Wednesday, September 30, 2009

When Good Pairs Go Bad

Keeping with my oft voiced analogy of trading and gold mining, you sometimes you find nuggets in unexpected places (and you also find Fool's Gold). Case in point . . today's post.
In my continuing exploration for high correlation Qs pairs I ran the Qs against the major commodity ETFs.
The results, unfortunately, are not encouraging for supplementing my current Qs pairs arsenal.
However, a quick look at the upper matrix will show that the majority of the 90% linearity correlations are linked to a single ETF. . . JJT. Now this looks promising. . . .

If we optimize JJT pairs opportunities in the larger pairs universe there appear to be some very high probability correlated and non-correlated linkages. These various pairs also offer a variety of lookback (N) periods and minimal drawdowns. So far, we're looking good.
But before going any further I've got to warn you that more than a few of these ETFs have pathetic volume, wide spreads and little or no open interest in the options. the REALLY bad news. These JJT pairs are utterly untradeable.
JJT's volume is low to none and the bid/ask spreads will take your breath away.
Forget about the JJT pairs for a moment . . . if you could just sell JJT at the ask and buy at the bid (or anything even close) you'd be very rich very quickly and be able to hang out with the glitterati in Monaco.
Likelihood of such a scenario? Sorry . . . Zip.
The lesson to be learned here is that all pairs have to be carefully deconstructed and evaluated with a variety of metrics to provide the greatest probability of real trading success.
Simply relying on high readings of linearity correlation and P&L is not enough.
On the other hand, there do appear to be a few robust trading pair candidates culled from the commodities matrix that offer some reasonable prospects.
We'll check out one of those tomorrow as the Qs basket devlopment continues.


Kristianson said...


How do you allocate your cash to
your trades? Do you do a dollar neutral or using the beta approach? You have to be very careful with some pairs if you are testing the leveraged versions as this could could look extremely good on back testing in terms of total return, and spreads and execution may be tight as well, but it may be too much volatility to stomach.

I just played with the FAS/GE spread the other day and had a quite rough time shorting FAS (as suggested by the system). I had a dollar neutral exposure, but did not realize (my stupidity of not looking at the beta that showed +3.5, i.e FAS is moving 3.5 times as much as GE). So being short FAS when it close up 10% for the day can be hair raising. Now this managed to open up lower the next day and sell off so I managed with some skill (I must pat myself on my back) and get out of FAS short early on and keep GE long to move up a bit after initial sell off and reduce my loss. In any case, I think it is important to also understand what you are trading and not be blinded by the total returns. This may have nothing to do with your excellent analysis in this case, however I just wanted to write about my personal experience.


bzbtrader said...

Excellent observations (and recovery) and greatly appreciated.
I have to caution readers that these pairs studies are exploratory in nature and not definitive or a recommendation to trade specific setups.
I have only been running these pair studies for a month with the stated goal of developing a set of pairs baskets to corral risk and miminize drawdown (my most important trading considerations).
We've touched on the issue of beta previously (briefly) but your experience and willingness to describe in detail how an unbalanced beta trade played out is a valuable lesson to be seriously heeded.

bzbtrader said...

I also not that the GE/FAS pair P&L line crossed below the R2 back on 8/1, suggesting that caution was warranted in going forward with any new trades in this pair.
One of the projects I'm exploring with Jeff is the evolution of the Pairs Trader into a robust real time platform using 5, 10 or 60 (or 65) minute bars. Still a ways
off, but such a fractal pairs model would introduce a whole new set of trading dynamics and risk management tools.

QuantWizards said...

A fractal model, as you say, would be possible to construct using, say IB feed. If there is enough interest in this that would be an interesting road to pursue. Again, be careful with those levered ETFs/ too much beta disparity!