101 Alphas Research Platform

A modular, robust, and educational platform for quantitative research based on the "101 Formulaic Alphas" paper (Kakushadze, 2016).

Key Features

• Data Agnostic Architecture: The system ingests standard "Long Format" data (Database/CSV style) and internally converts it to strictly aligned "Wide Format" matrices. This prevents common indexing errors and alignment bugs found in ad-hoc research scripts.
• Robust Engine: The MarketData container guarantees that all fields (Open, Close, Volume, etc.) share the exact same Time x Ticker index.
• Educational Frontend: A Streamlit UI that displays the Mathematical Formula (LaTeX), the Python Code, and the Visual Output side-by-step.
• Modular Operators: Core primitives (rank, delay, signedpower) are isolated in src/operators.py, making the code self-documenting.

Quick Start

1. Installation

The project uses pyproject.toml for dependencies.

pip install -r pyproject.toml
# OR if using uv (recommended)
uv pip install -r pyproject.toml

2. Run the App

streamlit run app.py

Architecture

• src/market_data.py: The core container. It takes raw data, pivots it, aligns indices, and provides safe accessors (e.g., data.close).
• src/operators.py: Vectorized implementations of the alpha primitives.
• src/definitions.py: The Alphas themselves, defined as functions of MarketData.
• src/data_loader.py: Adapter for loading data from DuckDB (or generating synthetic test data).

Extending

To add a new Alpha:

1. Open src/definitions.py.
2. Define a function alpha_XXX(data: MarketData).
3. Register it in ALPHA_REGISTRY with its LaTeX formula and description.

Robustness Notes

• Alignment: The MarketData class reindexes all inputs to a shared (Date, Ticker) grid upon initialization.
• Missing Data: Forward-filling and Backward-filling are applied by default to handle sparse data, ensuring Alphas don't break on isolated NaNs.