Page 1 of 1

Algo-trading system design question

Posted: December 29th, 2009, 3:42 pm
by Stale
Hi guys,I'm playing around on my spare time building a trading system. It is mostly to get my hands dirty on a variety of issues, from system design to programming and SQL. This will - if it gets that far - do some simple mean-reversion strategies and the like, with a low frequency.I've separated the system into components, using something like MVC, and were thinking of how the algos should be maintained. Should I store all the algos in the SQL db? Like, for the mean reversion pair, should I make a table listing all the pairs, with stocks, calibrated parameters etc. ? Another alternative would be to store a short description of the algo only, and use this to track performance. It would be nice to keep all data, but this will get time-consuming for many pairs/strategies and would mess up the design if I were to do several algos. Any clues appreciated,Stale

Algo-trading system design question

Posted: January 1st, 2010, 5:50 pm
by aiQUANT
1 question, what is this system ultimately supposed to do?

Algo-trading system design question

Posted: January 1st, 2010, 11:20 pm
by Stale
Hi,Thanks for your replies.Like Outrun states, searching for profitable algos and storing performance metrics, but also algo parameters, on these was one solution I was looking into. I'm just not sure whether this would demand too much logic, and thus would give me a hard time implementing this. I googled around some more, and found this thesis, ftp://ftp.cs.utexas.edu/pub/techreports ... tovich.pdf , which describes one possible solution. There is no intelligence for looking for algos, so I guess these are put into the system in some static fashion. Do you think this is the way to go?Cheers,Stale

Algo-trading system design question

Posted: January 22nd, 2010, 1:24 pm
by goekhan
No its not the way to go. You have to have a dynamic setting, either in the sence of constant recalibration and toolsthat test whether recalibration is necessary. But also your methodologies have to by dynamic as well.There are ways how you can algorithmize the evolution of your methodology instead of pure parameter-changes.And than you constantly do model selection.