How to use Kalman filter in the trading system.

Aleksandr Masterskikh 2017.10.12 16:04 #21

Very interesting example of using recursive Kalman filter.

In spite of many simplifications - "sketching" autoregression, using closing prices instead of matrix representation, we got a clear example of using the method.

Of course, this is only a small component of the trading system, as the task set (to filter out noise) is not completely solved - there are many entries in the flat zone (i.e. in the noise zone).

But this is the trouble of the traditional analysis industry - modern analysis methods have no unambiguous (in terms of the nature of the process) definition of the concepts "trend", "noise", "flat", so each author solves the filtering problem in his own way.

Discussion of article "Deep Discussion of article "How First sacred cow: "If

Dmitriy Gizlyk 2017.10.12 16:24 #22

Aleksey Vakhrushev:

Again, the red line was redrawn on the 0 bar, maybe it should be, I can send you a screenshot in the evening, I was looking at the minutes

Some glitch on the site, I was replying to your previous message, which for some reason is now empty, as well as my reply.

In the indicator, I deliberately did not do the redrawing. I will check it again.

Adding script (or anything) Interesting and Humour extern type variables discoveries

Aleksey Vakhrushev 2017.10.12 16:56 #23

This is my redraw

Dmitriy Gizlyk 2017.10.12 17:47 #24

Aleksey Vakhrushev:

This is how I have such an overdrawing

Can you specify the parameters of the indicator?

Aleksey Vakhrushev 2017.10.12 22:12 #25

Dmitriy Gizlyk:

Can you specify the parameters of the indicator?

Bars 144 Shift 10000

with standard settings it also redraws

Vasily Belozerov 2018.02.05 12:51 #26

Imagine you are about to read an article in a respected scientific journal. The article begins: The Earth is flat and stands on Three Whales. How would you feel about the rest of the words in the article? My point is that the author uses the phrase "extrapolation (prediction)". I remind you that prediction is a favourite pastime of astrologers and has nothing to do with the scientific approach. And extrapolation is just one of (and not the only) methods of forecasting, meaning "the extension of trends established in the past to the future period". And if we quote the footnote given by the author, we get that "The Kalman filter operates with the concept of a vector whose dynamics is described by probability densities". It turns out that ideally, this vector will show us only the probable direction. Even if the probability of the direction is 50/50 or 75/25, then besides the direction, we need the amplitude and time of entry. It turns out that we need to solve a problem like, X+Y+Z=17, and this problem cannot be solved in one action. And lastly. To quote the author: "On the chart of currency or stock rates we always see price fluctuations that differ in frequency and amplitude. So maybe create an amplitude-frequency demodulator and things will move forward? But as far as I'm concerned, I'd rather have an amplitude-phase one.

Discussion of article "Comparative From theory to practice Wave analysis

Vasily Belozerov 2018.02.05 13:18 #27

To give you an analogy. The location of your mobile phone can be determined with one tower, but only approximately. And if you use two towers, you will be easily found at the intersection of vectors. So Kalman Filter is just one vector (one tower). That's good, but it's not enough. At least two, and preferably three towers. The probability of detection increases many times over.

Help! That's the style! :) A little surprised :)