estimate_tvp: Estimation of time-varying parameter models with latent...

In gregorkastner/threshtvp: Bayesian Estimation of (Threshold) Time-Varying Parameter Models

Description

Estimates a time-varying parameter model with stochastic volatility and a set of latent thresholds that control the time-variation and account for model uncertainty for each parameter.

Usage

estimate_tvp(Y, X, save = 5000, burn = 5000, Q_prmean = 0.01, Qprvar = 30, theta_0 = 1, theta_1 = 1, prior_phi = c(25, 5), prior_mu = c(0, 10), tau0 = 0, sv_on = 0, TVS = 1, alpha1 = 0.01, alpha2 = 0.01, beta0 = 1.5, beta1 = 1)

Arguments

Y	Matrix of endogenous variables of dimension T x M. If M>1 then the algorithm estimates a TVP-VAR-SV with latent thresholds.
X	Matrix of explanatory variables of dimension T x K. Has to be supplied by the user.
save	Number of saved draws from the posterior. Default is 5000.
burn	Number of burn-ins. Default is 5000.
theta_0	A value between 0 and 1. A value of 1 implies that the full range of absolute parameter values is used to construct a prior and the proposal for the underlying MH algorithm for the variable inclusion threshold. Default is 1.
theta_1	A value between 0 and 1. A value of 1 implies that the full range of absolute parameter changes is used as a proposal for the MH algorithm for the threshold controlling the time variation. Default is 1.
prior_phi	A numeric vector of length two. Controls the prior moments of the Beta distribution used for the autoregressive parameter of a stochastic volatility model. Defaults to c(25,5)
prior_mu	A numeric vector of length two. Mean and variance for the normally distributed prior on the level of the log-volatily. Defaults to c(0,10)
sv_on	sv_on=1 turns on stochastic volatility as in Kastner & Frühwirth-Schnatter (2013). If sv_on=0 then a homoscedastic model is fitted with a standard inverted gamma prior on the innovation variance of the observation equation.
TVS	TVS=1 allows for shrinkage on the time-variation of the parameters. If TVS=0 the model of Nakajima/West (2013, JBES) arises.
alpha1	Scalar hyperparameter used for the Inverse Gamma prior on the innovation variance in the observation equation. Default is 0.001.
alpha2	Scalar hyperparameter used for the Inverse Gamma prior on the innovation variance in the observation equation. Default is 0.001.
beta0	Scalar hyperparameter for the Gamma prior on the threshold controlling the time-variation. Default is 1.5.
beta1	Scalar hyperparameter for the Gamma prior on the threshold controlling the time-variation. Default is 1.
kappa_l	Scalar hyperparameter. Specifies the prior shape and scale. Defaults to a_kappa=1.
zeta_l	Scalar hyperparameter. Specifies the prior shape and scale. Defaults to a_zeta=1.

Author(s)

Florian Huber

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--	or do  help(data=index)  for the standard data sets.

## The function is currently defined as
function (Y, X, save = 5000, burn = 5000, Q_prmean = 0.01, Qprvar = 30, 
    theta_0 = 1, theta_1 = 1, prior_phi = c(25, 5), prior_mu = c(0, 
        10), tau0 = 0, sv_on = 0, TVS = 1, alpha1 = 0.01, alpha2 = 0.01, 
    beta0 = 1.5, beta1 = 1) 
{
    parm_list <- list("save", "burn", "Q_prmean", "Qprvar", "theta_0", 
        "theta_1", "prior_phi", "prior_mu", "sv_on", "TVS", "tau0", 
        "alpha1", "alpha2", "beta0", "beta1", "nthin")
    if (length(ncol(Y)) == 0) 
        Y <- as.matrix(Y)
    K <- ncol(X)
    M <- ncol(Y)
    T <- nrow(Y)
    if (M > 1) {
        post_draws <- list()
        sfInit(parallel = TRUE, cpus = 4)
        sfExport(list = c(list("Y", "X", "KF", "MCMC_tvpm", "get_variable_incl", 
            "get_moving_incl"), parm_list))
        post_draws <- sfLapply(1:ncol(Y), function(i) MCMC_tvpm(Y, 
            X, nburn = burn, nsave = save, B0_inv = B0inv, Qprmean = Q_prmean, 
            Q_prvar = Qprvar, theta_incl = theta_0, theta_dyn = theta_1, 
            priorphi = prior_phi, priormu = prior_mu, sv_on = sv_on, 
            TVS = TVS, nr = i, tau_0 = tau0, a1 = alpha1, a2 = alpha2, 
            b0 = beta0, b1 = beta1))
        sfStop()
        get_var_post <- VAR_posterior(post_draws, nthin * save)
    }
    else {
        post_draws <- MCMC_tvpm(Y, X, nburn = burn, nsave = save, 
            B0_inv = B0inv, Qprmean = Q_prmean, Q_prvar = Qprvar, 
            theta_incl = theta_0, theta_dyn = theta_1, priorphi = prior_phi, 
            priormu = prior_mu, sv_on = sv_on, TVS = TVS, nr = 1, 
            tau_0 = tau0, a1 = alpha1, a2 = alpha2, b0 = beta0, 
            b1 = beta1)
    }
    return(list(posterior = post_draws, args = .construct.arglist(MCMC_tvpm)))
  }

gregorkastner/threshtvp documentation built on Feb. 26, 2021, 12:25 a.m.