Solving nonlinear stochastic optimization and equilibrium problems backwards

In a stochastic equilibrium model some stochastic processes are usually exogenously given, while others are either chosen optimally by agents or emerge from market equilibrium conditions. When we simulate such a model, often we aim at studying the relations among variables in the model as we vary parameters of policy and of behavior of economic agents. We are no more certain (indeed often less certain) of what is reasonable or interesting behavior for the exogenous variables (some of which may be unobservable) than of the variables chosen by agents or fixed in markets. It turns out that if we are flexible about which variables’ behavior we take as given in the model solution computation, freeing ourselves from the convention that the variables exogenous to the model economy must be taken as given in the simulation computations, great computational savings may result.