Pairs Trading Strategy Model

Do you trade in the direction of convergence or divergence of the pairs?

 

I think it just depends on your time frame you trade in. Let's say you do your research and conclude on a 5 year chart of a pair that the two stocks difference is "out of whack." Say...at an extreme relative to their past. Yet the fundamentals suggest a convergence back to parity over time, and the chart suggests you can enter a trade at this logical resistance...then yeah...I fade the trend (the trend which has taken it to the "out of whack point"). Am I losing you... :eek: But there's nothing saying you can't just wait for the trend to reverse first before guessing the trend is over -- right?

You form your fundamental and technical bias...then you wait for the trend to go in this bias and you strap it on! There's lots of ways to do this as you can see.

So, if you come to this same conclusion on your monthly or 60 minute charts the same game can be played. Fade or trend? It depends. The 60 min chart says the trend is starting, but the 2 minute chart says this is reeeeealy extreme here to fade IN THE DIRECTION of the 60 min chart...then I play the divergence.

I think there's a big problem with defining your bias time frame (say daily) but playing 2 min charts. They don't mesh. You should have 5 x the time frame. Find bias in a 10 min, yet trade the signals in a 2. Find bias in a weekly, yet trade the signals in the daily, etc.

Rambling...sorry.

In general, you'll find the normal pair trader FADES plays more than trades trends. We expect correlated stocks to come back to parity. But there is NO rule that says you can't or shouldn't play the divergence when your analysis suggests. As a matter of fact, there's tremendous money to be made doing just that...but you have to be open to the idea. Most pair traders aren't.

Hope that helps.

 

Looked at the www.pairsfinder.com site and its under construction. Does anyone know how long its been offline?
My free trial to www.pairfinder.com is almost up and I'm going to subscribe. I've used the "Pair Finder" section of their site to determine return correlation relationships between several stocks in the same industry. Return correlation seems to be a more accurate measure because it doesn't treat the moves of a $5 stock and a $20 stock the same, like price correlation.
The site also provides a market neutral hedge ratio that has helped me even out the risk of my shorts and longs.

 

I believe the earlier post said that "pairsfinder" and "pairfinder" are registered by the same person.

 

If there are 2 functions (in our case 2 stocks) then the best way to find out if the functions related to each other is to find correlation between them (that’s mathematics). You of course may try to find correlation between secondary functions and in some cases the result will be close.
For example, if there are 2 stocks (A and B) and during 3 days period they move like this:
(A): 3,6,9 and (B): 2,4,6 then both correlations (price and return) would give you 100%.
However you don’t trade secondary functions (in your case you trying to find correlations between 2 differentials divided by functions itself). You trade the stocks – primary functions. Even if the secondary functions will be highly correlated, stocks themselves might behave quite different.
Here are some examples:
Let say stock A goes like this: 10,1,2; And stock B goes like this: 10,9,18
I would not say these are highly correlated stocks. Return Correlation however would give you 100%; but Price correlation would give negative (!) 10%
Some say that the most important that Difference between stocks should be in some range. If it goes on edge, then, based on historical data, we might go into trade. Yes, we might, and in some cases we could make money. However, good thing about correlated stocks that you are protected (kinda) against market moves. If you trade correlated stocks you are market neutral (and then - range may go into account). Let’s take another example:
(A): 10,11,10,9,10…; (B): 10,9,10,11,10…
Difference would be between –2 and 2. However, these stocks correlated with negative 100%. Based on historic data we may trade this pair. But, in that case we as well might trade these stocks just by themselves. They don’t need a pair.
Can we make money here? Yes. Are we going to be protected against market moves – not in this “pair”.

 

Please help me with my definitions of Price and Return Correlations. I understood Price Correlation to be a $ for $ comparison of two individual stocks: (to use your explaination method) If stock (a) 3,6,9 and stock (b) 2,4,6 then stock (a) moves 3 then 3 and stock (b) has moves of 2 then 2. So in my understanding of Price Correlation, (b)/(a) would be 67% correlated. Your defintion seems to be a direct ratio comparison of (a) and (b) over a common time period: (b1)/(a1), (b2)/(a2), (b3)/(a3) then since the ratio result is constant throughout the time period, they are 100% Price Correlated.

Now for Return Correlation, in my understanding, is the change of (a) divided by the price of (a) compared to the change of (b) divided by the price of (b) over common time periods. If (a) is a $10 stock and moves 10% and (b) is $5 and also moves 10% then my definition of Return Correlation would result in 100% correlation and 50% price correlation.

I appreciate your reply and welcome your tutelage. I may e-mail the guys at www.pairfinder.com and see what their definitions are since I've been using their site for correlated pairs.

 

Quote from Fraze14:

Return correlation seems to be a more accurate measure because it doesn't treat the moves of a $5 stock and a $20 stock the same, like price correlation.

--------------------------------------------------------------------------------
In reply to Fraze14 -- again, price correlation can accurately measure movement IF you trade them with a factor. For stock A at $5.00 you buy 4x more than stock B at $20. Of course, these can move over and over again out of their ranges and re-capitalizing would make sense. But the same can happen with return correlation -- in that they can easily get unbalanced, and therefore not market neutral. As a professional daytrader, I could care less about % return. I measure things in cents and points just because it's easier to compute in my head. Each person may approach this differently. For investors, you may be more used to % return.

Like any other type of investment, two pairs are not created equal. One can be very tame and the other a wild child. % return to me seems to disguise some risk just because the relative movements seem so small.

I'll have to look at a few intraday in %return format to see how this adjusts my thinking, so that I can more clearly respond to it. I've been doing this for years, and frankly $ return works for me.

You might look for more info at: pairtrader.com, pairstrading.com, pairfinder.com, etc. I'm sure each of these have a different approach to it -- and like all trading strategies you'll have to grab on to one of them and just see how you like it.

sslavin

 

Ok -- I've looked at both a price difference chart and a % chart (or ratio chart) of MWD:MER and MWD:GS. I use a 1:1 differential for MWDMER and a 1:1.5 differential for MWDGS. This was on a 10 min chart over 5 days.

First off -- what would visually confuse me is the small numbers in % than float by. I'd have to really get used to this. We're talking hundreths of a %, yet for the differential this is points moving by. I watch 30 pairs a day and frequently have 10 or more on at one time. These similar numbers would likely confuse me. I'd have to set alerts for everything (sup/resistance) and not rely on my memory as to what number are important.

Second -- a $3.00 point move in MWDGS150 is the same as .02% move. For MWDMER100, this same .02% move is $1.00. So, in % terms you might be fooled into thinking that MWDGS is the same type of risk as MWDMER and let me assure you -- it's not. MWDGS is a much wilder pair to trade.

So, how do you address this issue again, and why are ratio charts your visual choice for pairs? Thanks.

Do you all trade stocks or the S&P's in % terms or % charts? I have yet to see a professional trader do this and I have a pretty vast library of books on traders. I see a reference from time to time when one compares returns of sectors against eachother, or when a fund says we're running 2% above the S&P's for the year. But intraday trading in % terms is a new one I'd have to look at.

Thanks. Good posts going.

sslavin

 

There is a formula for correlation for any 2 functions. It’s pure mathematics. This formula is not very complicated (no integrals or differentials involved) but not the easy one either. You can search for it on the Internet.
That’s not the point. The point is –what functions you use for this formula. In Price Correlation you use 2 functions – 2,4,6 and 3,6,9. Correlation for these 2 functions would be 100%.
In Return correlation you also use 2 functions – but the different ones. Because the stocks go 100% up on the thirst day and another 50% up on the third day in both cases – Return correlation would give you also 100% here.
But consider the previous example:
10,9,18 and 10,1,2 – if you looking for return correlation - first stock goes down 10% and then up 100% and the second stock goes down 90% and then up 100%. Believe it or not for these 2 functions Correlation (in this case Return Correlation) would be 100%.
I’m a programmer myself and trade pairs for fun. I found www.pairstrading.com back in December and since then I back tested their results since 1998. It’s seems like the long weekend already started, so early next week I’ll give some results.

 

Hi, all.
It looks like pairs trading model might do much better then combined with ADX indicator.
I’d like to back test it. Formula for ADX is pretty simple for a stock. However, what are High and Low for a pair? Obviously, they could not be derived from the Highs and Lows of 2 stocks that are in a pair. Do I really need all intraday data for all the stocks (I’m interested only in SP500)? Or I need it per minute / 10 minute / an hour? And if so, where can I get it? Any thoughts?
Thanks