@jsauter86 the genetic algo finds the best inputs using backtest information between 2009-2015. After that, it tests if those inputs hold in the "future", 2015-2020. If forward test fails, we know the good backtest was just luck ("overfitting").
https://en.wikipedia.org/wiki/Training,_validation,_and_test_sets
@jsauter86 fair enough, sounds good!
@jsauter86 I mean the optimization algorithm in the tester, you use that?
Recovery factor = net profit / biggest drawdown
@jsauter86 for me, recovery factor optimizing works better than just balance, but that might depend on the strategy
@jsauter86 thats a good result. Lots of trades, both longs and shorts, profit is likely not just coincidence.
From the deposit load I see the EA builds bigger and smaller positions, nothing too crazy.
Well done
@umbrellamaker there's a block "change timeframe for next blocks", that should work.
The enum_timeframes thing should also work... https://docs.mql4.com/constants/chartconstants/enum_timeframes
@seb-0 that long-only thing is a fair point. You know eurusd has risen in the past year, so long-only EAs will naturally work better in the backtest. However, you dont know if eurusd will rise in the future years. If you have to assume it rises, you could just go long and forget about the position.
Only a long-short system can prove you can make profit without knowing the direction beforehand.
Just playing devil's advocate here lol
@jsauter86 all right. My goal is to find something for all market conditions. As per my logic, the further an EA works to the past, the further it will work to the future.
@jsauter86 pretty good! How about 2013-2021 ?
2014 and 2017 saw some big trends on eurusd, that would really stress test the ea
@jsauter86 said in My Best Result Ever, Check this out guys.:
Im not a fan of 88% equity drawdown...
Didnt notice that, lol
Spreads are much larger during rollover, is it just that?
@lanrayy that is a good result. What is that instrument, "step index"? Can you really trade that index?
@jsauter86 I dont know... Try to make your EA simpler and test again -> if theres still difference, make the EA even more simplier and repeat. See if in the end, your EA buys every bar and the 2 PCs still behave different.
And hit the default settings button in mt5, just in case
@chrisl-0 so, keep track of highest recorded balance (I would do it that way), and if current balance is enough below that record, change your risk percent.
Use these blocks in your project, should work
https://fxdreema.com/shared/P9AoUisOd
@sm24 many times the tester finds just 1 or 2 really good trades with no drawdown and gets stuck with those. I think a good score would be recovery factor * trades, or sharpe * trades
Hi @fabiobioware.
I think I can help you with this, and also make your project much smaller. If you can share the whole thing.
@jsauter86 just a basic rule in statistics. More data means more proof. Less data means less proof.
If you cant have more trades in a year, then you should test multiple years.
I would love to see this same backtest result, but started 5 years earlier.