Hierarchy, polyarchy, and experimentalism in EU banking regulation: the Single Supervisory Mechanism in action

ABSTRACT This paper analyzes the Single Supervisory Mechanism (SSM) from three overlapping perspectives: first, as a centralized hierarchy, seeking to enforce uniform rules, standards, and procedures; second, as a polyarchic network, seeking to orchestrate cooperation between the ECB and national authorities; and finally, as an experimentalist organization, seeking to learn from diversity by adapting common rules and procedures to the specificities of individual banks, and revising them regularly through peer review of implementation experience at multiple levels. Drawing extensively on interviews and primary documents, the paper traces the evolution of the SSM’s institutional structures, decision-making processes, and organizational practices from its origins to the present. On this basis, the paper argues that the most encompassing perspective on the SSM is that of an experimentalist organization, which integrates and recasts key elements of the other two views into a more complete and dynamic analysis of its evolving architecture and practical operations.


Introduction
Nowhere has European integration advanced more rapidly over the past decade than in financial regulation generally and banking supervision in particular. But how should the Single Supervisory Mechanism (SSM) for Banking Union, the most important institutional innovation in this field, be analyzed in governance-theoretic terms, and what are its implications for the EU's broader governance trajectory? Is the SSM best understood as a centralized hierarchy, a polyarchic network, an experimentalist organization, or a distinctive combination of all three forms of governance?
The SSM has been termed 'the most significant integration step since the Maastricht Treaty' and 'a greater pooling of sovereignty than signing up to the Euro' (Mario Draghi quoted in Glöckler, Lindner, andSalines 2017, 1136;Sharon Bowles MEP quoted in McPhilemy 2014, 13). It has likewise been called a decisive step away from the EU's 'main established form of governance, with strong multi-level and heterarchic properties, towards government with a strong degree of centralization and hierarchy', replacing 'the soft governance of the previous period with legally binding and centralized enforcement and the European Commission. A full list of interviews can be found in the Appendix. The interviews are referred to in the text by codes, for which the key is provided in the Appendix. For reasons of space, only the most essential citations to these sources are provided in subsequent sections. Supporting evidence for empirical claims in the main text can be found in Section B of the Appendix. References to this material are indicated in the body of the text by bold numbers in square brackets.
The body of this paper analyzes the SSM from three governance-theoretic perspectives: first, as a centralized hierarchy, seeking to impose and enforce uniform rules, standards, and procedures across the Banking Union; second, as a polyarchic network, seeking to orchestrate intensive cooperation between the ECB and the NCAs; and finally, as an experimentalist organization, seeking to accommodate and learn from diversity by adapting common rules and procedures to the specificities of individual banks, and revising them regularly through peer review of implementation experience at multiple levels. The aim of the analysis is not only to adjudicate empirically between these apparently opposed views of the SSM, but also to show how they can be reconciled with one another in a more comprehensive analytical perspective, and to draw out the implications for EU financial regulation and European integration more generally. To prepare the theoretical ground for the application to the SSM of these three forms of governance, the next section briefly discusses their organizational characteristics, scope conditions, and incidence within the EU.

Hierarchy, polyarchy, and experimentalism in EU governance
Hierarchy is widely considered the standard form of governance in modern states. In the governance literature, hierarchy is often conflated with government, referring to any form of public action by legislative, executive, or judicial authorities aimed at directing the behavior of subordinate actors within their territory (e.g. Börzel 2010, 194-6;Bartolini 2011, 7). But from a theoretical perspective, the scope conditions for hierarchical governance are quite demanding, leading to frequent deviations from a pure principal-agent model. Thus the superordinate principal must have not only the formal authority to enact rules and issue commands to subordinate agents, but also the organizational capacity and coercive power to implement decisions and enforce compliance with them. And even where principals do have the formal authority and coercive power to sanction noncompliance, they often lack the panoramic knowledge necessary to formulate sufficiently precise goals and rules ex ante, together with the detailed monitoring capacity to hold agents accountable for carrying them out ex post. The more diverse and rapidly changing the governance field, the more difficult it is for principals to promulgate and enforce uniform rules through command-and-control methods (Sabel 2004;Kjaer 2004: ch. 4;Zeitlin 2008, 303-5, 2012, 174-5).
Where one or more of these scope conditions do not obtain, the result is forms of governance that depart to varying degrees from a centralized hierarchical model. Among the most widespread of these is street-level bureaucracy, where front-line agents responsible for regulatory enforcement and/or service provision reconcile conflicting rules and misfits with local circumstances by developing ad hoc workarounds and exercising personal judgment about how best to advance the underlying policy goals, often with the tacit toleration of their superiors, but without any explicit mechanism for revising disfunctional procedures and generalizing promising solutions (Lipsky 2010;Sabel 2013). Another prominent deviation from centralized hierarchy is polyarchic network governance, in which officials from lower-level governmental units -often supplemented by business and civil society stakeholders -participate directly in policy formulation, resulting in a closer fit of jointly agreed rules with local preferences and circumstances, which may be expected to smooth implementation and encourage compliance (Kjaer 2004, 34-52;Rittberger and Kohler-Koch 2006;Keast 2022). To ensure that these polyarchic networks reach public-regarding decisions, or remediate the consequences if they do not, principals may invoke the so-called 'shadow of hierarchy', by threatening to use legislative, administrative, or judicial measures to sanction abuses of delegated authority. But this shadow of hierarchy can only work effectively if the superordinate principal has the capacity to withdraw the authority delegated to subordinate agents and take over the relevant governance tasks itself, which is often far from the case, especially in multi-level systems like the EU (Héritier 2002a, 194;Héritier and Lehmkuhl 2008;Börzel 2010, 196-206;Bartolini 2011, 8;Sabel and Zeitlin 2008, 307-9). Where central principals lack such autonomous implementation and enforcement capacities, they often seek instead to 'orchestrate' the cooperation of lower-level actors in pursuit of common goals through some combination of positive and negative inducements (Abbott et al. 2015(Abbott et al. , 2020. Within the EU, there are few if any full-fledged examples of centralized hierarchical governance (pace Börzel 2010, 198-200). Legislative competences in most areas are shared between the EU and the member states, while in many others the EU is confined by the Treaties to a supporting or coordinating role. Even in areas where the EU has been granted exclusive competences, their practical exercise typically involves intensive forms of cooperation with member states. In trade, for example, member states must approve the Commission's negotiating mandate, and together with the Parliament ratify the proposed agreement (European Commission 2018). In competition, since the 2003 Modernization Regulation, the Commission has chosen to share responsibility for implementation and enforcement of EU regulation with a network of national authorities, whom it is obliged to consult on its own decisions (Svetiev 2020). Although the ECB has exclusive powers over monetary policy in the eurozone, and is widely considered an archetypal case of supranational centralization, the presidents of national central banks hold a (rotating) majority of voting rights in its Governing Council, even if they cannot be mandated by their governments (Moschella and Diodati 2020) In cases of persistent non-compliance with EU law, the Commission can bring infringement proceedings against member states in the European courts, whose use has become less frequent and more strategically selective over the past decade (Kelemen and Pavone 2021). But the Commission's limited size and organizational resources make this supranational body heavily dependent on cooperation with national administrations for the implementation of EU policies (Bauer and Trondal 2015: esp. chs. 1-5). Since the 1990s, such cooperation has been increasingly organized through European networks and agencies, in whose governance national regulators continue to play a key role (Thatcher and Coen 2008;Kelemen and Tarrant 2011;Everson, Monda, and Vos 2014;Blauberger and Rittberger 2015).
In some cases, such as the European Supervisory Authorities (ESAs) for banking and capital markets, these agencies can adopt binding standards and non-binding guidelines for the implementation of EU regulation. They may also issue instructions to NCAs and market participants to tackle emergency situations and breaches of EU law, as well as to resolve disputes between their members through binding mediation. But the ESAs' supervisory boards are composed of the heads of the NCAs themselves, who collectively take all decisions by majority vote, while their powers to impose binding decisions on their members are hedged round with conditions and have been scarcely used (Ferran 2012(Ferran , 2016Moloney 2018). At least for now, the ESAs, like other EU agencies, thus resemble more closely a polyarchic network than a centralized supranational hierarchy.
But centralized hierarchies and polyarchic networks do not exhaust contemporary forms of governance. An alternative model, which has gained ground widely over the past two decades within and beyond the EU, is experimentalist governance (XG), based on a recursive process of provisional goal setting and revision through comparative review of implementation experience in different local contexts. This alternative model, which reverses the key assumptions of hierarchical principal-agent governance based on a sharp separation between policy conception and administrative execution, likewise adds a dynamic feedback structure for improvement of central rules through learning from local application to the apparently fluid practices of polyarchic network governance Zeitlin 2008, 2012).
In its classic form, XG involves a multi-level architecture, whose four functional elements are linked in an iterative cycle. First, broad, open-ended goals and metrics for assessing their advancement are established jointly by some combination of 'central' and 'lower-level' actors (in the EU, the European institutions and the member states), in consultation with relevant civil society stakeholders. In regulatory domains, these goals are typically elaborated into framework rules and standards, some of which may be incorporated into legislation and made legally binding. Second, 'lower-level' actors (like national ministries and regulatory authorities) are given substantial discretion to pursue these goals in ways adapted to their local contexts, and to propose changes to the rules themselves in cases of serious misfit. But in return for this autonomy, these units must report regularly on their performance, and participate in a peer review where their results are compared with those of others pursuing different approaches to the same general ends. Where lower-level units are not making good progress toward the agreed goals, they are expected to take corrective measures, informed by the experience of their peers. The goals, rules, metrics, and decision-making procedures are then periodically revised in response to the problems and possibilities revealed by the review process, and the cycle repeats. For a diagrammatic representation of this architecture in the EU, see Figure 1 below.
In many cases, these experimentalist architectures are underpinned by 'penalty defaults': mechanisms that induce reluctant parties to cooperate in framework rule making and respect its outcomes, while stimulating them to propose plausible and superior alternatives, typically by threatening to reduce control over their own fate. In the EU context, such penalty defaults frequently involve court judgments or (threats of) Commission decisions, which oblige member states and/or private actors to explore how to pursue their preferred goals in ways compatible with the fundamental principles of European law, but without hierarchically imposing specific solutions (Sabel and Zeitlin 2008, 305-8, 2010, 13-16, 2012Zeitlin 2016, 3-4;Gerstenberg 2019;Svetiev 2020).
Like hierarchical governance, XG in this form also depends on a number of scope conditions. The first and most important is strategic uncertainty, where policy makers cannot define their precise goals or how best to achieve them ex ante, but must instead discover both in the course of problem solving, because they are operating in a turbulent, rapidly changing environment. A second is a polyarchic or multi-polar distribution of power, in which no single dominant actor can impose their own preferred solution without taking into account the views of others. A third is a high level of diversity, which increases the difficulty of adopting and enforcing uniform rules. A final scope condition concerns interdependence, which must be sufficient to motivate actors to collaborate in seeking joint solutions to common problems, but not so great as to preclude decentralized experimentation by local units, as in hierarchical governance, where high externalities are often adduced as a key motivation for centralized decision making and coercive enforcement (Sabel and Zeitlin 2012, 174-5;Rangoni and Zeitlin 2021, 823-4;Héritier 2002b).
Where these scope conditions are met, XG architectures have four fundamental advantages, relative both to centralized hierarchies and polyarchic networks, which help to explain their proliferation in contemporary governance. First, they accommodate diversity by adapting shared goals and rules to varied local contexts, rather than seeking to impose one-size-fits-all solutions. Second, they provide a mechanism for coordinated learning from local experimentation through disciplined comparison of different approaches to common overarching objectives. Third, the same processes of mutual monitoring, peer review, and joint evaluation that support learning from diverse experience also provide dynamic, non-hierarchical mechanisms for holding both central and lower-level actors accountable for their actions in pursuit of agreed goals. Finally, because both the goals themselves and the means for achieving them are explicitly conceived as Figure 1. EU XG as an iterative, multi-level architecture. Source: Zeitlin (2015, 2) provisional and subject to revision in light of experience, problems identified in one phase of implementation can be corrected in the next iteration.
Although XG architectures of this type are neither universal nor ubiquitous in the EU, they are widely diffused across a variety of policy domains. Well-documented examples include: regulation of competition, energy, telecommunications, and finance; food, drug, chemicals, and maritime safety; environmental protection; employment promotion and social inclusion; justice and home affairs; data privacy, anti-discrimination, and fundamental rights Zeitlin 2008, 2010). These architectures also play a growing part in EU external governance, where the revisable framework rules they generate are frequently extended to third-country actors (Zeitlin 2015). A typical pattern in recent years has been progressive formalization of EU regulatory networks, without full supranational centralization, as in the case of the ESAs. In some sectors, under conditions of high interdependence coupled with high uncertainty, concern for the integrity of integrated markets has led to the creation of a single set of harmonized but provisional rules, revisable through ongoing monitoring and review of implementation experience, as for example in chemicals and electricity regulation. These developments in turn raise the possibility, which I explore further in this paper, of the emergence in such cases of a simplified XG architecture, combining synchronic uniformity with diachronic revisability (Zeitlin 2016;Rangoni 2020;Rangoni and Zeitlin 2021;Zeitlin and Rangoni 2023).

The SSM as a centralized hierarchy
There is no shortage of evidence to support the view of the SSM as a centralized hierarchy. Foremost is its authority to grant and withdraw banking licenses: the power of life and death over credit institutions in participating member states. The SSM directly supervises the largest and most systemically important banks in the eurozone (currently 111), accounting for over 80% of banking assets. The SSM likewise has full authority over the designation of banks as SIs, along with far-reaching powers over other crucial matters such as the appointment and removal of bank directors, passporting, acquisition of qualifying holdings, own funds determinations, and adoption or modification of internal models. To streamline the process, routine decisions on many of these issues have been formally delegated by the SSM Supervisory Board (SB) to the heads of ECB work units. As part of the annual Supervisory Review and Evaluation Process (SREP) decision, the SSM can require SIs to hold additional capital to cover specific risks, as well as to revise their governance arrangements, planning processes, controls, and other internal systems (ECB 2018a: chs. 3-4;Petit 2019, 117-22;D'Ambrosio 2020, 39-60, 193-8).
The ECB can also take over supervision from NCAs of the remaining 2500 or so less significant institutions (LSIs) where it deems this necessary to 'ensure consistent application of high supervisory standards', and can demand any information it requires from these institutions via their supervisors. The ECB is authorized to exercise powers granted to national supervisors under both EU and national law over SIs, and to instruct NCAs on their use towards LSIs. Unlike other fields of EU multi-level regulation, the European Courts have determined that powers over banking supervision within the SSM have been delegated exclusively to the ECB rather than distributed between the ECB and the NCAs, which are responsible for 'decentralized implementation' under the former's control, including as regards supervision of LSIs (ECB 2018a: ch. 5;Smits 2017Smits , 2019Petit 2019, 123-8;D'Ambrosio 2020, 203-30;Schiammo 2021). 5 To carry out these tasks, ECB Banking Supervision has built up a substantial central organization directly employing some 1200 staff (ECB 2018a: ch. 5; 2021a: 78; https:// www.bankingsupervision.europa.eu/about/thessm/html/index.en.html). Compared to other EU agencies, the SSM is much more involved in executive and enforcement tasks in the member states. From the outset, it has been committed to developing 'intrusive and hands on supervision', aimed at implementing the single rulebook 'diligently and assertively' by ask[ing] the hard questions and challeng[ing] the responses where necessary" (2015, 5; 2018a: 4). At the heart of this process are the Joint Supervisory Teams (JSTs), which carry out the annual SREP exercise for each SI according to a standardized methodology assessing their business models, governance, risk management, capital, liquidity, and funding. The JSTs combine ECB staff and local supervisors from the NCAs in which the bank operates. The JST Coordinator, employed by the ECB, is responsible for proposing the SREP decision, based on standardized risk scores, benchmarked against those of other similar banks by the ECB's horizontal services, which can be adjusted upwards or downwards according to 'constrained supervisory judgment' within a limited bandwith (ECB 2018a(ECB , 11-14, 77-89, 2018b. A key purpose of this decision-making procedure is to reduce national bias and regulatory capture [1]. The JSTs are supported by an extensive program of on-site inspection missions, which conduct detailed investigations on the bank's premises of specific issues ranging from residential real estate or shipping loans to governance and operational risks, based on a targeted engagement level keyed to the size and complexity of each institution (ECB 2018c; ECB 1; NCA 1; NCA 3; NCA 11.4). To reduce information asymmetries further, the SSM has carried out a multi-year on-site comparative investigation of banks' internal models, analyzing not only their technical content but also their management processes for validation, monitoring, review, and revision (ECB 2021b) [2].
From the outset, the SSM has sought to increase harmonization and convergence of supervisory approaches across participating units. Among its first products was an internal 'Supervisory Manual' describing 'common processes, procedures and methodologies' for the ECB and the NCAs, aimed at 'ensuring that the same supervisory standards are applied across the banking union . . . complementary to the EBA's work in fostering harmonization within the EU' (ECB 2015(ECB , 33-34, 2018a [3]. The ECB has likewise sought to develop 'joint supervisory standards' to steer and harmonize NCA supervision of LSIs, including a modified version of the SREP methodology, which is being progressively rolled out across NCAs (ECB 2018a, 67-8;2020b). It also publishes detailed guidance documents and recommendations setting out its supervisory expectations for banks on a variety of issues, including non-performing loans, internal models, and internal capital and liquidity assessment processes, and dividend distributions during the Covid-19 pandemic. 6 The SSM has consistently sought to enhance the uniformity of the Single Banking Rulebook and harmonize its implementation at national level. Among its first major projects was a review of the numerous options and discretions (O&Ds) provided to competent authorities under EU legislation, aimed at reducing variations across the eurozone which were believed to distort competition and fragment the banking market. As a result, the ECB issued a detailed regulation and guide setting out which of the more than 120 O&Ds granted originally to the NCAs it would continue to apply under what conditions to SIs under its direct supervision. It then went on to publish a similar set of instruments requesting the NCAs to adapt these harmonized rules for the use of O&Ds in the supervision of LSIs (ECB 2015(ECB , 7, 2016a(ECB , 62-67, 2016b; Guideline (EU) 2017/697 of the ECB; Kudrna and Puntscher Riekmann 2018) [4].
Beyond these O&Ds, many supervisory rules and powers remain unharmonized across the Banking Union due to variations in the underlying national legislation. A notable case in point are the 'fit and proper' requirements for appointing and removing bank directors, which involve ex ante supervisory approval in some countries but only ex post assessment in others. Here, the SSM has published a guide aimed at establishing a common 'Europewide process' for such assessments, requesting banks to give prior notice of proposed appointments and enabling the ECB to provide advance feedback about any potential issues before the appointment takes effect (Lo Schiavo 2019b;ECB 2022, 56-57). SSM leaders have likewise highlighted overlaps and contradictions in the legal frameworks for crisis management, resolution, insolvency, and liquidation, pressing for revised legislation to harmonize the transposition of EU rules into national law and enhance centralized European coordination of bank liquidations (Nouy 2018b(Nouy , 2018cECB , 7, 2020aECB , 6-7, 2021aEnria 2020aEnria , 2021).

The SSM as a polyarchic network
But alongside these centralized hierarchical elements, the SSM also displays significant heterarchical features that support an alternative view of it as a polyarchic network. Foremost among these is the structure of decision making. All major decisions of the SSM must be approved by the Supervisory Board, where NCA representatives account for the overwhelming majority of votes (currently 21 out of 27) [5]. Formally, SB decisions must also pass via a non-objection procedure through the ECB Governing Council (GC), on which national central banks also have a structural majority, with a mediation panel to address possible concerns raised by NCAs. But so far, the GC has not intervened in SSM decisions and the mediation panel has never been activated, about which 'the SSM is quite proud' (Petit 2019, 112-15;NCA 9.3).
Although NCA representatives are legally obliged to perform their duties 'in the interest of the Union as a whole', they also keep a watchful eye on national interests and defend member state perspectives in the SB's deliberations. All NCAs interviewed had established SSM coordination units of varying sizes to prepare the work of their SB representative and review draft decisions proposed through the written approval procedure, which account for the majority of the SSM's decisions (ECB 2015(ECB , 11-12, 2021a [6]. At the outset, NCAs scrutinized all draft decisions for approval by written procedure closely, to safeguard national preferences and avoid undesirable precedents. While the intensity of such scrutiny has diminished, NCAs still carefully review and comment on these draft written decisions; where three or more object, the decision must be placed on the agenda of the SB. In practice, when NCAs submit comments through these written procedures, these are reviewed by the responsible ECB units and as far as possible incorporated in the final version. Because of the workload involved in reviewing these draft decisions, many on routine issues, NCAs have welcomed their delegation to ECB work units. But negative decisions cannot be taken under the delegation framework, and must be submitted for approval to the SB. All controversial decisions, irrespective of the procedure under which they are taken, thus come before the SB, where they may be challenged and debated by NCA representatives (ECB 6; ECB 11.1; NCA 4.2; NCA 7; NCA 9.1; NCA 11.1; European Commission 2017a: 6, 2017b: 20-21; Petit 2019, 115-23).
These governance arrangements in turn mean that all major SSM policies and decisions must be agreed between the ECB and a majority of NCAs [7]. To channel input from the NCAs, the SSM has established a dense web of networks bringing together ECB staff and national supervisors. Each ECB horizontal division has its own joint expert network with the NCAs to discuss and develop plans, policies, and procedures (ECB 1; ECB 2.1; ECB 8; ECB 10.1; ECB 11.1; ECB 12.1; NCA 4.1; NCA 5.1; NCA 7; NCA 9.2; NCA 11.3; NCA 11.4; Gren 2018, 295-7). All major SSM initiatives and documents, including preparation and revision of manuals and operational guides, are developed through joint working groups, task forces, and drafting teams convened by these networks, many of which are also led by NCA representatives (Gren 2018, 298-9). The purpose of these arrangements is to develop political buy-in for common positions as well as to tap into the detailed knowledge and expertise of the NCAs on complex technical issues [8]. NCAs, for their part, consider participation in these networks and joint drafting teams crucial to ensure that their perspectives are taken into account before any proposals go to the SB [9].
In the case of LSIs, the impetus towards cultivation of cooperative relations with the NCAs is even stronger. In the early years of the SSM, as one top official explained, 'certain stakeholders expected that the ECB could maybe intervene much, much more in some LSIs, the largest ones essentially', and 'there was a whole methodology developed for these high-priority LSIs, with . . . larger involvements.' But over time, the ECB's oversight approach has evolved away from such efforts to implement a sort of 'direct-indirect' supervision of individual LSIs towards 'a more system perspective' focused on the codevelopment of Joint Supervisory Standards with NCAs [10]. While the ECB is legally empowered to issue general instructions to NCAs on LSI supervision, they now find it more effective to rely more on informal guidance and persuasion than on binding instruments, not only because it creates less 'stress in the system', but also because the latter are slower and more difficult to change in response to implementation problems and new developments (ECB 10.1; ECB 10.2; NCA 8). To date, moreover, the ECB has never exercised its power to take over direct supervision of an LSI from an NCA against the latter's objections, which the SB Chair termed 'an exceptional response -a measure of last resort which should be considered only when all other appropriate supervisory measures have been unsuccessful' (Petit 2019, 125, n. 113;Gren 2020, 251).
But even for SIs, there are sharp constraints on the ECB's hierarchical powers. Thus, the ECB does not directly employ or control NCA supervisors participating in SI oversight. The numerical disparity is greatest in the case of on-site missions, where only 70 of the c. 1000 supervisors involved are employed by the ECB, and the head of mission typically works for the host NCA [11]. The balance is less skewed for the JSTs, where ECB staff comprised 37% of the 1318 FTEs involved in 2019 (ECB 2020a, 91). Here, too, however, the JST coordinator has limited direct authority over local supervisors, who do not necessarily work full-time for the team, and whose career prospects remain in the hands of their NCA employers [12]. While a protocol has been agreed whereby the JST coordinator can submit an annual performance evaluation for team members employed by the NCAs, their use remains up to the latter (Decision (EU) 2019/976 of the ECB; (Gren 2018, 273).
Hence each JST has a local coordinator responsible for reconciling tensions in work assignments and priorities between NCA managers and the ECB team coordinator. Although the ECB coordinator is formally empowered to propose the SREP decision, any dissenting opinions are also reported, and controversial cases must be resolved by the SB. Larger teams therefore have a 'core JST' of ECB and local coordinators, who interact frequently 'to discuss the most important topics' and hammer out the proposed SREP decision (ECB 3; NCA 6; NCA 4.1; NCA 4.2; NCA 9.2; NCA 11.4). During the SSM's early years, the degree of friction within JSTs often depended on the background and personality of the ECB coordinator, as well as on NCAs' willingness to embrace the new supervisory arrangements. But interviewees from both sides agree that these relationships have become much smoother and more cooperative. Crucial in this respect has been the mandatory rotation of coordinators and local team members every three to five years, which interviewees argue has fostered the circulation of ideas and experiences across JSTs, sanding off rough edges in relationships between ECB and NCA officials, while encouraging the diffusion of a common approach to decision making. The ECB itself has deliberately sought to stimulate this transnational socialization process through systematic secondments and exchanges of staff at multiple levels, along with a variety of training and experience-sharing forums (ECB 3; NCA 6; NCA 3; NCA 4.1; NCA 4.2; NCA 9.2; NCA 11.4; ECB 2022, 74-75).
Beyond the SSM, NCAs retain an independent voice on EU banking regulation through the EBA, with double majority voting arrangements to safeguard the interests of non-Banking Union Member States. The ECB participates in meetings of the EBA Board of Supervisors (BoS), but does not have voting rights there, unlike the NCAs (Ferran 2016). While the ECB seeks to coordinate with the NCAs before each BoS meeting, in order to develop a common position on matters of particular interest to the SSM, it cannot compel the NCAs to follow their lead [13]. More generally, SSM NCAs value the EBA as 'a different channel to communicate your stance, with full independence', especially on issues concerning smaller banks, and regularly take different positions from one another as well as the ECB within the BoS (NCA 9.2; NCA 5.1; NCA 5.2; NCA 2; EBA 1.1; COM 1) [14].
The EBA is responsible for drafting binding regulatory standards and 'soft law' guidelines on interpretation of EU legislation, while the SSM, like NCAs outside the Banking Union, is responsible for their implementation in supervising individual credit institutions. From this perspective, the SSM is itself expected to follow EBA supervisory guidelines, including on core activities such as the conduct of the SREP [15]. To influence EU regulation and ensure a productive division of labor between the two bodies, the ECB therefore cooperates closely with the EBA on key policy issues, such as the recommendations on suspension of bank dividend payouts during the Covid-19 pandemic [16].
At a higher level, the Council and Parliament retain the final word over revisions to the legislative framework of the EU's Single Financial Rulebook (level 1 rules), including the elimination of the remaining O&Ds vested with member states and other revisions to the crisis management, resolution, and liquidation regulations advocated by the SSM leadership [17]. The Commission and the EBA are likewise responsible for adopting and revising binding technical standards (level 2 rules) and soft law guidelines (level 3 standards), which the ECB, like the NCAs, is expected to implement, subject to the 'comply or explain' procedure in the latter case.

The SSM as an experimentalist organization
The polyarchic governance structure of the SSM thus strongly encourages the pursuit of a cooperative rather than a hierarchical approach by the ECB to joint supervision of eurozone banks with the NCAs. But how far can it be considered an experimentalist organization? As previous sections have shown, the SSM clearly diverges from the classic XG architecture outlined earlier in a number of significant respects. Rather than setting open-ended framework goals and giving lower-level actors substantial autonomy to pursue them in ways adapted to their local circumstances, the SSM has developed increasingly detailed and prescriptive rules and methods, which banking supervisors are expected to apply as consistently as possible across credit institutions and jurisdictions. Such supervisory convergence is considered crucial to advance the SSM's mission and strategic aims of 'contributing to the safety and soundness of credit institutions and the stability of the financial system' while 'promoting European financial integration', by reducing opportunities for regulatory arbitrage, removing national barriers to crossborder operations, and ensuring a level playing field and equal treatment of all eurozone banks (ECB 2018a, 4-6).
Within these limits, however, a closer look at the SSM 'in action' reveals the centrality to its organization of experimentalist practices of learning from diversity, peer review, and continuous revision of rules, methods, and procedures based on comparative evaluation of their implementation in different local contexts. The widespread adoption of such experimentalist practices, as this section will demonstrate, flows directly from the SSM's deliberate efforts to adapt its rules, methods, and procedures to eurozone banks' diverse business models and risk profiles on the one hand, and to continuously update and revise them in response to rapidly changing markets and technologies on the other.
Thus, despite the SSM's emphasis on regulatory harmonization and supervisory convergence, it does not seek to homogenize banks' business models nor impose a single 'one-size-fits-all' approach to their supervision across the eurozone [18]. The SSM supervisory model was explicitly designed to combine the 'deep specific knowledge of national supervisors with the broad-ranging experience of the ECB' through the JSTs and joint expert networks, while leveraging 'increased opportunities for benchmarking and peer comparison . . . to improve the tools of supervisory risk assessment, with due regard to the diversity of banks' business models' across the euro area (ECB 2015, 5). The core idea is thus to treat similar institutions similarly and different institutions differently across the Banking Union, irrespective of national origin, using 'a common methodology to provide a level playing field' for assessing each bank, while 'tailor [ing] . . . supervisory expectations to its specific situation' (Enria 2019a) [19].
The design of the SSM supervisory model was itself the outcome of an intensive process of joint deliberation and comparison of national practices by mixed teams of ECB and national officials (Gren 2018, 294-5) [20]. Beyond the JSTs, which were an entirely novel construction, among the most important design decisions taken in this phase was the generalization of extended on-site investigations, conducted on behalf of the JSTs by mainly local supervisors, but coordinated by the ECB. The SSM's on-site inspection (OSI) function was based on the practices of a few national authorities, notably the French, the Italian, and the German, 'with some minor variations in order to try to take the best out of each of those models', as one of its architects observed. But OSI was largely new to many NCAs, like the Dutch or the Irish, and differed significantly from that of others, like the Spanish, who as another ECB official remarked, 'did offsite supervision onsite', processing reporting that they were getting from the banks on the latter's premises (ECB 1; ECB 3; NCA 9.2; NCA 9.3; NCA 11.4; NCA 1; NCA 3; NCA 6). The SSM's comparative, numerical approach to supervision, focused on calculating specific ratios, benchmarking banks against peer groups, and identifying outliers, likewise drew heavily on the practices of certain NCAs 'who worked to a very large extent quantitatively beforehand', like the Spanish, French, and Italians, while representing a bigger change for others like the Germans and the Dutch, whose supervisory culture was more qualitative and principlesbased (NCA 6; ECB 6; COM 1; BaFin 2016: 22, 89-90; Del Barrio Arleo C 2020: ch. 3) [21]. The development of the JSTs themselves has similarly involved an intensive process of crossfertilization and mutual learning between supervisors from different national systems, who look 'with different eyes and different perspectives' at each other's entrenched practices, such as Dutch mortgage lending or German reliance on external auditors (Das 2014, 39-40) [22].
To foster this multi-perspectival approach to bank supervision, the SSM systematically combines multiple forms of comparison within and between institutions both nationally and cross-nationally. Thus, in preparing the SREP decision for each bank, JSTs draw not only on the reports of on-site investigations into key issues, but also on the work of specialized risk teams, often led by NCA experts, whose purpose is to 'connect the dots' on each major type of risk across the banking group as a whole. These experts from the different JSTs in turn participate in knowledge-sharing groups on specific topics such as liquidity or credit risk coordinated by the vertical DGs to which they belong, while interacting with the horizontal line services of the ECB on methodological issues. Each SSM Member State also has a Country Coordination Group, 'where all the JSTs . . . in that member state discuss, exchange views on common topics, which are relevant for these SIs.' In the French case, as one national official explained, topics discussed in these forums include specificities of the national banking market, such as regulated savings and mortgage guarantee schemes, which have 'helped other supervisors and the ECB colleagues to understand what is the real risk related to these instruments' (ECB 3;NCA 11.4). This distinctive combination of vertical and horizontal analysis, including benchmarking conducted by the ECB, in turn plays a crucial role in resolving disagreements about the SREP decisions between JST coordinators and NCAs [23].
Such horizontal comparisons, experience sharing, and peer review are even more important in on-site inspections, where the proportion of ECB staff, as we have seen, is much lower than in the off-site work of the JSTs. Thus, in addition to centralized quality assurance reviews of each report, the ECB has deliberately sought to promote 'mixed teams' and 'cross-border missions', comprising supervisors from multiple NCAs, which currently account for some 30% of the 160 or so on-site investigations conducted each year. Alongside regular experience-sharing workshops on specific topics, the ECB onsite function also organizes coordinated 'campaigns', where the same type of investigation is conducted in different banks more or less simultaneously, on topics such as residential real estate, commercial real estate, leveraged finance, market risk, or IT risk, 'in order to enable heads of missions to be able to share experiences more or less in real time with other colleagues . . . in order to make sure that the outcome will be as homogeneous as possible regardless of the fact that we are dealing with different countries and banks' (ECB 1; NCA 1). Behind all this coordinated experience-sharing is an explicit recognition that the only way to ensure consistent outcomes across the SSM is for peers to compare supervisory techniques and clarify through discussion the reasons for differences in their application to individual cases [24].
From the outset, the SSM set out to engage in 'forward-looking' supervision, aimed at identifying and addressing prudential risks and potential threats to financial stability in a timely manner, rather than 'looking backwards towards audited accounts', as had been the dominant practice in some Member States (ECB 2015: 5, 56, ECB 2018ainterview with two DNB officials, 27.5.14, quoted in Das 2014: Appendix 4). Hence the SSM Supervisory Manual was deliberately conceived as a 'living document', 'subject to continuous review and improvements on the basis of internal evaluations, internationally accepted benchmarks and international regulatory developments' and regularly 'updated to reflect new market developments and supervisory practices' (ECB Banking Supervision 2015;DNB interviews in Das 2014: Appendix 4;Gren 2018, 299-301). The Manual was comprehensively revised in the spring of 2015, following field tests covering more than half the significant banks in the eurozone, aimed at detecting and correcting problems in the SREP methodology, promoting 'learning by testing' by the JSTs, identifying further best practices and 'focus areas for the continuous methodology improvement process' (ECB 2015, 55-7) [25].
The SREP methodology is now updated on an annual cycle, based on inputs from joint NCA-ECB drafting teams, 'which allows us to make sure that all the evolution and changes in the regulation can be included in the process' [26] Unlike in street-level bureaucracies, where workarounds to dysfunctional rules typically remain clandestine, frontline supervisors can and do challenge specific procedures for the SREP assessment prescribed by the Manual, on the grounds that they do not fit the bank in question, leading to a discussion within the core JST, which may decide not to apply it. In many cases, the issue will then be raised in the divisional network, resulting in a possible revision of the Manual, e.g. to take account of national differences in bank board structures [27]. This recursive approach to revision and updating of methodologies based on frontline experience with their application in different local contexts is equally pronounced in on-site supervision, where the COI network convenes five to ten joint drafting teams of ECB and NCA officials per year to update the methodologies on specific risks, so that each component of the 1500-page on-site inspection guide will be updated at least every twothree years [28].
Beyond these intensive forms of peer review and recursive revision developed within the SSM's frontline units and horizontal services, the ECB has created an innovative Supervisory Quality Assurance (SQA) function as a 'second line of defense'. This SQA function conducts regular 'advisory reviews' addressing thematic issues across JSTs and ECB business areas, such as the implementation of the SREP and on-site inspections. It also conducts 'assurance reviews' to check the execution of tasks by the JSTs and NCAs against the manuals and regulations, including the 'common procedures' delegated to the national authorities such as preparation of fit and proper assessments, in which the latter are encouraged to participate (ECB 2018a, 47-8;ECB 2.1;ECB 2.2;ECB 2.3). SQA reviews combine horizontal data analysis, benchmarking, surveys, and interviews (with JSTs, OSI missions, ECB business areas, and NCAs), based on samples stratified according to criteria relevant to the specific investigation. The emphasis in these reviews is on identifying blind spots in the work of the frontline units, pushing them to justify why specific areas may have been neglected or given lower priority in SREPs and OSIs of individual banks. Their aim is thus to contribute both to improvements in the supervision of individual banks and to systematic improvements in the SSM's horizontal services (ECB 2.1;ECB 2.2;ECB 2.3) [29].
SQA reviews focus not only on identifying problems, but also on helping to develop solutions, in contrast to a classic internal audit. Reports on JSTs are anonymized, and the identity of the interviewees remains confidential, even to their immediate supervisors, in order to ensure the full cooperation of the frontline supervisors and the candor of the information provided. Reviews conclude with 'proposals for improvement' (rather than recommendations or 'corrective action requests' typical of internal audits), which are consensually agreed with the ECB business areas, based on 'a convincing effort' rather than 'an exercise of power'. SQA has become increasingly involved in helping these units to implement its improvement proposals through customized training activities, such as for instructing JSTs on how to do a good challenge to a bank's recovery plan. Its activities are thus explicitly forward-looking, aimed at contributing to continuous improvement in the SSM's operations and effectiveness (ECB 2.1;ECB 2.2;ECB 2.3) [30].
SQA activities can be understood as an experimentalist form of dynamic, forwardlooking accountability, in which officials are expected to explain how they have used their discretion within a given framework to advance the institution's goals, subject to review by peers knowledgeable enough to challenge such explanations. Where problems or performance failures are revealed by such reviews, officials are expected to explain what revisions in their internal policies and practices they are adopting to improve the situation, and to monitor the implementation and effectiveness of such measures -for which they can then be held accountable in the future (Zeitlin and Brito Bastos 2020).
Within a few years of its formation, the SSM had thus established an intensely recursive process for developing and revising policies, regulations, methodologies, and standards based on continuous review of frontline implementation, which as depicted in its Supervisory Manual closely resembles the classic XG cycle [31]. The recent reorganization of ECB Banking Supervision, carried out in the autumn of 2020, can itself be seen as a higher-level instance of such experimentalist recursive revision, aimed at by breaking down organizational silos, fostering closer cooperation and knowledge sharing between frontline and horizontal supervision, and clustering vertical oversight around specific banking business models [32].

Conclusions
This paper has analyzed the SSM from three distinct perspectives: as a centralized hierarchy, seeking to impose and enforce uniform rules and procedures across the Banking Union; as a polyarchic network, seeking to orchestrate intensive cooperation between the ECB and the NCAs; and as an experimentalist organization, seeking to accommodate and learn from diversity by adapting common rules and procedures to the specificities of individual banks, and revising them regularly through peer review of implementation experience at multiple levels. Each of these views, as we have seen, highlights important characteristics of the SSM, which need to be incorporated into any comprehensive analysis. But each successive perspective also illuminates critical features obscured by its predecessor, leading to a richer understanding of the SSM as a novel institution for European banking supervision. The most encompassing perspective, however, is that of the SSM as an experimentalist organization, which integrates and recasts key elements of the other two views into a more complete and dynamic analysis of its evolving architecture and practical operations.
The SSM was designed as a much more centralized institution than its predecessors, including the ESAs, while the ECB has been given far-reaching hierarchical authority over eurozone banks and NCAs by the Council and the Courts. But as a deeper investigation of the SSM's governance structure reveals, all major policies and decisions must effectively be agreed by the NCAs, while the supervision of individual banks depends in large measure on tasks carried out by national officials over whom the ECB has no direct hierarchical control. Even the most formally impressive powers granted to the ECB, such as the right to take over direct supervision of LSIs or to issue binding instructions to NCAs turn out on closer inspection to operate more as a form of experimentalist penalty default, a last-resort mechanism for inducing reluctant parties to collaborate in joint activities by threatening to impose consequences undesirable for both sides, than as an effective instrument of hierarchical control.
The SSM's polyarchic governance structure and institutional design mean that the ECB and the NCAs are ineluctably 'condemned to cooperate' with one another, a shared fate embodied in the dense web of expert networks, working groups, and drafting teams created to co-develop policies, procedures, and methodologies, as well as in the JSTs and on-site inspection missions themselves. But the ECB and the NCAs created these elaborate joint structures for feeding local supervisory knowledge into the development and application of common methodologies and procedures not merely because they felt obliged to do so politically, but also because they considered them functionally essential for tackling the diversity of business models and national conditions across the Banking Union, while adapting to rapid changes in financial markets, technologies, and lending practices. To support a forward-looking approach to supervision under conditions of high uncertainty, the SSM has accordingly established a remarkable array of experimentalist processes for recursive revision of policies, methodologies, and procedures through continuous peer review and benchmarking of implementation experience at multiple levels. Six years after its formation, moreover, the SSM applied these same experimentalist principles of recursive review and revision to its own organizational structure, in order to deepen collaboration between the JSTs, the NCAs, and the ECB's horizontal services, while enhancing its capacities to tailor supervision to banks' individual business models.
A crucial feature of the SSM, which may at first glance appear to conflict with this experimentalist perspective, is its commitment to the development of uniform rules, methodologies, and procedures, which supervisors are expected to apply as consistently as possible across banks and jurisdictions. Such uniformity and consistency, its leaders firmly believe, are crucial to advance the SSM's overarching goals of financial stability and market integration by reducing opportunities for regulatory arbitrage, removing barriers to cross-border operations, and ensuring equal treatment for all credit institutions within the Banking Union. The SSM does not seek to impose a 'onesize-fits-all' approach on eurozone banks, but rather to calibrate its supervision ever more finely to the latter's diverse business models and risk profiles, treating similar institutions similarly and different institutions differently, irrespective of national origin.
The design of the SSM's supervisory model, along with its core methodologies and procedures, was itself the result of an intensive process of comparison and 'learning from difference' by mixed teams of ECB and NCA officials from a wide range of national and professional backgrounds. The outputs of these methodologies and procedures, including the SREP assessments, are subject to intensive peer review and comparative benchmarking at multiple levels, aimed at clarifying the reasons for disagreements about their application to individual cases, and identifying blind spots, misfits, and possibilities for improvement, which should be addressed in subsequent iterations. In this process, frontline supervisors can and do challenge their applicability in particular cases, and feed in proposals for changes to manuals and operational guides through joint ECB-NCA networks, working groups, and drafting teams. The SSM's rules, methodologies and procedures, together with its broader policies and organization, are thus regularly updated and revised on the basis of learning from comparative review of frontline experience with their implementation in different local contexts, as in the classic XG architecture.
The major difference between the SSM and the classic XG architecture is its distinctive combination of synchronic uniformity with diachronic revisability, whereby rules and procedures are progressively specified, narrowing discretion for lower-level actors at any given moment, but remain fully contestable in light of local application, and subject to recursive revision in light of comparative implementation review. In this respect, the SSM represents an advanced case of the broader trend within the EU discussed earlier towards the emergence, under conditions of high interdependence coupled with high uncertainty, of simplified XG architectures based on a single set of harmonized but provisional rules, revisable through ongoing monitoring and review of local implementation. In such simplified XG architectures, framework rules and procedures may be progressively specified and discretion for lower-level actors at any given moment narrowed, but the rules and procedures themselves remain contestable in light of local application, while revisions over time based on learning from comparative review of implementation experience provide a crucial source of improvement and adaptability for the governance system as a whole. Such architectures have previously been identified in sectors like chemicals and electricity (Biedenkopf 2015;Rangoni 2020). Simplified XG architectures of this type may also become increasingly prevalent in other sectors of EU regulation subject to rapid and unpredictable changes in markets and technology, where concerns to promote a level playing field and prevent regulatory arbitrage are similarly strong, such as competition and telecommunications (Svetiev 2020;Mathieu and Rangoni 2019;Zeitlin and Rangoni 2023).
The case of the SSM supports the view that under such conditions, harmonized rules and supervisory practices can be accepted as effective and legitimate by Member States, provided that they are applied in contextually sensitive ways and revised regularly on the basis of local implementation experience, through deliberative procedures in which frontline officials themselves participate. In this sense, this case further suggests that far from being uniformity and experimentalism being antithetical to one another, diachronic experimentalism may actually be a necessary condition for synchronic uniformity of regulation within a diverse polity like the EU (Zeitlin and Rangoni 2023).