1. D-Wave Quantum Inc
  2. Community
  3. Coding Tips and Tricks

Assistance in Encoding an AI High-Frequency Trading Algorithm

Avatar
Daro G
Follow

I have built a AI High-Frequency Trading system whose logic is very similar to how a quantum computer operates. The trading system enables one to manage large (500 to more than 2000 symbols simultaneously) baskets of stocks in real time. The software is causing some problems when operated on a classical computer: the complexity and speed of the system is such that both hardware and the Linux operating system tend to fail on a regular basis. I am hoping that converting the system to run on a quantum computer might solve the problem.

I am looking for assistance/guidance in rewriting logic that uses AI and digital signal processing techniques to process probabilities related to optimizing the value of a basket of stocks in real time. This is a purely Alpha (profit) seeking algorithm that is stable across time and market conditions. Technical problems with hardware and software not designed to handle the processing load is the primary issue at this time.

If anyone is interested in being of assistance, please leave a comment below. I believe I need a means of encoding probability statements that can be solved in parallel, which is effectively what I am doing now with a large server. A quantum computer seems to be an ideal tool for this problem. However, the available programming tools are somewhat primitive with respect to the complexity of solving a large number of probability logic equations in real time.

0

Comments

2 comments
Sort by Date Votes
  • Avatar
    Alex K  (Report)

    Hi Daro,

    This sounds like a really interesting problem!

    It seems like you're approaching encoding your application onto the QPU from the right frame of mind - you will need to translate the problem to a binary quadratic model (QUBO/Ising) or constraint satisfaction problem in order to run it on the QPU. Have you taken a look at this guide on formulating a problem or this guide on solving problems on the D-Wave system? 

    Given the size and complexity of your problem I think the Hybrid framework may be a good candidate. It decomposes a problem and runs the sub-problems in parallel on classical and quantum resources. It might be similar to what you're doing with a large server. 

    Here are a few papers on using quantum computing in portfolio optimization. I don't think any of them use a purely alpha seeking algorithm, but there may be some other similarities or approaches that could help.

    Let me know if you have any questions!

     

    0
    Comment actions Permalink
  • Avatar
    Daro G  (Report)
    Alex,
     
    Thank you for your response. 
     
    I looked over the papers and discovered that I am solving a very different financial optimization problem than those considered. My algorithm is an Alpha "hunter seeker" strategy that identifies opportunities as they arise in the market and dynamically targets those investment opportunities in real time for cash profit. I am running at approximately 80% accuracy identifying and trading stock investment opportunities. Typical projected annual return exceeds 1000%, if I can succeed in getting the hardware to sustain the processing load.
     
    However, my hardware is having trouble keeping up with the software logic. One possible solution is to buy more and more servers. I am already have a Linux box with 64 CPUs, and am running into trouble. Another 50 CPUs might help in the short-term, but I believe I have a solution that would benefit from even more powerful computing resources, e.g., a quantum computer.
     
    I am thinking of building a solution---hardware and software---that I can market to institutions such as Goldman Sachs. One strategy is to team up with a quantum computing company. Alternatively, I can just buy more and more classical servers and spread the processing load as evenly as possible. My algorithm is capable of trading all of the securities of all markets worldwide in parallel with an average ROI per day of up to 5%, if I can solve the hardware problem.
     
    Do you have any suggestions? The problem is interesting from an academic perspective, but I am running a for-profit enterprise and need a workable solution at a cost-effective price. Please tell me if I am looking in the wrong direction for a practical solution.
    0
    Comment actions Permalink

Please sign in to leave a comment.

Didn't find what you were looking for?

New post