pypi / statsmodels

statsmodels

What changed in statsmodels 0.14 to 0.15, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-28.

4 breaking12 new APIs and options1 behavior and defaults2 fixes and other

See which of these apply to your code, ranked, in your agent:

npx @unstale/cli init

statsmodels 0.1517 changes

The minimum supported versions of NumPy, SciPy, and pandas are raised.
0.15.0release notes
Functions with variable-length tuple returns now return documented NamedTuple results.
0.15.0release notes
Randomness is controlled with rng across the package, replacing seed and random_state.
0.15.0release notes
Input validation is tighter across many string-valued options; invalid values now raise a clear ValueError instead of failing silently or with a confusing error.
0.15.0release notes

A new robust estimator is available.
0.15.0release notes
A new delete-k block jackknife estimator is available.
0.15.0release notes
A new Diebold-Mariano test is available.
0.15.0release notes
HurdleCountModel gains fit_regularized.
0.15.0release notes
A new abstracted formula engine supports patsy and formulaic as backends.
0.15.0release notes
The Hamilton filter is newly available.
0.15.0release notes
A new Jonckheere-Terpstra test is available.
0.15.0release notes
An L1-penalized GLM solver is newly available.
0.15.0release notes
A new Leybourne-McCabe test is available.
0.15.0release notes
A local false discovery rate correction is newly available.
0.15.0release notes
A new Pesaran-Timmermann test is available.
0.15.0release notes
Polars DataFrames and Series can be used as model input.
0.15.0release notes

Includes minor changes to improve future compatibility with NumPy.
0.14.6release notes
Fixes an issue with pandas 3.0.0 that prevented statsmodels from importing.
0.14.6release notes