Stock Movement Prediction That Integrates Heterogeneous Data Sources Using Dilated Causal Convolution Networks With Attention

The purpose of this research is to develop a high performing model for stock movement prediction utilizing financial indicators and news data. Until recently, the majority of prediction models have employed only the financial indicators, but they possess the risk of missing unconventional agitators that can be derived from other heterogeneous sources. To address this, few research studies began to explore the use of news data and other social features along with financial indicators. In this work, we propose a novel integrative approach to effectively blend views from the news and financial time series. We generate event-knowledge representations from news data by capturing direct and inverse relationships among event tuples and then apply the attention mechanism to infer inter-day relationships among the representations. To capture temporal dynamics of financial indicators, we further integrate an attention augmented dilated causal convolutional network. We report empirically that our model achieves a substantial 5% improvement from 68.81% to 74.29% in stock movement prediction for the Standard & Poor’s 500 (S&P500) index and companies over existing models.