When attempting to set an index using a Period rather than a Datetime, the performance significantly deteriorates. A small example:
import pandas as pa
obs = 1000
start_date = '1/1/1900'
df = pa.DataFrame(list(range(1, obs + 1)))
df['dateindex'] = pa.date_range('1/1/1900', periods=obs, freq='M')
df['periodindex'] = pa.period_range('1/1/1900', periods=obs, freq='M')
%timeit df.set_index('dateindex')
#1000 loops, best of 3: 444 µs per loop
%timeit df.set_index('periodindex')
#1000 loops, best of 3: 1.89 ms per loop
This difference in performance becomes especially frustrating with larger dataframes. In order for Period to be as useful as Datetime, these should be much closer in performance.
Comment From: jreback
this is a symptom of #7964 it is an object dtype now and is not first class. contributions along those lines are welcome.