SPO3

SPO3 is a research codebase for portfolio optimization with learning-based return prediction and decision-focused losses (including SPO+).

Quick Start

1. Install dependencies in your Python environment.
2. Prepare data using scripts in scripts/.
3. Run training and evaluation:

python run.py --config <path_to_config.yaml>

Project Structure

• models/: portfolio optimization models (e.g., Markowitz, CVaR).
• losses/: decision-focused loss implementations.
• predictors/: return prediction modules.
• utils/: training, tuning, backtesting, and data utilities.
• configs/: experiment configuration files.
• scripts/: data download and preprocessing helpers.

Notes

• Use batch_run.py or batch_run_baseline.py for batch experiments.
• Example data download instructions are in scripts/download_example.md.