Adaptive lookback indicators - page 16

New comment
 
MrM:
Simba, if I understand correctly you would only select the statistically significant cycles, advance their phases and combine them into 1 wave? Is that screenshot from your post above from a causal indi?

The screenshot is the FFT endpointed by mladen with different settings....For example use it on EURUSD H4 with 0,64,3,4,60 and you will see what I mean...there is a 17-20 bars cycle at EU H4,now, and using those settings the FFT gives back the slope of any cycle with periodicity 16 to 21,it signals entries with 1 to 2 bars of delay,which is very good and can be anticipated with experience and lower timeframes..And,disregard the past turns besides the nearest 2,this slope is valid as long as the cycle periodicity stays within 16-21,so,you want it to adapt to the actual cycles,no matter how ugly the indicator looks in the past.

I would select the right cycles,IMO,that means they would have to fullfill at least 2 conditions,passing Bartels and having good amplitude.

Combining them into 1 composite wave or using different waves are 2 options I would explore,personally I prefer single waves,they are easier to extrapolate into the next nesting area,but this is a matter of personal preferences and trading style.

S

Files:
screen_2.gif  54 kb
 
SIMBA:
For example use it on EURUSD H4 with 0,64,3,4,60 and you will see what I mean...there is a 17-20 bars cycle at EU H4,now, and using those settings the FFT gives back the slope of any cycle with periodicity 16 to 21,it signals entries with 1 to 2 bars of delay,which is very good and can be anticipated with experience and lower timeframes..And,disregard the past turns besides the nearest 2,this slope is valid as long as the cycle periodicity stays within 16-21,so,you want it to adapt to the actual cycles,no matter how ugly the indicator looks in the past.
" using those settings the FFT gives back the slope of any cycle with periodicity 16 to 21"

then why FilterAbove and FilterBellow 3 & 4 and not 16 & 21? Ideally this indicator would not only show the last 2 turns correctly..

Files:
fft_end-point.jpg  198 kb
 
 

Dear mladen

Could you make the SSA end-pointed in mode advanced just like the SSA advanced, in the way that it changes color acoording the trend. If you can please post it.

Thanks.

mladen:
MrM

This post is going to answer your question but I think it will answer some more (unasked by you or by anybody else so far, for which I thank all of you) questions too ...

I guess that you all know that a couple of days ago I posted the end-point SSA at the public thread. I assume that some of you guys were even angry at me why I did not post it here (in the advanced elite section) instead of posting it in the public section. Apart from the reasons stated at that post, will add some more arguments why I posted it there but also why am I posting this post here and not at the public section.

____________________________

For start : I already told over and over that centered triangular moving average and SSA are surprisingly similar in values (surprisingly because the math is completely different) Here is a comparison of a Lag 25, 1 computation SSA (blue) and 25 HalfLength centered triangular MA (orange)
As you can see only the last (extrapolated in TMA and last LAG bars in SSA) differ slightly but with time they tend to equalize. For every additional computation of SSA it is rather easy to to get the very similar TMA (for 2 computations simply divide SSA Lag by 2 and apply it to centered TMA and there you go). Now, you may be wondering why am I talking so much about similarities of SSA ant centered TMA. Here is the reason - here is how end-point centered triangular moving average (blue) looks like compared to centered triangular MA (orange)
Before I make a POINT here is some more explanation

As you know, end-pointing is rather a simple idea : for indicators that use future values in calculation we extract a "window" of data from the data we want to calculated end-point of, calculate it (by windowing we made sure that future data is not available to calculation) and use only the last value as a value of indicator, and that way we ensure it is causal (non-repainting, non-recalculating, whichever expression one likes more in this case - I think that repainting is a coding error and that is why I use recalculating for correctly coded indicators)

And in the end : the POINT. On the upper picture red line is the end-point version of centered triangular moving average. But, I did not make a new indicator, nor did I use some indicator made specially for that. That indicator is usually called linear weighted moving average (LWMA) and it is the centered triangular moving average end-pointed. Just add 1 to half length use it as LWMA length and you got it. Observe the rightmost value and you will see that they are the same (in data window you might get a difference that comes out from different precision, but rounded to 4 digits they are the same)

____________________________

Now I hope you know the answer why I posted the end-point SSA where I posted it. They differ, but not too much. But I also thought that this (the little secret) should be told here. They are not the same (due to different ways how the rightmost values are calculated) but are similar (as are TMA and SSA - not as much as when they "stabilize", but they are) Here is the comparison of end-point SSA and end-point centered TMA (so LWMA)
I do not know the state of end-point FFT (working on it in these days, but still have some issues that needs to be solved) but somehow I do not expect too much from it either. As useful as the original tools are, the end-point (in this case at least) is not exactly what we were expecting

____________________________

Now I hope that all of you reading this will understand why I post this post here. In one PM to MrM I told him that I think that learning process is far more valuable than the result itself. This time I decided to share what I have learn so far about this issue, but I decided to share it only here (I leave the myths to myths, but i think that here we do not need so many myths)

regards

Mladen
 

Hi MrM, any news with the patent code?

Regards,

- trendick

MrM:
I'm not writing the matlab code because I'm getting it from the US Patent . I'll post it when I get my CD-R...

About the lag: I know.. the lag would be the number of bars since the last extremum. It's the eternal problem of the right edge of the window..

Another interesting EMD-clone is Hilbert Vibration Decomposition
 
trendick:
Hi MrM, any news with the patent code?

I'll post it when I get it

 

Some more mods , posting the Stlm here my thinking is it is a digital adaptation., and also an non lag version of phase accumulation bbands!

 

xxxxxxxxxxxxxxxxx

 

xxxxxxxxxxxxxxxxxxxxxxxxx

 

Hi,

where can I find the Indicators ( page 138 )

end-point centered triangular moving average

centered triangular MA

Thanks

derumuro

Thanks a lot, of all your working !

1...91011121314151617181920212223...128
New comment