Байесовская идентификация структурных векторных авторегрессий
We propose a new method of Bayesian identification of a structural vector autoregression based on the Bayesian model averaging. As compared to the literature on Bayesian SVAR averaging, the proposed algorithm can identify not only recursive, but also cyclical models given that some conditions specified in the paper hold. Bayesian model selection is made within the set of distinguishable on data models. We use simulations to assess the performance of the algorithm. We also check sensitivity of the proposed algorithm with respect to true parameter values, number of observations, and with respect to the parameters of prior distribution.