Random Flow Theory and FOREX - page 76

 
IgorM:

Alas, for the first time I don't agree with you, it doesn't work in forex: if you enter the market every time by a TS signal, any system will sell out during a flat

....
It's not a problem if you lose less than you made during the trend.
 
sganalysis.com
 

I'm leaving a useful link to the Kalman filter, so it doesn't get lost. I hope it will help those who want to understand Kalman, and how to apply it to the market here in the thread has been written for a long time

http://habrahabr.ru/post/140274/

 
Prival:

I am leaving a useful link to the Kalman filter, so it doesn't get lost. I hope it will help those who want to understand Kalman, and how to apply it to the market has been explained in the thread for a long time.

http://habrahabr.ru/post/140274/


Alive! And not in the bathhouse! Haven't seen you in a while. Somehow I came across your free signals on instaforex. I think you have a website somewhere.
 
Prival:

I'm leaving a useful link to the Kalman filter, so it doesn't get lost. I hope it will help those who want to understand Kalman, and how to apply it to the market has been explained in the thread for a long time.

http://habrahabr.ru/post/140274/


Christ, even people like Prival demonstrate a phenomenal level of ignorance!

Using Kalman filter is a state space model. You cannot do without Kalman.

For all this there is a ready-made mathematics, e.g. EVews.

I enclose an overview of R packages on the topic of the branch. One should use ready-made, not re-inventing bicycles. And discuss the use of off-the-shelf stuff.

Files:
packages_r.zip  48 kb
 
faa1947:

Christ, even people like Prival demonstrate a phenomenal level of ignorance!
Using a Kalman filter is a state space model. There's nowhere without Kalman.
There are ready-made mathematics for all this, e.g. EVews.
I enclose an overview of R packages on the topic of the branch. You should use ready-made, not re-inventing bicycles. And discuss the use of off-the-shelf stuff.

Do you not understand what the article is about? The article explains the principle on the fingers.

Roughly speaking, to demonstrate multiplication 4 x 5.
you drew five rows of sheaves of 4 in a row and asked them to calculate that there are 20 of them.
The author knows that there is a calculator, an excel and a matlab...
 
jartmailru:
Did you not understand what the article was about? The article explains the principle on the fingers.

Roughly speaking, to demonstrate multiplication of 4 x 5
you drew 5 rows of sheaves of 4 pieces in a row and offered to calculate that there are 20.
The author knows that there is a calculator, and Excel, and matlab...


I understand everything perfectly.

The article is discussing the Kalman filter as such, which has the widest application.

I say, there is already a ready-made code to apply this filter to a quotient and one can and should discuss the results of applying this ready-made code to quotients, rather than reading scientific and cognitive articles with a subject area other than quotients. Understand, everything known about Kalman as applied to economics for 30 or 40 years: advantages and disadvantages, areas, limitations, formation of said matrices, etc. - is a ready-made code with instructions for application specifically to economic data.

Of course, it is possible to read an article and, after picking your nose, invent another bicycle with column exercises or a calculator. I don't know if Excel or Matlab have ready-made code, but the packages I mentioned have ready-made code for application of models, of which Kalman filter is a part. You take a cotier, take a model with default parameters and see what you get without thinking about Kalman's inner workings. And if you're not happy with the result, you start to delve into the model settings. It's just another level. But we are offered to read about Kalman's filter, and even with an example that has nothing to do with kotirs. And this is fundamental as kotirs are always non-stationary and for non-stationary time series Kalman's filter is especially good.

 
faa1947:


I understand everything perfectly.

The article is discussing the Kalman filter as such, which has the widest application.

I say, there is already a ready-made code to apply this filter to a quotient and one can and should discuss the results of applying this ready-made code to quotients, rather than reading scientific and cognitive articles with a subject area other than quotients. Understand, everything known about Kalman as applied to economics for 30 or 40 years: advantages and disadvantages, areas, limitations, formation of said matrices, etc. - is a ready-made code with instructions for application specifically to economic data.

Of course, it is possible to read an article and, after picking your nose, invent another bicycle with column exercises or a calculator. I don't know if Excel or Matlab have ready-made code, but the packages I mentioned have ready-made code for application of models, of which Kalman filter is a part. You take a cotier, take a model with default parameters and see what you get without thinking about Kalman's inner workings. And if you're not happy with the result, you start to delve into the model settings. It's just another level. But we are offered to read about Kalman's filter, and even with an example that has nothing to do with kotirs. And this is fundamental as kotirs are always non-stationary and for non-stationary time series Kalman's filter is especially good.


faa1947:


Gosh, even people like Prival display a phenomenal level of ignorance!

Using a Kalman filter is a state space model. There's nowhere without Kalman.

There are ready-made mathematics for all this, e.g. EVews.

I enclose an overview of R packages on the topic of the branch. You should use ready-made, not re-inventing bicycles. And discuss the use of off-the-shelf stuff.

I don't understand why the criticism. The man wrote an article about the Kalman filter and you say that everything is already known about it. The prival doesn't claim to have invented it. Now you need an article on how to use it correctly in forex.
 
faa1947:

You are wrong. I will try to write briefly about what.

1. You are right that the procedure for calculating the Kalman filter is known and has been known for a long time. The order of calculation and formulas themselves are given here in a branch (I even laid out two ways, depending on a priori data how to count these matrices), but...pay attention 4 years as lay, and no implementation not laid out in the code base.... just ask yourself a question why ? (although I could be wrong it's been a long time since I looked there) but what they sometimes send me on Skype can hardly be called a competent solution...

2) in order to properly use this filtration procedure, it needs to be understood, really understand how things work, otherwise it turns out to be nonsense.

Although the procedure is known (it is a sequence of matrix multiplication and operations on them), the matrices themselves must and should be set correctly. And these matrices (their dimensionality and structure) depend on the observed process and the characteristics of the observation (measurement accuracy).

4. For example, can you say with what precision in MT4 EURUSD is measured ? what is the observed noise value ? the noise characteristics ? is it white ? Without answering these questions you cannot correctly set the measurement matrix, what will you put there ? what numbers ?

5. but the most difficult, or rather the main one, is to specify matrix F. it defines all characteristics, whether the filter will work or not, etc. Before the filter will work, it is necessary to answer a lot of questions and put it all into it, and if everything is done correctly + the process is observable, it is possible. I repeat maybe it will work.

6. I'll try to explain it with an example. Here is a picture

http://charts.mql5.com/1/39/eurusd-m5-alfa-foreks.png

Each colored line is the output of a Kalman filter tuned to determine parameters of currency movements, i.e. to calculate USD movement characteristics all currency pairs containing USD are used, for each currency pair the (travelled distance, speed and acceleration) are additionally taken into account correlations between pairs... in total 7 currency pairs are considered per 3 stochastic diff. equations (they are in a branch here) + currency correlations - total matrix 22*22

Tell me, is there the same F matrix in those packs or is it different ? if even 1 digit in that matrix is different, then the filter outputs will be different...there's nothing to discuss...

So that's it. I apologise for the lack of brevity.

 
gpwr:

Alive! And not in a bathhouse! Haven't seen you in a while. I got your free signals on instaforex. I saw your site somewhere.


I can assure you it wasn't me. I never gave signals to anybody. I took part in some contests on Broko (but that was a long time ago). I do not have a website and was not. I am sorry somebody is using my nickname, I hope it is not for evil. If you suddenly stumble across this information, please tell me in a personal link, if you do not mind. Thanks in advance.

Z.I. Better also told me he saw my signals, but that can't be.