General-to-Specific Model Selection Procedures for Structural Vector Autoregressions

Structural vector autoregressive (SVAR) models have emerged as a dominant research strategy in empirical macroeconomics, but suffer from the large number of parameters employed and the resulting estimation uncertainty associated with their impulse responses. In this paper we propose general-to-specific model selection procedures to overcome these limitations. After showing that single-equation procedures are efficient for the reduction of the SVAR, but generally not for the reduction of its reduced form, the proposed reduction procedure is computer-automated using PcGets and its small-sample properties are evaluated in a realistic Monte Carlo experiment. The model selection procedure is shown to recover the DGP specification from a large unrestricted SVAR model with controlled size and power. The impulse responses generated by the selected SVAR are compared to those of the unrestricted and reduced VAR and found to be more precise and accurate. The proposed reduction strategy is then applied to the US monetary system considered by Christiano, Eichenbaum and Evans (1996). Although the selection process is hampered by the misspecification of the unrestricted VAR, the results are consistent with the Monte Carlo and question the validity of the impulses responses generated by the full system.