(A) bayesian analysis to detect structural changes in a simultaneous dynamic linear model

Abstract

It is important in view of model maintenance to detect structural changes in econometric or times series models. Bayesian or classical approaches have been done to identify structural change points, types and sizes. They are mostly based on an univariate analysis, a trend-cycle decomposition and a specification of random shocks using reduced-type models - state space or autoregressive integrated moving average models. These methods have been applied to macroeconomic time series and various types of shocks have been analyzed. Specially gross national or domestic products have been studied using state space models. In these analyses structural changes are classified into random shocks in a trend and a cycle, and they are estimated by using forecast errors. Detected random shocks have been interpreted as structural changes in a demand and a supply side respectively. However, it cannot be said that these analyses dealt with demand and supply structural changes directly. The reason is that a random shock affects not only the one time series in a model but also other related time series out of the model. In order to solve this problem structural changes are estimated and interpreted in multivariate and structural-type models where demand and supply equations determine gross domestic products and gross price level together. This approach can deal with demand and supply shocks directly and can be intimately related to macroeconomic policies by adding exogenous variables, for example money supply and government expenditure. A proposed simultaneous dynamic linear model is established considering integration and cointegration of the time series. In this model all variables have unit roots in autoregressive polynomials to reflect that a stochastic time trend dominates a deterministic time trend. The variables in a demand equation are cointegrated to reflect that a demand shock remains in each series temporarily whereas the variables in a supply or other equations are fir...