Solving large scale normalised rational expectations models

The novelty of the method presented here is, that it is applied to a small subset of model variables only, the so called feedback variables, that an approximate Jacobian is used and that a subperiod method is introduced. This leads to significantly smaller Jacobians and less calculations for solving a model. The feedback variables are determined by an ordering algorithm.

The paper describes the modification to Newton's method by describing an Extended Feedback Jacobian, introducing an approximate 'Shift' Jacobian to reduce calculation time and a subperiod method to reduce storage space for the Jacobian matrix. The method has been implemented at the CPB and used for its GAMMA model. This paper also reports the results of some experiments with Multimod mark III. These experiments show faster convergence than the Fair-Taylor method and significantly smaller matrices than the Stacked-Time method.