mixAR_sim: Simulate from MixAR models

In GeoBosh/mixAR: Mixture Autoregressive Models

Simulate from MixAR models

Description

Simulate from MixAR models

Usage

mixAR_sim(model, n, init, nskip = 100, flag = FALSE)

mixAny_sim(model, n, init, nskip=100, flag = FALSE,
                  theta, galpha0, galpha, gbeta)

Arguments

model	model from which to simulate, an object inheriting from class MixAR.
init	initial values, numeric vector.
n	size of the simulated series.
nskip	number of burn-in values, see Details.
flag	if TRUE return also the regimes.
theta	ma coef, a list.
galpha0	alpha0[k], k=1,...,g.
galpha	garch alpha.
gbeta	garch beta.

Details

mixAR_sim simulates a series of length nskip+n and returns the last n values.

mixAny_sim simulates from a MixAR model with GARCH innovations. mixAny_sim was a quick fix for Shahadat and needs consolidation.

The vector init provides the initial values for t=...,-1,0. Its length must be at least equal to the maximal AR order. If it is longer, only the last max(model@order) elements are used.

Value

a numeric vector of length n. If flag = TRUE it has attribute regimes containing z.

Examples

exampleModels$WL_ibm
## simulate a continuation of BJ ibm data
ts1 <- mixAR_sim(exampleModels$WL_ibm, n = 30, init = c(346, 352, 357), nskip = 0)

# a simulation based estimate of the 1-step predictive distribution
# for the first date after the data.
s1 <- replicate(1000, mixAR_sim(exampleModels$WL_ibm, n = 1, init = c(346, 352, 357), 
                      nskip = 0))
plot(density(s1))

# load ibm data from BJ
## data(ibmclose, package = "fma")

# overlay the 'true' predictive density.
pdf1 <- mix_pdf(exampleModels$WL_ibm, xcond = as.numeric(fma::ibmclose))
curve(pdf1, add = TRUE, col = 'blue')

# estimate of 5% quantile of predictive distribution
quantile(s1, 0.05)

# Monte Carlo estimate of "expected shortfall"
# (but the data has not been converted into returns...)
mean(s1[ s1 <= quantile(s1, 0.05) ])

GeoBosh/mixAR documentation built on Oct. 28, 2024, 7:01 p.m.