statsmodels Import Reference

The imports that succeed and then fail

This is the part no documentation-derived reference contains, because you cannot see it without running the code. Three of these four statements import cleanly. They hand back different objects, and the failure surfaces later — at the call, or at instantiation — by which point the import line looks innocent.

Every path, executed

Both the statements that work and the ones that do not. Failing paths are kept with their real exception text so you can match the traceback in front of you — a reference that lists only what works cannot be searched from an error message.

What coint_johansen actually returns

Run on 1,000 real daily closes for AAPL and MSFT (2022-08-24 to 2026-08-19), via coint_johansen(data, det_order=0, k_ar_diff=1). The attribute names below are the ones the returned object really carries, listed with dir().

from statsmodels.tsa.vector_ar.vecm import coint_johansen

# data: an (n_obs, n_series) array of LEVELS, not returns
res = coint_johansen(data, det_order=0, k_ar_diff=1)

# Trace test at the 95% level (column index 1 of the critical-value table)
for r in range(len(res.lr1)):
    reject = res.lr1[r] > res.cvt[r, 1]
    print(f"r <= {r}:  trace={res.lr1[r]:.4f}  crit95={res.cvt[r, 1]:.4f}  reject={reject}")

# The cointegrating vector, if you rejected r = 0
beta = res.evec[:, 0]

Method

Each import statement was executed in this interpreter and its outcome recorded. Failing paths are kept with their real exception text so a reader can match the traceback they are looking at. Return shapes come from calling the function on real daily price data, not from documentation.

This matters because import paths move between releases and documentation lags them. Anything on this page can be regenerated with python -m pipeline.statsmodels_imports, and every figure is tagged with the version that produced it — so when it goes stale, it goes visibly stale rather than quietly wrong.

Frequently asked questions