numeraire 0.2.0

A research framework for backtesting, comparing, and replicating empirical asset-pricing methods.

numeraire

A research framework providing a stable bedrock for backtesting, comparing, and replicating empirical asset-pricing / financial-econometrics methods (IPCA, VoC, KNS, 1/A, factor-model tests, …), extensible by design so new methods plug in as first-class extensions.

Documentation: https://py-numeraire.github.io/numeraire/

The numéraire is the reference unit against which all prices are measured. Core stays representation-agnostic: it defines capabilities (what a model can produce — weights, pricing, …), never a specific method's internal form, so linear-factor, nonlinear/RFF, neural, and distributional methods are all first-class.

Architecture (the boundary rule)

numeraire.core is exactly the modules that depend on no specific method and that every method depends on. Dependency arrows point toward core; core never imports a method, an adapter, or a reference library. This is enforced in CI by import-linter — the lint rule is the architecture (see pyproject.toml [tool.importlinter]).

src/numeraire/
  core/        # spine: DataView/Estimator/Splitter/Evaluator protocols, capabilities,
               # result schema, evaluator registry  (stable, strict-typed, high-coverage)
  adapters/    # thin wrappers making reference libraries conform to core — glue, not spine
  methods/     # published methods bundled as extensions (VoC, 1/A, classical tests, …)

Methods register via the numeraire.methods entry-point group, so external packages (numeraire-yourlab, numeraire-<method>) are first-class peers without editing core.

Install

uv sync --extra dev            # dev environment

Base install is the spine + native general evaluators only; method/adapter deps are extras.

Develop

uv run ruff check . && uv run ruff format --check .   # lint + format
uv run basedpyright src/numeraire/core                # strict types on core
uv run lint-imports                                   # architecture boundary
uv run pytest                                         # tests (public/synthetic data only)

Status

Pre-1.0, development. Usable via GitHub install; not on PyPI yet (the capability layer is expected to crystallize once three real adapters land). The spine (DataView, walk-forward OOS engine, Splitter, native evaluators) is in place; the first method (1/A conservative slope) is wired end-to-end.

License: BSD-3-Clause. Never commit CRSP/WRDS/proprietary data or credentials (data/, ref/, .env are git-ignored).