The Sultonov Regression Model (SRM) - claiming to be a mathematical model of the market. - page 44
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
Five parameters for 15 points is too much. The usual fitting.
Try the same equation with the same parameters on a thousand bars.
You seem to have a change of direction in your research. You used to make up difuras and try to explain the market.
Now you are just trying to describe it without even trying to explain it.
Five parameters for 15 points is too much. The usual fit.
Try the same equation with the same parameters on a thousand bars.
You seem to have a change of direction in your research. You used to make up difuras and try to explain the market.
Now you are just trying to describe it without even trying to explain it.
1.Correctly noticed, here is bare statistics and a by-product of another direction of research, not related to Forex. As you know, when estimating the coefficients of a linear equation with sets of variables by ANC method, Gauss indicated only a two-step method of gradually eliminating variables at the first step and finding coefficients at the second step, which is very labour-intensive and cumbersome. The second method is based on Cramer's method using determinants, which is no simpler than Gauss's method and has the same computational complexity, although it is more elegant. I have managed to simplify the method decisively and determine coefficients directly and the above example is a debut and I thought you'll pay attention, how I managed to find 5 coefficients while varying four variables simultaneously. Usually, for example, when planning an experiment, it is recommended to gradually determine coefficients with only one variable, setting the other variables at a constant level, you probably remember this not so successful hit of the 60's and 70's. Now really, I have no problem researching 1000 bars at the same time and I will do so. Just please tell me how to download the story directly into exel with commas, sorry, I'm a zero in the technique of using a comp. I am learning gradually and only what is needed at the moment. Please specify in great detail, down to the sequence of button presses.
2. No fit, but indeed the amount of data is not much, as, it was entered manually. However, what is remarkable is that this simple equation tries to describe the price variations as if it were periodic, although it is by no means such.
3. do you think it makes sense to add volumes to these four parameters, especially since they are available, although they are said to be biased?
4. noticed the difference in coefficients at OHLC, apparently from lack of data.
Just please tell me how to download the story directly into exel with commas, sorry, I'm a novice in computer techniques.
1. In the terminal, press F2. In the table that pops up, select a symbol and press "Export". We got a file.
We open the file in Excel. It looks like:
3. In Excel at the top of the tab "Data".
4. Highlight the desired section of data in the table.
5. Click on "Column by column" The Text Wizard will pop up.
6. In the first step we select "Delimited".
7. On the second step of the wizard you should additionally specify the comma separator.
8. On the third step:
8.1. For the first two columns specify the data format of the column "text".
8.3 We leave "General" for the rest, but open "more details" and put "point" as a place separator.
It should look like this
It takes only this equation
F=1.00010409798*CLOSE(-1)^0.999631066509
Trying to add another value leads to a degenerate (singular) matrix.
The fit is very good.
Dependent Variable: F
Method: Panel Least Squares
Date: 11/30/12 Time: 10:57
Sample: 1 2652
Periods included: 23
Cross-sections included: 113
Total panel (unbalanced) observations: 2538
Convergence achieved after 1 iteration
F=C(1)*CLOSE(-1)^C(2)
Coefficient Std. Error t-Statistic Prob.
C(1) 1.000104 0.000122 8222.019 0.0000
C(2) 0.999631 0.000511 1955.530 0.0000
R-squared 0.999342 Mean dependent var 1.266171
Adjusted R-squared 0.999342 S.D. dependent var 0.029512
S.E. of regression 0.000757 Akaike info criterion -11.53332
Sum squared resid 0.001454 Schwarz criterion -11.52872
Log likelihood 14637.78 Hannan-Quinn criterion. -11.53165
Durbin-Watson stat 1.951579
Dependent Variable: CLOSE
Method: Panel Least Squares
Date: 11/30/12 Time: 10:59
Sample: 1 2652
Periods included: 23
Cross-sections included: 113
Total panel (unbalanced) observations: 2538
Convergence achieved after 2 iterations
CLOSE=C(1)*F(-1)^C(2)
Coefficient Std. Error t-Statistic Prob.
C(1) 1.000222 0.000233 4283.747 0.0000
C(2) 0.999132 0.000981 1018.334 0.0000
R-squared 0.997578 Mean dependent var 1.266170
Adjusted R-squared 0.997577 S.D. dependent var 0.029520
S.E. of regression 0.001453 Akaike info criterion -10.22961
Sum squared resid 0.005354 Schwarz criterion -10.22501
Log likelihood 12983.38 Hannan-Quinn criterion. -10.22794
Durbin-Watson stat 1.294442
Here is the chart
We see spikes of around 100 pips. But a very decent histogram, though not normal
Odds = 14 pips.
But the confidence ellipse is depressing - we see extremely high correlation of our coefficients. This is the reason for the singularity of the matrix when additional variables are added.
I would refrain from using the above equation
I have tried to express the average forecast price of a future bar (F) through OHLC prices of previous bars as the following relationship, though I do not know if it has been tried before in such a form or not:
F=A*O^a1*H^a2*L^a3*C^a4,
where - A, a1, a2,a3,a4 are constant coefficients determined by the MNC Gaussian method and this is what we get for 15 bars of TF D1
Hence, the quotient can in principle be expressed by a single equation, but let's find out what the practical usefulness of this is. What are your views?
I did not find the time period from which you took the price sample, but the last 15 bars show the same picture (according to the above formula and derived coefficients):
The green МА with period = 1 is used to compare the forecast more clearly.
The price marks are drawn by a script (in the attachment).
I have not found the time period from which you took the price sample, but the last 15 bars show this picture (according to the formula and the coefficients you have derived):
The green МА with period = 1 is used to compare the forecast more clearly.
The price marks are drawn by a script (in the attachment).
Data used on D1 from 16. 09. 12 to 05. 10. 12
P.S. And if it were not the end of the month, the current candle would be "bearish"... :)))
Didn't notice right away that the coefficients are arranged in reverse order a4 -> a1. Then a month later the calculated coefficients are not "a finger in the sky"... ;)
P.S. And if it were not the end of the month, the current candle would be "bearish"... :)))
It takes only this equation
F=1.00010409798*CLOSE(-1)^0.999631066509
Trying to add another value leads to a degenerate (singular) matrix.
The fit is very good.
Dependent Variable: F
Method: Panel Least Squares
Date: 11/30/12 Time: 10:57
Sample: 1 2652
Periods included: 23
Cross-sections included: 113
Total panel (unbalanced) observations: 2538
Convergence achieved after 1 iteration
F=C(1)*CLOSE(-1)^C(2)
Coefficient Std. Error t-Statistic Prob.
C(1) 1.000104 0.000122 8222.019 0.0000
C(2) 0.999631 0.000511 1955.530 0.0000
R-squared 0.999342 Mean dependent var 1.266171
Adjusted R-squared 0.999342 S.D. dependent var 0.029512
S.E. of regression 0.000757 Akaike info criterion -11.53332
Sum squared resid 0.001454 Schwarz criterion -11.52872
Log likelihood 14637.78 Hannan-Quinn criterion. -11.53165
Durbin-Watson stat 1.951579
Dependent Variable: CLOSE
Method: Panel Least Squares
Date: 11/30/12 Time: 10:59
Sample: 1 2652
Periods included: 23
Cross-sections included: 113
Total panel (unbalanced) observations: 2538
Convergence achieved after 2 iterations
CLOSE=C(1)*F(-1)^C(2)
Coefficient Std. Error t-Statistic Prob.
C(1) 1.000222 0.000233 4283.747 0.0000
C(2) 0.999132 0.000981 1018.334 0.0000
R-squared 0.997578 Mean dependent var 1.266170
Adjusted R-squared 0.997577 S.D. dependent var 0.029520
S.E. of regression 0.001453 Akaike info criterion -10.22961
Sum squared resid 0.005354 Schwarz criterion -10.22501
Log likelihood 12983.38 Hannan-Quinn criterion. -10.22794
Durbin-Watson stat 1.294442
Here is the chart
We see spikes of around 100 pips. But a very decent histogram, though not normal
Odds = 14 pips.
But the confidence ellipse is depressing - we see extremely high correlation of our coefficients. This is the reason for the singularity of the matrix when additional variables are added.
I would refrain from using the above equation
Refrain from jumping to conclusions