Implementing Systematic Sensitivity Analysis Using GEMPACK

In economic simulation, results often hinge crucially on values of key exogenous inputs (the values of the parameters of the model and the shocks applied). Computational burden has, in the past, hindered systematic investigation of the impacts of variations in these key exogenous inputs. In this document, practical methods for conducting systematic sensitivity analysis for any model implemented using the GEMPACK suite of software are documented. The procedures described here are based on GTAP Technical Paper number 2 which sets out the theory behind the Gaussian quadrature methods on which the automated procedure is based. The procedures allow modellers to obtain estimates of the means and standard deviations of any endogenous variables of their model. The model only needs to be solved a relatively modest number of times (usually only 2N times if N exogenous inputs are varying); this is considerably fewer than the number of solves required by Monte Carlo methods. The procedure documented here fully automates solving the model as often as is necessary; once the user sets it up and starts it running, no further intervention is required. The document spells out the assumptions which must be made about the distribution of the exogenous inputs for the methods described to be valid. Five examples of systematic sensitivity computations are presented and the accompanying software allows modellers to work through these examples while reading the document. This should leave readers fully prepared to analyse the sensitivity of results for any model implemented in GEMPACK.