Financial news-based stock movement prediction using causality analysis of influence in the Korean stock market

With the advent of the Big Data era and the development of machine learning technologies, predicting stock movements by analyzing news articles, which are unstructured data, has been studied actively. However, so far no attempts have been made to utilize the asymmetric relationship of firms. Thus far, most papers focus on only the target firm, and few papers focus on the target firm and relevant firms together. In this article, we propose a novel machine learning model to forecast stock price movement based on the financial news considering causality. Specifically, our method analyzes the causal relationship between companies, and it accounts for the directional impact within the Global Industry Classification Standard sectors. In our proposed method, transfer entropy is used to find causality, and multiple kernel learning is used to combine features of target firm and causal firms. Based on a Korean market dataset and out-of-sample test, our experimental results reveal that the proposed causal analytic-based framework outperforms two traditional state-of-the-art algorithms. Furthermore, the experimental results show that the proposed method can predict the stock price directional movements even when there is no financial news on the target firm, but financial news is published on causal firms. Our findings reveal that identifying causal relationship is important in prediction problems, and we suggest that it is important to develop machine learning algorithms and it is also important to find connections with well-established theories such as the complex system theory.