Welcome to GPT-Augmented Momentum Strategy!
This course documents an end-to-end, audit-first workflow for developing a momentum-based stock selection strategy using a GPT-augmented research process.
What You'll Learn:
- Build a complete quantitative trading strategy from scratch
- Use GPT as a research copilot (not an oracle)
- Implement walk-forward validation to prevent data leakage
- Compare Long-Cash vs Long-Short strategy variants
- Create reproducible, auditable research workflows
What's Included:
- 8 production-ready Jupyter notebooks
- Complete dataset for backtesting
- GPT Strategy Book (80+ pages)
- All code templates and utilities
Prerequisites:
- Python programming experience
- Basic understanding of financial markets
- Familiarity with pandas and numpy
Episode 00: Quick Start Preview Get a rapid overview of the entire workflow before diving deep. This preview notebook walks through the complete pipeline in condensed form.
Duration: ~30 minutes
Level: Overview
Episode 01: Introduction & Setup
The Research Question and the GPT-Augmented Journey
In this episode, you'll learn:
- Why momentum was chosen as the test case
- The "copilot, not oracle" philosophy
- Setting up the research contract
- Defining success metrics (risk-adjusted returns)
Key Concept: GPT accelerates iterationβit does not decide credibility.
Episode 02: Data Preparation
Data Integrity Before Models
In this episode, you'll learn:
- Building the 25-stock universe + SPY benchmark
- Why Adjusted Close is mandatory
- Weekly resampling methodology
- Data quality audit checklist
- The importance of freezing your dataset
Key Concept: A dataset is only as credible as the audit that accompanies it.
Episode 03: Feature Engineering
Feature Engineering Under Timing Constraints
In this episode, you'll learn:
- Building momentum features (returns, volatility, moving averages)
- Strict timing constraints to prevent look-ahead bias
- Feature scaling and normalization
- The feature matrix structure
Key Concept: Every feature must be computable with information available at decision time.
Episode 04: Model Training
Framing the Modeling Problem
In this episode, you'll learn:
- Classification vs Regression framing
- Why XGBoost/Gradient Boosting
- Model configuration and hyperparameters
- Cross-validation approach
Key Concept: The model choice matters less than the methodology around it.
Episode 05: Backtesting Framework
When Results Look Too Good to Be True
In this episode, you'll learn:
- Building the unified backtest
- Detecting data leakage
- When to be suspicious of your results
- The audit-first mindset
Key Concept: Any result that looks "too good" is assumed wrong until proven otherwise.
Episode 06: Walk-Forward Validation
The Canonical Walk-Forward Architecture
In this episode, you'll learn:
- Walk-forward validation methodology
- Training window management
- Embargo periods to prevent leakage
- The canonical loop structure
Key Concept: Walk-forward validation simulates real trading conditions.
Episode 07: Results Analysis
Results, Interpretation, and the Economics of Credible Performance
In this episode, you'll learn:
- Interpreting backtest results
- Long-Cash vs Long-Short comparison
- Risk-adjusted performance metrics
- What the results actually mean
Key Concept: Results must be economically plausible, not just statistically significant.
Episode 08: Complete Pipeline
A New Paradigm for Quant Research
In this episode, you'll learn:
- Putting it all together
- The complete pipeline in one notebook
- Reproducibility and audit trails
- Next steps and extensions
Key Concept: Credible research, compressed in time.