Kevin Sheppard

Autocorrelated errors do not result in inconsistency of estimators as long as the relatively weak condition of E[X \epsilon]=0 is still true.

When you have a lagged dependent variable, then autocorrelation will almost always result in inconsistency. When you have only exogenous regressors, and if these are unconditionally uncorrelated with the residuals,...

One small issue, otherwise LGTM.

Model estimates for me without issue on Windows. ``` import pandas as pd import statsmodels.api as sm endog = pd.read_csv('my_data_sample.csv', index_col=[0])['1b'] order = (25, 1, 21) seasonal_order = (0, 0,...

This does not look liek a sensible model ``` ar.L1 0.002454 ar.L2 -0.001238 ar.L3 -0.004040 ar.L4 -0.003733 ar.L5 0.000023 ar.L6 0.003945 ar.L7 0.003790 ar.L8 0.000220 ar.L9 -0.003202 ar.L10 -0.003236 ar.L11...

> ```python > mod = sarimax.SARIMAX(np.zeros(10), order=(2, 0, 0)) > res = mod.filter([1, 2., 1.]) > ``` This throwing an error that suggests the matrix being factorized is exactly singular,...

I can't reproduce. When I put the binary in the XPATH location it works without errors.

x13as_ascii is the required binary.

A PR would be welcome.

The arch package has `engle_granger`, which has critical values for no trend.