Mixing Timeframes in Indicators

Release 1.3.0.92 brings up the possibility to have data (from either data feeds and/or indicators) from different timeframes mixed.

Background: Indicators are smart dumb objects.

They are smart because they can make complex calculations.
They are dumb because they operate with no knowledge of what sources are providing the data for the calculations

As such:

If the data sources providing the values have different timeframes, different lengths inside the Cerebro engine, the indicator will break.

Example of a calculation, in which data0 has a timeframe of days and data1 has a timeframe of months:

pivotpoint = btind.PivotPoint(self.data1)
sellsignal = self.data0.close < pivotpoint.s1

Here a sell signal is sought when the close is below the s1 line (1^st support)

Note

PivotPoint works in a larger timeframe by definition

This will in the past with the following error:

return self.array[self.idx + ago]
IndexError: array index out of range

And for a good reason: self.data.close provides values from the very 1^st instant, but PivotPoint (and hence the s1 line) will only deliver values once a full month has gone by, which is roughly equivalent to 22 values of self.data0.close. During this 22 closes there isn’t yet a value for s1 and the attempt to fetch it from the underlying array fails.

Lines objects support the (ago) operator (__call__ special method in Python) for deliver a delayed version of itself:

close1 = self.data.close(-1)

In this example the object close1 (when accessed via [0]) always contains the previous value (-1) delivered by close. The syntax has been reused to accomodate adapting timeframes. Let’s rewrite the above pivotpoint snippet:

pivotpoint = btind.PivotPoint(self.data1)
sellsignal = self.data0.close < pivotpoint.s1()

See how the () is executed with no arguments (in the background a None is being supplied). The following is happening:

pivotpoint.s1() is returning an internal LinesCoupler object which follows the rhythm of the larger scope. This coupler fills itself with the latest delivered value from the real s1 (starting with a default value of NaN)

But something extra is needed to make the magic work. Cerebro has to be created with:

cerebro = bt.Cerebro(runonce=False)

or executed with:

cerebro.run(runonce=False)

In this mode the indicators and late-evaluated automatic lines objects are executed step by step rather than in tight loops. This makes the entire operation slower, but it makes it possible

The sample script at the bottom which was breaking above, now runs:

$ ./mixing-timeframes.py

With output:

0021,0021,0001,2005-01-31,2984.75,2935.96,0.00
0022,0022,0001,2005-02-01,3008.85,2935.96,0.00
...
0073,0073,0003,2005-04-15,3013.89,3010.76,0.00
0074,0074,0003,2005-04-18,2947.79,3010.76,1.00
...

At line 74 the 1^st instance of close < s1 takes place.

The script also provides insight into the additional possiblity: couple all lines of an indicator. Before we had:

self.sellsignal = self.data0.close < pp.s1()

Being the alternative:

pp1 = pp()
self.sellsignal = self.data0.close < pp1.s1

Now the entire PivotPoint indicator has been coupled and any of its lines can be accessed (namely p, r1, r2, s1, s2). The script has only interest in s1 and the access is direct.:

$ ./mixing-timeframes.py --multi

The output:

0021,0021,0001,2005-01-31,2984.75,2935.96,0.00
0022,0022,0001,2005-02-01,3008.85,2935.96,0.00
...
0073,0073,0003,2005-04-15,3013.89,3010.76,0.00
0074,0074,0003,2005-04-18,2947.79,3010.76,1.00
...

No surprises here. The same as before. The “coupled” object can even be plotted:

$ ./mixing-timeframes.py --multi --plot

Full coupling syntax

For lines objects with multiple lines (for example Indicators like PivotPoint):

obj(clockref=None, line=-1)

clockref If clockref is None, the surrounding object (in the examples a Strategy) will be the reference to adapt larger timeframes (for example: Months) to smaller/faster timeframes (for example: Days)

Another reference can be used if desired

line

* If the default `-1` is given, all *lines* are coupled.

* If another integer (for example, `0` or `1`) a single line will be
  coupled and fetched by index (from `obj.lines[x]`)

* If a string is passed, the line will be fetched by name.

  In the sample, the following could have been done:

  ```python
  coupled_s1 = pp(line='s1')
  ```

For lines objects with a single line (for example line s1 from the indicator PivotPoint):

obj(clockref=None) (see above for clockref)

Conclusion

Integrated in the regular () syntax, data feeds from different timeframes can be mixed in indicators, always taking into account that cerebro needs to be instantiated or created with runonce=False.

Script Code and Usage

Available as sample in the sources of backtrader. Usage:

$ ./mixing-timeframes.py --help
usage: mixing-timeframes.py [-h] [--data DATA] [--multi] [--plot]

Sample for pivot point and cross plotting

optional arguments:
  -h, --help   show this help message and exit
  --data DATA  Data to be read in (default: ../../datas/2005-2006-day-001.txt)
  --multi      Couple all lines of the indicator (default: False)
  --plot       Plot the result (default: False)

The code:

from __future__ import (absolute_import, division, print_function,
                        unicode_literals)

import argparse

import backtrader as bt
import backtrader.feeds as btfeeds
import backtrader.indicators as btind
import backtrader.utils.flushfile


class St(bt.Strategy):
    params = dict(multi=True)

    def __init__(self):
        self.pp = pp = btind.PivotPoint(self.data1)
        pp.plotinfo.plot = False  # deactivate plotting

        if self.p.multi:
            pp1 = pp()  # couple the entire indicators
            self.sellsignal = self.data0.close < pp1.s1
        else:
            self.sellsignal = self.data0.close < pp.s1()

    def next(self):
        txt = ','.join(
            ['%04d' % len(self),
             '%04d' % len(self.data0),
             '%04d' % len(self.data1),
             self.data.datetime.date(0).isoformat(),
             '%.2f' % self.data0.close[0],
             '%.2f' % self.pp.s1[0],
             '%.2f' % self.sellsignal[0]])

        print(txt)


def runstrat():
    args = parse_args()

    cerebro = bt.Cerebro()
    data = btfeeds.BacktraderCSVData(dataname=args.data)
    cerebro.adddata(data)
    cerebro.resampledata(data, timeframe=bt.TimeFrame.Months)

    cerebro.addstrategy(St, multi=args.multi)

    cerebro.run(stdstats=False, runonce=False)
    if args.plot:
        cerebro.plot(style='bar')


def parse_args():
    parser = argparse.ArgumentParser(
        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
        description='Sample for pivot point and cross plotting')

    parser.add_argument('--data', required=False,
                        default='../../datas/2005-2006-day-001.txt',
                        help='Data to be read in')

    parser.add_argument('--multi', required=False, action='store_true',
                        help='Couple all lines of the indicator')

    parser.add_argument('--plot', required=False, action='store_true',
                        help=('Plot the result'))

    return parser.parse_args()


if __name__ == '__main__':
    runstrat()