Serving the Quantitative Finance Community

 
fiqureshi1
Topic Author
Posts: 4
Joined: April 9th, 2018, 7:40 pm

Stochastic Control in Algo Trading

April 9th, 2018, 8:14 pm

Hi All:

Nice to find this forum, and look forward to learning from the participants.

I am looking into the viability of commercializing some stochastic control based algo trading theory/research. The IP owner is a well credentialed, accomplished (and very smart) professor. The subject matter involves the use of stochastic control theory (stochastic calculus, dynamic programming etc) to find optimum trading strategies (with applications in optimum execution, stat arb etc).  The commercialization idea is to develop a software development "toolkit" that could be used by traders/firms to "optimize strategies in a more mathematically rigorous manner".  The main issue I see is whether the mathematical theory can be made practicable enough to be of value, and how it would fit into the overall "algo trading" picture.  
 
How is stochastic control being used by day traders, firms, institutions? Are there existing stochastic control based software packages? Any thoughts at all on the topic would be most appreciated!

Thanks in advance!
- Fas
 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Stochastic Control in Algo Trading

April 10th, 2018, 7:42 am

Hi Fas,
Many trading firms use stochastic control, but since its their core business they will have assembled dedicated teams in algo development and optimization, backtesting and their data management that goes with that. All these things make it hard to sell software packages, it would create an operational risk to outsource core activities. For stochastic control its common to use model free methods without explicit model of the dynamics or reward like reinforcement learning.
 
fiqureshi1
Topic Author
Posts: 4
Joined: April 9th, 2018, 7:40 pm

Re: Stochastic Control in Algo Trading

April 10th, 2018, 12:52 pm

Thanks outrun for the feedback. You seem to have a good understanding of the industry.

Would using a stochastic control software package in house be considered outsourcing? Lets assume we could get over the "outsourcing risk" concern, could a stand alone Stochastic Control module add value by letting firms try to find optimum strategies which could then be back tested using existing methods? 

Finally regrading your "model-free" point ... isn't stochastic control fundamentally about maximizing a reward function/performance criteria  ?

Thanks again.
-Fas
 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Stochastic Control in Algo Trading

April 10th, 2018, 3:21 pm

I think there are a couple of types companies you can analyse from a sales perspective?

1) Large Trading firms:
These firms have a long history of trading and modelling and they can't realy do their jobs without a good infrastructure and good models. They have large teams. I've worked with a couple and I was impressed by their models, hardware, and development processes. From a content point of view you would want to work with these companies (sophisticated models that are important to them). The backtesting is extremely difficult, the amount of data is huge and there is a need to optimize assume your actions impact the market. All this makes me thing that you can't solve their problems with a software module.

2) Small trading firms
These companies mostly use vendor software, have smaller teams and run simple models in niche markets. Their edge is knowledge about these markets. Their needs are fulfilled with Matlab, R, Python and its libs, maybe even Excel.

3) Asset managers:
They have large funds they move around slowly. They have skilled people but their focus is more on risk mangement, reporting, compliance, legal,.. and allocation is done based on expert opinion and meetings and not overly complicated models.

In general, I expect it'll very difficult as a small vendor like you to find a sweet spot between large and small companies and offer something important for them that they trust you with.

If I had to guess I see two possible paths;
1) My best guess would be to work together with a large trading firm as an advisor: see if you can help them improve their algorithms continuously in return for revenues. Large firms have good ties with universities and they nurture those. they are used to this.
2) write addins/libraries that compete with Matlab / Python optimizers for small trading firms... but there too you would need to make a strong case to make people switch,.. and you would need to learn to do sales and marketing as well! (trying to make people like your facebook page is much harder than stochastic optimization :-) )
 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Stochastic Control in Algo Trading

April 10th, 2018, 3:24 pm

..or work together with a large vendor: companies  like IBM, Bloomberg etc. and let them do the sales and service
 
fiqureshi1
Topic Author
Posts: 4
Joined: April 9th, 2018, 7:40 pm

Re: Stochastic Control in Algo Trading

April 10th, 2018, 8:13 pm

Outrun - Appreciate you taking the time to reply in depth.

What you say makes sense, and I agree with your perspective re: large sophisticated traders and smaller less sophisticated ones. Option 1 definitely would seem like the best angle. But I really like your idea about partnering with vendors. They are always looking to up-sell their clients. The offering could possibly be a toolkit with extensive "consulting" services under the partners brand (i.e. IBM). Speaking of which you have given me the idea to speak to a few contacts I have!!!!

There's still a lot of diligence to do before proceeding obviously. But if we do, I will def. save a spot for you on the eventual advisory board :)
 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Stochastic Control in Algo Trading

April 10th, 2018, 8:44 pm

Excellent! We can have monthly advisory meetings at exotics islands. 

IBM bought "Algorithmic" some time ago, a risk engine and reporting tool, and Ive seen that being used at an asset manager and also at a bank, that's one of their services in the financial sector.

Other cool option I once looked at is to deploy your code on Bloomberg terminals, a lot of potential clients would have a terminal https://www.bloomberg.com/professional/ ... r-program/
Besides that Bloomberg is also pushing for offering portfolio risk analytic services themselves, and they have a big team of quant working on that I've heard, ..which might also be an interesting group to talk, it's an ambitious project: 

https://www.bloomberg.com/professional/ ... analytics/

Lots of options..  I'm curious what will happen, keep us posted!
 
fiqureshi1
Topic Author
Posts: 4
Joined: April 9th, 2018, 7:40 pm

Re: Stochastic Control in Algo Trading

April 13th, 2018, 1:29 pm

Outrun- definitely good ideas. Algorithmic does bank risk software but it's exactly who I was thinking of. And Bloomberg I did not know about but seems like a really good idea to look into. (I'm also going to look into other similar products).

Will keep you posted on development. Thanks for the feedback. You definitely deserve first class tickets if this takes off