(I'm quite new to quant finance so I'm not sure if this is an eligible question.)

I've decided I want to backtest pairs trading on the Nordic stockmarket. So I would guess there exists different methods of selecting the pairs (not just guess, I've heard and read some about them). For example the 'most written about' selection based on cointegration which seems most straight forward (which was used by Gatev et al in their paper called Pairs Trading: Performance of a Relative-Value Arbitrage Rule), but also some extended versions of this based on forecasting (tried reading about it but it was a bit to advanced for me). I've also heard about pairs trading calender time (or something like that), which I tried to find some articles about without succeeding (my guess would be that the selection and trades are done 'in calender time', meaning they are based on opening/closing prices rather than intraday. Please correct me if I'm wrong).

So my question is basically what different methods of pairs selection exists? I wouldn't want a full description of them, just some keywords to search (it's hard to find something when you don't really know what you're looking for.. sort of).

Maybe if someone has come across some recent finding in this area or if someone who is actually working with this could give me some ideas on what is 'hot' (meaning what is used in the industry today) it would be great. I tried to contact a few hedgefunds, however they didn't really have time for students (maybe it was naive of me think they would).

Thank you in advance for your help!

  • 3
    $\begingroup$ I tried to contact a few hedgefunds, however they didn't really have time for students -- that's about the most honest assessment I've ever read on here from a non-quant. $\endgroup$ Commented Feb 11, 2013 at 20:20
  • $\begingroup$ What do you mean by that? =) $\endgroup$ Commented Feb 11, 2013 at 20:28
  • $\begingroup$ this is a related question: quant.stackexchange.com/questions/2753/… $\endgroup$
    – RoBiK
    Commented Feb 11, 2013 at 22:18
  • $\begingroup$ @RoBiK That related question wasn't answered either, which is what I suspect will happen here. $\endgroup$ Commented Feb 12, 2013 at 0:00

1 Answer 1


The basic idea of pair trading is to find two symbols (I'll use that to mean stocks, futures, anything you can trade) that historically have correlated price movements. Then if just one of them increases in price you short that symbol, and buy its pair, on the assumption that they will soon go back in sync.

However when you think there is a genuine reason for just one to move then you skip the trade; this is why trading stocks is nice, as most of the time when one stock moves but its pair does not, it is due to some news item. You can then quickly make a judgement how stock-specific it is (E.g. BP and the Deepwater accident affected BP more than SHEL).

You can find candidates with brute force: run correlations between all pairs of symbols in your database. But once you've found the correlation the next step is to find a story. An explanation for why their prices move in step. Without that you won't know when to ignore a move. If you cannot find a story either you have discovered data mining noise, or you haven't tried hard enough.

That is why usually you find pairs by starting with the story, and then looking for a correlation:

  • Two stocks in the same industry. The Wikipedia page has a few stock examples.
  • Two industries or stock market indices that have a lot in common. E.g. FTSE100 and FTSE250 correlate fairly well, because both are driven by the same general economic news.
  • You might also consider pair-trading one stock with its industry ETF.
  • You might even consider pairing an exchange rate with a stock or an index. E.g. the Nikkei 225 and USD/JPY. The story there being that as the yen weakens the big manufacturers become more competitive overseas. However raw material costs, in yen terms, also go up, so there is a counter-story there too.

IIRC, Ernie Chan's Quantitative Trading book had a chapter on pair trading.

  • $\begingroup$ Thank you for your answer. I like the "story-idea", sounds reasonable for trading in real time. However it feels like it's not easily applicable for computerbased technology, especially not on backtesting. $\endgroup$ Commented Feb 21, 2013 at 16:24
  • $\begingroup$ @GoodGuyMike Yes, news analysis can be included in automated trading, but that might not count as "easy". However your question was about pair discovery: my point was only look at symbol pairs that have something in common; anything else you discover will most likely be data mining noise. $\endgroup$ Commented Feb 22, 2013 at 0:32
  • $\begingroup$ Yeah, it's my understanding that news analysis (including 'news' from social media) in trading has been of much interest lately and I think it's quite interesting! However as you say it's not easily done and it's definitely out of my scope. But ok, it should be some logical explanation to why they are moving together, so trading two stocks from the same market would be the intuitive one I guess and also the easiest to incorporate in the trading algorithm. But you say anything else is most likely to be data mining noise. In the case of cointegration they talk about 'spuriously cointegrated' $\endgroup$ Commented Feb 22, 2013 at 12:51
  • $\begingroup$ so this would be considered data mining noise I guess? (Not sure if I should ask this question here, but lets try) I've been trying to understand what they mean by spuriously cointegrated, but I'm still a bit confused. Do they mean spuriously as in "there's no logical explanation why they should be cointegrated, but the results of analyzing the sample series still says that they are". Or do they mean that the test (now talking about the Engle-Granger test in particular) fails, so it says they are cointegrated but they're in reality not. I hope I'm not to confusing... $\endgroup$ Commented Feb 22, 2013 at 12:56
  • 1
    $\begingroup$ @cf16 Thanks; not being a mathematician I suspect I am using the word correlation sloppily: I mean finding a function of the price movements of two symbols that is mean-reverting. $\endgroup$ Commented Mar 12, 2013 at 1:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.