Author's dialogue. Alexander Smirnov. - page 15

New comment
 
AlGor писал (а): Off-topic, of course, but still interesting - LeoV, could you show a picture of CSSA Cycles indicator of the same developer (looks very nice on stocks)? I would like to see how it looks on Forex quotes.


Here is CSSA-Cycles. This is with default parameters. But otherwise - parameters need to be adjusted of course.......

 

At point N it matches minus distortions from MT4 chart.
has already thrown out the [100][21] array and a week's handiwork.)))

Respect to Math and Math.

 
Korey:

At point N it matches minus distortions from MT4 chart.
has already thrown out array [100][21] and a week of crude work.)))

Kudos to Math and Math.



So you don't have to dig around. Here is a piece of S. Bulashev's "Statistics for Traders" p.156.

Alexei (Mathematician), what do you think about it? I feel uncomfortable with it, it feels wrong.

 
LeoV, thank you !
It will be something to think about.
 
to Prival
Yes, you're right, experimentally with period 4 => Lrma[i+1]-Lrma[+2]==a
But I've thrown the [100][21] array and about three dozens of other indices will follow. Respect.
 
Prival:

Alexei (Mathematician), what do you think about it? It makes me feel uncomfortable, something is wrong.


I'm not Mathemat, but I will say. The average is calculated on an interval and should refer to the whole interval. Attributing it to a particular point in the interval is, in a sense, arbitrary and incorrect. Bulashev's interpretation - the mean of the function corresponds to the mean of the argument - is no more justified than any other interpretation. One could say, without any summation, that the midpoint of the interval is most justified because the future and the past are equal. Or one could say, on the grounds of causality, that the price at a given point in time depends only on the past and does not depend on the future and attribute its average value to the last point in the interval. From a physical (but not mathematical) point of view this makes more sense, but that is how phase lag occurs.
 
Reshetov:
LeoV:
Reshetov wrote (a): The whole point is that there is no applied sense from this adaptive steepness. If the indicator was extrapolating at least 1 bar ahead, then it would be worth the trouble. As it is, it is only of academic interest to nerds.

Absolutely agree. And for neural networks a small delay is nothing at all....

I take it back.

Here are the results of LRMA + neural network (constant lot strategy):


I will have to experiment with JMA

Yura, please put it in a separate thread. Put a little more detail on it. It is interesting. As you wish, you may call me a botanist or a pot (only do not put me into an oven :-) ). But do not drop the thread. In a correct argument sometimes the truth is born. With respect. Privat.

 
Prival писал (а): Here is a piece from S. Bulashev "Statistics for Traders" p.156

<Scan from Bulashev>

Alexei (Mathematician), what do you think about it? It makes me feel uncomfortable, something is wrong.

It's not a big deal. Mashka has a lag that is about half a period. This is it. This is how I calculate it (without theoretical justification, purely on a hunch): I build a function w[n] whose values are equal to the weights of each clause inside the smoothing window. For SMA it is just a constant. And then we calculate a point back in time where the area under the curve is exactly half the full area. It is exactly in the middle, i.e. (T-1) / 2.

For LWMA, by the way, the lag is smaller: Lag = (T-1) * (sqrt(2) - 1) / sqrt(2) ~ (T-1) / 3.42. This is due to the skewness of the weights to the right, i.e. to the current price.

For the EMA: it needs to be integrated in order to evaluate. The skewness is even larger.

Finally, for LRMA: Weighting function is a straight line with negative k-values in the left part of the window. That is why its lag is even less than LWMA's, but it still lags behind EMA on large windows.

If interested, I will calculate and post delays for EMA and LRMA.
 

The lag is different and at point N they converge. So far I am uncomfortable with this, I do not understand something. This is probably from the same area where I argued with mech.matzov, how to solve an integral in the form of ITO (- t ...0) or Stratonovich (- t /2 ...t /2) 'FR H-volatility'. Because the solutions are different (although the model is the same, there are both solutions on a simple model in the file), and I don't know which one is correct :-( .


Z.P. Yurixx and Mathemat.

After reading your posts, an idea came up, I'm writing it down so I won't forget it. To make an adaptive indicator based on FFT without re-drawing + triangular window with a peak at t=0, adaptation by a threshold that removes ADC noise. It is necessary to think about variation of window width.

 
Mathemat:

Finally, for LRMA: the scales function is a straight line with negative k-values on the left side of the window. So it has even less lag than LWMA, but still lags behind EMA on large windows.

If you interested, I'll calculate and post delays for EMA and LRMA.
Hi! Very interesting. At least approximately, without calculations. I'm really into mash-ups! I guess it must happen to anyone, and not just once.
Also, tell me please, at what window does EMA start winning lag over LRMA?
1...8910111213141516171819202122...44
New comment