Pandas: "data type datetime not understood" error in date offset operation

Question

I am trying to do date shift just as the answer in this post:

After pd.to_datetime(), the data type is datetime64[ns].

However I am receiving "data type 'datetime' not understood" error. The error comes from ops.py line 454:

     if (inferred_type in ('datetime64', 'datetime', 'date', 'time') or
            is_datetimetz(inferred_type)):

(Incidentally in debug mode, if I change to ('datetime64', 'a', 'b', 'c') it doesn't throw the error)

I am using pandas version 0.19.2 and numpy 1.13.1.