A1811
Title: Asymptotic properties of Bayesian inference for structural changes in multivariate regressions
Authors: Jaewon Lee - Korea University (Korea, South) [presenting]
Yunjong Eo - Korea University (Korea, South)
Abstract: Under a fairly general set of assumptions and a wide class of priors, we explore the asymptotic properties of Bayesian inference in multivariate regression models with multiple structural breaks, where changes can occur in both regression coefficients and the covariance matrix of the errors. We establish asymptotic equivalence between the highest posterior density (HPD) region and confidence sets for breakdates, along with boundedness on the joint marginal posterior distribution and a large-sample correspondence between the posterior density ratio and the likelihood ratio. Moreover, we validate a Bernstein-von Mises-type theorem for regression coefficients in the context of multivariate regressions with multiple breaks. The consequences of misspecifying the model are discussed. Our Monte Carlo analysis confirms the Bernstein-von Mises theorem and the similar behavior of HPD regions and inverted likelihood ratio confidence sets, support our findings.
