Gaussian Quadratures vs. Monte Carlo Experiments for Systematic Sensitivity Analysis of Computable General Equilibrium Model Results

Abstract

Third-order Gaussian quadratures (GQ) approximate the mean and variance of model results allowing for computationally inexpensive sensitivity analysis to uncertainty in exogenous parameters. Unfortunately, commonly used GQ approaches restrict the marginal distributions of both parameters and results sacrificing valuable distributional information. Using higher order quadratures, or incorporating more uncertain exogenous parameters, rapidly increases the sample size, undermining the rationale for using GQ. In contrast, Monte Carlo methods directly approximate the distribution of model outcomes without restrictive distributional assumptions on exogenous parameters. We argue that current computing capabilities allow for wider use of Monte Carlo methods for conducting stochastic simulations.