Model comparison for VAR/VEC models with stochastic volatility?

[vestedinterests]

Congrats and thanks for the package development! I think this is really one of the few options for estimating Vector Error Correction Models with stochastic volatility in R, works like a charm!

I've encountered an issue, however, when trying to use the summary.bvarlist on models with stochastic volatility.

The following works (taken from the vignette, substitung VAR for VEC with the default sigma priors):

library(bvartools)

set.seed(123456) # Set seed for reproducibility

data("e1") # Load data
data <- diff(log(e1)) * 100 # Obtain log-differences

# Use date up to 1978Q4
data <- window(data, end = c(1978, 4))

object <- gen_vec(data,
    p = 1:3,
    iterations = 5000, burnin = 1000
)

object <- add_priors(object,
    coef = list(v_i = 0, v_i_det = 0),
    sigma = list(
        shape = 0.01,
        rate = 0.01,
        mu = 0, v_i = 0.01,
        sigma_h = 0.05,
        constant = 1e-04
    )
)

object <- draw_posterior(object, mc.cores = 3)

summary(object)

  p s r        LL       AIC       BIC        HQ
1 1 1 1 -487.5117  983.0235  992.1302  986.6489
2 1 1 2 -519.6540 1049.3080 1060.6913 1053.8397
3 2 1 0 -472.2474  950.4948  957.3248  953.2138
4 2 1 1 -455.6306  925.2613  941.1980  931.6057
5 2 1 2 -501.8354 1019.6707 1037.8840 1026.9215
6 3 1 0 -450.6170  913.2341  926.8941  918.6722
7 3 1 1 -444.5118  909.0237  931.7903  918.0871
8 3 1 2 -501.3037 1024.6073 1049.6507 1034.5771

When estimating this model with sv = TRUE, however, I get the following error message.

object <- gen_vec(data,
    p = 1:3,
    iterations = 5000, burnin = 1000, sv = TRUE
)
# add priors and draw_posterior like above
summary(object)
> Error in matrix(NA, tt, draws) : non-numeric matrix extent
   In addition: Warning message:
   In cbind(temp_pars, object[[i]][[j]]) :
     number of rows of result is not a multiple of vector length (arg 1)

The same error message also comes up for for

object <- gen_var(data,
    p = 1:3,
    iterations = 5000, burnin = 1000, sv = TRUE
)

Is this a mathemical issue (impossibility), or rather a programmatic one? It would be helpful to make informed decisions about lag-length choice for those kind of VEC or VAR models.

R version 4.4.1 (2024-06-14)
Platform: x86_64-apple-darwin20
Running under: macOS 15.5

other attached packages:
[1] bvartools_0.2.4 Matrix_1.7-2    coda_0.19-4.1