mmmm..how so?..you long AMAT,short NVLS..news come out that AMAT is going to be bought @ +50% of current value. where is limited upside? looks like same s**t to me. if you short AMAT, long NVLS-the you lose 50%, if you long-you gain 50%. i'm wrong or missing something?
Well, if they are set up to fade a one or two sigma move - then yes, they would get crushed by any such event. So, he does make a valid point.
Our algorithm is mean reverting. It gets into a position when the spread deviates from the fair value by a certain amount. So let's say we're currently flat BAC-C, and news comes out that S&P will be downgrading BAC. The stock takes a huge hit while C remains stable. In that scenario, we will get long BAC and short C. As a trader/researcher, one of my roles is to analyze the impact of such events and figure out how best to reduce risk on such positions.
I find it interesting that you characterize this spread relationship as having a "fair value", a term generally reserved for pure arb where it holds true for equivalent trading instruments where mean reversion works. Spread trades between two stocks like this are stat arb, where mean reversion is just as speculative as taking a position in either outright.
Traditionally it is done for CFD ( convergence for difference ); which is, of course, essentially mean reversion to some sort of 'fair value'. These days, however, the simple pairs are usually much better at divergence - and the mean reversion will typically require multi-legged spreads and even baskets. It is truly all in the construction choices.
One of the challenges that I'm working on is the application of a stop-loss threshold to intraday equity pairs. For the most part, I'm opposed to using a strict dollar loss amount as a stop-loss. This is not entirely objective or quantitative, but there are varying types of divergences, and each should be treated somewhat differently. So for example, if stock A makes a very sharp rally in light of rumors about it being a target of an acquisition, and the algorithm goes short A and long B, you will be losing money at the onset. I don't think it makes sense though in such a situation to flatten the pair right away. On the other hand, if it's a slow trading day with low volatility and you see pairs where A is up 5% and B is down 2%, and the spread has been gradually drifting away from fair value, then it makes sense to at least flatten portions of it. I'm still working on this to gain more clarity, but in the current environment where the market is dominated by macro concerns over Europe, and stocks are at historically high correlation levels, I don't think it's prudent to just let the algos "run wild" without any manual intervention and risk management.
For statistical arbitrage trading, one may also want to look into using a platform that has dollar-based stops and take-profits; one that is able to close out both positions simultaneously once either of those conditions are met. Otherwise, there would be no other way (other than manually) to place automatic stops on this kind of a trade. If you do not know of a platform that can do this, hereâs one add-on for MT4 that can do it: http://www.globalfutures.com/platforms/expert-pad.asp
Interested in hearing if anyone follows the COT report, and if they incorporate it into their spread trading strategies. Have never delved too deeply into the COT reports previously but have started looking at them a little more and starting to think whether there isn't a way to apply them to spreads......