I know that the Metatrader 4 strategy tester is limited, and that backtesting as a whole is an unreliable affair, but I never thought it would be this difficult to get a remotely accurate result. Can anyone tell me what I’m missing here?
I have been using with Metatrader 4 for the last three years, and have been writing and demo-testing expert advisors for most of that time. Every single time a new terminal build comes out, I try out the strategy tester, but I have yet to get a valid result (by valid, I mean a set of tester trades that very closely approximates actual demo account trades that the advisor has made over the same timeframe in a live test).
So far I have tried:
I have been using with Metatrader 4 for the last three years, and have been writing and demo-testing expert advisors for most of that time. Every single time a new terminal build comes out, I try out the strategy tester, but I have yet to get a valid result (by valid, I mean a set of tester trades that very closely approximates actual demo account trades that the advisor has made over the same timeframe in a live test).
So far I have tried:
- ·All three emulation models (including “every tick” at 90% quality)
- ·About ten different data sources (Alpari’s Datacenter, Excel tick data found online, data gotten from MT4’s History Center using the “Download” button, etc.)
- ·Using the period converter to translate 1M data into other timeframes
- ·Various brokers' platforms
- ·Older Metatrader builds
- ·About a hundred different combinations of EA, timeframe, parameter, and testing date range, including EAs that trade on longer 4H-1D charts
My strategy tester, meanwhile, does one of the following:
- ·Makes absolutely no trades, ever
- ·Gives modeling quality below 90%
- ·“Forgets” months of data at a time
- ·Makes trades that seem legit, but actually have nothing to do with the live account results of the same EA over the same date range
And it seems that I’m not the only one who has this problem – whenever I visit a forum dedicated to EA development, I see other posters getting wildly different results by testing the same EAs on the same terminals with the same 90% modeling quality.
Sometimes I feel like I’m just feeding random numbers into a black-box random number generator that outputs some more random numbers and then stamps a big green 90% sticker on it, regardless of what it did. Can anyone relate?