Because correlation is not the appropriate measure; look up co-integration. Ignore corr and focus on back-test results. [Although most will likely ignore me instead!]
Hi, I am building my pair-trading system from the scratch. I am now coding correlation between 2 stock prices. My doubt now is, should i use Pearson Correlation or Spearman Rank Correlation ? Anyone knows what is the algorithm used on PairTradeFinder? Thanks
Knocks, Ok, I understand. However, I'm having trouble grasping how to identify pairs that are co-integrated. Co-related stocks can be found by industry and those I can grab in bulk and back test. But how do I identify co-integrated pairs? I found interesting comments in another forum and include two below: "The difficulty is finding these portfolios [of cointegrated pairs] that will maintain this stationary relationship for a long time out of sample. Unfortunatelly it is easy to find spurius cointegrations in the sample that won´t stand for long when you decide to go trading." and: "Cointegration is nice because it can capture the causal relationship between financial quantities (two stocks, one of which is traded actively and leads the other, for example) or complex dynamics (two stocks in the same business moving together, but then one takes market share from the other so they move apart). Correlation is popular in practice because 'it is there' i.e. it's the easiest way to model relationships, even though that model is not very stable." Co-integration vs co-relation: http://www.tradingmarkets.com/.site...ntegration-is-not-the-same-as-correlation.cfm Walt B
Co-integration isn't a difficult calculation per say but trying to do it an excel over a couple 100 stocks can be difficult. The comment is correct, sometimes you'll find IBM co-integrated with HNZ. Is that spurious or not? Depends on the trader I guess. Same with stationary systems, pairs will come and go. KO/PEP may have been a great pair at one point but not anymore. This is where fundi's and common sense come in to play. My suggestion is to get as much data as possible (10+ years if available) and concentrate on profitability first. Then add common sense to find best pairs. If you really want to perform co-integration without weeks of effort, try to get a copy of MatLab.
I do a lot of pairs trading with my own system. The basic concept is to try to capture some of the spread b/t the two. Where it really gets interesting is where you have a group that correlates so that you can swap substitutes for winners. IOW, if both legs rise, take the profit on the long leg and add a new long leg (different issue). Conversely, if the pair moves down, cover the short leg and replace it with another short. As a result, you're always booking profits which the market can't take away - as opposed to cutting a losing leg and holding the winning leg, which a gap will clobber. I shift my bias intraday (more long than short in an up move) but one it's a rule of thumb is to always get close to balancing out for overnight since one never knows what the morning will bring. Happy trading!
Interesting. I've had that same thought because I do have pairs set up with multiple instances between a single stock. I may consider this approach further.
closed very nice trade on thursday, significant profits on both sides Sold BKI @ 2.40 (went long 2.18) Covered IP @ 8.03 (went short 9.20)
John, I have read that stocks liquidity, and institutional ownership has some impact on pairtrading profit. The reason is obvious, market inefficiencies tend to be corrected much faster when the big players are involved. Do you have any idea if these considerations are correct? Noticed this impact on your trades? If you don't mind, i can analyze your trades historical data, and try to correlate the profits with stocks average volume, and institutional ownership. Best Regards and good trades!
Hey guys, I haven't been able to read the whole thread yet, so sorry if these questions have already been brought up but I was hoping if anyone in here trades/has traded pairs intraday they could answers the below questions... - How much movement are you looking to catch in these intraday pair trades? 1%? Less? More? Are they correleated pair trades when you are going intraday or not? - If you are only trying to catch one or two percentage points intraday, how have you found the B/A spreads effect you. For instance, do you only trade extremely liquid stocks or stocks above say $5.00(due to the automatic loss that could occur even if the stock B/A was a penny or two wide.). - Also how do commisions/haircuts/ and interest rates play a role since I am guessing you would have to trade large volume and use margin to make money pair trading intraday. Thanks for any help you can provide.