Package ‘bayesianVARs’

July 22, 2025

Title MCMC Estimation of Bayesian Vectorautoregressions
Version 0.1.5
Description Efficient Markov Chain Monte Carlo (MCMC) algorithms for the
fully Bayesian estimation of vectorautoregressions (VARs) featuring
stochastic volatility (SV). Implements state-of-the-art shrinkage
priors following Gruber & Kastner (2023) <doi:10.48550/arXiv.2206.04902>.
Efficient equation-per-equation estimation following Kastner & Huber
(2020) <doi:10.1002/for.2680> and Carrerio et al. (2021)
<doi:10.1016/[Link].2021.11.010>.
License GPL (>= 3)
URL [Link]
[Link]
BugReports [Link]
Depends R (>= 3.3.0)
Imports colorspace, factorstochvol (>= 1.1.0), GIGrvg (>= 0.7),
graphics, MASS, mvtnorm, Rcpp (>= 1.0.0), scales, stats,
stochvol (>= 3.0.3), utils
Suggests coda, knitr, rmarkdown, testthat (>= 3.0.0)
LinkingTo factorstochvol, Rcpp, RcppArmadillo, RcppProgress, stochvol
VignetteBuilder knitr
Config/testthat/edition 3
Encoding UTF-8
LazyData true
RoxygenNote 7.3.1
NeedsCompilation yes
Author Luis Gruber [cph, aut, cre] (ORCID:
<[Link]
Gregor Kastner [ctb] (ORCID: <[Link]
Maintainer Luis Gruber <[Link]@[Link]>
Repository CRAN
Date/Publication 2024-11-13 20:50:02 UTC

1
2 bvar

Contents
bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
coef . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
fitted.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
my_gig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
pairs_predict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
plot.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
plot.bayesianVARs_fitted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
plot.bayesianVARs_predict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
posterior_heatmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
predict.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
print.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
print.bayesianVARs_predict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
[Link].bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
[Link].bayesianVARs_predict . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
specify_prior_phi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
specify_prior_sigma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
stable_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
summary.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
summary.bayesianVARs_draws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
summary.bayesianVARs_predict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
usmacro_growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
vcov.bayesianVARs_bvar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
[.bayesianVARs_coef . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
[.bayesianVARs_draws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

Index 38

bvar Markov Chain Monte Carlo Sampling for Bayesian Vectorautoregres-

sions

Description
bvar simulates from the joint posterior distribution of the parameters and latent variables and returns
the posterior draws.

Usage
bvar(
data,
lags = 1L,
draws = 1000L,
burnin = 1000L,
thin = 1L,
prior_intercept = 10,
prior_phi = specify_prior_phi(data = data, lags = lags, prior = "HS"),
bvar 3

prior_sigma = specify_prior_sigma(data = data, type = "factor", quiet = TRUE),

sv_keep = "last",
quiet = FALSE,
startvals = list(),
expert = list()
)

Arguments
data Data matrix (can be a time series object). Each of M columns is assumed to
contain a single time-series of length T .
lags Integer indicating the order of the VAR, i.e. the number of lags of the dependent
variables included as predictors.
draws single integer indicating the number of draws after the burnin
burnin single integer indicating the number of draws discarded as burnin
thin single integer. Every thinth draw will be stored. Default is thin=1L.
prior_intercept
Either prior_intercept=FALSE and no constant term (intercept) will be in-
cluded. Or a numeric vector of length M indicating the (fixed) prior standard
deviations on the constant term. A single number will be recycled accordingly.
Default is prior_intercept=10.
prior_phi bayesianVARs_prior_phi object specifying prior for the reduced form VAR
coefficients. Best use constructor specify_prior_phi.
prior_sigma bayesianVARs_prior_sigma object specifying prior for the variance-covariance
matrix of the VAR. Best use constructor specify_prior_sigma.
sv_keep String equal to "all" or "last". In case of sv_keep = "last", the default, only
draws for the very last log-variance hT are stored.
quiet logical value indicating whether information about the progress during sampling
should be displayed during sampling (default is TRUE).
startvals optional list with starting values.
expert optional list with expert settings.

Details
The VAR(p) model is of the following form: y ′t = ι′ + x′t Φ + ϵ′t , where y t is a M -dimensional
vector of dependent variables and ϵt is the error term of the same dimension. xt is a K = pM -
dimensional vector containing lagged/past values of the dependent variables y t−l for l = 1, . . . , p
and ι is a constant term (intercept) of dimension M × 1. The reduced-form coefficient matrix Φ is
of dimension K × M .
bvar offers two different specifications for the errors: The user can choose between a factor
stochastic volatility structure or a cholesky stochastic volatility structure. In both cases the distur-
bances ϵt are assumed to follow a M -dimensional multivariate normal distribution with zero mean
and variance-covariance matrix Σt . In case of the cholesky specification Σt = U ′−1 D t U −1 ,
where U −1 is upper unitriangular (with ones on the diagonal). The diagonal matrix D t de-
pends upon latent log-variances, i.e. D t = diag(exp(h1t ), . . . , exp(hM t ). The log-variances
4 bvar

follow a priori independent autoregressive processes hit ∼ N (µi + ϕi (hi,t−1 − µi ), σi2 ) for i =
1, . . . , M . In case of the factor structure, Σt = ΛV t Λ′ + Gt . The diagonal matrices V t
and Gt depend upon latent log-variances, i.e. Gt = diag(exp(h1t ), . . . , exp(hM t ) and V t =
diag(exp(hM +1,t ), . . . , exp(hM +r,t ). The log-variances follow a priori independent autoregres-
sive processes hit ∼ N (µi +ϕi (hi,t−1 −µi ), σi2 ) for i = 1, . . . , M and hM +j,t ∼ N (ϕi hM +j,t−1 , σM
2
+j )
for j = 1, . . . , r.

Value

An object of type bayesianVARs_bvar, a list containing the following objects:

• PHI: A bayesianVARs_coef object, an array, containing the posterior draws of the VAR co-
efficients (including the intercept).
• U: A bayesianVARs_draws object, a matrix, containing the posterior draws of the contempo-
raneous coefficients (if cholesky decomposition for sigma is specified).
• logvar: A bayesianVARs_draws object containing the log-variance draws.
• sv_para: A baysesianVARs_draws object containing the posterior draws of the stochastic
volatility related parameters.
• phi_hyperparameter: A matrix containing the posterior draws of the hyperparameters of the
conditional normal prior on the VAR coefficients.
• u_hyperparameter: A matrix containing the posterior draws of the hyperparameters of the
conditional normal prior on U (if cholesky decomposition for sigma is specified).
• bench: Numerical indicating the average time it took to generate one single draw of the joint
posterior distribution of all parameters.
• V_prior: An array containing the posterior draws of the variances of the conditional normal
prior on the VAR coefficients.
• facload: A bayesianVARs_draws object, an array, containing draws from the posterior dis-
tribution of the factor loadings matrix (if factor decomposition for sigma is specified).
• fac: A bayesianVARs_draws object, an array, containing factor draws from the posterior
distribution (if factor decomposition for sigma is specified).
• Y: Matrix containing the dependent variables used for estimation.
• X matrix containing the lagged values of the dependent variables, i.e. the covariates.
• lags: Integer indicating the lag order of the VAR.
• intercept: Logical indicating whether a constant term is included.
• heteroscedastic logical indicating whether heteroscedasticity is assumed.
• Yraw: Matrix containing the dependent variables, including the initial ’lags’ observations.
• Traw: Integer indicating the total number of observations.
• sigma_type: Character specifying the decomposition of the variance-covariance matrix.
• datamat: Matrix containing both ’Y’ and ’X’.
• config: List containing information on configuration parameters.
bvar 5

MCMC algorithm
To sample efficiently the reduced-form VAR coefficients assuming a factor structure for the
errors, the equation per equation algorithm in Kastner & Huber (2020) is implemented. All
parameters and latent variables associated with the factor-structure are sampled using package
factorstochvol-package’s function update_fsv callable on the C-level only.
To sample efficiently the reduced-form VAR coefficients, assuming a cholesky-structure for the
errors, the corrected triangular algorithm in Carriero et al. (2021) is implemented. The SV param-
eters and latent variables are sampled using package stochvol’s update_fast_sv function. The
precision parameters, i.e. the free off-diagonal elements in U , are sampled as in Cogley and Sargent
(2005).

References
Gruber, L. and Kastner, G. (2023). Forecasting macroeconomic data with Bayesian VARs: Sparse
or dense? It depends! arXiv:2206.04902.
Kastner, G. and Huber, F. Sparse (2020). Bayesian vector autoregressions in huge dimensions.
Journal of Forecasting. 39, 1142–1165, doi:10.1002/for.2680.
Kastner, G. (2019). Sparse Bayesian Time-Varying Covariance Estimation in Many Dimensions
Journal of Econometrics, 210(1), 98–115, doi:10.1016/[Link].2018.11.007.
Carriero, A. and Chan, J. and Clark, T. E. and Marcellino, M. (2021). Corrigendum to “Large
Bayesian vector autoregressions with stochastic volatility and non-conjugate priors” [J. Economet-
rics 212 (1) (2019) 137–154]. Journal of Econometrics, doi:10.1016/[Link].2021.11.010.
Cogley, S. and Sargent, T. (2005). Drifts and volatilities: monetary policies and outcomes in the
post WWII US. Review of Economic Dynamics, 8, 262–302, doi:10.1016/[Link].2004.10.009.
Hosszejni, D. and Kastner, G. (2021). Modeling Univariate and Multivariate Stochastic Volatility
in R with stochvol and factorstochvol. Journal of Statistical Software, 100, 1—34. doi:10.18637/
jss.v100.i12.

See Also
• Helpers for prior configuration: specify_prior_phi(), specify_prior_sigma().
• Plotting: plot.bayesianVARs_bvar().
• Extractors: coef.bayesianVARs_bvar(), vcov.bayesianVARs_bvar().
• ’stable’ bvar: stable_bvar().
• summary method: summary.bayesianVARs_bvar().
• predict method: predict.bayesianVARs_bvar().
• fitted method: fitted.bayesianVARs_bvar().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)
6 coef

# Plot
plot(mod)

# Summary
summary(mod)

coef Extract VAR coefficients

Description
Extracts posterior draws of the VAR coefficients from a VAR model estimated with bvar().

Usage
## S3 method for class 'bayesianVARs_bvar'
coef(object, ...)

Arguments
object A bayesianVARs_bvar object obtained from bvar().
... Currently ignored.

Value
Returns a numeric array of dimension M × K × draws, where M is the number of time-series, K
is the number of covariates per equation (including the intercept) and draws is the number of stored
posterior draws.

See Also
summary.bayesianVARs_draws(), vcov.bayesianVARs_bvar().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Extract posterior draws of VAR coefficients

bvar_coefs <- coef(mod)
fitted.bayesianVARs_bvar 7

fitted.bayesianVARs_bvar
Simulate fitted/predicted historical values for an estimated VAR model

Description

Simulates the fitted/predicted (in-sample) values for an estimated VAR model.

Usage

## S3 method for class 'bayesianVARs_bvar'

fitted(object, error_term = TRUE, ...)

Arguments

object A bayesianVARs_bvar object estimated via bvar().

error_term logical indicating whether to include the error term or not.
... Currently ignored.

Value

An object of class bayesianVARs_fitted.

Examples

# Access a subset of the usmacro_growth dataset

data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Simulate predicted historical values including the error term.

pred <- fitted(mod, error_term = TRUE)

# Simulate fitted historical values not including the error term.

fit <- fitted(mod, error_term = FALSE)

# Visualize
plot(pred)
plot(fit)
8 pairs_predict

my_gig Draw from generalized inverse Gaussian

Description
Vectorized version of rgig

Usage
my_gig(n, lambda, chi, psi)

Arguments
n A single integer indicating the number of draws to generate.
lambda vector of shape parameters.
chi vector of shape/scale parameters. Must be nonnegative for positive lambdas and
positive else.
psi vector of shape/scale parameters. Must be nonnegative for negative lambdas and
positive else.

Value
Matrix of dimension c(n,m), where m is the maximum length of lambda, psi and chi.

Examples
gigsamples <- my_gig(2, c(1,1), c(1,1), c(1,1))

pairs_predict Pairwise visualization of out-of-sample posterior predictive densities.

Description
Pairwise visualization of out-of-sample posterior predictive densities.

Usage
## S3 method for class 'bayesianVARs_predict'
pairs(x, vars, ahead, ...)

Arguments
x An object of class bayesianVARs_predict obtained via predict.bayesianVARs_bvar().
vars Integer vector (or coercible to such) indicating which variables to plot.
ahead Integer vector (or coercible to such) indicating which step ahead to plot. max(ahead)
must be smaller equal to dim(x$predictions)[1].
... Currently ignored!
plot.bayesianVARs_bvar 9

Value
Returns x invisibly.

Note
Note that that bayesianVARs_predict can also be used withing plot.bayesianVARs_bvar().

See Also
Other plotting plot.bayesianVARs_bvar(), plot.bayesianVARs_fitted(), plot.bayesianVARs_predict()
posterior_heatmap().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Simulate from posterior predictive

predictions <- predict(mod, ahead = 1:3)

# Visualize
pairs(predictions, vars = 1:3, ahead = 1:3)

plot.bayesianVARs_bvar
Plot method for bayesianVARs_bvar

Description
Visualization of in-sample fit. Can also be used to display prediction intervals of future values.

Usage
## S3 method for class 'bayesianVARs_bvar'
plot(
x,
predictions = NULL,
quantiles = c(0.05, 0.5, 0.95),
dates = NULL,
n_col = 1,
...
)
10 plot.bayesianVARs_fitted

Arguments

x An object of class bayesianVARs_bvar obtained via bvar().

predictions Optional array of out of sample predictions, e.g. obtained via predict.bayesianVARs_bvar().
quantiles numeric vector indicating which quantiles to plot.
dates optional vector of dates for labelling the x-axis. The default values is NULL; in
this case, the axis will be labeled with numbers.
n_col integer indicating the number of columns to use for plotting.
... Currently ignored!

Value

Returns x invisibly.

See Also

Other plotting plot.bayesianVARs_fitted(), plot.bayesianVARs_predict(), pairs.bayesianVARs_predict(),

posterior_heatmap().

Examples

# Access a subset of the usmacro_growth dataset

data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Simulate from posterior predictive

predictions <- predict(mod, ahead = 1:3)

# Visualize
plot(mod, predictions = predictions)

plot.bayesianVARs_fitted
Visualization of in-sample fit of an estimated VAR.

Description

Visualization of in-sample fit of an estimated VAR.

plot.bayesianVARs_fitted 11

Usage
## S3 method for class 'bayesianVARs_fitted'
plot(
x,
dates = NULL,
vars = "all",
quantiles = c(0.05, 0.5, 0.95),
n_col = 1L,
...
)

Arguments
x A bayesianVARs_fitted object.
dates optional vector of dates for labelling the x-axis. The default values is NULL; in
this case, the axis will be labeled with numbers.
vars character vector containing the names of the variables to be visualized. The
default is "all" indicating that the fit of all variables is visualized.
quantiles numeric vector indicating which quantiles to plot.
n_col integer indicating the number of columns to use for plotting.
... Currently ignored.

Value
returns x invisibly

See Also
• fitted method for class ’bayesianVARs_bvar’: fitted.bayesianVARs_bvar().
• Other plotting plot.bayesianVARs_bvar(), plot.bayesianVARs_fitted(), plot.bayesianVARs_predict(),
pairs.bayesianVARs_predict(), posterior_heatmap().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Simulate predicted historical values including the error term.

pred <- fitted(mod, error_term = TRUE)

# Visualize
plot(pred)
12 plot.bayesianVARs_predict

plot.bayesianVARs_predict
Fan chart

Description
Visualization of (out-of-sample) predictive distribution.

Usage
## S3 method for class 'bayesianVARs_predict'
plot(
x,
dates = NULL,
vars = "all",
ahead = NULL,
quantiles = c(0.05, 0.25, 0.5, 0.75, 0.95),
n_col = 1L,
first_obs = 1L,
...
)

Arguments
x An object of type bayesianVARs_predict obtained via predict.bayesianVARs_bvar().
dates optional vector of dates for labeling the x-axis. The default values is NULL; in
this case, the axis will be labeled with numbers.
vars character vector containing the names of the variables to be visualized. The
default is "all" indicating that all variables are visualized.
ahead Integer vector (or coercible to such) indicating which step ahead to plot. max(ahead)
must be smaller equal to dim(x$predictions)[1].
quantiles numeric vector indicating which quantiles to plot.
n_col integer indicating the number of columns to use for plotting.
first_obs integer indicating the first observation to be used for plotting.
... Currently ignored!

Value
Returns x invisibly!

See Also
Other plotting plot.bayesianVARs_bvar(), plot.bayesianVARs_fitted(), pairs.bayesianVARs_predict()
posterior_heatmap().
posterior_heatmap 13

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Simulate from posterior predictive

predictions <- predict(mod, ahead = 1:3)

# Visualize
plot(predictions, vars = 1:3, ahead = 1:3)

posterior_heatmap Posterior heatmaps for VAR coefficients or variance-covariance ma-

trices

Description
Posterior heatmaps for VAR coefficients or variance-covariance matrices

Usage
posterior_heatmap(
x,
FUN,
...,
colorbar = TRUE,
xlabels = NULL,
ylabels = NULL,
add_numbers = FALSE,
zlim = NULL,
colspace = NULL,
main = "",
[Link] = 0.75,
[Link] = 1,
[Link] = 1,
asp = NULL
)

Arguments
x An array of dimension a × b × draws, where a × b is the dimension of the
parameter to visualize and draws is the number of posterior draws.
FUN The summary function to be applied to margins c(1,2) of x. E.g. "median",
"mean", "IQR", "sd" or "var". apply(x, 1:2, FUN, ...) must return a ma-
trix!
14 posterior_heatmap

... optional arguments to FUN.

colorbar logical indicating whether to display a colorbar or not. Default is TRUE.
xlabels ylabels=NULL, the default, indicates that the names of the dependent variables
will be displayed. ylabels="" indicates that no ylabels will be displayed.
ylabels xlabels=NULL, the default, indicates that the labels of all covariables (the lagged
values of the dependent variables) will be displayed. xlabels="lags" indicates
that only the lags will be marked. xlabels="" indicates that no ylabels are
displayed.
add_numbers logical. add_numbers=TRUE, the default indicates that the actual values of summary
will be displayed.
zlim numeric vector of length two indicating the minimum and maximum values for
which colors should be plotted. By default this range is determined by the max-
imum of the absolute values of the selected summary.
colspace Optional argument.
main main title for the plot.
[Link] The magnification to be used for y-axis annotation relative to the current setting
of cex.
[Link] The magnification to be used for colorbar annotation relative to the current set-
ting of cex.
[Link] The magnification to be used for the actual values (if add_numbers=TRUE) rela-
tive to the current setting of cex.
asp aspect ratio.

Value
Returns x invisibly.

See Also
Other plotting plot.bayesianVARs_bvar(), plot.bayesianVARs_fitted(), plot.bayesianVARs_predict(),
pairs.bayesianVARs_predict().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(100*data, sv_keep = "all", quiet = TRUE)

# Extract posterior draws of VAR coefficients

phi_post <- coef(mod)

# Visualize posterior median of VAR coefficients

posterior_heatmap(phi_post, median)

# Extract posterior draws of variance-covariance matrices (for each point in time)

predict.bayesianVARs_bvar 15

sigma_post <- vcov(mod)

# Visualize posterior interquartile-range of variance-covariance matrix of the first observation
posterior_heatmap(sigma_post[1,,,], IQR)

predict.bayesianVARs_bvar
Predict method for Bayesian VARs

Description
Simulates from (out-of-sample) predictive density for Bayesian VARs estimated via bvar() and
computes log predictive likelhoods if ex-post observed data is supplied.

Usage
## S3 method for class 'bayesianVARs_bvar'
predict(
object,
ahead = 1L,
each = 1L,
stable = TRUE,
simulate_predictive = TRUE,
LPL = FALSE,
Y_obs = NA,
LPL_VoI = NA,
...
)

Arguments
object A bayesianVARs_bvar object, obtained from bvar().
ahead Integer vector (or coercible to such), indicating the number of steps ahead at
which to predict.
each Single integer (or coercible to such) indicating how often should be drawn from
the posterior predictive distribution for each draw that has been stored during
MCMC sampling.
stable logical indicating whether to consider only those draws from the posterior that
fulfill the ’stable’ criterion. Default is TRUE.
simulate_predictive
logical, indicating whether the posterior predictive distribution should be simu-
lated.
LPL logical indicating whether ahead-step-ahead log predictive likelihoods should
be computed. If LPL=TRUE, Y_obs has to be specified.
Y_obs Data matrix of observed values for computation of log predictive likelihood.
Each of ncol(object$Yraw) columns is assumed to contain a single time-series
of length length(ahead).
16 predict.bayesianVARs_bvar

LPL_VoI either integer vector or character vector of column-names indicating for which
subgroup of time-series in object$Yraw a joint log predictive likelihood shall
be computed.
... Currently ignored!

Value
Object of class bayesianVARs_predict, a list that may contain the following elements:
• predictions array of dimensions c(length(ahead), ncol(object$Yraw), each * dim(object$PHI)[3])
containing the simulations from the predictive density (if simulate_predictive=TRUE).
• LPL vector of length length(ahead) containing the log-predictive-likelihoods (taking into
account the joint distribution of all variables) (if LPL=TRUE).
• LPL_univariate matrix of dimension c(length(ahead), ncol(object$Yraw) containing
the marginalized univariate log-predictive-likelihoods of each series (if LPL=TRUE).
• LPL_VoI vector of length length(ahead) containing the log-predictive-likelihoods for a sub-
set of variables (if LPL=TRUE and LPL_VoI != NA).
• Yraw matrix containing the data used for the estimation of the VAR.
• LPL_draws matrix containing the simulations of the log-predictive-likelihood (if LPL=TRUE).
• PL_univariate_draws array containing the simulations of the univariate predictive-likelihoods
(if LPL=TRUE).
• LPL_sub_draws matrix containing the simulations of the log-predictive-likelihood for a subset
of variables (if LPL=TRUE and LPL_VoI != NA).

See Also
stable_bvar(), plot.bayesianVARs_predict(), pairs.bayesianVARs_predict().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Split data in train and test

train <- data[1:(nrow(data)-4),]
test <- data[-c(1:(nrow(data)-4)),]

# Estimate model using train data only

mod <- bvar(train, quiet = TRUE)

# Simulate from 1-step to 4-steps ahead posterior predictive and compute

# log-predictive-likelihoods
predictions <- predict(mod, ahead = 1:4, LPL = TRUE, Y_obs = test)

# Summary
summary(predictions)

# Visualize via fan-charts

plot(predictions)
print.bayesianVARs_bvar 17

# In order to evaluate the joint predictive density of a subset of the

# variables (variables of interest), consider specifying 'LPL_VoI':
predictions <- predict(mod, ahead = 1:4, LPL = TRUE, Y_obs = test, LPL_VoI = c("GDPC1","FEDFUNDS"))
predictions$LPL_VoI

print.bayesianVARs_bvar
Pretty printing of a bvar object

Description

Pretty printing of a bvar object

Usage

## S3 method for class 'bayesianVARs_bvar'

print(x, ...)

Arguments

x Object of class bayesianVARs_bvar, usually resulting from a call of bvar().

... Ignored.

Value

Returns x invisibly.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Print model
mod
18 [Link].bayesianVARs_bvar

print.bayesianVARs_predict
Print method for bayesianVARs_predict objects

Description
Print method for bayesianVARs_predict objects.

Usage
## S3 method for class 'bayesianVARs_predict'
print(x, ...)

Arguments
x A bayesianVARs_predict object obtained via predict.bayesianVARs_bvar().
... Currently ignored!

Value
Returns x invisibly.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Split data in train and test

train <- data[1:(nrow(data)-4),]
test <- data[-c(1:(nrow(data)-4)),]

# Estimate model using train data only

mod <- bvar(train, quiet = TRUE)

# Simulate from 1-step ahead posterior predictive

predictions <- predict(mod, ahead = 1L)
print(predictions)

[Link].bayesianVARs_bvar
Print method for summary.bayesianVARs_bvar objects

Description
Print method for summary.bayesianVARs_bvar objects.
[Link].bayesianVARs_predict 19

Usage
## S3 method for class 'summary.bayesianVARs_bvar'
print(x, ...)

Arguments
x A summary.bayesianVARs_bvar object obtained via summary.bayesianVARs_bvar().
... Currently ignored!

Value
Returns x invisibly!

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate model
mod <- bvar(data, quiet = TRUE)

# Print summary
summary(mod)

[Link].bayesianVARs_predict
Print method for summary.bayesianVARs_predict objects

Description
Print method for summary.bayesianVARs_predict objects.

Usage
## S3 method for class 'summary.bayesianVARs_predict'
print(x, ...)

Arguments
x A summary.bayesianVARs_predict object obtained via summary.bayesianVARs_predict().
... Currently ignored.

Value
Returns x invisibly.
20 specify_prior_phi

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Split data in train and test

train <- data[1:(nrow(data)-4),]
test <- data[-c(1:(nrow(data)-4)),]

# Estimate model using train data only

mod <- bvar(train, quiet = TRUE)

# Simulate from 1-step ahead posterior predictive

predictions <- predict(mod, ahead = 1L)
sum <- summary(predictions)
print(sum)

specify_prior_phi Specify prior on PHI

Description
Configures prior on PHI, the matrix of reduced-form VAR coefficients.

Usage
specify_prior_phi(
data = NULL,
M = ncol(data),
lags = 1L,
prior = "HS",
priormean = 0,
PHI_tol = 1e-18,
DL_a = "1/K",
DL_tol = 0,
R2D2_a = 0.1,
R2D2_b = 0.5,
R2D2_tol = 0,
NG_a = 0.1,
NG_b = 1,
NG_c = 1,
NG_tol = 0,
SSVS_c0 = 0.01,
SSVS_c1 = 100,
SSVS_semiautomatic = TRUE,
SSVS_p = 0.5,
HMP_lambda1 = c(0.01, 0.01),
HMP_lambda2 = c(0.01, 0.01),
specify_prior_phi 21

normal_sds = 10,
global_grouping = "global",
...
)

Arguments
data Optional. Data matrix (can be a time series object). Each of M columns is
assumed to contain a single time-series of length T .
M positive integer indicating the number of time-series of the VAR.
lags positive integer indicating the order of the VAR, i.e. the number of lags of the
dependent variables included as predictors.
prior character, one of "HS", "R2D2", "NG", "DL", "SSVS", "HMP" or "normal".
priormean real numbers indicating the prior means of the VAR coefficients. One single
number means that the prior mean of all own-lag coefficients w.r.t. the first
lag equals priormean and 0 else. A vector of length M means that the prior
mean of the own-lag coefficients w.r.t. the first lag equals priormean and 0
else. If priormean is a matrix of dimension c(lags*M,M), then each of the M
columns is assumed to contain lags*M prior means for the VAR coefficients of
the respective VAR equations.
PHI_tol Minimum number that the absolute value of a VAR coefficient draw can take.
Prevents numerical issues that can appear when strong shrinkage is enforced if
chosen to be greater than zero.
DL_a (Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. If the
argument global_grouping specifies e.g. k groups, then DL_a can be a numeric
vector of length k and the elements indicate the shrinkage in each group. A ma-
trix of dimension c(s,2) specifies a discrete hyperprior, where the first column
contains s support points and the second column contains the associated prior
probabilities. DL_a has only to be specified if prior="DL".
DL_tol Minimum number that a parameter draw of one of the shrinking parameters of
the Dirichlet Laplace prior can take. Prevents numerical issues that can appear
when strong shrinkage is enforced if chosen to be greater than zero. DL_tol has
only to be specified if prior="DL".
R2D2_a (Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. If the
argument global_grouping specifies e.g. k groups, then R2D2_a can be a nu-
meric vector of length k and the elements indicate the shrinkage in each group.
A matrix of dimension c(s,2) specifies a discrete hyperprior, where the first
column contains s support points and the second column contains the associ-
ated prior probabilities. R2D2_a has only to be specified if prior="R2D2".
R2D2_b (Single) positive real number. The value indicates the shape parameter of the in-
verse gamma prior on the (semi-)global scales. If the argument global_grouping
specifies e.g. k groups, then NG_b can be a numeric vector of length k and the
elements determine the shape parameter in each group. R2D2_b has only to be
specified if prior="R2D2".
22 specify_prior_phi

R2D2_tol Minimum number that a parameter draw of one of the shrinking parameters of
the R2D2 prior can take. Prevents numerical issues that can appear when strong
shrinkage is enforced if chosen to be greater than zero. R2D2_tol has only to be
specified if prior="R2D2".
NG_a (Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. If the
argument global_grouping specifies e.g. k groups, then NG_a can be a numeric
vector of length k and the elements indicate the shrinkage in each group. A ma-
trix of dimension c(s,2) specifies a discrete hyperprior, where the first column
contains s support points and the second column contains the associated prior
probabilities. NG_a has only to be specified if prior="NG".
NG_b (Single) positive real number. The value indicates the shape parameter of the in-
verse gamma prior on the (semi-)global scales. If the argument global_grouping
specifies e.g. k groups, then NG_b can be a numeric vector of length k and the
elements determine the shape parameter in each group. NG_b has only to be
specified if prior="NG".
NG_c (Single) positive real number. The value indicates the scale parameter of the in-
verse gamma prior on the (semi-)global scales. If the argument global_grouping
specifies e.g. k groups, then NG_c can be a numeric vector of length k and the
elements determine the scale parameter in each group. Expert option would be
to set the scale parameter proportional to NG_a. E.g. in the case where a discrete
hyperprior for NG_a is chosen, a desired proportion of let’s say 0.2 is achieved
by setting NG_c="0.2*a" (character input!). NG_c has only to be specified if
prior="NG".
NG_tol Minimum number that a parameter draw of one of the shrinking parameters of
the normal-gamma prior can take. Prevents numerical issues that can appear
when strong shrinkage is enforced if chosen to be greater than zero. NG_tol has
only to be specified if prior="NG".
SSVS_c0 single positive number indicating the (unscaled) standard deviation of the spike
component. SSVS_c0 has only to be specified if prior="SSVS". It should be
that SSV Sc0 ≪ SSV Sc1 ! SSVS_c0 has only to be specified if prior="SSVS".
SSVS_c1 single positive number indicating the (unscaled) standard deviation of the slab
component. SSVS_c0 has only to be specified if prior="SSVS". It should be
that SSV Sc0 ≪ SSV Sc1 !
SSVS_semiautomatic
logical. If SSVS_semiautomatic=TRUE both SSVS_c0 and SSVS_c1 will be
scaled by the variances of the posterior of PHI under a FLAT conjugate (de-
pendent Normal-Wishart prior). SSVS_semiautomatic has only to be specified
if prior="SSVS".
SSVS_p Either a single positive number in the range (0,1) indicating the (fixed) prior
inclusion probability of each coefficient. Or numeric vector of length 2 with
positive entries indicating the shape parameters of the Beta distribution. In that
case a Beta hyperprior is placed on the prior inclusion probability. SSVS_p has
only to be specified if prior="SSVS".
HMP_lambda1 numeric vector of length 2. Both entries must be positive. The first indicates the
shape and the second the rate of the Gamma hyperprior on own-lag coefficients.
HMP_lambda1 has only to be specified if prior="HMP".
specify_prior_phi 23

HMP_lambda2 numeric vector of length 2. Both entries must be positive. The first indicates the
shape and the second the rate of the Gamma hyperprior on cross-lag coefficients.
HMP_lambda2 has only to be specified if prior="HMP".
normal_sds numeric vector of length n, where n = lagsM 2 is the number of all VAR
coefficients (excluding the intercept), indicating the prior variances. A single
number will be recycled accordingly! Must be positive. normal_sds has only
to be specified if prior="normal".
global_grouping
One of "global", "equation-wise", "covariate-wise", "olcl-lagwise"
"fol" indicating the sub-groups of the semi-global(-local) modifications to HS,
R2D2, NG, DL and SSVS prior. Works also with user-specified indicator ma-
trix of dimension c(lags*M,M). Only relevant if prior="HS", prior="DL",
prior="R2D2", prior="NG" or prior="SSVS".
... Do not use!

Details
For details concerning prior-elicitation for VARs please see Gruber & Kastner (2023).
Currently one can choose between six hierarchical shrinkage priors and a normal prior: prior="HS"
stands for the Horseshoe-prior, prior="R2D2 for the R2 -induced-Dirichlet-decompostion-prior,
prior="NG" for the normal-gamma-prior, prior="DL" for the Dirichlet-Laplace-prior, prior="SSVS"
for the stochastic-search-variable-selection-prior, prior="HMP" for the semi-hierarchical Minnesota
prior and prior=normal for the normal-prior.
Semi-global shrinkage, i.e. group-specific shrinkage for pre-specified subgroups of the coefficients,
can be achieved through the argument global_grouping.

Value
A baysianVARs_prior_phi-object.

References
Gruber, L. and Kastner, G. (2023). Forecasting macroeconomic data with Bayesian VARs: Sparse
or dense? It depends! arXiv:2206.04902.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Horseshoe prior for a VAR(2)

phi_hs <- specify_prior_phi(data = data, lags = 2L ,prior = "HS")

# Semi-global-local Horseshoe prior for a VAR(2) with semi-global shrinkage parameters for
# cross-lag and own-lag coefficients in each lag
phi_hs_sg <- specify_prior_phi(data = data, lags = 2L, prior = "HS",
global_grouping = "olcl-lagwise")

# Semi-global-local Horseshoe prior for a VAR(2) with equation-wise shrinkage

24 specify_prior_sigma

# construct indicator matrix for equation-wise shrinkage

semi_global_mat <- matrix(1:ncol(data), 2*ncol(data), ncol(data),
byrow = TRUE)
phi_hs_ew <- specify_prior_phi(data = data, lags = 2L, prior = "HS",
global_grouping = semi_global_mat)
# (for equation-wise shrinkage one can also use 'global_grouping = "equation-wise"')

# Estimate model with your prior configuration of choice

mod <- bvar(data, lags = 2L, prior_phi = phi_hs_sg, quiet = TRUE)

specify_prior_sigma Specify prior on Sigma

Description
Configures prior on the variance-covariance of the VAR.

Usage
specify_prior_sigma(
data = NULL,
M = ncol(data),
type = c("factor", "cholesky"),
factor_factors = 1L,
factor_restrict = c("none", "upper"),
factor_priorfacloadtype = c("rowwiseng", "colwiseng", "normal"),
factor_priorfacload = 0.1,
factor_facloadtol = 1e-18,
factor_priorng = c(1, 1),
factor_priormu = c(0, 10),
factor_priorphiidi = c(10, 3),
factor_priorphifac = c(10, 3),
factor_priorsigmaidi = 1,
factor_priorsigmafac = 1,
factor_priorh0idi = "stationary",
factor_priorh0fac = "stationary",
factor_heteroskedastic = TRUE,
factor_priorhomoskedastic = NA,
factor_interweaving = 4,
cholesky_U_prior = c("HS", "DL", "R2D2", "NG", "SSVS", "normal", "HMP"),
cholesky_U_tol = 1e-18,
cholesky_heteroscedastic = TRUE,
cholesky_priormu = c(0, 100),
cholesky_priorphi = c(20, 1.5),
cholesky_priorsigma2 = c(0.5, 0.5),
cholesky_priorh0 = "stationary",
specify_prior_sigma 25

cholesky_priorhomoscedastic = [Link](NA),
cholesky_DL_a = "1/n",
cholesky_DL_tol = 0,
cholesky_R2D2_a = 0.4,
cholesky_R2D2_b = 0.5,
cholesky_R2D2_tol = 0,
cholesky_NG_a = 0.5,
cholesky_NG_b = 0.5,
cholesky_NG_c = 0.5,
cholesky_NG_tol = 0,
cholesky_SSVS_c0 = 0.001,
cholesky_SSVS_c1 = 1,
cholesky_SSVS_p = 0.5,
cholesky_HMP_lambda3 = c(0.01, 0.01),
cholesky_normal_sds = 10,
expert_sv_offset = 0,
quiet = FALSE,
...
)

Arguments
data Optional. Data matrix (can be a time series object). Each of M columns is
assumed to contain a single time-series of length T .
M positive integer indicating the number of time-series of the VAR.
type character, one of "factor" (the default) or "cholesky", indicating which de-
composition to be applied to the covariance-matrix.
factor_factors Number of latent factors to be estimated. Only required if type="factor".
factor_restrict
Either "upper" or "none", indicating whether the factor loadings matrix should
be restricted to have zeros above the diagonal ("upper") or whether all elements
should be estimated from the data ("none"). Setting restrict to "upper" of-
ten stabilizes MCMC estimation and can be important for identifying the fac-
tor loadings matrix, however, it generally is a strong prior assumption. Setting
restrict to "none" is usually the preferred option if identification of the factor
loadings matrix is of less concern but covariance estimation or prediction is the
goal. Only required if type="factor".
factor_priorfacloadtype
Can be "normal", "rowwiseng", "colwiseng". Only required if type="factor".
"normal": Normal prior. The value of priorfacload is interpreted as the
standard deviations of the Gaussian prior distributions for the factor load-
ings.
"rowwiseng": Row-wise Normal-Gamma prior. The value of priorfacload
is interpreted as the shrinkage parameter a.
"colwiseng": Column-wise Normal-Gamma prior. The value of priorfacload
is interpreted as the shrinkage parameter a.
For details please see Kastner (2019).
26 specify_prior_sigma

factor_priorfacload
Either a matrix of dimensions M times factor_factors with positive elements
or a single number (which will be recycled accordingly). Only required if
type="factor". The meaning of factor_priorfacload depends on the set-
ting of factor_priorfacloadtype and is explained there.
factor_facloadtol
Minimum number that the absolute value of a factor loadings draw can take.
Prevents numerical issues that can appear when strong shrinkage is enforced if
chosen to be greater than zero. Only required if type="factor".
factor_priorng Two-element vector with positive entries indicating the Normal-Gamma prior’s
hyperhyperparameters c and d (cf. Kastner (2019)). Only required if type="factor".
factor_priormu Vector of length 2 denoting prior mean and standard deviation for unconditional
levels of the idiosyncratic log variance processes. Only required if type="factor".
factor_priorphiidi
Vector of length 2, indicating the shape parameters for the Beta prior distribu-
tions of the transformed parameters (phi+1)/2, where phi denotes the persis-
tence of the idiosyncratic log variances. Only required if type="factor".
factor_priorphifac
Vector of length 2, indicating the shape parameters for the Beta prior distribu-
tions of the transformed parameters (phi+1)/2, where phi denotes the persis-
tence of the factor log variances. Only required if type="factor".
factor_priorsigmaidi
Vector of length M containing the prior volatilities of log variances. If factor_priorsigmaidi
has exactly one element, it will be recycled for all idiosyncratic log variances.
Only required if type="factor".
factor_priorsigmafac
Vector of length factor_factors containing the prior volatilities of log vari-
ances. If factor_priorsigmafac has exactly one element, it will be recycled
for all factor log variances. Only required if type="factor".
factor_priorh0idi
Vector of length 1 or M, containing information about the Gaussian prior for
the initial idiosyncratic log variances. Only required if type="factor". If an
element of factor_priorh0idi is a nonnegative number, the conditional prior
of the corresponding initial log variance h0 is assumed to be Gaussian with mean
0 and standard deviation factor_priorh0idi times sigma. If an element of
factor_priorh0idi is the string ’stationary’, the prior of the corresponding
initial log volatility is taken to be from the stationary distribution, i.e. h0 is
assumed to be Gaussian with mean 0 and variance sigma2 /(1 − phi2 ).
factor_priorh0fac
Vector of length 1 or factor_factors, containing information about the Gaus-
sian prior for the initial factor log variances. Only required if type="factor".
If an element of factor_priorh0fac is a nonnegative number, the conditional
prior of the corresponding initial log variance h0 is assumed to be Gaussian with
mean 0 and standard deviation factor_priorh0fac times sigma. If an element
of factor_priorh0fac is the string ’stationary’, the prior of the corresponding
initial log volatility is taken to be from the stationary distribution, i.e. h0 is
assumed to be Gaussian with mean 0 and variance sigma2 /(1 − phi2 ).
specify_prior_sigma 27

factor_heteroskedastic
Vector of length 1, 2, or M + factor_factors, containing logical values in-
dicating whether time-varying (factor_heteroskedastic = TRUE) or constant
(factor_heteroskedastic = FALSE) variance should be estimated. If factor_heteroskedastic
is of length 2 it will be recycled accordingly, whereby the first element is used
for all idiosyncratic variances and the second element is used for all factor vari-
ances. Only required if type="factor".
factor_priorhomoskedastic
Only used if at least one element of factor_heteroskedastic is set to FALSE.
In that case, factor_priorhomoskedastic must be a matrix with positive en-
tries and dimension c(M, 2). Values in column 1 will be interpreted as the shape
and values in column 2 will be interpreted as the rate parameter of the corre-
sponding inverse gamma prior distribution of the idiosyncratic variances. Only
required if type="factor".
factor_interweaving
The following values for interweaving the factor loadings are accepted (Only
required if type="factor"):
0: No interweaving.
1: Shallow interweaving through the diagonal entries.
2: Deep interweaving through the diagonal entries.
3: Shallow interweaving through the largest absolute entries in each column.
4: Deep interweaving through the largest absolute entries in each column.
For details please see Kastner et al. (2017). A value of 4 is the highly recom-
mended default.
cholesky_U_prior
character, one of "HS", "R2D2", "NG", "DL", "SSVS", "HMP" or "normal". Only
required if type="cholesky".
cholesky_U_tol Minimum number that the absolute value of an free off-diagonal element of an
U -draw can take. Prevents numerical issues that can appear when strong shrink-
age is enforced if chosen to be greater than zero. Only required if type="cholesky".
cholesky_heteroscedastic
single logical indicating whether time-varying (cholesky_heteroscedastic =
TRUE) or constant (cholesky_heteroscedastic = FALSE) variance should be
estimated. Only required if type="cholesky".
cholesky_priormu
Vector of length 2 denoting prior mean and standard deviation for unconditional
levels of the log variance processes. Only required if type="cholesky".
cholesky_priorphi
Vector of length 2, indicating the shape parameters for the Beta prior distribu-
tions of the transformed parameters (phi+1)/2, where phi denotes the persis-
tence of the log variances. Only required if type="cholesky".
cholesky_priorsigma2
Vector of length 2, indicating the shape and the rate for the Gamma prior distri-
butions on the variance of the log variance processes. (Currently only one global
setting for all M processes is supported). Only required if type="cholesky".
28 specify_prior_sigma

cholesky_priorh0
Vector of length 1 or M, containing information about the Gaussian prior for the
initial idiosyncratic log variances. Only required if type="cholesky". If an
element of cholesky_priorh0 is a nonnegative number, the conditional prior
of the corresponding initial log variance h0 is assumed to be Gaussian with
mean 0 and standard deviation cholesky_priorh0 times sigma. If an element
of cholesky_priorh0 is the string ’stationary’, the prior of the corresponding
initial log volatility is taken to be from the stationary distribution, i.e. h0 is
assumed to be Gaussian with mean 0 and variance sigma2 /(1 − phi2 ).
cholesky_priorhomoscedastic
Only used if cholesky_heteroscedastic=FALSE. In that case, cholesky_priorhomoscedastic
must be a matrix with positive entries and dimension c(M, 2). Values in column
1 will be interpreted as the shape and values in column 2 will be interpreted as
the scale parameter of the corresponding inverse gamma prior distribution of the
variances. Only required if type="cholesky".
cholesky_DL_a (Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. A matrix
of dimension c(s,2) specifies a discrete hyperprior, where the first column con-
tains s support points and the second column contains the associated prior prob-
abilities. cholesky_DL_a has only to be specified if cholesky_U_prior="DL".
cholesky_DL_tol
Minimum number that a parameter draw of one of the shrinking parameters of
the Dirichlet Laplace prior can take. Prevents numerical issues that can appear
when strong shrinkage is enforced if chosen to be greater than zero. DL_tol has
only to be specified if cholesky_U_prior="DL".
cholesky_R2D2_a
(Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. A matrix
of dimension c(s,2) specifies a discrete hyperprior, where the first column con-
tains s support points and the second column contains the associated prior proba-
bilities. cholesky_R2D2_a has only to be specified if cholesky_U_prior="R2D2".
cholesky_R2D2_b
single positive number, where greater values indicate heavier regularization.
cholesky_R2D2_b has only to be specified if cholesky_U_prior="R2D2".
cholesky_R2D2_tol
Minimum number that a parameter draw of one of the shrinking parameters of
the R2D2 prior can take. Prevents numerical issues that can appear when strong
shrinkage is enforced if chosen to be greater than zero. cholesky_R2D2_tol
has only to be specified if cholesky_U_prior="R2D2".
cholesky_NG_a (Single) positive real number. The value is interpreted as the concentration pa-
rameter for the local scales. Smaller values enforce heavier shrinkage. A matrix
of dimension c(s,2) specifies a discrete hyperprior, where the first column con-
tains s support points and the second column contains the associated prior prob-
abilities. cholesky_NG_a has only to be specified if cholesky_U_prior="NG".
cholesky_NG_b (Single) positive real number. The value indicates the shape parameter of the in-
verse gamma prior on the global scales. cholesky_NG_b has only to be specified
if cholesky_U_prior="NG".
specify_prior_sigma 29

cholesky_NG_c (Single) positive real number. The value indicates the scale parameter of the
inverse gamma prior on the global scales. Expert option would be to set the scale
parameter proportional to NG_a. E.g. in the case where a discrete hyperprior
for NG_a is chosen, a desired proportion of let’s say 0.2 is achieved by setting
NG_c="0.2a" (character input!). cholesky_NG_c has only to be specified if
cholesky_U_prior="NG".
cholesky_NG_tol
Minimum number that a parameter draw of one of the shrinking parameters
of the normal-gamma prior can take. Prevents numerical issues that can appear
when strong shrinkage is enforced if chosen to be greater than zero. cholesky_NG_tol
has only to be specified if cholesky_U_prior="NG".
cholesky_SSVS_c0
single positive number indicating the (unscaled) standard deviation of the spike
component. cholesky_SSVS_c0 has only to be specified if choleksy_U_prior="SSVS".
It should be that SSV Sc0 ≪ SSV Sc1 !
cholesky_SSVS_c1
single positive number indicating the (unscaled) standard deviation of the slab
component. cholesky_SSVS_c1 has only to be specified if choleksy_U_prior="SSVS".
It should be that SSV Sc0 ≪ SSV Sc1 !
cholesky_SSVS_p
Either a single positive number in the range (0,1) indicating the (fixed) prior
inclusion probability of each coefficient. Or numeric vector of length 2 with pos-
itive entries indicating the shape parameters of the Beta distribution. In that case
a Beta hyperprior is placed on the prior inclusion probability. cholesky_SSVS_p
has only to be specified if choleksy_U_prior="SSVS".
cholesky_HMP_lambda3
numeric vector of length 2. Both entries must be positive. The first indicates
the shape and the second the rate of the Gamma hyperprior on the contem-
poraneous coefficients. cholesky_HMP_lambda3 has only to be specified if
choleksy_U_prior="HMP".
cholesky_normal_sds
2
numeric vector of length M 2−M , indicating the prior variances for the free off-
diagonal elements in U . A single number will be recycled accordingly! Must be
positive. cholesky_normal_sds has only to be specified if choleksy_U_prior="normal".
expert_sv_offset
... Do not use!
quiet logical indicating whether informative output should be omitted.
... Do not use!

Details
bvar offers two different specifications for the errors: The user can choose between a factor
stochastic volatility structure or a cholesky stochastic volatility structure. In both cases the distur-
bances ϵt are assumed to follow a M -dimensional multivariate normal distribution with zero mean
and variance-covariance matrix Σt . In case of the cholesky specification Σt = U ′−1 D t U −1 ,
where U −1 is upper unitriangular (with ones on the diagonal). The diagonal matrix D t de-
pends upon latent log-variances, i.e. D t = diag(exp(h1t ), . . . , exp(hM t ). The log-variances
30 specify_prior_sigma

follow a priori independent autoregressive processes hit ∼ N (µi + ϕi (hi,t−1 − µi ), σi2 ) for i =
1, . . . , M . In case of the factor structure, Σt = ΛV t Λ′ + Gt . The diagonal matrices V t
and Gt depend upon latent log-variances, i.e. Gt = diag(exp(h1t ), . . . , exp(hM t ) and V t =
diag(exp(hM +1,t ), . . . , exp(hM +r,t ). The log-variances follow a priori independent autoregres-
sive processes hit ∼ N (µi +ϕi (hi,t−1 −µi ), σi2 ) for i = 1, . . . , M and hM +j,t ∼ N (ϕi hM +j,t−1 , σM
2
+j )
for j = 1, . . . , r.

Value
Object of class bayesianVARs_prior_sigma.

References
Kastner, G. (2019). Sparse Bayesian Time-Varying Covariance Estimation in Many Dimensions
Journal of Econometrics, 210(1), 98–115, doi:10.1016/[Link].2018.11.007
Kastner, G., Frühwirth-Schnatter, S., and Lopes, H.F. (2017). Efficient Bayesian Inference for Mul-
tivariate Factor Stochastic Volatility Models. Journal of Computational and Graphical Statistics,
26(4), 905–917, doi:10.1080/10618600.2017.1322091.

See Also
specify_prior_phi().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# examples with stochastic volatility (heteroscedasticity) -----------------

# factor-decomposition with 2 factors and colwise normal-gamma prior on the loadings
sigma_factor_cng_sv <- specify_prior_sigma(data = data, type = "factor",
factor_factors = 2L, factor_priorfacloadtype = "colwiseng", factor_heteroskedastic = TRUE)

# cholesky-decomposition with Dirichlet-Laplace prior on U

sigma_cholesky_dl_sv <- specify_prior_sigma(data = data, type = "cholesky",
cholesky_U_prior = "DL", cholesky_DL_a = 0.5, cholesky_heteroscedastic = TRUE)

# examples without stochastic volatility (homoscedasticity) ----------------

# factor-decomposition with 2 factors and colwise normal-gamma prior on the loadings
sigma_factor_cng <- specify_prior_sigma(data = data, type = "factor",
factor_factors = 2L, factor_priorfacloadtype = "colwiseng",
factor_heteroskedastic = FALSE, factor_priorhomoskedastic = matrix(c(0.5,0.5),
ncol(data), 2))

# cholesky-decomposition with Horseshoe prior on U

sigma_cholesky_dl <- specify_prior_sigma(data = data, type = "cholesky",
cholesky_U_prior = "HS", cholesky_heteroscedastic = FALSE)

# Estimate model with your prior configuration of choice

mod <- bvar(data, prior_sigma = sigma_factor_cng_sv, quiet = TRUE)
stable_bvar 31

stable_bvar Stable posterior draws

Description
stable_bvar() detects and discards all posterior draws of an bayesianVARs_bvar object that do
not fulfill the stability condition: A VAR(p) model is considered as stable only if the eigenvalues of
the companion form matrix lie inside the unit circle.

Usage
stable_bvar(object, quiet = FALSE)

Arguments
object A bayesianVARs_bvar object obtained via bvar().
quiet logical indicating whether informative output should be omitted.

Value
An object of type bayesianVARs_bvar.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Discard "unstable" draws

stable_mod <- stable_bvar(mod)

summary.bayesianVARs_bvar
Summary method for bayesianVARs_bvar objects

Description
Summary method for bayesianVARs_bvar objects.
32 summary.bayesianVARs_draws

Usage
## S3 method for class 'bayesianVARs_bvar'
summary(object, quantiles = c(0.025, 0.25, 0.5, 0.75, 0.975), ...)

Arguments
object A bayesianVARs_bvar object obtained via bvar().
quantiles numeric vector which quantiles to compute.
... Currently ignored!

Value
An object of type summary.bayesianVARs_bvar.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate model
mod <- bvar(data, quiet = TRUE)

# Summary
sum <- summary(mod)

summary.bayesianVARs_draws
Summary statistics for bayesianVARs posterior draws.

Description
Summary statistics for bayesianVARs posterior draws.

Usage
## S3 method for class 'bayesianVARs_draws'
summary(object, quantiles = c(0.25, 0.5, 0.75), ...)

Arguments
object An object of class bayesianVARs_draws usually obtained through extractors
like coef.bayesianVARs_bvar() and vcov.bayesianVARs_bvar().
quantiles A vector of quantiles to evaluate.
... Currently ignored.
summary.bayesianVARs_predict 33

Value
A list object of class bayesianVARs_draws_summary holding
• mean: Vector or matrix containing the posterior mean.
• sd: Vector or matrix containing the posterior standard deviation .
• quantiles: Array containing the posterior quantiles.

See Also
Available extractors: coef.bayesianVARs_bvar(), vcov.bayesianVARs_bvar().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]
# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Extract posterior draws of VAR coefficients

bvar_coefs <- coef(mod)

# Compute summary statistics

summary_stats <- summary(bvar_coefs)

# Compute summary statistics of VAR coefficients without using coef()

summary_stats <- summary(mod$PHI)

# Test which list elements of 'mod' are of class 'bayesianVARs_draws'.

names(mod)[sapply(names(mod), function(x) inherits(mod[[x]], "bayesianVARs_draws"))]

summary.bayesianVARs_predict
Summary method for bayesianVARs_predict objects

Description
Summary method for bayesianVARs_predict objects.

Usage
## S3 method for class 'bayesianVARs_predict'
summary(object, ...)

Arguments
object A bayesianVARs_predict object obtained via predict.bayesianVARs_bvar().
... Currently ignored!
34 usmacro_growth

Value
A summary.bayesianVARs_predict object.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Split data in train and test

train <- data[1:(nrow(data)-4),]
test <- data[-c(1:(nrow(data)-4)),]

# Estimate model using train data only

mod <- bvar(train, quiet = TRUE)

# Simulate from 1-step ahead posterior predictive

predictions <- predict(mod, ahead = 1L)
summary(predictions)

usmacro_growth Data from the US-economy

Description
21 selected quarterly time-series from 1953:Q1 to 2021:Q2. From FRED-QD data base (McCracken
and Ng, 2021). Release date 2021-07. Data is transformed to be interpreted as growth-rates (first
log-differences with the exception of interest rates, which are already growth rates).

Usage
usmacro_growth

Format
A matrix with 247 rows and 21 columns.

Source
Raw (untransformed) data available at [Link]
mccracken/fred-databases, [Link]
[Link].

References
McCracken, M. W. and Ng, S. (2021). FRED-QD: A Quarterly Database for Macroeconomic Re-
search, Review, Federal Reserve Bank of St. Louis, 103(1), 1–44, doi:10.20955/r.103.144.
vcov.bayesianVARs_bvar 35

vcov.bayesianVARs_bvar
Extract posterior draws of the (time-varying) variance-covariance
matrix for a VAR model

Description

Returns the posterior draws of the possibly time-varying variance-covariance matrix of a VAR esti-
mated via bvar(). Returns the full paths if sv_keep="all" when calling bvar(). Otherwise, the
draws of the variance-covariance matrix for the last observation are returned, only.

Usage

## S3 method for class 'bayesianVARs_bvar'

vcov(object, t = seq_len(nrow(object$logvar)), ...)

Arguments

object An object of class bayesianVARs_bvar obtained via bvar().

t Vector indicating which points in time should be extracted, defaults to all.
... Currently ignored.

Value

An array of class bayesianVARs_draws of dimension T ×M ×M ×draws, where T is the number

of observations, M the number of time-series and draws the number of stored posterior draws.

See Also

summary.bayesianVARs_draws, coef.bayesianVARs_bvar().

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]
# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Extract posterior draws of the variance-covariance matrix

bvar_vcov <- vcov(mod)
36 [.bayesianVARs_draws

[.bayesianVARs_coef Extract or Replace Parts of a bayesianVARs_coef object

Description
Extract or replace parts of a bayesianVARs_coef object.

Usage
## S3 method for class 'bayesianVARs_coef'
x[i, j, ...]

Arguments
x An object of type bayesianVARs_coef.
i indices
j indices
... further indices

Value
An object of type bayesianVARs_coef.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Extract coefficients, which are of class bayesianVARs_coef

phi <- coef(mod)
phi[1,1,1]

[.bayesianVARs_draws Extract or Replace Parts of a bayesianVARs_draws object

Description
Extract or replace parts of a bayesianVARs_draws object.

Usage
## S3 method for class 'bayesianVARs_draws'
x[i, j, ...]
[.bayesianVARs_draws 37

Arguments
x An object of type bayesianVARs_draws.
i indices
j indices
... further indices

Value
An object of type bayesianVARs_draws.

Examples
# Access a subset of the usmacro_growth dataset
data <- usmacro_growth[,c("GDPC1", "CPIAUCSL", "FEDFUNDS")]

# Estimate a model
mod <- bvar(data, sv_keep = "all", quiet = TRUE)

# Extract coefficients, which are of class bayesianVARs_draws

phi <- coef(mod)
phi[1,1,1]
Index

∗ datasets specify_prior_phi(), 5, 30
usmacro_growth, 34 specify_prior_sigma, 3, 24
[.bayesianVARs_coef, 36 specify_prior_sigma(), 5
[.bayesianVARs_draws, 36 stable_bvar, 31
stable_bvar(), 5, 16
bvar, 2 stochvol, 5
bvar(), 6, 7, 10, 15, 17, 31, 32, 35 summary.bayesianVARs_bvar, 31
summary.bayesianVARs_bvar(), 5, 19
coef, 6 summary.bayesianVARs_draws, 32, 35
coef.bayesianVARs_bvar(), 5, 32, 33, 35 summary.bayesianVARs_draws(), 6
summary.bayesianVARs_predict, 33
fitted.bayesianVARs_bvar, 7
summary.bayesianVARs_predict(), 19
fitted.bayesianVARs_bvar(), 5, 11
update_fast_sv, 5
my_gig, 8
usmacro_growth, 34
pairs.bayesianVARs_predict
vcov.bayesianVARs_bvar, 35
(pairs_predict), 8
vcov.bayesianVARs_bvar(), 5, 6, 32, 33
pairs.bayesianVARs_predict(), 10–12, 14,
16
pairs_predict, 8
plot.bayesianVARs_bvar, 9
plot.bayesianVARs_bvar(), 5, 9, 11, 12, 14
plot.bayesianVARs_fitted, 10
plot.bayesianVARs_fitted(), 9–12, 14
plot.bayesianVARs_predict, 12
plot.bayesianVARs_predict(), 9–11, 14,
16
posterior_heatmap, 13
posterior_heatmap(), 9–12
predict.bayesianVARs_bvar, 15
predict.bayesianVARs_bvar(), 5, 8, 10, 12,
18, 33
print.bayesianVARs_bvar, 17
print.bayesianVARs_predict, 18
[Link].bayesianVARs_bvar, 18
[Link].bayesianVARs_predict, 19

rgig, 8

specify_prior_phi, 3, 20

38