Event-Driven Learning of Systematic Behaviours in Stock Markets

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 markets. Motivated by this, we leverage financial event streams to train a classification neural network that detects latent event-stock linkages and stock markets' systematic behaviours in the U.S. stock market. Our proposed pipeline includes (1) a combined event extraction method that utilizes Open Information Extraction and neural co-reference resolution, (2) a BERT/ALBERT enhanced representation of events, and (3) an extended hierarchical attention network that includes attentions on event, news and temporal levels. Our pipeline achieves significantly better accuracies and higher simulated annualized returns than state-of-the-art models when being applied to predicting Standard&Poor 500, Dow Jones, Nasdaq indices and 10 individual stocks.

Hybrid Information Mixing Module for Stock Movement Prediction

“…Research [6], [21] has been conducted to predict stock price movement by identifying correlations between companies using the graph network. Many eventdriven methods [16], [27], [38] predict recent stock price movement trends using events extracted from news and social media. In addition, several studies [39], [40] have analyzed temporal factors of the stock market and extracted the characteristics of these temporal factors to reflect their effects on stock market fluctuations in predicting stock movement.…”

“…Fundamental analyses use other variables, such as text and company event information, along with historical prices [6], [17]. First, social mood determines the investment behavior of stock investors and corporate managers [38]; thus, text information analyses that determine buying and selling by stock investors must be studied. Investors primarily gauge social mood via text information, such as news or social media, and reflect it in their investment behavior.…”

“…They proposed a bi-level event detection model using global and local information to identify corporate events. Wu et al [38] used a financial event stream to train classification neural networks containing a combined event extraction method, BERT/ALBERT, and extended hierarchical attention networks to detect latent eventstock links and systematic stock market behavior. Additionally, Xu et al [47]proposed a relational event-driven stocktrend forwarding framework by constructing a stock graph considering the influence of event information in the stock market and designing a new propagation layer to spread the event information effect from related stocks.…”

Change-Oriented Summarization of Temporal Scholarly Document Collections by Semantic Evolution Analysis

“…For example, in [61,62], a BERT model based on MIMIC-III dataset [63] is built to improve the temporal extraction task on medical annotated data called THYMES [64]. Such works are also observed in [65] based on story corpus in ROCStories [66], and as temporal event extraction in financial related news [67]. From a high-level of view, it can be generally stated that pre-trained language models are usually being used as a method for text vectorization, in specific domain, and for specific application.…”