Toolkit julia source codes and example outputs for Bayesian (Structural) Vector Autoregressive (VAR) models. There are identification strategies in multivariabe time series analysis that requires bayesian framework, such as,
- Dynamic probabilistic forecasting estimations
- Stochastic volatility
- Time-varying parameters
- 'Big-data' or large dimension models
- Structural identification (partial-equilibrium)
A lot of the research and source codes are mainly written in MATLAB. The purpose of this repository is to direct-transalte those source codes from academic research and codes publicly available into open-sourced julia programming languge. This repository does not claim original authorship of the algorithms translated and used, and I recommend users to see the original research cited below.
This repository also includes applications of the models, such as estimation of impulse responses, foreasting, and scenarios/simulations.
| Source code | Model framework | References |
|---|---|---|
|
Bayesian AR application with state-space stochastic volatility | Mein: technically not multivariate but a good baseline use case |
|
Bayesian Structural Time Series model | Mein: technically not multivariate but a good baseline use case |
|
Bayesian VAR model with Gibbs sampling (Minnesota-prior) | Gary Koop and Dimitris Korobilis replication of Christiano et al. (2016) |
|
Bayesian VAR model with varying prior for extreme episodes | Cascaldi-Garcia - Pandemic prior |
|
Time-varying parameter VAR with stochastic volatility (Code directly available; application example of macroeconomic consumer sentiment) | Harron Mumtaz replication code |
|
Bayesian Structural Vector Autoregressive Model with Sign Restriction | Baumeister and Hamilton (2015) replication code |
|
Mixed-frequency VAR model | Harron Mumtaz replication code |
|
Dynamic Factor Model |