BUG: to_records() fails for empty MultiIndex

[fersarr]

Calling to_records() on an empty dataframe after a group-by results in a ValueError:

In [1]: import pandas as pd
In [2]: mydf = pd.DataFrame({'A':[10,-10], 'B':[400,400], 'C':["bla","bla"]})
In [3]: new_pd2 = pd.DataFrame(mydf.groupby(['B','C'])['A'].agg(lambda x: x.sum()))
In [4]: new_pd3 = new_pd2[new_pd2['A'] != 0 ]
In [5]: new_pd3.to_records()
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-ce958ec44125> in <module>()
----> 1 new_pd3.to_records()

/.../lib/python2.7/site-packages/pandas-0.22.0+ahl1-py2.7-linux-x86_64.egg/pandas/core/frame.pyc in to_records(self, index, convert_datetime64)
   1233         return np.rec.fromarrays(
   1234             arrays,
-> 1235             dtype={'names': names, 'formats': formats}
   1236         )
   1237 

/.../.eggs/numpy-1.14.2-py2.7-linux-x86_64.egg/numpy/core/records.pyc in fromarrays(arrayList, dtype, shape, formats, names, titles, aligned, byteorder)
    605 
    606     if dtype is not None:
--> 607         descr = sb.dtype(dtype)
    608         _names = descr.names
    609     else:

ValueError: 'names', 'formats', 'offsets', and 'titles' dict entries must have the same length

Problem description

Calling to_records() on an empty dataframe after a group-by results in a ValueError. I expect this to not throw and return the records.

pandas==0.22.0
numpy==1.14.2

The same code above without column 'C' (using 2 columns) works fine.

The problem is here:
https://github.com/pandas-dev/pandas/blob/master/pandas/core/frame.py#L1419

return np.rec.fromarrays(
            arrays,
            dtype={'names': names, 'formats': formats}
        )

where names and formats have different lengths and make numpy upset.

This seems to work in pandas==0.16.2 and returns the following

In [5]: new_pd3.to_records()
Out[5]: 
rec.array([], 
      dtype=[('B', '<i8')])

[chris-b1]

In particular the issue seems to be an empty MultiIndex

In [35]: df = pd.DataFrame(columns=['A'], index=pd.MultiIndex([['a'], ['b']], labels=[[], []]))

In [36]: df.to_records()
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)

You can workaround this with index=False, but would also welcome a PR to fix.

In [39]: df.to_records(index=False)
Out[39]: 
rec.array([],
          dtype=[('A', 'O')])

[fersarr]

@chris-b1 giving it a go here, but Im not very familiar with MultiIndex
#21082