numeraire-dataset#

Open, reproducible data loaders and builders for the numeraire research framework. This package ships code, not data: it fetches public (and, with your own credentials, licensed) sources and cleans them into tidy, point-in-time tables the framework consumes — so the cleaning is auditable and re-runnable, and no licensed data is ever redistributed inside a wheel.

from numeraire_dataset import load_ff_factors, load_ff_portfolio_view, load_gw_view

ff = load_ff_factors()                       # tidy frame: date, mkt_excess, smb, hml, risk_free
assets, assets_vintage = load_ff_portfolio_view(portfolio_set="industry_10")
view, vintage = load_gw_view(start_date="1926-07-01", end_date="2020-12-31")
# feed `view` straight into numeraire's backtest; `vintage` stamps the run's provenance

The *_view helpers return a numeraire TimeSeriesView together with a data_vintage stamp, so a downstream result can always trace back to the exact snapshot it was built from.

Two layers#

Sources

tidyfinance-primary loaders for standard sources (Fama–French, Goyal–Welch, and more) — thin functions returning tidy frames, with an optional point-in-time view helper that bridges into numeraire. The documented daily-Mom coverage gap uses the official Ken French zip directly and pins its content digest. Paired Pastor–Stambaugh loaders likewise pin exact official author bytes and explicitly mark those revised histories as non-PIT snapshots.

Builders

self-built ETL for what tidyfinance does not cover — a vintage-aware FRED-MD builder (reference period × vintage × series, with stationarity transforms applied at build time) and, with your own credentials, WRDS panels.

Where to go next#