Proc spectra is used to perform frequency domain analysis of time series data. Time domain (Box-Jenkins) analysis of time series is done using Proc Arima.
The main objective of frequency domain analysis is to identify periodicities in the time series. The modeling options available in proc arima are not available within proc spectra.
Typical invocation is:
proc spectra
data=skin
center
whitetest
p
s
out=spec;
var sun; /* required */
weights 1 2 3 2 1;
run;
The center option subtracts the mean of the time series from the
observations before computing the periodogram.
The whitetest option performs Fisher's test for white noise.
The p option computes the periodogram of the series.
The s option computes the estimate of the spectral density of
the series. If the s option is used, the weights option is required.
The out= option creates a SAS data set containing the
output of the analysis. If the p option is specified, the
periodogram values are included in the out data set; if the s
option is specified, the estimates of the spectral density are
included. The periodogram and spectral density estimates are in
variables named p_01, p_02,... and s_01, s_02,...
according to the order in which variables are listed on the var=
statement. The out data set includes variables freq and
period to allow for easy plotting of the periodogram or
spectral density as a function of either frequency or period.
The var= statement specifies the variable(s) to be analyzed.
This statement is required.
The weights= statement is required if the s option is used.
Weights proportional to those specified are used to define a
window and smooth the periodogram to produce the estimate of
the spectral density.
For further information, see the SAS/ETS Guide.
Copyright © 1997 by Jerry Alan Veeh. All rights reserved.