This paper argues that, in studying the monetary policy transmission process, more emphasis should be given to the systematic portion of policy behavior and correspondingly less to random shocks basically because shocks account for a very small fraction of policy-instrument variability. Analysis of the effects of the systematic part of policy requires structural modelling, rather than VAR procedures, because the latter do not give rise to behavioral relationships that can plausibly be regarded as policy-invariant. By use of an illustrative open- economy structural model based on optimizing analysis, and considering variants, the paper characterizes the effects of policy parameter settings by means of impulse response functions and root-mean-square statistics for target errors. Different models give different answers to questions about the effects of systematic policy, so procedures for scrutinizing model specification are essential. In this regard, it is argued that vector autocorrelation functions, augmented by variance statistics for each of a model's variables, seem more promising than impulse response functions because the latter require shock identification, which is inherently a difficult process....
