Download MetaTrader 5

Fast Fourier Transform - Cycle Extraction - page 5

To add comments, please log in or register
codobro
53
codobro  

Thank you SIMBA,

My first question is why have you chosen 720? You had mentioned using 200-210 bar samples previously. What do you think of using around 11.08.08, the last trough?

I have loaded the Goertzel indicator, squared=false with a 720 sampleset. Measuring from the highest amplitude line:

2008.06.08 16:00

Amplitude 5.2746

Phase 2.1042

PeakPer 0.0514

The amplitude fits within yours, and the peak from the sampleset looks within yours as well. What of my phase reading? It looks dramatically different, though I could understand having a larger room for discrepant phase value. Would you consider this too much discrepancy for effective application?

I certainly think phase has relevance, allowing for attenuation. I'm thinking about Ehlers mention of recuding phase lag. How would this be properly applied?

I have 36.504, which corresponds to the spectral analysis of 720 inside the digital filter program.

However loading all data shows that as a trough, as well as 25. 47 seems to correlate in the same way, but with near 0 amplitude. Very interesting numbers, I used 660 bars as well, 25 and 47 seem to show up, and 35/37.

codobro
53
codobro  

Ssa-mtm

About the toolkit, barnix posted a good use. Here is a little explaination:

SST-MTM is a toolkit, like an addon, for a program like Grace.

Grace has to be compiled to run, which is for Linux, but can be run under Win32. Grace FAQ (for Grace-5.1.22)

Cygwin is a program that runs Linux functions under Win32 and grace comes with it. Cygwin FAQ

If you do get this, a few things to note about installation:

Do not install to C:\cygwin for example. Instead, use something like C:\CYGWIN_INSTALL\Cygwin and use this for local install as well. It will prevent conflicts from Window's root files.

Under Install options, Catagory, the X11 Catagory has Grace as an optional install.

You'll need Spectra_st_linux32.exe if you plan to use this method. It seems less complex than installing another OS.

SIMBA
2078
SIMBA  

Thanks for fuel

Variety of sources is the fuel of achievement...;)

Codobro and barnix,

Thanks for your comments on the SSA toolkit installation.

Codobro:Just try it and see if you find it useful,different datasets might give us slightly different phases...if you have similar period and amplitude,that looks robust enough to be useful..

My question about phase was regarding to its uses as a trigger...where is the trigger point?

720 is a good number when working with cycles,since essentially we are trying to convert price changes into degrees of a circle that moves forward in time,so 360,720,etc are good starting points...I have tried to use 1440,for h4 and couldn`t,so I stick to 720....You can use 500,450 or whatever you want,there really isn`t any rational reason to stick with 720,besides the fact that,in my experience ,I get better results with that period...you just try and decide about your choice period....since Goertzel works well with noisy data the longer the dataset the better....if the indicator worked with 7200 points,I would run a walk forward in steps of either 720 or 360 periods and check for persistence between the full series(7200) and all its 360/720 subsets...the problem is that the longer cycles probably will not appear on the 360 subsets

I have used and still use 200,250 bars with Mesa to see if they confirm or not the full history analysis..basically do this kind of analysis for 5 September onwards (GBPJPY,GBPUSD,etc)to see if there are hints of a change in paradigm as price evolution may suggest,but this is a secondary type analysis,when it strongly differs from full history analysis you can run 2 simultaneous cycle models and either stay out of the market or decide which of the 2 cyclic models(full history or 200/250 bars) is more in sync with the market....of course filtering this short term analysis with Goertzel helps in eliminating the random cycles from the equation.

barnix
1310
barnix  

Gbpusd 1440

^

|

|

+---------------------------------->

X Axis (cycles/day)

barnix
1310
barnix  

Used Time Series

^

|

|

+---------------------------------->

X Axis (cycles/day)

Files:
barnix
1310
barnix  

Gbpusd 1440 _2

^

|

|

+---------------------------------->

X Axis (cycles/day)

barnix
1310
barnix  

Gbpusd 1440 _3

codobro
53
codobro  

Thank you Simba, I was trying to use the phase in some sort of 360' offset. I hadn't thought to apply it to the amount of bars

What about taking a higher resonant cycle, a lower resonant cycle, taking the lag and smoothing it with the averaged phase? This has roughly been my application from the Ehlers paper.

I have to test some more of what I'm wondering.

Thank you barnix for posting those images. I have installed cygwin with grace, but have not run it yet.

SIMBA
2078
SIMBA  
barnix:
Gbpusd 1440 _2

^

|

|

+---------------------------------->

X Axis (cycles/day)

Very interesting... barnix,For your MTM analysis there are 3 peaks above the 99% confidence level..approx at 0.035,0.135 and 0.49...Since your data is daily..that means cycle periods of approx 30,7.5 and 2 days...Can you confirm?

SIMBA
2078
SIMBA  

Barnix ssa

barnix:
Gbpusd 1440 _3

Barnix,

Could you apply Mesa to the reconstructed SSA series ?

To add comments, please log in or register