Multi-Graph Convolutional Network for Relationship-Driven Stock Movement Prediction

Abstract

Stock price movement prediction is commonly accepted as a very challenging task due to the volatile nature of financial markets. Previous works typically predict the stock price mainly based on its own information, neglecting the cross effect among involved stocks. However, it is well known that an individual stock price is correlated with prices of other stocks in complex ways. To take the cross effect into consideration, we propose a deep learning framework, called Multi-GCGRU, which comprises graph convolutional network (GCN) and gated recurrent units (GRU) to predict stock movement. Specifically, we first encode multiple relationships among stocks into graphs based on financial domain knowledge and utilize GCN to extract the cross effect based on these pre-defined graphs. To further get rid of prior knowledge, we explore an adaptive relationship learned by data automatically. The cross-correlation features produced by GCN is concatenated with historical records and fed into GRU to model the temporal dependency of stock prices. Experiments on two stock indexes in China market show that our model outperforms other baselines. Note that our model is rather feasible to incorporate more effective stock relationships containing expert knowledge as well as learn relationship on the basis of data dynamically.

Stock Movement Prediction Based on Bi-typed Hybrid-relational Market Knowledge Graph via Dual Attention Networks

“…In general, traditional SMP methods mainly can be categorized into two classes: technical analysis and fundamental analysis, according to the different types of the available stock own information they mainly used. Another major aspect for yielding better stock prediction is to utilize the stock connection information [Chen et al 2018;Cheng and Li 2021;Sawhney et al 2020b;Ye et al 2020]. We review them in the following.…”

“…Li et al [2020a] proposed a LSTM Relational Graph Convolutional Network (LSTM-RGCN) model, which models the connection among stocks with their correlation matrix. Ye et al [2020] encoded multiple relationships among stocks into graphs based on financial domain knowledge and utilized GCN to extract the cross effect based on these pre-defined graphs for stock prediction. Sawhney et al [2020b] proposed a spatio-temporal hypergraph convolution network for stock movement forecasting.…”

“…In particular, we construct a more comprehensive heterogeneous market graph that contains explicit relations, implicit relations and executive relations. (2) Different from most existing models that can only model homogeneous stock graph [Chen et al 2018;Cheng and Li 2021], or heterogeneity of stock explicit relations [Feng et al 2019b;Sawhney et al 2021Sawhney et al , 2020aYe et al 2020], which fall down in modeling heterogeneity of entities in real market, we propose a novel dual attention networks that is able to model bi-typed heterogeneous entities and hybrid-relations in newly constructed market graph of stock for its spillover effects. (3) To the best of our knowledge, this work is the first attempt to study stock movement prediction via heterogeneous GNNs.…”

“…3.1.1 Bi-typed Entities. Most existing SMP methods solely learn from company relationships in market [Kim et al 2019;Li et al 2020a;Ye et al 2020]. In fact, in most stock markets there are also significant associated executives for listed companies, with rich associative information about these companies [Cai et al 2016 3.1.2 Hybrid-relations.…”

“…Hence, the datasets are roughly balanced. Some indicators [Sousa et al 2019;Ye et al 2020] are selected to demonstrate the effectiveness of the proposed method, i.e. Directional Accuracy (DA), Precision, (AUC), Recall, F1-score.…”