to_datetime returns NaT for old dates with coerce=True

[vfilimonov]

When use coerce=True in pandas.to_datetime old dates (before 1677-09-22) return NaT. The same dates without coerce are parsed correctly:

print pd.to_datetime('13000101', format='%Y%m%d', coerce=True)
print pd.to_datetime('13000101', format='%Y%m%d')

returns

NaT
1300-01-01 00:00:00

pandas: 0.15.2
numpy: 1.9.1

[shoyer]

I believe this is intentional, although it is poorly documented. The return value in the second case is a datetime.datetime object rather than a pandas.Timestamp object. @jreback thoughts?

[shoyer]

To be clear, this (old dates) is a case where to_datetime cannot turn the date into a Timestamp, which is the normal result. So it's a choice (controlled here by coerce) between returning a pandas compatible NaT object or plain python datetime object.

[jreback]

this is correct, the 2nd is a datetime.datetime (Timestamp is a sub-class of this).

not a bug. let's turn this to a further update to the docs (see #8921).

@vfilimonov want to do this?