Var::drawrescov appears to return identical draws to Var::drawcoefs for a Sims-Zha Normal-Wishart BVAR

[finparis]

I am estimating a Bayesian VAR with 4 endogenous variables and drawing from
the posterior distributions using drawcoefs and drawrescov. I believe I
may have found a bug (or at least behavior that contradicts the
documentation) specific to the Sims-Zha Normal-Wishart prior.

Model specification:

var var_l1.ls 1 1 dtx_3m spread g_v_yoy soldep_p @ c
var_l1.bvar(prior=sznw, initcov=diag, nsumcoef, ninitobs, mu1=0, lambda1=0.2) 1 1 x y z w

- 4 endogenous variables, 1 lag, constant included
- Prior: Sims-Zha Normal-Wishart (sznw), with both dummy-observation
options (nsumcoef, ninitobs) enabled

Commands run:

var_l1.drawcoefs(draws=10, seed=1234567) mCoefDraws
var_l1.drawrescov(draws=10, seed=1234567) mSigDraws

Observed:
- Both mCoefDraws and mSigDraws are 10 x 20 matrices.
- After inspecting the two matrices element by element, they contain
EXACTLY THE SAME VALUES. mSigDraws does not appear to contain any
residual-covariance-related information; it is indistinguishable from
mCoefDraws (which appears to contain estimated-coef-related information).

Expected, from the help documentation:
- drawcoefs: "Each row of the matrix will contain the vec'd
coefficients from a single draw."
- drawrescov: "Each row of the matrix will contain the vec'd form of
the covariance matrix from a single draw."

These are described as draws from two distinct posterior distributions
(coefficients vs. residual covariance), so I would not expect the two
output matrices to be identical.

However, given that the two outputs are not just similarly sized but numerically identical, this
looks more likely to be an implementation issue specific to
prior=sznw, where drawrescov may not be correctly separating out the
residual covariance component of the posterior draw.

As there seems to be a bug with this drawrescov command, is there an alternative, reliable way to draw from
the posterior residual covariance distribution for a Sims-Zha Normal-Wishart BVAR in the EViews 13 version?

FB

[EViews Gareth]

Might be worth downloading the EViews 14 demo and trying it there.