Results too good to be true?

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

  1. Here is a simple MACD crossover system:

    Buy=Cross(MACD(),Signal() );
    Sell=Cross(Signal(),MACD() );

    Testing a year of 5 minute data on YM gives excellent results:

    Profit factor 2.07
    Payoff ratio: 1.98
    Net profit/drawdown: 20:1
    % Winners: 51%

    The results remain excellent with just about any N minutes. This is non-optimized.

    What am I missing here? It seems too good to be true.
  2. Do a search for macd crossover on this forum. This has been discussed a number of times.
  3. jsmith


    Did you take into consideration Commissions + Slippage
  4. I only found 2 threads with 'macd crossover' in the subject. Neither got into backtesting.

  5. My first thought is your backtesting formula is off.
  6. JackR


    You do not indicate the MACD parameters. Were they standard?

    What was your backtesting platform?

    I do not trade the YM but I believe it trades in $5.00 increments.
    Can you trade between those increments?
    Did your backtesting engine assume you could?
    Did you use market or limit orders?
    Did your backtesting engine assume you got the trade if the trade after the signal touched the next $5 increment or did the YM have to trade through that increment to the next increment to ensure a "real" trade?

    You tested with 5 minute data. Can you test to see if your buy price would have been available in the x next minutes?

    Obviously slippage plays an important part in addition to the above backtesting possibilities.
  7. Thanks for the post.

    Yes, they were standard: 12, 26 ...

    Correct, 1 tic = $5. No, you can't trade in between and the software did not assume so. It is Amibroker ( The orders are to buy/short on the open of the next bar or to sell/cover on the close of the next bar.

    I have a generous allowance for slippage and commission.

  8. How many trades? I guess > 1000...
    with realistic slippage and so frequent trading I can't imagine how could you get PF >2
  9. Hi Brooks,

    As you ask me, my opinion is difficult to make automatically due to different type of crossover in MACD.

    In MACD are very important two aspects: The different data compression to be considered and the Divergence.

    As you see in the next charts, the successful crossover is also in divergence.




  10. JackR



    Only thing you didn't mention and I didn't ask in my last post was the inclusive dates of the one year timeframe. I'm assuming you used the last twelve months. Do you have sufficient data to move the timeframe around to cover different markets (trending + and -, choppy)?

    You may have hit the perfect time for the crossover technique to work and, of course, if you commit money to it, it will fail.

    #10     Jul 11, 2005