microprediction/microprediction

If you can measure it, consider it predicted

What it solves

This repository serves as a central hub for a collection of quantitative finance, time series analysis, and statistical prediction tools. It addresses the need for fast, reliable, and often incremental (online) methods for distributional prediction, covariance estimation, and portfolio optimization.

How it works

The project consists of a suite of of independent, well-tested packages and research projects, including:

  • Skaters: A tool for online univariate time series distributional prediction.
  • Precise: Incremental covariance and correlation estimators that handle high dimensions and variable changes.
  • Thurstone: A package for calculating multi-entrant contest probabilities and inverse calibration using a fast ability transform.
  • Humpday: A collection of derivative-free (black box) optimization methods implemented in pure Python and JavaScript.
  • Allocation: Online portfolio methods, including novel Schur and Thurstone approaches.

Who it’s for

Quantitative analysts, applied mathematicians, and developers working in quantitative finance, sports analytics, and time series forecasting.

Highlights

  • Zero-dependency packages: Several tools are designed to run in Pyodide or as pure JavaScript ports for high portability.
  • Online Learning: Strong focus on incremental estimators and prediction methods that update as new data arrives.
  • SOTA Performance: Claims state-of-the-art reliability for univariate time series distributional prediction.
  • Black-box Optimization: Provides a comprehensive set of derivative-free optimization methods in one place.

Related

  • Project
    google/tf-quant-finance

    A TensorFlow-based library for quantitative finance that provides high-performance mathematical methods and pricing models with hardware acceleration.

  • Project
    DataDog/toto

    A foundation model for multivariate time series forecasting optimized for observability metrics, providing zero-shot and probabilistic predictions.

  • Project
    facebookresearch/Kats

    A comprehensive toolkit for time series analysis that provides a unified framework for forecasting, anomaly detection, and feature extraction.

  • Project
    Nixtla/statsforecast

    A high-performance library for univariate time series forecasting that provides fast, scalable implementations of statistical and econometric models.

  • Project
    Nixtla/hierarchicalforecast

    A library for probabilistic hierarchical forecasting that ensures consistency across different levels of data aggregation using statistical and econometric reconciliation methods.