def test_clustered_cov_debiased(data):
c = ClusteredCovariance(data.x,
data.y,
data.z,
data.params,
debiased=True,
clusters=data.clusters)
assert c.debiased is True
assert c.config["debiased"] is True
assert_equal(c.config["clusters"], data.clusters)
ngroups = len(np.unique(data.clusters))
sums = np.zeros((ngroups, data.nvar))
xe = data.xhat * data.e
for i in range(len(data.clusters)):
sums[data.clusters[i]] += xe[i]
op = np.zeros((data.nvar, data.nvar))
for j in range(len(sums)):
op += sums[[j]].T @ sums[[j]]
# This is a strange choice
s = (op / data.nobs * ((data.nobs - 1) / (data.nobs - data.nvar)) *
ngroups / (ngroups - 1))
assert_allclose(c.s, s)
assert_allclose(c.cov, data.vinv @ s @ data.vinv / data.nobs)
cs = str(c)
assert "Debiased: True" in cs
assert "Num Clusters: {0}".format(len(sums)) in cs
assert "id" in c.__repr__()
def test_clustered_cov_errors(data):
with pytest.raises(ValueError):
ClusteredCovariance(data.x,
data.y,
data.z,
data.params,
clusters=data.clusters[:10])
def test_asymptotic(self, data):
c = ClusteredCovariance(data.x, data.y, data.z, data.params,
clusters=data.clusters)
assert c._kappa == 1
assert c.debiased is False
assert c.config['debiased'] is False
assert_equal(c.config['clusters'], data.clusters)
assert_allclose(c.s2, data.s2)
sums = np.zeros((len(np.unique(data.clusters)), data.nvar))
xe = data.xhat * data.e
for i in range(len(data.clusters)):
sums[data.clusters[i]] += xe[i]
op = np.zeros((data.nvar, data.nvar))
for j in range(len(sums)):
op += sums[[j]].T @ sums[[j]]
s = op / data.nobs
assert_allclose(c.s, s)
assert_allclose(c.cov, data.vinv @ s @ data.vinv / data.nobs)
cs = str(c)
assert 'Debiased: False' in cs
assert 'Num Clusters: {0}'.format(len(sums)) in cs