Spectral Analysis*

[NicolasR]

This thread is about the Spectral Analysis add-in which calculates various spectral analysis tools for time series :series:. It's an improved version of the periodogram add-in, it has new methods to calculate spectal density and a significance test signal that could be used as a white noise test.

Regards, Nicolas Ronderos Pulido

[ecofin]

It is exactly what I need, Thank you for all your efforts, I'm :D .

[trubador]

Yes, a very comprehensive and useful add-in. Thanks for sharing...

[felix.casares]

i don´t know much about spectral analysis, can i use one of any elements ofs this to make hp filter for daily frequency time serie?

please help

[NicolasR]

The Hodrick Prescott filter is a low pass filter, that is, it preserves the low frequency components in the spectrum (long run). Therefore, you can use the estimated spectrum of your series to see if there is any significant peak at the low frequency components, if there is not, the smoothed series with de HP filter would be based on artificially inflated frequency components.

Hope it helps.

Regads.

[NicolasR]

The version 2.0 of the spectral analysis add-in is now available. It contains tests for detecting periodic components, white noise and Gaussian white noise, also it performs an automatic procedure for extracting cycles of a time series, named significant pass filter (SPF). The add-in could also be used to perform dynamic Fourier analysis, the program file can be found below and in the add-in file. In the PDF document of the add-in are showed the spectral theory of economic series and some examples using the add-in applied to data of the Industrial Production Index, the Chicago Board Options Exchange S&P 100 volatility index (VXO) and the exchange rate Euro/Dollar.

Regads.

Code: Select all

mode quiet '---------------------------------------------------------------Dynamic Spectra Inputs---------------------------------------------------------------- !step=1 'Step parameter for the rolling !window=100 'Number of data that will be used to estimate the spectrum '-----------------------------------------------------------------Static Spectral Inputs------------------------------------------------------------------ %series="vxo" %s_window="Bartlett-Priestley" 'Periodogram, Hamming, Hann, Bartlett, Parzen, Truncated, Daniell, Bartlett-Priestley !criteria=0.05 'Selection criteria for the spectral filter %log_scale="N" 'Y/N !truncation=20 'Truncation point '---------------------------------------------------------------------------------------------------------------------------------------------------------------- !obs=@obs({%series}) !start=@ifirst({%series}) !end=@ilast({%series}) !nrolls=(!obs-!window)/!step !loop=0 for !i=0 to !obs-!window step !step !loop=!loop+1 %first = @otod(!start+!i) %last = @otod(!start+!i+!window) smpl {%first} {%last} if %log_scale="N" then {%series}.spectral({%s_window},table,truncation=!truncation,criteria=!criteria) else if %log_scale="Y" then {%series}.spectral({%s_window},table,truncation=!truncation,criteria=!criteria,log) endif endif for !j=1 to !window/2 table(!window/2+1,!nrolls+2) dynamic(1,!loop+1)=@pagesmpl table(!window/2+1,!nrolls+2) dynamic(!j+1,1)=data(!j+1,2) table(!window/2+1,!nrolls+2) dynamic(!j+1,!loop+1)=data(!j+1,4) next for !j=!start to @obsrange if !criteria<1 then if sf_wsum(!j)<>NA then vector(@obsrange,1) cycle(!j,1)=sf_wsum(!j) endif else if mf_wsum(!j)<>NA then vector(@obsrange,1) cycle(!j,1)=mf_wsum(!j) endif endif next if !criteria<1 then %s_m="s" else %s_m="m" endif if %s_window="Periodogram" then delete data {%s_m}f_wsum {%s_window}01 else if %s_window="Bartlett-Priestley" then %priestley=@right(%s_window,9) delete data periodogram_{%priestley} {%s_m}f_wsum else delete data periodogram_{%s_window} {%s_m}f_wsum endif endif scalar roll=@floor((!loop/!nrolls)*100) !roll_=roll statusline Rolling Percentaje: (!roll_%) next smpl @all mtos(cycle,stochastic_cycle) delete roll cycle '----------------------------------------------------------------------------------------------------------------------------------------------------------------

[lofficer]

I have version 2.0 of spectralanalysis. In order to do cross-spectral analysis, I need sine and cosine coefficients by frequency, not just plotted. Please help. Explain how I can do this without entering your program. If I alter your program, there will be future issues when you alter your program and I want to stick to your program. Thanks.

[NicolasR]

Hi lofficer,

The add-in only performs univariate spectral analysis, in fact I´m working on a cross spectral analysis add-in I belive it will be ready in a couple of months.

[lofficer]

Thanks. I look forward to your cross-spectral program.

[clubmed]

thanks for the STR Add-in it's very useful
i have two suggestions,
1-can you improve this Add-in to calculate (views the series of cumulative and frequency) and show the Cumulative spectral distribution (graph).
2-EViews 9 beta can now estimate the ARFIMA model, can you write an Add-in test of GPH (Geweke-Porter-Hudak test long memory).

Kind Regards. MED

[NicolasR]

The cumulative spectrum could be calculated simply with the command @cumsum(%spectrum_name). The test sounds interesting but it is not in my plans, sorry.

Best regards.

Nicolas.

[StoyanKiryazov]

Hi Nicolas,

many thanks for the great add-in. It is really helpful. May I ask if you have any progress with the cross-spectral tools? A few months ago you mentioned you were working on this.

Thank you.

Kind regards,
Stoyan

[nicotrader]

Hi all, i have a problem with spectral analysis add in.
I work with high frequency financial observation.
Whit more 10.0000 observation the system returns error "OUT OF MEMORY MATRIX"
Can you fix this problem?

[NicolasR]

Hi,

If you are working with 100.000 observations the add-in will not work. However if you click options-->General options-->Advanced system options and change the memory reserved for windows to 200 you will probably can work with 12000 observations (in a 32 bit version). Increase the temporal aggregation will afect the short-run cycle that you might find, In financial data I doubt that there is a cicle of this type. What are you trying to find or prove?

Regards.

[nicotrader]

Nicolas i m work with high frequency data (5min rivelation), in one day, for exp. Sep500 future there are 250 5min rivelation multiply for cycle of 30 day or more and get a 7500 observation, multiply for back test 2 years at least.....

I think software should allow econometric analysis of time series high frequency, everything else is the past.
I urge you to implement the software. I paid this software 1750 Dollars!