disable probabilistic capability of StatsForecastMSTL

sktime/forecasting/statsforecast.py

@@ -796,8 +796,8 @@ class StatsForecastMSTL(_GeneralisedStatsForecastAdapter):
         # estimator type
         # --------------
         "ignores-exogeneous-X": True,
-        "capability:pred_int": True,
-        "capability:pred_int:insample": True,
+        "capability:pred_int": False,
+        "capability:pred_int:insample": False,
         "python_dependencies": ["statsforecast>=1.2.0"],
     }
 
@@ -815,6 +815,14 @@ def __init__(
 
         super().__init__()
 
+        # adapter class sets probabilistic capability as true
+        # because level is present in statsforecast signature
+        # happens in _check_supports_pred_int method
+        # manually overriding this temporarily
+        self.set_tags(
+            **{"capability:pred_int": False, "capability:pred_int:insample": False}
+        )
+
         from sklearn.base import clone
 
         if trend_forecaster:

sktime/forecasting/tests/test_statsforecast.py

@@ -14,6 +14,9 @@
 from sktime.tests.test_switch import run_test_for_class
 
 
+@pytest.mark.skip(
+    reason="probabilistic capability of StatsForecastMSTL is disabled, see #5703, #5920"
+)
 @pytest.mark.skipif(
     not run_test_for_class(StatsForecastMSTL),
     reason="run test only if softdeps are present and incrementally (if requested)",

sktime/tests/_config.py

@@ -189,6 +189,8 @@
     "LTSFDLinearForecaster": ["test_predict_time_index_in_sample_full"],
     "LTSFNLinearForecaster": ["test_predict_time_index_in_sample_full"],
     "WEASEL": ["test_multiprocessing_idempotent"],  # see 5658
+    # StatsForecastMSTL is failing in probabistic forecasts, see #5703, #5920
+    "StatsForecastMSTL": ["test_pred_int_tag", "test_statsforecast_mstl"],
 }
 
 # We use estimator tags in addition to class hierarchies to further distinguish