Algorithmic Trading Implementation Help

Discussion in 'Automated Trading' started by wfeagin3, Nov 13, 2012.

  1. wfeagin3


    I need some opinions on my situation. I have an algorithm that I have developed on my own and now have started some backtesting. No one has given me any advice or input in developing this algorithm.

    However, I am working with a friend of mine that is more educated with computers ( he has a masters in computer science). So I have consulted with him on some coding specifically with historical data retrieval. He has not seen my algorithm or any code that I have created (using Matlab currently).

    The results from the backtests show real promise of producing real money. At this point, all my friend has provided is purely some IT help and consultation. If everything stays the same from this point to the time where the algorithm is profittable, what is adequate compensation for his contribution? Keep in mind that he had zero input with developing the algorithm.

    It was my hope that we would have a partnership. He would teach me all the computer knowledge and I would teach him how to write algorithms because I believe he is more than capable. However, he is not contributing like I thought he would be at this point. So I don't want to show him the algo since he isn't providing anything of equal value.

    I need to guard my algo for scalability reasons in the future (i.e. if 2 people run the same algo, the order size is limited).

    The next thing on the to-do list is to put the algo in the optimal laugage which I had planned to let my friend do. Now I am having second thoughts.

    Any comments or opinions would be greatly appreciated. Thanks!
  2. Contract and pay him a flat few for his work thus far and negotiate an ongoing rate for support.
  3. Mr_You


    First off, just because a strategy appears profitable in backtesting doesn't mean it will be profitable. Questions you need to ask yourself:

    - Have you eliminated the possibility that your strategy is overly curve fitted to the period of time and instrument you are backtesting?
    - Is your strategy profitable across multiple time frames and multiple instruments?
    - Have you performed out-of-sample walk forward testing?
    - Have you performed out-of-sample walk forward testing using a live data feed?
    - Have you performed live trading with real money using a low tick value/low risk instrument?

    I'm personally not convinced a strategy will be profitable until all the above is completed.

    To answer your questions...

    I would pay him an hourly rate of $20-$40.

    Now another question... What is this "optimal language" you speak of? Personally I would recommend implementing the strategy in a trading platform vs shooting for a specific language. PM me if you would like further assistance privately.

    Good luck!
  4. If it's not high frequency, just run it out of Matlab. No need to change language platform. Run it at Interactive Brokers, as there is a decent Matlab-to-IBAPI connector.

    Run it against the IB papertrading account first for a few months, just in case realtime performance doesn't really match your backtest.
  5. You want his skills to make money for u, but u are afraid he'll take your algo and run away to make money. YET, all u asking for is just "some computer knowledge", again for you to make money.

    Wont he be at the losing end ? Would he be stupid enuff to slave for u ?
  6. bluetrin


    I guess the OP better find a very good programmer with good IT skills and propose a partnership, in the long run it is probably the only way to have a serious business ...
  7. Mr_You


    If the OPer is coding strategies in Matlab I'm not sure why he needs any programming help or even IT help considering most of his questions could probably be answered with a Google search.
  8. There's two excellent points of advice in this thread already: first, what appears in backtesting, may well not appear in real cash trading. Second, fire it up against an API as mentioned, using paper money, and worry about who to trust and where to stack all the gold, later. : )

    1. If you're in a market that's so thinly traded it cannot support more than one retail trader, you're on a knife's edge.

    2. If your algo is fitted to a specific initial instrument such that it cannot be adapted to any other instrument, it will also not fit that initial instrument at all times of the year and in all market conditions, so again, it's questionable that you've got a robust system.

    If you ignore those two points and proceed, you need only denature the algo, then have any number of third party developers (they are legion ... just look at them on the many "me code you long time" and trade-a-long sites ...) do the work either in whole or give one developer one component and another developer another component ... then they have no way of knowing what they're building.

    3. If you can produce the functional decomposition to be able to compartmentalize your algo, you're going to have to go back to points 1 and 2 to reflect upon what you've really built.

    4. Thinking about your partnership with a developer, there's one hard lesson to learn about business partnerships: if you cannot trust the other party, you don't have a partnership.

    I'd suggest going in with something like a 50:50 revenue share for 5 years and you retain written and agreed intellectual property rights to all future applications, with boilerplate (and ironclad! : ) language for non-compete, etc. You can get this in any employment contract for technology R&D -- it's binding, but only bound by your willingness to protect your property and that can mean anywhere in the world. But if you've invented the money-printer algo, then it will be worth paying 50% to get it printing and you'll then have ample war chest funds to protect it.

    Well, good luck with the testing. I trust you're already underway with IB by the time, if ever, you read this post ... if you do have the license to print money ... hey buddy, can you spare a dime? ... and by "dime" I mean $10 million ...
  9. bluetrin


    If you give me 10mio $, I will be able to start a fund with 3 other partners. We have a strategy which will return 15% annually ...

    Spare a dime here : )
  10. Humpy


    Also get him to put in writing that he will only use your algo for himself ( can't stop him anyway )
    #10     Nov 22, 2012