User’s Guide : Multiple Equation Analysis : Bayesian VAR Models : Estimating a Bayesian VAR in EViews
  
Estimating a Bayesian VAR in EViews
 
Prior Type
Hyper-parameters
Options
To estimate a Bayesian VAR in EViews, click on Quick/Estimate VAR... or type var in the command window to bring up the VAR Specification dialog. Select the Bayesian VAR as the VAR type in the radio buttons on the left-hand side of the dialog.
The dialog will change to the BVAR version of the VAR Specification dialog. As with a standard VAR, you may use the Basics page to list of endogenous variables, the included lags, and any exogenous variables, and to specify the estimation sample:
The three BVAR specific tabs, Prior type and Hyper-parameters and Options, allow you to customize your specification. The following discussion of the settings on these three tabs assumes that you are familiar with the basics of the various prior types and associated settings. For additional detail, see “Technical Background”.
Prior Type
The Prior type tab lets you specify the type of prior you wish to use, whether to include dummy observations, and options for calculating the initial residual covariance matrix.
You may use the drop-down menu to choose between Litterman/Minnesota, normal-Flat, normal-Wishart, independent normal-Wishart, Sims-Zha normal-flat, Sims-Zha normal-Wishart, and Giannone, Lenza & Primiceri priors.
The Dummy observations check boxes allow you to include dummy/additional observations to the data matrices of the VAR. The Sum-of-coefficients setting adds observations to account for possible unit-root issues, while the Dummy-Initial-observations setting adds observations to account for possible cointegration issues (see “Dummy Observations”).
For the priors other than normal-Flat or normal-Wishart, you may select options for estimating the initial (or prior) residual covariance matrix. A dropdown menu allows you to specify the type of estimate used to calculate the initial covariance:
Univariate AR estimates a univariate AR model (with number of lags matching those specified for the VAR) for each endogenous variable, then constructs the residual covariance matrix as a diagonal matrix with diagonal elements equal to the residual variance from the estimated univariate models.
Diagonal VAR uses the covariance from an estimated classical VAR model, but zeros out the off-diagonals.
Full VAR uses the covariance from an estimated classical VAR model.
AR(1) is computed in the same way as the Univariate AR, but only one lag is used.
The Exogenous variables radio buttons specify whether to include any exogenous variables specified in the VAR in the calculation of the initial covariance estimates.
If you select either of the Dummy observations choices, the Include dummy observations in calculation checkbox specifies whether to include those observations in the initial covariance calculation.
The Degree of freedom correction selection determines whether to use a degrees-of-freedom correction in the covariance estimate.
If the Giannone, Lenza & Primiceri prior type was selected in the drop-down menu, the Optimize checkbox selects whether to use the initial covariances as hyper-parameters, or as the starting values for an optimized hyper-parameter selection.
Finally, the Sample: box is used to specify the sample that is used to estimate the covariance. If left blank, the same sample as the overall VAR estimation.
Hyper-parameters
The Hyper-parameters tab allows specification of the hyper-parameters of the prior distribution.
If the Giannone, Lenza & Primiceri prior type was selected on the Prior type tab, the Optimize checkbox selects whether to use the values as hyper-parameters, or as the starting values for an optimized hyper-parameter selection.
Note tat only hyper-parameters that are available for the current prior type and choice of initial dummy observations are available for selection.
Options
The Options tab offers options for the Giannone, Lenza & Primiceri (GLP) prior and the inverse normal-Wishart prior. The GLP prior requires optimization of the hyper-parameters, and so options relating to the optimization algorithm; Convergence tolerance and Maximum iterations. The latter requires estimation through a Gibbs Sampler, and so offers options for the number of draws from the sampler, the percentage of draws to discard as burn-in draws, and the random number generator’s seed value.