Risk control in auto-trading and HFT?

Discussion in 'Automated Trading' started by mizhael, Sep 24, 2010.

  1. Hi all,

    Could anybody please share your experiences and point me to resources and discussions about risk control in auto-trading and HFT? I mean risks such as market crash, crappy data, etc.

    Thank you!
     
  2. HFT is over-rated and not nearly as profitable as everyone thinks.

    Box simple strategies that can be traded manually. Don't even attempt to get into the HFT space - if you aren't already there its too late. The barrier to entry (true HFT) is too high.

    risk controls = build an "oh shit" button/connection tool so your stuff gets flat if something happens, worst case phone into your broker/execution and tell them to get you flat.
     
  3. I read somewhere that a HFT firm in California only had $10 million in capital and yet they were providing liquidity for 3% of the daily SPY or something... I definitely wouldn't call that risk control.
     
  4. Yep, I was asking about what are the possible "oh shit" situations and how people commonly tackled the problems, esp. the pre-trade risk control measures...

    Any further thoughts? Thanks
     
  5. Best risk control method would be to forget about this facade. Barring the cost of entry, which includes opportunity cost among other things, there is really no way to "simulate" these opportunities.

    There is still plenty of opportunity in larger timeframes throughout the marketplace to take advantage of. If you think you have an idea for HFT, and cannot come up with 1-5 minute programs that trade 10-25+ times a day that yield substantially well, you probably should just either :

    a) hang yourself

    b) go back to the call center making $11.50 / hr
     
  6. If you can't figure out THIS...
    Then back to the phone bank.
     
  7. Depends. Risk Management and Control is about loss limits and risk per trade. If a HFT system makes 500 trades per hour, risk parameters can be within limits AND it can provide a significant percentage of the volume. Harder thing is to come up with such an algorithm without the use of unfair edges (like flash orders).
     
  8. Any more thoughts?
     
  9. There really is no excuse for a loss of connectivity. You should never be running ATS on a retail (or god forbid wireless of any sort) connection.

    Pay for Five 9's (99.999%) uptime at your ISP + get power/backup and set it & forget it.

    If the shit hits the fan you will have a phone to call your back office to get you flat... If your phones/cell phones don't work and you are in the USA chances are its either New Years Eve or the NYSE is limit down and you have bigger issues than getting flat.

    My "oh shit" button is only in simulation and for debugging and as the strategy runs live for the first time. I built a small script that I use on a second machine and if the program runs away from me I click the button and it gets me flat. I never use it.