A fundamental point to understand is that any type of model linkage is fraught with difficulty, since major model differences exist in terms of the data, the assumed be- havioural parameters and the underlying structural mech- anisms of the models. This, however, does not mean that such a linkage should not be attempted, but, rather, one should have realistic expectations of what can be achieved when trying to harmonise different modelling approaches.
It is well known that PE and CGE models have structural differences, in terms of both the data and the behavioural elements (i.e. explicit or implicit elasticities), that can generate divergent results, while precedents in the literature even show that CGE and PE models can generate contradictory findings for the same scenario.14 Although this is recognised within the modelling community, in the policy arena it can often be hard to reconcile the findings
of both types of model when presenting a consistent argument for a given policy reform.
In the past, DG AGRI-commissioned research established a ‘soft’ model linkage (see, for example, Nowicki et al., 2007; Nowicki et al., 2009; Helming et al., 2010), such that different types of models generated a mutually consistent storyline. Typically, a soft linkage is driven by an ad hoc assessment of the overall results (i.e. are the models broadly telling the same story?), while one plays to the strengths of each model to serve as a source of input to the other. For example, the CGE model, with an explicit or endogenous treatment of factor markets, world trade and macroaggregates, could conceivably be used within a PE model. Similarly, the sectoral detail and econometric foundation in supply response that serves some PE models well could be employed to assess and improve the veracity of CGE model results.
The advantage of the soft approach is that it is relatively straightforward to implement in terms of the necessary modelling modifications. On the other hand, the ‘soft’ approach adopted in the Scenar 2020 project through the linkage of variables was, as noted above, implemented on more of an ad hoc basis rather than by following a systematic framework. Thus, subject to the prejudices of the model scenario (i.e. the scenario design, the type of shocks, etc.), the choice of variable linkage could conceivably vary considerably.
In Philippidis et al. (2017), a ‘test bed’ study, which considered a class of ‘soft’ model linkage between CAPRI and MAGNET, was carried out as a preparatory step of the Scenar 2030 study. The aim of the study was to ascertain the extent to which the MAGNET model results diverge between two experiments, a ‘standard’ MAGNET experiment and a scenario where MAGNET implements agri-food output results from the CAPRI model directly by performing a closure swap with a Hicks neutral output productivity variable, while sectoral prices are allowed to be continually adjusted endogenously.
The aim of the exercise was to ascertain the extent to which MAGNET model results diverge from the ‘standard’ results when soft-linking to CAPRI and to assess the degree of compromise required in MAGNET to accommodate said changes. It was found that the standard MAGNET model and the CAPRI model predictions implemented in MAGNET ‘more often than not’ predict the same signs for output. In the EU-15, however, where a more significant number of agri-food sectors are linked, there are quite a few sign differences between CAPRI predictions and those of the standard MAGNET model, while in the non-EU regions, the level of convergence is generally good. This evidence suggests that there is a need to have some form of linkage between the models, especially if the focus is on the EU. The choice of a ‘hard’ linkage to forge a union between the structural or behavioural elements of the model (see, for example, Britz & Hertel, 2011; Pelikan, et al., 2015) becomes appealing because it follows a very specific methodological approach, but it requires considerably more modelling expertise to implement, while the potential robustness of the two models being linked is, at the current time, under scrutiny and far from certain.15 In the abovementioned papers, an elegant method for structurally linking CAPRI to a specific GTAP model version was applied. The approach does not have to impose heavy restrictions in either of the two models (especially if, in CAPRI, one does not pass back crop supply prices from the GTAP model). Such an approach would be worth pursuing for potential policy-orientated work, and a ‘pilot’ study is currently under development involving CAPRI and MAGNET modellers.
These tests and literature reviews give insights into the scientific dimension of how to link models. As research and trials with large-scale models, mainly with CAPRI and MAGNET, are still ongoing, Scenar 2030 takes a more pragmatic approach. In a follow-up study, it is hoped that a more sophisticated, ‘hard-linkage’ approach will be pursued to link CAPRI and MAGNET.
Chapter 4 further outlines the detailed implementation of the baseline and the scenarios in Scenar 2030.
15 Within the two cited studies, the policy shocks were very discrete, while a more aggressive set of policy shocks (i.e. projections, etc.) which are typically used to characterise policy outlooks have, hitherto, not been attempted.