Solving and Estimating Finite Mixture Models in Parallel

In economics, numerical optimization is usually carried out using a package designed to optimize a black-box function f(x). Using a general-purpose package has many advantages, but it ignores the fact that objective functions in economics and econometrics have much in common. In particular, economic objectives are often formed from the solution of several smaller problems. This structure concentrates the computational cost of evaluating f(x) and they it limits the interactions among elements of x. Exploiting this structure can decreases the amount of time required to maximize the function. Furthermore, it can allow for the increased gains from parallel execution. This paper describes how to modify standard generic optimization code to exploit common elements of economic objective. In particular, the common element is solving and estimating models based on a finite mixture of heterogeneous agents The gains in computational efficiency under serial and parallel execution are described and some leading examples of optimization problems in economics and econometrics are mapped into the general framework.