BUG: unstack with missing levels results in incorrect index names

doc/source/whatsnew/v1.2.0.rst

@@ -697,6 +697,8 @@ Indexing
 - Bug in setting a new label on a :class:`DataFrame` or :class:`Series` with a :class:`CategoricalIndex` incorrectly raising ``TypeError`` when the new label is not among the index's categories (:issue:`38098`)
 - Bug in :meth:`Series.loc` and :meth:`Series.iloc` raising ``ValueError`` when inserting a listlike ``np.array``, ``list`` or ``tuple`` in an ``object`` Series of equal length (:issue:`37748`, :issue:`37486`)
 - Bug in :meth:`Series.loc` and :meth:`Series.iloc` setting all the values of an ``object`` Series with those of a listlike ``ExtensionArray`` instead of inserting it (:issue:`38271`)
+- Bug in :meth:`MultiIndex.remove_unused_levels` drops NaN when level contains NaN (:issue:`37510`)
+- Bug in :meth:`MultiIndex.remove_unused_levels` was dropping missing values when levels contain ``NaN`` which set by ``set_levels`` (:issue:`37510`)
 
 Missing
 ^^^^^^^
@@ -827,6 +829,7 @@ Reshaping
 - Bug in :func:`merge_ordered` returned wrong join result when length of ``left_by`` or ``right_by`` equals to the rows of ``left`` or ``right`` (:issue:`38166`)
 - Bug in :func:`merge_ordered` didn't raise when elements in ``left_by`` or ``right_by`` not exist in ``left`` columns or ``right`` columns (:issue:`38167`)
 - Bug in :func:`DataFrame.drop_duplicates` not validating bool dtype for ``ignore_index`` keyword (:issue:`38274`)
+- Bug in :meth:`DataFrame.unstack` with missing levels led to incorrect index names (:issue:`37510`)
 
 Sparse
 ^^^^^^

doc/source/whatsnew/v1.3.0.rst

@@ -267,7 +267,7 @@ Groupby/resample/rolling
 Reshaping
 ^^^^^^^^^
 
--
+- Bug in :meth:`DataFrame.unstack` with missing levels led to incorrect index names (:issue:`37510`)
 -
 
 Sparse

pandas/core/indexes/multi.py

@@ -1979,6 +1979,9 @@ def remove_unused_levels(self):
             has_na = int(len(uniques) and (uniques[0] == -1))
 
             if len(uniques) != len(lev) + has_na:
+
+                if lev.isna().any() and len(uniques) == len(lev):
+                    break
                 # We have unused levels
                 changed = True
 

pandas/tests/indexes/multi/test_sorting.py

@@ -271,3 +271,28 @@ def test_argsort(idx):
     result = idx.argsort()
     expected = idx.values.argsort()
     tm.assert_numpy_array_equal(result, expected)
+
+
+def test_remove_unused_levels_with_missing():
+    # GH 37510
+    df1 = DataFrame(
+        {
+            "L1": [1, 2, 3, 4],
+            "L2": [3, 4, 1, 2],
+            "L3": [1, 1, 1, 1],
+            "x": [1, 2, 3, 4],
+        }
+    )
+    df1 = df1.set_index(["L1", "L2", "L3"])
+    new_levels = ["n1", "n2", "n3", None]
+    df1.index = df1.index.set_levels(levels=new_levels, level="L1")
+    df1.index = df1.index.set_levels(levels=new_levels, level="L2")
+
+    result = df1.unstack("L3")[("x", 1)].sort_index().index
+    expected = MultiIndex(
+        levels=[["n1", "n2", "n3", None], ["n1", "n2", "n3", None]],
+        codes=[[0, 1, 2, 3], [2, 3, 0, 1]],
+        names=["L1", "L2"],
+    )
+
+    tm.assert_index_equal(result, expected)