GCN-based stock relations analysis for stock market prediction

Area of Science:

• Quantitative Finance
• Machine Learning
• Econometrics

Background:

• Traditional stock price prediction models often overlook inter-stock relationships, focusing solely on individual historical data.
• Prior research indicates that incorporating stock correlations can significantly enhance predictive performance.

Purpose of the Study:

• To propose a novel unified time-series relational multi-factor model (TRMF) for stock price prediction.
• To develop a self-generating relations (SGR) algorithm for automatic extraction of relational features.
• To integrate stock relations with multi-dimensional features for improved price forecasting.

Main Methods:

• Development of the TRMF model incorporating a self-generating relations (SGR) algorithm.
• Integration of automatically extracted relational features with other multi-dimensional stock data.
• Experimental validation using NYSE and NASDAQ datasets.
• Comparison against established methods like Attention Long Short-Term Memory (Attn-LSTM), Support Vector Regression (SVR), and the multi-factor framework (MF).

Main Results:

• The TRMF model demonstrated superior performance compared to Attn-LSTM, SVR, and MF.
• The proposed model achieved a higher expected cumulative return rate.
• The TRMF model exhibited a lower risk of return volatility.

Conclusions:

• The TRMF model effectively leverages inter-stock relationships for enhanced price prediction.
• Integrating relational features alongside multi-dimensional data offers significant advantages over existing methods.
• The model provides a promising approach for improving stock market forecasting accuracy and risk management.