dfosr_nig: MCMC Sampling Algorithm for the Dynamic Function-on-Scalars...
In drkowal/dfosr: Dynamic Function-on-Scalars Regression

Runs the MCMC for the dynamic function-on-scalars regression model based on an reduced-rank expansion. Here, we assume the factor regression has AR(1) errors. This particular sampler loops over the k=1,...,K factors, so the sampler is O(T*K*p^3) instead of O(T*(K*p)^3). For this variation, we assume normal-inverse-gamma priors on the innovations.

dfosr_nig(Y, tau, X = NULL, K = NULL, use_dynamic_reg = TRUE,
  nsave = 1000, nburn = 1000, nskip = 2, mcmc_params = list("beta",
  "fk", "alpha", "mu_k", "ar_phi"), X_Tp1 = 1, use_obs_SV = FALSE,
  includeBasisInnovation = FALSE, Con_mat = NULL)

`Y`	the `T x m` data observation matrix, where `T` is the number of time points and `m` is the number of observation points (`NA`s allowed)
`tau`	the `m x d` matrix of coordinates of observation points
`X`	the `T x p` matrix of predictors; if NULL, only include an intercept
`K`	the number of factors; if NULL, use SVD-based proportion of variability explained
`use_dynamic_reg`	logical; if TRUE, regression coefficients are dynamic (with random walk models), otherwise independent
`nsave`	number of MCMC iterations to record
`nburn`	number of MCMC iterations to discard (burin-in)
`nskip`	number of MCMC iterations to skip between saving iterations, i.e., save every (nskip + 1)th draw
`mcmc_params`	named list of parameters for which we store the MCMC output; must be one or more of "beta" (dynamic factors) "fk" (loading curves) "alpha" (regression coefficients; possibly dynamic) "mu_k" (intercept term for factor k) "ar_phi" (AR coefficients for each k under AR(1) model) "sigma_et" (observation error SD; possibly dynamic) "Yhat" (fitted values) "Ypred" (posterior predictive values) "Yfore" (one-step forecast; includes the estimate and the distribution)
`X_Tp1`	the `p x 1` matrix of predictors at the forecasting time point `T + 1`
`use_obs_SV`	logical; when TRUE, include a stochastic volatility model for the observation error variance
`includeBasisInnovation`	logical; when TRUE, include an iid basis coefficient term for residual correlation (i.e., the idiosyncratic error term for a factor model on the full basis matrix)
`Con_mat`	a `Jc x m` matrix of constraints for the loading curves such that `Con_mat*fk = 0` for each loading curve `fk`; default is NULL for no constraints.