An Introduction to Systematic Sensitivity Analysis via Gaussian Quadrature

Abstract

Economists recognize that results from simulation models are dependent, sometimes highly dependent, on values employed for critical exogenous variables. To account for this, analysts sometimes conduct sensitivity analysis with respect to key exogenous variables. This paper presents a practical approach for conducting systematic sensitivity analysis, called Gaussian quadrature. The approach views key exogenous variables as random variables with associated distributions. It produces estimates of means and standard deviations of model results while requiring a limited number of solves of the model. Under mild conditions, all of which hold with respect to the GTAP model, there is strong reason to believe that the estimates of means and standard deviations will be quite accurate.