Machine Learning in ProOrder ProRealTime
- This topic has 454 replies, 32 voices, and was last updated 2 years ago by
.
-
boxsize=atr[period](close)*constant ??
With period being decent number-100 or 200??
Constant could be linear or changes with the size of atr to give an exponential effect?
Hate to say this but I have to-but my experience-i don’t do intrabar systems, only inter bar systems. I’ve found I can’t trust single bar open/closes for accuracy. I could be wrong but I am now very cautious of a system with predominantly same bar open/closes. Better off seeing the market action in front of you, and not trusting how it works being hidden in a single candle. Not to detract from any of the great work here that is definently worth paying attention to.
1 user thanked author for this post.
It’s possible @nonetheless, but according to that tbt email I got from Anton (see my earlier post) there is enough data for daily. I also found out that the spread for Coffee KC is normally around 20 not 60.
60 was used in the following optimisations tests on the Dow, non of which threw up any tbt warnings.I used Dow 1 Hour, Daily and Monthly with that huge 60 point spread (from when we were testing the Coffee KC). It wasn’t intentional but I guess at least it’ll cover the worst market conditions and slippage. Pls see images.
I could not get the following to produce profits:
123456789101112//c1 = renkoMax + boxSize//c2 = renkoMin - boxSize// Conditions to enter long positionsIf close = renkoMax + boxSize thenBuy N CONTRACT at marketEndIf// Conditions to enter short positionsIf close = renkoMin - boxSize thenSellshort N CONTRACT at marketEndIfand have used the original buy and sell Stop conditions. Judging from these solid results for the Optimised BoxSize range of numbers, I’d say that if you put them into Excel and draw a scatter graph and pick a middle value from these very stable range of values they’d be safe to use on the Demo going forward.
I’ve attached the itf because it’d be interesting to see if they cause tbt warnings on the Daily charts for anyone? I expect they will. Particularly for the 2018-2020 range. Please let me know, (and if you see any errors too), I’d like to get to the bottom of this, (how can I put if tactfully?) “problematic” daily tbt warnings. Cheers.
1 user thanked author for this post.
Good morning,
Be careful before you write anything thinking about the account you have subscribed to your broker.
If it is a guaranteed stop account: what is written below will not work.123456once renkoMax = ROUND(close / boxSize) * boxSizeonce renkoMin = renkoMax - boxSizebuy at renkoMax + boxSize stop //Buy at Close + 40 (because renkomax = close).sellshort at renkoMin - boxSize stopand ptrailling
1 user thanked author for this post.
Attached is a Renko ML2: 2 x Machine Learning Algos: ValueX for the BoxSize and ValueY for the Trailing Stop.
I am not sure how much you can trust any of these results sometimes, because I have literally got quite different results using the exact same parameters (I screenshot everything as form of keeping track of data).
In the past I’ve tended to keep the BoxSizes and Trailing Stop Loss small between 10 and 150 but when you increase, eg the Trailing Stop range, like from 10 to 150 up to a new setting of 200 up to max 300 the profit might go down but drawdown has also gone down massively for some of my Dow backtests. I think ML1 is simpler and more profitable, I think the less parameters you have the more likely the system won’t suffer curve fitting. When I, (if I), get the Robustness Tester to be able to switch on an off using tradeon = 1, and without having to // all the code, I’ll do the VRT tests.
I’ve think I may have an answer to the Tick by Tick Warnings: Because of keeping track with screen-shoting, I think, the tbt warnings are at times being tripped by the settings and not because it’s a daily TF.
For example having used the larger Trailing Stop at 200 to max 300 and had no trouble backtesting Dow Daily 2015 to present, when I went back the smaller previous ranges (10 to 150 for both the Stop and the BoxSize) I immediately got a tbt warning. This happened with optimisations of the BoxSize when I used small starting values of 10 or 50 (to 300 in steps of 10) but not with 200 (to 300) but it didn’t happen when I replaced the ValueY with 10 (for a trailing stop loss)? When I went back and tried my ValueX = BoxSize (as opposed to Trailing Stop) system the only way I could get Dow Daily tbt warnings to stop was when I increased Starting Value to 200 (having tried 10, then 100 first). The Tick Mode was 579. However on the £/$ I got no tbt warning despite the Stop Loss starting at 10!?
The same thing happened when I changed the Ehler’s Universal Oscillator bandedge setting from 50 to 200, I got a tbt warning.
IG have confirmed there is enough daily tbt data: “Daily view: as much data as possible.” (Typically back to Aug 2010 and besides PRT will warn you when there isn’t enough data).Who knows why this is happening?
Bottom line though: If the curve looks too good to be true (i.e. perfect line from the bottom left to top right like a lot of the ValueX Trailing Stops test can show), then it likely is. Those Stops probably won’t happen as perfectly in live trading. Just a hunch unless anyone has live data?
1 user thanked author for this post.
Thinking about this version. Maybe it be possible separate boxes for long & short and add those to ml valuex/y?
1 user thanked author for this post.
Find exclusive trading pro-tools on 
