Event-Driven Learning of Systematic Behaviours in Stock Markets

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2434–2444 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2434 Event-Driven Learning of Systematic Behaviours in Stock Markets Xianchao Wu NVIDIA xianchaow@nvidia.com, wuxianchao@gmail.com Abstract It is reported that financial news, especially financial events expressed in news, provide information to investors’ long/short decisions and influence the movements of stock mar- kets. Motivated by this, we leverage finan- cial event streams to train a classification neu- ral network that detects latent event-stock link- ages and stock markets’ systematic behaviours in the U.S. stock market. Our proposed pipeline includes (1) a combined event ex- traction method that utilizes Open Informa- tion Extraction and neural co-reference reso- lution, (2) a BERT/ALBERT enhanced repre- sentation of events, and (3) an extended hi- erarchical attention network that includes at- tentions on event, news and temporal levels. Our pipeline achieves significantly better ac- curacies and higher simulated annualized re- turns than state-of-the-art models when be- ing applied to predicting Standard&Poor 500, Dow Jones, Nasdaq indices and 10 individual stocks. 1 Introduction It is widely reported that financial news, especially financial events expressed in the news, influence the movements of stock markets (Ding et al., 2014, 2015, 2016; Hu et al., 2018; Ding et al., 2019; Huang et al., 2019; Glasserman et al., 2019). For example, one company’s releasing of new prod- ucts brings novel time-sensitive factors to the stock movement of that company and further to the whole market; changing of interest rates from the central bank of the country bring changes of currency flu- idity invested in the stock market. Frequently, good news brings increased estimation of the future value of the target company and consequently a higher stock price tendency. Warren Buffett said in his interview1 that he fre- 1https://www.youtube.com/watch?v=Pqc5 6crs56s quently spent 5 hours reading news and financial reports to manage his portfolios. Regarding the rich information included in daily published news articles, we are encouraged to read them not by our- selves but by employing deep learning algorithms to guide our investments. One solution is to auto- matically collect thousands of news published ev- ery day, extract financial events from the news, and rank the importance of the events to predict market behaviours. In particular, we are aiming at quan- tifying the latent relevance of between financial event streams and target stocks’ price volatilities. However, this is not a trivial task and there are a number of challenges. First, it is ambiguous to define good or bad news. For example, bad news for one company could be worse news to its downstream supply-chain partner companies yet good news to its competitors. It is time-consuming and untrackable to annotate news manually and train a sentiment analysis model on it, consider- ing that the generalization of the model to novel financial events is fragile. Second, how shall we express the impact of the news published in dif- ferent days? Generally, news articles have their individual and accumulated influences to the in- vestors. For …

Content truncated to 500 words