Multi-Scenario Sensitivity Analysis with a very large CGE Model

The narrowing of GDP per capita gaps between sub-national regions in Europe through economic growth may depend on distance and degree of economic integration, as reflected in the parameter settings and market structure assumptions of a spatial computable general equilibrium (CGE) model. In this paper we develop and apply a new method of sensitivity analysis that allows for the interactions between key parameters and is parsimonious in the number of model runs combining policy shocks. The application results in a mapping of all NUTS2 regions showing overall sensitivity and the extent to which regions are prone to converge to the average GDP per capita of the country and the EU as a whole. Spatial CGE models pose a number of specific challenges to systematic sensitivity analysis that are turned into numerical advantages in the method applied, in particular by allowing multiple model run processes to be executed in parallel. The practice of borrowing econometric estimates from empirical growth studies and setting elasticities which are the same for all regions is overridden by a process in which information on the interaction within and between policy shocks and behavioural parameter changes is gradually extracted from the iterations with the model. The results will give an indication of which regions are most sensitive to moves towards greater integration in either direction, good or bad. As last step of SA with the CGE model of NUTS2 regions, we ran the model 27 times (3**n,n=3) for the three combinations of the three most influential elasticities, varying each of them by +/-10%, for the same TFP shock as in previous exercise. Analysing the results for each combination of elasticities' values, we analysed the average of all macroeconomic indicators in all regions, normalising the value of each variable to its post-simulation mean. After measuring the average relative variation of each macroeconomic variable after the 27 model runs, we generated an overall average per parameter combination following the same weighting procedure as before (based on the inverse of the average of the absolute values of pairwise correlations). This exercise permitted us to identify the elasticities associated with the highest and the lowest impact on the macroeconomic outcomes for the same shock