Thanks for posting the Excel VBA corr. matrix code. Along sililar ideas.... It wouldnt surprise me that a time series of daily or weekly correlation between two products can provide trade signals for mean reversion.
Spread trading is for those who do not understand which direction a market may be headed. Spread trading while having a high probability of winning, generally wins small. It is to be avoided.
This is a common point of confusion with regard to two different statistical concepts, correlation vs cointegration. Mean reversion requires only cointegration. You can have two time series that are uncorrelated and exhibit high mean reversion.
And to add another point to my last post, high correlation often works against mean reversion when the price series are moving together in synchronicity and not diverging/converging from each other. High correlation in the markets has reduced the viability of mean reversion trading (aka, pairs trading, statistical arbitrage) over the past decade. Most spread traders have transitioned their trading from mean reversion to trend (relative strength) because trend aligns well with correlation and they find that correlated markets have more persistent trends. In my own trading, I'm still an MR guy. But I'm in the minority.
I used to trade them when I traded prop. I have since moved on and now trade more volatile spreads as these offer better risk/reward in my opinion. My main reason for moving on to other products / strategies is the combination of ZIRP and the fact that markets in general have become more efficient and, as a result, these trades are increasingly less interesting as stand alone strategies. That said, they are still very important in a STIR (or other spreads) trader's arsenal as a way to for example defend positions or change from one position into another.
thank you for clarifying the differences between a condor and a box. There is very little information about this in the public domain.
I am maths challenged so i am not inclined to run coint. tests. I do it backwards. If i run a simple 2sd xover mean reversion on 10,0000 equity spreads over many years, some generalities appear to be true to me. 1)High degree of cointegration. 2)Reversion of correlation (it was temporarily broken providing the trade signal) to a mean. From that i was inferring that correlation itself may be a trading signal but probably in support of other signals. Possibly 2sd price excursions with 2sd correlation excursioms is better than just The 2sd price excursions. I cannot back this up with test results however and have no need or wish to prove it. An anaology to me is that cointegration is a pre requisite on a larger timeframe and correlation,a pre requisite as well, on a shorter timeframe, where one finds opportunity.
As an example to help to illustrate my point, the attached chart shows two time series. They are actually two sine waves and one is displaced (shifted forward) from the other. But you can imagine they are two price series for the sake of discussion. The correlation between these two series is calculated at ZERO. By definition, this means they are perfectly UNcorrelated and move in the same direction half the time and the opposite direction the other half of the time. Also note they are perfectly cointegrated as they always move back to an equilibrium (mean) state. If these were real market prices, you could trade them as a mean reverting pair and do very well. You may also have a perfectly correlated pair (correlation = ONE) that is mean reverting. You can simply imagine two price series moving in parallel but with different magnitudes of movement so they diverge and converge. The takeaway here is that mean reversion can occur regardless of the degree of correlation. So then why do traders focus on correlation? Because it is intuitively easy to understand, easy to calculate, and is readily available in market analysis tools. The appeal of correlation also favors trend trading a spread (instead of mean reversion) because it's easier to find candidates that are fundamentally related and statistically correlated. By comparison, cointegration is far more complex. These differences are also reflected in the "edge" for each style of trading. For trend trading a spread, the candidates are easy to find using correlation but the challenge (edge) is in your timing technique for entry/exit. It's this timing technique that is usually regarded as proprietary by spreaders who trade the trend. By comparison, when trading mean reversion the timing technique is easy using any common measure of variance to identify divergence. But the identification of candidate pairs is the challenge (edge) since there aren't many markets that mean revert on a sustainable basis due to correlation effects and arbing. Consequently, mean reversion traders usually regard their market selection as proprietary.