Pandas P&L rollup to the next business day

user_4046516 0
Viewed 179

I'm having a hard time trying to do this efficiently. I have some stocks and daily P&L info in a dataframe. In reality, I have millions of rows of data so efficiency matters a lot! The Dataframe looks like :

-------------------------------
| Date       | Security | P&L |
-------------------------------
| 2016-01-01 | AAPL     | 100 |
-------------------------------
| 2016-01-02 | AAPL     | 200 |
-------------------------------
| 2016-01-03 | AAPL     | 300 |
-------------------------------
| 2016-01-04 | AAPL     | -200 |
-------------------------------

All, I want to do is roll the P&L over to the next business day (exclude all US holidays and weekends) So, the resultant Dataframe looks like this:

-------------------------------
| Date       | Security | P&L |
-------------------------------
| 2016-01-04 | AAPL     | 400 |
-------------------------------

I'm looking for an efficient way to achieve this. I do have thousands of securities and over 5 yrs of data to process so brute force can't work, unfortunately!

Thanks in advance and highly appreciate any pointers on this!

quantitative-finance pandas-groupby python-datetime pandas python
2 Answers

We can create the DataFrame of business dates then merge_asof. Then we can group on this to get the sums. 

import pandas as pd
from pandas.tseries.holiday import USFederalHolidayCalendar

#df['Date'] = pd.to_datetime(df.Date)
date_min = '2015-01-01'
date_max = '2016-12-31'

cal = USFederalHolidayCalendar()
holidays = cal.holidays(date_min, date_max).tolist()
df2 = pd.DataFrame({'bdate': pd.bdate_range(date_min, date_max, 
                                            holidays=holidays, freq='C')})

res = pd.merge_asof(df, df2, left_on='Date', right_on='bdate', direction='forward')
#        Date Security  P&L      bdate
#0 2016-01-01     AAPL  100 2016-01-04
#1 2016-01-02     AAPL  200 2016-01-04
#2 2016-01-03     AAPL  300 2016-01-04
#3 2016-01-04     AAPL -200 2016-01-04

res.groupby(['Security', 'bdate'])['P&L'].sum()
#Security  bdate     
#AAPL      2016-01-04    400
user_43333590

IIUC you can do something like:

import pandas as pd
from pandas.tseries.holiday import USFederalHolidayCalendar
import numpy as np

date_min = '2015-01-01'
date_max = '2016-12-31'

cal = USFederalHolidayCalendar()
holidays = cal.holidays(date_min, date_max).tolist()

df = pd.DataFrame({"Date":pd.date_range(date_min, date_max)})
df["Security"] ="APPL"
df["P&L"] = np.random.randint(-1000, 1000, len(df))

df[~df["Date"].isin(holidays)].groupby("Security")\
                              .agg({"Date":"max",
                                    "P&L":"sum"})\
                              .reset_index()
user_48193760
Related