By the way, if you happen to succeed at something, it is better not to post it in articles and codes, I think you can see why.
Don't forget that people with special education are playing against us and what we are discussing here was chewed up for them in the first classes. Let's not be so proud of our phantom knowledge.
By the way, if you happen to succeed at something, it is better not to post it in articles and codes, I think you can see why.
Don't forget that people with special education are playing against us and what we are discussing here was chewed up for them in the first classes. Let's not be so proud of our phantom knowledge.
Models devoted to publicity lose effectiveness very quickly. This is an oft-discussed fact, confirmed more than once and by more than one person. Who plays there is less important :)
-Aleksey-: Models devoted to publicity lose effectiveness very quickly. This is an oft-discussed fact, confirmed more than once and by more than one person. And who plays there is less important :)
I know that.
I'm trying to discuss the methodology of model building here, and I believe that the lifetime of a model is exactly one step. That's why I've posted the formula for a perfectly chic mashup.
-Aleksey-: I'm interested in this question. Even if you mature for tester checks and the result is 51/49 simplistically on trades and 51/49 on points to the plus, you will have to wait about 100 trading days to realize the paltry stat advantage.
It will take not 100 trades at all, but many times that, tens of thousands, to grasp such statistical advantage to be able to talk about its significance.
Let the deals be N. The stat advantage (2% * N) must be at least twice as big as sqrt(N). And in this case we will be about 95% sure of the statpremium significance.
faa: That's why I posted the formula for the totally awesome waving.
What is your 97% quality of this machine (if you are talking about HP)? Is there a formula?
Mathemat: Ahh, now I see. 97% is just the R-squared of the model, not the HP quality.
For the rest of the readers, we would like to point out:
The R-square, also called the measure of certainty, is a measure of the quality of the regression obtained. This quality is expressed by the degree of consistency between the raw data and the regression model (estimated data). The measure of certainty is always within the interval [0;1].
In our case, the regression is 97% consistent with the quoted data.
As interest in the thread has waned, I repeat my post from Friday:
Closing the forecast for Friday on Close. Here is the result:
Fact
Value
Change
Forecast
Forecast
Error
Forecast
Error
Change
Change
Forecast
Forecast
for
Open
prices
to
based on
in pips
based on
in pips
forecast
forecast
on EURUSD
by DX
date
dates
eurusd
DX
on eurusd
on DX
matched?
matched?
2011.11.08 23:59
1,383
2011.11.09 23:59
1,3524
-0,0306
2011.11.09 23:59
1,3798
56
1,3663
67
-0,0032
-0,0167
Yes
Yes
2011.11.10 23:59
1,361
0,0086
2011.11.10 23:59
1,3613
60
1,3742
70
0,0089
0,0218
Yes
Yes
2011.11.11 23:59
1,3778
0,0168
2011.11.11 23:59
1,3541
59
1,3766
71
-0,0069
0,0156
No
Yes
2011.11.14 23:59
1,3624
-0,0154
2011.11.14 23:59
1,3676
59
1,3673
69
-0,0102
-0,0105
Yes
Yes
2011.11.15 23:59
1,3525
-0,0099
2011.11.15 23:59
1,3650
59
1,3634
69
0,0026
0,0010
No
No
2011.11.16 23:59
1,3455
-0,0070
2011.11.16 23:59
1,3529
57
1,3627
69
0,0004
0,0102
No
No
2011.11.17 23:59
1,3468
0,0013
2011.11.17 23:59
1,3446
57
1,3521
70
-0,0009
0,0066
No
Yes
2011.11.18 23:59
1,3514
0,0046
2011.11.18 23:59
1,3422
55
1,3479
70
-0,0046
0,0011
No
Yes
Some conclusions:
1. The DX prediction is much better than the lagged EURUSD itself
2. The results of the forecast are qualitative (matched - not matched) and do not take into account the MM and the spread. For example, on the last day, Friday, calculated price = 1.3514 and High = 1.3613. When using the DX forecast the potential profit was 100 pips higher. On the other hand, Low=1.3447, and using an unsuccessful EURUSD forecast using the sliding trawl for SL, the loss would have been minimal.
3. The table presented cannot be the basis for using the model due to the small sample size. The need to use a tester is obvious to all. Such a possibility is available. The corresponding code is laid out in the attachment to my article. But I will not do it, as in my opinion the model is not ready and needs to be finalised before final testing.
My plan is as follows:
1. I finish making predictions.
2. I suggest that everyone who is interested:
a) discuss these results
b) modernise this model.
c) offer your models
3. I am ready to implement the results of discussion and upgrades in code and post the results.
Let me remind you of the type of models:
a) For EURUSD on lags: EURUSD = hp(-1 to -4) + hp_d(-1 to -2)
b) For DX:
DXM = 1/DX - we use the inverse of the quotient
EURUSD = DXM_HP(-1 TO -4) + DXM_HP_D(-1 TO -2)
In these formulas HP is the Hedrick-Prescott indicator, and HP_D is the residual = kotir - indicator. The bars in brackets are the bars before the current one, (-1 to -4) means the last 4 bars.
The real equation after evaluating the coefficients at variables looks as follows:
If it is standard deviation (s.c.o.), and the forecast value itself is a normally distributed value with exactly that s.c.o., then it would be a good idea to ignore all forecasts that are less modulo than at least two s.c.o. Then if the modulus of the forecast is less than two s.c.o. (somewhere around 118 points) there's about 95% probability we won't make a mistake attributing the forecast value to zero.
It turns out that a forecast whose modulo value is less than 2 s.c.o. should be considered uninteresting (it is a zero movement forecast).
Isn't the forecast itself a mathematical expectation of the model? In this case the magnitude of the error is irrelevant, because the average gain of the model will always be positive (m.o. > 0) and equal to the magnitude of the prediction * number of predictions. Well yes, a large error will increase the variance of the results, but no more than that.
Allow the use of cookies to log in to the MQL5.com website.
Please enable the necessary setting in your browser, otherwise you will not be able to log in.
You are signed in. Two-factor authentication has been disabled. Set it up again and scan a new QR code, or continue with two-factor authentication disabled.
By the way, if you happen to succeed at something, it is better not to post it in articles and codes, I think you can see why.
Don't forget that people with special education are playing against us and what we are discussing here was chewed up for them in the first classes. Let's not be so proud of our phantom knowledge.
By the way, if you happen to succeed at something, it is better not to post it in articles and codes, I think you can see why.
Don't forget that people with special education are playing against us and what we are discussing here was chewed up for them in the first classes. Let's not be so proud of our phantom knowledge.
Models devoted to publicity lose effectiveness very quickly. This is an oft-discussed fact, confirmed more than once and by more than one person. And who plays there is less important :)
I know that.
I'm trying to discuss the methodology of model building here, and I believe that the lifetime of a model is exactly one step. That's why I've posted the formula for a perfectly chic mashup.
It will take not 100 trades at all, but many times that, tens of thousands, to grasp such statistical advantage to be able to talk about its significance.
Let the deals be N. The stat advantage (2% * N) must be at least twice as big as sqrt(N). And in this case we will be about 95% sure of the statpremium significance.
faa: That's why I posted the formula for the totally awesome waving.
What is your 97% quality of this machine (if you are talking about HP)? Is there a formula?
What is your 97% quality of this mach (if you are talking about HP)? Is there a formula?
Personally rewritten from the post above for you:
EURUSD = -1552.7613734*DXM_HP(-1) + 4731.89082764*DXM_HP(-2) - 4360.68995095*DXM_HP(-3) + 1287.82064375*DXM_HP(-4) - 98.9244837504*DXM_HP_D(-1) - 131.011472103*DXM_HP_D(-2)
HP is a fraction of that.
Name me any indicator in the code base for which the R-square is known
Ahh, now I see. 97% is just the R-squared of the model, not the HP quality.
For the rest of the readers, we would like to point out:
The R-square, also called the measure of certainty, is a measure of the quality of the regression obtained. This quality is expressed by the degree of consistency between the raw data and the regression model (estimated data). The measure of certainty is always within the interval [0;1].
In our case, the regression is 97% consistent with the quoted data.
As interest in the thread has waned, I repeat my post from Friday:
Closing the forecast for Friday on Close. Here is the result:
Some conclusions:
1. The DX prediction is much better than the lagged EURUSD itself
2. The results of the forecast are qualitative (matched - not matched) and do not take into account the MM and the spread. For example, on the last day, Friday, calculated price = 1.3514 and High = 1.3613. When using the DX forecast the potential profit was 100 pips higher. On the other hand, Low=1.3447, and using an unsuccessful EURUSD forecast using the sliding trawl for SL, the loss would have been minimal.
3. The table presented cannot be the basis for using the model due to the small sample size. The need to use a tester is obvious to all. Such a possibility is available. The corresponding code is laid out in the attachment to my article. But I will not do it, as in my opinion the model is not ready and needs to be finalised before final testing.
My plan is as follows:
1. I finish making predictions.
2. I suggest that everyone who is interested:
a) discuss these results
b) modernise this model.
c) offer your models
3. I am ready to implement the results of discussion and upgrades in code and post the results.
Let me remind you of the type of models:
a) For EURUSD on lags: EURUSD = hp(-1 to -4) + hp_d(-1 to -2)
b) For DX:
DXM = 1/DX - we use the inverse of the quotient
EURUSD = DXM_HP(-1 TO -4) + DXM_HP_D(-1 TO -2)
In these formulas HP is the Hedrick-Prescott indicator, and HP_D is the residual = kotir - indicator. The bars in brackets are the bars before the current one, (-1 to -4) means the last 4 bars.
The real equation after evaluating the coefficients at variables looks as follows:
EURUSD = -1552.7613734*DXM_HP(-1) + 4731.89082764*DXM_HP(-2) - 4360.68995095*DXM_HP(-3) + 1287.82064375*DXM_HP(-4) - 98.9244837504*DXM_HP_D(-1) - 131.011472103*DXM_HP_D(-2)
Anyone interested - take part in an econometrics exercise!I don't know what this error is.
If it is standard deviation (s.c.o.), and the forecast value itself is a normally distributed value with exactly that s.c.o., then it would be a good idea to ignore all forecasts that are less modulo than at least two s.c.o. Then if the modulus of the forecast is less than two s.c.o. (somewhere around 118 points) there's about 95% probability we won't make a mistake attributing the forecast value to zero.
It turns out that a forecast whose modulo value is less than 2 s.c.o. should be considered uninteresting (it is a zero movement forecast).
Isn't the forecast itself a mathematical expectation of the model? In this case the magnitude of the error is irrelevant, because the average gain of the model will always be positive (m.o. > 0) and equal to the magnitude of the prediction * number of predictions. Well yes, a large error will increase the variance of the results, but no more than that.
Isn't the prediction itself a mathematical expectation of the model
Everything is fine if the error is stationary. Many times I have written and given graphs of the error, which have a very convoluted appearance.