Integration is by far one of the most important ideas in economics (Machlup 1977). After WWII, it diffused broadly across the globe (at last), deepened, and developed. Tinbergen famously contrasted positive to negative integration. Negative integration meant the removal of trade barriers, while positive integration signified the creation of new, common institutions (Tinbergen 1954). Balassa (1961) extended this analysis with his influential ‘integration stages’ framework. Lawrence (1996) distinguished between shallow and deep integration with the former associated with traditional trade agreements affecting tariffs, and deep integration with trade agreements that go beyond traditional areas and affect policies and regulations (Fernandes et al. 2021).
In this column, we argue that what we are currently observing is better described as ‘institutional integration’. Until the late 1990s, integration was mostly trade-centred while it now can be better described as institutions-centred. Institutional integration means that countries delegate to supra-national institutions some political control over selected policies. These new policy areas include social, labour, competition, environmental, and technological concerns, to name a few. Moreover, these policy areas have gone well beyond those covered by traditional trade agreements to the point that one may question whether trade remains the sole main driving force behind integration initiatives or whether this role is now shared with global value chains (GVCs), migration, capital flows, and foreign direct investment (FDI) (Baldwin 2011). The clearest example of institutional integration is the EU. Head and Meyer even argue that the EU is today more integrated than the US: “EU economic integration now matches or even beats the equivalent measure for the 50 American states” (2021, p. 45).
No way, Norway
How can one estimate the benefits from choosing institutional over deep integration? The identification strategy we put forward in a recent contribution (Campos et al. forthcoming) takes advantage of a unique natural experiment: Norway fulfilled all EU entry requirements, completed accession negotiations, accepted founding membership in the European Economic Area (EEA, which gave it unrestricted access to the Single Market), but in a 1994 referendum decided to reject full-fledged EU membership.
According to the European Commission, Norway was as ready to join the EU in 1994 as the other candidates then (Sweden, Finland, and Austria). The discovery of oil (natural gas) reserves preceded the first (second) EU referendum in 1972 (in 1994), so Norway is the only country to have voted twice to reject EU membership (Archer 2005). Because it rejected full-fledged membership in 1994, Norway enjoys the benefits from deep economic integration with the EU but not the benefits from institutional integration (here proxied by EU membership, as of the mid 1990s).
A synthetic difference-in-differences approach
The results in Campos et al. (forthcoming) represent one of the first applications of the synthetic difference in differences (SDID) approach (Arkhangelsky et al. forthcoming). The SDID approach combines attractive features of two widely used empirical methods, namely, the difference-in-differences (DID) and the synthetic control method (SCM) (Abadie 2021). SDID improves pre-exposure matching which lessens the reliance on parallel trends assumptions that concern much of the DID debate, while it also allows for valid large-panel inference which by its turn remains a concern in SCM applications.
A first important difference between the SCM and the SDID is that the latter includes unit fixed effects which allow the construction of reliable counterfactuals also when there are important differences in the levels of the outcome between treated and control units. This feature is important in our case. The pre-1995 productivity levels for most Norwegian regions are higher than those for the control regions. A second important advantage is that the SDID allows constructing standard errors for the point estimates of the effects. Third, it corrects for both unit and time weights, typically assigning larger weights to the years close to the end of the pre-treatment period, reducing the incidence of past shocks for the construction of the counterfactuals. A fourth key advantage is related to the fact we can use the SDID method to estimate the treatment effects in case of multiple (as in DID) as well as in case of a single treated unit (as in SCM) to assess heterogeneous effects.
New synthetic difference-in-differences estimates of the benefits of institutional integration
We uncover large and significant benefits from institutional integration in terms of productivity growth. Using regional and sectoral data, we are able to construct synthetic counterfactuals for Norwegian regions to evaluate actual post-1995 enlargement outcomes. Our estimates indicate that had Norway chosen institutional integration in 1995, instead of pursuing ‘purely’ deep economic integration, the average Norwegian region would have experienced an additional 0.6 percentage points in yearly average productivity growth. This is large as productivity growth is normally between 1.5 and 2% per annum (Syverson 2011). Moreover, we find the effects of not joining the EU vary across sectors (and regions), with large negative effects estimated for the industrial sector, while positive for non-tradable sectors.
We first estimate counterfactual series of productivity, by sector, in the multiple treated regions case. This means that, by sector, all the 19 NUTS 3 Norwegian regions are jointly pooled in the treatment group and compared to the counterfactual obtained by the donor pool of all the 75 NUTS 3 regions of Austria, Finland, and Sweden. Focusing on productivity (gross value added (GVA) per worker) the estimated effect of non-EU membership on Norwegian regions indicates that Norwegian regions experienced an average loss after 1995 in terms of GVA per worker of about €2,355. This estimate, however, is not statistically significant. In light of the heterogeneity across sectors and regions we encounter, the fact that the overall effect is not statistically significant may not be surprising. In particular, the industrial sector shows a large and statistically significant loss with respect to the counterfactual, both in terms of productivity levels and growth rates.
Figure 1 Individual treatment effects for gross value added (GVA) per employee, NUTS 3 region, and sector
Notes: Confidence interval at 5%. Data is for the total regional economy (all sectors aggregated, abbreviated as Total), regions, and for six broad sectors (NACE Rev. 2) as follows: Agriculture, forestry, and fishing (abbreviated with Agriculture); Manufacturing; Construction; Wholesale, retail, transport, accommodation & food services, information and communication (WRTAFIC); and Financial & business services (FBS).
Figure 1 shows the heterogeneity of the effects reporting, for region-sector pairs, the estimated effects and their confidence intervals. Concerning the total economy, 12 out of 19 regions display a negative impact and this is statistically significant for eight of them. For the remaining seven regions, we find a positive impact but which is only statistically significant for two of them. These mixed results for the total regional economies may not be that surprising given how heterogeneous the estimated effects are across sectors: a mix of positive and negative but mostly statistically non-significant impacts are obtained for both the agriculture and wholesale and retail sectors, mostly positive (and in part significant) for the financial and business service sector, negative but mostly non-significant for the construction sector, while for industry we obtain mostly negative and statistically significant results (with Akershus and Oslo displaying instead positive and significant impacts).
Figure 2 shows how these effects evolved over time. For post-treatment periods, it presents the yearly kernel density computed on individual region-sector pairs. Effects are more concentrated right after the decision of not joining the European Union and spread afterwards.
Figure 2 Kernel estimates computed on yearly estimated treatment effects
Notes: Kernel estimates computed on yearly estimated treatment effects for gross value added (GVA) per employee, NUTS 3 region, and sector. Data is for total regional economy (all sectors aggregated, abbreviated as Total), regions, and for six broad sectors (NACE Rev. 2) as follows: Agriculture, forestry, and fishing (abbreviated with Agriculture); Manufacturing; Construction; Wholesale, retail, transport, accommodation & food services, information and communication (WRTAFIC); and Financial & business services (FBS).
This column argues that the concept of deep integration is beginning to lose its explanatory power and this is in large part because of the deepening of integration itself. So many policy areas are now covered that one should ask whether trade remains such a powerful driving force of integration initiatives. We suspect that integration, especially at the frontier (which is the EU), now tends to proceed not through trade agreements but instead uses them as a common platform to focus on an array of other issues. What we observe at the frontier of integration is that trade issues are not as central as before and they have been replaced by regulations, policy coordination, and institutions (Campos et al. 2019.) The earlier we start thinking about these changes, the more we will be able to influence this process of integration driven by changes in institutions to maximise the benefits we show it can generate.
Abadie, A (2021), “Using Synthetic Controls: Feasibility, Data Requirements, and Methodological Aspects”, Journal of Economic Literature 59: 391–425.
Archer, C (2005), Norway Outside the European Union: Norway and European Integration from 1994 to 2004, London: Routledge.
Arkhangelsky, D, S Athey, D A Hirshberg, G W Imbens and S Wager (forthcoming), “Synthetic Difference in Differences”, American Economic Review.
Balassa, B (1961), The Theory of Economic Integration, New York: Routledge.
Baldwin, R (2011), “21st Century Regionalism and Global Trade Governance”, VoxEU.org, 23 May.
Campos, N F, F Coricelli and E Franceschi (forthcoming), “Institutional Integration and Productivity Growth in Europe: Synthetic Differences-in-Differences Evidence from the 1995 Enlargement of the European Union”, European Economic Review.
Campos, N F, F Coricelli and L Moretti (2019), “Institutional Integration and Economic Growth in Europe”, Journal of Monetary Economics 103: 88–104.
Dhingra, S, R Freeman and H Huang (2021), “The Trade and Welfare Benefits of Deep Trade Agreements”, VoxEU.org, 21 January.
Fernandes, A, N Rocha and M Ruta (2021), “The Economics of Deep Trade Agreements”, VoxEU.org, 23 June.
Head, K and T Mayer (2021), “The United States of Europe: A Gravity Model Evaluation of the Four Freedoms”, Journal of Economic Perspectives 35(2): 23–48.
Lawrence, R Z (1996), Regionalism, Multilateralism, and Deeper Integration, Brookings Institution Press.
Machlup, F (1977), A History of Thought on Economic Integration, New York: Springer.
Syverson, C (2011), “What Determines Productivity?”, Journal of Economic Literature 49: 326–365.
Tinbergen, J (1954), International Economic Integration, Amsterdam: Elsevier.