linearmodels/unknown-cov-type
| Rule | unknown-cov-type |
|---|---|
| Pack | linearmodels |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
The model does not take this cov_type, such as the statsmodels names "HC1" or "cluster", so fit raises. Use a name the model takes, such as "robust" or "clustered".
result = PanelOLS(y, X, entity_effects=True).fit(cov_type="HC1")iv = IV2SLS(y, exog, endog, instruments).fit(cov_type="nonrobust")result = PanelOLS(y, X, entity_effects=True).fit(cov_type="robust")iv = IV2SLS(y, exog, endog, instruments).fit(cov_type="unadjusted", debiased=True)What the rule reports
Section titled “What the rule reports”The rule reports a cov_type string that the model doesn't take, such as "HC1", "HAC",
"cluster" or "nonrobust" of statsmodels, "clustered" on FamaMacBeth or IVSystemGMM,
"kernel" on BetweenOLS, and "unadjusted" on a factor model.
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
On OLS, IV2SLS, IVLIML and AbsorbingLS, the unsafe fix writes "robust" for "HC0", "robust", debiased=True for "HC1" and "unadjusted", debiased=True for "nonrobust", in a call that doesn't pass debiased. The call raised before the fix. Any other name has no fix, because the code doesn't show which name of the model is meant.
7.1 raises UnknownOptionError, which is both a ValueError and a KeyError, with the list of
names the model takes, and 7.0 raised ValueError or KeyError. The lists were read from the
error for each of the 19 model classes, and each name on a list was then run on its class:
| Models | Names |
|---|---|
PanelOLS, PooledOLS, RandomEffects, FirstDifferenceOLS |
unadjusted, homoskedastic, conventional, robust, heteroskedastic, clustered, kernel, dk, driscoll-kraay, ac, autocorrelated |
BetweenOLS |
unadjusted, homoskedastic, conventional, robust, heteroskedastic, clustered |
FamaMacBeth |
unadjusted, homoskedastic, robust, heteroskedastic, kernel |
OLS, IV2SLS, IVLIML, AbsorbingLS |
unadjusted, homoskedastic, homo, robust, heteroskedastic, hccm, kernel, clustered, one-way, and the four class names such as KernelCovariance |
IVGMM, IVGMMCUE, IVJIVE |
unadjusted, homoskedastic, robust, heteroskedastic, kernel, clustered |
SUR, IV3SLS |
unadjusted, homoskedastic, robust, heteroskedastic, kernel, hac, clustered |
IVSystemGMM |
unadjusted, homoskedastic, robust, heteroskedastic, kernel, hac |
TradedFactorModel, LinearFactorModel, LinearFactorModelGMM |
robust, heteroskedastic, kernel |
Since 7.1 the names are not case sensitive, so the rule takes a known name in any case.
On OLS with the mroz data, the standard errors of linearmodels were equal to the ones of statsmodels
to 1e-13 for each pair below, and the message for the instrument models gives the same pairs:
| linearmodels | statsmodels |
|---|---|
"robust" |
"HC0" |
"robust" with debiased=True |
"HC1" |
"unadjusted" with debiased=True |
"nonrobust" |
"clustered" with clusters= and debiased=True |
"cluster" with the same groups |
The panel models default to debiased=True, and their default standard errors were equal to the ones
of statsmodels.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
PanelOLS, PooledOLS, RandomEffects and FirstDifferenceOLS take "unadjusted", "robust", "clustered", "kernel" and "ac", so cov_type=$C raises an error at fit. The statsmodels names, such as "HC1", "HAC", "cluster" and "nonrobust", are not among them.Browse the other rules of the linearmodels pack.