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. Despite their advantages, just-identified SVAR models suffer from (i) the great number of parameters (“curse of dimensionality”), (ii) the resulting uncertainty as-sociated with impulse responses, (iii) the existence of alternative observationally-equivalent just-identified models and (iv) the lack of identification of the imposed causal ordering of the variables of the system. In this paper we propose general-to-specific reductions of just-identified SVAR mod-els to overcome these limitations. We show that the computer-automated model selection algorithm embodied in PcGets (see Krolzig and Hendry, 2001) can be used for an efficient implementation of the underlying methodology. Since jointly selecting and diagnostic testing eludes theoretical ana-lysis, we evaluate the proposed reduction and identification strategy by simulation. The application of the proposed reduction strategy to a US monetary system demonstrates the feasibility of PcGets for the analysis of large macroeconomic data sets