks in terms of risk pro le and measure the impact of nancial news using abnormal return which adjusts for market movement and individual stock's systematic risk. In addition, many past studies use advanced machine learning techniques Thus, this study aims to investigate the impact of nancial news in terms of risk-adjusted returns. In addition, there is an increasing trend to use advanced machine learning algorithms such as neural networks which lacks interpretability. Given that nancial news in uences the market through new information, this study aims to improve the interpretability of the prediction model by using the topic modelling technique Latent Dirichlet Allocation, which captures underlying topics in news articles. The model is also made more robust by applying ensemble learning techniques, which often improve prediction performance as compared to single prediction model. In particular, this study looks at the nancial news of individual S&P 500 stocks published by a reputable publisher, to predict which news are relevant in terms of the abnormal returns of the stocks in the following trading day. The results indicate that the use of Latent Dirichlet Allocation and ensemble methods improve the interpretability of the model as well as prediction performance.