BUG: Fix predict when exog or endog is None

Author: bashtage

linearmodels/iv/model.py

@@ -50,6 +50,7 @@
     OneWayClusteredWeightMatrix,
 )
 from linearmodels.iv.results import IVGMMResults, IVResults, OLSResults
+from linearmodels.panel.utility import InvalidFormulaError
 from linearmodels.shared.exceptions import IndexWarning, missing_warning
 from linearmodels.shared.hypotheses import InvalidTestStatistic, WaldTestStatistic
 from linearmodels.shared.linalg import has_constant, inv_sqrth
@@ -291,15 +292,34 @@ def predict(
                 "Predictions can only be constructed using one "
                 "of exog/endog or data, but not both."
             )
-        if exog is not None or endog is not None:
+        if exog is not None:
             exog = IVData(exog).pandas
+        if endog is not None:
             endog = IVData(endog).pandas
-        elif data is not None:
+        if data is not None:
+            if endog is not None or exog is not None:
+                raise ValueError(
+                    "Predictions can be constructed using either exog and endog "
+                    "or data, but not both."
+                )
             parser = IVFormulaParser(self.formula, data, eval_env=eval_env)
             exog = parser.exog
             endog = parser.endog
-        else:
-            raise ValueError("exog and endog or data must be provided.")
+            if exog is None and exog is None:
+                raise InvalidFormulaError(
+                    f"Parsed formula ({self.formula}) has no exog and no endog. One "
+                    f"of these must be included in the formula to make a prediction."
+                )
+        if all(a is None for a in (exog, endog, data)):
+            raise ValueError("At least one of exog, endog, or data must be provided.")
+        if exog is None:
+            assert endog is not None
+            exog = IVData(exog, nobs=endog.shape[0]).pandas
+            exog.index = endog.index
+        if endog is None:
+            assert exog is not None
+            endog = IVData(endog, nobs=exog.shape[0]).pandas
+            endog.index = exog.index
         assert exog is not None
         assert endog is not None
         if exog.shape[0] != endog.shape[0]:

linearmodels/panel/utility.py

@@ -24,6 +24,7 @@
 __all__ = [
     "AbsorbingEffectError",
     "AbsorbingEffectWarning",
+    "InvalidFormulaError",
     "_drop_singletons",
     "_py_drop_singletons",
     "_remove_node",
@@ -688,3 +689,7 @@ def generate_panel_data(
     clusters = concat([vc1_df, vc2_df], sort=False)
     data = concat([y_df, x_df], axis=1, sort=False)
     return PanelModelData(data, w_df, other_eff, clusters)
+
+
+class InvalidFormulaError(Exception):
+    pass

linearmodels/tests/iv/test_formulas.py

@@ -210,7 +210,9 @@ def test_predict_formula(data, model_and_func, formula):
     assert_frame_equal(pred, pred2)
     assert_allclose(res.fitted_values, pred)
 
-    with pytest.raises(ValueError, match=r"exog and endog or data must be provided"):
+    with pytest.raises(
+        ValueError, match=r"At least one of exog, endog, or data must be provided"
+    ):
         mod.predict(res.params)
 
 
@@ -404,3 +406,13 @@ def test_formula_categorical_equiv(data, model_and_func, dtype):
         "x2",
         "x3",
     ]
+
+
+def test_predict_no_rhs(data, model_and_func):
+    model, _ = model_and_func
+    mod = model.from_formula("y ~", data)
+    res = mod.fit()
+    pred0 = res.predict()
+    pred1 = res.predict(data=data)
+    pred1.columns = pred0.columns
+    assert_frame_equal(pred0, pred1)

linearmodels/tests/iv/test_results.py

@@ -1,7 +1,7 @@
 from numpy import asarray
 from numpy.testing import assert_allclose
 from pandas import DataFrame
-from pandas.testing import assert_series_equal
+from pandas.testing import assert_frame_equal, assert_series_equal
 import pytest
 
 from linearmodels.iv.data import IVData
@@ -96,3 +96,36 @@ def test_predict_no_selection(data, model):
     res = mod.fit()
     with pytest.raises(ValueError, match=r"At least one output must be selected"):
         res.predict(fitted=False, idiosyncratic=False, missing=True)
+
+
+@pytest.mark.parametrize("include_vars", ["exog", "both", "endog"])
+def test_fitted_predict_combinations(data, model, include_vars):
+    args = (data.dep,)
+    if include_vars in ("exog", "both"):
+        args += (data.exog,)
+    else:
+        args += (None,)
+    if include_vars in ("endog", "both"):
+        args += (data.endog, data.instr)
+    else:
+        args += (None, None)
+
+    mod = model(*args)
+    res = mod.fit()
+    assert_series_equal(res.idiosyncratic, res.resids)
+    y = mod.dependent.pandas
+    expected = asarray(y) - asarray(res.resids)[:, None]
+    expected = DataFrame(expected, y.index, ["fitted_values"])
+    assert_frame_similar(expected, res.fitted_values)
+    assert_allclose(expected, res.fitted_values)
+    pred = res.predict()
+    pred2 = res.predict(exog=args[1], endog=args[2])
+    pred2.columns = pred.columns
+    assert_frame_equal(pred, pred2)
+    nobs = res.resids.shape[0]
+    assert isinstance(pred, DataFrame)
+    assert pred.shape == (nobs, 1)
+    pred = res.predict(idiosyncratic=True, missing=True)
+    nobs = IVData(data.dep).pandas.shape[0]
+    assert pred.shape == (nobs, 2)
+    assert list(pred.columns) == ["fitted_values", "residual"]