If there *is* a difference, then something is wrong with the approach in the backtest program vs what would be reality.
What you describe is a traditional walk forward test, which the BT engine can do automatically for you. This is better than nothing, but a BT cannot replicate all real trading conditions. You can see this for yourself: if you have a couple of months of demo data, try running a backtest over the same period and you will usually get a different result. The demo data is more accurate due to many factors that BT cannot account for.
As for Grahal’s theory, I think he’s prob right – walk back is better than walk forward – but in my book, forward testing in demo is the only meaningful oos test.
For what it’s worth, I have had this on demo for almost a month with v good results; personally I like to see at least 3 months before going live.
Think of the attached algo as partly- baked bread, you have to finish cooking it at home (in demo) !
Yes … we miss out on optimising for the market cycles / price action existing today, last week , last month etc.
Thank you, GraHal. Yes, I would agree with that. It will be closer to the reality of today, although still no guarantees for tomorrow.
Far better to leave an unoptimised period at the beginning of backtest (10k bars, 50k bars, whatever)
That is indeed what I do always. I do it this way :
- Optimise what you want in the last 200K bars.
- When you think you did right (not over-optimise) then draw in the 1M bars from historical data. Do NOT optimize that at all.
- When the result over 1M approaches 5 times the result over the 200K bars, go for it. When far less (up to losses) come form it, you know you over-optimised the 200K bars.
The 200K is arbitrary of course, but obviously resembles the amount available outside of historical data.
As Always a very interesting strategy from nonetheless, thanks for sharing!
I started it in demo when it was posted but it has now been stopped 3 times “This system was stopped because one of its orders was rejected more than the allowed number of retries.”
Is it working for everyone else?
For me it is working properly without any error since 9th sept2021, in a loosing row though.
Are you running it in Demo Ryugin ?
Yes, it keeps working properly for me in demo account without changing anything of the code.
It works … but does it win too? Can you post the latest results please?
Sorry, what I posted before is not correct. I’ve running this algo on a mini real account I usually use for live test and the algo have been active since 2nd sept. I am attaching the results obtained (0,5 batch size).
@Ryugin
Do not embed pics and files in your post, as this slows down the loading of pages and the forum overall.
Attaching them is enough.
Thank you 🙂
Does anyone have this strategy in demo or live? And what are the results?
I have a new version on demo, I’ll post the results at the end of november.
How does ur results look like nonetheless
It’s a bust I’m afraid, not something I intend to pursue.
I think if you move the timeframes up so that the smallest timeframe is M1 or something. It could look better.
