DislikedThis will be helpful so as more of us have this skill set, we all can work more collectively in Optimizing and Testing this EA.Ignored
http://www.strategybuilderfx.com/sho...t=optimization
More detail information:
http://www.strategybuilderfx.com/sho...t=optimization
You may also reading this book for overall concept:
Design, Testing, and Optimization of Trading Systems
ISBN: 0471554464
Refer to MT4 official site for step-by-step tutorial:
http://articles.mql4.com/192
Programming consideration in testing:
http://articles.mql4.com/72
If you are interesting about internal algorithm of tester:
http://articles.mql4.com/134
A sample of complete MT4 EA development cycle:
http://articles.mql4.com/163
Since there are too many things to be care when you are doing optimization, I think here are some things you must keep in mind:
1. Get historical M1 tick data as long as you can.
2. Watch out the timezone difference between your M1 tick data and your broker's server. Difference timezone may produce difference result sometime.
3. Make sure each testing produce 90% modeling quality result, and don't trust any result if MQ<90%.
4. Beside to M1 tick data, make sure you have also download all timeframe (M5, M15, H1, Hxxx.....) data from your broker. If the downloaded data cannot cover your testing period (ex: you want test 2 year, but broker only provide 6 months data), you may produce all timeframe data from your M1 historical tick via period_convert script which build-in with your MT4.
5. Make sure your computer has fastest cpu and largest memory, or you may waste you life....
6. Do not use OVER-OPTIMIZED result (someone call it CURVE-FITTED). Since over-optimized result only perform good to historical data, and will make very bad performance in live trade.
7. More longer testing period (ex: more than 3 year) will produce more reliable result.
The hardest part is: how do I know which result is over-optimized?
I have no answer too, but if your result can be used with difference pairs (or testing period) with only a little parameter shifting, that may be a good result. In other word, if you get very difference result with difference data pattern, or get very weird parameter number, especially if you get very HIGH PROFIT with only FEW TRADE times, that may be a very bad result.
By the way, someone's optimization result may not fit to your live trade style, since you may using: difference historical data, difference broker timezone, difference broker price, difference risk strategy.....I have found many times that someone published optimized parameter cannot produce same result with my backtesting.
- Ray Tracy -
Novice forex trader from GMT+8