The identification of significant terms in a flexible metamodel using spectral analysis

Complex economic systems are frequently modeled using flexible functional forms. Flexible functional forms have desirable mathematical and statistical properties, and are analogous to simulation metamodels. The estimation of these flexible metamodels can lead to serious statistical problems. These problems are mitigated by reducing the number of terms in the model specification. The Schruben-Cogliano response surface methodology is a method for identifying the significant terms in a simulation metamodel. In this paper, the ability of the Schruben-Cogliano methodology to alleviate the estimation problems associated with flexible cost functions is explored. The method is applied to the simulation model of multi-product tomato processing plant. The results indicate that the Schruben-Cogliano method identifies the metamodel specification with the highest goodness of fit.