Centralized and Decentralized Strategies for Multi-Country Macroeconometric Modelling. The Case of the European Commission (1978-1991)

How did macroeconometric models address interdependencies across national economies? What were the distinctive theoretical, empirical, and technical challenges raised by modelling economic interdependence? This contribution to the history of macroeconomics aims at documenting how these two questions have been addressed by macroeconomists through multi-country macroeconometric models. We focus more specifically on the multi-country macroeconometric models developed in the 1970s and 1980s by macroeconomists working for the European Commission (EC) Directorate General for Economic and Financial Affairs (DG II hereafter). The DG II started with developing the EUROLINK model (in 1978), then the COMPACT model (1986), the QUEST I model (1988), and the QUEST II model (1991). This case study allows us to answer the two questions above and to address a third question: What reasons motivated this sequence of models? We argue that the development of different models was both inspired by debates within the DG II (notably the integration of the models to the DG II short and medium-term forecasting and policy analysis routines) and by external factors, notably the development of macroeconometric models (both single-country and multi-country) in Europe and in the United States. Our history of multi-country macroeconometric models built by the DG II uncovers two alternative strategies for multi-country modelling. The first approach is what we label the ‘decentralized' approach: the multi-country model is built by connecting, for instance through trade equations, single-country macroeconometric models, developed already (and independently) by national teams. This modelling strategy was initiated in 1968 by Project LINK, a U.S.-based initiative (under the aegis of Lawrence Klein); then, the same strategy was also applied by (among the most knowns) the Federal Reserve Board Multi-Country Model (MCM), the OECD Interlink model, and the EPA World Model of the Japan Planning Agency. The DG II as well initially embraced this decentralized approach by building the EUROLINK model (1978-1983), which brought together existing national models for Italy, France, West Germany, and the UK. The second and alternative approach is what we call the ‘centralized' approach: the multi-country model is built by one single team, designing every ‘country-block' of the model. Such approach was notably developed in the mid-1970s by macroeconomists based in Brussels, in particularly André Dramais (Université Libre de Bruxelles). From the mid-1980s, the DG II shifted definitely to this latter approach, by building COMPACT, then QUEST—the approach that is still favoured today. Our paper, relying notably on archives, tells the history of this shift, highlighting the theoretical, empirical, and computational arguments within the DG II that brought the adoption of the centralized strategy. We illustrate how the main difficulty with the decentralized approach chosen by EUROLINK was the heterogeneity of national models—heterogeneity across several dimensions, including notably their different theoretical specifications, their sizes, and hence the diverse level of details in representing each national economy. Such a complexity made it difficult for macroeconomists at the DG II to understand the economic mechanisms at work in EUROLINK simulations; moreover, complexity limited the possibilities for DG II macroeconomists to adapt EUROLINK specifications to their own needs for forecasting and simulation. Conversely, the centralized approach had been considered as allowing ‘full control' ("intellectual command") on modelling choices and uses for DG II macroeconomists.