BUG: Concatenating DataFrames with NaT TZ Timestamps results in incorrect dtype

[tonytao2012]

Code Sample, a copy-pastable example if possible

ts1 = pd.Timestamp(pd.NaT, tz='UTC')
ts2 = pd.Timestamp('2015-01-01', tz='UTC')

df1 = pd.DataFrame([[ts1]])
df2 = pd.DataFrame([[ts2]])

result = pd.concat([df1, df2])

result[0]
Out[6]: 
0                          NaT
0    2015-01-01 00:00:00+00:00
Name: 0, dtype: object

expected = pd.DataFrame([[ts1], [ts2]])

expected[0]
Out[8]: 
0                         NaT
1   2015-01-01 00:00:00+00:00
Name: 0, dtype: datetime64[ns, UTC]

Problem description

Concatenating a DataFrame containing NaT with a tz specified with another DataFrame containing an actual date with tz specified results in an object dtype instead of the expected datetime64[ns, UTC] dtype.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.6.final.0
python-bits: 64
OS: Windows
OS-release: 10
machine: AMD64
processor: Intel64 Family 6 Model 158 Stepping 9, GenuineIntel
byteorder: little
LC_ALL: None
LANG: en
LOCALE: None.None

pandas: 0.24.0.dev0+708.gce1f81f8b
pytest: 3.8.2
pip: 10.0.1
setuptools: 40.2.0
Cython: 0.28.5
numpy: 1.15.2
scipy: 1.1.0
pyarrow: None
xarray: 0.10.9
IPython: 6.5.0
sphinx: 1.8.1
patsy: 0.5.0
dateutil: 2.7.3
pytz: 2018.5
blosc: None
bottleneck: 1.2.1
tables: 3.4.4
numexpr: 2.6.8
feather: None
matplotlib: 3.0.0
openpyxl: 2.5.5
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: 1.1.1
lxml: 4.2.5
bs4: 4.6.3
html5lib: 1.0.1
sqlalchemy: 1.2.12
pymysql: 0.9.2
psycopg2: None
jinja2: 2.10
s3fs: 0.1.6
fastparquet: 0.1.6
pandas_gbq: None
pandas_datareader: None
gcsfs: 0.1.2

[TomAugspurger]

I think the issue is in the DataFrame constructor?

In [48]: df1.dtypes
Out[48]:
0    datetime64[ns]
dtype: object

That said, there's no way for the constructor to infer datetime64[ns, UTC] as the correct dtype. You would need df1 = pd.DataFrame([[ts1]], dtype='datetime64[ns, UTC]')

[mroeschke]

I think is the intended behavior.

When concating two different dataframes with different timezone information (in this case None and UTC, we aim to preserve the timezone information on both sides so the object type is expected. One can use to_datetime to get back a datetime64[ns, UTC] dtype after the operation