Confidence intervals for computable general equilibrium models
MetadataShow full item record
Computable general equilibrium (CGE) models have expanded from being a simple theoretical tool to a widely accepted policy evaluation tool. Despite recognizing that model parameters involve uncertainty, virtually all modelers report their results without confidence intervals. This obscures the uncertainty inherent in the models and gives the impression that the results are far more certain than they actually are. CGE models with calibrated parameters and econometric CGE models using only the mean value of the parameters share a common flaw: their results are point estimates only, with no indication of the range of possible variation. A better analysis would include confidence intervals that communicate the underlying uncertainty. This would allow the policy makers to understand the precision of the results. In this dissertation a tractable formal technique for calculating confidence intervals is presented. The results from this approach are compared with sensitivity analysis, an alternative method sometimes used for assessing uncertainty. It is shown for the models presented that sensitivity analysis can produce misleading results and that the confidence intervals are feasible to compute and qualitatively superior. Next, the technique is applied to an econometric intertemporal general equilibrium model of the US economy to examine a current policy issue. The strong form of the double dividend hypothesis, which asserts that revenue-neutral substitution of an environmental tax for a distortionary income tax can improve welfare, is tested. The intertemporal equivalent variation (EV) for the policy is calculated. Unlike other studies, however, the 95 percent confidence interval for the EV is presented. The mean EV is slightly negative but the confidence interval is large and includes zero, so the model neither supports nor rejects the double dividend hypothesis. In addition, the short-run and the long-run intratemporal EV is calculated and compared to the intertemporal EV. The result implies that the long-run result supports the double dividend hypothesis even though the short-run does not. Finally, I present a detailed analysis of the general equilibrium effects that yield these distinct and contradictory results. In sum, this dissertation provides an econometric view of CGE modeling and statistical testing of CGE results that is acceptable to econometricians. It attempts to answer criticisms of CGE modeling and the wider challenge to empiricism in economics (Whalley, 1985).