Definition of spread in pair trading

[garchbrooks]

I'm curious what all the spins are on this definition. I was reading a paper on optimizing some stochastic control problem, and the paper didn't bother to specify where the anchor point in the control problem was, whereas that book by Mark Conway decided to use returns from a daily close, etc.

Would some of you care to clarify what "spread" is to you mathematically, and what you think the best ways to select anchor points are?

 

[bone]

The spread as it applies to relative value trading to me at least is the valuation differential between two or more financial instruments.

The 'anchoring point' as you describe can mean different things to different people. Some traders prefer to make constant, 'on-the-fly' currency and volatility adjustments to each spread leg instrument in order to maintain as constant a 'zeroing effect' as possible. Their preference is to always maintain a +10 to zero to -10 scale for the purposes of CFD (mean reversion). The majority of these traders are in the equity shares space. If certain spread combinations can be efficiently zeroed out on the fly they make excellent stat arb candidates.

Many spread trade combinations shouldn't be 'anchored' in the sense that currency and volatility adjustments are not frequently required - especially those combinations with high correlations and suitable z-scores between +1 and +3. Examples might include the Five Year versus Ten Year Treasury Yield Spread adjusted to DV01.

Many floor traders and floor brokers 'anchored' spread trades by
quoting them in terms of net change from the previous day's settlement prices.

Other trades, like the Heating Oil Crack Spread, are quoted by the simple price differential and the leg weightings are never changed.

 

[garchbrooks]

Thank you very much for your informative reply, bone.

 

[black diamond]

Hi Garch,

The spread is whatever your model says is mean reverting and stationary that you want to trade. I think the "best" spreads are price ratios because if you are trading a price ratio, you can put equal dollars into each side and it is self-rebalancing and roughly market neutral. You can also trade a price difference model without rebalancing, but you have to trade a fixed number of shares on each side instead of a fixed dollar amount, and I also don't like that approach as much because I don't think that kind of relationship is as logical so I would worry more about the model working in past data by coincidence. You can do other stuff, but then whatever you find has be good enough to either overcome rebalancing costs/hassle or be worth living with more risk.

A simple example of a price ratio model would be Price of Stock A/Price of Stock B = X, open a trade when the ratio is outside X +/- 2 standard deviations and close when it is X again. You get X from historical data, and you have to crank through a lot of pairs to find some that "work".

If you are new at this, one warning is you have to adjust the prices somehow so something like a dividend or split doesn't trigger a trade.