By Francisco Blasques
This paper proposes a functional specification approach for dynamic stochastic general equilibrium (DSGE) models that explores the properties of the solution method used to approximate policy functions. In particular, the solution-driven specification takes the properties of the solution method directly into account when designing the structural model in order to deliver enhanced flexibility and facilitate parameter identification within the structure imposed by the underlying economic theory. A prototypical application reveals the importance of this method in improving the specification of functional nonlinearities that are consistent with economic theory. The solution-driven specification is also shown to have the potential to greatly improve model fit and provide alternative policy recommendations when compared to standard DSGE model designs.
Traditionally, we specify a model, calibrate it and then apply a solution method. The latter has an impact on the result, though. For example, if a solution uses polynomial functions and it only preserve the properties of functions locally around the steady state, there is no need to use functional forms that have required properties beyond locally, especially if global properties impose additional unwelcome constraints. This means that functional-form choice depends on the solution method. And as the example in the paper shows, it can matter.