Stochastic resonance - page 14
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 wise Chinese say "it is foolish to look for a black cat in a dark room, especially where there is none". So, colleagues, I seem to have found a cat in a black room, at least it meows and demands a sausage. The side channel search algorithm developed for the goals in previous posts had to be discarded, replacing it with an old work in progress. I am sharing the first and possibly "extreme" results on this subject. The figure shows a scatter plot of the normalized lifetime (x-axis) and noise level (y-axis) of a 5000-sample plot investigated for a certain class of side channels:
All in all, everything is visible, the final conviction that there is a pattern came after subsequent correlation analysis in STATISTICA and MineSet for each major class. Yes, this is only preliminary data, requires detailed study and revision of my plans for free time, but it is possible that the great Chaos "joked". Let me remind you that normalised lifetime is the ratio of the number of counts a channel has lived to its original length, i.e. if a channel has a lifetime value of 2, it means that it has lived two more of its original length (note the x-axis). For the other classes of side channels the approximate appearance of the function is retained, only the noise intensity values change. The most interesting results are the noise intensities calculated (approximately) by the method suggested by Yury, but normalized to a standard deviation of amplitude, taken to some power. According to preliminary checks, the value of degree has a dependence on aggregated values of the channel. Yes and the degree Yuri wrote about is also important.
Thought for a long time about the dilemma, whether to publish detailed results or not. The decision is the following - I am not ready yet, so I am relieved of my duties as the "lead researcher". Roughly speaking, I am beginning, or rather, as usual, continuing with my individual programme.
AAB, rsi, Candid I feel a bit guilty towards you, for your willingness to work as a team and my opportunism. But don't misunderstand: Earlier I wrote about the model for stochastic resonance proposed by Mathemat- ik I have been using for a long time in my system, about the parameters of which I have managed to tell a little and I already know a lot about this model. Why, in fact, I grabbed this idea, because I had completely forgotten about the noise. To publish detailed results means to reveal important details, which took many years to find. If you want to continue research in this direction, I recommend rereading Yuri's posts and leafing through a physics textbook, namely the "waves" section (you probably don't need to, you remember everything anyway). Anyway, I am sort of here, always ready to help, especially if I do not know anything about it. :o)
PS: AAB, don't erase MathCAD, it's a good tool. If you master it you will get a huge time advantage, at least for research, over the most proficient MQL programmer. Yes and you may always need it.
Hooray! The result is there, congratulations . Good luck to you, about MathCAD thank you, but I'm already addicted to Matlab. When I decided to use a tool without connection to a certain DC and its products, I looked around and chose Matlab. I have looked around and chose Matlab, whether it is good or bad may show only with time and work with it.
Thank you, and good luck to you and good trends. I also use MathLab sometimes, but mostly MathCAD, which I'm really fond of. And I thought I'll make you an optional task to test one forecasting hypothesis (maybe somebody did it already) on basis of inputted instruments, but I can't get used to it :o)
PS: Here's another nice picture for reflection, and no errors. On the x-axis you plot consecutive samples of the investigated range, on the y-axis you plot the length of the found side channels on these samples. "Width of the squiggles" is determined by the tolerance of the list of criteria classifying the side channels.
NB: Candid, pay attention to this picture. Maybe it will help you find your own patterns on the issue you are discussing. You just have to consider a little more than you can see with the naked eye.
And I thought I would "charge you as an elective" to learn MathCAD to check one forecasting hypothesis (maybe it's already been done) on the basis of the tools, which I can't get to :o).
Leave it to . aab dog doma.net.ua
I thought I'd "give you an optional task" to master MathCAD and test a forecasting hypothesis (maybe someone did it already) on the basis of tools, which I can't get to :o).
Entrust . aab dog doma.net.ua
AAB, rsi, Candid feel a bit guilty towards you, for your willingness to work as a team and my opportunism...
AAB, rsi, Candid feel a bit guilty towards you, for your willingness to work as a team and my opportunism...
Thanks, good luck to us all.
The published needs to be checked more than once, as I wrote, the Great Chaos may have just played a funny joke on me. About the inspiration, I see it a little differently. Still convinced of the existence of stochastic resonance as a phenomenon in the market, but making an adequate model, I'm afraid, will never work. And here it's not even a question of complexity, after all we can get professional mathematicians involved, but of what we want out of it. In other words, "WHY?". Model a simple stochastic resonance system in any software product, MathCAD, MathLAB, MT, it doesn't matter at all. What is important is that in fact it will not have any predictive properties. And therefore I suppose the practical application of this direction is in development of a tool, probably more adequate, more precise, than numerous methods of calculation of volatility, which I do not like so much. Most likely this tool must according to a "tedious" algorithm search for the side channels (flat), hold them for some time and control the noise level, as soon as the level reaches the optimal value, then ... then it actually does not mean anything, except that one of the channel development limitations says about a high degree of a local trend appearance, and not just a trend, but comparable to the size of the spreading within the channel limits. Probably, it is possible to calculate the probability of occurrence of this local trend, but it is not possible to calculate "where" this trend will be directed, at least according to this model, but noise. In this case a wonderful joke comes to mind: after trading the nervous trader and the eternally calm analyst are going home, and in the lift the trader yells at the analyst "Now will you at least say up or down?
PS: I just forgot to add, as if it is not a trade secret, and my colleagues probably did not ask because of politeness. Anyway, if interested, I can share the view of the signal. The view is quite funny - complete absence of it, i.e. the price range was taken as noise.
NB: Candid, turn your attention to this picture. Perhaps it will help you find your own patterns on the issue you are discussing. You just have to consider a little more than you can see with the naked eye.
Generally speaking, it is a dangerous and thankless task to interpret the results without knowing in detail how they were obtained :). The picture is quite interesting. IMHO, it is absolutely necessary to understand "on the fingers" the reason of bifurcation of all channels, either it is something fundamental, or a mistake in the algorithm. To a question about the reason of actually one-step occurrence of channels: channels proceed only to the future or to the future and the past as well?
NB: Candid, pay attention to this picture. Maybe it will help you to find your own regularities in the problem you are discussing. You just need to take into account a bit more than can be seen with the naked eye.
Generally speaking, it is a dangerous and thankless task to interpret the results without knowing in detail how they were obtained :). The picture is quite interesting. IMHO, it is absolutely necessary to understand "on the fingers" the cause of bifurcation of all channels, either it is something fundamental, or a mistake in the algorithm. To the question about the reason of actually one-step emergence of channels: channels continue only to the future or both to the future and to the past?
In anticipation of this assumption, I specifically added to my post earlier in the day:
The channels only continue into the future (roughly speaking, marked the beginning of a flat), there is a variant with a full channel search (no difference). In terms of fundamentals, I can't say anything, as Yuri rightly wrote - apparently some tricky phenomenon, but it's there. Not all channels are bifurcated there, you may get a more complex pattern, it all depends on the depth of the search, and it's very small in the picture. Not all channels are there, you can see that if you tighten up the search conditions, there will be even fewer channels.
The purpose of this picture is not to look for errors where they don't exist but to remind (it was discussed on a neighboring forum) that you have to take into account channels structure: nesting and hierarchy for your task :o)
I've been reading here, I've been reading.... It's interesting. I come back to noise and signals from time to time myself.
I'll download Matcad for now (the channel is thin :)) maybe someone will try to use the following on the basis of theirs.
I have approached the problem in a peasant way: if I look through the price fluctuation spectrum on the history (I looked at M1 on GBPJPG) then the spikes are around 3, 5, 7 , 13, etc.
True, the amplitude and the peaks are a bit floating from day to day. I just plotted a sine wave with these periods and added them up. And it turns out (purely visually so far) that each movement
is a spike or a dip. Can I use this as a sub-threshold signal?