class TestGLMGaussHACGroupsum(CheckDiscreteGLM):
cov_type = 'hac-groupsum'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {
'time': pd.Series(np.tile(np.arange(7), 5)[:-1]),
# time index is just made up to have a test case
# check for GH#3606
'maxlags': 2,
'use_correction': 'hac',
'df_correction': False
}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
cls.res1b = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
def test_kwd(self):
# test corrected keyword name
assert_allclose(self.res1b.bse, self.res1.bse, rtol=1e-12)
class TestGLMGaussHACPanel(CheckDiscreteGLM):
cov_type = 'hac-panel'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {
'time': np.tile(np.arange(7), 5)[:-1],
# time index is just made up to have a test case
'maxlags': 2,
'kernel': sw.weights_uniform,
'use_correction': 'hac',
'df_correction': False
}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
cls.res1b = mod1.fit(cov_type='nw-panel', cov_kwds=cls.cov_kwds)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
def test_kwd(self):
# test corrected keyword name
assert_allclose(self.res1b.bse, self.res1.bse, rtol=1e-12)
class TestGLMGaussHAC(CheckDiscreteGLM):
cov_type = 'HAC'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {'maxlags': 2}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
class TestGLMGaussClu(CheckDiscreteGLM):
cov_type = 'cluster'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {'groups': group}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls): # TODO: de-dup with other setup_classes
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
class TestGLMGaussNonRobust(CheckDiscreteGLM):
cov_type = 'nonrobust'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {}
fit_kwargs = {"cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
class TestGLMGaussHACUniform2(TestGLMGaussHACUniform):
# GH#4524
cov_type = 'HAC'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {"kernel": sw.weights_uniform, "maxlags": 2}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(**cls.fit_kwargs)
mod2 = OLS(endog, exog)
# check kernel as string
kwds2 = {'kernel': 'uniform', 'maxlags': 2}
cls.res2 = mod2.fit(cov_type=cls.cov_type, cov_kwds=kwds2)
class TestGLMGaussHAC2(CheckDiscreteGLM):
cov_type = 'HAC'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {'kernel': 'bartlett', 'maxlags': 2}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
# check kernel specified as string
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False,
cov_type=cls.cov_type,
cov_kwds={'maxlags': 2})
class TestGLMGaussHACUniform(CheckDiscreteGLM):
cov_type = 'HAC'
mod_kwargs = {"family": families.Gaussian()}
cov_kwds = {'kernel': sw.weights_uniform, 'maxlags': 2}
fit_kwargs = {"cov_type": cov_type, "cov_kwds": cov_kwds}
@classmethod
def setup_class(cls):
mod1 = cls.model_cls(endog, exog, **cls.mod_kwargs)
cls.res1 = mod1.fit(disp=False, **cls.fit_kwargs)
mod2 = OLS(endog, exog)
cls.res2 = mod2.fit(disp=False, **cls.fit_kwargs)
# for debugging
cls.res3 = mod2.fit(cov_type=cls.cov_type, cov_kwds={'maxlags': 2})
# TODO: Should something be done with res3?
def test_cov_options(self):
# check keyword `weights_func
kwdsa = {'weights_func': sw.weights_uniform, 'maxlags': 2}
res1a = self.res1.model.fit(cov_type=self.cov_type, cov_kwds=kwdsa)
res2a = self.res2.model.fit(cov_type=self.cov_type, cov_kwds=kwdsa)
assert_allclose(res1a.bse, self.res1.bse, rtol=1e-12)
assert_allclose(res2a.bse, self.res2.bse, rtol=1e-12)
# regression test for bse values
bse = np.array([2.82203924, 4.60199596, 11.01275064])
assert_allclose(res1a.bse, bse, rtol=1e-6)
assert res1a.cov_kwds['weights_func'] is sw.weights_uniform
kwdsb = {'kernel': sw.weights_bartlett, 'maxlags': 2}
res1a = self.res1.model.fit(cov_type='HAC', cov_kwds=kwdsb)
res2a = self.res2.model.fit(cov_type='HAC', cov_kwds=kwdsb)
assert_allclose(res1a.bse, res2a.bse, rtol=1e-12)
# regression test for bse values
bse = np.array([2.502264, 3.697807, 9.193303])
assert_allclose(res1a.bse, bse, rtol=1e-6)