Part 3: Machine Learning
Goal for this stage: From Models to Agents. Understand the proper use of machine learning in quantitative trading, and how to evolve from "prediction models" to "decision-making Agents."
Lesson List
| Lesson | Topic | Deliverables |
|---|---|---|
| Lesson 09 | Supervised Learning in Quantitative Trading | ML Strategy Framework, IC/IR Evaluation |
| Lesson 10 | From Models to Agents | Single Agent Strategy Module |
Background Knowledge
| Document | Description | Suggested Reading Time |
|---|---|---|
| LLM Research in Quantitative Trading | Latest applications of large language models in quantitative trading | 15 min |
| Triple Barrier Labeling Method | Defining ML labels using take-profit and stop-loss | 15 min |
| Time Series Cross-Validation (Purged CV) | Preventing information leakage in time series | 15 min |
| Reinforcement Learning in Trading | Combining RL algorithms with trading | 20 min |
| Alternative Data (NLP and Satellite) | Non-traditional data sources: text sentiment, satellite imagery, etc. | 15 min |
| Meta-Labeling Method | Secondary model for predicting signal reliability | 15 min |
| Feature Engineering Common Pitfalls | Future leakage, overfitting, and other common mistakes | 10 min |
| Limitations of ML in Finance | Core challenges: low signal-to-noise ratio, distribution drift, etc. | 15 min |
| Model Architecture Selection Guide | LSTM/GRU/Transformer/CNN comparison, RL algorithm selection | 20 min |
| Model Drift and Retraining | K-S/CUSUM drift detection, retraining triggers and strategies | 20 min |
| MLOps Infrastructure | Feature Store, Model Registry, Drift Monitor | 30 min |
| Frontier ML and RL Methods (2025) | SOTA techniques: Decision Transformer, AlphaAgent, GNN, Diffusion Models | 30 min |
After Completing This Stage
You will be able to:
- Understand why "predicting price" is the wrong objective
- Design proper labels (Triple Barrier Method)
- Understand the core components of an Agent: State, Action, Reward, Environment
- Build a single Agent strategy that generates trading signals
Next Stage
→ Part 4: Multi-Agent Systems - Building collaborative Agent systems