Backtesting results

Discussion in 'Automated Trading' started by vayu, Sep 10, 2012.

  1. vayu


    Suppose you developed an ATS. What backtesting results do you consider as appropriate to start the real money trading?
    What are your criteria to start trading with a new system?

  2. If the forward test equity profile has a reasonable match to that seen in the back test, it may be time to go live..... on a tight leash.

    It's all about the OOS
  3. Mr_You


    * Equity curve looks good. No huge draw downs.
    * Strategy works on multiple instruments and multiple time frames. For example, as of just last night, I'm working on one that works (is profitable) on ES, CL, NQ, TF, YM, etc. And its profitable (albeit with different levels of profitability, but still profitable) when I run it on multiple higher time frames (such as 60 minutes for entries) and multiple lower time frames (such as 10 minutes for exits).
    * All of the above from hundreds to thousands of trades over months or years.

    The above means its probably not curve fitted and a good start to consider Sim/Paper trading the system with live data and/or on an instrument with low tick values. Unless you've got big balls you wouldn't want to immediately start trading the system on the ES for example.

    PM me if you want to hear about a good backtesting webinar on another forum.
  4. dom993


    Several aspects - but first, your backtesting itself:

    - it *MUST* include commissions
    - it *MUST* provide realistic fills (for stocks nowadays this is next to impossible, for futures you still have a chance if you can guarantee your backtesting engine only provide fills once price trade *through* the Limit)
    - it *MUST* provide realistic slippage (on liquid futures, 1-tick for stops & 2-ticks for Market orders should be fair)
    - it *MUST* use quality historical data

    I used to say backtest over at least 3 years & 300 trades (per market), but it turns out to be a pretty optimistic picture ... I am more on a 5 years & 1000 trades today.

    Unless your system trades only Long or Short, it should use the same parameter values for Long & Short, and provide roughly the same amount of trades / win% / P/F / avg/trade ... for both. Of course, this is an idealized view, and it wouldn't be uncommon to have 30 to 50% difference between Longs & Shorts.
    (of course, at some point you'll have enough experience to know when & why shorts should be treated differently from longs)

    Now, for the backtesting results:
    - backtesting results are just that, backtesting results. If the system doesn't perform well in backtest, of course it won't perform well in live trading. But even if the system backtests well, there is no guarantee - ever - it will perform to any similar level of performance in live trading.

    - assume that your net result & average per trade are cut in half. Assume further that the max drawdown gets multiplied by 2. If you can't see yourself still trading that system at 90% of that drawdown, then you and your system are not ready for live trading.

    Reality check:
    - do you understand why your system has an edge?
    - do you know how to monitor the underlying conditions for that edge?
    If you cannot say "yes" without a doubt to both questions, you probably won't have the psychological strength to trade that system even at 50% of its backtesting drawdown (which would be only 25% of what you need to be prepared - financially & psychologically - to go with that system)

    Keep us posted...
  5. That's a warning sign, if tested over significant amounts of time.