New pairs trading software draws attention

G

gwb-trading

Interesting Article from a third party news source...

New pairs trading software draws attention

Louis Capital Markets is known for its quantitative equity trading. This year, it launched an updated version of a long/short equity model for trading equity pairs. The model signals when pairs of like-industry stocks are deviating from their historical averages and suggests when that will reverse. Since then, as interest in algorithmic trading has heated up, interest has ratcheted up. The fact is that hedge funds are hungry for new ideas. Anything that offers some differentiation will get some looks.

Finding the Perfect Pair
Louis Capital Markets develops quant model to sift through 180,000 equity pairs and select the top 15.

WallStreetTech Article on Finding the Perfect Pair
 
Interesting post. Intelligent product. I would comment:

(1) Mean reversion of hedged pairs is a very simple strategy...
And it can be very profitable if done right.

But often people do not chose/screen pairs in an optimal way.
The big mistake is overestimating the relationship between 2 securities...
Or being unaware of the pitfalls indigenous to a particular class of stock...
For example, I could give you a list of 10 deadly pitfalls pertaining only to convertible securities.
Using just corelation with convertibles will get you crushed... and so forth with other classes.

(2) This works in temporarily inefficient markets.
One can argue that liquid markets do not have inefficiencies...
So making the cutoff at $20M/day in trading volume defeats the purpose.
My cutoff is $100K/day.

(3) Giving the same numbers to 250 institutional clients... would obviously work against this product.

(4) If you don't get source code... or it's not FULLY customizable... then coding your own is well worth it.
2-3 programmers at 75K/year plus expenses...
Could code and test a proprietary equivalent in 12-24 months.

Any decent size hedge fund should be able to roll their own.
As opposed to paying 5 figures/year forever.
 
From the "article" (read: paid advertisement):

To reduce the universe of 180,000 pairs down to about 200 to 300 remaining pairs, LCM applies a number of criteria. For example, pairs must be in the same industry and stocks must be liquid with at least $20 million in U.S. trading volume. Further, LCM defines the different correlation averages for each industry sector.

"Correlation is really the key data in the model -- the data that drives the long/short opportunity," says LCM's Gaubert. For instance, the oil and gas industry has an 85 percent correlation at minimum, while Internet stocks have a 40 percent correlation.
More...

This is a fundamentally flawed model. It is also 20 years or more behind current techniques for modeling non-stationary time series. IMO, its worth less than a blazing stack of pesos.

-segv
 
What would you say are some up to date ways of modeling nonstationary time series?


(This is a serious question, not a troll.)
 
Quote from segv:

From the "article" (read: paid advertisement):



This is a fundamentally flawed model. It is also 20 years or more behind current techniques for modeling non-stationary time series. IMO, its worth less than a blazing stack of pesos.

-segv
More...

Actually...
My proprietary software and trading is based on the similar principles...
And I have been consistently profitable for > 10 years.

This is basic quant analysis...
And not bogus like TA or astrology or seances with the dead.

Main difference is that I'm super, laser targeted...
While they are supplying a broad, generic product that might be useful...
But MUST be integrated with a lot of other systems...
And run by an experienced professional.

** To address your point **...
I have always found it a waste of time to employ mathematical complexity...
Higher than the level of simple corelation and linear equations.

I can and have tried...
But the importance of execution and professional decision making...
Outweighs any precision you might squeeze out...
By, for example, using non-linear equations/regression or exotic time series analysis.

Another way of saying it...
The ** high noise ** of the trading environment...
Overwhelms hair-splitting math/stats techniques.

Another way of saying it...
It doesn't matter if the "fair value" you come up with is $10.00 or $10.05...
It's how well you exploit/trade the market inefficiency.
 
Quote from DrChaos:

What would you say are some up to date ways of modeling nonstationary time series?
(This is a serious question, not a troll.)
More...

Sorry, but I cannot comment on this topic, other than to point you towards a starting point. Start here and work your way forward:

Engle, RF and CWJ Granger, 1987, Co-Integration and Error Correction: Representation, Estimation, and Testing, Econometrica 55, 251--276.

-segv
 
Quote from HoundDogOne:

But the importance of execution and professional decision making...
Outweighs any precision you might squeeze out...
By, for example, using non-linear equations/regression or exotic time series analysis.
More...

Execution and sound decision making are essential parts of any successful strategy. Successful execution and sound decision making with a fundamentally flawed model will still result in a negative expectation.


Another way of saying it...
The ** high noise ** of the trading environment...
Overwhelms hair-splitting math/stats techniques.
More...

I am quite happy that so many people have so little faith in such pursuits. In any case, the software touted by this article is in all likelihood worthless.

-segv
 
You must log in or register to reply here.
Back
Top