Global identification of linearized DSGE models

This paper introduces a time domain framework to analyze global identification of stochastically nonsingular DSGE models. A formal identification condition is established that relies on the restrictions linking the observationally equivalent minimal state space representations and on the inherent constraints imposed by them on deep model parameters. We next develop an algorithm that checks global identification by searching for observationally equivalent model parametrizations. The algorithm is efficient as the identification conditions it employs shrink considerably the space of candidate deep parameter points and does not require solving the model at each of these points. We also derive two complementary necessary conditions for global identification. Their usefulness and the working of the algorithm are illustrated with an example.