Author's dialogue. Alexander Smirnov. - page 42

New comment
 
Mathemat:
Prival:
Mathemat:

Quadratic Regression MA = 3 * SMA + QWMA * ( 10 - 15/( N + 2 ) ) - LWMA * ( 12 - 15/( N + 2 ) )

QWMA( i; N ) = 6/( N*(N+1)(2*N+1) ) * sum( Close[i] * (N-i)^2; i = 0...N-1 ) (a wizard with quadratic weights).

I got other formulas.

where

Exactly the same formulas, thank you, Prival. Give me similar ones with respect to the dummies.


Given similar ones (the answer is the same) + reduced the number of operations, here is the final expression

the difference I meant in QWMA calculation I have i^2, you have (N-i)^2. Double-check that.

 
Prival:

If you know the current value of coefficients A and B in a linear regression, can you calculate the RMS

here are the formulas

coefficient A

coefficient B

Hm, what do you mean it's better in the morning, but here's the formula :): SCO^2 = (Sum(Y*Y) - A*Sum(X*Y) - B*Sum(Y))/(N-2). It includes SMA, LWMA, and a mean from squares of prices not yet developed in this approach. It is essential that X should vary from 0 to N-1.
Prival:

I have i^2, you have (N-i)^2. Double-check this.

Of course, for another direction X will be different A and B. But the regression line itself and the RMS will still coincide. If all is correct of course.

P.S. I have redirected QWMA to LWMA. I continue to confuse in terms :)
 
Prival: unlike what I meant in the QWMA calculation I have i^2, you have (N-i)^2. Double-check that.
It depends on the numbering of the counts(closing prices). If as in MT4, then the formula is like mine, and if the last bar (zero) has a number N, then like yours.
 

Gentlemen ichmo error - 0 bar is always zero, and N is extreme in the sample, regardless of where to count from right or left (this is an array), although I understand what you mean, and I think you know what I mean. correct i^2. It would not be correct to use the coefficient (N-1)^2 (instead of 1^2) on the 1st bar, is this a mistake or am I deriving something wrong.

I'll send you the RMS later and double-check it, the result is disappointing, but it's what I was saying RMS(Y) is directly proportional to RMS(X) and if we don't pay attention to random value of X axis we step on it, at least for more than one time (at least for me). Everything is interconnected :-(.

Mathematician, let's make something clear with notation, you know English, I'm much worse. That's why I suggest to double-check cubic approximation and make it coherent, since everybody understands SMA, but it's necessary to determine how to calculate QWMA. Here is a new branch. Because Smirnov is not topical now, again we are already in the thicket :-)

 
Hmmm, what does it mean to have a better time in the morning, but here is the formula :) : СКО^2 = (Sum(Y*Y) - A*Sum(X*Y) - B*Sum(Y))/(N-2). It includes SMA, LWMA, and the unlearned average from price squares in this approach. It is essential that X should vary from 0 to N-1.
I understand that in this formula there is a division by N-2, i.e. an attempt to get an unbiased estimation? Something like that is confusing, it seems easier from 1 to N, and dividing by N-1, then it seems classic + there are some programmers calculations at 0 bar do not recognize :-) (thank goodness they don't use bars like MN for trading :-)))),
 
Prival:
I understand that the formula for RMS^2 has a division by N-2, i.e. trying to get an unbiased estimate ?
No. N-2 instead of N is actually a consequence of replacing the expectation by the mean in the real calculations. And "from 0 to N-1" is a choice of direction and origin for the X-axis. Depending on their choice, expressions may become simpler or more complex. With such a choice, the expression for RMS becomes as I wrote, i.e. it is very simple and fits wonderfully into the boosted algorithm for calculating the sliding LR. Once again I would like to emphasize an important thing to be reconciled with :) The values of regression coefficients will depend on the choice of direction and datum for X, but the line on the chart will ultimately be the same. And, consequently, the RMS for Y-mu will not depend on the choice of direction and datum for X.
P.S. It has nothing to do with the zero bar. I just assume that for the first bar X=0. If I were calculating the zero bar, I would take X=0 for the zero bar. If I were starting LR from the 10th bar, I would assign X=0 to the 10th bar.
 
I will also say this: If the RMS is the standard deviation from the Ah+B line, then divide by N. If the RMS is the root-mean-square error of the regression, then divide by N-2. For price charts, however, I think this is an insignificant subtlety.
 
lna01:
I will also say this: If the RMS is the standard deviation from the Ah+B line, then divide by N. If the RMS is the root-mean-square error of the regression, then divide by N-2. However, for price charts, I think it's an insignificant subtlety.

This is probably the most accurate way. That is not in relation to the number of regression points, but in relation to the number of degrees of freedom.
 

Is there any way of contacting the author, Alexander Smirnov? My Facebook handle is 311652834

 
LeoV:

Is there any way of contacting the author, Alexander Smirnov?

Perhaps by e-mail: smirnov_dntu@ukr.net
1...35363738394041424344
New comment
Reason: