BMTAR (version 0.1.1)

mtarstr: Estimation of structural parameters of MTAR model

Description

Estimate structural and non-structural parameters of a MTAR model when the number of regimes is fixed.

Usage

mtarstr(ini_obj, level = 0.95, niter = 1000, burn = NULL, chain = FALSE,
r_init = NULL, parallel = FALSE)

Arguments

ini_obj

class “regime_inipars” object, here specificate in pars: l known, orders not known. Not NULL. Default for l = 2, orders = list(pj = c(2,2)) and method = 'KUO'

level

numeric type, confident interval for estimations. Default 0.95

burn

numeric type, number of initial runs. Default NULL (30% of niter)

niter

numeric type, number of runs of MCMC. Default 1000

chain

logical type, if return chains of parameters. Default FALSE

r_init

numeric type of length l - 1. If r not known, starting value of the chain. Default NULL

parallel

logical type, if package parallel should be used. Default FALSE

Value

Return a list type object of class “regime_model”

Nj

number of observations in each regime

estimates

list for each regime with confident interval and mean value of the parameters

regime

“regime” class objects with final estimations

Chain

if chain TRUE list type object with parameters chains

fitted.values

matrix type object with fitted.values of the estimated model

residuals

matrix type object with residuals of the estimated model

logLikj

log-likelihood of each regime with final estimations

data

list type object $Yt and $Ut = (Zt,Xt)

r

final threshold value estimation with acceptance percentage

orders

list type object with names (pj,qj,dj) final estimations

Details

If the number of regimes \(l\) is known or fixed, we can estimate other structural parameters of the MTAR model: Thresholds(\(r_1,\cdots,r_{l-1}\)), and autoregressive orders(\(p_j,q_j,d_j\)). Of course, the non-structural parameters are also estimated. The problem of estimation the autoregressive orders is addressed to the problem of Bayesian variable selection in regression using Gibbs Variable selection(GVS) or Kuo and Mallick Methodologies. Samples of the full conditional distribution for Threshold values are extracted using Random Walk Metropolis-Hastings Algorithm.

References

Calderon, S. and Nieto, F. (2017) Bayesian analysis of multivariate threshold autoregress models with missing data. Communications in Statistics - Theory and Methods 46 (1):296<U+2013>318. doi:10.1080/03610926.2014.990758.

Examples

# NOT RUN {
data("datasim")
data = datasim
# KUO method
initial = mtarinipars(tsregime_obj = data$Sim,method = 'KUO',
list_model = list(pars = list(l = 2),orders = list(pj = c(2,2))))
# }
# NOT RUN {
estruc = mtarstr(ini_obj = initial,niter = 500,chain = TRUE)
autoplot.regime_model(estruc,1)
autoplot.regime_model(estruc,2)
autoplot.regime_model(estruc,3)
autoplot.regime_model(estruc,4)
autoplot.regime_model(estruc,5)
# }
# NOT RUN {
# method can also be 'SSVS'
# }