[Paper Review] HATS: A Hierarchical Graph Attention Network for Stock Movement Prediction

Raehyun Kim, Chan Ho So+4 more|arXiv (Cornell University)|Aug 7, 2019

Stock Market Forecasting Methods24 references111 citations

TL;DR

HATS introduces a hierarchical graph attention network that selectively aggregates information from multiple relation types to predict stock movements and market index movements, outperforming baselines.

ABSTRACT

Many researchers both in academia and industry have long been interested in the stock market. Numerous approaches were developed to accurately predict future trends in stock prices. Recently, there has been a growing interest in utilizing graph-structured data in computer science research communities. Methods that use relational data for stock market prediction have been recently proposed, but they are still in their infancy. First, the quality of collected information from different types of relations can vary considerably. No existing work has focused on the effect of using different types of relations on stock market prediction or finding an effective way to selectively aggregate information on different relation types. Furthermore, existing works have focused on only individual stock prediction which is similar to the node classification task. To address this, we propose a hierarchical attention network for stock prediction (HATS) which uses relational data for stock market prediction. Our HATS method selectively aggregates information on different relation types and adds the information to the representations of each company. Specifically, node representations are initialized with features extracted from a feature extraction module. HATS is used as a relational modeling module with initialized node representations. Then, node representations with the added information are fed into a task-specific layer. Our method is used for predicting not only individual stock prices but also market index movements, which is similar to the graph classification task. The experimental results show that performance can change depending on the relational data used. HATS which can automatically select information outperformed all the existing methods.

Motivation & Objective

• Investigate how different types of corporate relations affect stock movement prediction.
• Develop a model that selectively aggregates information from multiple relation types.
• Enable both individual stock movement prediction and market index (graph-level) prediction using relational data.

Proposed method

• Propose HATS, a hierarchical graph attention network that computes relation-type specific summaries and fuses them into node representations.
• Use a two-layer hierarchical attention mechanism: first layer attends over neighbors within each relation type, second layer attends over relation types to form aggregated node representations.
• Initialize node features with a feature extraction module (LSTM for stock prediction, GRU for index prediction).
• Incorporate a task-specific module for individual stock prediction via a softmax classifier on updated node representations.
• Apply graph pooling (mean pooling) to obtain graph representations for index prediction, combined with the graph-encoded features for final prediction.

Experimental results

Research questions

• RQ1 Which types of relational data most benefit stock movement prediction?
• RQ2 Can a model selectively aggregating information across relation types outperform baselines in both node and graph (index) prediction tasks?
• RQ3 How does the hierarchical attention mechanism affect the utilisation of relation types and neighbor information?

Key findings

• HATS outperforms existing baselines on stock movement prediction when using selectively chosen relations.
• The method achieves higher performance when useful relations are included; irrelevant relations can degrade accuracy.
• For index movement prediction, graph pooling combined with relational features improves predictive capability compared to baselines.
• The study demonstrates that performance varies with the quality and relevance of relational data.
• The paper reports a substantial improvement in practical metrics: Sharpe ratio increases by 19.8% and F1-score by 3% over baselines.