9.2. Rolling returns

Takeaway

When you want rolling returns over some calendar unit (hours, days, weeks, months, year), in realistic data with many firms and returns missing for parts of the calendar (e.g. no weekend trading, no after-hours trading):

  1. df['newVarName'] = df.set_index('date').groupby('asset')['ret'].rolling(window=<dateoffset>,1).apply(lambda x: np.prod(1+x)-1).values

  2. Do not use window=# unless you are sure your data is perfectly formed. window=5 fails in real data because stock markets close for the weekend.

  3. The <dateoffset> could be ‘7D’, ‘24H’, ‘1M’ or many other things - see this link.

  4. The docs for the rolling function provide extra functionality.

Suppose we have an asset-timeperiod dataset, in long/tidy form. For example, this is a firm-day trading dataset:

import pandas as pd
import numpy as np

df = pd.DataFrame({'Firm': [1,1,1,1,1,1,1,2,2,2,2,2,2,2],
                  'Date' : ['1/11/22','1/12/22','1/13/22','1/14/22','1/18/22','1/19/22','1/30/22',
                            '1/11/22','1/12/22','1/13/22','1/14/22','1/18/22','1/19/22','1/30/22'],
                  'ret' : [.01,.01,.01,-.0294,.01,-0.01,.03,.02,-.02,.03,-.03,.04,-.04,.05] })
df['Date'] = pd.to_datetime(df.Date,format="%m/%d/%y")
df
Firm Date ret
0 1 2022-01-11 0.0100
1 1 2022-01-12 0.0100
2 1 2022-01-13 0.0100
3 1 2022-01-14 -0.0294
4 1 2022-01-18 0.0100
5 1 2022-01-19 -0.0100
6 1 2022-01-30 0.0300
7 2 2022-01-11 0.0200
8 2 2022-01-12 -0.0200
9 2 2022-01-13 0.0300
10 2 2022-01-14 -0.0300
11 2 2022-01-18 0.0400
12 2 2022-01-19 -0.0400
13 2 2022-01-30 0.0500

9.2.1. The problem

Let’s create a variable containing the cumulative return over the last week. Let’s stipulate that if we have less than fives days of returns for a firm at any point, we just use what we have.

9.2.2. Working the problem

This one-liner is very close to correct:

df.groupby('Firm')['ret'].rolling(5,1).apply(lambda x: np.prod(1+x)-1)

It’s not correct, but let’s start here because it’s easier to understand. Let me separate that code up and add comments for each part:

(df.groupby('Firm')['ret']         # for each firm, grab the ret var
 .rolling(window=5,min_periods=1)  # in each row, use the prior 1-5 obs (including the current one)
 .apply(lambda x: np.prod(1+x)-1)  # R[0,T] = prod{ R(0), ..., R(T) } - 1
)

To add that variable to our dataset, df['varname'] = <the above> doesn’t quite work. A small workaround does it: add .values to the end of the line. (It returns just the column of numbers and ignores the weird index the code above makes.)

So let’s look at the result:

df['ret1week_wrong'] = df.groupby('Firm')['ret'].rolling(5,1).apply(lambda x: np.prod(1+x)-1).values
df[:7] # just look at the top of the dataset
Firm Date ret ret1week_wrong
0 1 2022-01-11 0.0100 0.010000
1 1 2022-01-12 0.0100 0.020100
2 1 2022-01-13 0.0100 0.030301
3 1 2022-01-14 -0.0294 0.000010
4 1 2022-01-18 0.0100 0.010010
5 1 2022-01-19 -0.0100 -0.009990
6 1 2022-01-30 0.0300 0.009614

Warning

The result above is wrong!

  • 1/17 was a holiday, so the return for 1/18 should start with 1/12 (and only use 4 rows).

  • 1/30 should only include itself since we are missing dates.

Let’s fix that. The documentation mentions that the window length can use a time period!

So instead of window=5, we can use window='7D'!

Except there is a final complication: After df.groupby('Firm')['ret'], the date variable is gone. And rolling() needs that, So we need to keep the date variable around somehow.

The last fix: Add .set_index('Date') before groupby, so that the date variable will stick around as the index.

9.2.3. The solution

Put all that together:

  1. Put the date in the index

  2. “For each asset”

  3. Grab the returns for the prior week (using the dates to allow for holidays)

  4. Cumulate the returns

  5. Add that variable to the dataset

We can write this in one line:

df['ret1week'] = df.set_index('Date').groupby('Firm')['ret'].rolling('7D',1).apply(lambda x: np.prod(1+x)-1).values

9.2.4. The answer

Note

Look at the last three dates for each firm, and you’ll see how incorrect the window=# approach is.

df.set_index(['Firm','Date'])
ret ret1week_wrong ret1week
Firm Date
1 2022-01-11 0.0100 0.010000 0.010000
2022-01-12 0.0100 0.020100 0.020100
2022-01-13 0.0100 0.030301 0.030301
2022-01-14 -0.0294 0.000010 0.000010
2022-01-18 0.0100 0.010010 0.000010
2022-01-19 -0.0100 -0.009990 -0.019792
2022-01-30 0.0300 0.009614 0.030000
2 2022-01-11 0.0200 0.020000 0.020000
2022-01-12 -0.0200 -0.000400 -0.000400
2022-01-13 0.0300 0.029588 0.029588
2022-01-14 -0.0300 -0.001300 -0.001300
2022-01-18 0.0400 0.038648 0.018283
2022-01-19 -0.0400 -0.022449 -0.002499
2022-01-30 0.0500 0.047377 0.050000