bsvars v2.0.0

[donotdespair]

The new version 2.0.0 of the bsvars package for Bayesian Estimation of Structural Vector Autoregressive Models in R is now available on CRAN!

The official CRAN page is: https://cran.r-project.org/package=bsvars

Package description

The 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. All models include three-level equation-specific local-global hierarchical prior distribution for the estimated level of shrinkage for autoregressive and structural parameters.

The new version 2.0.0

This version is not backward compatible, so you might need to rewrite your existing scripts. This is because the new version:

• includes a new estimate method replacing the estimation functions for individual models estimate_bsvar_*(). To estimate the model just use the estimate routine:

library(bsvars)                                                   # upload the package
data(us_fiscal_lsuw)                                              # load the data
specification  = specify_bsvar_sv$new(us_fiscal_lsuw, p = 2)      # specify the model and set seed
posterior      = estimate(specification, 10000)                   # estimate the model THAT'S IT!

• it also facilitates the use of the |> pipe for simplified workflows. The script above can be rewritten as:

library(bsvars)                                                   # upload the package
data(us_fiscal_lsuw)                                              # load the data
us_fiscal_lsuw |>
   specify_bsvar_sv$new(p = 2) |>                                 # specify the model
   estimate(S = 10000) _> posterior                               # estimate the model VOILA!

• provides routines for fast computations for impulse responses, forecast error variance decompositions, historical decompositions, computations of structural shocks, fitted values, conditional standard deviations, and regime probabilities for MS and MIX models,
• implements a new three-level equation-specific local-global hierarchical prior distribution for the estimated level of shrinkage for autoregressive and structural parameters,
• includes a new model similar to that by Chan, Koop, Yu (2022)
• implements even faster random number generators to speed things up even more!
• and many more that can be found at NEWS

Further developments

The package is under intensive development, and more functionalities will be provided soon! To see the package ROADMAP towards the next version 2.1.0.

Join the conversation!

Have a question, or suggestion, or wanna get in touch? Join the package DISCUSSION forum.

Greetings, Tomasz @donotdespair

---

This discussion was created from the release bsvars v2.0.0.

[donotdespair]

Hi Dear Team @MarkosFarag @YuFuOliver @lcq110 @aschnuecker (apologies for tagging individuals - no GH enterprise yet 😉)

I invite you to join the conversation about your use, suggestions, and questions about the bsvars package here.

I can't wait for what you are up to!

Cheers, T @donotdespair

[donotdespair]

Hey @MarkosFarag

I remember about your email. I was just waiting for the new package version release as now, I can provide help that would be linked to this version and would not have to be adjusted soon after the release. The new package version is not backwards compatible so a new script is needed. I'm getting on it soon!

Cheers, T @donotdespair

[MarkosFarag]

Hi Prof. Tomasz, Thanks a lot for the update. I'm looking forward to hearing back from you regarding the issue I have with the code. Kind regards, Markos

…

On Tue, Oct 24, 2023 at 8:50 AM Tomasz Woźniak ***@***.***> wrote: Hey @MarkosFarag <https://github.com/MarkosFarag> I remember about your email. I was just waiting for the new package version release as now, I can provide help that would be linked to this version and would not have to be adjusted soon after the release. The new package version is not backwards compatible so a new script is needed. I'm getting on it soon! Cheers, T @donotdespair <https://github.com/donotdespair> — Reply to this email directly, view it on GitHub <#39 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOCQ7YW5BIHUWCEEIFJXUBLYA5QMLAVCNFSM6AAAAAA6NDLTR2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TGNRWGU3TC> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[donotdespair]

Hey Marcos,

How are you? It took me a while but I have it! I rewrote your script so it's adjusted for version 2.0.0 of the package. Your problems came from using version 1.0.0 functions for the developer's version of the package 1.0.1.9000 which was an early version of 2.0.0. And 2.0.0 is not backward compatible.

But there you go. You have a working script and I believe that I included hints regarding how to use other compute_* functions as well.

I also included lots of comments, e.g., regarding your identification of B. It is not identifying the homoskedastic system, and one should expect problems there. But your data is heteroskedastic, and you should use the bsvar model with stochastic volatility anyway. This would identify the model and let you use your B matrix.

Let me know if you have more questions! And thank you for your interest in this package! Let me know how your research report progresses!

The sample script for a heteroskedastic model is:

install.packages("bsvars") # please confirm that version 2.0.0 is installed
library(bsvars)
Data <- read_excel("Dataset.xlsx")
y = as.matrix(Data[,2:7])     # TW: data prepaation for bsvars (only matrices supported ATM)

B = matrix(
  c(T,T,T,T,T,T,F,T,F,F,F,T,T,T,T,F,T,T, F,F,F,T,T,T,F,F,F,T,T,T, T, T,T,T,T,T), 
  6,6);B                      

# a heteroskedastic BVAR specification with your B:
###############################################################
specification = specify_bsvar_sv$new(y, p = 4, B = B)
set.seed(123)
specification |>
  estimate(10000) |>
  estimate(20000, thin = 2) -> posterior     # this run will save the final run in object posterior
# This posterior is normalised!

plot.ts(t(posterior$posterior$B[6,,])) 
# the plot above reveals that the normalisation did not work that well and your 
# posterior is still multimodal. That's bc of your identification of B. 
# Try running longer final MCMC

irfs    = compute_impulse_responses(posterior, horizon = 20)
irfs_med = apply(irfs, 1:3, median)
plot.ts(t(irfs_med[,1,])) # irfs to the first shock

volatility = compute_conditional_sd(posterior)
plot.ts(t(apply(volatility, 1:2, median))) # plot of the estimated conditional standard deviations

Greetings, T @donotdespair