Results too good to be true?

Discussion in 'Strategy Building' started by BrooksRimes, Jul 10, 2005.

  1. I am somewhat embarrassed to say that I found Amibroker was using the open price of the bar being tested for the condition rather than the open of the next bar, by default. When changed to the open of the next bar, the system became unprofitable.

    Amibroker does not seem to have the checks and balances that I was used to from working with Easy Language and TS which uses the next bar's open.

    Sorry for the confusion and thanks to all who posted. There is a lesson in here for me and maybe others.

    I am a big fan of AB, but would change how some things work with backtesting. You can specify things in code and specify the same things in Setting windows and its your responsibility to make sure it works the way you intended. There are many backtesting features and it is powerful but it can also be complex and difficult to understand, IMHO.
     
    #11     Jul 11, 2005
  2. Div_Arb

    Div_Arb

    Are there any good sites out there to back-test a strategy on?

    What software is the best for back testing?
     
    #12     Jul 14, 2005
  3. bobbyod

    bobbyod

    I've found Wealth-Lab to be much more versatile and easier to learn than AB.

    www.wealth-lab.com
     
    #13     Jul 14, 2005
  4. mogul

    mogul

    I find tradestation paints a very accurate picture of the outcome, if you go back and verify sample trades more closely

    personally I don't trust amibroker for it's backtesting
     
    #14     Jul 14, 2005
  5. You may want to start a separate thread for this question. You should also search for 'backtesting' - its probably been asked and answered before.

    There are many choices: Tradestation, Ensign, eSignal, Amibroker, Wealthlabs, etc. Features vary. A key question is how difficult is the programming language to learn. If you are not a detail person and don't have a little programmer in you, you will not enjoy it.

    For Wealthlabs, you must be a Fidelity brokerage customer.

     
    #15     Jul 14, 2005
  6. I think what you actually did is called " Self-fulfilling strategy" which tradestation don't allow that.

    That means referencing tomorrow or next bars OHLC and
    testing/generating order on the OHLC of current bar.

    I personably think Amibroker is garbage, my reason is by
    looking at it's syntax and build-in functions ( though i have
    no experience with it's performance).
     
    #16     Jul 14, 2005
  7. To address some misunderstandings:

    1. AmiBroker's default setting is not to trade on open with zero delay. The default is to trade on CLOSE.
    So you needed to change the settings by yourself to open/no delay.

    2. As for "not allowing the user to do something". Well in any language (including EasyLanguage) there are many ways to code things which are not possible in real trading. The reasoning behind leaving the user an option to specify trading on open with zero delay is that the user's code may already inducing delays on its own so no additional delay is needed. For example:
    If your code is like this:
    Buy = Ref( Cross( C, MA(C, 15 ), -1 );
    it implements checking for cross condition in PREVIOUS bar, so buy signal for TODAY's open will be generated when Close crosses above MA YESTERDAY.
    This is perfectly valid and no additional delay is necessary.

    What I want to point out is that in ANY langauge the USER needs to understand his OWN code and it has nothing to do with the tool he/she uses.
    Of course AmiBroker has some tool (Tools->Check in the Editor) that helps the user to spot some obvious cases of referncing future, but no automatic tool can really replace user's own thinking.

    3. As for comments about AFL language : take a look for example at the coding for Stocks&Commodities Traders Tips http://www.traders.com/Documentation/FEEDbk_docs/Archive/042005/TradersTips/TradersTips.html and compare the lengths of formulas for different platforms (including Tradestation, WealthLab, and others) to AFL coding and you will quickly find out that AFL codes are usually 2-3 times shorter than others doing exactly the same.
     
    #17     Jul 21, 2005