Serving the Quantitative Finance Community

 
Antwen
Topic Author
Posts: 1
Joined: October 1st, 2019, 8:50 am

How to determine basket of stocks in pair trading.

October 1st, 2019, 9:00 am

Hi all,

Just as the titled says. With regards to pair trading, does anyone know a methodology in determining the best combination of basket of stocks to pair trade.

That is pair trading 'N' basket of stocks to 'M' basket of stocks. How do you determine N and M.

I've tried searching online but to no avail.

Thanks.

Edit: Just to add abit more colour. Suppose I've tested cointegration for 1 to 1 pairs of stocks in a huge sample of stocks. Of those cointegrated pairs I've back and forward tested with my rule based trading system. But I want to take it one step further by pair trading between basket of stocks. The only way I can think is to run every possible combination of baskets, test cointegration and back/forward test. Which would either make my pc explode or take a very long time. Can I know beforehand (using a metric or methodolgy) that adding 'this' stock into the basket is beneficial or not.
 
User avatar
MartinGale7
Posts: 70
Joined: April 1st, 2011, 7:42 am

Re: How to determine basket of stocks in pair trading.

October 7th, 2019, 10:16 am

Your easiest way is to take a large universe of stocks. Measure and rank correlations over a long timeframe (eg weekly). Measure and rank correlations over short timeframe (eg hourly) too. Subtract the two ranks. You will find the most cointegrated pairs will have good price correlation over the long term but not over the short term.

That's what I do. Hope it helps.
 
User avatar
JohnLeM
Posts: 379
Joined: September 16th, 2008, 7:15 pm

Re: How to determine basket of stocks in pair trading.

October 7th, 2019, 10:38 am

Hello, 

I am not sure that this helps for pair-trading, but I did some related works in a more complex situation : find the best basket for Autocalls, based over some criteria as coupon values.
You can have a look here for some resuts, where we tested any combinations of Autocalls written on any basket of N shares among M. This is feasible for M <=14 (2^14 = 16384 possible combinations).

For larger basket, there is a variant that is faster to reach a given optimality criteria : use some kind of descent algorithm on a graph to reach the criteria, the graph being defined by every combinations.

Hope it helps !