Adaptive digital filters - page 5
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
to Northwind
I understood everything about zig-zag, really simple, thank you. Regarding the link, my approach was much more modest, I did not think about the mood of the change, and approached the task as described in Schwagger's parable - BETTER. Buy cheap - sell expensive. And this is effective mainly at extremes :o). A few more words about an old strategy. For its implementation you only need to:
The rest is simple, at the moment when an extremum of the current wave A is fixed and wave B begins to form, we can get its statistical estimate of the development. The notion of an extremum is used similarly, i.e., to fix the extremum, we need to wait for one more complete countdown. Here everything is tricky, and a kind of "memory" starts working. It is very similar to an experiment, when only one cartridge is inserted into a revolver, the cylinder is unwound and the trigger is pulled (after each pull, the cylinder is not unwound again, i.e. In other words, I know that there will be a maximum or minimum (depending on what was before), I know below what level the new extremum will not rise/fall, I know how many counts it passes (I pull the trigger) - as a result, I can forecast a new wave (for skeptics +1 count greatly changes the probabilistic state of the system). All this nonsense begins to work very well only for certain classes of waves, and they do not occur that often.
Plus an alternative forecast based on the phenomenological dependence of parameter X (which I wrote about in a neighboring thread) and some additional method as a check. Getting the characteristics of the predicted wave by the parameters of the current one, I can immediately see its "place in the statistics" and draw some conclusions.
Besides, I found out that "memory" of the waves obtained by LPF filtering is 3-5 at the most, and the prediction depth is three waves maximum:
What am I talking about? Oh yes, all this actually works, but the main problem is that LPF has a huge number of input parameters and their small change leads to "shifts" in static data and empirical formulas. It's not a fact that you've found the best/optimal ones. And so it is the adaptive filter that is needed to be completely happy. If it is well designed, coefficients at the very beginning can be set to zero - it should equalize them by itself. Roughly speaking, this is the situation with this idea.
PS: In general I'm ready to take part in its development (if it is interesting of course), I have something to share.
PS: Prival, I'm a bit confused, were you deprived of a bonus for writing the word "radar" on the traders' forum?
I'm out of 13, for about this. Look at your post attached file name dsp11. The number 11 I discard, and write in russkim dsp. Better yet capital letters DSP(For Official Use), but he got caught creep put classified information + something about the military academy spoke about + apparently engaged in science. Yes even a creep and does not lay out their works, some Tikhonov VI refers to and his work posted. He referred to Tikhonov's work and posted it.
It's just that there are such freelance FSB officers, one of them caught my posts, so he ratted them out. And the melon came to the academy already from above. And I proved that the files that I posted it all on the Web lying + the book is for free sale + what I wrote there is nothing secret, and everything is known, you just need to be able to read. I proved it, but what is the point? We are forbidden to use the Web :-(. But they have to report to the top, an internal investigation has been held, so it's a full-time job, measures have been taken, fools have been punished, etc.. I even think they will get a bonus for their hard work. And that "vigilante" will also get his 30 silver coins. And all out of my 13th bonus. Damn them to hell.
That's it gentlemen and ladies :-(.
NorthernWind
I know a very good filter. It even has a name Kalman filter, I really want to run it. I like it very much, but it takes a long time to kick it until it works :-) And kick it right :-) + when it starts all the same there are pitfalls that you need to know how to circumvent. I know this filter and I have often used it in real research, but I cannot implement Forex into it.
grasn
beautiful, clarify some things I don't understand.
1. What do you mean by "+1 count", what is a count in your terminology. (1 tick or something else).
2. ""memory" of the waves obtained by LPF filtering is no more than 3-5". what 3-5 periods ?
3. "The LPF has a huge number of input parameters" please list the inputs.
4. and I think the most important thing is what the filter should be adaptive to? Can you make it a little simpler, at least what characteristics should it have?
to Prival
I am not particularly surprised. I work on several projects with secret state agencies, but it seemed to me that people do not reach this level, but no, there are exceptions. What is unclear is how the ferret got wind of the mql Web site? Ah, most likely traced via links if from work. What an attentive FSB team you have, even empathetic ones.
Don't get upset, just to add that it's not the 1930s...
Preferably start reading here: 'Stochastic Resonance' (post grasn 28.10.2007 13:26)
one count is roughly one bar of some kind. I use a clock, for my case 1 count is one hour. In this context I meant the number of counts (bars) passed since the extremum fixation.
I meant the wave itself, as if measuring them in units. I searched for regularities using the Data Mining technique having collected all the statistics on the waves. It turned out that at the moment when the extremum is fixed (the previous wave has finished), there is a possibility to predict maximum three future waves (B, C, D in the picture). They are influenced by 3 to 5 previous waves. Let me remind you, a wave is a piece of BP.
By analogy: one can suggest by the ACF how many counts BP can be predicted. Similar here, but only how many waves in total can be statistically predicted ahead, based on waves past.
My LPF has a number of parameters of five:
that's a lot to find the optimum
Exactly for these reasons I could not complete my AF. To complete this model you need an AF with one parameter, or we'll gather the statistics in the next life.
Bars are a tradition based on opportunity. In fact, the technology to do away with bars and move to other forms of presentation is not very long ago. The count goes back years, so let's not be too harsh on them.
And to all concerned about noise in the market (heh, I just do not know where to throw this information). In the attached file there is a link (not on the WWW sense) to the work of De Meyer B., Moussa Saley H. On the Strategic Origin of Brownian Motion in Finance. - Internat. J. Game Theory, 2002, v. 31, p. 285-319.
So "" ... believe that the Brownian component in the evolution of prices at financial markets is caused by different awareness of market participants of events affecting prices. "" .
Is the practical application of this assumption possible ? Good luck.
2 grasn
I don't even know what to say. The thing is, VLF and Waves are so far away from what I do, that I can't make any recommendations. I can only wish you good luck.
SZZ, on the alpari forum BQQ somewhere laid out why in his opinion, as an expert in DSP, DSP methods are difficult to use in the market. Quite clearly, in my opinion.
2 Prival
I have heard about Kalman's filter but never dealt with it in detail. It seems to be statistically optimal, it certainly inspires some optimism, on the other hand how many of them already were, great technologies from other areas of science, which do not work on the market. One has to watch.
2 AAB
This is the emergence of information technology. Plus there are topics, you can search by words, "market own time", "operating time", etc.
to North Wind.
Similarly, I am working on another model at the moment. Its pluses are good statistical advantage in predicting price levels and complete absence of any parameters whatsoever. But on the other hand it has high drawdowns and difficulties in calculation of stop levels. And the considered model, I so suspect, should have its pluses, because it actually predicts the trajectory of movement, and this is interesting to me in combination with the basic model. Anyway, I'll get to it sometime. Special thanks for the wish.
PS: If it is not a trade secret, tell me what you work on, at least conceptually. I am very interested :o)
I came to the same conclusions myself and therefore I use only LPF for wave extraction - no more (no worse than a zigzag). I don't have any illusions about using DSP. :о)
to North Wind
PS: If it's not a trade secret, tell us what you're working on, at least conceptually. Very interesting :o)
In general I can't tell much, only in general. First: the market, it's not Forex DTs, what is it? - If your vast majority of the market is not Forex brokerage companies that you are used to, you may find some classical forex brokerage systems that will suit your own temperament. The main thing is that it must be a growing market with lots of opportunities. Secondly, no complicated mathematical methods, in principle, and no unnecessary theoretical calculations, everything should be subordinated to one goal - profit and fast. Third: we need a simple and rather consistent concept of market life, upon which everything else is based. Fourth: as part of the concept, the market is all about general trends, and everything revolves around these general trends. Fifth: The basis of the approaches is the task of making the process work, given that the process will then work out again, within the framework of the general tendencies. Sixth: So far playing on the market has not been able to replace my main profession. That's how it is, in general terms.
Integer, you mean this JMA - 'JMA'?