linearmodels

7.0 · active · verified Wed Apr 15

linearmodels is a Python library that extends `statsmodels` with advanced econometric models, including Panel data models (Fixed Effects, Random Effects), Instrumental Variable (IV) estimators (2SLS, GMM), Factor Asset Pricing models, and System Regression models (SUR, 3SLS). It is currently at version 7.0 and sees active development with several major/minor releases per year.

Warnings

breaking Starting with version 5.0, the variable ordering in formulas processed by `from_formula` is preserved as it appears in the formula, rather than being sorted alphabetically. This can lead to changes in coefficient ordering or model interpretation if relying on the previous sorting behavior.
Fix: Review models created with formulas to ensure variable order is as expected. Explicitly define variable order if sensitive.
breaking Version 5.0 and 7.0 significantly increased the minimum required versions for Python, NumPy, SciPy, pandas, statsmodels, and formulaic. Older environments may not be compatible.
Fix: Upgrade your environment to meet the new minimum requirements: Python >=3.10, NumPy >=1.22, SciPy >=1.8, pandas >=1.4, statsmodels >=0.13, formulaic >=1.0.
gotcha The name for clustered covariance was corrected from 'cluster' to 'clustered' in version 7.0. Using the old name will likely result in an error.
Fix: Update `cov_type='cluster'` to `cov_type='clustered'` when specifying clustered standard errors.
gotcha The library transitioned to `formulaic` as the preferred formula parser. While `patsy` might have been implicitly used or supported in older versions, `formulaic` is now the standard for R-style formula parsing.
Fix: Ensure `formulaic` is installed and that formulas conform to its expected syntax. Be aware that `linearmodels.future.ordering` is a no-op in 5.0+ and has no effect on variable ordering.
gotcha Estimating models with rank-deficient regressors can lead to unreliable estimates. `linearmodels` includes a rank check by default, but it can be skipped using `rank_check=False`.
Fix: Always ensure your regressor matrix is full rank, especially when using `rank_check=False`. Carefully inspect model output and diagnostics for signs of multicollinearity or rank deficiency.

Install

pip install linearmodels PyPI

Imports

PanelOLS
```
from linearmodels import PanelOLS
```
IV2SLS
```
from linearmodels.iv import IV2SLS
```
SUR
```
from linearmodels.system import SUR
```

PooledOLS

from linearmodels.panel import PooledOLS

RandomEffects

from linearmodels.panel import RandomEffects

Quickstart

This quickstart demonstrates how to load panel data, structure it for `linearmodels` using a pandas MultiIndex, and fit a basic PanelOLS model with entity fixed effects and clustered standard errors.

import numpy as np
import pandas as pd
from linearmodels.datasets import grunfeld
from linearmodels.panel import PanelOLS

data = grunfeld.load_pandas().data
data.year = data.year.astype(np.int64)
# Create a MultiIndex (entity - time) for panel data
data = data.set_index(['firm', 'year'])

# Define dependent and independent variables
dep = data.invest
exog = data[['value', 'capital']]

# Initialize and fit the PanelOLS model with entity effects
mod = PanelOLS(dep, exog, entity_effects=True)
res = mod.fit(cov_type='clustered', cluster_entity=True)

print(res)

view raw JSON →