Quantopian's community platform is shutting down. Please read this post for more information and download your code.
Back to Community
Any comment on this backtest result for JPM/GS Pairs Trading?

Hi

I am trying to see performance of JPM/GS pairs trading, not to be confused, I have changed sid accordingly & my ticker1 is JPM & ticker2 is GS.

Any feedback for improvement?

3 responses

I suggest you to record the current value, mean (rolling window), +/- deviation (rolling window) of spread, rather than only zscore. Because zscore always looks very good, but what you really trade is spread. My experience is single pair can hardly achieve very good performance because there are lots of news on each stock in the pair. For example, JPM acquired Bear Stearns in 2008 and this event will have a large impact on stock price of JPM but not GS. It would be better if you diversify on lots of pairs to reduce risks on corporate news.

Indeed; my biggest trouble with finding any pair trading scheme that works is most of the mean reversion of the z-score is caused by the mean moving, not the spread reverting. I asked about this in https://www.quantopian.com/posts/model-slash-beta-slippage-in-dynamic-spread-mean-reversion-models but nobody had any thoughts. :)

Also, one important indicator for goodness of mean reversion strategy is lambda, or half-life of mean reversion. You can access more details about it in Ernest Chan's book Algorithmic Trading. Large absolute value of lambda, or small half-life means big profit and stable performance.