bsvars

Bayesian Estimation of Structural Vector Autoregressive Models

This package provides efficient algorithms for Bayesian estimation of Structural Vector Autoregressive (SVAR) models via Markov chain Monte Carlo methods. A wide range of SVAR models is considered, including homo- and heteroskedastic specifications and those with non-normal structural shocks.

Installation

Just open your R and type:

install.packages("bsvars")

The package is under intensive development. To the the newest changes install it by typing:

devtools::install_git("https://github.com/bsvars/bsvars.git")

Start your Bayesian analysis of data

The beginnings are as easy as ABC:

# upload the package
library(bsvars)

# upload data
data(us_fiscal_lsuw)
 
# specify the model and set seed
specification  = specify_bsvar_sv$new(us_fiscal_lsuw, p = 4)
set.seed(123)
 
# run the burn-in
burn_in        = estimate(1000, specification)

# estimate the model
posterior      = estimate(50000, burn_in$get_last_draw(), thin = 10)

# Now, you're ready to analyse your model!

Starting from bsvars version 2.0.0 a simplified workflow using the |> is possible:

library(bsvars)
set.seed(123)
us_fiscal_lsuw |>
  specify_bsvar_sv$new(p = 4) |>
  estimate(S = 1000) |> 
  estimate(S = 50000) |> 
  compute_impulse_responses(horizon = 8) -> irfs

Progress bar

MCMCs for multivariate dynamic structural models might take a while to run. bsvars relies on efficient algorithms and fast codes developed using cpp code via Rcpp and RcppArmadillo packages to cut the time of computations by orders of magnitude. Still, while waiting these a few minutes, you can track the progress by looking at the beautiful progress bar:

License

This package is distributed under license GPL (>= 3)