Temporal and Heterogeneous Graph Neural Network for Financial Time Series Prediction (2305.08740v1)

Abstract: The price movement prediction of stock market has been a classical yet challenging problem, with the attention of both economists and computer scientists. In recent years, graph neural network has significantly improved the prediction performance by employing deep learning on company relations. However, existing relation graphs are usually constructed by handcraft human labeling or nature language processing, which are suffering from heavy resource requirement and low accuracy. Besides, they cannot effectively response to the dynamic changes in relation graphs. Therefore, in this paper, we propose a temporal and heterogeneous graph neural network-based (THGNN) approach to learn the dynamic relations among price movements in financial time series. In particular, we first generate the company relation graph for each trading day according to their historic price. Then we leverage a transformer encoder to encode the price movement information into temporal representations. Afterward, we propose a heterogeneous graph attention network to jointly optimize the embeddings of the financial time series data by transformer encoder and infer the probability of target movements. Finally, we conduct extensive experiments on the stock market in the United States and China. The results demonstrate the effectiveness and superior performance of our proposed methods compared with state-of-the-art baselines. Moreover, we also deploy the proposed THGNN in a real-world quantitative algorithm trading system, the accumulated portfolio return obtained by our method significantly outperforms other baselines.

Temporal and Heterogeneous Graph Neural Network for Financial Time Series Prediction

The paper introduces a novel approach for predicting stock price movements using a Temporal and Heterogeneous Graph Neural Network (THGNN), designed to navigate the challenges inherent in financial time series data and stock market prediction. The focus is on improving accuracy in predictions by leveraging graph neural networks that incorporate dynamically changing company relations based on historical price data rather than relying on static, handcrafted graphs or NLP-based constructions.

Methodology and Contributions

The primary innovation of this paper is the introduction of THGNN, which models the inter-company relations as a dynamic and heterogeneous graph. This approach addresses existing limitations in financial prediction models that use static graphs with fixed relationships, which may not adequately reflect the rapidly shifting dynamics in stock markets. Specifically, the paper contributes:

1. Graph Construction and Modeling: Relations among companies are represented in a temporal graph where edges signify positive or negative correlations derived from historical price sequences with two types of relationships — positive and negative. This auto-generation process circumvents traditional manual or NLP-based methods, focusing directly on price movement correlations.
2. Encoder Design: The Transformer encoder is employed to extract temporal features from historical price data, effectively capturing underlying trends and dependencies crucial for accurate prediction.
3. Two-stage Attention Mechanisms: The model deploys temporal and heterogeneous graph attention networks, allowing for adaptive learning and inference of relationships among companies across time. Temporal attention helps adjust node importance dynamically, while heterogeneous attention balances the influence of different relationship types in the graph.
4. Extensive Evaluation: Empirical evaluations conducted on S{content}P 500 and CSI 300 indices show the superior performance of the proposed method compared to state-of-the-art baselines, both in terms of prediction accuracy and portfolio optimization metrics.

Numerical Results and Implications

The experiments demonstrate that THGNN significantly outperforms various models both in terms of prediction accuracy and investment return metrics such as ARR, ASR, CR, and IR. This indicates not only a higher prediction reliability but also a better portfolio performance, thus implying possible real-world applications in quantitative trading and investment strategies.

The deployment of the model within a real-world trading platform further underscored its practical utility, where it consistently achieved higher cumulative portfolio returns compared to baseline financial prediction models.

Speculations on Future Developments

The application of graph neural networks in financial predictions showcases the potential of adaptive learning and dynamic relationship modeling, hinting at further refinements and enhancements in AI-driven investment applications. Future research could explore deeper integration of alternative data sources like news and social media into dynamic graph constructions. Moreover, the capability of THGNN to adapt to changes in graph structures presents opportunities for enhanced data mining and risk management solutions tailored to specific industry conditions and geopolitical events.

In conclusion, the paper lays a strong groundwork for the integration of graph-based methodologies in financial forecasting tasks, potentially influencing strategies in algorithmic trading and financial decision-making processes. The dynamism and adaptability of the proposed model emphasize a shift towards more nuanced, responsive AI applications in finance, capable of managing complexity and volatility inherent in global markets.