Pairs trading research and execution on backtrader - any tips?

tommark

Hi, I am new to backtrader. Wonder if there is any existing frameworks on pair trading which I can learn and build from, in python / backtrader space.

https://www.bloomberg.com/professional/blog/pair-trading-its-complicated-part-i/
https://www.bloomberg.com/professional/blog/pair-trading-its-complicated-part-iii-this-is-not-your-standard-algorithm/

Paska Houso

There is a Pair Trading contributed sample for backtrader in the sources under samples/contrib/pair-trading

But I recall that the OLS technique it uses may be broken due to changes in the OLS API. This thread tackled it a few days ago:

https://community.backtrader.com/topic/734/ols_slope_interceptn

Apparently add_constant was missing after some API got updated.

tommark

Can you please elaborate on the above please. Or anyone else with the knowledge.

When I run 'pair-trading.py' script, yes I get the error 'NameError: name 'prepend_constant' is not defined', as it brings up the 'ols.py' script.

Diving into the problem, it stems from the class below. Pardon the code rendering, what should I do to render the code display properly?

Is there any copy and paste solution I can use to get it working? Think the issue lies in 'sm.add_constant(p1, prepend=prepend_constant)'

I have referred to https://stackoverflow.com/questions/2820453/display-html-code-in-html and https://community.backtrader.com/topic/734/ols_slope_interceptn/7 but to no avail.

class OLS_Slope_InterceptN(PeriodN):
'''
Calculates a linear regression using statsmodel.OLS (Ordinary least
squares) of data1 on data0
Uses pandas and statsmodels
Use prepend_constant to influence the paramter prepend of
sm.add_constant
'''
_mindatas = 2 # ensure at least 2 data feeds are passed

packages = (
    ('pandas', 'pd'),
    ('statsmodels.api', 'sm'),
)
lines = ('slope', 'intercept',)
params = (
    ('period', 10),
    ('prepend_constant', True),
)

def next(self):
    p0 = pd.Series(self.data0.get(size=self.p.period))
    p1 = pd.Series(self.data1.get(size=self.p.period))
    p1 = sm.add_constant(p1, prepend=prepend_constant)
    slope, intercept = sm.OLS(p0, p1).fit().params

    self.lines.slope[0] = slope
    self.lines.intercept[0] = intercept

tommark

Took me 2 full days to realize that I only needed to restart my kernel to reset the namespace after making the edits

jxm262

hi @tommark , I'm getting the same error as you mentioned preprend_constant is not defined.

Can you please let me know what you meant by restarting the kernel in your case? I was assuming you meant restart the actual python interpreter (or your machine)? I'm not actually running on a Jupyter notebook or anything, so was a bit confused on your comment.

Anyway, after looking at @jirathh 's last comment here - https://community.backtrader.com/topic/734/ols_slope_interceptn/12, I took a wild guess and changed that line offending line to this -

p1 = sm.add_constant(p1, prepend=True)

Seems to work, but I haven't dug into any details on the actual root cause yet.

tommark

change
sm.add_constant(p1, prepend=prepend_constant)
to
sm.add_constant(p1, prepend=self.params.prepend_constant)

the 'self.params' should be intact

jxm262

@tommark said in Pairs trading research and execution on backtrader - any tips?:

self.params

ah perfect, that works.

thanks @tommark