The effect of observables, functional specifications, model features and shocks on identification in linearized DSGE models

Both the investment adjustment costs parameters in Kim (2003) and the monetary policy rule parameters in An & Schorfheide (2007) are locally not identifiable. We show means to dissolve this theoretical lack of identification by looking at (1) the set of observed variables, (2) functional specifications (level vs. growth costs, output-gap definition), (3) model features (capital utilization, partial inflation indexation), and (4) additional shocks (investment-specific technology, preference). Moreover, we discuss the effect of these changes on the strength of parameter identification from a Bayesian point of view. Our results indicate that researchers should treat parameter identification as a model property, i.e. from a model building perspective.