Generate Stationary Gaussian Process Using Hosking's Method

Uses exact time-domain method from Hosking (1984) to generate a simulated time series from a specified autocovariance sequence.

hosking.sim(n, acvs)

Arguments

n	Length of series.
acvs	Autocovariance sequence of series with which to generate, must be of length at least n.

Value

Length n time series from true autocovariance sequence acvs.

References

Hosking, J. R. M. (1984) Modeling persistence in hydrological time series using fractional differencing, Water Resources Research, 20, No. 12, 1898-1908.

Percival, D. B. (1992) Simulating Gaussian random processes with specified spectra, Computing Science and Statistics, 22, 534-538.

Author

Brandon Whitcher

Examples

dB <- function(x) 10 * log10(x)
per <- function (z) {
  n <- length(z)
  (Mod(fft(z))^2/(2 * pi * n))[1:(n%/%2 + 1)]
}
spp.sdf <- function(freq, delta, omega)
  abs(2 * (cos(2*pi*freq) - cos(2*pi*omega)))^(-2*delta)
data(acvs.andel8)
n <- 1024
if (FALSE) {
z <- hosking.sim(n, acvs.andel8[,2])
per.z <- 2 * pi * per(z)
par(mfrow=c(2,1), las=1)
plot.ts(z, ylab="", main="Realization of a Seasonal Long-Memory Process")
plot(0:(n/2)/n, dB(per.z), type="l", xlab="Frequency", ylab="dB",
     main="Periodogram")
lines(0:(n/2)/n, dB(spp.sdf(0:(n/2)/n, .4, 1/12)), col=2)
}