The measurement equation is a VAR with time-varying coefficients: yt=Xt′βt+At−1Ht1/2εt, where yt is the n×1 vector of observables, Xt=In⊗(1,yt−1′,…,yt−p′) is the Kronecker-structured regressor block, βt stacks all VAR coefficients at time t, and εt∼N(0,In). The factored error structure At−1Ht1/2 separates the contemporaneous interactions (in At) from the idiosyncratic shock variances (in Ht).
The transition equations are random walks. Coefficients: βt=βt−1+ηt, ηt∼N(0,Q). Free elements of At: αt=αt−1+ζt, ζt∼N(0,S). Log-volatilities: lnhit=lnhi,t−1+σiνit, νit∼N(0,1). The random-walk specification is parsimonious - it avoids estimating mean-reversion parameters - but implies that the coefficients can wander arbitrarily far from their initial values, which occasionally produces explosive draws that must be discarded.
Estimation proceeds via Gibbs sampling. The algorithm cycles through: (i) draw β1:T from p(β1:T∣y1:T,A1:T,H1:T,Q) using the Carter-Kohn simulation smoother; (ii) draw α1:T from p(α1:T∣y1:T,β1:T,H1:T,S) using the same algorithm on the partially orthogonalized residuals; (iii) draw each lnhi,1:T from p(lnhi,1:T∣⋅) using the Kim, Shephard, and Chib (1998) mixture-of-normals approximation; (iv) draw the hyperparameters Q, S, σ12,…,σn2 from their conditional posteriors (inverse-Wishart or inverse-gamma). Convergence is monitored via trace plots and the Geweke (1992) diagnostic.