HierarchicalBVARResult.beta_draws

HierarchicalBVARResult.beta_draws: NDArray[float64]

(S, w, k) independent posterior draws of the coefficient matrix.

Each draw is laid out exactly as beta_mean, and draw s was generated conditional on sigma_draws[s] from the matrix-normal conditional \(B \mid \Sigma, Y \sim \mathcal{MN}(\bar B, K^{-1}, \Sigma)\). The draws are i.i.d., not a chain: no burn-in or thinning was applied and none is needed. Every propagating method – irf(), irf_bands(), forecast_paths(), posterior_replications() – iterates over this axis, so its length S is n_kept.