What is the meaning of the term non-lagging (in relation to the indicator)? - page 7

 
Yuriy Asaulenko:
A properly done EMA has a group delay about a third smaller than the SMA.
The standard EMA as an average is not done correctly. To see this, all you have to do is draw the graphs on the step.

Ah, what about publishing the correct EMAs of e.g. second/fifth orders. And from the "future" it will be possible to take the geometric mean. :)

It has already been discussed:

Forum on trading, automated trading systems and strategy testing

Moving averages lag

Aleksey Panfilov, 2018.02.21 06:01

Judging by the name, the formula must derive from proportions Y0/Y1=Y1/Y2 or Y0 * Y1^(-2) * Y2=1 (not from differences: Y0-Y1=Y1-Y2 or Y0-2*Y1+Y2=0), but as said on the forum error is not visually noticeable and calculation is harder.

Y0 * Y1^(-2) * Y2=1

Y0 * Y2^(-3) * Y3^(2)=1

Y0 * Y72^(-73) * Y73^(72)=1

Y72 = ( Y0 * Y73^(72) )^(1/73) - the root of degree 73 probably requires a decent amount of computer resources.

Who can do it?
 
Aleksey Panfilov:

Why don't we publish the correct EMA of e.g. second/fifth orders. And from the "future" we can take the geometric mean. :)

Already discussed:

Who will do it?

From the future? - I'll pass.))

In general, anyone can do it, but there's no point in MA being bigger than the 2-3rd order.

 
Yuriy Asaulenko:

From the future? - I'll pass.)

)))

Well at least just throw the correct EMA into the code base? )))

You can average (by interpolation of one point on each new bar) and predict the variants of exponents for each bar. But to tell the truth, the spread would be huge, such a saw-toothed pattern we would get. ))

But from this saw we may take the geometrical mean. It smoothes much smoother than the arithmetic one and corresponds to the nature of rates better. )

 
Aleksey Panfilov:

)))

But the exponent can be averaged (by interpolating one point on each new bar), as well as predicted variants of exponent development for each bar. But the scatter will be maddening, such a saw-toothed pattern will appear. ))

But from this saw we may take the geometrical mean. It smoothes much smoother than the arithmetic one and corresponds to the nature of the rates better. )

MA of 2-3 orders may be used as a basis for forecasting. But they cannot do it on their own. I was using 5 min on 5 min on 1m timeframe and it worked more or less ok. When forecasting on 1 min a candle is approximately equal to noise - it doesn't work.

 
Yuriy Asaulenko:

MAs of order 2-3 can be used as a basis for forecasting. But they are not capable of doing that on their own. I've been doing it on 5 min NS, on 1m timeframe - it works more or less fine. When predicting at 1 min candle is about equal to noise - not good enough.

This is a very wrong approach. There is no noise. Use all available market information from the price indicator to your advantage.
 
Yuriy Asaulenko:

MAs of order 2-3 can be used as a basis for forecasting. But they are not capable of doing that on their own. I've been doing it on 5 min NS, on 1m timeframe - it works more or less fine. When predicting on 1 min candle is about the same as the noise - it doesn't roll.

Most probably it is.

But the correct EMA is just for the love of art. ))

Here are the coefficients for leverage 72:

72:    1    -73              72
72:    1    -2701            5328         -2628
72:    1    -67525           199800       -197100         64824
72:    1    -1282975        5061600       -7489800        4926624    -1215450 
 
Aleksey Panfilov:

)))

At least just the correct EMA to throw into the code base? )))

And there is, one of the first options, back in 2008, but not very correct yet either. ))

 
Aleksey Panfilov:

Here are the ratios for shoulder 72:

This I don't understand. I only follow your topic in general, without going into details. I have enough of my own cockroaches)).

 
Yuriy Asaulenko:

This I do not understand. I only follow your topic in general, without going into detail.

In that topic the difference analogue, that is, the development of an arithmetic progression, and I would like to see it based on a geometric progression.

All the source code is on this page. You just need to program it.

You can easily rework theseMT5,MT4.
 
Aleksey Panfilov:

The topic is a difference analogue, i.e. the development of an arithmetic progression, but I would like to see one based on a geometric progression.

All the sources are on this page. You just need to program them.

You can easily redesign theseMT5,MT4.

I make classical recursive filters - all sorts of Butterworths, Chebyshevs, ... and their modifications. There coefficients are calculated according to known methods, which are not very simple, I can't say.

Your filters are a completely different field and their design is not based on AFC-frequency response. This I can't imagine.

In fact, I've gone to Python.