- Title of your submission: Predicting Liquidity-Aware Bond Yields using Causal GANs and Deep Reinforcement Learning with LLM Evaluation
- Team Members: Jaskaran Singh Walia, Aarush Sinha, Srinitish Srinivasan, Srihari Unnikrishnan
- All team members agree to abide by the Hackathon Rules
- This AAAI 2025 hackathon entry was created by the team during the period of the hackathon, February 17 – February 24, 2025
- The entry includes a 2-minute maximum length demo video here: Link
Predicting Liquidity-Aware Bond Yields using Causal GANs and Deep Reinforcement Learning with LLM Evaluation
This project presents a novel approach to predicting liquidity-aware bond yields by integrating three cutting-edge technologies:
- Synthetic Data Generation with Causal GANs – Models realistic bond yield time-series data while preserving causal dependencies.
- Deep Reinforcement Learning (Soft Actor-Critic - SAC) – Enhances synthetic data quality through self-learning feedback loops.
- Predictive Modeling with Fine-tuned LLMs – Utilizes a fine-tuned Qwen2.5-7B model to extract actionable trading signals, risk assessments, and volatility forecasts.
- High-Fidelity Synthetic Bond Yield Data with Causal GANs
- Self-Optimizing Data Refinement using Deep RL (SAC)
- LLM-Driven Market Insights for intelligent trading strategies
- Robust Performance Evaluation with Mean Absolute Error (MAE) metrics
For detailed experimental results, refer to Results/README.md.
├── GANS/ # Causal GAN implementation for synthetic bond yield data
├── LLMs/ # Fine-tuned Qwen2.5-7B for predictive modeling
├── MAE/ # Scripts for Mean Absolute Error computation
├── Results/ # Evaluation scripts, logs, and results summary
├── DATA/ # CSV files and datasets used in the project
├── Paper.pdf # Full research paper (Introduction, Methodology, Results)
└── README.md # Project documentation
- Extensive quantitative and qualitative evaluation results are available in the Results folder.
- Benchmarked against real-world bond yield data with low MAE scores.
- The LLM predictions align with historical financial trends and trader insights.
git clone https://github.com/AAAI-2025-Hackathon/team_38.git
cd team_38- GANS: Run synthetic data generation scripts.
- LLMs: Experiment with predictive models.
- MAE: Evaluate model performance.
- Results: Access detailed findings and analysis.
Follow the setup and execution instructions inside each folder’s README.md.
For questions or collaborations, reach out to the team via email or check the repository for further details.
🚀 Let’s redefine liquidity-aware bond yield prediction together!