How far to backtest intraday?

Discussion in 'Strategy Development' started by BrooksRimes, Jun 9, 2010.

  1. For an intraday system such as based on 5 minute bars that generates a few trades per day, how far should it be backtested?

    This is for futures (ES, NQ, TF, YM ZB).

  2. Retief


    Years, and the longer the better.
  3. For 5 minutes, I would use 3-5 years. For 1 minute, I would use 1-2 years.

    You need to factor the frequency of trading. Less trades, require more back testing.
  4. Six months. Things change. Strategies breathe.
  5. Retief


    Cockroaches haven't changed in hundreds of millions of years.
  6. Agreed. But market makers are slightly higher on the evolutionary scale.
  7. Retief


    The point about the cockroaches, is that a system that keeps working through all kinds of market environments is going to be something simple. Something like a cockroach that eats crumbs and scuttles out of the way when larger animals are about. So it should be back testable over a long time.

    Can you really trust a system that has only been back tested over six months? A long-only system would likely back test pretty well from March 9, 2009 through the present. It would probably suck during 2008.
  8. ET99


    intraday? like scalping?
    backtest it for 1 busy week.
    if you can't make money after a week, your system does not work. No amount of extended backtest will help.
    If it makes money after a week, then extend the test to a month.
  9. vikana

    vikana Moderator

    Somewhat depending on trading frequency, I see no reason not to test on 2007 through today. That includes a wide range of markets.
  10. Retief


    What platform are you using for the back testing? I have not found it to be that reliable. Actual trading varies significantly, and the shorter the time frame, the worse it gets. I can't imagine relying on a back test over just one week.

    Back test a martingale over a week that was choppy with no clear trends developing. Now use that martingale during a week when a trend occurs and see what happens.

    Edit: Maybe back testing isn't very useful, sort of like trying to back test a poker strategy.
    #10     Jun 9, 2010