BUG: pct_change wrong result when there are duplicated indices

doc/source/whatsnew/v1.0.0.rst

@@ -765,6 +765,7 @@ Numeric
 - Bug in :class:`NumericIndex` construction that caused :class:`UInt64Index` to be casted to :class:`Float64Index` when integers in the ``np.uint64`` range were used to index a :class:`DataFrame` (:issue:`28279`)
 - Bug in :meth:`Series.interpolate` when using method=`index` with an unsorted index, would previously return incorrect results. (:issue:`21037`)
 - Bug in :meth:`DataFrame.round` where a :class:`DataFrame` with a :class:`CategoricalIndex` of :class:`IntervalIndex` columns would incorrectly raise a ``TypeError`` (:issue:`30063`)
+- Bug in :meth:`Series.pct_change` and :meth:`DataFrame.pct_change` when there are duplicated indices (:issue:`30463`)
 - Bug in :class:`DataFrame` cumulative operations (e.g. cumsum, cummax) incorrect casting to object-dtype (:issue:`19296`)
 
 Conversion

pandas/core/generic.py

@@ -9901,11 +9901,11 @@ def pct_change(self, periods=1, fill_method="pad", limit=None, freq=None, **kwar
             data = self.fillna(method=fill_method, limit=limit, axis=axis)
 
         rs = data.div(data.shift(periods=periods, freq=freq, axis=axis, **kwargs)) - 1
-        rs = rs.loc[~rs.index.duplicated()]
-        rs = rs.reindex_like(data)
-        if freq is None:
-            mask = isna(com.values_from_object(data))
-            np.putmask(rs.values, mask, np.nan)
+        if freq is not None:
+            # Shift method is implemented differently when freq is not None
+            # We want to restore the original index
+            rs = rs.loc[~rs.index.duplicated()]
+            rs = rs.reindex_like(data)
         return rs
 
     def _agg_by_level(self, name, axis=0, level=0, skipna=True, **kwargs):

pandas/tests/frame/methods/test_pct_change.py

@@ -76,3 +76,21 @@ def test_pct_change_periods_freq(
         rs_freq = empty_ts.pct_change(freq=freq, fill_method=fill_method, limit=limit)
         rs_periods = empty_ts.pct_change(periods, fill_method=fill_method, limit=limit)
         tm.assert_frame_equal(rs_freq, rs_periods)
+
+
+@pytest.mark.parametrize("fill_method", ["pad", "ffill", None])
+def test_pct_change_with_duplicated_indices(fill_method):
+    # GH30463
+    data = DataFrame(
+        {0: [np.nan, 1, 2, 3, 9, 18], 1: [0, 1, np.nan, 3, 9, 18]}, index=["a", "b"] * 3
+    )
+    result = data.pct_change(fill_method=fill_method)
+    if fill_method is None:
+        second_column = [np.nan, np.inf, np.nan, np.nan, 2.0, 1.0]
+    else:
+        second_column = [np.nan, np.inf, 0.0, 2.0, 2.0, 1.0]
+    expected = DataFrame(
+        {0: [np.nan, np.nan, 1.0, 0.5, 2.0, 1.0], 1: second_column},
+        index=["a", "b"] * 3,
+    )
+    tm.assert_frame_equal(result, expected)

pandas/tests/series/methods/test_pct_change.py

@@ -68,3 +68,12 @@ def test_pct_change_periods_freq(
         rs_freq = empty_ts.pct_change(freq=freq, fill_method=fill_method, limit=limit)
         rs_periods = empty_ts.pct_change(periods, fill_method=fill_method, limit=limit)
         tm.assert_series_equal(rs_freq, rs_periods)
+
+
+@pytest.mark.parametrize("fill_method", ["pad", "ffill", None])
+def test_pct_change_with_duplicated_indices(fill_method):
+    # GH30463
+    s = Series([np.nan, 1, 2, 3, 9, 18], index=["a", "b"] * 3)
+    result = s.pct_change(fill_method=fill_method)
+    expected = Series([np.nan, np.nan, 1.0, 0.5, 2.0, 1.0], index=["a", "b"] * 3)
+    tm.assert_series_equal(result, expected)