Leveraging BiLSTM-GAT for enhanced stock market prediction: a dual-graph approach to portfolio optimization
This paper presents a dual-graph BiLSTM-GAT-AM model for enhanced stock market prediction and optimized portfolio management.
What it examines
The paper introduces a novel BiLSTM-GAT-AM model that combines bidirectional LSTM with graph attention networks using dual graphs based on technical and fundamental similarities. It aims to improve stock market prediction and portfolio optimization by capturing both temporal dynamics and inter-company relationships for better decision-making.
What it concludes
The dual-graph approach improved prediction accuracy and portfolio performance, effectively merging technical and fundamental insights. Results suggest applications in real-time trading, portfolio optimization, and risk analysis, with future work recommended on integrating more data sources and real-world market testing.
Evidence objects
Researchers introduce a novel dual-graph model combining BiLSTM-GAT-AM architecture with technical and fundamental analyses, using DTW-based graph for price movements and industry similarity graph for longer-term relationships, enhancing stock predictions.
key_findings bullet 1 · key_findings · validation V0
Extensive backtesting on S&P500 data reveals significantly improved Sharpe ratios compared to traditional methods, with rigorous ablation studies and multiple evaluation metrics like Mean Squared Error and Mean Absolute Error.
key_findings bullet 2 · key_findings · validation V0
Despite strong performance, the study notes potential weaknesses, including heavy reliance on historical data and operational complexity, while offering a notably promising framework bridging predictive accuracy with actionable trading decisions.
key_findings bullet 3 · key_findings · validation V0
Integrating BiLSTM with Graph Attention Networks, the paper introduces a dual-graph approach that captures technical and fundamental relationships, offering a distinct, unified framework for portfolio optimization and market prediction. Its originality and novelty, exemplified by the $DualGraph$ structure, create compelling insights that remarkably advance quantitative and computational finance research.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … portfolio and conduct backtesting on the test dataset. Empirical results indicate that our portfolio … of our approach in stock market prediction and portfolio optimization. …
Source row: 1198 · abstract type: snippet