Update v3
– indicator included in the code and not called anymore, in order to easily optimize pivotbar and lookback values (thanks to Grahal for the idea)
– included security exits when bullish become bearish and bearish become bullish
– new graphic with optimizable values highlighted
Tried to optimize on us500 as a first attempt on another asset, it didn’t go badly at all
Don’t need further days on demo testing to understand that this was highly optimized for the last week and now it’s trash (someone said short timeframe system problematics?)
That’s why i’m working on v4 with the intent to have it FULLY working under machine learning; created 1s 5s and 10s to run for the rest of the week.
I will update you on results, for now here’s the differences between the non dynamic and dynamic version on DOW, looks promising?
Update v4
– full implementation of machine learning on EVERY value, the only thing you have to change is positionsize (big thanks to Paul). Values reset each day in order to have no correlation with some kinds of optimization of the dynamic values adjustement.
– bearish and bullish detection converted from exponential average to moving average
Backtest on positionsize=0.2, resulting on a 50% gain
Still trying to manage some holes in position opening before considering it for demo forward testing (see attached)
looks promising! when you say 50% gain, do you mean the improvement add by ML?
I mean the gain in reference to the position size (118€ gain with a 240€ position size)
If you are curious, here’s the results of the non dynamic version with values optimized on the last week.
Sorry, I still dont get it. You mean €240 in working capital + a €118 profit?
The ML improvement is v impressive!
Thanks for this work !
I like the < 1 min stratégie. I gonna try this !
Did you try a trailing ?
The system started scalping again after 19:00 and ended up in gain also today, 5/5; maybe it’s ready for demo testing 🙂
Further updates at the end of the week.
Did you try a trailing ?
The system already have a trailing stop
PaulParticipant
Master
@Francesco Hi that pic looks promising!
a little thing I prefer myself. You have flatbefore/after.
Perhaps use i.e. time < 21u to open a trade at the latest and flat after 22u.
Nice to see you took the ML code and implemented it! For me I didn’t have the success I wanted with it so far.
with losses, I looked at the bar when the signal came. If it closed at the high and it gave a short signal then it is ignored, because that’s a sign of conflict. It removed some bad trades in my strategy (5s).
Looking forward to your update!
Hey Paul, thanks for your comment.
I tried to implement your different time management, but it involves unwanted trades before 9am.
Btw i’m happy you’re also developing a short timeframe system.. who knows that one day we’re gonna make them work properly 🙂
This will be my last version of this, it worked good through past two weeks but this one went particulary bad.
I don’t know if it’s a moment or it’s a failing system, but for sure i don’t think i will never put low tf systems on my live account; maybe they’re gold miner or maybe not, but there is too much instability despite the complete automation with the ml.
You are free to put it in demo forward if you like, to play with ml values to let it perform better, or put it into live account if you like to risk, but i’m just sharing this for knowledge purposes.
(If you want to play with it, i found that splitting the system into s/r bounce and s/r breakout versions can perform good, maybe running it together)
That’s a shame, I was really hoping the ML could save it, but as you know I came to the same conclusion about <1m TFs
Personally, I think 2m is the sweet spot. It’s what I’ve always used for day-trading and it just feels right to me. Allows for early entry and no long wait for the trail to start. 1 year back test is not great but at least it can be considered as “meaningful”.
Oh well, on to the next one. Have a great weekend.
