Stochastic resonance - page 28
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 mission trip was put on hold for a while. People have honestly admitted that there are some holidays coming up, and after them will come an off-duty state.
to Yurixx
∫ (ξ^a)*exp(–B*(ξ^b)) dξ = –1/(B*b) * (X^(a–b+1))*exp(–B*(X^b)) + (a–b+1)/ (B*b) *∫ (ξ^(a–b))*exp(–B*(ξ^b)) dξ
Not a very big expert in mathematical analysis, frankly speaking not at all, but I couldn't take this integral, neither MathCAD, nor "Mathematics". But maybe I misunderstood something or misread an expression.
In my opinion, this is the simpler of the two. A small part of code that calculates the normalization coefficients depending on t/f and averaging parameters is embedded into init() of an indicator or an Expert Advisor. How can you do it in the Championship if the Expert Advisor is not on your PC and the history is loaded there in the unknown volume?
But those are trivial matters. The issue is more serious. Do you have to recalculate these ratios every time you change a symbol, t/f, etc., either by hand or using Matkad :-)? ? Don't you get tired of it? Or create a database for all symbols, t/f, smoothing parameters etc ? ? :-)
There's nothing difficult about it, you only need to recalculate once for the current quote environment and that's enough, the calculation time will take no more than a gradient descent.
It's even more ridiculous to simply declare that the distribution is like that, without any proof. Yes, you're right, the real quotes have a completely different distribution than the one used as the basis of your proof.
No I haven't. But I noticed long ago that the rationing problem has been solved in many ways for a long time, for example I use a method described in the statistical section "planning of experiment". At least I have no problem with rationing. In addition, all the listed pros and applications are highly questionable:
4. In neural networks, which I know nothing about, it is necessary to normalize the data. Going beyond the conditional range causes neuro-brains to lose their neuro-coverage.
Perhaps this way of normalisation will prove to be more useful in some cases than what is currently used.
In neural networks in which I probably still "do not understand enough" there is no problem of rationing (where did you get this problem?), moreover, they require exactly metrological accuracy of this operation, and it is simply impossible to get it by your way.
Suggestion:
OK, let's do it this way. You tell me what input parameters are needed for the calculation. I choose an arbitrary series, calculate the required parameters and tell you. After that, we agree on the value of an arbitrary sliding window. You publish the calculated values of min and max for this window. In my turn, I will publish the exact values and we will compare them.
Honesty guaranteed, you can not doubt.
PS:
The programmer is not the one who programs everything that comes to hand.
Just like a man is not the one who drinks everything that burns and eats... everything that moves.
Yuri, what is the reason for that? As a consequence of the proof? Have you decided to confirm it, like Ales did? Though, knowing you, it is doubtful...but nevertheless
To Neutron
Thanks Seryoga, I will have a look.
Yuri, what is this written for? As a corollary to the evidence? Have you decided to assert yourself like Ales? Though, knowing you, it is doubtful...but nevertheless
There's some aggression in your tone, Sergei. I don`t think you should view what I`ve published through the prism of this condition. You don't like my suggestions? Don't take them into your practice. Did you not understand something in my explanations? But that's no reason to get annoyed. I don't understand a lot of your methods and I don't have any problem with it. You don't want it? Fine, don't worry about it.
Have you found any mistakes? Then point them out. From your whole answer I saw only one place which can be called an objection:
You probably aren't familiar with the way mathematics works. It postulates objects and then examines them. It's the same here. I originally postulated a type of distribution function and then refined it to suit my needs. And didn't you notice the phrase that the curve of the model distribution function corresponds perfectly to the experimental one ? What more proof do you want?
Your phrase "real quotes it's quite different" implies that you know which one ? Then give it here. What do quotes have to do with it? Who told you that I work with quotes? From my writing it follows that I work with data of an indicator, which I need to normalize. Unfortunately, you don't know anything about the distribution of values for this indicator. So what are you objecting to?
In general, there are two types of real objections: 1. The incorrectness of theoretical construction. 2. The incorrectness of the application. On the first point, as far as I understand, you do not have any valid arguments. And the second point is an elastic one. It is difficult for you to judge the application to my problem. And the application to your tasks is your business. If you do not need such features (and few people do, it is a very particular case), then you should not pay so much attention.
By the way, about neural nets. When we feed data to the grid input, we need to clearly specify its range. This is impossible for price and most indicators. This is the reason for my comment. May be it is wrong but it is only an opinion about the possible use of the method.
Sergey, you are able to do a lot of things that I am not. But it is not a reason for me to take a dislike to it. On the contrary, it is a reason for the greater respect.
could not take this integral
The expression is absolutely correct, I just checked it myself (it's not the gradient descent method though, just the method of integration by parts). It is sufficient to simply put the exponent under the differential, compensating for this operation by the degree of the independent variable before the differential.
The problem with the other is that, as correctly noted here, the distribution is arbitrarily chosen. But I would still look at the results of what Yurixx suggested for a practical test of these formulas. After all, the main criterion of truth is practice, even if the final conclusions are not entirely correct.
I would still like to clarify what is the X-value in your problem,Yurixx...
IMHO, for price, a single distribution is of no value. At different points in time, different distributions. For example, the same channel is a temporarily stable distribution of a certain kind. At transitional moments, it is impossible to determine which one it will end up in, there are always alternative scenarios. We enter just counting on some particular distribution with positive MO (or use it so as to get +MO). Of course we can mix them all up for arbitrary range and pick similar distribution, and then use as you suggest: reduction to universal standard, normalization... But the purpose of indicators and other tools is to detect the moments when a particular distribution may occur (or continue), which we can use, but not to predict some global market condition. For this purpose an indicator shouldn't mix several previous distributions, and it's not clear with what period: we took some from this distribution, some from the other one and some from the third one. IMHO, there is a need for separation, perhaps even post factum, but not mixing. And then either use the revealed distribution in the hope that it is still preserved, or wait for the new one at transient moments and use it. In the latter case, the previous (completed) distribution may also determine the potential for the future to be used.
The first two sentences articulate a concept which is fundamentally different from mine. I, on the contrary, believe that the market is one and consequently its distribution is a separate phenomenon, although in some periods this distribution is dominated by different (but its own) aspects.
To be clear, the proposed distribution function is a special case. It suits me fine for my problem, but it does not have to suit everyone at all. I wrote at the end about the difficulties of application. In point 3 it says "investigate the statistics of the series". What do you think it is for ? I think to find out if this distribution function is suitable or not.
And about the "universal standard" there is also a discrepancy in terms. You talk about distribution, I talk about standardization of a range of values (just) for which normalization is the easiest and most effective way.
Your thoughts about the purpose of indicators and all that follows are certain notions that gradually turn into vague fantasies. I dare not challenge them. Unfortunately, it has nothing to do with the topic of the paper, the ideas expressed, application considerations and the rest. So I'd love to respond to those statements, but I don't know how to relate it to my writing. :-)
grasn
Could you please provide a picture (histogram) of the distribution under study. There is also a distribution defined on the interval 0 to ... . http://avs.cde.spbstu.ru/str/HTML/pag/1/23.htm
Maybe it's a good one for you. Look, it's the last one in this article. I'll get home. I'll find a program in Matkad where I worked with data, check them for compliance with this law by various criteria chi-square, Neuman Pearson. I'll send it to you if you need it.
could not take this integral
The expression is absolutely correct, I just checked it myself (it's not the gradient descent method though, just the method of integration by parts). It is sufficient to simply put the exponent under the differential, compensating for this operation by the degree of the independent variable before the differential.
The problem with the other is that, as correctly noted here, the distribution is arbitrarily chosen. But I would still look behind the results of what Yurixx suggested on practical testing of these formulas. After all, the main criterion of truth is practice, even if the final conclusions are not entirely correct.
I would still like to clarify what is the X-value in your problem,Yuurixx...
I can't believe it ....
And who said that this integral is taken by gradient descent method ? I wrote clearly: "we take it piece by piece". What place, if not a secret, do you guys read. :-)))
As for the rest, your post, Mathemat, is very optimistic. It's already clear that you're better than " both MathCAD and Mathemat." Significantly better ! It's not for nothing that I'm sceptical about these contrivances. You can't argue with them, like you ...
The X value in my problem is a set of values of some indicator.
My own practice showed that this method solved my problem. I am satisfied with the results.
Your thoughts on the purpose of the indicators and all that follows are certain perceptions that gradually devolve into vague fantasies. I dare not challenge them. Unfortunately, it has nothing to do with the theme of the work, with the ideas expressed, with the considerations of application and everything else. So I'd love to respond to those statements, but I don't know how to relate it to my writing. :-)
Before we do anything with indicators or anything else, we need to define what we expect from them. This will determine which problems need to be solved within the framework of the task at hand and which are farfetched.
"TA indicators are virtually independent of t/f, as far as I understand it, this is a property of their statistics. But if the problem of smoothing could be solved, then maybe something new could be gained from them."
What new things do you want from the indicators, for example?
Your thoughts on the purpose of the indicators and all that follows are certain perceptions that gradually devolve into vague fantasies. I dare not challenge them. Unfortunately, it has nothing to do with the theme of the work, with the ideas expressed, with the considerations of application and everything else. So I'd love to respond to those statements, but I don't know how to relate it to my writing. :-)
Before we do anything with indicators or anything else, we need to define what we expect from them. This will determine which problems need to be addressed within the task at hand and which are farfetched.
"TA indicators are virtually independent of t/f, as far as I understand it, this is a property of their statistics. But if the problem of smoothing could be solved, then maybe something new could be gained from them."
What new do you want to get from indicators for example?
When it comes to standard TA indicators, not much. But it deserves attention too. I have already posted the pictures. There are RSI charts for two different periods and my comments on it. If to normalize RSI so that its magnitude does not depend on the smoothing period, it may be used more effectively. The same applies to some other indicators.
It's OK, Yurixx, I'm not attributing that phrase to you. As for being skeptical about the gadgetry... I have Maple installed at home, sometimes it really helps me, including symbolic computations. However, I haven't used it for a long time.
to Yurixx
Yury, I apologize if I offended you in any way, I didn't have any such thoughts, there's no animosity, on the contrary, I see an intelligent, interesting and patient interlocutor.
On the first point I stumbled on the integral, I wrote about it, therefore, further just take my word for it, but it is, of course, my problem, take this integral and check the entire proof. The second point is what caused the reaction, as you think, aggressive. On duty, worked on a project where the customer required optimizing their inventory for main production based on the probability of failure of components (units, assemblies). They are quite expensive and the task was academic. We had an entire research institute working with us, which undertook the theoretical justification. The result was an excellent theoretical model, but it did not work at all in practice, all because the actual distribution turned out to be just a little bit different.
About the quotes I didn't assume, but read your posts. Here's one of them, a bit taken out of context: "...The original series is prices. It must have a non-normal distribution. I wrote about normal, because many things can be calculated analytically for it and because the real distribution can be approximated with a certain accuracy by normal.... " That was my further understanding that X is a price series. If the series X is generally something abstract, but subject to a given distribution, then there is no problem. Are you sure, for example, that a stochastic obeys an "assigned" distribution?
Apparently my stupidity does not allow me to realize the depth of this NOT problem, as well as in other points.
I am amazed ....
And who says that this integral is taken by the gradient descent method ? I wrote clearly: "take it piece by piece". Which place, if not a secret, do you guys read. :-)))
I don't think I wrote that I tried to take the integral by the gradient method,
I think you're right, as always :o)
to Mathemat
Thanks for the explanation, I just wrote it wrong originally, got confused in brackets and mixed up the integration limits :o.
to Prival
Sorry, I don't really understand what for and why, i.e. what's the point of your request?
to Neutron
It is an interesting indicator, I will be able to study it in details in the evening, now I'm at work and can only theorize. Here it is: 'Stochastic Resonance';;;; I have published the first material on the patterns found. But the subtlety is that it appears very well only on some classes of channels, on others - complete Chaos. These channels, artistically speaking, can be counted on the fingers. I wrote below that the results were obtained with no signal at all, i.e. the price range was taken as noise.