GTAP Resources: Resource Display
| GTAP Resource #3565 |
|---|
|
"Systematic Sensitivity Analysis for GTAP – Where We’ve Been and Where We’re Going" by Preckel, Paul, Monika Verma, Thomas Hertel and Will Martin Abstract We never precisely know the values of the exogenous parameters in policy models. Values of these parameters have an impact on the results of a policy analysis, and we need to know how robust results are to variation in these parameters. DeVuyst and Preckel [1997] proposed using Gaussian quadrature and demonstrated its efficacy for assessing the mean and variance of a policy outcome. The method is implemented within the GEMPACK and has been used with the GTAP model for several years. Users have noted two shortcomings with the method: a) the initial implementation allowed for only two cases of stochastic relationships (either perfect correlation or stochastic independence) and b) the sample points for the model inputs do not range very far from the mean – plus or minus only a small multiple of the standard deviation of the parameter. Artavia et al. [2009] note there are multiple approaches to incorporating correlation between parameters and highlight two, based Cholesky decomposition and Eigen-system decomposition, and recommend the one that is computationally simplest. We examine the performance of these two procedures, and find that the computational performance is similar. The narrow nature of the sampling is likely a problem is when the model has many policy or technology regimes. For highly nonlinear models, we expect that sampling more broadly could improve our estimates of the statistics of model results. We present a new quadrature that contains double the points of the original quadrature, but samples as broadly as desired. Simulation methods are used to compare this broader sampling quadrature with Monte Carlo estimates of the statistics of model outputs. The impact of broader sampling appears ambiguous and model specific. However, the models employed do not exhibit regime switching, and further work is needed. Finally, we provide more extensive comparisons of Gaussian quadrature based sensitivity analysis than have been previously published. |
| Resource Details (Export Citation) | GTAP Keywords | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
- Software and modeling tools |
| Attachments |
|---|
|
If you have trouble accessing any of the attachments below due to disability, please contact the authors listed above.
Public Access  2011 Conference Paper (195.9 KB) Replicated: 0 time(s)Restricted Access No documents have been attached. Special Instructions No instructions have been specified. |
| Comments (0 posted) |
|---|
|
You must log in before entering comments.
No comments have been posted. |
Last Modified: 9/15/2023 1:05:45 PM
