TSR - resuscitating trading systems - page 3
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
The method, on the other hand, is elementary:
A kind of diversification, which is not really it. But it is also a method. Why not?
The point here will be to choose the intervals to optimise. Because to divide it into equal parts is out of the question)) Let's say there are 2 global market phases now - within a specific phase, a specific strategy is optimal. But we don't have a filter for which strategy to apply now (i.e. which phase of the market now). Then the best solution would be to cross these two systems - even if the second (in the wrong phase) rejects some trades or vice versa, the result will still be positive.
In this case, the luck of the expirience can be explained by a fortuitous division of history. At other times it may not be so :) I.e. the important parameter here is to divide the story into chunks to fit - how to divide. Random partitioning can give random profits even on the right wall
A little bit about the different TCs, though:
The market is changing, that is obvious and there is no arguing with this fact.
However, I dare say the cause and effect relationship is evolving (say the market is changing) towards the future rather than the past.
....
PS I was not objecting to the method, but to the practice of using OOS DO Sample. (remark made for particularly gifted nerds, not Reshetov)
Did these connections come out of nowhere on Sample and go changing towards the future? Or they possibly have arisen before Sample and smoothly evolved there and therefore they will be for some time in this or that form and before Sample and after. ? Ok, let's forget about "before Sample", and take the following way: we take a period of 5 months, train EA for 1,2,4,5 months, this is Sample. The 3rd month is "Simple". Is it reasonable?
Our task is to find the patterns, not to fit the Expert Advisor with excessive degrees of freedom to the curve. OOS serves to help us distinguish one from the other. In fact, blindly both "before" and "after" are wrong. The learning period was UP-trend, at OOS even if "after" the DOWN-trend fell out. The SLO was a failure. But it will go up again, and the TS may show itself even better. In short, everyone is wrong according to the measure of their "special talent".
But yes it's just off topic, but on the topic as for me and for those who ask "how the method differs from the crossing of two different TS, fitted at different intervals? But I'm sure it may open "the whole world" for someone, as the world of neural networks was once opened for me, in fact, not at all by neural network advisor AI of the same author, sending everyone to Job)
Figar0 has suggested a good alternative, namely OOS between two Sample. I will try it. What about "before" or "after": sometimes results look better "before" and "after". The truth is somewhere in the middle.
I've tried both between and breaking up into (many and not so many, specially selected and random) plots, tested then on demo, real. The result is the same - it's all a lottery. Quite sparing and even acceptable. But the answer to whether this particular TS will be profitable or not, these methods do not provide.
As for your new Expert Advisor, I have not viewed it yet.
A little bit about the different TCs, though:
1. Yes.
2. Yes.
3. Not necessarily. The proposed method "finds it by itself". That is actually the point. You may think that it is just a way of finding it. And it is automatic.
4. It won't work. They are both unprofitable outside of OOS/.
5. The fifth is indeed desirable. But not to the point of fanaticism. In general, the essence of the idea is that the "abstracted true" component of optimized strategies will add up under such mutual filtering, while the "noise-adjustment" component will not add up. The result is an increase in "specific profitability".
A little bit about the different TCs, though:
1. I can tell you a terrible secret right away. One and the same TS will be most correlated and consistent in trading signals. And the hallmark of a false signal is the presence of inconsistency in different optimization results. Another thing is that a false signal can be consistent or inconsistent, so it is not a trivial task to eliminate them completely. And if we take different TS, for example, one that trades by trend, the second one counter-trends, we will get a fable by the name of "a swan, a crawfish and a pike" by grandfather Krylov.
2. The adjustment is not obligatory. I.e. if TS passes one way or another the OOS tests, then it is even better for this method and much more reliable. I took knowingly fitting TS only for the sake of example to demonstrate how to adequately filter signals even from shitty chases. Because even a fool may make profit without adjustment, i.e. this variant is not a good example. But now there is no more necessity in searching for a line between adjustment and its lack by way of empty forum flooding, because the regularities can be found even from adjusted results, what was publicly shown.
the point here will be to choose the intervals to optimise. .................
As for your new EA, haven't looked at it yet.
That's where you should have started: I haven't read it, but I've given my own opinion on it.
No. That's not true. That's not the point. And the choice of intervals is not particularly important. Although... the more random, the better. :)
can you prove it? or blah, blah, blah? :)
can you prove it? or blah, blah, blah? :)
Considering you were the first to make a statement, you're the one to prove it. You can start now.
I'm off for popcorn...
;)