Methods of carrying out a rolling forward - page 10
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
When working with WF, it is only a question of resources, and a very strong one.
If we see that forwards suck over several sites (even 3-6 months), we stop the process and sort out the problems.
A good system has very few optimisable parameters (ideally 1-2).
But if you want to make an automatic WF-based wrap-around for anything and everything, it's a brick-and-mortar job, not a building one.
The key is in the simplified signal calculation scheme.
It doesn't help. Everything has already been simplified, the main thing is to remember to count the commission;) In general, I think it is right to wish success to each other
By the way I have in the input of about 140 000 000 + main "strategies", each having only 50-500 passes, and WF as the key to all this (+ a lot of tricks and other small things), but to calculate it is simply not realistic, even to save a report on each system is already problematic
Lack of resource is simply catastrophic, if we check WF on something specific then again we are chopping on the very idea of WF from the shoulder
If we see that forwards suck in several sections (even 3-6 months), we stop the process and solve the problems.
A good system has very few optimisable parameters (ideally 1-2).
But if you want to make an automatic WF-based wrap-around for everything and everything, it's a brick-and-mortar job, not a construction project.
So if I take a previously profitable system and watch WF change mytake profit level, is that cool? That would ruin the whole idea.
WF's idea is simple. You have a different idea - to feed in rubbish and get candy.
It doesn't help. Everything has already been simplified, the main thing is to remember to count the commission;) In general, I think it is right to wish success to each other
By the way I have in the input of about 140 000 000 + main "strategies", each having only some 50-500 passes, and WF as the key to all this (+ a lot of tricks and other small things), but just do not really count it, even save a report on each system is already problematic
Lack of resource is simply catastrophic, if we check WF on something specific, then again we cut off the very idea of WF from the shoulder
It doesn't help. Everything has already been simplified, the main thing is to remember to count the commission;) In general, I think it is right to wish success to each other
By the way I have in the input of about 140 000 000 + main "strategies", each having only some 50-500 passes, and WF as the key to all this (+ a lot of tricks and other small things), but just do not really count it, even save a report on each system is already problematic
Lack of resource is simply catastrophic, if we check WF on something specific, then again we are cutting off the very idea of WF from the shoulder
The idea of WF is simple. You have a different idea - to feed in rubbish and get candy.
What makes you think that you are better than a machine at figuring out what is rubbish and what is sweet - and if it is, then why trust the computer at all?
If the strategy has only 1-2 parameters - it is already predetermined, and it is just not suitable in principle for such decisions. If a strategy is not predefined (not designed for a specific situation), WF selects the set, which is profitable, and then checks it, if everything is ok, it is success and then starts over with another time step (if the strategy is predefined - then logically 70% of all passes will be profitable, so what for does WF need)
Everything you have written shows a complete misunderstanding of the issue.
You have a new way of doing things, invented by you.
If anything, here's a good phrase regarding WF:
WF testing allows us to develop a trading system while maintaining a reasonable 'degree of freedom'.
I will answer by Urain's words You have a new way, invented by you
I don't even want to argue. But making WF for single studies (individual values) is a dead end IMHO ;)
Although any way of testing gives exactly what is put in it and neither more nor less, and for the masses the other and not to realize, all as always.
I don't even want to argue. But making WF for single studies (individual values) is a dead end IMHO ;)
Although any way of testing gives exactly what is put in it and neither more nor less, and for the masses there is nothing else to be realized, everything is as usual.
Rather, I agree with this position. I tried to do "divide testing" when for different variants of code I singled out "responsible", as I thought, variables and optimized only them, saving time and compared forwards, turning a blind eye to the overall result - but it didn't work. Apparently in a good system its parts are interdependent and unbalance in one leads to failure of the whole system.
(That's why, by the way, there is no working system with a small number of variables. It is like an aeroplane - if at least one function does not adjust to the flight conditions, sooner or later it will crash. And though its function is outwardly simple, take-off and landing, the more complex the plane, the more factors it takes into account, the more reliable it is).
On the other hand, it is clear that some variables are responsible for the long-term "memory" of the system, some for the operational. So, in the long run, I have already figured out how to split them into groups with different optimization durations. But in the meantime the backing for all variables is the same - it's not right. But volking-forward is able to solve this problem too.
Another issue is that many things can be sacrificed to speed up the process. For example,
1. there's no sense in racing on ticks. Testing in VF is mostly relative in nature and super-accurate matching of environment conditions is not necessary to select the best option. I.e. a good solution is still a good solution.
2. Variables can be optimized one by one, one by one, not all of them. In this case the cloud goes Then the cloud goes its own way and you can more or less do without agents. That said, the quality of optimization is no different, I've checked. It's probably because during volkiness a separate fragment of history gets optimized for a certain variablemore than once. And uncaught interaction of variables in this point-situation always has a chance to manifest itself in the next step.
3. it is possible to catch unsuccessful solutions in the process and move on to the next task - here there is room for creativity (comparison with the benchmark, reaction to an obvious minus, etc.).