EViews.com

This thread is about Periodogram* Add-in which calculates the estimated spectrum of a time series series object.

Very nice and useful add-in. Thanks for sharing.

Hi

I've just tried this add-in with US industrial production data. I couldn't replicate the periodogram in the accompanying documentation either in scale or peaks. I didn't get the nice sharp well-defined peaks seen in graph 5.

I calculated the percentage MoM change as 100*dlog(indpro) from 1919 to November 2013 (which seems to be how it was calculated in the documentation).

Any ideas?

Hi,

I do it with genr ip_mom=@pc(ip).And the series source is: http://www.econstats.com/ipcu/ip_m1.htm the nsa.

Got it and replicated your result.

Hi

I should have added that I used SA data in my first estimation. I probably didn't read the documentation carefully enough (a bad habit I have).

thank's for this add-in , just i have one suggestion if in next version of Periodogram (update version) can calculte spectral density and show the spectral density like oxmetrics software.
best regard.

I don’t know OxMetrics. But the Periodogram is the nonparametric estimator of the population spectrum, perhaps what is needed is the estimated periodogram with kernel (different windows, I'm working on it). I'm not sure if you were talking about this. I would appreciate if you send me papers or literature if not. Thanks.

thank's for your reply
yes I'm talking about the different windows
for OxMetrics 6 it will plot spectral density and periodogram in separate graphs for the serie it use probably Parzen windows (see the picture), the question is I can do periodogram and spectral density with defferent windows?
I will send you some papers in next time
best regards. ecofin

this is a papers of spectral density see chapter 7 page 121 the analysis of time series an introduction 6th ed, Chris Chatfield, CHAPMAN AND HALL/CRC, I have some suggestions if you can work some different windows like:
a/Parzen
b/Tukey-Hanning (Tukey-Hann)
c/Tukey-Hamming
d/Daniell
e/Quadratic Parzen
f/Bartlett
g/Truncated

with a more options like:
1/number of points (intervals) like: auto or custom, series length.
2/bandwith: auto or custom, log scale.
3/calculated: omega, scaled frequency, periods, spectral density (in series output)
if there are other windows I'm appreciate to add it in next version of the add-in.
it's just my opinion

some books:
1-Time Series Analysis and Its Applications With R Examples 3rd ed, Robert H. Shumway David S. Stoffer, Springer. (chapter 4)
2-Time Series Analysis With Applications in R, 2nd ed, Jonathan D. Cryer Kung-Sik Chan, Springer. (chapter 13 and 14)
3-Introductory Time Series with R, 1st ed, Paul S.P. Cowpertwait · Andrew V. Metcalfe, Springer. (chapter 9)
4-Time Series: Theory and Methods, 2nd ed, Peter J. Brockwell Richard A. Davis, Springer. ( chapter 4 and 10)
5-Introduction to Time Series and Forecasting, 2nd ed, Peter J. Brockwell Richard A. Davis, Springer. (chapter 4)
6-the spectral analysis of time series: probability and mathematical statistics volume 22, Acadimic Press, Lambert H. Koopmans.

finally the next verion of add-in will contain: data periodogram and estimate of A and B (like this version), and different windows (spectral density) with options.

best regrads. ecofin

thanks for the papers, suggestions and ideas. I will finish soon. thank you again.

it's me that I must thank you again
other books if you are interested for mutivariate series in spectral density
1-Spectral Analysis and time series: univariate series, volume 1, M. B.Priestley, ACADIMIC PRESS 1981.
2-Spectral Analysis and time series: multivariate series, volume 2, M. B.Priestley, ACADIMIC PRESS 1981.
3-Univariate Time Series in Geosciences: Theory and Examples, Hans Gilgen, Springer 2006. (chapter 6-7-8-9).

I have simple suggestion, if the next version will calculate and graph periodogram and spectral density with different windows for equation object for ARMA model and for group series, for all objects for example: LogL object state space object...etc?
kind regards.

Hi,

I don't understand the suggestion. Do you mean for the residuals from an equation or state-space representation? or for a series simulated with an ARMA(p,q) data generating mechanism?

Regards.

This add-in calculates the estimated spectrum of a series object. if the next version will support equation object, state space object, logL object (this is my suggestion).
for example I estimate ARMA(2,1) next I chose PROC next add-in next periodogram, the same for state space and logL objects.
have you understand me.
kind regrads