Futures and Spot Compensation

Release 1.9.32.116 adds support for an interesting use case presented in the Community

Start a trade with a future, which includes physical delivery
Have an indicator tell you something
If needed be, close the position by operating on the spot price, effectively canceling the physical delivery, be it for receiving the goods or for having to deliver them (and hopefully making a profit)

The future expires on the same day the operation on the spot price takes place

That means:

The platform is fed with data points from two different assets
The platform has to somehow understand the assets are related and that operations on the spot price will close positions open on the future

In reality, the future is not closed, only the physical delivery is compensated

Using that compensation concept, backtrader adds a way to let the user communicate to the platform that things on one data feed will have compensating effects on another. The usage pattern

import backtrader as bt

cerebro = bt.Cerebro()

data0 = bt.feeds.MyFavouriteDataFeed(dataname='futurename')
cerebro.adddata(data0)

data1 = bt.feeds.MyFavouriteDataFeed(dataname='spotname')
data1.compensate(data0)  # let the system know ops on data1 affect data0
cerebro.adddata(data1)

...

cerebro.run()

Putting it all together

An example is always worth a thousand posts, so let’s put all the pieces together for it.

Use one of the standard sample feeds from the backtrader sources. This will be the future

Simulate a similar but distinct price, by reusing the same feed and adding a filter which will randomly move the price some points above/below, to create a spread. As simple as:

# The filter which changes the close price
def close_changer(data, *args, **kwargs):
    data.close[0] += 50.0 * random.randint(-1, 1)
    return False  # length of stream is unchanged

Plotting on the same axis will mix the default included BuyObserver markers and therefore the standard observers will be disabled and manually readded to plot with different per-data markers
Positions will be entered randomly and exited 10 days later

This doesn’t match future expiration periods, but this is just putting the functionality in place and not checking a trading calendar

!!! note

  A simulation including execution on the spot price on the day of
  future expiration would require activating `cheat-on-close` to
  make sure the orders are executed when the future expires. This is
  not needed in this sample, because the expiration is being chosen
  at random.

Notice that the strategy

buy operations are executed on data0
sell operations are executed on data1

class St(bt.Strategy):
    def __init__(self):
        bt.obs.BuySell(self.data0, barplot=True)  # done here for
        BuySellArrows(self.data1, barplot=True)  # different markers per data

    def next(self):
        if not self.position:
            if random.randint(0, 1):
                self.buy(data=self.data0)
                self.entered = len(self)

        else:  # in the market
            if (len(self) - self.entered) >= 10:
                self.sell(data=self.data1)

The execution:

$ ./future-spot.py --no-comp

With this graphical output.

And it works:

buy operations are signaled with a green triangle pointing upwards and the legend tells us they belong to data0 as expected
sell operations are signaled with an arrow pointing downwards and the legend tells us they belong to data1 as expected
Trades are being closed, even if they are being open with data0 and being closed with data1, achieving the desired effect (which in real life is avoiding the physical delivery of the goods acquired by means of the future)

One could only imagine what would happen if the same logic is applied without the compensation taking place. Let’s do it:

$ ./future-spot.py --no-comp

And the output

It should be quite obvious that this fails miserably:

The logic expects positions on data0 to be closed by the operations on data1 and to only open positions on data0 when not in the market
But compensation has been deactivated and the intial operation on data0 (green triangle) is never closed, so no other operation can never be initiated and short positions on data1 start accumulating.

Sample Usage

$ ./future-spot.py --help
usage: future-spot.py [-h] [--no-comp]

Compensation example

optional arguments:
  -h, --help  show this help message and exit
  --no-comp

Sample Code

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

import argparse
import random
import backtrader as bt


# The filter which changes the close price
def close_changer(data, *args, **kwargs):
    data.close[0] += 50.0 * random.randint(-1, 1)
    return False  # length of stream is unchanged


# override the standard markers
class BuySellArrows(bt.observers.BuySell):
    plotlines = dict(buy=dict(marker='$\u21E7$', markersize=12.0),
                     sell=dict(marker='$\u21E9$', markersize=12.0))


class St(bt.Strategy):
    def __init__(self):
        bt.obs.BuySell(self.data0, barplot=True)  # done here for
        BuySellArrows(self.data1, barplot=True)  # different markers per data

    def next(self):
        if not self.position:
            if random.randint(0, 1):
                self.buy(data=self.data0)
                self.entered = len(self)

        else:  # in the market
            if (len(self) - self.entered) >= 10:
                self.sell(data=self.data1)


def runstrat(args=None):
    args = parse_args(args)
    cerebro = bt.Cerebro()

    dataname = '../../datas/2006-day-001.txt'  # data feed

    data0 = bt.feeds.BacktraderCSVData(dataname=dataname, name='data0')
    cerebro.adddata(data0)

    data1 = bt.feeds.BacktraderCSVData(dataname=dataname, name='data1')
    data1.addfilter(close_changer)
    if not args.no_comp:
        data1.compensate(data0)
    data1.plotinfo.plotmaster = data0
    cerebro.adddata(data1)

    cerebro.addstrategy(St)  # sample strategy

    cerebro.addobserver(bt.obs.Broker)  # removed below with stdstats=False
    cerebro.addobserver(bt.obs.Trades)  # removed below with stdstats=False

    cerebro.broker.set_coc(True)
    cerebro.run(stdstats=False)  # execute
    cerebro.plot(volume=False)  # and plot


def parse_args(pargs=None):
    parser = argparse.ArgumentParser(
        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
        description=('Compensation example'))

    parser.add_argument('--no-comp', required=False, action='store_true')
    return parser.parse_args(pargs)


if __name__ == '__main__':
    runstrat()