mfbvar: mfbvar: A package for mixed-frequency Bayesian vector...

In mfbvar: Mixed-Frequency Bayesian VAR Models

Description

The mfbvar package makes estimation of Bayesian VARs with a mix of monthly and quarterly data simple. The prior for the regression parameters is normal with Minnesota-style prior moments. The package supports either an inverse Wishart prior for the error covariance matrix, yielding a standard normal-inverse Wishart prior, or a time-varying error covariance matrix by means of a factor stochastic volatility model through the factorstochvol-package package.

Specifying the prior

The prior of the VAR model is specified using the function set_prior. The function creates a prior object, which can be further updated using update_prior. The model can be estimated using the steady-state prior, which requires the prior moments of the steady-state parameters. The function interval_to_moments is a helper function for obtaining these from prior intervals.

Estimating the model

The model is estimated using the function estimate_mfbvar. The error covariance matrix is given an inverse Wishart prior or modeled using factor stochastic volatility. If the former is used, mdd can be used to estimate to the marginal data density (marginal likelihood).

Processing the output

Plots of the output can be obtained from calling the generic function plot (see plot-mfbvar). If factor stochastic volatility is used, the time-varying standard deviations can be plotted using varplot. Predictions can be obtained from predict.mfbvar.

mfbvar documentation built on Feb. 10, 2021, 5:12 p.m.