Abstract
Spatial CGE models rely on detailed multiregional input–output (MRIO) tables. This paper compares two different approaches to compiling MRIO tables for Austria – an algorithm-based approach that regionalizes national input–output tables (IOT) and generates trade estimates using a predefined set of regional variables (i.e. Horridge’s algorithm), and a hybrid approach that uses as much regional and interregional data as possible. We investigate whether we observe differences in CGE simulation results that use them. Results from an aggregate simulation are surprisingly similar. So the algorithmic approach is, in fact, effective in making an MRIO from a national IOT. But noticeable differences appear at the sectoral level. They seem mainly due to differences in calibration rather than in regionalization.
Acknowledgments
The authors gratefully acknowledge helpful comments received from the editor and two anonymous referees.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 Data recently published by Statistik Austria should be considered as a remarkable exception.
2 Here, the former is a required but not sufficient condition for the latter.
3 Israilevich et al. (Citation1996) evaluate the impact of different sources of regional IOT on simulation results, applying an econometric-input-output model of the Chicago economy. In another paper, Sargento et al. (Citation2012) make a similar exercise for the European Union (treating the member states as its regions).
4 For more details about TERM model see Horridge (Citation2011).
5 As a result, national totals can change slightly during the balancing procedure.
6 For details see Fritz et al. (Citation2005).
7 Following other papers Bayesian techniques were applied (e.g. Rodrigues, Citation2014).
8 Still, excess supplies must be exported or satisfy demands from other regions (outflows) and excess demands must be satisfied by inflows and imports.
9 For details concerning interregional trade estimation in TERM, see Horridge (Citation2011).
10 In all of the cases we also performed long-run simulations. To save space, we do not report the results here as they are in line with the short-run ones. They can be obtained from the authors upon request.
11 As suggested by Jensen (Citation1980), high partitive accuracy is hardly achievable in the regional context given the data constraints. This, however, does not impede the accomplishment of holistic accuracy at the aggregated level.
12 The number was chosen to provide easy comparison with IO multipliers.