Skip to content

MFDRO

Point-in-time multi-frequency distribution disagreement, with the evidence needed to audit every estimate.

MFDRO turns a daily panel of simple returns into a reproducible path of Wasserstein dispersion estimates. It keeps the scientific configuration, frequency construction, random seeds, window checks, and skipped formations visible.

Get started Understand the signal

daily returns → empirical frequency measures → common scale → barycenter → ρ and √ρ → audit trail

Install

python -m pip install mfdro

MFDRO supports Python 3.10 and newer. See Getting started for a pinned installation and a complete first estimate.

What you get

  • A small, explicit API

    Start from MultiFrequencySignal, choose a validated SignalConfig, and receive labelled pandas outputs. Use a preset for exploration or declare every scientific choice yourself.

    First estimate →

  • Two to many frequencies

    Daily, weekly, monthly is only the default. Any ordered grid with at least two measures is supported; resampling boundaries and effective horizons are explicit.

    Configure frequencies →

  • Point-in-time diagnostics

    Inspect the complete formation schedule before paying for transport geometry. Warm-up, calendar mismatch, and insufficient aggregation samples are machine-readable.

    Diagnose inputs →

  • Portable research artifacts

    Save estimates, audits, skipped dates, the exact configuration, a manifest, and SHA-256 checksums without using unsafe pickle files.

    Preserve results →

Scope, without ambiguity

MFDRO estimates cross-frequency distribution disagreement. rho is a non-negative squared dispersion and sqrt_rho is its square root. Neither is an expected return, a buy/sell recommendation, nor an automatically calibrated DRO radius.

The package deliberately does not download or clean production data, infer an investable universe, solve a portfolio problem, or simulate execution. Those steps have different assumptions and belong in explicit downstream components. See the end-to-end workflow and backtesting boundary.

Alpha research software

The implementation is tested and typed, but the public API is not frozen across minor releases. Pin an exact package version and retain exported results for any research archive.

Choose your next page