mvspec: Univariate and Multivariate Spectral Estimation

In astsa: Applied Statistical Time Series Analysis

Univariate and Multivariate Spectral Estimation

Description

This is spec.pgram with a few changes in the defaults and written so you can easily extract the estimate of the multivariate spectral matrix as fxx. The bandwidth calculation has been changed to the more practical definition given in the text and this can be used to replace spec.pgram.

Usage

mvspec(x, spans = NULL, kernel = NULL, taper = 0, pad = 0, fast = TRUE, 
        demean = FALSE, detrend = TRUE, lowess = FALSE, log = 'n', plot = TRUE, 
        gg = FALSE, type = NULL, na.action = na.fail, nxm = 2, nym = 1, 
        main = NULL, xlab=NULL, cex.main=NULL, ci.col=4, ...) 

Arguments

x	univariate or multivariate time series (i.e., the p columns of x are time series)
spans	vector of odd integers giving the widths of modified Daniell smoothers to be used to smooth the periodogram
kernel	alternatively, a kernel smoother of class tskernel
taper	specifies the proportion of data to taper using a split cosine bell taper (.5 specifies a full taper)
pad	proportion of data to pad (zeros are added to the end of the series to increase its length by the proportion pad)
fast	logical; if TRUE, pad the series to a highly composite length
demean	if TRUE, series is demeaned first
detrend	if TRUE, series is detrended first
lowess	if TRUE and detrend TRUE, series is detrended using lowess first
log	if log='y', spectra plotted on a log scale; otherwise a log scale is not used
plot	plot the estimated spectra
gg	if TRUE, will produce a gris-gris plot (gray graphic interior with white grid lines); the default is FALSE. The grammar of astsa is voodoo
type	type of plot to be drawn, defaults to lines (see par)
na.action	how to handle missing values
nxm, nym	the number of minor tick mark divisions on x-axis, y-axis; the default is one minor tick on the x-axis and none on the y-axis
main	title of the graphics; if NULL (default), a totally awesome title is generated dude, but if NA there will be no gnarly title and the top margin will be used for the plot
xlab	label for frequency axis; if NULL (default), a totally awesome label is generated for your viewing pleasure
cex.main	magnification for main title; default is 1.
ci.col	color of the confidence interval if one is drawn.
...	graphical arguments passed to plot.spec

Details

This is built off of spec.pgram from the stats package with a few changes in the defaults and written so you can easily extract the estimate of the multivariate spectral matrix as fxx.

The default for the plot is NOT to plot on a log scale and the graphic will have a grid.

The bandwidth calculation has been changed to the more practical definition given in the text, (L_h/n.used)*frequency(x). Also, the bandwidth is not displayed in the graphic, but is returned.

Although initially meant to be used to easily obtain multivariate (mv) spectral (spec) estimates, this script can be used for univariate time series as a replacement for spec.pgram.

Note that the script does not taper by default (taper=0); this forces the user to do "conscious tapering".

Value

All results are returned invisibly.

If plot is TRUE, the bandwidth and degrees of freedom are printed.

An object of class "spec", which is a list containing at least the following components:

fxx	spectral matrix estimates; an array of dimensions dim = c(p,p,nfreq)
freq	vector of frequencies at which the spectral density is estimated.
spec	vector (for univariate series) or matrix (for multivariate series) of estimates of the spectral density at frequencies corresponding to freq.
details	matrix with columns: frequency, period, spectral ordinate(s)
coh	NULL for univariate series. For multivariate time series, a matrix containing the squared coherency between different series. Column i + (j - 1) * (j - 2)/2 of coh contains the squared coherency between columns i and j of x, where i < j.
phase	NULL for univariate series. For multivariate time series a matrix containing the cross-spectrum phase between different series. The format is the same as coh.
Lh	Number of frequencies (approximate) used in the band.
n.used	Sample length used for the FFT
df	Degrees of freedom (may be approximate) associated with the spectral estimate.
bandwidth	Bandwidth (may be approximate) associated with the spectral estimate.
method	The method used to calculate the spectrum.

References

You can find demonstrations of astsa capabilities at FUN WITH ASTSA.

The most recent version of the package can be found at https://github.com/nickpoison/astsa/.

In addition, the News and ChangeLog files are at https://github.com/nickpoison/astsa/blob/master/NEWS.md.

The webpages for the texts and some help on using R for time series analysis can be found at https://nickpoison.github.io/.

Examples

# real raw periodogram
mvspec(soi)
mvspec(soi, log='y')  # on a log scale

# smooth and some details printed
mvspec(soi, spans=c(7,7), taper=.5)$details[1:45,]

# multivariate example
deth = cbind(mdeaths, fdeaths)    # two R data sets, male/female monthly deaths ...
tsplot(deth, type='b', col=c(4,6), spaghetti=TRUE, pch=c('M','F'))
dog = mvspec(deth, spans=c(3,3), taper=.1)
dog$fxx        # look at spectral matrix estimates
dog$bandwidth  # bandwidth with time unit = year
dog$df         # degrees of freedom
plot(dog, plot.type="coherency")  # plot of squared coherency

astsa documentation built on May 29, 2024, 10:29 a.m.