BUG: DataFrame.min/max with axis=1 and uniform datetime64[ns, tz] types does not return NaNs

doc/source/whatsnew/v0.24.0.rst

@@ -1647,6 +1647,7 @@ Timezones
 - Bug in :meth:`DataFrame.any` returns wrong value when ``axis=1`` and the data is of datetimelike type (:issue:`23070`)
 - Bug in :meth:`DatetimeIndex.to_period` where a timezone aware index was converted to UTC first before creating :class:`PeriodIndex` (:issue:`22905`)
 - Bug in :meth:`DataFrame.tz_localize`, :meth:`DataFrame.tz_convert`, :meth:`Series.tz_localize`, and :meth:`Series.tz_convert` where ``copy=False`` would mutate the original argument inplace (:issue:`6326`)
+- Bug in :meth:`DataFrame.max` and :meth:`DataFrame.min` with ``axis=1`` where a :class:`Series` with ``NaN`` would be returned when all columns contained the same timezone (:issue:`10390`)
 
 Offsets
 ^^^^^^^

pandas/core/frame.py

@@ -49,6 +49,7 @@
     find_common_type)
 from pandas.core.dtypes.common import (
     is_dict_like,
+    is_datetime64tz_dtype,
     is_object_dtype,
     is_extension_type,
     is_extension_array_dtype,
@@ -7390,7 +7391,9 @@ def f(x):
             return op(x, axis=axis, skipna=skipna, **kwds)
 
         # exclude timedelta/datetime unless we are uniform types
-        if axis == 1 and self._is_mixed_type and self._is_datelike_mixed_type:
+        if (axis == 1 and self._is_datelike_mixed_type
+                and (not self._is_homogeneous_type
+                     and not is_datetime64tz_dtype(self.dtypes[0]))):
             numeric_only = True
 
         if numeric_only is None:

pandas/tests/reductions/test_reductions.py

@@ -135,6 +135,19 @@ def test_nanops(self):
         assert obj.argmin(skipna=False) == -1
         assert obj.argmax(skipna=False) == -1
 
+    @pytest.mark.parametrize('op, expected_col', [
+        ['max', 'a'], ['min', 'b']
+    ])
+    def test_same_tz_min_max_axis_1(self, op, expected_col):
+        # GH 10390
+        df = DataFrame(pd.date_range('2016-01-01 00:00:00', periods=3,
+                                     tz='UTC'),
+                       columns=['a'])
+        df['b'] = df.a.subtract(pd.Timedelta(seconds=3600))
+        result = getattr(df, op)(axis=1)
+        expected = df[expected_col]
+        tm.assert_series_equal(result, expected)
+
 
 class TestSeriesReductions(object):
     # Note: the name TestSeriesReductions indicates these tests