Samples the degrees of freedom parameter of standardized and homoskedastic t-distributed input variates. Marginal data augmentation (MDA) is applied, tau is the vector of auxiliary latent states. Depending on the prior specification, nu might not be updated, just tau.
update_t_error( homosked_data, tau, mean, sd, nu, prior_spec, do_tau_acceptance_rejection = TRUE )
homosked_data | de-meaned and homoskedastic observations |
---|---|
tau | the vector of the latent states used in MDA. Updated in place |
mean | the vector of the conditional means // TODO update docs in R |
sd | the vector of the conditional standard deviations |
nu | parameter nu. The degrees of freedom for the t-distribution. Updated in place |
prior_spec | prior specification object. See type_definitions.h |
do_tau_acceptance_rejection | boolean. If |
The function samples tau and nu from the following hierarchical model: homosked_data_i = sqrt(tau_i) * (mean_i + sd_i * N(0, 1)) tau_i ~ InvGamma(.5*nu, .5*(nu-2)) Naming: The data is homoskedastic ex ante in the model, mean_i and sd_i are conditional on some other parameter in the model. The prior on tau corresponds to a standardized t-distributed heavy tail on the data.
Other stochvol_cpp:
update_fast_sv()
,
update_general_sv()
,
update_regressors()