TradeMaster: A Holistic Quantitative Trading Platform Empowered by Reinforcement Learning
We introduce TradeMaster, an open-source RL-powered quantitative trading platform covering the full pipeline from data processing to live deployment, accepted at NeurIPS 2023.
We introduce TradeMaster, a holistic open-source platform for reinforcement learning-based quantitative trading, accepted at NeurIPS 2023.
⭐ 2.7k GitHub stars
Motivation
Applying reinforcement learning to financial trading is deceptively difficult. Beyond model design, researchers must wrangle heterogeneous data sources, build realistic market simulators, implement fair evaluation protocols, and eventually bridge the gap to live deployment. Most existing work handles these concerns in isolation — making it nearly impossible to reproduce results or compare methods fairly.
TradeMaster provides a unified, end-to-end pipeline that handles every stage, so researchers can focus on the algorithms rather than the infrastructure.
Pipeline
TradeMaster covers the full quantitative trading workflow:
1. Data Processing
- Multi-source financial data ingestion: price, volume, news, alternative data
- Standardized preprocessing and feature engineering pipelines
- Support for equities, cryptocurrencies, and futures markets
2. Environment Simulation
- Realistic market simulators with transaction costs, slippage, and liquidity constraints
- Multiple task formulations: single-asset trading, portfolio management, order execution, market making
3. RL Agent Training
- Plug-and-play support for major RL algorithms: PPO, SAC, TD3, A2C, and more
- Hierarchical RL support for multi-timescale trading strategies
- Distributed training infrastructure for large-scale experiments
4. Evaluation
- Standardized benchmark metrics: Sharpe Ratio, Maximum Drawdown, Annualized Return, Calmar Ratio
- PRUDEX-Compass framework for systematic, multi-dimensional strategy evaluation
- Cross-market and cross-asset generalization testing
5. Deployment
- Live trading interface bridging simulation to real brokerage APIs
- Real-time monitoring and risk management hooks
Supported Tasks
| Task | Description |
|---|---|
| Algorithmic Trading | Single-asset buy/sell/hold decision making |
| Portfolio Management | Multi-asset allocation optimization |
| Order Execution | Optimal execution of large orders with minimal market impact |
| Market Making | Simultaneous bid/ask quoting strategies |
| High-Frequency Trading | Intraday strategies at tick-level granularity (see also EarnHFT) |
Connection to the AI4Finance Ecosystem
TradeMaster is the foundation of our broader AI4Finance research program. Subsequent works build directly on it:
- EarnHFT (AAAI 2024) — hierarchical RL for high-frequency trading
- EarnMore (WWW 2024) — maskable stock representation for customizable portfolio management
- FinAgent (KDD 2024) — multimodal foundation agent for financial trading
- FinWorld (KDD 2026) — next-generation all-in-one financial AI platform
Links
- Paper: NeurIPS 2023
- GitHub: TradeMaster-NTU/TradeMaster
