BUG: Min/max does not work for dates with timezones if there are missing values in the data frame

doc/source/whatsnew/v1.4.0.rst

@@ -549,7 +549,7 @@ Missing
 ^^^^^^^
 - Bug in :meth:`DataFrame.fillna` with limit and no method ignores axis='columns' or ``axis = 1`` (:issue:`40989`)
 - Bug in :meth:`DataFrame.fillna` not replacing missing values when using a dict-like ``value`` and duplicate column names (:issue:`43476`)
--
+- Bug in :meth:`DataFrame.max`, :meth:`DataFrame.min`, :meth:`Series.max` and :meth:`Series.min` when called on datetime columns with timezone aware data and missing elements (:issue:`27794` and :issue:`44196`)
 
 MultiIndex
 ^^^^^^^^^^

pandas/core/frame.py

@@ -9979,7 +9979,11 @@ def _get_data() -> DataFrame:
                 data = self._get_bool_data()
             return data
 
-        if numeric_only is not None or axis == 0:
+        if (
+            numeric_only is not None
+            or axis == 0
+            or (name in ["max", "min"] and axis == 1)
+        ):
             # For numeric_only non-None and axis non-None, we know
             #  which blocks to use and no try/except is needed.
             #  For numeric_only=None only the case with axis==0 and no object

pandas/tests/frame/test_reductions.py

@@ -17,7 +17,9 @@
     DataFrame,
     Index,
     MultiIndex,
+    PeriodDtype,
     Series,
+    Timedelta,
     Timestamp,
     date_range,
     isna,
@@ -756,7 +758,7 @@ def test_operators_timedelta64(self):
         # excludes numeric
         with tm.assert_produces_warning(FutureWarning, match="Select only valid"):
             result = mixed.min(axis=1)
-        expected = Series([1, 1, 1.0], index=[0, 1, 2])
+        expected = Series([])
         tm.assert_series_equal(result, expected)
 
         # works when only those columns are selected
@@ -1763,3 +1765,141 @@ def test_prod_sum_min_count_mixed_object():
     msg = re.escape("unsupported operand type(s) for +: 'int' and 'str'")
     with pytest.raises(TypeError, match=msg):
         df.sum(axis=0, min_count=1, numeric_only=False)
+
+
+def test_timezone_min_max_with_nat():
+    # GH#27794
+    df = pd.DataFrame(
+        {
+            "A": pd.date_range(start="2018-01-01", end="2018-01-03", tz="UTC"),
+            "B": pd.date_range(start="2018-01-01", end="2018-01-02", tz="UTC").insert(
+                2, pd.NaT
+            ),
+        }
+    )
+
+    expected = pd.Series(
+        [
+            pd.Timestamp("2018-01-01", tz="UTC"),
+            pd.Timestamp("2018-01-02", tz="UTC"),
+            pd.Timestamp("2018-01-03", tz="UTC"),
+        ],
+    )
+    result = df.min(axis=1)
+    tm.assert_series_equal(result, expected)
+
+    expected = pd.Series(
+        [
+            pd.Timestamp("2018-01-01", tz="UTC"),
+            pd.Timestamp("2018-01-02", tz="UTC"),
+            pd.Timestamp("2018-01-03", tz="UTC"),
+        ],
+    )
+    result = df.max(axis=1)
+    tm.assert_series_equal(result, expected)
+
+
+def test_min_max_timestamp_timezone_nat():
+    # GH#44196
+    rng_with_tz = pd.date_range(
+        start="2021-10-01T12:00:00+02:00", end="2021-10-02T12:00:00+02:00", freq="4H"
+    )
+    df_with_tz = DataFrame(
+        data={"A": rng_with_tz, "B": rng_with_tz + pd.Timedelta(minutes=20)}
+    )
+    df_with_tz.iloc[2, 1] = pd.NaT
+
+    result = df_with_tz.max(axis=1)
+    expected = pd.Series(
+        [
+            pd.Timestamp("2021-10-01T12:20:00+02:00"),
+            pd.Timestamp("2021-10-01T16:20:00+02:00"),
+            pd.Timestamp("2021-10-01T20:00:00+02:00"),
+            pd.Timestamp("2021-10-02T00:20:00+02:00"),
+            pd.Timestamp("2021-10-02T04:20:00+02:00"),
+            pd.Timestamp("2021-10-02T08:20:00+02:00"),
+            pd.Timestamp("2021-10-02T12:20:00+02:00"),
+        ]
+    )
+    tm.assert_series_equal(result, expected)
+
+
+def test_timezone_min_max_both_axis():
+    rng_with_tz = pd.date_range(
+        start="2021-10-01T12:00:00+02:00", end="2021-10-02T12:00:00+02:00", freq="4H"
+    )
+    df_with_tz = DataFrame(
+        data={"A": rng_with_tz, "B": rng_with_tz + pd.Timedelta(minutes=20)}
+    )
+    df_with_tz.iloc[2, 1] = pd.NaT
+
+    result = df_with_tz.max(axis=1)
+    expected = df_with_tz.T.max(axis=0)
+
+    tm.assert_series_equal(result, expected)
+
+    result = df_with_tz.min(axis=1)
+    expected = df_with_tz.T.min(axis=0)
+
+    tm.assert_series_equal(result, expected)
+
+
+def test_min_max_timedelta64_nat():
+    df = DataFrame(
+        [
+            [Timedelta(minutes=20), Timedelta(days=2), Timedelta(seconds=3)],
+            [Timedelta(minutes=2, seconds=2), Timedelta(days=2, minutes=30), pd.NaT],
+        ]
+    )
+    expected = pd.Series(
+        [Timedelta(minutes=2, seconds=2), Timedelta(days=2), Timedelta(seconds=3)]
+    )
+    result = df.min(axis=0)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.min(axis=0), df.T.min(axis=1))
+
+    expected = pd.Series([Timedelta(seconds=3), Timedelta(minutes=2, seconds=2)])
+    result = df.min(axis=1)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.min(axis=1), df.T.min(axis=0))
+
+    expected = pd.Series(
+        [Timedelta(minutes=20), Timedelta(days=2, minutes=30), Timedelta(seconds=3)]
+    )
+    result = df.max(axis=0)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.max(axis=0), df.T.max(axis=1))
+
+    expected = pd.Series([Timedelta(days=2), Timedelta(days=2, minutes=30)])
+    result = df.max(axis=1)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.max(axis=1), df.T.max(axis=0))
+
+
+def test_min_max_perioddtype_nat():
+    df = DataFrame(
+        [
+            [PeriodDtype(freq="20m"), PeriodDtype(freq="1h"), PeriodDtype(freq="1d")],
+            [PeriodDtype(freq="25m"), PeriodDtype(freq="2h"), pd.NaT],
+        ]
+    )
+
+    expected = Series([])
+    result = df.min(axis=0)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.min(axis=0), df.T.min(axis=1))
+
+    expected = Series([])
+    result = df.min(axis=1)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.min(axis=1), df.T.min(axis=0))
+
+    expected = Series([])
+    result = df.max(axis=0)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.max(axis=0), df.T.max(axis=1))
+
+    expected = Series([])
+    result = df.max(axis=1)
+    tm.assert_series_equal(result, expected)
+    tm.assert_series_equal(df.max(axis=1), df.T.max(axis=0))

pandas/tests/indexes/datetimes/test_date_range.py

@@ -746,7 +746,7 @@ def test_cached_range_bug(self):
         assert len(rng) == 50
         assert rng[0] == datetime(2010, 9, 1, 5)
 
-    def test_timezone_comparaison_bug(self):
+    def test_timezone_comparison_bug(self):
         # smoke test
         start = Timestamp("20130220 10:00", tz="US/Eastern")
         result = date_range(start, periods=2, tz="US/Eastern")