Methods of carrying out a rolling forward - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
In my opinion, the coolest way to implement WF is to make strategies in the form of indicators.
The base indicator gives universal opening/closing signals. Another universal indicator builds a balance line using the basic indicator.
Everything else can be easily attached on top. Moreover, the processing speed will be one order (ki) faster than that of the tester.
I must immediately say that we are talking only about strategies that enter the market by signal, not using a zero bar for analysis, without grids, martins and the like.
True, there are questions about loading/unloading
Disagree all you like, do as you please, this thread is about Walk-Forward, so if you don't want to talk about it, let me ask you from here
Let me disagree. Optimisation over the entire 43 year range is to determine, once and for all, the parameters of the TS "on average". Then, the TC is run 43 times each year. It is assumed that, the TC is bound to encounter abnormal market conditions. Your misconception is that, on the history you cannot meet the cases that will be met in the future. Since the TS copes with all instances of history, I am sure it will cope with the future, with only minor variations.
By defining the TC parameters on the whole history, you have deprived your TC of the chance to meet an unoptimised future. Your TC has already once processed all areas, including abnormal normal etc., which has reflected in its settings The point of volking is that we are simulating the onset of the future on past material. I.e. what you are doing is not volkking forwards, but something else. Volkking works like this.
Forward is usually WAY worse than backing and there are fewer deals on it.
By defining the TC parameters on the whole history, you have deprived your TC of the chance to meet an unoptimised future. Your TC has already once processed all areas, including abnormal normal etc., which has reflected in its settings The point of volking is that we are simulating the onset of the future on past material. I.e. what you are doing is not volkking forwards, but something else. Volkking works like this.
In the tester pictures it looks like this.
Forward is usually WAY worse than backing and there are fewer trades on it.
This is a matter of taste, TC logic and the researcher's views on the problem of TC optimization. What is detrimental for any TS? Right, a large drawdown or maximum drawdown. When you look at the whole history, you put in the TS to work in the future the maximum drawdown, which ever was on the history. A small backtest will not reveal it. Any unfavourable part of the history can become a prototype of the situation in the future. True, in my case the effectiveness of the TS will be significantly lower for the sake of reliability than the continuous search for variants according to the scheme laid out by you.
The point is that for a wolfing forward, the principle I outlined must be followed. As for the specifics - i.e. back-to-forward ratio, history duration, optimization criteria etc. - of course everyone does it as he/she sees fit. -Of course, everyone does it as he sees fit.
In my opinion, the coolest way to implement WF is to make strategies in the form of indicators.
The base indicator gives universal opening/closing signals. Another universal indicator builds a balance line using the basic indicator.
Everything else can be easily attached on top. Moreover, the processing speed will be one order (ki) faster than that of the tester.
I must immediately say that we are talking only about strategies that enter the market by signal, not using a zero bar for analysis, without nets, martins and the like.
True, there are questions about loading/unloading
Could you give us more details, step by step?
Igor Volodin:
On the next iteration the EA is allowed to trade M days (on the OOS of the first leg), this trading result we remember.Описал свою схему работы Walk Forward с использованием штатного тестера тут.
I've described my Walk Forward scheme using an in-house tester here.
Making your own genetic algorithm? That's a lot of work that's already done in the in-house tester. I think Metacquotes spent more than a hundred hours developing it.
I think it's easier to run genetic optimization in the tester 10-20 times with a third-party program. But Youri Tarshecki has some kind of auto-optimizer.
And it's easy to run it manually 20 times, which is what I do.
For example, the most important and most unclear to me is the criterion of selecting one of the optimization variants (winner set) to be launched. After all, it is what will influence the success of actual trading.
4 years ago they started digging into volking trading..... and very hard.
And I have questions for you, what do you want from volking? To find out if the system works so you don't have to test it on a demo?
It's cool, but you should try it on demo anyway. If you hope that the proto system will work with volking, it is not so, he often says that everything is wrong during the extensive testing.
We are giving a huge sample to volking that just can't be reduced (to set direction), otherwise the whole principle of volking will break on the fly. And the reason why 80% of all calculations will be wasted is because of peculiarities of working with agents. I.e. when from first three days we understand that the result will be worse than the one we have we will continue testing until the end.
Do you have any idea how generalized the strategy for volking trading must be? - There will be more parameters to optimize than when you manually try to optimize passes predetermining profitable ones - otherwise everything