a completely random process and FOREX. - page 7

New comment
 
Mathemat:
It depends on how one looks for these Fibs. If in the same way as Swannell, i.e. analyzing only waves of the same order, then one really cannot see any special "resonances" there: smooth p.d.f. without any prominent convexities. And if one searches through Fib clusters from waves of different orders, something might come out. I haven't found one yet :)
If you remember, in the thread https://forum.mql4.com/ru/9325/page3#50924, post 15.11.2007 16:25, I wrote about one of the approaches, a kind of coarse dipstick. I think there was a faint indication there of a 50% level being highlighted. But it can't be linked to Fibo specifically, by scaling we just get degrees of twos (including Murray octaves). Yes and the evidence is not too confident.
 
I don't yet have any clear idea how to check the significance of these Fibs. Either by a wavelet transform or some other way. The situation is naturally exacerbated by the apparent fractality of the plots (in the sense of self-similarity). I haven't heard of any such test anywhere. Swannell's attempt doesn't seem convincing at all, as it only covers one layer of the whole phenomenon. At the same time, if one manages to do something like that, it will be almost the first clear indication of the apparent non-randomness of the quotes: the generator made by D.Will shows Fibs only by chance.
 
Here is what I think on the subject.
Of course Forex market cannot be called chaotic, almost everything here is subject to human laws (economics + politics), but I think that the form of quotations has an absolutely random character, or is formed in a similar way. On this issue I personally found it useful to read the link: http://monetarism.ru/search.pl?topic=6 provided by Zhunko.
I.e. the global movement is defined by funds, central banks and largest banks of the world, of course competing with each other (like in war), but the fluctuations inside the movement are more chaotic, because of the larger number of participants (market makers, banks, large investors) and the conclusion is even less coherence and the desire not to burn and earn (the risks are higher than in Big Uncles)... Going even lower (brokerage companies) we get almost a random character of movements ...
P.S. The conclusion for myself, I made not so long ago, but it's already bringing profit, only work in a big trend ( at least a day).
P.p.s. Regarding Fibo levels and others... they work with the same probability as the normal averages (only because of the greater complexity many seem to be the panacea), and as my esteemed KimIV said: It's hard to understand simplicity. Man has a tendency to complicate things. He doesn't want to believe that a simple thing works and wants to spoil it by making it more complicated. I use very simple algorithms. Sometimes, a buy or sell signal of an EA is less than a dozen short lines of code. But I have been working on these lines for over a year. This is the symbiosis of simplicity and complexity!
 

As for how to make quotations not take negative values, you just need to generate % increments instead of absolute ones.

As for the fact that the generator has a limited memory and repeats cyclically, it depends on the generator. There are some that are shifted by timer, others depending on CPU load, etc., etc.

It is the brain's skill to see levels and trend lines and it will find them on any data. When you have a hammer in your hand, everything seems like a nail. You have to check and understand what you want to use in trading, then similarity is of no importance :)

You can encode and represent as a similar chart anything: a novel "War and Peace", a digital photo, a favorite song, etc. Everything will be very similar and again the person wishing will find levels and what he has learned to distinguish, distributions of increments will be by the same functions (so coded :)), however it will not be the same and if you want you can restore the original.

 
Anyway, the constructed process is
1st order linear regression process.

AR(1)

y(n+1)=y(n)+e(n). where e(n) is normal noise with m.o. and std.
 
D.Will писал (а):
Anyway, the process we've built is
is a 1st order linear regression process.

AR(1)

y(n+1)=y(n)+e(n). where e(n) is normal noise with m.o. and std.

It's understandable.

But ichmo, you have to start at the other end. Prove that this phrase of yours is true "expectation 0. variance 0.0077. these parameters are similar to the real eurusd.
(See the first post) . A rigorous mathematical proof is needed. Proof which is very similar is not exactly something on which to base any conclusions

 
IMHO, there's no need to reinvent the wheel. There is a wonderful thing called G.A.R.C.H. in MATLAB. Toolbox - just a tool for studying financial time series. See e.g. here: http://www.mathworks.com/access/helpdesk/help/toolbox/garch/
 
Prival:
D.Will wrote (a):

In general. the constructed process is

is a 1st order linear regression process.



AR(1)



y(n+1)=y(n)+e(n). where e(n) is normal noise with m.o. and std.




That's understandable.



But ichmo, you have to start at the other end. Prove that this phrase of yours is true "expectation 0. variance 0.0077. these parameters are similar to the real eurusd.

(See the first post) . A rigorous mathematical proof is needed. Proof which is very similar is not exactly something on which to base any conclusions




Proof of what?
Parameters 0 and 0.0077 are taken from 1D EurUsd. for 2002-2004.

I don't know if I should explain. that generating AR(1) with parameters e(0,0.0077). showed those very pictures.
Clearly, it is stationary and ergodic, unlike the real market. (non-stationary and not ergodic).

With AR(1) with white noise a very interesting result was obtained. which I'm still digesting -).

And what I said in the 1st post, that dependence is very similar to forex and one can find different patterns there.
what is wrong here?

the conclusion remains the same forex is similar to the PRNG. the only difference is that the market
1. non-stationary 2. non-ergodic 3. partially deterministic.
I mean the market.

Of course, to say this is the same as to say nothing.

But to me, once again, the nature of different market figures has become clear.

Or did you have something else in mind?
 
alexjou:
IMHO, there's no need to reinvent the wheel. There is a wonderful thing called G.A.R.C.H. in MATLAB. Toolbox - just a tool for studying financial time series. See, for example, here: http://www.mathworks.com/access/helpdesk/help/toolbox/garch/.

Thank you. It's under my nose and I don't know.

GARCH is I know it's a non-linear regression model.
AR-MA-ARMA-ARIMA-NARX-ARCH-GRACH.

I took a look at the market modelling package.

I don't really understand this approach.

You take a pair, take the first difference, build an autocorrelation (correlation can only estimate linear dependence)
Then one of models, e.g. ARMA, is built.
But these equations include e(t). This somehow stops me from further study.



Have you worked with it?
 
Due to lack of time, I haven't got into GARCHe thoroughly, I'm just about to. I've read the pdf manuals for versions 13 and 14, mainly the introductory part, which explains what they are. So far, I've got some vague idea of how to use GARCH + ANFIS together.
123456789
New comment