A sensitivity study on identification schemes of the structural vector autoregression

摘要

Current study is motivated initially by a casual observation that the lag length in VAR causes drastic fluctuations in the estimation of structural parameter matrix in Structural VAR model under Gordon & Leeper's 1994 JPE approach, thus tampering inference made concerning monetary policy effects based on impulse response functions and forecast error variance decomposition.;With the ever increasing popularity of VAR or SVAR in macroeconomic research to draw inference and implications of monetary policy rules, it becomes undoubtedly important to know and make certain that the inference is robust to identification schemes employed in SVAR. We use various measures to assess the variability of structural matrix relative to lag length specification in the estimation procedure of VAR. A Monte Carlo simulation study is carried out in an attempt to separate the variability of structural parameters induced by the particularity of the data series from the variability introduced by the identification schemes.;The Monte Carlo study confirms the sensitivity of Gordon & Leeper's 1994 JPE structural approach. We also contrast their approach to Christiano, Eigenbaum & Evans (1996)'s identification scheme and conclude that the latter is more robust in inferences.