Bayesian VAR prior dummies: BVAR package

The BVAR package can be used to create the basic types of prior dummy observations when estimating Bayesian VAR models. The dummy observations are passed in the VAR/estimate function through the 'BVAR=' option.

Constructing dummy observations

• covmat - Covariance matrix prior dummy observations for BVARs.
• litterman - Litterman's prior dummy observations for BVARs.
• sumofcoeff - Doan et al sum-of-coefficient prior dummy observations for BVARs.
• uncmean - Unconditional-mean dummy (or Sims' initial dummy) observations for BVARs.
• user - User-supplied prior dummy observations for BVARs.

Weights on prior dummy observations

The prior dummies produced by litterman, uncmean, sumofcoeff can be weighted up or down using the input argument Mu. To give the weight a clear interpretation, use the option 'stdize=' true when estimating the VAR. In that case, setting Mu to sqrt(N) means the prior dummies are worth a total of extra N artifical observations; the weight can be related to the actual number of observations used in estimation.

Getting help on BVAR functions

help BVAR
help BVAR/function_name