Algorithmic Design and Beowulf Cluster Implementation of Stochastic Simulation Code of Stochastic Simulation Code for Large Scale Non Linear Models

Anderson & Moore describe a powerful method for solving linear saddle point models. The algorithm has proved useful in a wide array of applications including analyzing linear perfect foresight models, providing initial solutions and asymptotic constraints for nonlinear models. However, many algorithmic design choices remain in selecting components of a nonlinear certainty equivalence equation solver. This paper describes the present state of development of this set of tools. The paper descibes the results of simulation experiments using the FRBUS quarterly econometric model and the Canada Model. The paper provides data characterizing the impact of solution path length, initial path guess, terminal constraint strategy and strategies for exploiting sparsity on computation time, solution accuracy and memory requirements. The paper compares algorithm performance on traditional unix platform with our recent Beowulf Cluster Parallel Computation Implementation.