We need a second head, or even two, like the Garrynych kite. - page 9
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
1) I altered a lot of TS on my custom indicators, but one problem I cannot overcome, TS shows high efficiency, but on short intervals of history. I do not do optimization in principle and it is difficult to optimize my indicators, too many variable parameters, I created new and new TS according to different strategies trying to find a stable one in a wide range, I have not succeeded yet.
2) I decided to try to make the most primitive one on one machine, the working range is much wider, but the performance, compared to previous versions, is very low. Question in the studio: is there any prospect of refining such a TS, or according to the experience of experienced people such a TS has no chance?
I don't lose Faith and Hope and Love will come when I do!
:)
Love can come without it... Love is a spiritual thing, you know, not a material thing... :)
Maybe you're looking for love and not the grail after all. :) You'd better define your true goals first. :)
Both are lacking in adaptability, in my opinion. And it can take a lifetime to refine (which may turn out to be useless).
I support you if by "adaptivity" you mean enabling/disabling the signals depending on the number and profitability of results! Once there was an interesting topic, I do not remember the name of it, the man showed the effect of this very switch by the example of a strategy that neither earned nor lost money. He traded two types of lots: either 1.00 or 0.01. He virtually plotted the curve of profit in points and if the profit was above the curve - he traded with 1.00 lot, otherwise 0.01.
No, my point is different. You're talking about filtering, and it doesn't matter if it's filtering of what - signals or position size.
I'm talking about adaptability. The ability of TS to change. The adaptivity can be divided into two types:
a) change of parameters is continuous (continuity of course implies change of TS parameters at each bar, less discreteness is impossible to achieve because of the discrete nature of the signal itself)
b) change of parameters after a certain period of time.
The author obviously faces a choice - either to work on further detailing of TC parameters and try to "consider everything" (again, to consider nuances of the historical section only), or to try to make TC coarser (adepts of this approach think that the less parameters a TC has, the better). Both approaches have disadvantages: the first has low robustness, the second has low profitability. In general, I do not understand the desire of many people to test with constant lots and constant TP and SL. You know how it is called.
I emphasize that we're talking about adaptivity, not over-optimization (performing optimization after some time). Although even over-optimisation is better than nothing.
joo:
or try to coarsen TC (adherents of this approach believe that the less parameters a TC has, the better).
Reducing parameters is not always associated with "coarsening" the TS. Simply the parameter can be calculated adaptively instead of using some constant for all cases. Hence it is easy to understand that exactly the constant approach is more problematic since it will require new constants to fill in holes elsewhere.
Also it is easy to understand that such patching of all holes at code level at some interval of history in principle does not differ from fitting to history.
No, that's not what I mean. You're talking about filtering, and it doesn't matter whether it's filtering signals or position size.
I'm talking about adaptability. The ability of the TS to change. The adaptivity can be divided into two types:
a) continuous change of parameters (continuity of course implies a change of TS parameters at every bar, less discreteness is impossible to achieve due to the discrete nature of the signal itself)
b) change of parameters after a certain period of time.
It's clear that the author has to decide - either to work on further refinement of TS parameters and try to "consider everything" (again, to consider nuances of the history interval only), or to make TS coarser (followers of this approach think that the less parameters in the TS, the better). Both approaches have disadvantages: the first has low robustness, the second has low profitability. In general, I do not understand the desire of many people to test with constant lots and constant TP and SL. You know how it is called.
I emphasize that we're talking about adaptivity, not over-optimization (performing optimization after some time). Although even over-optimisation is better than nothing.
You make a good point, and that's the way I'm trying to go. I try to find the strategy, that allows to make adaptation of TS in a wider range, but so far my grey matter is not enough to fill all the gaps.
Primitive approaches do not provide the necessary efficiency, and the TC identified at the beginning of the topic shows this.
The creation of complex systems with a large number of logical units, which would be trained to recognise and process individual market situations, dramatically increases the effectiveness of TS, but the flipside of this model becomes the complexity of adjustment, and even more so the adaptation of these systems, because each unit essentially has its own parameters for adjustment. Creating more and more functional blocks it is still impossible to take into account the infinite variety of market situations, and if we make the blocks static, in the best case they will stop working when the situation changes, but often small changes in the market character do not disable them, but distort their work and they form false signals.
Attempts to solve the problem of adaptation in traditional ways, by the principle of strategy tester operation inside the TS itself, retraining it at certain intervals in virtual runs by changing the combination of parameters being tuned, is not possible due to a large number of optimization parameters reflecting the operation of each block, and even genetic algorithms will not help in this case. Besides, it is optimization with all its consequences, not pure adaptation, and optimization cannot reflect current changes on the market, it always lags behind at least by the time spent on the optimization; in addition, the resulting parameters are very averaged over the optimization range, which means they will not fit the optimal ones at any given time.
The way out of this impasse is the creation of feedback adaptation systems, such as in oscillators, where feedback stabilises the frequency and phase of the signal, or in amplifiers, where negative feedback equalises the amplitude-frequency response in a given range. Although the signal within this range may vary widely in both amplitude and frequency, just one node of operation keeps the system within the specified limits, regardless of the nature of the changes in the signal.
I have been struggling with this problem for a long time and intuitively feel that there is a solution, I have been going round in circles around it, but I cannot get hold of any reference point, as Archimedes said, "Give me a fulcrum and I will overturn the earth", so I cannot find any such fulcrum. I have therefore raised this topic, it is difficult to find a solution to complex TS, so I propose to start at the most primitive one on one machine.
The way out of this impasse is to create feedback adaptation systems, such as in oscillators, where feedback stabilises the frequency and phase of the signal, or in amplifiers, where negative feedback equalises the amplitude-frequency response within a given range. Although the signal within this range may vary widely in both amplitude and frequency, just one functional unit keeps the system within the specified limits regardless of the nature of the signal variations.
Crap, at the most interesting point...didn't get it... :( people, what's in the paragraph, in a nutshell...
Angela is probably referring to a corrective RC chain in the feedback (if you take the amplifier). But, gentlemen, there are no such simple solutions on the market. There is a very simple thing, which for some reason is difficult for most people to understand: there is a cause - there is a consequence. Trying to anticipate (adapt to the cause???)) will get you nowhere. The only thing left for us to do is to react as quickly as possible to the changing context - not an easy task, but, unlike trying to anticipate the price, solvable.
It's not like I was talking about anticipation. It was about adaptability. In your terminology - "react as quickly as possible to the changing context". That's what I meant anyway, I don't know if that's what Angela meant.
Certainly, Andrew. Following, not anticipating. But, you must agree, it's not unreasonable to emphasise it once again. With persistent persistence people try precisely to guess, rather than to understand what is happening at a particular moment. Hence the constant nonsense about TA not working and other nonsense.
So... I was only answering the question about the OOS.))