TST: Test GMM

Complete initial testing of linear factor GMM

Author: bashtage

linearmodels/asset_pricing/covariance.py

@@ -31,7 +31,7 @@ def __init__(self, xe, *, jacobian=None, inv_jacobian=None,
         self._jac = jacobian
         self._inv_jac = inv_jacobian
         self._center = center
-        if (jacobian is None and inv_jacobian is None)\
+        if (jacobian is None and inv_jacobian is None) \
                 or (jacobian is not None and inv_jacobian is not None):
             raise ValueError('One and only one of jacobian or inv_jacobian must be provided.')
         self._debiased = debiased
@@ -41,6 +41,16 @@ def __init__(self, xe, *, jacobian=None, inv_jacobian=None,
         else:
             self._square = inv_jacobian.shape[0] == inv_jacobian.shape[1]
 
+    def __str__(self):
+        return self.__class__.__name__
+
+    def __repr__(self):
+        return self.__str__() + ', id: {0}'.format(hex(id(self)))
+
+    @property
+    def config(self):
+        return {'type': self.__class__.__name__}
+
     @property
     def s(self):
         """
@@ -142,6 +152,18 @@ def __init__(self, xe, *, jacobian=None, inv_jacobian=None,
             if bandwidth < 0:
                 raise ValueError('bandwidth must be non-negative.')
 
+    def __str__(self):
+        descr = ', Kernel: {0}, Bandwidth: {1}'.format(self._kernel,
+                                                       self.bandwidth)
+        return self.__class__.__name__ + descr
+
+    @property
+    def config(self):
+        out = super(KernelCovariance, self).config
+        out['kernel'] = self._kernel
+        out['bandwidth'] = self.bandwidth
+        return out
+
     @property
     def kernel(self):
         """Kernel used in estimation"""

linearmodels/asset_pricing/model.py

@@ -252,7 +252,8 @@ def fit(self, cov_type='robust', debiased=True, **cov_config):
                        rsquared=r2, total_ss=total_ss, residual_ss=residual_ss,
                        param_names=param_names, portfolio_names=self.portfolios.cols,
                        factor_names=self.factors.cols, name=self._name,
-                       cov_type=cov_type, model=self, nobs=nobs, rp_names=self.factors.cols)
+                       cov_type=cov_type, model=self, nobs=nobs, rp_names=self.factors.cols,
+                       cov_est=cov_est)
 
         return LinearFactorModelResults(res)
 
@@ -418,7 +419,6 @@ def fit(self, cov_type='robust', debiased=True, **cov_config):
         and ``bandwidth`` (a positive integer).
         """
         # TODO: Refactor commonalities in estimation
-        # TODO: Verify GLS implementation
         nobs, nf, nport, nrf, s1, s2, s3 = self._boundaries()
         excess_returns = not self._risk_free
         f = self.factors.ndarray
@@ -493,7 +493,8 @@ def fit(self, cov_type='robust', debiased=True, **cov_config):
                        rsquared=r2, total_ss=total_ss, residual_ss=residual_ss,
                        param_names=param_names, portfolio_names=self.portfolios.cols,
                        factor_names=factor_names, name=self._name,
-                       cov_type=cov_type, model=self, nobs=nobs, rp_names=rp_names)
+                       cov_type=cov_type, model=self, nobs=nobs, rp_names=rp_names,
+                       cov_est=cov_est)
 
         return LinearFactorModelResults(res)
 
@@ -719,9 +720,11 @@ def fit(self, center=True, use_cue=False, steps=2, disp=10, max_iter=1000,
                        options={'disp': bool(disp), 'maxiter': max_iter})
         params = res.x
         last_obj = res.fun
+        iter = 1
         # 3. Step 2 using step 1 estimates
         if not use_cue:
             for i in range(steps - 1):
+                iter += 1
                 g = self._moments(params, excess_returns)
                 w = weight_est.w(g)
                 args = (excess_returns, w)
@@ -735,6 +738,7 @@ def fit(self, center=True, use_cue=False, steps=2, disp=10, max_iter=1000,
                 if np.abs(obj - last_obj) < 1e-6:
                     break
                 last_obj = obj
+
         else:
             args = (excess_returns, weight_est)
             obj = self._j_cue
@@ -752,7 +756,6 @@ def fit(self, center=True, use_cue=False, steps=2, disp=10, max_iter=1000,
                           df=self.factors.shape[1], **cov_config)
 
         full_vcv = cov_est.cov
-        self._testing = {'s': cov_est.s}
         sel = slice((n * k), (n * k + k + nrf))
         rp = params[sel]
         rp_cov = full_vcv[sel, sel]
@@ -789,7 +792,8 @@ def fit(self, center=True, use_cue=False, steps=2, disp=10, max_iter=1000,
                        rsquared=r2, total_ss=total_ss, residual_ss=residual_ss,
                        param_names=param_names, portfolio_names=self.portfolios.cols,
                        factor_names=self.factors.cols, name=self._name,
-                       cov_type=cov_type, model=self, nobs=nobs, rp_names=rp_names)
+                       cov_type=cov_type, model=self, nobs=nobs, rp_names=rp_names,
+                       iter=iter, cov_est=cov_est)
 
         return GMMFactorModelResults(res)
 

linearmodels/asset_pricing/results.py

@@ -30,6 +30,7 @@ def __init__(self, res):
         self._cols = ['alpha'] + ['{0}'.format(f) for f in self._factor_names]
         self._rp_names = res.rp_names
         self._alpha_vcv = res.alpha_vcv
+        self._cov_est = res.cov_est
 
     @property
     def summary(self):
@@ -110,7 +111,9 @@ def summary(self):
                             headers=header,
                             title=title)
         smry.tables.append(table)
-        smry.add_extra_txt(['See full_summary for complete results'])
+        smry.add_extra_txt(['Covariance estimator:',
+                            str(self._cov_est),
+                            'See full_summary for complete results'])
 
         return smry
 
@@ -201,7 +204,7 @@ def tstats(self):
     @property
     def cov_estimator(self):
         """Type of covariance estimator used to compute covariance"""
-        return self._cov_type
+        return str(self._cov_est)
 
     @property
     def cov(self):
@@ -262,6 +265,7 @@ def residual_ss(self):
 class GMMFactorModelResults(LinearFactorModelResults):
     def __init__(self, res):
         super(GMMFactorModelResults, self).__init__(res)
+        self._iter = res.iter
 
     @property
     def std_errors(self):
@@ -273,3 +277,7 @@ def std_errors(self):
         se = np.r_[ase, se[:nloadings]]
         se = se.reshape((nportfolio, nfactor))
         return pd.DataFrame(se, columns=self._cols, index=self._portfolio_names)
+
+    def iterations(self):
+        """Number of steps in GMM estimation"""
+        return self._iter

linearmodels/tests/asset_pricing/test_linear_factor_gmm.py

@@ -67,35 +67,30 @@ def test_linear_model_gmm_moments_jacobian(data):
     get_all(res)
 
 
-@pytest.mark.skip
 def test_linear_model_gmm_smoke_iterate(data):
     mod = LinearFactorModelGMM(data.portfolios, data.factors)
     res = mod.fit(cov_type='robust', disp=5, steps=20)
     get_all(res)
 
 
-@pytest.mark.skip
 def test_linear_model_gmm_smoke_risk_free(data):
     mod = LinearFactorModelGMM(data.portfolios, data.factors, risk_free=True)
     res = mod.fit(cov_type='robust', disp=10)
     get_all(res)
 
 
-@pytest.mark.skip
 def test_linear_model_gmm_kernel_smoke(data):
     mod = LinearFactorModelGMM(data.portfolios, data.factors)
     res = mod.fit(cov_type='kernel', disp=10)
     get_all(res)
 
 
-@pytest.mark.skip
 def test_linear_model_gmm_kernel_bandwidth_smoke(data):
     mod = LinearFactorModelGMM(data.portfolios, data.factors)
     res = mod.fit(cov_type='kernel', bandwidth=10, disp=10)
     get_all(res)
 
 
-@pytest.mark.skip
 def test_linear_model_gmm_cue_smoke(data):
     mod = LinearFactorModelGMM(data.portfolios, data.factors, risk_free=True)
     res = mod.fit(cov_type='robust', disp=10, use_cue=True)