BVARBUILDPRIORMN Procedure

@BVARBuildPriorMN builds a "Minnesota" prior for a VAR model using repeated calls to the @BVARBuildPrior procedure. After @BVARBuildPriorMN, you need to use @BVARFinishPrior to convert the working matrices to the form needed for multivariate Gibbs sampling.

See GIBBSVARBUILD.RPF for an example of its use.

@BVARBuildPriorMN( options ) hbpriors hpriors

@BVARFinishPrior hbpriors hpriors bprior hprior

Parameters (required)

hbpriors	VECT[VECT] accumulation of precision matrix \(\times\) prior mean vectors for the elements of the prior for each equation.
hpriors	VECT[SYMM] accumulation of the precision matrices for the elements of the prior for each equation

Output Parameters for @BVARFinishPrior (required)

bprior	Stacked VECTOR of prior means for the equations of the model.
hprior	VECT[SYMM] accumulation of the precision matrices for the elements of the prior for each equation

Options

MODEL=MODEL for which the prior is being constructed

SCALEFACTORS=VECTOR of OLS autoregression scale factors

[CALCULATE]/NOCALC

If you need to set up several priors based upon a single set of data, you can use NOCALC to avoid doing the calculation of the cross product matrix and the scale factors. You need to have the scale factors the first time through. Then use the same option on the future calls.

EQUATION=equation defining the form of the estimated equation

INIT/[NOINIT]

Initialize (dimension and zero out) the hbprior and hprior

TIGHT=overall tightness parameter [.20]

OTHER=relative tightness on other variables [1.0]

MATRIX=RECTANGULAR array of weights for general handling of variables.

CONTIGHT=relative tightness on deterministic variables [flat]

DECAY=harmonic decay exponent (std dev is \(l^{-decay}\)) [0=no decay]

MEAN=prior mean for first lags for all equations [1.0]

MVECTOR=VECTOR of prior means of first lags of each equation (overrides MEAN if used).