Graph-augmented reinforcement learning for stock trading
A TD3 trading agent that uses Node2Vec embeddings of how stocks relate, outperformed RL baselines on Dow 30 data, and explains its own strategy.
- Capability
- Applied ML and research
- Role
- Researcher, MSc thesis
- Domain
- Graduate research, quantitative finance (Research and public sector)
- Period
- Sep 2022 – Sep 2024
Problem
RL trading agents usually treat each stock independently, even though stocks move together by sector and supply chain. Their strategies are also hard to explain, which makes them hard to trust.
What I built
- A graph-augmented TD3 agent: Node2Vec embeddings of the relationships between stocks feed the agent's state.
- Sector clustering and technical indicators to improve risk-adjusted returns.
- Post-hoc visualizations that make the agent's strategy auditable.
Outcome
- Outperformed RL baselines on Dow 30 data.
- MSc in Computer Science (AI specialization) with a 4.0 GPA and an Ontario Graduate Scholarship.
Capabilities
Stack
- Python
- TD3
- Node2Vec
- Reinforcement learning
Skills shown
- Python
From the log
- 3 Sep 2024
Completed an MSc thesis on graph-augmented reinforcement learningAn RL agent is a sequential decision-maker, like an LLM agent. Designing rewards, comparing against baselines and explaining behavior is the same discipline I now apply to agent evaluation. - 6 Sep 2022
Started graduate research on reinforcement learning for tradingThe research foundation in agents, reward design and evaluation that the rest of this log builds on.