fixed issues leading to failing tests

statsmodels/tsa/base/mlemodel.py

@@ -52,7 +52,7 @@ def __init__(
         self._fixed_params_index = None
         self._free_params_index = None
 
-    def prepare_data(data):
+    def prepare_data(self, data):
         raise NotImplementedError
 
     def clone(self, endog, exog=None, **kwargs):
@@ -118,7 +118,7 @@ def fix_params(self, params):
             self._params_index[key] for key in self._fixed_params.keys()
         ]
         self._free_params_index = list(
-            set(np.arange(self.k_params)).difference(self._fixed_params_index)
+            set(np.arange(k_params)).difference(self._fixed_params_index)
         )
 
         try:
@@ -183,15 +183,6 @@ def param_names(self):
                 names = []
             return names
 
-    @classmethod
-    def from_formula(
-        cls, formula, data, subset=None, drop_cols=None, *args, **kwargs
-    ):
-        """
-        Not implemented for state space models
-        """
-        raise NotImplementedError
-
     def _wrap_data(self, data, start_idx, end_idx, names=None):
         # squeezing data: data may be:
         # - m x n: m dates, n simulations -> squeeze does nothing
@@ -267,7 +258,11 @@ def _hessian_finite_difference(self, params, approx_centered=False,
                                 epsilon=epsilon, kwargs=kwargs,
                                 centered=approx_centered)
 
-        return hessian / (self.nobs_effective)
+        if hasattr(self, "ssm"):
+            nobs_effective = self.nobs - self.ssm.loglikelihood_burn
+        else:
+            nobs_effective = self.nobs
+        return hessian / nobs_effective
 
     def _hessian_complex_step(self, params, **kwargs):
         """
@@ -281,7 +276,11 @@ def _hessian_complex_step(self, params, **kwargs):
         hessian = approx_hess_cs(
             params, self.loglike, epsilon=epsilon, kwargs=kwargs)
 
-        return hessian / (self.nobs_effective)
+        if hasattr(self, "ssm"):
+            nobs_effective = self.nobs - self.ssm.loglikelihood_burn
+        else:
+            nobs_effective = self.nobs
+        return hessian / nobs_effective
 
 
 class StateSpaceMLEResults(tsbase.TimeSeriesModelResults):
@@ -330,17 +329,15 @@ def __init__(self, model, params, normalized_cov_params=None, scale=1.0):
 
         # Dimensions
         self.nobs = self.model.nobs
-        self.k_params = self.model.k_params
+        if hasattr(model, "k_params"):
+            self.k_params = self.model.k_params
 
         self._rank = None
 
     @cache_readonly
     def nobs_effective(self):
         raise NotImplementedError
 
-    @cache_readonly
-    def df_resid(self):
-        return self.nobs_effective - self.df_model
 
     @cache_readonly
     def aic(self):
@@ -530,7 +527,7 @@ def test_normality(self, method):
                 d = 0
             output = []
             for i in range(self.model.k_endog):
-                if hasattr(self, "fiter_results"):
+                if hasattr(self, "filter_results"):
                     resid = self.filter_results.standardized_forecasts_error[
                         i, d:
                     ]

statsmodels/tsa/exponential_smoothing/ets.py

@@ -1396,6 +1396,8 @@ def __init__(self, model, params, results):
         # degrees of freedom of model
         k_free_params = self.k_params - len(self.fixed_params)
         self.df_model = k_free_params + 1
+        self.df_resid = self.nobs_effective - self.df_model
+
 
         # standardized forecasting error
         self.mean_resid = np.mean(self.resid)

statsmodels/tsa/statespace/mlemodel.py

@@ -24,9 +24,9 @@
 import statsmodels.tsa.base.prediction as pred
 
 from statsmodels.base.data import PandasData
-import statsmodels.tsa.base.tsa_model as tsbase
 from statsmodels.tsa.base.mlemodel import (StateSpaceMLEModel,
                                            StateSpaceMLEResults)
+import statsmodels.tsa.base.tsa_model as tsbase
 
 from .news import NewsResults
 from .simulation_smoother import SimulationSmoother
@@ -183,9 +183,6 @@ def initialize_statespace(self, **kwargs):
         # Other dimensions, now that `ssm` is available
         self.k_endog = self.ssm.k_endog
 
-        # Effective number of observations
-        self.nobs_effective = self.nobs - self.ssm.loglikelihood_burn
-
     def _get_index_with_final_state(self):
         # The index we inherit from `TimeSeriesModel` will only cover the
         # data sample itself, but we will also need an index value for the
@@ -1981,6 +1978,13 @@ def impulse_responses(self, params, steps=1, impulse=0,
                 irfs = pd.DataFrame(irfs, columns=self.endog_names)
         return irfs
 
+    @classmethod
+    def from_formula(cls, formula, data, subset=None):
+        """
+        Not implemented for state space models
+        """
+        raise NotImplementedError
+
 
 class MLEResults(StateSpaceMLEResults):
     r"""
@@ -2045,6 +2049,7 @@ def __init__(self, model, params, results, cov_type=None, cov_kwds=None,
         k_free_params = self.params.size - len(self.fixed_params)
         self.df_model = (k_free_params + self.k_diffuse_states
                          + self.filter_results.filter_concentrated)
+        self.df_resid = self.nobs_effective - self.df_model
 
         # Setup covariance matrix notes dictionary
         if not hasattr(self, 'cov_kwds'):

statsmodels/tsa/statespace/sarimax.py

@@ -1801,6 +1801,8 @@ def __init__(self, model, params, filter_results, cov_type=None,
         super(SARIMAXResults, self).__init__(model, params, filter_results,
                                              cov_type, **kwargs)
 
+        self.df_resid = np.inf  # attribute required for wald tests
+
         # Save _init_kwds
         self._init_kwds = self.model._get_init_kwds()
 
@@ -1876,10 +1878,6 @@ def __init__(self, model, params, filter_results, cov_type=None,
         # Handle removing data
         self._data_attr_model.extend(['orig_endog', 'orig_exog'])
 
-    @cache_readonly
-    def df_resid(self):
-        return np.inf  # attribute required for wald tests
-
     def extend(self, endog, exog=None, **kwargs):
         kwargs.setdefault('trend_offset', self.nobs + 1)
         return super(SARIMAXResults, self).extend(endog, exog=exog, **kwargs)