How to price an hft blackbox algo if you have to sell it?

Aleph · Post by **Aleph** » Thu Jun 20, 2013 9:03 am

It's easy now to find shops with good infrastructure who will give you a large-ish cut (50%) and leave you to be. I think you want to be a bit careful with how much of the details you reveal as I have heard that if you're a one trick pony and a bit too open they some have been rumored to pass on details to other desks on a smaller cut, and then on down the line..

Along those lines:

"Trading strategies don't fetch high prices, and there is a reason for that. Especially High-Frequency strategies tend to break without much notice. "

This is key. Don't be a guy with -a- strategy, be the guy who creates strategies. Those people are far fewer and more valuable. Most people with -a- strategy usually got it by working with someone who built it and really understands it and think the ones who 'take' it and leave believe they can replicate it elsewhere. They rarely can. The devil is always in the un-sexy, seemingly unimportant details.

If you're the only who's able to build new things, someone will eventually try to copy it and leave. It won't really matter though as they'll do a poor job and it'll decay in a few months leaving them with nothing.You''ll already be on to the next one.

veserog88 · Post by **veserog88** » Fri Jun 21, 2013 7:45 pm

Thanks for the input, Aleph, it makes a lot of sense.

ESMaestro · Post by **ESMaestro** » Sat Jun 22, 2013 12:36 am

Nm.

#edited

Tradenator · Post by **Tradenator** » Sun Jun 23, 2013 11:51 am

Another approach to the valuation issue is a factor model using listed managers. Factors to consider include AUM and length of track record (longevity). I would think that the AUM factor convolutes a mixture of performance plus distribution. To back out the distribution component you can try to get data on listed managers that are more about distribution than actual trading. You might also consider adjusting for the difference between a listed manager and the specific type you are dealing with, such as a family office or prop shop. Such a value would likely carry wide error bars around the result.

I think FDAX's advice on being positive and getting things going quickly is very important. I am not a lawyer, so definitely seek advice on the following thoughts which are not legal advice in any way. They are just my thoughts, without knowledge of your circumstances and requirements.

To placate yourself and your existing partners without selling the algo, you can set up a holding company and license the software to the manager with exclusivity and with the source code held in escrow with an agent you trust. To protect the licensee, they will likely want the license to be perpetual, but it can be for, say, N years instead. People need to see that the system will be around for a long time, otherwise they won't invest. They should also ask for minimum performance/drawdown to be agreed (don't overpromise and under-deliver). To protect the licensor there should be IP protection clauses (no attempt to recreate...). The agreement should include tough clauses that trigger or nullify escrow if either side violates the license. IMHO the escrow agreement is best done separately, and referred to in the license agreement. The license should cover not just the code, but all of the IP, using words such as "knowhow" to include the human element.

EDIT: If you license any software, make sure that anything it is built on allows you to license the results. Check any open source, Microsoft, Sun, etc licenses that cover what you have built before spending too much on lawyers. I like open source, but the licenses aren't always amenable to commercial application so be careful and get good advice.

veserog88 · Post by **veserog88** » Fri Jun 28, 2013 8:16 am

Tradenator, thanks a lot for the super informative post. I've just read it, and it is quite a coincidence but after all I and my partners are now following the route you described almost word by word.