Abstract:
Identifiability of the parameters is an important precondition for consistent estimation of models designed to describe empirical phenomena. Nevertheless, many estimation exercises proceed without a preliminary investigation into the identifiability of its models. As a consequence, the estimates could be essentially meaningless if convergence to the 'true' parameters is not guaranteed in the pertinent problem. We provide some evidence here that such a lack of identification is responsible for the inconclusive results reported in recent literature on parameter estimates for a certain class of nonlinear behavioral New Keynesian models. We also show that identifiability depends on the subtle details of the model structure. Hence, a careful investigation of identifiability should preceed any attempt at estimation of such models.
