Introduction

In the last decade, five Eurozone governments received financial assistance from the International Monetary Fund (IMF) and the European Union (EU), on the condition that a long list of policies were implemented. Despite the widespread media coverage, leaks and memoirs, there is actually very little empirical research about how domestic governments have dealt with conditionality.

In our last book, co-authored by Catherine Moury, Stella Ladi, Daniel Cardoso and Angie Gago (Moury et al. 2021), we systematically analysed the room for manoeuvre of bailed out governments, to what extent and how MoU’s were implemented, and how resilient those reforms had been for the five Eurozone bail-out countries. To do so, the book relies on qualitative process-tracing and on a database that we present in this explanatory note. This note is organized as follows. First, we briefly present the main findings of the book. We then turn to presenting, in detail, the database and some key quantitative findings deriving from it. The last section discusses the advantages and limitations of the database, and how our methodology and data could be relevant for a comparison with other conditionality mechanisms, such as the Recovery and Resilience Fund.

Room for manoeuvre of bailed out governments

Our book covers the period during and beyond the crisis (2007–2020). The case selection includes the Eurozone’s five bailed out countries (Greece, Ireland, Portugal, Spain and Cyprus). Even though in Spain the conditions only concerned the financial sector, the Commission (mandated by the Council) started a programme of strict monitoring of fiscal consolidation measures and structural reforms in 2011. It explicitly linked its recommendations to a change in the disbursement of the loan installments.

In our book, we opted for a mixed-methods approach. On the one hand, we qualitatively explored in each country the constraints of conditionality, the use of these constraints, and the motivation and willingness to reverse what was done during the programme. Validity was ensured by triangulating different types of data, including 129 face-to-face interviews. On the other hand, as we explain in detail below, we collected quantitative data on the fulfilment of the Memoranda of Understanding, to constitute the Data Base on EcoNomic Adjustment Policies (ENAP) database.

The main finding of our analysis is that Eurozone bail-outs were not a diktat, but rather a two-sided process in which executives negotiated with international lenders. Our book shows that governments were constrained during the financial crisis, by the conditionality associated with the loans, but also before the bail-out by the European Commission, their peers in the Council, by the increasing divergence of yields between Eurozone countries, and by the European Central Bank (ECB).

Yet, the book also shows that bailed out governments kept leverage. This is the case, first, because the lenders—in particular the more experienced IMF—pay high value to ‘ownership’, i.e. a country's interest ‘in pursuing reforms independently of any incentives provided by multilateral lenders’ (Drazen 2002, p 37). Programme ownership, indeed, raises the probability of the programme’s success and, hence, avoids a waste of resources (Drazen 2002; Khan and Sharma 2001, p. 28).

Second, domestic leverage emerges from the fact that international lenders need insider knowledge to design efficient plans. As one IMF official explained:

‘It’s not like the Troika comes with a list of things and then we get the money. It doesn’t make sense partly because only the government knows how things work’ (quoted by Moury and Freire 2013, p. 46).

Indeed, informants from all sides explained to us that negotiators from the Troika were open to the solutions proposed by governments, as long as they reached the general objectives. Governments felt they could take the initiative to insert specific policies in the MoU or its revisions. Interviewees from all countries further reported instances of policies that the lenders wanted but which were eventually dropped from the Memoranda. An Irish high official, for example, explained to us: ‘the Irish side had many red lines which demarcated things that we wouldn't agree on (…) these were not public and were not crossed’ (Moury et al. 2021, Chapter 4).

In a similar vein, we show that domestic actors sometimes successfully resisted the implementation of some MoU policies they disliked; and that lenders were sometimes willing to turn a blind eye and let non implementation go unsanctioned. This often happened towards the end of the programme when the economic conditions improved and/or when trust was high or even when the Troika was suffering from ‘reform fatigue’. In particular, in Greece, an official from the Troika recalls about the third bail-out: ‘(they were) a mix of various things that the government wants to reverse and for which the Institutions and especially the Commission has given up and says; “ok, go ahead”’ (Moury et al. 2021, chapter 7).

We further demonstrate in our book that conditionality was to some extent useful to executives. This is because only a small number of people within the governments were fully informed of the Troika’s red lines, and the latter, in some instances, exploited this asymmetry of information when presenting their plan to the parliamentary opposition or stakeholders. For example, in Greece, many of the issues about structural reforms raised by the Troika in the first bail-out were already in PASOK’s electoral programme and the Greek government saw the MoU as an opportunity to ensure their implementation (Ladi 2014; PASOK 2009). A key Greek Minister in the negotiations described MoU measures like the public administration reform and the Kallikrates local government reform as ‘self-evident and not imposed by the Troika’.

Similarly, ministers from Portugal told us that the bail-out made the reduction in severance payment possible, as well as the decrease in the rents (excessive profits) enjoyed by the pharmaceutical and electricity companies; and that they had been wanting to do this for a long time. In Spain, informants from both centre-left and centre-right governments told us that they welcomed the reform of the pension system. Similarly, in Cyprus, our interviewees from all major parties stated that the social benefits system needed rationalization. More generally, a government Minister from the Anastasiades government stated that the MoU ‘could thus be considered as part of and indeed as a centrepiece of our governmental programme’ (Moury et al. 2021, chapter 7). In Ireland, respondents acknowledged that the liberalisation of legal services had been on their agenda for some time. A high state official, for example, noted that ‘any good finance ministry will look at what are all the things that we would like to do that we haven't been able to do for a number of years and there's a crisis moment in which you can do them, for example taxation on property’. Another senior civil servant told us that it was not the Troika but the government that had requested the reduction in the minimum wage (Moury et al. 2021, chapter Four).

As a result of this leverage, elections during a bail-out changed the terms of conditionality, albeit to a limited extent. For example, in Portugal, elections took place immediately after the Socialist party had negotiated the MoU and, during the economic adjustment program, new measures—such as a new section on deregulation, a new judiciary reform and the further liberalisation of the labour code—were added to the reviews at the request of the centre-right government. In Ireland, the change in government—formed just after the bail-out negotiations—led to a reversal of the minimum wage cut and an extension of the fiscal consolidation period. In Greece, the SYRIZA government did not manage to change the conditionality of the programme but attempted to implement a ‘Parallel programme’ focusing on alleviating some of the social inequalities arising from austerity.

This finding consolidates the strand of literature which demonstrates the relevance of governing parties’ ideology and electoral motivations during moments of external conditionality imposed by the EU (for example Dyson and Featherstone 1996; De la Porte and Natali 2014), the IMF (i.e. Pop-Eleches 2008) or the Troika (amongst others Moury and Standring 2017; Walter 2016; Hick 2018, Bulfone and Tassinari 2021). Particularly related to this research, Genovese and Hermida-Rivera (2022) recently coded all conditions in the quarters of the 12 bail-outs in Europe during the period 2008–2015 and showed that governments obtain fewer ‘harsh’ measures in areas that are salient for their core voters.

Our book thus further contributes to nuancing the argument that the conditions of international loans are a ‘diktat’. We show that conditionality narrows, but does not annihilate, political choice, and that voting still matters. Even if strongly constrained, governments in all cases had alternatives to balance the accounts and room for manoeuvre as regards the structural reforms.

However, we also show in the book that this room of manoeuvre existed as long as bailed out governments were minimally credible. Building credibility during the crisis was a complex, multi-faceted issue. The improvement or deterioration of economic fundamentals (deficit, debt, etc.) was one factor, but our research reveals that the perception of these changes was also paramount. A good example of this is the successful strategy pursued by the Irish, Portuguese and Cypriot ministers to ‘under-promise and over-deliver’, that is, to purposefully keep expectations low as regards the reduction in their deficit so that they could then do better than expected.

Credibility also derived from a given government’s capacity to implement the agreed deals, itself facilitated by political cohesion and stability. Credibility also interlinks with ideological congruence and personal trust, that is, the good faith and competence of the interlocutor as perceived by the lenders. This was particularly visible in Greece, where the credibility of the government (and hence its leverage) varied at different times depending on the success and speed of the implementation of reforms.

Presentation of the database

We collected quantitative data on the implementation of the MoU and on the resilience of measures taken during the crisis and collated them into a database in order to offer some measurements and comparisons. This coding, as well as the full project, was funded by the Portuguese Science foundation (Democracy in times of crisis: Power and Discourse in a three-level game, 2014–2017, Project Reference: PTDC/IVC-CPO/2247/2014). It is worth keeping in mind that the database was created with the aim of providing additional empirical support to the process-tracing (including 129 interviews), and hence, some choices had to be made about coding.

We first collected the documents produced by the IMF and the EC including both the original MoU’s and their revisions (and for Spain also the documents produced by the EC on the monitoring pursued in parallel to the bail-out, in the context of the European Semester). The Memorandum of Understanding between a given country and the lenders is constituted by two documents: the Memorandum of Understanding with the IMF in which the most important measures are described and the Memorandum of Understanding with the European Commission (which is annexed to the IMF-MoU, and is called: ‘technical Memorandum of Understanding’). The two documents focus on the same topics and include roughly the same objectives and policies, but the former is less detailed and (around by half) shorter than the latter. Typically, the IMF document would identify the reform that should be made, and the technical annex (and the MoU with the EU) would divide those reforms into several detailed measures.

In those two MoUs, and in the EU monitoring documents for Spain, we coded every testable commitment for policy action. In order to do that, we used the methodology that pledge scholars have developed for identifying (and testing) the fulfilment of party pledges in electoral platforms. This method, to which the first author of our book collaborated in elaborating, has been tested for its reliability with satisfactory results (see Thomson et al. 2017 and its Appendix).

Given that we were looking at 7 bail-outs in total, we made the choice to code only the main policies/reforms in the documents in its review. Hence, when the different steps and calendarization of a specific policy or reforms were detailed in the MoU or its annex, they were not coded separately. As discussed in the conclusion, such a choice—if it gives a good idea of the fulfilment of the MoU’s—does not offer a fine-grain measurement of whether every single measure has been fulfilled (on time) or not.

Pledges scholars distinguish between pledges about ‘action’ (e.g. we will decrease taxes by 5%) and pledges about ‘outcome’ (e.g. we will reduce the number of people living in poverty). While the first are directly dependent on the government’s action, the other depends on the success and appropriateness of the government’s policies. In each MoU, they were indeed a series of deficit reduction objectives, which were always completed by a series of concrete policy actions on how to reach them.

Given the limited resources, it made sense for us to code only commitments in the MoU that were about actions and not outcomes. Thus, for example, all the statements about deficit reduction were not included in our database, while the policies or reforms listed to reach those objectives were coded. This choice was made because our focus was the study of the governments’ influence on policy choices and structural reforms and the way these were reflected in the conditionality. The cost-cutting exercise and macro-economic decisions deriving from the austerity dogma had already been extensively discussed in the literature (i.e. Blyth 2013). We wanted to add depth and to understand the policies stemming from this primary decision towards austerity. As noted in the conclusion, however, our database could be completed by a coding of those deficit reduction outcomes.

A second key feature of the pledge scholars’ method is to code only policy commitments that are testable, that is, for which the coder could gather evidence so as to objectively measure whether the action was either accomplished or not (‘narrow definition of pledges’, see appendix of Thomson et al. 2017). Hence, vague statements such as: ‘the government will ensure that the most vulnerable people are not hurt too much by the spending cuts’ was considered too subjective for testing and hence not coded. By contrast, statements such as ‘salaries of the civil servants earning less than 1000 € will not be cut’ would be coded as a policy commitment.

Once testable policy commitments were coded, we continued to follow the pledge scholars’ method to check for the fulfilment. For this, we verified whether the conditions had been fully, partially or not implemented by the end of the Economic Adjustment Programme. This entailed looking for an action by the government implementing each of the specific policy statements previously coded. Although the reports by the EC and the IMF were very useful, we very often had to consult government official websites and legislation, the press, as well as other documents or interview transcripts in order to be sure that implementation had actually occurred.

We finally strengthened Thomson’s method by having a more senior researcher check systematically all statements already coded, and correct them if appropriate (see below for inter-coder reliability tests). Doing so, we discovered that inter-coder reliability was much higher when we merged the categories of full and partial fulfilment (95% for three categories versus 73% for two categories). This is because, in roughly 20% of the (often most complex) cases, the researcher can see that an action has been taken, but evaluating with absolute certainty whether the commitment is partially or fully fulfilled would require days of investigation—something that we could not do for each of the written commitments.

For this reason, although the categories are kept in the database, in our book, we presented and discussed the results with the merging of partial and full fulfilment. That was also a choice operated for the same reasons by some of the pledge scholars (Moury 2011; Artes 2013).

In order to explore the resilience of the policy commitments that were implemented, we coded whether the policies implemented during the bail-out were kept or reversed. Coding reversals is innovative and makes our database original since coding often stops at the end of the Economic Adjustment Programmes. Reversal occurred when an action changed the reform or policy towards the status quo ex ante (that is, the situation before the bail-out). We looked at decisions taken by national executives or legislators, but also by the judiciary or local governments, and we distinguished between full and partial reversals. Full reversals occurred when there was a return of the policy to the status quo ex-ante, and partial reversal that resulted from a movement of the policy towards the status quo ex-ante without reaching it. As it happened for pledge fulfilment, inter-coder reliability was higher with two categories (95%) than for the 3 (73%).

Because some reforms preceded the MoU by a few months, we also coded in all countries the major reforms taken by governments in the period between May 2009—that corresponded to the U-turn of the European Commission in its neo-Keynesian stance towards a preoccupation with public debt (Moury et al. 2021, chapter 1)—and the period of the Economic Adjustment Programme. The documentary sources for the period before the MOU were the country reports produced by the European Commission in this period, relevant legislation, media reports, politician’s memoirs and interviews. In those, we listed every structural reform or cost-cutting decision taken by the country in question.

An important strength of the database is that we also coded a series of explanatory variables. We coded the policy sector (according to a list of sectors, see appendix); the type of reform (structural reforms versus reforms aimed at cutting spending in the short or medium term); whether the reversal of the new policy would have brought diffuse or concentrated costs and whether the reversal of the policy would have brought diffuse or concentrated benefits; the timing of reversal, and the number of veto players (calculated adding the number of parties in governments, 1 if the government was a minority, 1 if the president has a power of veto). The codebook and all inter-coder reliability results—which are all more than very satisfactory—are described in the appendix.

Database findings

MoU’s are less similar than you would think

In Table 1, we present the total number of testable policy commitments for each country per policy field, and we show that the MOUs are not ‘one size fits all’. The number of conditions as well as the spread between different sectors, and between structural versus cost-cutting reforms, differs.

Table 1 Number of testable policy commitments in IMF MoU, per country and policy field
Full size table

We can see that the number of conditions for Ireland is much lower than for the other countries. Different explanations for longer MoUs have been advanced in the literature, such as the economic weakness of a country prior to the crisis (Copelovitch 2010). This appears to be true in our cases. Ireland was already a very liberalized country. Given that the growth of the finance and real estate sectors meant that the economy was particularly susceptible to global shocks, it had, since 2008, re-prioritized its export-led growth model through policies aimed at attracting foreign investment in high tech sectors and high value goods such as pharmaceuticals (Dukelow 2015). Moreover, in the months preceding the request of the bail-out, Ireland had passed 32 measures with the aim of reducing its expenditures and increasing taxation (while the number of such measures for the other bailed out countries were less than 5 for each).

In our book, the process-tracing also uncovers other explanations that had not yet been explored. Trust seems to be of paramount importance, insofar as when lenders do not believe that the decision-makers in a given borrower country act in good faith, they will tend to detail the policies in writing (and hence insert more conditions). Another variable suggested by our process-tracing analysis is the fact that some measures were set out in detail in the original MoU or its review at the request of the government itself, to more easily circumvent the opposition to those reforms. In Spain, for example, the interviewees from the governments mentioned that they included the liberalization of some professions in the MoU hoping to bypass veto players. In Portugal, some parts of the labour code were specified in great detail in one of the reviews at the request of the government and for the same reason.

In the last row of Table 1, we additionally show the proportion of commitments which are structural reforms that are defined as reforms that are designed to boost productivity and improve the supply-side functioning of the economy, in the medium and long term. They include labour code reforms, reduction in business and trade union monopolies, decreased regulation and red-tape, privatization, removal of licensing, improvements to the functioning of the court, and enforcement of property rights (Rodrik 2009). Non-structural commitments were decisions aimed primarily at cutting spending in the short or medium term. We can see that, with the exception of Cyprus which consensually opted for cost-cutting reforms affecting its welfare provision, roughly half or more of the commitments were structural reforms. We introduced this distinction because many cost-cutting policies (such as special taxation or freezing of salaries or recruitment) were not aimed to be permanent, and to the contrary of structural reforms which were not supposed to be reversed after the bail-out.

MoUs are implemented to a large extent

The ENAP also shows that despite the struggles between national executives and the Troika, a large number of MoU reforms had been (at least partially) implemented. Indeed, the spending cuts and structural reforms made by the five countries in our analysis were mostly in line with the MoUs.

In Table 2, we present the number (and percentage) of conditions included in MoUs, whether or not they were (at least partially) implemented. It shows that non-implementation is relatively rare: 84% on average and at least 75% of the conditions of the MoUs were at least partially implemented during the bail-outs. This is in part because governments ‘own’ some of the reforms included in the MoUs, but also because not implementing the MoU is costly. We do, however, observe variation in the fulfilment rate: Ireland and Spain are the champions of fulfilment (more than 90%), followed by Greece (86%), Cyprus (85%) and Portugal (75%). The Spanish and Irish rates are easily explained by the fact that there was a great congruence of views between the centre-right governments and the lenders (see Chapter five Moury et al. 2021). Moreover, Ireland was already very liberalized (see Chapter 3). In addition, the Irish MoU contains less specific policy statements than its counterparts (19 versus 75 for Portugal, for example); it focuses more on objectives to be reached than specific measures. In Cyprus, there was a great internal consensus about the need to introduce the reforms (Chapter 6).

Table 2 Number and percentage of reforms in MoUs at least partially implemented
Full size table

The results for Greece and Portugal are more surprising. In Greece, although the high fulfilment rate goes against conventional wisdom, it is linked to the fact that the breadth and the duration of the programmes were much greater than in the other countries and, therefore, some reforms started in the first MoU and were completed during the second and third ones. Additionally, when conditions were not met immediately, they were broken down into smaller measures so that their fulfilment could be more easily monitored. Finally, the Greek case shows that the country’s disagreements with the Troika were very often about the pace and depth of the reforms rather than the substance (e.g. there was agreement on the need for the pension system reform but the Greek governments could not fulfil the reform at the fast pace required by the Troika, Chapter 2).

Taking a closer look at the Greek data, we calculated the difference between the implementation rates of each MoU. Whereas the fulfilment of the first MoU was high (39/48, or 81% of the conditions were at least partially fulfilled at the end of the first programme), the scores are lower at the end of second (41/66 or 62%) and very high for the third MoU (59/64, 92%). The SYRIZA government’s good fulfilment of the MoU supports the idea that left-wing politicians may be able to introduce structural reforms or spending cuts more easily than their rightist counterparts, as their proposals might face less opposition—this is the ‘it takes a Nixon to go to China’ argument (Cukierman and Tommasi 1998). In the same vein, Gunaydin (2018) shows that left-wing governments reform the labour market more successfully than centre/right-wing governments in the presence of militant labour unions.

Finally, Portugal had the lowest fulfilment score, despite the fact that ministers generally agreed with the lenders about the country's diagnostics and prognosis (see chapter five of the book). One obvious reason for this lower fulfilment is the fact that the Constitutional Court blocked the adoption of 9 measures in the MoU; if this had not been the case, Portugal’s implementation rate would have been more similar to the other countries.

More generally, our process-tracing shows that there are several other reasons than the constitutional court for non-implementation. First, sometimes governments reverse their decisions in the face of social discontent (as the case in Portugal for the payment by the employees of a larger share of social security or in Greece the reversal of the labour reserve system for the public sector). Second, sometimes decisions are reversed during the bail-out after an election. This was the case seven times in Greece, and once in Ireland. For example, in Greece, the Hellenic National Broadcasting Corporation re-opened when SYRIZA took over. In Ireland, the 2010 minimum wage cut agreed with the Troika was reversed in 2011 by a new government. Finally, our five cases show that conditions were often only partially met because of the strength of special interests—see for example the liberalisation of professions in Greece, Ireland and Spain; or the decrease in rents in the energy sector in Portugal.

The qualitative chapters in our book also show, however, that lenders were sometimes willing to turn a blind eye and let non-implementation go unsanctioned. This often happened towards the end of the programme when the economic conditions improved and/or when trust was high or even when the Troika was suffering from ‘reform fatigue’. In Spain, for example, the pension sustainability factor was approved but never implemented. In Cyprus and Ireland, a privatization programme included in the MoU was never fully executed and, similarly, many of the privatizations in Greece are still pending. Lastly, the Portuguese government resisted pressure from lenders to find an alternative to the Constitutional Court’s annulment of a permanent reform of the pension system.

Reforms adopted during conditionality are quite resilient

Our qualitative data additionally show that, after a bail-out, executives were both able and willing to reverse what was done under conditionality. They were able to do so because the lenders do not have the same ‘sticks’ and ‘carrots’ as during the programme (Drazen 2002; Vreeland 1999). In Table 3, we present the absolute number and percentage of the measures passed since May 2009 that were reversed once the programmes ended (and until December 2019). Therefore, Table 3 not only shows the reforms included in the MoUs, but also those passed by governments just before the bail-out, under pressure from external actors—and that is why the total number of reforms is higher to the total number of MoU commitments which had been implemented. As written above, this is especially the case in Ireland, where 32 decisions had been taken by Irish authorities just before asking for a bail-out.

Table 3 Number and percentage of reversals of reforms taken since 2009 after the exit of the bail-out
Full size table

We can see the extent of reversals varies across countries. In particular, post-programme reversals are rareFootnote 1 in Greece and Cyprus. This is not surprising in Greece given that it terminated the third bail-out in August 2018, just 16 months before the end of our analysis. Furthermore, post-conditionality scrutiny has been stricter in Greece than in the other four countries because the Greek loan was the biggest of all and there were many more reforms to be followed up by the government in order to guarantee a long-lasting economic recovery. However, some reversals had been planned in the 2020 state budget prepared by the SYRIZA-led coalition government, and were subsequently implemented by the Conservative government led by New Democracy. This is the case, for example, of the reductions in selected VAT rates, the re-introduction of the 13th month payment for pensioners, and a reversal of an earlier reform of survivors’ pensions (Chapter 2). In Cyprus, the rarity of reversals could also be explained by the fact that the bail-out terminated later than in Portugal, Spain and Ireland, but also by political continuity—as Nicos Anastasiades has been the Cypriot President since early 2013. Nevertheless, some reversals were also made there—such as a progressive reversal of the cuts in public servants’ wages (Chapter 6).

In Ireland, Spain, and above all in Portugal, post-programme reversals were more frequent. In Portugal, in particular, 50% of the measures taken under conditionality were later dismantled, mostly by the minority socialist government that was supported by three radical left parties that had put reversals as a condition for their support (Moury et al. 2019, 2021, De Giorgi and Cancela 2019). In Spain, the new minority Socialist government had promised during its campaign to pass many reversals, but political instability prevented it from passing many of them in the period under study (Chapter 5). In Ireland, there were many reversals despite relative political stability (Fine Gael governed both during and after the bail-out, first in coalition and then alone): most of the public sector cuts and some social benefits cuts were, at least, partially reversed. In this country, many reforms had been taken before the bail-out, and hence most of the measures adopted during the crisis were related to fiscal consolidation, which means that they were also easier to reverse in the post-programme period.

Table 3 further distinguishes structural reforms and other types of measures and shows that out of the 58 reversals, only ten were for structural reforms. This number is mainly driven by the Portuguese case (7 out of 10), where many parts of the labour code reforms were eventually reversed (although the central parts remain, see Moury et al. 2019, 2021). On average, the great majority (91%) of the structural reforms enacted during the crisis remained. For example, the reforms of the banking system in Cyprus, the introduction of the Public Revenue Independent Authority in Greece and the public administration reform, or the pension age increase in Ireland, were all kept. These figures and the information in the case studies testify to the governments’ unwillingness to reverse structural reforms.

Finally, in Table 4, we look at the absolute number and percentage of reversals per type of reform. Here, we see that reversals are less likely when the reforms 1) were structural and involved concentrated costs, and more likely when they concentrated benefits and diffused costs. These differences are statistically significant. The findings speak to the conclusion reached in the literature that governments make cost–benefit calculations when deciding which measures to reverse (Independent Evaluation Office 2003; Rickard and Caraway 2019; Moury and Afonso 2019; Moury et al. 2019; Branco et al. 2019).

Table 4 Absolute number and percentage (in parenthesis) of reversals per type of reform and statistical significance
Full size table

Our data thus shows that since the end of the Eurozone bail-out programmes, executives have been able to reverse what was done during the programme. However, these reversals were partial and limited. Reversals mostly concerned visible spending cuts and tax increases, while less noticeable changes in taxation were often kept. We also show that the great majority of structural reforms were maintained, and thus that conditionality has at least a medium-term effect. The bail-out has therefore brought about a profound and durable change in the socio-economic structure of the society.

Conclusions: limitations and future uses of the database

The ENAP database has shown that economic adjustment programs do have a deep and long-lasting effect in the countries in which they are conducted. Most of the policies and reforms included in the documents are fulfilled, and they are quite resilient (even though it is truer for structural reforms than for spending cuts). In our book, we show that a main reason for this implementation and resilience is that governments, to some extent, ‘own’ the policies of the Memoranda of Understanding. Even if they were clearly constrained, national executives were indeed far from powerless: in all our cases, governments of bailed out countries were able to advocate, resist, shape or roll back some of the policies demanded by external actors.

The release of the ‘ENAP’ database for free use, available here (supplementary material), for other researchers to use and complete, aims to contribute to the production of new academic work in at least three directions. The central focus is expected to be conditionality, its resilience and possible reversals in a comparative perspective.

The first direction could be to complete the data base with more cases or variables. The ‘ENAP’ indeed was constituted with the aim of giving quantitative support to the process-tracing in the five countries. For meaningful use of the quantitative data by itself, some more policies and variables could be inserted. A first step, for example, would be to include commitments about outcomes (such as deficit reduction). Similarly, one could code, for the general policies and reforms, every specific action and its calendarization. The timing of reversals could be added, as well. A third possibility in this direction would be to update the database beyond 2019 and provide a medium-term assessment of reversals or resilience of reforms in all five member-states covered by the ‘ENAP’.

Another step forward in the direction of completion would be to code more explanatory variables. For example, researchers could systematically code the explanation for non-implementation or reversals. One could, also, codify further information about the winners and losers of the reforms. We already coded whether the losers and winners were diffused or concentrated, but it would also be very interesting to identify whether those reforms hurt or favour the core voters of the governing parties. Such coding will allow to investigate further the role of governing parties’ ideology and electoral motivations not only during the formulation of the MoU and its reviews (as done already by Genovese and Hermida-Rivera 2022 and Bulfone and Tassinari 2021), but also after the programme. In that line, a codification of the material and instrumental power of the losers and winners of the reforms would be of great importance to scholars interested in the influence of firms and interest groups.

A second direction of research would be to use our data and coding methodology to enable comparisons of past conditionality exercised in the EU such as the ‘accession’, or EU cohesion funding conditionality, taking into account the different nature and features of these types of conditionality. Similarly, data could be collected for previous instances of IMF lending to EU member-states (i.e. Hungary, Latvia and Romania) at the beginning of the global financial crisis and the conditionality attached to these agreements. An extension of the database to EU member-states that were not members of the Eurozone when they received assistance would allow for conclusions about the significance or not of Eurozone membership for the implementation and resilience of reforms.

A third and forward-looking use of our methodology and database could be its use as a blueprint for the creation of a database recording the conditionality of the National Recovery and Resilience Plans and the implementation of their measures. The Recovery and Resilience Facility offers funding conditional to reforms for all EU member-states in order to assist them in their recovery from the Covid-19 pandemic economic slowdown. The comparison with previous, more targeted forms of conditionality, such as the Eurozone crisis conditionality or the ‘accession’ conditionality, could prove very interesting and allow for conclusions with policy significance for the future relationship between the EU institutions and the member-states.