Hey there, I'm a newbie to pairs and stat arb. I'm a trader at a prop firm and we have a group that runs a pair trading approach and does quite well. Some of their guidance and insights have led me to investigate this type of approach. Just made my way through Ganapathy Vidyamurthy's book and it was a good read. Also read Ernie Chan's Quantitative methods book and found it useful. Their holding periods span weeks and months so it is pretty capital intensive and not what I am looking for at the moment. I am looking at an intraday approach. What I am working on, is a form of sector arbitrage. Taking industries within sector ETF's, finding the most optimal stocks, finding optimal hedge ratio's to trade against the ETF. This is assuming that co-integration exists between stocks and their component etf's of course. In theory, by finding the optimal weightings of component stocks we should have created a synthetic asset to trade against the ETF. So when a sector or group of stocks get skewed, I can trade my synthetic asset against the broader ETF. I am looking at using some different factors to create the asset, but intuition tells me simplicity is key here. So this is what I am working on at the moment, just wondering if anyone has done this type of work on the intraday time frame?
The testing is going to be done regardless lol. I haven't seen anything out there about this type of approach so was just curious if anyone had gone this route before
You can have a look at the Pairs Trading Strategy Journal. http://www.elitetrader.com/vb/showthread.php?s=&threadid=134253
most of ET trying to make a living scalping SP\CL fututers..your stuff is wayy too complicated for them..
For what I know, this is what Chan is doing in his book and the timeframe does not matter. Dont get excited too soon. You are walking on a paved route, not off the beaten path.
Yea I've looked though most of the threads on the internet pertaining to this and didn't get a whole lot. Oh well, next week will be pretty hectic. Will start some testing and maybe report back if I find anything.
You should find that correlation is a time varying value. And therefore you cointegration value should also change over time. If you find a process that is well and constantly cerellated with the same cointegration vector (IVV/SPY, ) the bad news is that it is already traded by other guys that have more ressources than you. Then the execution time or the comish will make a difference.
hmmm yea your probably right, I'm sure this has been done before. In fact, thats why I posted here. Also, I have Chan's book and access to his website. Looks to me like he is just finding the most cointegrated pairs and then just regressing them to find the optimal weightings. Nothing about specific baskets chosen because of some factor.