Get different Standard Errors from AbsorbingLS and PanelOLS

[huicaikrz]

I do fixed effect regression (time effect+entity effect) with AbsorbingLS as well as PanelOLS, I get same coefficient while the standard errors seems to be quite different. I tried different settings, like "clustered" with entity and time effect / robust/
unadjusted, but they are always different.

I checked source code and it seems that standard errors of the two models are relaized in different py file. May I know the difference? Since I thought that the stand errors should be same.

Thanks.

[bashtage]

They are different. Which standard errors are you using for the Panel model? Can you post a code example that shows the differences you are seeing?

[huicaikrz]

Here's the sample code. I use the sample dataset from statsmodels and test three different settings of standard errors. They are all quite different. Among the three settings, I am quite interested in standard errors with "cov_type='clustered'".

from linearmodels import AbsorbingLS, PanelOLS
import numpy as np
import pandas as pd

from statsmodels.datasets import grunfeld
data = grunfeld.load_pandas().data
data.year = data.year.astype(np.int64)
data = data.set_index(['firm','year'], drop=False)
mod_panel = PanelOLS(data.invest, data[['value','capital']], entity_effects=True, time_effects=True)
res_panel1 = mod_panel.fit(cov_type='clustered', cluster_entity=True, cluster_time=True)
res_panel2 = mod_panel.fit(cov_type='robust')
res_panel3 = mod_panel.fit(cov_type='unadjusted')

cats = pd.DataFrame(pd.Categorical(data[col]) for col in ["firm", "year"])
mod_absorb = AbsorbingLS(data.invest, data[['value', 'capital']], absorb=cats)
for col in cats.columns:
    cats[col] = cats[col].astype('category').cat.codes
res_absorb1 = mod_absorb.fit(method='lsmr', cov_type='clustered', clusters=cats)
res_absorb2 = mod_absorb.fit(cov_type='robust')
res_absorb3 = mod_absorb.fit(cov_type='unadjusted')

print(res_panel1.summary, "\n\n")
print(res_absorb1.summary, "\n\n")
print(res_panel2.summary, "\n\n")
print(res_absorb2.summary, "\n\n")
print(res_panel3.summary, "\n\n")
print(res_absorb3.summary, "\n\n")`

[bashtage]

Thanks for the code. I can see they are different, but will need more time to determine. I suspect that the issue is a DoF correction.