test_script_mskf.R

In mskf: Markov Switching Kalman Filter

testfunction <- function() {
	simres = replicate(100, {
		set.seed(123)
		# parameters
		phi01 = 0       # constant in regime 1
		phi02 = 3       # constant in regime 2
		phi11 = 0.5     # AR(1) parameter in regime 1
		phi12 = 0.3     # AR(1) parameter in regime 2
		rsd1  = 1       # residual standard deviation in regime 1
		rsd2  = 1       # residual standard deviation in regime 2

		# number of time point and burn in
		nt = 1350
		burn.in = 100

		# Markov transition matrix
		p = matrix(c(.7,.3,.2,.8), 2, 2)
		lp = eigen(p)$vectors[,1]
		lp = lp / sum(lp) # equilibrium regime distribution

		true.par = c(phi01, phi02, phi11, phi12, rsd1, rsd2, log(p[1,1] / p[2,1]),
								 log(p[2,2] / p[1,2]))

		# Create a discrete Markov series
		r = sample(1:2, 1, p = lp);
		for (t in 2:(nt + burn.in))
			r[t] = sample(1:2, 1, prob = p[,r[t - 1]]);
		y = rnorm(1)
		for (t in 2:(nt + burn.in))
			y[t]  = if (r[t] == 1)
				phi01 + phi11 * y[t - 1] + rnorm(1,0,rsd1)
		else
			phi02 + phi12 * y[t - 1] + rnorm(1,0,rsd2)
		y <- y[-(1:burn.in)]


		# Specify the model
		skel = mskfModelSkeleton(y, ne = 1, nm = 2)
		start = c(
			`1` = .05, `2` = 3, `3` = .51, `4` = .22, `5` = 1.5, `6` = 2,
			`7` = 1, `8` = .5
		)
		mod = mskfModel(
			skel, pap = diag(7:8), paK = 5:6, pac = 1:2, paH = 3:4, start = start,
			upper = 50, lower = -50
		)
		fitres <- mskf(mod,control = list(trace = 3,maxit = 2000),debug = 0)
		fitres
	})
	simres
}