Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 68

New comment
 
alexeymosc:
The data is in the attachment. I worked with the quantized series (far right).

Here's the result.

A very strange graph. Trimmed. It looks like the calculations were done with limited accuracy.

Statistics

Very funny.

ACF

Date: 10/14/12 Time: 11:58

Sample: 1,272

Included observations: 3271

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

| | | | 1 -0.059 -0.059 11.332 0.001

| | | 2 -0.053 -0.057 20.704 0.000

| | 3 0.025 0.019 22.820 0.000

| | 4 0.005 0.005 22.908 0.000

| | 5 -0.062 -0.059 35.486 0.000

| | | 6 0.007 -0.000 35.639 0.000

| | | 7 -0.038 -0.045 40.475 0.000

| | 8 0.032 0.030 43.845 0.000

| | 9 -0.007 -0.008 44.004 0.000

| | 10 0.025 0.026 46.003 0.000

| | | 11 -0.033 -0.032 49.674 0.000

| | 12 0.048 0.043 57.372 0.000

| | 13 0.002 0.006 57.382 0.000

| | 14 -0.032 -0.028 60.736 0.000

| | 15 -0.033 -0.033 64.288 0.000

| | 16 0.047 0.034 71.425 0.000

| | 17 -0.004 0.007 71.469 0.000

| | 18 -0.039 -0.037 76.462 0.000

| | 19 -0.004 -0.008 76.520 0.000

| | 20 0.017 0.004 77.426 0.000

| | | 21 -0.046 -0.040 84.377 0.000

| | 22 0.020 0.013 85.636 0.000

| | 23 0.006 0.006 85.767 0.000

| | 24 -0.010 -0.010 86.089 0.000

| | | 25 -0.001 -0.004 86.090 0.000

| | | 26 -0.022 -0.028 87.663 0.000

| | 27 0.025 0.031 89.677 0.000

| | | 28 -0.022 -0.028 91.250 0.000

| | 29 0.028 0.029 93.841 0.000

| | 30 0.009 0.011 94.135 0.000

| | 31 0.007 0.015 94.290 0.000

| | 32 0.004 0.001 94.350 0.000

| | | 33 -0.007 -0.009 94.501 0.000

*| | *| | 34 -0.092 -0.085 122.33 0.000

| | | 35 0.010 -0.006 122.66 0.000

| | | 36 0.008 0.003 122.89 0.000

The last column is the probability of correlation. Zero.

This data is of no interest - loss of precision. The analysis is nothing, just a figure.

 
Avals:

What is ZZ according to Pastukhov? Pastukhov investigated kagi/renko in the classical construction. This rule (2H) does not apply to ZZ exactly. There is a dependence on the value of the knee in points.
Yes, we are talking about H-volatility.
 
VNG: You can maximise by investigating the structure of the engine, decomposing it into ranges, getting statistics of transitions to senior (or junior) ranges.

hmm, did this - visually it looks like this:

http://imglink.ru/pictures/14-10-12/6038b20b9bfbd1e06c08e649623cca4b.jpg

http://imglink.ru/pictures/14-10-12/47b7615b511f6b8a6f3b638a2fcda38b.jpg

Each colored triangle is the TF from right to left of M1,M5 to MN relative to the vertical line that simulates the observer's view of the history, the history in the form of ranges of High and Low extremum/historical max/min

I unloaded into Statistica as an alphabet, yes there are repeated areas/words, even for 2-3 TFs, but repeatability is not periodic, periods of repeatability vary from 2 months to several years

 
VNG:


I am also "you" to me, if there are no objections.

Why not? Is there a reasoning behind it?


the abstract SB will have the same thing
 
HideYourRichess:
Yes, it's about H-volatility.

it's different there (on the getch chart)
 
alexeymosc:
The data is in the attachment. I worked with the quantized series (far right).

I'll take the usual increments for the opener.

Much more interesting. Statistics

ACF

Date: 10/14/12 Time: 12:05

Sample: 1 3272

Included observations: 3271

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

| | | | 1 -0.063 -0.063 13.075 0.000

| | | 2 -0.033 -0.037 16.554 0.000

| | 3 0.017 0.013 17.558 0.001

| | 4 -0.000 0.001 17.558 0.002

| | | 5 -0.043 -0.043 23.757 0.000

| | | 6 -0.003 -0.009 23.788 0.001

| | 7 -0.024 -0.028 25.722 0.001

| | 8 0.022 0.019 27.264 0.001

| | 9 -0.005 -0.004 27.338 0.001

| | 10 0.032 0.032 30.668 0.001

| | 11 -0.027 -0.025 33.069 0.001

| | 12 0.051 0.048 41.461 0.000

| | 13 0.011 0.016 41.861 0.000

| | 14 -0.020 -0.014 43.111 0.000

| | | 15 -0.040 -0.040 48.488 0.000

| | 16 0.047 0.039 55.873 0.000

| | 17 -0.003 0.006 55.900 0.000

| | 18 -0.054 -0.051 65.566 0.000

| | 19 0.006 0.000 65.688 0.000

| | 20 0.013 0.004 66.214 0.000

| | 21 -0.053 -0.047 75.446 0.000

| | 22 0.025 0.015 77.560 0.000

| | 23 0.014 0.014 78.179 0.000

| | | 24 -0.009 -0.008 78.465 0.000

| | 25 -0.003 -0.005 78.490 0.000

| | 26 -0.024 -0.030 80.367 0.000

| | 27 0.018 0.022 81.400 0.000

| | | 28 -0.006 -0.007 81.522 0.000

| | 29 0.017 0.016 82.452 0.000

| | 30 0.008 0.013 82.657 0.000

| | | 31 -0.002 0.005 82.675 0.000

| | 32 0.010 0.004 83.006 0.000

| | | 33 -0.024 -0.025 84.980 0.000

*| | *| | 34 -0.083 -0.079 107.74 0.000

| | | 35 0.005 -0.011 107.82 0.000

| | | 36 0.022 0.014 109.37 0.000

Probability of no correlation. Initially there is some correlation, but not significant.


 
alexeymosc:
The data is in the attachment. I worked with the quantized series (far right).

I'll take the usual increments for the opener.

Much more interesting. Statistics

ACF

Date: 10/14/12 Time: 12:05

Sample: 1 3272

Included observations: 3271

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

| | | | 1 -0.063 -0.063 13.075 0.000

| | | 2 -0.033 -0.037 16.554 0.000

| | 3 0.017 0.013 17.558 0.001

| | 4 -0.000 0.001 17.558 0.002

| | | 5 -0.043 -0.043 23.757 0.000

| | | 6 -0.003 -0.009 23.788 0.001

| | 7 -0.024 -0.028 25.722 0.001

| | 8 0.022 0.019 27.264 0.001

| | 9 -0.005 -0.004 27.338 0.001

| | 10 0.032 0.032 30.668 0.001

| | 11 -0.027 -0.025 33.069 0.001

| | 12 0.051 0.048 41.461 0.000

| | 13 0.011 0.016 41.861 0.000

| | | 14 -0.020 -0.014 43.111 0.000

| | | 15 -0.040 -0.040 48.488 0.000

| | 16 0.047 0.039 55.873 0.000

| | 17 -0.003 0.006 55.900 0.000

| | 18 -0.054 -0.051 65.566 0.000

| | 19 0.006 0.000 65.688 0.000

| | 20 0.013 0.004 66.214 0.000

| | 21 -0.053 -0.047 75.446 0.000

| | 22 0.025 0.015 77.560 0.000

| | 23 0.014 0.014 78.179 0.000

| | | 24 -0.009 -0.008 78.465 0.000

| | 25 -0.003 -0.005 78.490 0.000

| | 26 -0.024 -0.030 80.367 0.000

| | 27 0.018 0.022 81.400 0.000

| | | 28 -0.006 -0.007 81.522 0.000

| | 29 0.017 0.016 82.452 0.000

| | 30 0.008 0.013 82.657 0.000

| | | 31 -0.002 0.005 82.675 0.000

| | 32 0.010 0.004 83.006 0.000

| | | 33 -0.024 -0.025 84.980 0.000

*| | *| | 34 -0.083 -0.079 107.74 0.000

| | | 35 0.005 -0.011 107.82 0.000

| | | 36 0.022 0.014 109.37 0.000

Probability of no correlation. Initially there is some correlation, but not significant.


 
VNG:


Can you tell me more about it?

The algorithm is stated in this sentence

The Expert Advisor counts the number of Zig-Zag knees (not less than Pips) and saves it to the file

I'm sorry, I did not look at the code, but it follows from this sentence that the number of passes to calculate the number of knees should be equal to the number of pips on a minute TF from the maximum price range over the history.

More details? graphs and so on? - It was a long time ago, and only a speculative conclusion has survived. I was satisfied with it because it corresponds, unlike the numerological botany, to the general ideas about the processes on the market - they are different at different levels. In general terms, on small levels there is a tendency towards "returns" (which HFT exploits, and is itself involved in creating this effect), on large levels there is a tendency towards "trendiness" (long-term investments). Somewhere in the middle is what Pastukhov describes as 2H - in my understanding something similar to martingale or "efficient market". The second point is that level boundaries are not constant, i.e. it is impossible to plot one chart and say that it will always be like this. The composition and nature of bidders is constantly changing, so everything else changes accordingly. And so on.
 
alexeymosc:
The data is in the attachment. I worked with the quantized series (far right).

Reduce the window. Large window - the limit theorem begins to work. But we are entering the market for a limited period of time.

Window=100. Graph:

ACF

Date: 10/14/12 Time: 12:11

Sample: 1 100

Included observations: 99

Autocorrelation Partial Correlation AC PAC Q-Stat Prob

.|. | .|. |. 1 0.001 0.001 3.E-05 0.996

.|. | .|. | 2 0.036 0.036 0.1371 0.934

*|. | *|. | 3 -0.148 -0.148 2.4225 0.489

.|. | .|. | 4 -0.047 -0.048 2.6516 0.618

*|. | *|. | 5 -0.132 -0.124 4.5037 0.479

.|* | |00 .|* |01 6 0.135 0.121 6.4763 0.372

*|. | *|. | 7 -0.096 -0.109 7.4812 0.381

.|. | .|. | 8 0.023 -0.021 7.5395 0.480

*|. | .|. | 9 -0.073 -0.050 8.1324 0.521

.|* | |00 .|* 10 0.105 0.083 9.3778 0.497

.|. | .|. |. 11 -0.018 0.002 9.4136 0.584

.|. | .|. | 12 0.034 -0.028 9.5449 0.656

.|. | .|* | 13 0.060 0.109 9.9605 0.697

.|. | .|. |. 14 0.062 0.049 10.418 0.731

.|. | .|. | 15 -0.053 -0.021 10.750 0.770

*|. | *|. | 16 -0.103 -0.132 12.038 0.741

.|. | .|. | 17 -0.036 0.018 12.196 0.788

*|. | *|. | 18 -0.111 -0.103 13.712 0.748

.|. | .|. | 19 -0.028 -0.062 13.812 0.795

.|. | .|. | 20 0.030 -0.004 13.923 0.834

.|. | *|. | 21 -0.045 -0.087 14.187 0.861

.|. | .|. | 22 -0.008 -0.002 14.196 0.894

.|* | |00 .|* 23 0.124 0.076 16.219 0.846

.|. | .|. |. 24 0.021 0.014 16.280 0.878

.|. | .|. | 25 -0.025 -0.059 16.364 0.904

.|. | .|. | 26 0.041 0.069 16.591 0.921

.|. | .|. | 27 0.046 0.073 16.879 0.934

*|. | .|. | 28 -0.074 -0.062 17.640 0.935

.|. | .|. | 29 0.038 0.056 17.848 0.947

.|. | .|. | 30 -0.039 -0.010 18.071 0.957

.|. | .|. | 31 0.023 0.069 18.151 0.968

.|. | .|. | 32 -0.014 -0.015 18.179 0.976

.|. | .|. | 33 0.021 -0.030 18.245 0.982

.|. | .|. | 34 -0.041 -0.031 18.505 0.986

.|. | .|. | 35 -0.019 -0.038 18.559 0.990

.|. | .|. | 36 -0.029 -0.043 18.697 0.992

The picture has changed dramatically. The likelihood of no correlation is very high.

It remains to be compared with TI. And figure out what we are talking about.

 
Avals:

it's different (on the getch chart)
I see. Well, what can I say - I somehow trust the H-Volatility more than the getch. ;) At least with Pastukhov it is clear where the legs grow from and what the ideas are.
1...6162636465666768697071727374
New comment