Modelling monetary policy using SPSA-based neural networks

Simultaneous perturbation stochastic approximation based neural networks (SPSA-NN) is introduced for modeling economic policy for the first time. A simulation method is used to compare the performances of two monetary policy models for the US economy pursuing a targeted objective. In the first method a one-step linear feedback policy is used, and in the second method a policy based on SPSA-NN is used. It is shown that SPSA-NN policy is much faster to learn the system, and once it is learned, the policy is quick to adjust to the changes.