World beating online covariance and portfolio construction.

The microprediction/precise repository provides tools for online covariance estimation and portfolio construction, aiming for superior performance. It targets quantitative analysts and researchers seeking advanced, incremental methods for financial modeling and optimization. The library offers novel approaches to portfolio theory, potentially enhancing investment strategies.

How It Works

The core innovation is the "Schur Complementary" portfolio construction method, which leverages block matrix inversion to unify top-down and bottom-up approaches. This technique is presented as advantageous for its insights into portfolio optimization. The package implements various "skaters" for incremental covariance estimation and "managers" for portfolio weight calculation, with an emphasis on parameter-free methods where possible.

Quick Start & Requirements

Installation is via pip install precise or directly from the repository (pip install git+https://github.com/microprediction/precise.git). It supports Python 3.11 and earlier. Key dependencies include ecos (consider conda install ecos if issues arise) and osqp (refer to CVXPY issue #1190 for potential system-specific problems). Official documentation and listings of available covariance estimators and portfolio managers are linked within the README.

Highlighted Details

• Claims "world beating" performance in online covariance and portfolio construction.
• Features the novel "Schur Complementary" portfolio construction methodology.
• Methodology is linked to insights from the "M6 contest" and powers a "Portfolio Theory GPT".
• Focuses on forecasting future realized covariance, acknowledging empirical noise.

Maintenance & Community

The project welcomes pull requests and is part of the broader "microprediction" initiative. Specific community channels, active maintainers, or sponsorship details are not explicitly detailed in the README.

Licensing & Compatibility

The project is distributed under the MIT License, which generally permits broad commercial use and integration.

Limitations & Caveats

The library is not intended for high-precision covariance calculations but rather for forecasting realized covariance with awareness of data noise. Some methods are parameter-free, while others require specific parameters. The project explicitly states it is not investment advice. Support is confirmed for Python 3.11 and earlier.