Report on the Dynamics of Differentiated Integration

This report examines the dynamics of differentiated integration. In particular, it studies whether differentiated integration is a path-dependent process, in which earlier differentiation increases the likelihood of additional differentiation in the future. In a set of analyses on the EU’s treaty-based and legislative differentiation and the Eurozone crisis, this report provides consistent evidence for path-dependence. The analysis of treaty reforms shows that the extent of pre-existing differentiation in a policy area is correlated significantly with differentiations in subsequent treaty reforms – even when controlling for exogenous factors of differentiated integration. In turn, differentiation in EU legislation follows differentiation in EU treaties. Whereas such path dependence only takes place in the area of core state power integration, it is clearly the most important factor driving legislative differentiation. Finally, in the Eurozone crisis, the differential impact of the crisis on euro area and non-euro area countries motivated the euro area countries to adopt far-reaching centralizing reforms and financial commitments that non-euro area members refused or were excluded from. In sum, differentiated integration has not only been pronounced in policy domains related to core state powers – such as macroeconomic and security policies – when they were initially integrated, but the legal gap between insiders and outsiders has grown over time.


Introduction
Is differentiated integration a 'slippery slope'? According to an important concern about DI, once the EU embarks on differentiation in a given policy area or with a specific member state as a strategy to overcome negotiation deadlocks, DI will not only harden into permanent divides among the member states, but also gain momentum over time. Put differently, is differentiation a dynamic process?
Dynamic differentiation refers to a temporal process in which earlier differentiations change the probability of subsequent differentiations, ceteris paribus. Generally, the feedback mechanism in dynamic processes comes in two major variations. Positive feedback processes are self-reinforcing. They increase the future probability of an event: the more differentiations already exist, the more will be negotiated in the future. By contrast, negative feedback processes are self-undermining and thus decrease the future probability of an event. If differentiation was self-undermining, we would observe the fewer future differentiations, the more had been negotiated in the past. Positive feedback processes are also known as path-dependent; negative feedback processes as backlash. If we start from the normative position that uniform integration is desirable, the major concern is path-dependent differentiation. A negative feedback process would correct itself in the direction of uniform integration if differentiations get out of hand. By contrast, path dependence implies that a return to uniform integration requires a major exogenous shock or intervention.
In a previous report (Schimmelfennig and Winzen 2020a, D4.3/D4.4), we argued that differentiated European integration does not provide any obvious evidence for this concern in general. For one, whereas the number of differentiations has increased in the course of European integration, it has remained stationary relative to the growth of the EU's membership, policy portfolio and legal production. In EU treaty law, the level of differentiationtaking into account the much larger number of member states and integrated policiesis about the same as in 1958 when the Treaties of Rome entered into force. In secondary legislation, we even found a downward trend that lasted until 2008. More recently, we observe a rise in legislative differentiations. Yet the current level of secondary-law differentiation is still below the level observed before the mid-1990s.
Moreover, differentiated integration in the EU is predominantly temporary or 'multi-speed' differentiation. Most differentiations negotiated in the history of the EU by far have expired after a reasonable period of time. In the case of treaty-based differentiations, only one third of the differentiations negotiated in the course of European integration were still in force at the end of 2019, and those differentiations that have already expired have lasted less than six years on average. We did not find either that differentiations have become more long lasting over time.
We also observed that a significant minority of differentiations has proven durable, defying the logic of multi-speed differentiation. These differentiations only account for a sixth of the treaty-based differentiations, but they cluster in specific member states and important areas of European integration: the monetary and interior policies that were the most important additions to the EU beyond the internal market. The most durable differentiations are the Eurozone non-membership of the UK, Denmark, Sweden, the Czech Republic, Hungary, and Poland; the Schengen non-membership of the UK, Ireland, and Cyprus; Denmark's opt-out from the common defence policy; exemptions from the free movement of capital (mainly relating to foreign ownership of land property) for Denmark, Estonia, Malta, and Hungary; and exemptions from Justice and Home Affairs for Denmark, the UK, and Ireland. As such, however, durable differentiation does not constitute evidence for a dynamic, self-reinforcing process of differentiated integration in the EU.
Yet our aggregate descriptive findings may be misleading. For one, path-dependent developments for specific countries and individual policies may be hidden in the large number of cases we study. In addition, the fact that most differentiations have expired after a few years and that their relative number has not increased does not exclude that these differentiations increased the probability of further Robert Schuman Centre for Advanced Studies Working Papers differentiation as long as they were in forceand that the number of differentiations would be lower today otherwise.
This report presents three analyses of differentiated integration that aim to identify the pathdependent component in differentiated integration more precisely. The first focuses on the effect of existing differentiations in EU treaty making. The second looks at the impact of treaty differentiation on differentiation in the legislations of the EU. And the final analysis is a case study of differentiation in the Eurozone crisis. 1 All three analyses conclude that path dependence is a relevant and consistent factor in differentiated integration. However, path dependence appears to be contained within those policy sectors and countries that experienced high initial and durable differentiated integration. Simply put, path-dependent differentiation is a correlate of the European integration of core state powers, and it is driven either by Eurosceptic member states that refuse to participate or by new member states deemed unfit to participate.

Path-dependence in treaty-based differentiated integration
The first part of the report examines differentiated integration in EU treaties from the Treaty of Maastricht (1992) to the 2016 intergovernmental agreement on the Single Resolution Fund, the latest reform treaty. The unit of analysis is the 'differentiation opportunity', a combination of member state, policy area and reform treaty. Each treaty is preceded by an intergovernmental conferences, during which representatives of the member states negotiate changes in a varying number of policy areas. Whereas a few treaties focus on a single policythe Schengen Agreement or the Treaty on the European Stability Mechanism are cases in pointreforms of the Treaty on European Union from Amsterdam to Lisbon cut across many domains. In the Maastricht Treaty itself, 28 policy areas were reformed. In each policy area, individual member states may negotiate exemptions and derogations. At Maastricht, the 12 member states were thus faced with 336 differentiation opportunities corresponding to an equal number of observations in the data. To establish whether a member state concluded the negotiations on any of the negotiated policies with an opt-out, the analysis uses the EUDIFF1 dataset.
Importantly, in every treaty negotiation, we only code an opt-out if a country is exempt from new rules. If a treaty creates new rules in a policy area, and a country fully participates in these rules, we do not code a differentiation, even if the country has active opt-outs in this policy area from previous reforms. Moreover, we also include new policy areas. For instance, the Maastricht Treaty included the Common Foreign and Security Policy, Economic and Monetary Union (EMU) and Justice and Home Affairs, for which there could not have been a DI precedent. Both design choices make sure that we obtain a conservative estimate of the path-dependent effects of differentiated integration.
For the same reason, it is important to control for other potential causal conditions of new differentiations. Without doing so, a new differentiation in a policy area, in which a member state already has a differentiation, may appear as a case of path dependence, whereas in fact both the earlier and the new differentiation are causally unrelated and driven by exogenous factors. We therefore include those factors that have been shown to co-vary with initial differentiations for member states. If we find that existing differentiations are significantly correlated with new ones after including these exogenous factors in the analysis, we can claim path dependence with less uncertainty.
These exogenous factors are wealth, identity, core state powers, Euroscepticism and ratification by referendum (Schimmelfennig and Winzen 2020b;Winzen and Schimmelfennig 2016). First, heterogeneity of wealth is a major factor in DI. Both the wealthiest and the poorest member states are 1 The three analyses build on, modify and update earlier work published in Schimmelfennig and Winzen (2019) for the first analysis and Schimmelfennig and Winzen (2020b) for the second and third ones. Please refer to these works for more detailed information on the theoretical background, the variables, the measurements, the research design, and for robustness checks.

Report on the dynamics of differentiated integration
European University Institute 3 particularly likely to be integrated differentially. Wealthy member states sometimes refuse to participate in additional steps of integration because they can afford to stay out and are concerned about disproportionate financial burdens; by contrast, less affluent member states sometimes are excluded because the other member states are concerned about their capacity to participate effectively. We measure heterogeneity of wealth at the country-level based on the absolute distance of a country from average EU wealth in terms of gross national income (GNI) per capita using World Bank data.
Second, member states with more exclusive national identities are more likely to opt out of treaty reforms aiming to deepen integration than states with a population that sees itself at least partly as 'European'. Here, we measure the percentage of Eurobarometer respondents in each member state feeling 'exclusively national'. Third, policies belonging to the sovereignty-sensitive class of 'core state powers' (Genschel and Jachtenfuchs 2014;Rittberger et al. 2014) are particularly likely to exhibit differentiation. We code those manually. The same is true for government-level Euroscepticism, which we measure as the seat-weighted average EU position of all governing parties, based on the Chapel Hill expert survey data (Bakker et al. 2015). Moreover, we code whether a member state ratified a treaty by referendum, because failed treaty referendums have been an important pathway to differentiated integration. Finally, we code our main variable of interestexisting differentiationat the policy-level as a count of pre-existing national opt-outs in a policy area at the time of a reform treaty negotiation. This coding not only tells us whether a policy was differentiated at all, but also to what extent it was differentiated. Figure 1 reports the main results of a regression analysis of our dataset of differentiation opportunities. The 'full model' is a standard logistic regression model including our main variable of interest alongside the control variables. The 'RE model' makes an effort to control for unobserved treaty, policy and country differences by using a multilevel structure with random effects for these clusters.
The analysis of both models shows that the effect of existing DI is robust. Even when controlling for country wealth, exclusive national identities, the Eurosceptic positions of governing parties, the use of referendums and the type of policy (core state powers), existing DI retains a positive and significant effect on the probability of differentiation in the same policy area. Robert Schuman Centre for Advanced Studies Working Papers Yet the analysis also shows that path dependence is one effect among many and not the dominant one. The integration of core state powers creates the largest effect on differentiated integration, and the heterogeneity of wealth is at least equally relevant as existing differentiation. Yet the other control factors have smaller and less certain effects than existing DI, and neither wealth nor core state powers explain away the knock-on effects of established precedent.  We thus conclude that path-dependent differentiation plays a relevant role in EU treaty making. Optouts and exclusions agreed in earlier treaty negotiations significantly increase the probability that subsequent treaties will produce more of the same. This probability rises with the number of previous differentiations and independently of those factorssuch as disparities in national wealth, identities and Euroscepticismthat generated differentially integrated policy areas in the first place. In the next section, we examine whether this finding holds for legislative differentiation as well.

Path dependence in EU legislative differentiation
Differentiations in treaties may not only have an effect on differentiation in subsequent treaties, but also on the legislation that is based on these treaties. The treaties ('primary law') define general competences, goals and principles. Legislation ('secondary law') then elaborates substantive rules on this basis. In this system, three pathways to legislative differentiations are conceivable. First, treaty-based differentiation produces legislative differentiation in the secondary law that has its legal basis in the differentiated treaty articles, and it typically concerns those member states that are exempted or excluded from the treaty article. This would constitute positive feedback or path dependence.
Alternatively, however, legislation may compensate or circumvent treaty-based differentiation by creating bridges over the treaty-based gap (Adler-Nissen 2009). This would amount to negative feedback or legislative backlash against primary-law differentiation. This expectation is not unreasonable given that legislation does not require domestic ratification in the member states and is more limited in scope than treaty reform. It is therefore less likely to bring up salient issues of sovereignty and produce public attention and scrutiny. Especially in those cases in which differentiation was enforced against the will of the government or political elites more broadly, as in failed referendums on treaty ratification, the governments and MEPs of opt-out countries may also have an interest in mitigating the divide through legislative alignment. Finally, of course, legislative differentiation (or unity) might arise independently of the treaties and differentiated legislation might generate its own path dependenceby creating a precedent for subsequent legislation in the same area. In this section, we will only examine treaty effects, however.
How strongly treaty opt-outs influence legislative differentiation depends on their design. In principle, the more discretion legislators have, the more legislative differentiation depends on bargaining rather than treaty rules. Some treaty opt-outs exempt countries from entire policy regimes. The key cases encompass differentiated integration in the Eurozone, Schengen area, and interior justice. EU legislation in these areas exempts some countries because they have treaty opt-outs. Yet, other treaty differentiations exclude countries from some but not all facets of a policy. This variation in the design of treaty opt-outs suggests a strong but imperfect correspondence between treaty and legislative differentiation. Moreover, if policy-wide treaty opt-outs become more common, the link between treaty and legislative differentiation should become stronger.
Nevertheless, even policy-wide treaty differentiation does not fully determine legislative differentiation. First, there are not many policy-wide exemptions. The aforementioned examples are the only cases. More importantly, even these far-reaching treaty exemptions leave room for re-interpretation in the legislative process (Adler-Nissen 2009). This is clear in the design of the British opt-outs from Justice and Home Affairs policies. The British government was able to decide on a case-by-case basis whether to participate in new laws (opt-in). There are more examples. Britain opted out only of the free travel rules of the Schengen area, but it participated in the Schengen Information System for law enforcement. Furthermore, countries that have not adopted the Euro are excluded from most legislation on the governance of the common currency. However, some rules in this area, such as on banking Robert Schuman Centre for Advanced Studies Working Papers regulation, may be applicable to these countries, rendering the legislative reach of their treaty opt-outs a matter of bargaining rather than design.
Finally, in the EU as elsewhere, treaty negotiators do not anticipate all conflicts of interest or deliberately avoid controversial matters in the hope to resolve them in day-to-day decision-making. This further enhances the incompleteness of the EU treaties and means that governments might seek opt-outs not specified by the treaties. First, even if governments do not disagree over treaty principles, they might clash over their specification in legislation. These disagreements might trace back to the same factors as treaty differentiation, yet only surface when concrete rules rather than general principles are at stake. Second, developments at the level of the treaties might create problems in the legislative process and encourage differentiation.
We explore our arguments in an analysis of legislative differentiation opportunities, which we define in parallel to the differentiation opportunities in treaty negotiations (Section 2). Legislative differentiation opportunities arise from the EU's adoption of legislation during which a country could seek an opt-out. Whereas governments can in principle seek opt-outs after the law is in force, e.g. through amending legislation, initial legislative differentiation is the most important case. The EUDIFF2 dataset tells us about the legislative acts that are differentiated, their legal basis in treaty law, and the member states that use these differentiations.
On the side of explanatory variables, the analysis is interested in the relationship between treaty-level developments and legislative differentiation. To this end, we identify whether a country has a treaty optout in the issue-area in which a new piece of legislation falls. The operationalization of the remaining variables follows that of the previous section. However, since this analysis covers a longer time framefrom 1973 when the first Eurobarometer surveys became available-several differences warrant discussion. First, we operationalize country-level wealth as GDP per capita in 1000 United States (US) Dollars, in 2005 prices, based on data by Gleditsch (2002). The previous section relied on GNI per capita but we can cover a longer period using GDP.
Again we measure exclusive identity as the share of Eurobarometer respondents saying that they feel 'exclusively national'. However, the Eurobarometer did not ask this question before the 1990s. In the period with data for both variables, we therefore regress identity on Euroscepticism and country fixed effects. This model accounts for about 80 percent of the variation in identity during the 1990s and 2000s. Then we out-of-sample predict identity for the 1970s and 1980s. This predicted identity correlates at .91 with observed identity during the 1990s and 2000s.
Our operationalization of government party support for European integration draws on data from the comparative party manifesto project rather than the Chapel Hill expert surveys (Budge et al. 2001;Volkens et al. 2012). The manifesto project covers the post-war period whereas the Chapel Hill data start in the mid-1980s. The manifesto project classifies quasi-sentences in partisan election manifestos in terms of their policy focus and tonality. We measure party support for the EU based on a party's share of positive quasi-sentences on the EU relative to negative quasi-sentences. Then we create a seatweighted average governmental position on the EU.
Finally, as an additional control variable, we include the absolute distance of a government from the EU's average left-right position. EU law-making invokes socio-economic conflicts in addition to questions of sovereignty (Hix et al. 2007). Possibly, governmental distance from the Union's left-right centre of gravity inspires differentiation demands. Relying on the left-right scale of the manifesto project, we measure each government's position as the seat-weighted average position of the governing parties. Then, we take the distance of a country from the EU average. In addition, we controlled for the time since accession for new member states. We lagged the explanatory variables by two years as it typically requires two years for new EU legislation to enter into force.
We begin with a bivariate analysis. Here we find a strong increase in path-dependent legislative differentiation since the 1990s. Since the 1990s, member states have used nearly 15 percent of their

Report on the dynamics of differentiated integration
European University Institute 7 opportunities to opt out of legislation in areas in which they had treaty differentiations (Figure 3). This share is below one percent in areas not covered by treaty differentiation. Before the 1990s, treaty differentiations, then exclusively the result of enlargement treaties, also encouraged legislative opt-outs but far less frequently. This periodization coincides with the integration of core state powers, which we found to be particularly influential for the path-dependent differentiation of treaty law as well.

Figure 4 The effect of treaty-based differentiations on legislative differentiations Robert Schuman Centre for Advanced Studies Working Papers
The multivariate analysis compares three models that cover different time periods and policy contexts. The first examines differentiation in market, expenditure, and regulatory policies from 1973 to 1990, the early history of European integration. The second and third models cover the years from 1993 to 2012. Whereas the second model encompasses legislation in market, expenditure, and regulatory policies, the third focuses on core state powers. 2 The main finding is a strong relationship between treaty and legislative differentiation in core state powers after 1990. This relationship is much weaker in the market, expenditure, and regulatory domains before or after 1990. Before 1990, it did not matter strongly whether or not a treaty differentiation existed in a given area. If so, this changed the predicted probability of a legislative differentiation for the affected country by about 1 percentage point. The results differ starkly in the area of core state powers in which the EU obtained competences after 1990. In this area, a treaty differentiation is associated with more than a 60 percentage point increase in the median probability that the member state obtains an opt-out from subsequent legislation. In other words, after 1990, a close correspondence between treaty and legislative differentiation has emerged in the domain of core state powers.
In contrast with differentiation in EU treaty-making, existing (treaty-based) differentiation is the dominant factor in legislative differentiation. No other factor comes even close to the substantive significance of treaty-level DI when it comes to explaining the occurrence of legislative differentiation in the area of core state powers. This observation should not be exaggerated, however. In the analysis of treaty differentiations, we included policy type (core state powers) as a variable where it explained the largest share of the variation. By contrast, in the analysis of legislative differentiation, we used it to split the population of cases but did not enter it into the analysis. It therefore did not suppress the size of the effect of existing DI. Had we analysed treaty articles concerning core state powers exclusively and thereby taken 'core state powers' out of the model, we would have likely found a larger effect of existing DI for treaty differentiations as well.

Path dependence in the Eurozone crisis
Monetary policy has always been the most differentiated policy area in European integration. Monetary policy cooperation started outside the treaty framework with the Currency Snake in 1972, which was superseded by the European Monetary System (EMS) in 1979. Both regimes set narrow bands in which national currencies were allowed to float. Yet only five member states (Belgium, Denmark, Germany, Luxembourg, and the Netherlands) participated continuously in the Snake and the EMS. Economic and Monetary Union (EMU) has continued this pattern. Monetary policy was the first treaty-based EU policy from which member states -Denmark and the UKwere granted a full and formal opt-out, and it is the policy area with the most non-participants among the member states. In addition, EMU is an area of durable differentiation. The British and Danish opt-outs are the longest-running differentiations in the EU treaties; and for most of the new member states having joined since 2004, non-participation in EMU has been the most durable treaty-based differentiation.
The case of the Eurozone crisis (roughly lasting between 2009 and 2015) is thus a most likely case for path-dependent differentiation. If it is correct that a policy will experience more differentiation, the more differentiated it already is, then we should see path-dependence most clearly in EMU. Moreover, we would expect that crises amplify this effect. Crises expose the vulnerabilities and deficits of the integrated policy and put insider countries (Eurozone member states in this case) under pressure to move ahead with integration to fix the deficits and avoid the costs of breakdown. Outsiders (non-euro area member states) do not experience the same pressures. In addition, the different constellations of preferences within insider and outsider groupsinsiders are more willing, and outsiders are more reluctant, to consent to further integrationcan create a situation in which insiders move ever further 2 Our analysis omits the years between 1990 and 1993 in which German re-unification led to exceptionally many new differentiations. It also excludes Luxembourg, which is an extreme outlier in terms of GDP per capita. ahead with integration and make it ever more unlikely for outsiders to opt in (Jensen and Slapin 2012). In other words, the initial reasons for which countries have opted out from or been excluded from more integration, are likely to be reproduced and strengthened by crises in the differentiated groups of member states. The Eurozone crisis thus allows us to see how a differentiated policy develops under stress, and to consider whether differentiated policy is a slippery slope towards an increasingly loose or divided union.
Path dependence in a differentiated policy area has three observable implications. First, path dependence reproduces the original membership patterns. Participants in the integrated policy participate in new steps of integration, whereas the exempted or excluded countries do not. Insiders will not become outsiders, and outsiders will not become insiders (unless they had been preparing to join the insiders anyway). Second, the institutional gap between insiders and outsiders widens, because institutional developments in the differentiated policy, which respond to crisis shocks, will be limited to the insiders. Third, because of policy externalities that need to be managed to consolidate the integrated policy, DI will expand beyond its original policy area into new, formerly uniformly integrated policy areas. This is in essence what happened in the Eurozone crisis. The Eurozone countries not only strengthened EMU in order to avoid a risky and costly breakup of the Eurozone but also developed a 'banking union' in order to prevent future sovereign debt crises. The banking union was built as a differentiated policy on top of the EU's uniform financial market regulation. In terms of membership patterns, the Eurozone did not suffer any exits but admitted the Baltic countries, which had been on track to join before the crisis hit. On the other hand, the non-euro area countries refrained from participation in any institutional reforms that would have reduced their fiscal and financial autonomy and entailed potentially costly international assistance.
Most fundamentally, the Eurozone as such managed to avoid disintegration and the exit of individual Eurozone countriesin spite of frequent expectations of collapse and hard bargaining by the German and Greek finance ministers, Schäuble and Varoufakis, including threats of 'Grexit' (Varoufakis 2017). By contrast, the Baltic countries -Estonia, Latvia, and Lithuaniajoined the Eurozone between 2011 and 2015. These accessions do not count as effects of the Euro crisis, however: the Baltic countries had intended and prepared for joining the Eurozone since their EU accession in 2004. By contrast, the Euro crisis did not persuade any hitherto sceptical country to adopt the Euro. Rather, the crisis reinforced popular scepticism in the outsider countries (Hobolt 2014).
Moving beyond Eurozone membership, the institutional and legislative reforms introduced during the Eurozone crisis, clearly separate Eurozone members and non-members. Table 1 gives an overview of the most important pieces of treaty law and secondary legislation responding to the financial and Euro crisis. It shows a pervasive pattern: Eurozone outsiders have generally supported treaty revisions and legislation strengthening the fiscal and financial integration of the Eurozone, and have participated in reforms that aim at better macroeconomic coordination and financial market regulation, but have not taken part in policies entailing major financial commitments, supranational centralization and sanctions. As a result, the reforms have increased differentiation between Eurozone insiders and outsiders. Whereas outsiders have had an interest in the preservation and consolidation of the Eurozone and have therefore agreed to the necessary policy changes for the Eurozone, they have been unwilling to reduce their own autonomy and enter into strong legal or financial commitments. Robert Schuman Centre for Advanced Studies Working Papers The European Financial Stability Facility (EFSF) and the European Stability Mechanism (ESM), which requires its members to pay in a capital stock of 80 billion euros and has a lending capacity of a minimum of 500 billion, produce considerable liability risks to which only Eurozone countries have committed themselves. The Euro Plus Pact of March 2011, which not only commits member states to sustainable public finances but also to strengthening competitiveness, employment, financial stability, and tax policy coordination uses the non-hierarchical Open Method of Coordination. For this reason, it was joined by six non-Eurozone countries.
By contrast, the six-pack and two-pack legislation as well as the Treaty on Stability, Coordination and Governance (TSCG) reduce national budget autonomy and increase the sanctioning power of the EU. Whereas the legislative acts were agreed by all member states, and the TSCG with the exception of the Czech Republic and the UK, their provisions distinguish clearly between obligations of Eurozone and non-Eurozone countries. Outsiders are exempt from the introduction of a national balanced budget rule and correction mechanism and from the sanctioning mechanism for deficit countries. Conversely, outsiders accepted those rules and procedures that do not go beyond macroeconomic coordination and supranational monitoring and that stop short of sanctions.
Regarding financial supervision, all EU member states participate in the ESFS, a conglomerate of four agencies created in 2010 in response to the financial crisis. The System has focused on and elaborated the so-called 'Single Rulebook' for the EU financial market, a set of prudential standards for financial institutions, in four legal acts that entered into force in 2013 and 2014. By contrast, its supervisory role is limited. The System relies mainly on the coordination and cooperation of national supervisory authorities. Decisions are taken collectively by representatives of national supervisors. Because the ESFS does not envisage supranational centralization, all EU member states were willing to participate. This changed when the regime of financial supervision was upgraded to the European Banking Union.
The legal acts establishing the banking union were again agreed by all EU member states. Whereas Eurozone countries are mandatory full members of the banking union, non-Eurozone countries are not. They may, however, voluntarily participate by establishing 'close cooperation' between their national regulatory authorities and the European Central Bank (ECB). Participation in the Single Resolution Mechanism (SRM) follows from participation in the Single Supervisory Mechanism (SSM). Only member states that subordinate their national banking regulation to the SSM are permitted to join the SRM and benefit from the joint resolution fund. By the same token, SSM countries participate automatically in the SRM.
Bulgaria and Croatia were the first non-euro area member states to enter into close cooperation with the banking union in 2020. For both countries, however, close cooperation is a stepping stone towards full membership in EMU and came at the same time as the entry of their national currencies into the Exchange Rate Mechanism II with the Euroone of the treaty-based convergence criteria for joining the Eurozone.
By contrast, the opt-out countries have been among the strongest opponents of the banking union. Sweden and the UK even refused to sign the SRF agreement. The discussion about the banking union in these countries matches their original Euroscepticism and sovereignty and autonomy concerns that had motivated the original decision to opt out. In the UK, participation in the banking union was ruled out from the beginning. Rather, the British government was most concerned with preventing the SSM and SRM from constraining its own autonomy and dominating the ESFS. For this reason, the British government was at the forefront of introducing safeguards in the European Banking Authority, such as the triple majority requirement for binding decisions, that would prevent a majority of Eurozone countries from imposing decisions on a majority of non-Eurozone countries (Howarth and Quaglia 2014: 115-116).
Whereas there was more discussion on the pros and cons of participation in Sweden, a very important reason for opting-out was that the Swedish government objected to being a second-class member of the banking union by being subjected to ECB decisions it could not influence (Spendzharova and Bayram 2016). In addition, Ferran reports that while the Danish central bank recommended joining the banking union, several Danish parties argued that this 'would be equivalent to participation in the euro area Robert Schuman Centre for Advanced Studies Working Papers and … therefore have to be preceded by a referendum' (Ferran 2014: 15, fn. 31). In a typically pathdependent manner, these reasons for Danish and Swedish reluctance show how a prior decision (not to join the euro area) constrained a subsequent decision (to join the banking union). It is also in line with the path-dependency argument that Denmark, which participates in the European Exchange Rate Mechanism (ERM II)a halfway participation in the Eurozonehas signed the SRF agreement and adopted a wait-and-see rather than a clearly opposing position to the banking union.

Conclusions
Is differentiated integration an endogenous, dynamic process? Do existing differentiations in EU treaties and legislation change the probability of differentiation in subsequent treaties and legislative acts? Finally, does differentiation put European integration on a slippery slope leading to 'ever looser union'? This report attempts to provide answers to these questions based on the analysis of encompassing datasets of differentiations in EU primary (EUDIFF1) and secondary law (EUDIFF2) and a case study of the effects of the Eurozone crisis on differentiation.
The overall answer is a qualified yes. We find, indeed, that existing treaty-based differentiations increase the probability of differentiation in subsequent treaty reforms and legislation in the same policy area. This effect is consistent across model specifications and robust against the inclusion of factors that have proven relevant in the explanation of the initial treaty differentiations, above all, member state wealth, identity, Euroscepticism and policy type (core state power vs. market-related policies). The case study on the Eurozone crisis confirms this result. Both new treaties and new legislation triggered by the crisis reproduce the original spilt between insiders and outsiders. Crisis-induced reform has not only widened the institutional gap in monetary and fiscal policy integration, but it has also spilled over into financial market integration, which had been integrated uniformly in the ESFS.
We add two qualifications to these findings. First, path-dependent differentiation is limited to domain of core state power integration, in particular, EMU, Schengen and the Area of Freedom, Security and Justice (AFSJ). The European integration of these policies started as differentiated integration and is characterized by durable and policy-wide treaty differentiations. Each reform of these policies, and their legislative implementation, is highly likely to reproduce these fundamental divides. The Eurozone crisis case analysed in detail in this report is thus a most likely case for path-dependent differentiation but cannot be generalized to other policies.
Second, path dependence is one factor among many that affect differentiations in EU treaties. Fundamentally, DI reflects increasing and persistent heterogeneity in integration attitudes, preferences and capacities among the member states. Even in the absence of path dependence, this structural divergences would have created additional differentiations in treaty reforms. In legislative differentiation, however, treaty-based differentiation is the dominant driver behind the increase in legislative differentiation that we have witnessed in the area of core state power policies.
In sum, differentiated integration is not a slippery slope. Most differentiation in European integration is temporary and does not create discernible knock-on effects. And in the core state power policies that do exhibit significant path-dependent differentiation, it mainly translates persistent structural differences between the member states into treaty and legislative outcomes.