A Thousand Cuts: Social Protection in the Age of Austerity 0190637730, 9780190637736, 9780190637750

Citation preview

A Thousand Cuts

A Thousand Cuts Social Protection in the Age of Austerity ALEX ANDROS KENTIKELENIS AND T HOM A S S T U B B S

Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and certain other countries. Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America. © Oxford University Press 2023 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by license, or under terms agreed with the appropriate reproduction rights organization. Inquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above. You must not circulate this work in any other form and you must impose this same condition on any acquirer. Library of Congress Cataloging-in-Publication Data Names: Kentikelenis, Alexandros, author. | Stubbs, Thomas (College teacher), author. Title: A thousand cuts : social protection in the age of austerity / Alexandros Kentikelenis and Thomas Stubbs. Description: New York, NY : Oxford University Press, [2023] | Includes bibliographical references and index. Identifiers: LCCN 2023006076 (print) | LCCN 2023006077 (ebook) | ISBN 9780190637736 (hardback) | ISBN 9780190637750 (epub) Subjects: LCSH: International Monetary Fund. | Structural adjustment (Economic policy)—Social aspects. | Social policy. | Public welfare. Classification: LCC HD87.K457 2023 (print) | LCC HD87 (ebook) | DDC 338.9—dc23/eng/20230209 LC record available at https://lccn.loc.gov/2023006076 LC ebook record available at https://lccn.loc.gov/2023006077 DOI: 10.1093/​oso/​9780190637736.001.0001 Printed by Sheridan Books, Inc., United States of America

Στη Φρειδερίκα Νομικού, που μου έμαθε το πραγματικό νόημα της οικογένειας —​ΑΚ For Isobel, Jacob, and Nevyn —​TS

Contents Acknowledgments 

ix

1. Introduction 

1

PA RT I .   4 0 Y E A R S O F S T RU C T U R A L A DJ U S T M E N T 2. The Evolution of IMF Conditionality 

19

3. How to Evaluate the Effects of IMF Conditionality 

40

PA RT I I .   S O C IA L P R O T E C T IO N A N D S T RU C T U R A L A DJ U S T M E N T 4. Conditionality and Health Policy 

77

5. Conditionality and Income Inequality 

110

6. Conditionality and Health Outcomes 

134

PA RT I I I .   L O O K I N G F O RWA R D 7. The IMF and the COVID-​19 Response 

173

8. The Future of IMF Lending: A Better Way? 

189

Appendix: A New Dataset on Conditionality, 1980–​2019  References  Index 

201 227 265

Acknowledgments This book is the outcome of more than a decade’s worth of intellectual exchange and extensive collaboration between its two authors. Our starting point was the observation that the types of policies that were being proposed and implemented in the aftermath of the Global Financial Crisis that started in 2007 bore striking similarities to those that were proposed and implemented after every other major crisis since the 1980s: deregulation, liberalization, privatization. Academic scholarship and policy research had long cast serious doubts on whether such policy packages offered a reliable and durable way forward for countries in crisis. Yet we saw the promotion of the same set of policies by the “experts” of global economic governance over and over again—​from Athens to Accra, from Buenos Aires to Bangui, from Kingston to Kigali. Were the critics wrong? We set out to definitively resolve these debates by collecting a lot of new data and developing new methodological approaches to capture the impact of austerity policies. To complete this endeavor we accrued debts to many scholars, policymakers, and activists, who generously gave their time and energy to discuss our findings, debate our conclusions, and open doors for our research. Here, we limit ourselves to thanking those colleagues without whose input this book would not have been possible. Lawrence King provided encouragement in the early stages of the project and worked with us to publish the initial set of findings. Through generous funding by the Institute for New Economic Thinking and the Cambridge Political Economy Society Trust we were fortunate enough to be able to build a larger research team at the University of Cambridge. Bernhard Reinsberg and Timon Forster joined us on this project and their contributions to the research underpinning this book were invaluable. We draw—​directly and indirectly—​on the findings of our team throughout this volume. We were also privileged to have received guidance, support, and friendly criticisms from David Adler, Sarah Babb, Cornel Ban, David Brady, Miriam Brett, Emma Burgisser, Daniela Gabor, Kevin Gallagher, Ilene Grabel, Jo Marie Griesgraber, Neesha Harnam, William Kring, Valentin Lang, Christina Laskaridis, Martin McKee, Isabel Ortiz, Rebecca Ray, Leonard Seabrooke, David Stuckler, and Robert Wade. The collective input of this group strengthened our arguments and prompted us to think about how abstract processes occurring at the global level impact the lives of people on the ground. At Oxford University Press, we are grateful to James Cook for believing in this project and tolerating delays as we changed jobs, moved countries, and weathered a global pandemic.

x Acknowledgments On a personal level, Alexandros is grateful to Giovanni Menegalle, Domna Maria Michailidou, Philipa Mladovsky, Michał Murawski, and Charlotte Kühlbrand for the many hours spent discussing the impacts of economic crises and their reverberations on politics and society, and to Bart van der Heide for helping him maintain sanity and motivation while wrapping up this manuscript amidst much uncertainty. Thomas thanks Roisin Orchard and William Cochrane for their emotional and material support.

1

Introduction On Monday, June 4, 2018, Jordan’s prime minister Hani Mulki was tendering his resignation after just two years in office. His government had been rocked by five days of intense protests throughout the country, culminating in several thousand demonstrators gathering outside his office the night before. At issue were the introduction of extensive budget cuts and steep tax hikes that the government had signed on to as part of its loan agreement with the International Monetary Fund (IMF). Jordan had already implemented increases in general sales taxes and removed bread subsidies following the IMF’s policy recommendations, but this time the population vehemently resisted. Protesters, egged on by parliamentarians, took to the streets and demanded change: the IMF-​backed reforms were unwelcome. Responding to these calls, the new government promptly announced their intention to revisit the terms of the agreement and dispatched a team to seek a new deal with the IMF. A few months and minor concessions later, the controversial tax reforms were passed by parliament in November. Protesters gathered yet again calling for change, but it was now too late. Under the pressure of deteriorating economic conditions, the government had capitulated to IMF demands for a rapid scale-​back of public debt from 94 percent of gross domestic product (GDP) to 77 percent in just three years. The funds raised through budget cuts and tax hikes would be directed to that goal. Around the same time, on the other side of the world, Argentina’s government, facing economic crisis, went back to the doors of the IMF on Tuesday, May 8, 2018, to request financial assistance. This came almost 15 years after Argentina’s last encounter with the IMF’s policy prescriptions that left nearly a fifth of the workforce unemployed. Unsurprisingly, the Argentine people were not content. Protests erupted all over the country against the steep budget cuts that the government pledged in exchange for the IMF loan. The 2.7 percent deficit in 2018 was supposed to become a 0.5 percent surplus within three years. Protests intensified in subsequent months as the government prepared to enact these cuts. On July 9, Argentina’s Independence Day, protestors mobilized again with signs reading “Independence Cannot Be Negotiated” and “No to the IMF.” That September, general strikes saw hundreds of thousands block the main streets surrounding parliament in Buenos Aires. Despite protests, however, the government pressed ahead. By the end of the year, Argentina was already implementing an array of unpopular measures—​slashing energy subsidies, hiking taxation A Thousand Cuts. Alexandros Kentikelenis and Thomas Stubbs, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/​oso/​9780190637736.003.0001

2  A Thousand Cuts on cooperatives, reducing fiscal transfers to provinces, and instituting a hiring freeze for public sector employees. In both cases, the IMF stands out as the key actor pushing countries to adopt tough budget cuts and structural reforms that fundamentally reshape their policy environments in exchange for providing much-​needed loans at preferential terms. This loans-​for-​reforms practice is known as “conditionality” and is one of the most controversial outputs of any international organization, as it restructures political-​economic environments among borrowing countries. The IMF’s loans are intended for countries facing external shocks or unsustainable debt burdens, thereby placing the organization at the center stage of global economic governance as a key “financial firefighter.” In turn, conditionality is used both to avert moral hazard—​that is, the risk that countries continue to adopt unsustainable policies if they can always anticipate yet another IMF bailout with limited strings attached—​and to ensure that the funds are actually repaid to the organization. Conditionality has afforded the IMF substantial domestic policy influence, with major implications for social policy. Countries like Jordan and Argentina resort to the IMF because the alternative—​defaulting on external debt—​can be much worse for their citizens. It would result in international lenders withholding further credit, which could plunge governments, firms, and households into crisis (Roos, 2019). For example, a default would ripple through the wider economy by provoking capital flight and a collapse of domestic banks; exporters and importers would lose access to trade credit, resulting in shortages of necessary goods; producers would find it difficult to obtain foreign or domestic investment, and would lay off workers; and households would struggle to obtain credit for consumption (Roos, 2019). Faced with such bleak options, most governments still choose the IMF. What has been the scale and scope of IMF conditionality over the past four decades, and how has it impacted social policies and outcomes? These questions are important because they highlight the role of the IMF—​primarily an economic institution—​in materially shaping the social conditions of its borrowers. This has obvious distributional implications, as IMF policy advice explicitly or implicitly promotes the interests of some social groups at the expense of others. How budget cuts are distributed is, after all, a core political question. But the IMF’s impact on social protection systems also shapes economic conditions in the long run, as adverse effects on these systems reverberate across time: they shape the income, health, and living conditions of individuals, and—​by extension—​key attributes of the future workforce of countries in crisis. Tracing the many links between austerity and social protection is the task of the present volume. But before outlining the ambition and contributions of the book, some opening remarks on the histories of the IMF and austerity are in order.

Introduction  3

A Brief History of the IMF In the global economy, no nation is an island. Each depends for many of the things it consumes on imports from other countries, and on revenues from the goods it exports to other countries. There is also a constant migration of capital flows (in the form of investment or loans) back and forth across national borders. When a country’s export revenues and financial inflows add up to less than the cost of imports and capital outflows, the country is said to have a balance of payments deficit. This causes its central bank to lose reserves, and perhaps ultimately leads to a devaluation in its national currency. Orderly, incremental currency adjustments are routine events. Sometimes, however, there are big destabilizing shocks—​a spike in the price of imports, or a crisis in the national financial system—​that can cause large, disorderly devaluations, with a host of undesirable side effects: escalating debt, the mass exodus of nervous investors, spiraling inflation, and the contagious spread of financial instability to other countries. Enter the IMF as the most important global financial firefighter to provide loans to governments to manage and contain balance of payments crises. Among the multitude of multilateral bureaucracies, the IMF stands out for the power it has over borrowing countries and for its role as the focal institution of global economic governance (see Box 1.1). The roots of the organization lie in the Bretton Woods Conference in July 1944, when representatives from

Box 1.1  Activities of the IMF The IMF engages in three main operational activities. First, the organization conducts regular surveillance missions, in which it monitors economic performance and risks at the national, regional, and global levels. These missions form the basis of nonbinding policy advice communicated to countries via annual discussions, called “Article IV consultations.” The IMF’s second core activity is the provision of technical assistance and training on economic issues to central banks, finance ministries, and statistical agencies. It is delivered via a combination of short-​term staff missions, long-​term in-​country placements of resident advisors, and regional capacity development centers. Third, the IMF lends to countries experiencing balance of payments crises. In exchange for financial support, borrowing countries must agree to a package of obligatory policy reforms, or conditionality, administered through a lending program lasting from six months to four years. Loan disbursements are phased over the duration in tranches, contingent upon the implementation of policy reforms assessed on a quarterly or biannual basis.

4  A Thousand Cuts 44 nations gathered in a mountain resort in New Hampshire to negotiate the foundations of the post–​Second World War economic order (Mazower, 2012). The towering figure in these discussions was renowned British economist John Maynard Keynes, even though the ultimate outcome was much closer to the policy preferences of the United States, then already emerging as the world’s largest economy and creditor (Ikenberry, 1992; Steil, 2013). In response to appeals for a system of global financial and monetary governance, the IMF was established in 1945. Its original role was to oversee the system of pegged exchange rates of member governments and make financial resources available on advantageous economic terms to countries facing balance of payments crises. But following the shift to floating exchange rates in 1973, only the second facet endured. Today, the IMF has an estimated one trillion US dollars of lending firepower (Gallagher et al., 2020). In principle, all member-​states can turn to it for financial support, but in practice high-​income countries rarely do anymore—​ with the notable exceptions of Cyprus, Greece, Iceland, Ireland, and Portugal in the aftermath of the 2007–​2008 Global Financial Crisis. The main purpose of the IMF’s support is to prevent financial crises in one country spreading to its trading partners, forestall sharp or disorderly devaluations, and allow countries to keep servicing their external debts. In doing so, the organization became infamous for using these resources as a lever for inducing governments to implement policy reforms that are timetabled in lending agreements and assessed on a regular basis. Nonimplementation can result in delays in loan disbursements and—​ultimately—​the suspension of lending altogether. But the nature of the conditions required of countries has changed substantially over the years. Until the 1980s, the IMF primarily required reforms to fiscal, monetary, and exchange rate policies with the aim of controlling inflation, stabilizing currencies, and reaching sustainable balance of payments (Dell, 1981, 1982; Diaz-​Alejandro, 1981; Williamson, 1983). These reforms appeared in loan documents as a series of quantifiable targets, such as reductions in the fiscal deficit and money supply. While potentially painful to local populations, IMF intervention was short-​term in nature—​usually about one year in duration—​and did not disrupt the relative role of states and markets in domestic economies, a matter considered beyond the IMF’s mandate (Finch, 1983). In the mid-​1980s, however, the IMF introduced a series of lengthier lending programs—​typically up to three years in duration—​targeting structural change. The term “structural adjustment” became shorthand for an extensive range of reforms in these programs to promote fundamental, comprehensive, and enduring overhaul of a borrowing country’s policy arrangements. Against a background of protracted economic crises in the 1980s and 1990s, structural adjustment programs became ubiquitous, achieving notoriety for requiring developing countries to implement market-​ liberalizing reforms

Introduction  5 (Chang & Grabel, 2004). The IMF’s advocacy of an erstwhile narrow set of reforms—​mostly on fiscal and monetary policy—​thus expanded to a much wider remit: the elimination of barriers to trade and the movement of international finance to facilitate access to international markets and promote foreign direct investment; repeal of government rules, regulations, and checks and balances surrounding economic activity—​such as industry entry criteria and labor standards—​to abolish perceived inefficiencies in the functioning of the private sector; and the selling of state-​owned enterprises and natural resources to the private sector, with the hope of improving the economic management of these industries (Summers & Pritchett, 1993; Toye, 1994; Williamson, 1990). The augmentation of the remit of IMF activities to target policy areas that were hitherto under much greater state control was seen as evidence of “mission creep” and prompted intense controversies (Babb & Buira, 2005). On the one hand, these policies were seen as an undue imposition of a radical free market agenda on countries caught up in a crisis (Simmons et al., 2008). The fact that it was an international organization dominated by the Global North (see Box 1.2) forcing countries in the Global South to implement extensive reforms that expanded the remit of markets and reshaped state infrastructures led observers to draw parallels to the colonial era, only now rich countries were operating more stealthily through the cloak of a venerable multilateral institution (Browne, 1984). But it was not only the ideological orientation of these policies that attracted attention. The types of reforms promoted by the IMF sought to irrevocably alter the policy environments of its borrowers by reducing governments’ options on how to deal with economic predicaments. Once many of the IMF-​mandated reforms had been implemented, they were—​by design—​very difficult to reverse, as they create their own self-​enforcing dynamics (Appel & Orenstein, 2016; Stallings & Peres, 2011). For instance, attempts by countries like Argentina or Ecuador to renationalize national resources or companies sold off as part of IMF conditionality in the past have been met with vehement resistance from the multinational corporations that now own them and resulted in years of expensive litigation. The promise of the IMF’s painful reforms was that they would be justified in the long run by sustained economic growth, which generally failed to materialize. This was most emphatically the case in sub-​Saharan Africa, where the 1980s were dubbed a “lost decade” following a succession of IMF-​endorsed crisis management measures that precipitated negative growth rates and triggered rapid rises in poverty (Harrison, 2010; van de Walle, 2001). Faced with disconfirming evidence, the IMF’s view was that their programs had paid insufficient attention to the institutions that allow markets to function, such as laws and judicial systems, but that the underlying market-​liberalizing impetus was essential correct. By the 1990s, conditionality had expanded further to include rule

6  A Thousand Cuts

Box 1.2  Governance of the IMF The IMF is located in Washington, DC, and headed by a managing director who is, by convention, a European national—​since October 2019 Kristalina Georgieva from Bulgaria. It is staffed by technocrats trained in neoclassical economics at elite universities, typically in the United States or United Kingdom. The IMF’s highest decision-​making body is the Board of Governors, where all its 190 members are represented, and which meets twice a year. In terms of day-​to-​day operations, the organization is governed by an executive board that meets up to three times per week, composed of 24 executive directors that are appointed by member countries to decide on a range of key issues, including the approval of loans and the establishment of organizational policies. The institutional setup of the IMF reinforces dominance by the United States and other economically powerful countries that contribute to its resource base. In contrast to many other international organizations, their voting power is linked to this dominance. The United States holds the largest block of voting shares (16.5% of votes in 2021), followed by Japan (6.2%), China (6.1%), Germany (5.3%), France (4.0%), United Kingdom (4.0%), Russia (2.6%), and Saudi Arabia (2.0%). These eight shareholders appoint their own executive directors, whereas other countries must form constituencies (except for Syria, which is represented by the Russian executive director). The most important decisions, such as changes to the mandate, require a supermajority of 85%, giving veto power to the United States. Resistance to organizational changes that would grant greater influence to developing countries remains a major source of contention and conflict (Vestergaard & Wade, 2013; Wade, 2013a, 2013b; Wade & Vestergaard, 2015).

of law and governance issues, and also become a staple vehicle for implementing the transition to capitalism in postcommunist countries. The IMF was extensively criticized in subsequent years, especially following their handling of financial crises in Mexico, East Asia, Russia, and Argentina (Babb & Carruthers, 2008; Stiglitz, 2002; Wade & Veneroso, 1998). By the early 2000s, in response to such criticism, the IMF pledged to strengthen the pro-​poor orientation of their programs (IMF, 2001b). We evaluate this track record later in this book.

Austerity and Its Discontents A narrow interpretation of the term “austerity” refers to a reduction in the government budget balance. The budget or fiscal balance equates to the difference

Introduction  7 between what a government spends and the revenues it collects. A reduction in this balance, otherwise known as “fiscal consolidation,” occurs when governments cut expenditure and when they generate more revenues, such as by raising taxes. From a pragmatic standpoint, many economists—​including IMF staff—​view austerity as a desirable policy prescription in times of balance of payments crisis because it frees up government resources to repay external debts and replenish international reserves (Ban & Patenaude, 2019; Chwieroth, 2010; Momani, 2005; Nelson, 2014a). Ideologically speaking, austerity also guards against having a big interventionist state, seen as anathema by orthodox economists, for whom the market is thought to be able to provide services to the population more efficiently (Ban, 2016). The idea goes that austerity is a form of voluntary deflation that allows a government to adjust via reductions in wages, prices, and public spending to restore competitiveness and inspire business confidence (Blyth, 2013). As austerity advocate John Cochrane of Stanford University proclaimed, “Every dollar of increased government spending must correspond to one less dollar of private spending. Jobs created by stimulus spending are offset by jobs lost from the decline in private spending. We can build roads instead of factories, but fiscal stimulus can’t help us to build more of both” (cited in Blyth, 2013). A wide mix of policies have been historically used to consolidate budgets (Blyth, 2013; Ortiz & Cummins, 2019, 2021). Cuts or caps on the government wage bill reduce or freeze the salaries and number of public sector workers. Reducing or removing social welfare programs—​occasionally big budget items even in countries in the Global South that still spend considerably less than high-​income countries—​can free-​up funds. This includes overhauls of public health and education systems, like raising fees and introducing co-​payments for patients or students, which reduce public spending. Labor market flexibilization reforms help limit salary adjustments and decentralize collective bargaining, which encourages precarization of employment and further depresses incomes. Privatizations of state-​owned enterprises can both remove expensive institutions from the budget balance sheet and generate short-​term revenues that can limit budgetary pressures. Finally, increases in consumption taxes on basic goods and services, such as value-​added taxes or excise taxes, has very commonly been a staple policy to bolster revenues. In rich countries, the onset of austerity measures is commonly associated with the ascendancy of right-​wing governments in the 1980s—​most notably, Margaret Thatcher in the United Kingdom and Ronald Reagan in the United States—​that had ideological predispositions toward reducing the role of the state in the economy and expanding the remit of markets (Blyth, 2013; Harvey, 2005). In developing countries, however, it was often the IMF that advised governments to undertake austerity reforms, either as part of its regular surveillance missions or when countries had to sign up to its “structural adjustment programs” to

8  A Thousand Cuts borrow money. For this reason, the term “austerity” is sometimes used synonymously with “structural adjustment,” even though in practice IMF programs can entail a broader set of policies. As Larry Summers and Lant Pritchett explained, these programs were designed to target “the four ‘-​ations’—​stabilization, liberalization, deregulation, privatization” (Summers & Pritchett, 1993). That is, they went beyond simply balancing the budget in the short term, as a narrow understanding of austerity implies, but were more ambitious in their nature. In this book we adopt this wider understanding, and use the terms “austerity” and “structural adjustment” interchangeably in reference to the broader suite of measures, unless stated otherwise. Austerity policies have been consistently criticized for prioritizing short-​term fiscal objectives over longer-​term social investments such as health. For example, civil society organizations contend that decades of austerity promoted by the IMF—​as well as its sibling institution, the World Bank—​have undermined social protection, impeding the ability of governments to respond to major shocks, including the COVID-​19 pandemic (Griffiths & Todoulos, 2014; Ortiz & Cummins, 2019; Ortiz & Stubbs, 2022). Further, close observers claim that austerity had a detrimental impact on the population—​that inequalities grew, and millions were pushed into poverty, with women particularly affected (UN Women, 2015). Some countries, like Pakistan and Jordan, have seen the emergence of widespread social mobilization to fight the introduction of austerity. Evidence on worldwide social protests suggests that 53 percent of them were resisting austerity measures and over 10 percent explicitly targeted at IMF conditionality (Ortiz et al., 2022). In this book, we weigh in on controversies surrounding the social consequences of IMF-​mandated austerity. These controversies have a long history (Payer, 1974). Critics argue that—​by design or by omission—​IMF conditionality can result in extensive collateral damage in developing countries (Babb, 2005), and that a social development policy agenda was never institutionalized within the organization even though it has become a widely recognized international priority (Vetterlein, 2010). But despite early recognition by the IMF that—​in the words of its then managing director—​“adjustment that pays attention to the health, nutritional and educational requirements of the most vulnerable groups is going to protect the human condition better than adjustment that ignores them” (de Larosiere, 1986), criticisms of its policy advice only intensified. Trailing the expansion of the IMF’s remit into an ever-​growing number of policy areas, a new generation of studies furnished additional evidence of the links between IMF-​ designed reforms—​like privatization, trade and financial liberalization, and the marketization of social policies—​and adverse social consequences (Colclough & Manor, 1991; Stewart, 1991; Toye, 1994; van der Hoeven & Stewart, 1994). The IMF responded to these criticisms by claiming that its modus operandi

Introduction  9 now incorporates due attention to social protection issues (Fedelino et al., 2006; Gupta, 2015, 2017; IMF, 2008, 2015b; Martijn & Tareq, 2007; McDonald, 2007; Verhoeven & Segura, 2007; IMF, 2019b). Who is correct? Is the organization right to proclaim that it has addressed criticisms and adapted the policies attached to its financial assistance packages vis-​à-​vis social protection? Or are critics accurate in their portrayal of the IMF as “going backwards” (Griffiths & Todoulos, 2014)? Tackling these contested questions requires new data, the use of appropriate quantitative methods, and extensive reanalysis of the impact of IMF-​mandated reforms on public policies and the human condition. This is the task for the remainder of the book. In taking on these questions, we build on scholarship across the social sciences that has highlighted the connections and tensions between global processes and local outcomes, between economic reform and social change, between supposedly apolitical technocratic knowledge and attempts by peoples and movements to defend rights and livelihoods (e.g., Almeida, 2014; Almeida & Martín, 2022; Ghodsee & Orenstein, 2021; Kentikelenis, 2018; Orenstein, 2008; Pfeiffer & Chapman, 2010; Walton & Seddon, 1994).

The Ambition and Contributions of This Book Writing a book on structural adjustment in 2023 might appear anachronistic to some readers. Indeed, the IMF jettisoned this term altogether at the turn of the millennium and has carefully distanced itself ever since. The official narrative emanating from the institution is that they “do not do that anymore” (IMF, 2014k). But a starting point for our analysis is that taking the IMF narrative—​or any self-​serving narrative by an international organization, for that matter—​at face value is inappropriate. To be sure, “structural adjustment programs” were replaced by “poverty reduction and growth programs”—​but does that really mean that the latter were substantially new creatures, and ultimately delivered poverty reduction and growth? These are empirical questions, with important real-​world implications. After all, the IMF’s conditions fundamentally reshape domestic policy environments in borrowing countries, thereby having a direct impact on socioeconomic development and long-​run development trajectories. However, answering these empirical questions has been marred with many difficulties until very recently. Most notably, social scientists did not have access to detailed and easy-​to-​use data on conditionality (Vreeland, 2006), thereby leading to highly imprecise measurement. For example, scholars studying the impact of IMF programs on different indicators tended to use a binary variable for the presence of a program or not in a given country in a given year. This assumes that all IMF programs are alike, when in fact they contain different types of

10  A Thousand Cuts conditions: some programs may emphasize labor market reforms, others might prioritize privatizations, and some may focus on rule of law and anti-​corruption issues. Pretending they are all alike impedes a fine-​grained evaluation of which precise policies help or hurt in relation to social outcomes. This volume makes three key contributions. First, we present a comprehensive new dataset of IMF conditionality, which we also make publicly available at www.imf​moni​tor.org. Based on archival material on the IMF’s lending operations, we identified all policy conditionality in loan agreements between 1980 and 2019. In total, we retrieved over 6,100 loan-​related documents, from which we extracted 65,707 individual conditions applicable across 132 countries. We then classified these conditions into eight mutually exclusive policy areas. The detailed codebook is available in the appendix. Through this data, we can initially show that “structural adjustment” is not a term with delimited use to describe 1980s and 1990s conditionality, but still highly relevant. There is a clear family resemblance of the reforms introduced in the past and those advocated today, albeit with modifications that we trace and clarify. Second, we introduce a novel methodological approach to capture the effects of IMF programs. Our approach is an improvement on existing methods because it allows us to isolate where an effect is derived from among several types of IMF conditions, promising greater nuance for our findings and their policy implications. For instance, we can establish whether it is trade liberalization or state-​owned enterprise reform that influences social outcomes, whereas earlier studies could only identify that IMF intervention had an overall effect. Our approach also allows us to make causal—​rather than correlational—​arguments by addressing statistical biases that have hamstrung earlier analyses: that governments choose to participate in IMF programs reflecting, inter alia, the severity of economic crises and the political will of leaders to address the situation; and that once participating in a program, governments receive more or less conditions depending on domestic characteristics that also affect social outcomes, such as the extent to which democratic rights are respected. Finally, with these two tools, we can turn to making a range of substantive contributions based on new analyses on the impact of IMF programs on social protection. Our findings provide grounds for continued skepticism about the role of conditionality. Across a range of empirical analyses, we show that IMF programs have led to the introduction of regressive public policies, which in turn have adverse impacts on social outcomes. Relatedly, we document that while the organization has made some strides in the introduction of pro-​social spending measures, these are often cosmetic changes that are frequently not enforced. In short, there is still scope for extensive reforms in IMF programs if they are to underpin equitable socioeconomic development, especially in the context of the postpandemic recovery.

Introduction  11 A note on what this book does not attempt is also relevant. We focus on the activities of the IMF due to its key role in influencing policy design at the level of central government decision-​making. Its mandated reforms directly target tax and spending policies designed by finance ministries, and spell out structural changes in most major areas of economic policy. In doing so, the IMF shapes the policy space available to decision-​makers across levels of government. But the IMF is not the only organization with the power to do so. The World Bank and the European Union can also attach conditions to their support, sometimes in cooperation with the IMF (Broome, 2013; Kranke, 2020; Lütz & Kranke, 2014). In particular, the World Bank has also had a role in designing structural adjustment programs for low-​and middle-​income countries around the world since the 1980s. Yet for the Bank, such programs—​now revamped as “development policy lending”—​form a relatively small share of its lending activities, with the bulk going to finance large projects like infrastructure or health facilities. Even so, criticisms of the role of World Bank projects and policies abound, but these receive extensive coverage elsewhere (e.g., Cormier & Manger, 2022; Gabor, 2021; Malik & Stone, 2018; Noy, 2017, 2021; C. L. Shandra et al., 2011).

Outline The goal of this book is to explore how the IMF has influenced social protection worldwide. Without denying the importance of other areas of IMF activity (such as economic surveillance and technical assistance), we turn our attention to the decisive role played by the organization’s practice of conditionality in shaping public policies and social outcomes. In Chapter 2, we situate our inquiry within contemporary scholarly debates and media criticism over the extent to which the IMF’s official narrative of organizational reform matches reality. Since the onset of the Global Financial Crisis in 2007–​2008, a number of analysts have observed a “new, cuddly IMF” (Wolf, 2011). This has been coupled with a proliferation of self-​congratulatory organizational factsheets, discussion notes, and public statements telling us that the IMF is now a changed institution. If the hype is to be believed, the IMF took on board the many criticisms of yesteryear and reformed its practices. The pièce de résistance of the new IMF is an organizational mantra of enabling policy flexibility for borrowing countries, a marked shift from the rigid structural reforms that shackled governments and were so notorious as to elicit a “phobia of the IMF” among people in countries participating in structural adjustment programs (IMF, 2014k). The chapter sets out to examine whether IMF rhetoric on policy flexibility for its borrowers matches the actual conditions attached to the organization’s loans. In so doing, the chapter offers useful context for Part II

12  A Thousand Cuts of the book by providing an overview of IMF conditionality, including its organizational and technical apparatuses. Using our original data, we focus on whether the IMF has evolved to allow for more policy space by empirically exploring the patterns of conditionality in loan agreements between 1980 and 2019. What does the available evidence show? It provides little support for arguments claiming a fundamental transformation of practices by the IMF. Instead, we find that the scale of organizational change was both modest and short-​lived. In the immediate aftermath of the Global Financial Crisis the scope of reforms was somewhat reduced, but as the IMF solidified its role as the central institution charged with crisis management, its programs re-​incorporated many of the reforms that it claimed to no longer advocate. To explain this finding, we conceive of the IMF as an open system influenced both by external environments and internal pressures and demands that give rise to a decoupling between stated goals and actual practices of conditionality. Put simply, what the IMF says is not necessarily what the IMF does. Before proceeding with the analyses, it is essential to scrutinize the methods social scientists use to evaluate the effects of IMF conditionality. Chapter 3 takes on this task, arriving at a methodological template applied to the quantitative studies undertaken in Part II of this book. To understand the issues involved in pinpointing effects of IMF intervention on social outcomes, we consider the oft-​used analogy of the IMF as a doctor giving treatment to a patient. Countries opt into an IMF program because they are suffering from macroeconomic maladies. The IMF prescribes a series of conditions to treat the patient, or ailing economy. If we fail to account for the fact that the patient—​or country—​was initially feeling poorly, then we underestimate the effectiveness of the doctor’s—​ or IMF’s—​treatment. Scholars are, of course, acutely aware of such issues and have developed complex statistical solutions as a result, techniques we consider in the chapter. Yet these studies rely on a crude indicator of IMF intervention that statistically treats all programs as if they are identical, since data on conditionality was not yet available for this earlier generation of studies. Once we consider information on conditionality, additional steps must be included in statistical analyses to ensure that we capture the effect of IMF intervention and not pre-​existing circumstances that prompted the country’s government to seek a program. Thus, we develop a sophisticated state-​of-​the-​art statistical strategy capable of testing the effects of IMF conditionality. In so doing, we can account for the two main sources of statistical bias. First, countries decide to enter an IMF program based on an array of domestic and global economic, political, and social characteristics, some of which cannot be directly observed or measured, which may in turn influence the outcome of interest. And second, these characteristics may also determine the number and type of conditions included in the IMF program. By statistically controlling for these factors, we are able to ascertain

Introduction  13 whether there is truly a causal relationship between IMF conditionality and social policy and outcomes, rather than being a mere correlation. Chapter 4 focuses on health systems and is the first of three chapters in Part II that apply our statistical strategy to examine the social consequences of IMF conditionality. Strengthening public health systems is central to achieving universal health coverage and to minimizing the fall-​out of global health emergencies, witnessed so acutely during the COVID-​19 pandemic. In setting fiscal priorities for borrowing countries through the practice of conditionality, the IMF has emerged as a key player in shaping the design and implementation of health policies and systems throughout the world. While the IMF has long been criticized for impeding the development of public health systems in borrowing countries, the organization has persistently argued to the contrary, claiming their programs have in fact strengthened public health systems. We wade in on this debate by focusing on cross-​national and historical comparisons of government health spending in response to IMF conditionality. The results demonstrate that IMF conditionality has a robust and powerful negative influence on government health spending. These effects are mostly channeled through labor-​ related reforms, such as wage and employment ceilings, which limit the ability of governments to hire and retain civil servants, including key medical personnel. We also find some evidence that fiscal, revenue, and tax reforms—​such as ceilings on general government expenditures and floors on the fiscal balance—​ shrink fiscal space for spending in the health sector. We then show how these channels of influence bear out on the ground in an analysis of additional archival documents for West Africa, a region that stands out due to the sheer magnitude of health and developmental challenges and near-​constant presence of IMF programs over the past three decades. Chapter 5 evaluates the impact of IMF conditionality on income inequality. The IMF has in recent years styled itself as a champion of reducing inequalities, representing a sharp departure from the reputation of the organization. According to its critics, the IMF has increased social inequalities through its own policy advice. Both sides of the argument have been guilty of resorting to selective anecdotes—​rather than rigorous scientific inquiry—​to consolidate their positions. Given the centrality of the IMF in guiding post-​COVID economic recovery, it is crucial to obtain something approaching closure on the IMF’s record on income inequality. We therefore set out to test these competing claims systematically across a broad range of countries and years. Our results validate the position of the critics: IMF conditionality drives increases in income inequality, as measured by the income Gini. These findings are driven by reforms pertaining to the liberalization of flows of goods and capital as well as currency changes. Workers’ rights are often adversely affected by trade liberalization, as firing is made easier and employment relations favor business interests rather than

14  A Thousand Cuts worker protections; and the expansion of portfolio investments from liberalizing capital flows can increase market volatility and amplify financial crises, disproportionately affecting poorer individuals. Using an auxiliary dataset measuring the intensity of fiscal adjustment required by countries participating in IMF programs, we also show that IMF fiscal consolidation—​conditions calling for cuts to government spending and increases in revenues—​fostered inequality by concentrating income to the richest 10 percent of the population, with middle-​ class and low-​income earners accruing the biggest losses. This occurred via wage, employment, and pension cuts for civil servants, as well as through rises in value-​added taxes over income and corporate taxes. Has IMF conditionality ultimately harmed population health? Chapter 6, which is the last of the three chapters in Part II, investigates this issue. Apart from any impact that conditionality has on government health spending, increasing attention has focused on the effects of broader public policies on the social determinants of health, understood as the circumstances “in which people are born, grow, live, work, and age” (Commission on Social Determinants of Health, 2008, p. 1). The chapter initially focuses on two key measures of health outcomes: a universal health coverage index based on a series of risk-​ standardized death rates from 32 causes, and the neonatal mortality rate. Based on the most rigorous statistical techniques available, the evidence shows that across multiple decades throughout the world conditionality reduced coverage of essential health services and increased neonatal mortality, regardless of how much a government spends on health. This effect is derived from state-​owned enterprise privatization and labor conditions, which affect health systems indirectly by weakening state capacity to deliver a range of effective public health services and interventions, as well as depriving public sector workers of high-​ quality healthcare access where such employment benefits are withdrawn. These conditions have a cascading impact on key social determinants of health, as changing work patterns and unemployment lead to stress and social exclusion, affecting access to food and fueling self-​destructive behaviors such as alcoholism. We also find a detrimental effect linked to health sector restructuring, as well as price increases for basic needs goods like food, water, public transport, and shelter, access to which influences key social determinants of health like stress and early life experiences. To further explore the effects of IMF conditionality on health outcomes, the chapter then draws on data for 23 additional indicators featured as health targets nested within Sustainable Development Goal 3 (i.e., to ensure healthy lives and promote well-​being for all at all ages). Of these, 10 were influenced adversely by IMF conditionality: maternal mortality, under-​five mortality, malaria cases, hepatitis B incidence, vaccination coverage, health worker density (nurses and midwives, physicians, and summed across cadres), deaths from poisoning, and deaths attributed to unsafe water, sanitation, and hygiene.

Introduction  15 The only indicator where we detected a beneficial impact was for the prevalence of smoking, a likely product of conditions calling for increases in excise and value-​added taxes, thereby shrinking demand for tobacco due to rising prices. Chapter 7 grapples with the question of how the IMF has changed since the onset of the COVID-​19 pandemic. The IMF stressed the importance of avoiding a divergent recovery from the pandemic, where some countries race ahead and others fall further behind. To this end, IMF managing director Kristalina Georgieva called for an economic recovery from the pandemic that should not witness budget cuts, but rather investments in employment and human capital (Georgieva, 2021b). Given these statements, are our findings on the detrimental social impact of IMF conditionality still relevant in a post-​COVID world? Based on public spending projections from the IMF’s highly influential World Economic Outlook report, we find—​as we did in Chapter 2 for an earlier era—​that the IMF’s proclamations cannot be taken at face value. A total of 86 countries are projected to face fiscal contractions by 2023—​that is to say, they will be spending less than their 2010s average—​exposing 2.3 billion people to the socioeconomic consequences of budget cuts. These projections become prescriptions for many developing countries, especially IMF borrowers where they get inscribed as conditions. We also present new data on IMF financing approved to combat COVID-​19 and subsequent economic recovery. While no-​ conditionality facilities were the prevalent lending instrument for the initial phase of the pandemic, these are being replaced by lending arrangements that mandate the introduction of policy reforms—​those that have been the topic of analyses throughout this book. To better understand the requirements of post-​ COVID IMF programs, we examine the precise content of IMF conditionality in three of these programs: Ecuador, Kenya, and Madagascar. We find that these programs continue to include steep austerity measures and structural reforms like privatizations or civil service restructuring. The narrative on the abandonment of austerity is therefore dubious, as there is every indication that the IMF is returning to its business-​as-​usual approach. Mistakes of the past—​such as conditions that have reduced government health spending, increased income inequality, and detrimentally impacted health outcomes—​may well be repeated without decisive action to forge an alternative approach. Looking forward, what is to be done? In the concluding Chapter 8, we elaborate on how the IMF can effectively live up to its promise and potential by encouraging more effective engagement with social protection policies and avoiding the adverse socioeconomic consequences that we document in this book. In the immediate aftermath of the Global Financial Crisis of 2007–​2008, commentators remarked on what appeared at the time as a period “pregnant with new development possibilities” (Grabel, 2011, p. 805), which—​as our data shows—​never came to fruition. And while international political economic history does not

16  A Thousand Cuts repeat itself, it does sometimes rhyme. Early in the COVID-​19 pandemic, no-​ conditionality IMF facilities supported rapid and substantive increases in health and social protection spending; the longer-​term outlook presents a bleak picture of austerity redux. But a window of opportunity does present itself. This window is closing but has not shut completely. Among governments and international institutions, social policy is receiving renewed attention and there has become a much clearer recognition that “we are only as strong as the weakest health system,” as UN Secretary General Antonio Gutiérrez put it (United Nations, 2020). The chapter contemplates how IMF programs will interface with social protection policies in a status quo scenario. We then home in on a fledgling area of IMF engagement that is likely to become the dominant battleground for social protection issues: climate change adaptation and mitigation policies. Finally, we reflect on whether the IMF’s mandate is consistent with achieving a socially just world economy (it is!), before proposing concrete recommendations on how the organization can move toward addressing the criticisms raised in this book.

Part I

40 YE A R S OF ST RU CT U R A L A DJ U STMENT

2

The Evolution of IMF Conditionality After half a century of undisputed dominance as the core actor in global economic governance, the IMF’s role and credibility was seriously questioned in the first two decades of the twenty-​first century. In the early 2000s, the IMF’s handling of economic crises and the subsequent design of reform programs came under fire for failing to deliver stability and growth. Prominent critics, like Nobel prize-​winning economist Joseph Stiglitz (2002), forcefully argued that IMF involvement actually worsened crises in a range of low-​and middle-​ income countries, which gradually turned their back on the institution. By the mid-​2000s demand for its services was at historic lows, resulting in staff layoffs and widespread doubts about its future (Helleiner & Momani, 2007). Noting this demise, anthropologist and activist David Graeber (2008, p. 16) pronounced that “the IMF is rapidly approaching bankruptcy, and it is a direct result of the worldwide mobilization against them. To put the matter bluntly: we destroyed it.” But such statements proved premature. In the aftermath of the 2007–​2008 Global Financial Crisis, G20 leaders committed $750 billion to enable the IMF to step up its operations. Reflecting the significance of this cash injection, its managing director at the time, Dominique Strauss-​Kahn, could safely proclaim that “the IMF is back” (Walker, 2009). Between 2009 and 2019, the organization made 229 new loans to 87 countries, reinforcing its standing as the central actor in global economic governance and crisis management. The IMF’s recent revival has been accompanied by an unfamiliar discourse. Following the Global Financial Crisis, the re-​emergence of the IMF was also coupled with a sea change in the organization’s rhetoric on the ways in which it offers financial assistance. As former managing director Christine Lagarde told journalists, “Structural adjustments? That was before my time. I have no idea what it is. We do not do that anymore. No, seriously, you have to realise that we have changed the way in which we offer our financial support” (IMF, 2014i). The oft-​repeated “we do not do that anymore” mantra is far removed from what observers have long come to expect from the IMF, and is a clear departure from the reputation that normally precedes it—​prescriptions of fiscal austerity, trade and capital account liberalization, public sector layoffs, and privatizations of state-​owned enterprises (Babb & Buira, 2005; Babb & Kentikelenis, 2018; Gallagher, 2015b; Stiglitz, 2002). Instead, the organization now acknowledges the importance of countercyclical spending to sustain economic activity (Berg et al., A Thousand Cuts. Alexandros Kentikelenis and Thomas Stubbs, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/​oso/​9780190637736.003.0002

20  40 Years of Structural Adjustment 2009; IMF, 2009b, 2020p), the potential utility of capital controls (IMF, 2010a, 2011; Ostry et al., 2010), the perils of high income inequality (Dabla-​Norris et al., 2015; Fabrizio et al., 2017; IMF, 2014b, 2017a; Ostry et al., 2014), and the adverse consequences of inadequate social protection policies (IMF, 2014h, 2015b, 2019b; Lagarde, 2019). Taken at face value, such policy advice would amount to a monumental shift in the IMF’s practices. Some scholars have identified the potential onset of an organizational recalibration in these developments (Ban, 2015; Ban & Gallagher, 2015; Broome, 2015; Gallagher, 2015a; Grabel, 2011, 2017, 2019). But there is reason for skepticism. Several studies suggest that the practice of IMF programs in recent years largely reflects business as usual (Brunswijck, 2018; Gabor, 2010; Griffiths & Todoulos, 2014; Güven, 2012; Stubbs & Kentikelenis, 2018b; Weisbrot et al., 2009). For example, the authors of this book revealed that of the 48 programs commencing between January 2010 and May 2016, 58 percent were fiscally contractionary in the initial year of the program, and even more became contractionary in their second, third, or fourth years (Stubbs & Kentikelenis, 2018b). Similarly, Brunswijck (2018) revealed that 23 out of 26 programs that were approved in 2016 and 2017 were pro-​cyclical. Griffiths and Todoulos (2014, p. 4) also pointed to “widespread and increasing use of controversial conditions in politically sensitive economic policy areas.” The IMF has challenged the accuracy and veracity of such conclusions (e.g., Gupta et al., 2018; Gupta & Shang, 2017; IMF, 2017c; Yukhananov, 2014). Until now, however, short timeframes and small samples have limited our ability to evaluate these competing claims. This chapter sets the context for Parts I and II of the book by providing an overview of IMF conditionality. It begins by describing analytical approaches to the study of operational change—​or lack thereof—​by international public bureaucracies, before focusing on a long-​standing controversy: Has the policy content of IMF programs evolved to allow for more policy space? To explore this issue as well as questions posed in subsequent chapters, we collected relevant archival material on the IMF’s lending operations and identified all policy conditionality in IMF loan agreements between 1980 and 2019. In total, we retrieved over 6,100 loan-​related documents, from which we extracted 65,707 individual conditions applicable across 132 countries. Using our original data, we present in this chapter general trends in the application of conditionality over the past four decades. The chapter establishes that in the immediate aftermath of the Global Financial Crisis the scope of reforms was somewhat reduced, but what appeared at the time as a period “pregnant with new development possibilities” (Grabel, 2011, p. 805) never came to fruition. Instead, as the IMF solidified its role as the central institution charged with crisis management and built-​up organizational self-​confidence anew, its programs reincorporated many of the reforms that it claimed to no longer advocate. The most recent data from 2019 show sharp

The Evolution of IMF Conditionality  21 increases both in the total number of conditions and in the array of policy areas under reform.

Organizational Change or Organized Hypocrisy? The apparent distance between the IMF’s rhetoric—​ broadcasted in the organization’s conditionality reviews (IMF, 2012b, 2019a), factsheets (IMF, 2014h, 2015b), speeches (Lagarde, 2015, 2019), and responses to critics (Gupta, 2010, 2017)—​and its actual practice of conditionality can be understood as processes of “paradigm maintenance” and “organized hypocrisy” (Güven, 2012; Lipson, 2007; Wade, 1996; Weaver, 2008). Through this rubric, contemporary organizations are viewed as open systems influenced both by external environments and internal pressures and demands (Pfeffer & Salancik, 1978; Scott, 2014). As a result, a “decoupling” can emerge between stated goals and actual practices of organizations (Meyer & Rowan, 1977). This occurs as the outward communications of an organization highlight change and reform—​ commonly in response to criticisms and challenges—​while in fact the underlying practices show limited or transient evidence of change. Since evidence of decoupling can threaten legitimacy and resources on which organizations are dependent (Powell & DiMaggio, 1991), they attempt to mask this apparent disjuncture by engaging in organized hypocrisy (Brunsson, 1989, 1993). Indeed, it is necessary for organizational survival: “Without hypocrisy, one party or interest would be completely satisfied and all others completely dissatisfied. With hypocrisy, several parties and interests can be somewhat satisfied” (Brunsson, 2007, p. 116). In other words, to maintain their standing and relevance, organizations need to respond to pressures, but they may do so in a ceremonial—​rather than substantive—​manner (Meyer & Rowan, 1977; Scott, 2014). These insights from organizational sociology have been introduced to the study of international public bureaucracies, which face a range of pressures from their principals (states), organizational environments, internal bureaucratic or technocratic agendas, and—​occasionally—​the public. Catering to contradictory demands would obfuscate bureaucratic action, therefore a degree of hypocrisy has come to be anticipated by observers of these organizations (Bukovansky, 2010; Lipson, 2007; Mundy & Menashy, 2014; Weaver, 2008). These accounts have distinguished between different gradients of hypocrisy: from blatant violations of established policies, to greyer areas where rhetoric is inadequately, superficially, or haphazardly translated into practice. Yet despite the widespread practice of organized hypocrisy, intergovernmental organizations are highly sensitive to such accusations, insofar as these can erode their legitimacy (Seabrooke, 2010).

22  40 Years of Structural Adjustment Two responses can be deployed by intergovernmental organizations to defend against such challenges. One path involves accepting criticism in a way that necessitates corrective action. A World Health Organization (WHO) mishap is revealing of the reluctance of international organizations to follow this route. Responding to criticisms over its role in the 2014–​2016 West African Ebola crisis, the organization issued an uncharacteristically strongly-​worded statement: “We did not work effectively in coordination with other partners, there were shortcomings in risk communications and there was confusion of roles and responsibilities” (The Guardian, 2015). The initial reaction by observers was that “by using strong and clear language the WHO’s leadership had decided to take a new approach to dealing with the criticisms” (Grepin, 2015). This was not to last. Within hours, the WHO edited the passage into more circumspect language: “We have learned the challenges of coordination. We have learnt to recognise the strengths of others, and the need to work in partnership when we do not have the capacity ourselves” (WHO, 2015b). This move from a blunt acceptance of organizational failure to bland, unobjectionable language that could be used for almost any situation reveals the perceived delegitimizing effect that the first statement would have had. Senior officials took down the statement due to concerns that honesty about shortcomings would intensify challenges. The alternative to honesty is to respond to criticisms and accusations by adding a layer of hypocrisy. In what follows, we examine the latter process, whereby maintaining myths about actual practices over extended periods of time necessitates intensification of efforts. We identify this layering process as an escalating commitment to organized hypocrisy: rhetorical commitments are overlaid with a rebranding of activities, issuing factsheets showcasing a biased and/​or inaccurate assessment, or instituting new policies that are never adequately incorporated into organizational practice. This path of action entails considerable sunk costs and explains behavior-​persistence (Simon, 1997), since backtracking after years (or decades) of ceremonial commitment to a goal can result in significant reputational damage and legitimation crises. To support this argument, we revisit a key debate surrounding the practice of IMF conditionality. In particular, we examine the relationship between conditionality and development policy space. We understand policy space as a government’s ability to select the policy instruments through which they address their economic problems, free from coercive conditionality. Insofar as IMF adjustment programs are indicators of underlying economic trouble, policy space is already expected to be somewhat limited. But even under constraining economic conditions, policy options remain. According to our understanding, then, governments with IMF programs have greater policy space when they can choose exactly how to mend external or budgetary imbalances, but less

The Evolution of IMF Conditionality  23 policy space when conditionality specifies the policy instruments by which they must address these imbalances (for instance, via increasing value-​added tax or privatizing natural resources). Our understanding of development policy space applies to countries at all income levels that are under distress and IMF supervision, and is similar to that of other scholars in this field (Chang, 2006; Cooper, 1968; DiCaprio & Gallagher, 2006; Grabel, 2011; Wade, 2003). Like these accounts, we acknowledge that open economies do not have absolute policy autonomy, as they need to respond to changes in the global economic environment. Consequently, the question is not whether or not these countries have to adapt, but the terms and conditions in which they do so.

Mission Creep Among the IMF’s various activities—​including collecting and publishing data, conducting surveillance of its members’ economic policies, providing technical assistance, and carrying out research (Babb & Kentikelenis, 2018)—​ lending programs to countries in economic trouble have attracted the most attention. In exchange for financial support, borrowing countries agree to implement a package of obligatory policy reforms, or conditionality, phased over one or more years. In turn, the implementation of conditionality is assessed on a quarterly or biannual basis and determines the disbursement of IMF funds in tranches. The content of conditionality has been among the most controversial outputs of any intergovernmental organization (Williamson, 1983). Reviewing the full range of debates around these issues is beyond the scope of this chapter and available elsewhere (Babb & Carruthers, 2008; Dreher, 2009; Guitián, 1995; Kapur, 2005; Nelson, 2014a, 2018; Nowzad, 1981; Spraos, 1986; Stubbs & Kentikelenis, 2018a; Vreeland, 2003; Williamson, 1983), and debates surrounding the social consequences of conditionality are covered in Part II of this book. Here, we revisit a persistent controversy: mission creep in the IMF’s policy advice. The IMF’s purpose, as stated in its Articles of Agreement, is to provide member-​states “with opportunity to correct maladjustments in their balance of payments without resorting to measures destructive of national or international prosperity.” Excluding a short period after its establishment, the IMF has practiced conditional lending in pursuit of this objective. For the first four decades of its operations, conditionality included a set of predictable reforms that focused almost exclusively on budget deficit reductions, restrictive monetary policy, and exchange rate devaluations (Dell, 1981, 1982; Diaz-​Alejandro, 1981; Williamson, 1983); that is, areas understood to be covered by the IMF’s mandate. At the same time, the IMF was mandated to remain neutral vis-​à-​vis its borrowers’ economic

24  40 Years of Structural Adjustment and social objectives—​an approach known as the “doctrine of economic neutrality” (Finch, 1983). For example, according to this approach, while the IMF provided support for a country’s macroeconomic adjustment efforts, “how the government brought down the deficit, by raising taxes or cutting expenditure, and the particular taxes or expenditure at issue [remained] the government’s responsibility” (Polak, 1991, p. 39). In other words, IMF programs did not make explicit attempts—​at least via conditionality—​to change the underlying structure of borrowing countries’ economies. However, in the 1980s, responding to political pressures and new economic ideas, the IMF’s programs expanded in policy content to include a range of “structural” conditions, moving well beyond what the IMF’s founders had prescribed in the Articles of Agreement (Babb & Buira, 2005; Kentikelenis & Babb, 2019). The era of so-​called structural adjustment saw the involvement of the IMF in sensitive policy areas, such as the privatization of state-​owned enterprises, trade and financial liberalization, and economic deregulation (Summers & Pritchett, 1993; Toye, 1994; Williamson, 1990). Over subsequent years, the remit of conditionality expanded even further to cover a growing array of policy areas, including social policy, labor market reforms, and “good governance” (Babb, 2013; Chang, 2006; Serra & Stiglitz, 2008; Stiglitz, 2002). Scholars understood the widening number of policy areas that IMF-​prescribed policies covered—​the scope of conditionality (Stone, 2008)—​as “mission creep” (Babb & Buira, 2005). These arguments suggest that the IMF’s conditionality moved beyond its core mandate of economic issues and “into areas that properly belong in the realm of politics” (Stiglitz, 2002, pp. 44–​45), thereby challenging national sovereignty and domestic autonomy to design policy (Babb & Carruthers, 2008; Krasner, 1999; Przeworski & Vreeland, 2000). In particular, after the organization’s handling of the Asian financial crisis of the late 1990s, a growing chorus of academics and policymakers strongly criticized the organization for advocating reforms in many and disparate policy areas, frequently with little direct relevance to the IMF’s core areas of expertise (Feldstein, 1998; Goldstein, 2001; Meltzer, 2000; Radelet & Sachs, 1998; Stiglitz, 2002). Following this torrent of high-​profile criticism, the IMF itself acknowledged the sprawling of conditionality in an excessive number of policy areas and embarked on a long process of streamlining (IMF, 2001c). New lending programs would, in the IMF’s words, afford “policy space” to governments as long as they reached “the standard that members’ policies must meet in order to qualify for Fund support” (Koeberle et al., 2005, p. 35). The IMF intended new principles underpinning the design of conditionality to mark a break from the past: “ownership” of reforms, “tailoring” to country specificities, “parsimony” and “clarity” in conditionality, and “coordination” with other international organizations (Koeberle et al., 2005). Yet beyond rhetoric, the IMF’s Independent

The Evolution of IMF Conditionality  25 Evaluation Office—​established by the IMF’s executive board in 2000 to provide impartial assessments of the organization’s operations (Lissakers et al., 2006)—​ raised doubts about the extent of genuine change in lending operations, as conditionality remained “very detailed, not obviously critical, and often felt to be intrusive” (IEO, 2007a, p. vii). Responding to the new round of criticisms, the IMF produced a series of self-​congratulatory reports highlighting the apparent overhaul of past practices: “Programs are helping countries to weather the worst of the [global financial] crisis, [ . . . and] program design has learned the lessons of the past” (IMF, 2009b, 2009f, pp. 43–​44). Indeed, “flexibility” was the term touted in IMF reports to describe its policy advice in the early crisis period (IMF, 2009b, 2009a, 2009f, 2009d, 2009c). Continuing on the same theme, in 2012, the organization’s review of conditionality—​the clearest articulation of IMF staff ’s house view—​concluded that “program conditionality has generally been appropriately streamlined, even-​handed, and tailored to country needs” (IMF, 2012b, p. 4, 2012a). In the IMF’s most recent conditionality review, the organization conceded a rise in the number of structural conditions had occurred during the 2010s, but that this “reflect[ed] the rising structural challenges [ . . . and] remained largely focused on the Fund’s core areas of responsibility” (IMF, 2019a, p. 2). In short, current IMF rhetoric conceives of recent programs as—​for the most part—​parsimonious, flexible, and within the organization’s core areas of expertise, thereby addressing past criticisms over its own mission creep and its undue restrictions on borrowing countries’ policy space.

Development Policy Space Under IMF Programs Consistent across IMF programs has been the use of conditionality as a compliance mechanism, as well as the free-​market orientation of these policies (Babb & Carruthers, 2008; Greer, 2014; Kentikelenis, 2017; Pfeiffer & Chapman, 2010). Yet experiences of conditionality also diverge in important ways across time and space. To investigate the evolution of conditionality, we draw on IMF archival documents covering the years 1980 to 2019. In total, we searched 6,100 loan-​ related documents to extract 65,707 conditions spread across 132 countries. These documents consisted of IMF staff reports, national governments’ Letters of Intent, and accompanying Memoranda of Economic and Financial Policies, which specify conditionality. Documents are generally updated with additional conditions on three-​or six-​month review cycles (although delays are common), and the successful conclusion of reviews determines the disbursement of loan tranches. In this section, we focus on the number of conditions applicable per country-​year; we discuss methodological issues related to data collection in the appendix.

26  40 Years of Structural Adjustment Before presenting our findings, a word of caution is warranted. Reporting the number of conditions, while revealing, is an imperfect measure of conditionality because it tells us nothing about the difficulty of implementing any particular condition. For example, a condition stipulating the privatization of a state-​ owned enterprise is qualitatively distinct from one introducing a value-​added tax. Despite this, previous studies have shown the measure is a suitable proxy for the intrusiveness of conditionality (Beazer & Woo, 2016; Caraway et al., 2012; Chwieroth, 2015; Copelovitch, 2010b; Dreher et al., 2009, 2015; Dreher & Jensen, 2007; Dreher & Vaubel, 2004b; Gould, 2003; Rickard & Caraway, 2014), and it has been fruitfully employed in the IMF’s own studies (Bulír & Moon, 2004; Ivanova et al., 2006). We proceed with this indicator to yield inferences about important aspects of the IMF’s policy advice: the evolution of conditionality, changes to the use of so-​called structural conditions, and patterns in the breadth of policy areas targeted by IMF conditionality. In Figure 2.1, we present a world map showing countries’ total number of conditions applicable in all IMF loans between 1980 and 2019. The countries shaded in the darkest color have the highest number of conditions over the period, those where the total number of conditions is greater than 1,050. The countries shaded in the lightest color have the lowest conditions, at 350 or less. There are also two shades of color in-​between: a somewhat lighter shading covering countries with conditions between 350 and 700, and a somewhat darker shading covering countries with conditions between 700 and 1,050. Last, there are countries without any shading, reflecting no conditions at all during the period. Armenia, Kyrgyzstan, Malawi, Pakistan, Romania, and several West African countries emerge as countries with highest overall conditionality burdens: all had repeat loans that carried a high degree of conditionality. Furthermore, we observe widely diverging experiences with IMF conditionality across countries. For example, Mauritania had IMF programs active for 30 of the 40 years covered, carrying a total of 1,175 conditions. Other countries had only brief encounters with the Fund, which is also reflected in relatively limited conditionality. For instance, South Africa only had a one-​year Stand-​by Arrangement—​ the IMF’s staple lending facility—​with 11 conditions attached between 1982 and 1983. Lithuania, with 417 conditions, held the median number of conditions out of countries that had at least one. Most advanced nations did not have any conditions during the period covered because they did not borrow from the IMF, although Cyprus, Greece, Iceland, Ireland, and Portugal are notable exceptions. Table 2.1 offers some indication of how conditionality has evolved in the period covered by our data. Besides the mean and median number of conditions, also listed are the number of countries participating in an IMF program and the three countries with the highest number of conditions for each year. It shows that the median number of conditions in IMF programs gradually increased from

Source: Authors’ dataset

Figure 2.1  Total conditions, 1980–​2019

Number of conditions 1050–1400 700–1050 350–700 1–350 None

28  40 Years of Structural Adjustment Table 2.1  Descriptive statistics on conditionality Year

Mean

Median Programs Highest conditionality in . . .

1980

12.1

12

33

Bolivia (28), Turkey (25), Congo Dem. Rep. (21)

1981

11.6

12

41

Madagascar (27), Turkey (22), Jamaica (21)

1982

11.9

12

36

Jamaica (27), Uganda (23), Cote d’Ivoire (20)

1983

14.7

15

53

Dominican Republic (33), Argentina (30), Turkey & Jamaica (28)

1984

18.6

18

41

Jamaica (50), Brazil (49), Cote d’Ivoire (32)

1985

20.7

23

35

Jamaica (42), Argentina (36), Ghana (32)

1986

21.1

18.5

42

Morocco (65), Zambia (63), Congo Dem. Rep. (40)

1987

27.3

28

35

Tanzania (49), Niger (42), Tunisia, Senegal & Gabon (40)

1988

25.3

24

45

Tanzania (57), Togo (56), Tunisia (49)

1989

28.8

30

50

Pakistan (87), Nepal (56), Tunisia (52)

1990

26.6

25

49

Pakistan (61), Tanzania (59), Gabon (54)

1991

27.0

26.5

52

Tunisia (52), Rwanda & Senegal (46)

1992

29.9

29

53

India (62), Gabon (56), Pakistan (54)

1993

27.1

28

51

Mauritania (73), Egypt (56), Burkina Faso (53)

1994

34.3

34

65

Mauritania (76), Kyrgyz Republic (70), Albania (66)

1995

36.8

37

67

Mauritania (88), Ukraine (78), Armenia (72)

1996

43.9

42.5

68

Azerbaijan (93), Russian Federation (87), Georgia (70)

1997

40.8

40.5

64

Kazakhstan (99), Bulgaria (89), Azerbaijan (88)

1998

43.0

42.5

62

Russian Federation (143), Ukraine (103), Indonesia (77)

1999

43.0

39.5

64

Ukraine (148), Bulgaria (99), Moldova (89)

2000

40.4

42.5

66

Kyrgyz Republic (97), Bulgaria (87), Romania (79)

2001

42.0

40

66

Pakistan (105), Turkey (80), Armenia, Rwanda & Ukraine (72)

2002

42.5

40

59

Romania (114), Pakistan (98), Turkey (86)

2003

45.9

48

57

Romania (114), Pakistan (87), North Macedonia (80)

2004

43.4

47

59

Romania (126), Serbia (87), Nicaragua (74)

The Evolution of IMF Conditionality  29 Table 2.1 Continued Year

Mean

Median Programs Highest conditionality in . . .

2005

44.5

43

47

Serbia (122), Romania (100), Senegal (83)

2006

39.7

41

43

Cameroon (79), North Macedonia & Congo Rep. (65)

2007

39.4

39

38

Cameroon (84), Dominican Republic (77), North Macedonia (67)

2008

29.2

31.5

48

Cameroon (75), Moldova (62), Haiti (60)

2009

31.9

33

53

Cote d’Ivoire (70), Afghanistan (53), Central African Republic (51)

2010

32.3

34

61

Cote d’Ivoire & Ghana (60), Tajikistan (58)

2011

32.2

32

51

Haiti (60), Tajikistan (58), Ghana (54)

2012

32.2

35

45

Greece (62), Afghanistan (60), Cote d’Ivoire (59)

2013

34.1

34.5

40

Bosnia and Herzegovina (85), Bangladesh (69), Cote d’Ivoire (68)

2014

36.1

43

37

Bosnia & Herzegovina (92), Jamaica (63), Ukraine & Greece (61)

2015

31.2

32.5

40

Serbia (65), Jamaica (62), Pakistan (60)

2016

33.9

32.5

40

Ghana (67), Serbia (61), Bosnia and Herzegovina (60)

2017

35.2

37.5

44

Bosnia and Herzegovina (92), Ghana (75), Jamaica (65)

2018

38.6

44

38

Bosnia and Herzegovina (97), Cameroon (68), Ghana (64)

2019

34.8

38

41

Burkina Faso (67), Cameroon (56), Malawi (54)

Total

33.2

32

Pakistan (1,303), Romania (1,272), Ghana (1,246)

12 per calendar year in the early 1980s, reaching a median of circa 42 conditions applicable per year by 1996, where it remained until 2007. Subsequently, we observe a reduction in the median of conditions to about 32 in the immediate aftermath of the Global Financial Crisis. In the most recent decade, the median has been fluctuating year-​on-​year between 32 to 44 conditions. Despite the apparent drop in the median number of conditions since the mid-​2000s, we note an increase in high conditionality programs. Afghanistan, Bangladesh, Bosnia, Burkina Faso, Cameroon, Cote d’Ivoire, Ghana, Greece, Haiti, Jamaica, Pakistan, Serbia, and Ukraine have all carried 60 or more conditions at least once since 2009. In 2018, the median of conditions jumped to 44, the third highest such

30  40 Years of Structural Adjustment figure in our dataset after 2004 (median of 47 conditions) and 2003 (median of 48 conditions). Moreover, several Eastern European and Central Asian countries represent high-​lying outliers, peaking at over 140 conditions for Russia in 1998 and Ukraine in 1999. Despite the fact that most postcommunist countries started implementing IMF reform programs in the early 1990s, only in the latter half of the decade did conditionality expand greatly in degree and scope. This evidence documents a slight reduction in IMF program conditionality over the period following the onset of the Global Financial Crisis, even though it remained considerably higher than the 1980s and early 1990s. The most recent data reveal a rising trend for the burden of conditionality since 2008. At the same time, while we notice this increase in conditionality, there has also been a marked decline in the number of lending programs (new or in progress). A spike in postcrisis lending was followed by a decline in demand for IMF services. Of the 41 lending programs active in 2019, 23 were to sub-​Saharan African countries. Lending to Europe and Central Asia, Latin America and the Caribbean, and East Asia and the Pacific—​common in the past—​has declined considerably. Only 11 countries from these three regions had IMF programs in 2019, compared to 35 countries in 2000.

The Rise, Fall, and Return of Structural Conditionality The technical apparatus of conditionality has also evolved over time, with direct implications for the policy space available to borrowing countries. To determine the disbursement of loan tranches, IMF lending programs make use of two types of conditions: quantitative and structural (Bird, 2009; IMF, 2015a). Originally, IMF conditions only appeared in the form of quantifiable macroeconomic targets (e.g., limits to government borrowing). While such conditions still form the bulk of conditionality up to the present, they only specify the policy ends rather than the means. That is to say, although quantitative conditions may be overly restrictive, in theory governments can pursue a range of alternative policies to meet them. In contrast, structural reforms clearly specify means that contribute to meeting the macroeconomic targets and other objectives. In this section, we examine evidence on structural conditions; that is, IMF-​ mandated reforms aimed at “reducing or dismantling government-​imposed distortions or putting in place various institutional features of a modern market economy” (Goldstein et al., 2003, p. 366). Such reforms have commonly aimed at altering the underlying structure of an economy; for instance, by privatizing state-​owned enterprises, legislating central bank independence, deregulating labor markets, or restructuring tax systems. These conditions can be distinguished vis-​à-​vis their strictness. Binding conditions—​known as prior actions

The Evolution of IMF Conditionality  31 and structural performance criteria—​form sine qua non criteria for continued financial assistance, and apply to areas that IMF staff consider “crucial to the success of the program” (Leckow, 2002, p. 3). In cases of nonimplementation, countries must request a waiver by the IMF’s executive board, an action that could damage countries’ reputation in international markets (Bird & Rowlands, 2002; M. Edwards, 2005; IMF, 2009d; Mody & Saravia, 2006; Reinsberg, Stubbs et al., 2021a, 2021b). In contrast, nonbinding structural conditions—​known as benchmarks—​are also envisaged to apply only to policy areas critical to the objectives of lending programs, but rely on the discretionary assessment by staff where nonimplementation occurs rather than needing a waiver from the executive board (Goldstein, 2001; Guitián, 1995). As shown in Figure 2.2, structural conditionality has evolved over time. Let us focus initially on the grey bars, which plot the total number of structural conditions from 1980 to 2019. They were initially incorporated into programs at very low levels in the mid-​1980s. Within a decade, the use of such policies expanded rapidly and peaked at about 16 structural conditions on average in 1998 and 1999. That peak also marked the height of criticisms of the IMF’s policy advice, and eventually resulted in the launch of the streamlining initiative (IMF, 2001c). Since then, the mean of structural conditions per year remained around 12, followed by a sharp drop in 2008, partly explained by the phasing out of structural performance criteria (IMF, 2009d), indicated by the dotted line. The latter 20

Number of conditions

15

10

5

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

0

Total

Prior Actions

Structural Performance Criteria

Structural Benchmarks

Figure 2.2  Mean of structural conditions in IMF programs Source: Authors’ dataset

32  40 Years of Structural Adjustment was followed by an upsurge in structural benchmarks, shown by the dashed line. The trend for these benchmarks has been to increase in the last decade (see also IEO, 2014). In 2018, more structural benchmarks were included in IMF programs than in any other year. Another development has been the gradual reduction of prior actions—​the strictest type of condition represented as the solid line—​between 1999 and 2007. In 2012, however, we observe a reversal of the downward-​sloping trend. Overall, the total number of structural conditions reached 11.2 in 2018, equivalent to the mean of the 2001–​2007 period. These trends suggest that structural adjustment is not a policy fad of the past with no relevance to contemporary IMF practices. The organization’s programs still incorporate a considerable number of structural conditions, and the total number of such conditions still far exceeds that observed in the pre-​1994 period. The emphatic return of structural conditionality in recent years calls into question the IMF’s “we do not do that anymore” rhetoric.

Scope of Conditionality While trends in the total number or type of conditions are suggestive, assessing IMF conditionality also requires exploring their distribution in different policy areas—​the so-​called scope of conditionality (Stone, 2008). Such metrics speak directly to issues of mission creep discussed above, as critics argue that IMF conditionality has veered into an ever-​growing number of policy areas. We identified the scope of reforms by classifying the policy areas of all conditions for the period 1980 to 2019 into mutually exclusive categories, summarized in Table 2.2. Following conventions in the literature, these are separated into core and non–​core policy areas (IEO, 2007b). Core policy areas include the following: external debt issues; financial sector, monetary policy, and Central Bank issues; fiscal issues, revenues, and taxation; and external sector (trade and exchange system). Together, core policies account for 86.7 percent of the total number of conditions. The remaining 13.3 percent of conditions pertain to non–​core policy areas: state-​owned enterprise privatization, reform, and pricing; labor issues (public and private sector); institutional reforms; and poverty reduction policies. What do such conditions look like in practice? We take the example of Cameroon, which entered into a 36-​month program in end-​October 2005, to illustrate how IMF conditionality became part of our data collection and coding effort. The program featured a wide array of quantitative performance criteria, including, inter alia, floors on the non–​oil primary budget balance (classified as “fiscal issues, revenues, and taxation”), ceilings on new medium-​and long-​ term nonconcessional external debt contracted or guaranteed by the central

The Evolution of IMF Conditionality  33 Table 2.2  Categorization of policy areas

Core policy areas

Non–​core policy areas

Policy area description

Number of conditions

External debt issues Debt management and external arrears

18,056

Financial sector, monetary policy, and Central Bank issues Financial institution regulation, financial SOE privatization, treasury bills, interest rates, Central Bank regulation, money supply, and domestic credit

16,490

Fiscal issues, revenues, and taxation Expenditure administration, fiscal transparency, audits, budget preparation, domestic arrears, and fiscal balance, customs administration, tax policy, tax administration, and audits of private enterprises

16,764

External sector (trade and exchange system) Foreign reserves, trade liberalization, exchange rate policy, capital account liberalization, and foreign direct investment

5,659

State-​owned enterprise privatization, reform, and pricing Nonfinancial SOE privatization (including liquidation and bankruptcy proceedings), SOE restructuring, subsidies, price liberalization, audits, marketing boards, and corporatization and rationalization

3,721

Labor issues (public and private sector) Wage and employment limits, pensions, and social security institutions

2,129

Institutional reforms Judicial system reforms, anti-​corruption measures, competition enhancement, private sector development, devolution, sectoral policies, social policies (excluding poverty reduction policies), price increases for food, water, public transport, or other basic needs goods, land registries, granting of property rights, environmental regulations, and access to commons

1,582

Poverty reduction policies Poverty Reduction Strategy Paper development, increases in social sector spending, and implementation of social safety nets

1,306

Total number of conditions

65,707

government (“external debt issues”), and ceilings on the increase in net claims of the banking system on the central government (“financial sector, monetary policy, and Central Bank issues”) (IMF, 2005a). The program also incorporated structural conditions spanning a variety of policy areas, such as the finalization

34  40 Years of Structural Adjustment and adoption of a privatization strategy of Cameroon Telecommunications (“state-​owned enterprise privatization, reform, and pricing”), preparation of a diagnostic study on the civil service remuneration system (“labor issues”), the publication of judicial decisions and sanctions aimed at combating corruption on the government’s website (“institutional reforms”), and the implementation of an electronic one-​stop shop to facilitate foreign trade (“external sector”) (IMF, 2005a, 2006c, 2007a). To investigate how the policy mix of conditions in IMF programs evolved over time, Figure 2.3 shows the percentage of conditions in non–​core policy areas per year. Let us focus initially on the overall length of the bar for each year, which represents the sum of each of the four non–​core policy areas by stacking them on-​top of each other. An initial observation is that the expansion of conditionality into non–​core areas began in the mid-​1980s. This was prompted by the push of the United States for programs to include a structural, supply-​side orientation and the associated introduction of the Structural Adjustment Facility (Babb, 2009; Boughton, 2001; Toye, 1994). The proportion of conditions in non–​core areas peaked in 1999 at about 20 percent of conditions, fluctuated for several years, and then dipped markedly from 2005 to 2009, before gradually returning to peak levels by 2019. We examine the evolution of IMF-​mandated reforms in different policy areas in turn, supplementing the quantitative data with additional country-​specific evidence from recent IMF conditionality.

Percent of conditions

20

15

10

0

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

5

SOE reforms

Labour issues

Institutional reforms

Poverty reduction policies

Figure 2.3  Conditionality in non–​core policy areas Source: Authors’ dataset

The Evolution of IMF Conditionality  35 First, a recurring controversy surrounding IMF programs concerns their purported bias against state-​owned enterprises (SOEs), considered as prone to mismanagement and inefficiencies (see Toye, 1994). These are represented by the dark-​grey portion of each bar in Figure 2.3. Reforms in this sector emerged in the early 1980s but entailed less than 3 percent of all conditions until 1987. From the late 1980s to early 1990s, they composed about 5 percent of all conditions. Consistent with the timing of the postcommunist transitions (and the IMF’s heavy involvement in the process), SOE-​ related reforms then increased throughout the mid-​1990s, reaching a height of about 10 percent of conditions by the late 1990s. In the most recent decade, about 4 to 6 percent of conditions stipulate SOE reforms and privatizations. On the one hand, these declines could correspond to a narrowing of the scope of conditionality, as per IMF rhetoric (IMF, 2012a). On the other hand, it may also be the case that after decades of countries corporatizing, privatizing, or liquidating publicly-​owned firms at the behest of the IMF (Brune et al., 2004), there is less room for further such reforms. Recent key examples of SOE-​related reforms include the adoption of a privatization strategy in Congo-​Brazzaville (IMF, 2019h), the privatization of Serbian electricity, gas, railways, and road companies (IMF, 2015c, 2016b), and a range of reforms to government enterprises in Barbados, Pakistan, and Sri Lanka (IMF, 2016d, 2018a, 2019g). Beyond these cases, IMF conditionality in Eurozone countries has also mandated privatizations or SOE reforms. Second, the IMF has long taken an interest in so-​called institutional reforms aimed at promoting a “business-​friendly” environment, notwithstanding doubts over whether certain such arrangements are desirable or exportable (Stiglitz, 2002). In Figure 2.3, these are indicated by the mid-​grey portion of each bar. Conditions in this broad policy area covered reforms to legal systems, competition law, business regulation, environmental policies (including natural resources), and social protection issues. They became more common during the early 1990s and remained at around 3 to 4 percent of conditions from 1995 up to 2007, before dipping to about 1 to 2 percent of conditions after the Global Financial Crisis, as showcased in the IMF’s review of conditionality (IMF, 2012a). Since 2012, institutional reforms have consistently made up 2 to 3 percent of conditions. For instance, the 2014 Tunisian adjustment program mandated passing a “decree for implementing the new investment code in line with the objective of protecting market access, reducing restrictions on investments, and rationalizing of incentives” (IMF, 2014j). Similar reforms were designed for Cote d’Ivoire, Guinea, Grenada, Niger, Suriname, and Tunisia. Ukraine’s recent experience of conditionality also stipulated that “the government will adopt an action plan to eliminate, streamline, simplify and clarify the legislative and regulatory frameworks governing economic activity” (IMF, 2014k). Likewise, the Armenian parliament was asked to adopt regulatory reforms in a range of policy

36  40 Years of Structural Adjustment areas, including entrepreneurship, customs, and social issues (IMF, 2014a). Most recently, the Honduran government was called upon to submit legislation requiring regulatory agencies to streamline requirements for obtaining business permits (IMF, 2019e). Third, labor issues are captured by the light-​grey portion of the bars in Figure 2.3. After a period of relative indifference until the early 1990s, where such conditions never reached more than 2 percent of all conditions, labor-​ related conditions reached between 4 to 6 percent over the 1994–​2007 period. A peak for these types of reforms is noted in 2005, incorporating 6.1 percent of all conditions, primarily reflecting conditions limiting the public sector wage bill. Since then, labor-​related conditionality has declined, with circa 1 to 3 percent of conditions advocating such reforms in the past decade. According to IMF staff, wage bill ceilings have been discontinued (Gupta, 2015), and labor market policies are peripheral to many of its programs (IMF, 2014h). At a surface level, our findings reflect these statements. Yet a close examination of IMF conditionality in developing countries reveals that the IMF’s involvement on issues directly affecting labor is decidedly more than peripheral, despite claims to the contrary. For instance, wage bill ceilings—​while used less than in the past—​have still been part of IMF programs in Ghana (2015–​2018), Honduras (2014–​2017), and Moldova (2016–​2019). With the exception of Ghana’s binding conditions, these were incorporated into IMF programs in a nonbinding form (benchmarks) but were still considered by IMF staff as critical enough for inclusion into conditionality. Furthermore, Serbia’s IMF-​designed labor-​related reforms included “a reduction of at least 14,500 permanent employees” in 2015 (IMF, 2015c), as well as further policy changes that affected civil servants (IMF, 2016b). Civil service restructuring was also required in, inter alia, Burkina Faso and Liberia (IMF, 2018b, 2019f). Moreover, a number of countries’ adjustment programs included extensive pension reforms. Romania’s program targeted pensions, including a 15 percent cut and a number of further changes to the system that would reduce payouts and raise the retirement age (IMF, 2010h). The measures proved controversial (Reuters, 2010), and were struck down by the country’s constitutional court (IMF, 2010h). Yet a few months later, the same reform package was re-​introduced as a binding prior action stipulating the “parliamentary approval of pension reform legislation” (IMF, 2010g), and was eventually passed despite objection from the Romanian president (IMF, 2010e). Similarly, Serbia’s conditionality introduced a pension freeze between 2009 and 2011 (IMF, 2009e, 2010f), and increased retirement ages (IMF, 2010d). More recently, cost-​cutting pension reforms were required in Barbados and Tunisia (IMF, 2018a, 2018d). The IMF programs in Eurozone countries have also relied on labor-​related reforms, including on deregulation, government wage spending, and social security systems. For

The Evolution of IMF Conditionality  37 instance, Greece’s conditionality stipulated extensive labor market liberalization. The country’s program included reforms—​often as prior actions—​to the collective bargaining system, the precedence of firm-​level (as opposed to sectoral) agreements, and the reduction of minimum wages and employee dismissal costs (IMF, 2010c, 2010b, 2012c, 2014c). Similarly, Portugal’s adjustment program stipulated increases to the retirement age, weakening of collective bargaining, and the introduction of a “public administration labor law that will aim at aligning current public employment regime to the private sector rules . . . , and termination of tenure” (IMF, 2012d, 2013a, 2014e). Finally, a notable development has been the rapid rise in the inclusion of poverty reduction conditions, which typically take the form of government spending minima on health and education. These are shown as black portions of the bars in Figure 2.3. While these comprised less than 0.1 percent of all conditions until 1995, since then the use of these instruments has steadily increased. The IMF has long claimed that it is concerned about the social consequences of its policy advice (Camdessus, 1998; de Larosiere, 1986; de Rato, 2006), and that its programs include adequate safeguards to ensure the protection of the poor (Gupta et al., 2000; IMF, 1995a, 2008, 2013c, 2014g, 2015b). Indeed, poverty reduction was recognized in the name of the organization’s standard lending facility for low-​ income countries, the Poverty Reduction and Growth Facility (until 1999 known as the Enhanced Structural Adjustment Facility, and since 2009 known as the Extended Credit Facility). Countries receiving financial support through this instrument are required to produce Poverty Reduction Strategy Papers (IEO, 2004), and the design of conditionality also incorporated explicit measures to maintain or increase social expenditures. In the past decade, the IMF has expended even greater attention on poverty reduction (IMF, 2017c, 2018c, 2019b), and poverty conditions have risen from 3.9 percent in 2010 to 8 percent of conditions in 2019. This rapid expansion of “pro-​poor policies” is often welcomed by the IMF’s critics as a positive development. However, as we document in Chapter 4, there is reason for skepticism: our findings suggest that such policies have been inadequately and haphazardly integrated in program design, are often not implemented, and have been used by the IMF as a smokescreen to deflect criticism. In sum, the evidence presented in this section reveals that only limited progress has been made in trimming down conditionality. While IMF programs are not as all-​encompassing as they were during the 1994–​2007 period, still more than 10 percent of conditions stipulate reforms to state-​owned enterprises, labor issues, and institutional environments; and poverty reduction policies—​even if welcome—​now make up 8 percent of conditions alone. The decline in the scope of conditionality vis-​à-​vis peak years for state-​owned enterprise and institutional reforms are, in part, explained by the years a country has spent under IMF

38  40 Years of Structural Adjustment programs. Repeat borrowers have already had to implement extensive conditionality over the years prior to the crisis, and—​as a result—​there is less need from the IMF’s point of view for such conditions to be introduced in recent programs. Furthermore, the brief overview of recent IMF conditionality on social protection and labor issues suggests that the organization’s claims of fundamental changes are exaggerated—​a theme we return to in subsequent chapters. Finally, the recent incidence of high-​conditionality programs and the increases in structural conditionality across IMF loans represent a key discrepancy between rhetoric and practice by the organization.

Conclusion In this chapter, we used our new data on IMF conditionality to revisit a key debate over the organization’s policy advice to its borrowers: whether it was engaging in mission creep through the widening number of policy areas that IMF-​prescribed policies covered. We showed that the IMF engages in the production and maintenance of myths about its actual practices—​or organized hypocrisy—​occurring through the rebranding of existing practices and the addition of token gestures to placate critics, without altering the underlying premises of reform design. Our findings suggest that the IMF’s claim that programs now “creat[e]‌policy space” by exhibiting “responsive design and streamlined conditionality” (IMF, 2009b, p. 1) is not an accurate representation of reality. The period of organizational insecurity preceding the Global Financial Crisis was linked to changes in the IMF’s policy prescriptions, reflected in declines in conditions attached to its loans. However, as the crisis progressed, the IMF re-​emerged as a central organization dealing with the policy response, and—​following this boost in organizational self-​confidence—​the downward trend in conditionality reversed. Reflecting on the IMF’s own claims to have changed, Financial Times columnist Martin Wolf (2011) had a pessimistic premonition that the “new, cuddly” approach “can’t possibly last”—​this is confirmed by our data. In particular, structural conditionality has returned as a key component of IMF programs, and its scope has been widening in recent years. In short, we showed how the maintenance of business-​ as-​usual practices remained tenable by adding ever-​more layers of ceremonial reforms and rhetoric. We note three limitations of our argument in this chapter. First, we have purely examined the “supply” of conditionality and weighed it against rhetoric. It is possible that some governments requested specific policy conditions to overcome domestic opposition to reforms (Güven, 2012; Vreeland, 2003), and that some programs were interrupted or not fully implemented (Arpac et al., 2008). Second, we focus our analysis mostly on the structural reform component of IMF

The Evolution of IMF Conditionality  39 programs. The IMF’s macroeconomic policy advice—​for example, on the speed or extent of fiscal consolidation—​is equally important. This has been explored by other recent studies that found evidence of pro-​cyclical policy advice that reduced fiscal policy space for governments (Grabel, 2011; Ortiz & Cummins, 2019; Weisbrot et al., 2009). Finally, conditionality represents only the most visible set of IMF-​induced policy constraints facing the organization’s borrowers. IMF programs also include noncoercive advice in the form of technical assistance or policy recommendations. The extent to which such advice is actually taken on-​board remains unclear. Our study has provided evidence of important gaps between rhetoric and practice in key aspects of the IMF’s lending activities. Intergovernmental organizations produce hypocrisy to safeguard their legitimacy, gain access to resources, and placate critics. As they do so, they become locked-​in on a specific course of action, which in turn yields a layering process of ceremonial reforms. The IMF’s involvement in low-​income countries illustrates this point. In the late 1990s, following criticisms of the organization’s structural adjustment programs, the IMF rebranded its lending instrument as the Poverty Reduction and Growth Facility, and introduced social spending floors to allay concerns over insufficient attention to the social consequences of its programs. Then, in the early 2000s, the IMF’s rhetoric emphasized its commitment to creating policy space via streamlined conditionality. Finally, after the onset of the Global Financial Crisis, IMF programs were purportedly flexible and tailored to country needs, not bearing any semblance to structural adjustment policies advocated in the past, as Christine Lagarde’s quote in the introduction to this chapter suggests. These iterative layers of rhetoric and ceremonial reforms are—​in part—​attempts to maintain the myth of a fundamentally reformed institution. In Part II of this book, we present evidence that directly addresses the question of whether or not the IMF has engaged in organized hypocrisy in relation to social protection. Before then, we describe in the next chapter our methods for evaluating the effects of IMF conditionality.

3

How to Evaluate the Effects of IMF Conditionality In this book we wish to evaluate the effects of the policy reforms, or conditionality, attached to IMF programs on a series of social policies and outcomes. In attempting to analyze the effects of these conditions, there are a number of methodological difficulties that must be overcome. The most obvious conundrum is the fact that countries turn to the IMF for financial support precisely because they are experiencing major economic problems. It would come as no surprise, then, if countries participating in IMF programs are also facing deteriorating social conditions. To deduce from this observation that IMF programs adversely impact social outcomes would be, as James Vreeland (2003, p. 4) eloquently put it, “akin to concluding that aspirin causes headaches or that doctors hurt their patients.” People visit the doctor because they are sick. Countries call on the IMF because they are—​in economic terms—​sick. If we fail to account for the fact that the patient was initially feeling poorly, we may end up underestimating the effectiveness of the doctor’s treatment, or even conclude that the treatment hurts the patient. The same is true of the IMF. If we fail to account for the initial economic, political, and social characteristics of the country, then we might also misestimate the impact of the IMF’s treatment. Scholars have developed complex statistical solutions to mitigate this issue, techniques we consider in this chapter. Yet most of this work relies on a crude aggregated indicator of IMF intervention that measures whether or not a country is under a program in a given year. This means that all programs are treated statistically as if they are identical. Our data, introduced in Chapter 2, enables analysts to undertake disaggregated statistical tests of the impact of conditionality. Being able to perform these more fine-​grained evaluations matters. For borrowing countries, their policy mix of conditionality determines their mode of integration into the world economic system, and their ability to provide basic services to their population. Despite such far-​ranging implications, much debate has taken place on the basis of haphazard or inadequate empirical data. Accounting for IMF program heterogeneity, as we do throughout this book, is a step toward rectifying this gap in knowledge. Nevertheless, these data advances introduce new methodological challenges, and tackling them necessitates substantive updates to the established methods. A Thousand Cuts. Alexandros Kentikelenis and Thomas Stubbs, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/​oso/​9780190637736.003.0003

How to Evaluate the Effects  41 As an illustration, let us return to the doctor analogy. Imagine a patient enters the doctor’s office complaining of headaches, a sore throat, and a blocked nose. Following a physical examination, the doctor might diagnose the patient as suffering from a common cold and recommend painkillers and a few days of rest as treatment. Alternatively, a patient may arrive at the doctor’s office complaining of breathing troubles, chest pains, and spells of dizziness. The doctor might diagnose a serious illness, or even multiple illnesses, and prescribe a smorgasbord of powerful pills over the course of the year. When the IMF engages with a struggling economy, the same logic applies in principle: an economy facing multiple severe economic problems (serious illness) may receive more conditions (many pills) than one with relatively mild concerns. If we fail to control for the type and extent of these initial economic problems, then we might also misestimate the impact of the IMF’s conditions. To tackle these issues, we need to first grapple with a range of methodological questions and develop appropriate analytical strategies to appropriately evaluate the impact of IMF-​mandated policy reforms. Consequently, the discussion is highly technical, although—​hopefully—​still accessible. Readers less inclined to extensive treatments of quantitative methodologies can skip straight to the discussion of our preferred approach or even to the subsequent chapters containing the findings of our analyses. The chapter begins with a critical appraisal of existing approaches to studying the effects of IMF programs. The chapter then arrives at its central task: to develop a statistical strategy capable of testing the effects of IMF conditionality per se. In doing so, our approach must address the main sources of statistical bias. On the one hand, bias stems from the IMF participation decision. Countries select into programs based on an array of domestic and global economic, political, and social characteristics, some of which cannot be directly observed or measured. These characteristics may in turn influence the outcome of interest. On the other hand, bias is linked to conditions included in the program. The above characteristics may also determine whether a country gets particular conditions in greater or lesser degrees—​that is, the number of conditions applicable either in total, across condition types (i.e., quantitative or structural), or in a given policy area. The chapter concludes by describing the basic methodological template applied to analyses undertaken in Part II of this book.

Established Approaches and Their Limitations How can one establish the impact of IMF intervention on a country’s social, political, and economic performance? A naïve strategy might be to measure performance before a country enters the IMF program and then measure it again

42  40 Years of Structural Adjustment once it has ended; hence, any change in the outcome could be attributed to the introduction of the program. Some of the earliest evaluations of IMF programs followed just that intuition (Pastor, 1987; Reichmann & Stillson, 1978). The most obvious problem with this “before-​and-​after” approach is that factors outside of the program may also change over the course of the program, which in turn impact the outcome of interest (Vreeland, 2003). For example, world prices of commodities the borrowing country exports may decline during the course of the program, reducing economic performance. Using the before-​and-​after approach, we would conclude that it was IMF intervention that resulted in worse economic performance, rather than the collapse in world prices. Another simple strategy adopted in early studies, the “with-​without” approach, attempts to account for this possibility by comparing the performance of countries with programs to countries without them over the same period (Edwards & Santaella, 1993; Gylfason, 1987). But this method neglects the fact that changes in performance may reflect the circumstances which lead governments to enter into IMF programs in the first place (Vreeland, 2003). Countries entering into programs are systematically different from countries that do not, which affects subsequent performance—​the selection issue. These early methods reflect, more broadly, the limitations of “bivariate” statistical approaches to policy evaluation. They look at the relationship between only two variables—​IMF participation status and the outcome of interest. As an illustration of the issues involved, consider the relationship between participation in an IMF program and government social spending for the year 2000. The average level of social spending for countries participating in IMF programs in 2000 was 8.7 percent of GDP, whereas countries not participating spent 10.3 percent of GDP. While it is indeed possible that IMF programs caused reductions in government social spending in the year 2000, the evidence presented thus far is only suggestive. It is also possible that there is a confounding variable driving this association that is negatively correlated with social spending and positively correlated with IMF participation, such as the extent of democracy or the level of economic development. In order to establish whether there is truly a causal relationship, rather than being a mere correlation, we must statistically control for a host of factors that may also affect social spending. To garner a better understanding of the causal impact of IMF programs, analysts have since moved on to incorporate multivariate statistical methods that correct for the determinants of selection and the changing domestic and international circumstances during the program. These methods allow for the isolation of the effect of programs by statistically controlling for potential confounding variables. There exists already a wealth of recent statistical research on the consequences of IMF programs using a multitude of different multivariate approaches to address the selection issue (Abouharb & Cingranelli, 2009;

How to Evaluate the Effects  43 Bas & Stone, 2014; Nelson & Wallace, 2017; Nooruddin & Simmons, 2006; Oberdabernig, 2013; Steinwand & Stone, 2008). Nevertheless, commonalities do exist between them. For instance, each technique involves the assembly of “panel” data: a dataset in which multiple characteristics for each country—​the dependent, explanatory, and control variables—​are observed across multiple years (e.g., all countries for a period of two decades), where the unit of analysis is the country-​year. It then entails calculating the average treatment effect: the difference in average outcomes between observations (i.e., country-​years) in the treatment group (IMF program participation) and observations in a control group (no IMF participation), controlling for a host of potential confounders. Strategies for estimating the average treatment effect of IMF participation on some outcome variable—​such as economic growth, foreign direct investment, or social expenditures—​must not only confront the selection question in terms of observable characteristics of a country. They must also contend with characteristics that are not directly observable or measurable (Przeworski & Vreeland, 2000; Stone, 2008; Vreeland, 2003). To illustrate, consider again our earlier example where we are concerned that the same forces influencing a country’s decision to participate in an IMF program might also affect government social expenditure. Some of these forces are observable; they are measured by the World Bank and can be included in our list of control variables: changes in GDP per capita, the balance of payments, the government fiscal balance, and so on. Other factors are not directly observable, such as the political will or motivation of a government to implement reforms (Vreeland, 2003). Some countries are poorly motivated. They may never go to the IMF even when in economic turmoil because they do not take their socioeconomic circumstances seriously. Other countries are highly motivated. They approach the IMF even when in good socioeconomic health because they want to ensure their policies are appropriate. If countries that are more motivated to participate in IMF programs are also more motivated to implement social sector reforms than countries who do not participate, then country motivation levels affect social spending independently of the IMF’s treatment. Failure to account for motivation would thus erroneously attribute its effect on social spending to IMF participation. Given these potential sources of bias, more advanced statistical strategies are needed to correctly identify the effect of IMF programs. In performing these techniques, scholars also rely on a broad-​brush binary indicator for whether or not a country is under an IMF program in a given year as a measure of the organization’s engagement. A country under an IMF program in a particular year is coded with a 1, while a country not on a program is coded with a 0. Using this so-​called binary or dummy variable for IMF intervention is problematic for a couple of reasons. For one, as mentioned above, measuring IMF intervention in this way means that all programs are treated

44  40 Years of Structural Adjustment statistically as homogeneous. This belies the evidence in Chapter 2, showing major discrepancies in program design across different countries and over time. For example, Guatemala in 2004 faced only one condition, a structural reform related to the financial sector. In the same year, Romania faced 126 conditions spanning seven out of the eight policy areas. Of these, 51 were structural conditions and 75 were quantitative conditions. In 2014, Bosnia and Herzegovina had 92 conditions across seven policy areas, 72 of which were quantitative, while Benin had a single structural condition on the preparation of tax inspection reports. An additional reason the dummy variable approach for IMF intervention is problematic is because analysts are unable to separate the effects of conditionality from other pathways of program influence. There are a multitude of channels by which IMF programs could plausibly influence social policies and outcomes. First, the actual money the IMF disburses to ailing economies may have an impact via a “resource effect” (Dreher, 2006). The added financial resource could be used to boost expenditures to meet social priorities, although there is some evidence that the repayment of external debt and replenishment of foreign exchange reserves is prioritized instead (Gould, 2003; Ooms & Schrecker, 2005a; Stuckler et al., 2011; Stuckler & Basu, 2009). Alternatively, IMF credit may reduce government incentives to reform policy by propping up failing or nondemocratic regimes that have no interest in investing in social services for the population, as was the case under Joseph Mobutu’s three-​decade dictatorship of the Democratic Republic of Congo (Hanlon, 2006; Vreeland, 2007). Second, the availability of IMF money might influence social policy before it has even disbursed because of moral hazard (Dreher & Vaubel, 2004a; Dreher & Walter, 2010; Vaubel, 1983). In this context, moral hazard refers to a situation where a borrowing government gets involved in riskier economic activity precisely because it knows that the IMF will protect them against the most damaging consequences by lending them more money. Third, the very presence of an IMF program can catalyze additional aid from donor governments and mobilize financial flows from private international investors (Bird & Rowlands, 2002, 2007; Chapman et al., 2017; IMF, 2004d; Stubbs et al., 2016). Underpinning this claim is the idea that IMF programs serve as a stamp of approval: they transmit positive signals to donors and investors about the borrowing country’s commitment to policy reform, information that might have been difficult and expensive to collect and interpret ad hoc. To the extent that donors reward recipients that demonstrate good governance and appropriate institutions with additional aid (Berthélemy, 2006), IMF programs could bolster resources to support social spending, although there is no guarantee that those funds will actually reach the social sector (Lu et al., 2010; Stuckler et al., 2011). If international investors interpret IMF lending as reducing the risk of

How to Evaluate the Effects  45 deepening economic crisis, they too may ramp-​up financial inflows into the country, offering a potential source of additional tax revenues. A fourth channel of IMF program influence is through scaled-​up technical assistance and policy advice (Broome & Seabrooke, 2015). IMF policy training functions as a mechanism of socialization that imparts particular sets of policy norms to national officials, which in turn may impact decisions about social policy. Notwithstanding the issues surrounding the use of a dummy variable to represent IMF intervention, recent multivariate strategies for addressing the selection issue can be grouped into three categories: matching methods, instrumental variable approaches, and Heckman estimation.1 We discuss the benefits and shortcomings of each approach below.

Matching Methods Matching is a nonparametric statistical approach that is widely used in studies exploring the impact of IMF programs (Atoyan & Conway, 2006; Bal Gündüz, 2016; Garuda, 2000; Hardoy, 2003; Nelson & Wallace, 2017). It addresses the issue of selection bias arising from observable characteristics of a country by pairing observations (country-​years) with similar context but different IMF participation status. The approach does not, however, offer a solution to selection bias arising from unobservable characteristics, and—​strictly speaking—​should only be used when the decision of a country to participate or not in an IMF program can be fully accounted for by observed factors (Hardoy, 2003). As noted earlier, this is likely never the case. Even so, matching methods are still widely adopted because they depend on fewer distributional and modelling assumptions than parametric approaches—​for instance, that data approximates a normal distribution and that relationships are linear—​and do not necessitate an arduous search for a valid excludable instrumental variable (Copelovitch, 2010a; Hardoy, 2003), described further below. In pairing countries at a similar likelihood of participation, matching approaches identify an average treatment effect on the treated (Wooldridge, 2010). This effect is distinct from the average treatment effect described earlier. The average treatment effect answers the following question: What would be the effect of participating in an IMF program for a randomly selected country compared to the outcome if it does not participate? The average treatment effect is therefore the mean impact of intervention on countries against a counterfactual 1 Martin Steinwand and Randall Stone (2008) provide a comprehensive review of an earlier wave of studies on IMF program effects that correct for selection bias.

46  40 Years of Structural Adjustment or unobserved version of such countries where their participation decision is different to what actually occurred. But this approach does not consider whether or not countries included in the analysis would ever select into an IMF program in the first place (Hardoy, 2003). Consider an example: New Zealand is an economically developed, stable democracy that has never been on an IMF program and is unlikely to consider it in the foreseeable future; therefore, it might make little sense to think about a counterfactual New Zealand in 2014 that does participate compared to the observed New Zealand in 2014 that does not participate. Furthermore, if New Zealand ever did participate, the impact of IMF intervention might systematically differ from the impact of more frequent recipients precisely because the IMF views it as a highly irregular occurrence and so treats the country differently. In contrast, the average treatment effect on the treated answers the following question: What would be the effect of participating in an IMF program for a randomly selected country from a group of countries that are likely to participate compared to what the outcome would have been if it had not participated? The average treatment effect on the treated thus identifies the mean effect of IMF intervention for countries that actually participate in a program. Matching proceeds in three steps. An initial step entails calculating the probability of each country in a particular year participating in an IMF program, after controlling for observable economic and political factors. This is typically achieved by estimating a probit statistical model. The next step involves generating matches of similar probabilities—​or propensity scores—​between participating and nonparticipating observations (country-​years) to construct a treatment and control group. For instance, let us assume for a moment that country selection is driven only by the government fiscal balance. We could then match Tanzania in 2009 with Kenya in 2010. Both cases were experiencing moderate fiscal deficits, at −4.42 percent of GDP in Tanzania-​2009 and −4.41 percent of GDP in Kenya-​2010, but only the former participated in an IMF program. In our hypothetical example, Tanzania-​2009 thus enters the treatment group, while Kenya-​2010 enters the control group, in effect acting as a counterfactual for Tanzania-​2009. This process repeats until all treatment and control country-​ years are paired. In practice, these matches can be constructed using different matching techniques—​such as nearest-​neighbor, interval, or kernel matching—​ and various tolerance levels (Morgan & Winship, 2007). The most commonly used matching technique in studies on the effects of IMF programs is nearest-​ neighbor matching, which attempts matches in terms of the absolute distance between propensity scores, subject to the goal of minimizing the sum of all distances over all possible sets of matches. In this context, the choice of tolerance level determines the absolute distance that propensity scores must be equal or less than before two observations are matched, with unmatched observations

How to Evaluate the Effects  47 excluded from subsequent analyses. As the final step, the average treatment effect on the treated is then calculated as the difference in means between treatment and control groups for the outcome of interest. Despite the merits of matching, important limitations remain. Since the method relies only on observable determinants of IMF participation, it can actually accentuate selection bias (Dreher, 2006; Przeworski & Vreeland, 2000; Vreeland, 2003). We described earlier how unobserved political will of governments can bias results. To see how this plays out when matching, let us consider another potential unobservable determinant of IMF participation: domestic constituencies in some countries may have more confidence that the government negotiated for the most lenient conditions the IMF would allow than in other countries (Vreeland, 2003). If confidence plays a role in the government’s decision to select into a program—​perhaps because governments are more likely to participate in IMF programs when domestic constituencies are unlikely to place all the blame on the government—​and confidence also affects the country’s performance on the outcome of interest, then even Tanzania-​2009 and Kenya-​ 2010 may not represent a match. Indeed, if only governments with high confidence sign agreements and governments with low confidence do not, then there will be no matching counterfactuals. And if one proceeds with matching on observables regardless, then effects of the program will be confused with those of unobserved confidence. Aside from the obvious shortcoming, there are also no set benchmarks for suitable levels of tolerance. Should differences between matching pairs of country-​years be minimized? Or should differences between matching pairs of country-​years be maximized to the extent that all observations can be matched? The former would result in exclusions of observations due to incomplete matching, and the latter may result in unsuitable matches. This decision is left to the idiosyncratic preferences of each analyst.

Instrumental Variable Approaches Instrumental variable approaches can overcome biases linked to both observable and unobservable factors that determine selection into IMF programs. To comprehend how instrumental variable approaches can account for selection bias, one must understand the meaning of an “error term.” In statistical analysis, the error term represents explanatory factors that are unobserved—​that is, they were not included in the statistical model—​and are assumed to be random disturbances. Each observation for the outcome of interest thus has an explicit error term attached to it: the difference between what the statistical model predicted and what was actually observed. For example, in generating a

48  40 Years of Structural Adjustment statistical model to predict the effects of IMF programs on government social spending, we would also generate error terms accounting for unobserved factors that determine such spending. The problem we face when it comes to selection bias is that the error terms are not truly random. They are correlated instead with factors that have not been included in the statistical model. Political will might be an example of this if our outcome of interest is government social spending. As mentioned earlier, if political will also impacts the decision to participate in an IMF program, then the effect of political will would be erroneously attributed to IMF participation. Error terms become important once we come to consider the intuition behind correcting for selection bias due to these unobservables using instrumental variable approaches. Initially, a selection equation is used to predict a country’s decision to participate in an IMF program. In generating a statistical model to predict IMF participation, we also generate error terms which account for unobserved factors that determine selection. We suspect these error terms will be correlated with nonrandom unobservable factors such as political will or confidence. As noted above, when examining the effects of IMF programs on government social spending, we also generate these error terms. In other words, where there is selection bias, the errors from the estimation of selection are correlated with the errors from the estimation of government social spending (Vreeland, 2003). Once a correlation between the error terms is detected, one can remove the effects of nonrandom selection on the outcome of interest. In the IMF literature, this is typically achieved via two-​stage least squared (2SLS) estimation (Barro & Lee, 2005; Butkiewicz & Yanikkaya, 2005; Dreher, 2006; Dreher & Gassebner, 2012; Easterly, 2005; Moser & Sturm, 2011; Oberdabernig, 2013; Steinwand & Stone, 2008), three-​stage least squared (3SLS) estimation (Butkiewicz & Yanikkaya, 2005; Dreher, 2006; Nsouli et al., 2006), and system generalized method-​of-​moments (GMM) estimation (Clements et al., 2013; Dreher & Gassebner, 2012; Dreher & Walter, 2010; Mukherjee & Singer, 2010). In 2SLS, the initial selection equation predicting IMF participation (i.e., the first stage) is calculated via Ordinary Least Squares (OLS) statistical estimation and should include the same control variables that will feature in the statistical model for the outcome of interest. In addition, at least one variable needs to be added to the selection equation to act as an instrumental variable. In order to serve as an instrument, the variable must fulfill two main criteria: it must not affect the outcome of interest except via IMF participation (the exclusion criterion), and it must be partially correlated with IMF participation once the other variables are controlled for (the relevance criterion). It is worth noting that the exclusion criterion is an untestable assumption, instead judged by the cogency of the analyst’s reasoning, whereas the relevance criterion can be directly tested. By

How to Evaluate the Effects  49 including an instrument, selection bias—​the correlation between observed IMF participation and the error term—​is essentially purged from predicted values of IMF participation. These predicted values—​instead of the actually observed values—​and a set of control variables that account for observed confounders for the outcome of interest are then included in a subsequent outcome equation that is estimated using OLS (i.e., the second stage). Assuming selection bias originates only from the observed IMF participation variable, then error terms in the outcome equation will not be correlated with factors excluded in the statistical model; that is to say, error terms will be random. Identifying an instrumental variable is notoriously difficult. While past research has relied on a range of political economy variables, by far the most popular instrument for IMF participation is United Nations General Assembly (UNGA) voting similarity with the United States (Dreher & Gassebner, 2012; Steinwand & Stone, 2008; Woo, 2013). The argument for the instrument’s relevance is that all else being equal, countries that vote similarly to the United States are more likely to participate in IMF programs. Research by Strom Thacker (1999) underpins this claim. He showed that shifting United Nations voting pattern alignment toward the United States increases a country’s chances of receiving a loan from the IMF, reasoning that the United States government pressures the IMF to approve loans on favorable terms to politically friendly countries. To fulfill the exclusion criterion, UNGA voting similarity with the United States must also not affect the outcome variable except via IMF participation. This may or may not be a valid assumption depending on the outcome of interest. For example, if the outcome is levels of democracy then the UNGA instrument is not excludable (Nelson & Wallace, 2017), since democratic states exhibit similar voting patterns to those cast by the United States (Carter & Stone, 2015). More recent research casts further doubt on the validity of the UNGA instrument. Specifically, a frequently overlooked criterion that is required for an instrument to be valid is that its effect on IMF participation must be representative of all programs, not just subpopulations of them. Axel Dreher and colleagues (2018) show that this is not the case. Politically motivated programs tend to be less effective than nonpolitically motivated ones. It therefore follows that studies trying to identify the causal effects of IMF intervention by instrumenting for program participation using UNGA alignment with the United States represent the lower bound of the true effect of intervention. An extension to the 2SLS instrumentation technique is 3SLS. This computationally more complex estimation method extends the 2SLS procedure by incorporating information from cross-​correlations of the error terms in a system of simultaneous equations (Zellner & Theil, 1962). In so doing, it can estimate the effects of IMF intervention with greater precision (Barro & Lee, 2005; Nsouli et al., 2006). But like 2SLS, 3SLS still requires a valid instrument to produce

50  40 Years of Structural Adjustment unbiased estimates, so the analyst is not absolved of the potentially vexing search to find one. System GMM estimation emerged more recently as a potential solution to the inherent difficulties in finding valid instruments (Arellano & Bond, 1991; Arellano & Bover, 1995; Blundell & Bond, 1998). Unlike 2SLS and 3SLS instrumental variable approaches, this technique does not assume that valid instruments are available outside the immediate dataset of IMF variables. It instead uses internally derived instruments based on earlier values of IMF participation. System GMM proceeds by estimating a system of two simultaneous equations (Roodman, 2009a, 2009b). The first is a “differences” equation. Here, explanatory and control variables enter the model as the changes in their value over the previous year (i.e., first-​differences), and previous years of IMF participation (usually two or more years) act as an instrumental variable for the change in IMF participation. The second is a “levels” equation. Here, explanatory and control variables enter the model as their actual value for the year (i.e., levels), and the previous year’s change in value over two-​years earlier (i.e., lagged first-​differences) act as an instrumental variable for the contemporaneous level of IMF participation. Despite its advertised flexibility, system GMM estimation makes heroic assumptions about the underlying data. It assumes that the correct model for the outcome of interest is one where changes are a function of past trends, and that lagged first-​differences can predict present levels. Again, this may or may not be a valid assumption depending on the outcome of interest. System GMM estimation also assumes that first-​differences of instruments are uncorrelated with time-​invariant country-​level characteristics (i.e., country fixed effects), such as legal origin, geography, or ethnolinguistic fractionalization (Roodman, 2009a, 2009b; Stuckler et al., 2012). For this assumption to hold, country fixed effects and first-​differences of IMF participation must offset each other for each country in the dataset. It requires that “throughout the study period, [countries] sampled are not too far from steady states, in the sense that deviations from long-​run means are not systematically related to fixed effects” (Roodman, 2009b, p. 128). Whether this criterion holds depends on the sample of countries and time periods included, but is unlikely to be met in the context of IMF intervention. A separate line of critique is that system GMM estimation is too sensitive to numerous decisions on statistical minutiae—​for example, the number of previous years the instrument should cover, whether instruments should be collapsed, and whether estimation is one-​or two-​step—​none of which have a clear theoretical basis when studying IMF intervention (Stuckler et al., 2012). Even David Roodman (2010), who developed the statistical package to compute system GMM, acknowledges that these choices matter: they can make estimates more or less valid, and they can make certain tests of that validity stronger or weaker.

How to Evaluate the Effects  51 Most alarmingly, Roodman (2009a, p. 156) concludes that “the estimators carry a great and under-​appreciated risk: the capacity by default to generate results that are invalid and appear valid.”

Heckman Estimation Heckman estimation is another popular statistical technique that is able to account for unobservable factors driving selection into IMF programs. The intuition behind the Heckman correction for selection bias is similar to instrumental variable approaches. Recall that if we generate a statistical model to predict government social spending, then the error terms will not be random because they will be correlated with nonrandom factors not included in the model, such as political will or confidence. This means we would erroneously attribute their effects to IMF participation. But whereas in instrumental variable approaches the predicted values of IMF participation, purged of selection bias, are subsequently included in the outcome model, for Heckman estimation the actually observed values of IMF participation are used instead, and the nonrandom assignment of countries into participating and nonparticipating groups is treated as an omitted variable problem (Heckman, 1979). Bias due to nonrandom country selection into IMF programs is controlled for by including the “inverse-​Mills ratio” in the outcome equation. The inverse-​ Mills ratio can be understood as “the marginal probability of misclassifying an observation given where the observation lies in the overall distribution of observations” (Vreeland, 2003, p. 116). In effect, the inverse-​Mills ratio represents a catch-​all term that captures the qualities that make a country prone to selection. These values are generated by estimating a probit statistical model predicting a country’s IMF program participation status. Importantly, as with instrumental variable approaches, the participation equation also requires an instrumental variable that influences selection into IMF programs but not the subsequent outcome of interest (Lang, 2021). The inverse-​Mills ratio generated from this process is then added as a control variable to the outcome equation estimated using OLS, along with a set of additional control variables that account for observed confounders for the outcome of interest. The parameters capturing the influence of the inverse-​Mills ratio in this estimation of program effects also indicate if there is an association between the selection and performance error terms (Vreeland, 2003). A negative relationship between the inverse-​Mills ratio and, say, government social expenditure indicates that unobserved variables that make IMF participation more likely are associated with lower spending; a positive relationship indicates that unobserved variables that make IMF participation more likely are associated with higher spending.

52  40 Years of Structural Adjustment The broader appeal of Heckman estimation over instrumental variable approaches is that it can measure the effects with greater precision when the selection variable is dichotomous, as is the case when a dummy variable is used to represent IMF intervention (Wooldridge, 2015). To this end, two main variants of Heckman estimators have been deployed in IMF literature: a standard Heckman model (Bas & Stone, 2014; Przeworski & Vreeland, 2000; Vreeland, 2003), and the control function approach (IEO, 2003; Mukherjee & Singer, 2010; Nooruddin & Simmons, 2006; Oberdabernig, 2013). For standard Heckman models, the outcome equation is limited to observations only in years where countries have selected into an IMF program. Although this approach cannot directly estimate the effect of participating in an IMF program, it can do so indirectly by estimating another statistical model for observations in years where countries have not selected into IMF participation. Then, one need only calculate the weighted difference for the entire sample of the selection-​corrected effect size for observations featuring IMF participation with the selection-​corrected effect size for observations not featuring participation (Vreeland, 2003). Conversely, a control function approach includes all observations in the outcome equation regardless of whether or not the country selected into an IMF program (Wooldridge, 2015), and can thereby directly estimate the effect. This approach is frequently mislabeled as a Heckman model in the IMF literature because it draws on insights from Heckman’s work vis-​à-​vis the source of omitted variable bias. Throughout, we refer to it by its proper terminology—​a control function approach. Notwithstanding its widespread use, Heckman estimation is not without limitations. The most obvious concern is the lack of a readily available instrument for IMF participation, as we described earlier in relation to instrumental variable approaches. Beyond this, the precision of Heckman estimation depends on the variance of the inverse-​Mills ratio, which is determined by the predictive capacity of the probit statistical model (Winship & Mare, 1992). That is to say, it depends on having correctly specified the participation equation. A final, related drawback is linked to a peculiar feature of the probit model: it is unable to facilitate the inclusion of country fixed effects—​a series of country dummy variables that control for time-​invariant country-​level characteristics—​without introducing new sources of statistical bias, due to the so-​called incidental parameter problem (Greene, 2004). Yet by not including these country fixed effects in the probit model, bias might be introduced to the analysis if time-​invariant country-​level characteristics do in fact affect the IMF participation decision. The analyst therefore faces no easy solution, since the introduction of any source of bias is likely to be deemed unsatisfactory given that it could undermine the credibility of the results. On balance, none of the methods described here are perfect, but some clearly perform better than others in dealing with the problem of selection

How to Evaluate the Effects  53 bias. Matching methods are the least palatable option due to their inability to address selection bias arising from unobservable characteristics. System GMM estimation—​an instrumental variable approach using internal instruments—​ can also be discounted because it carries stringent assumptions that are untenable in all but the most exceptional of circumstances. We are thus left with external instrumental variable approaches (i.e., 2SLS and 3SLS), and Heckman estimation (i.e., standard Heckman and control function approach). Of these, our favored strategy is external instrumental variable approaches. This is because we would rather minimize the potential bias that could be introduced by excluding country fixed effects in the participation equation over the efficiency gains that Heckman estimation achieves for dichotomous variables like IMF participation. In what follows, we develop a new technique for examining the effects of IMF conditionality that builds on external instrumental variable approaches. In so doing, we identify an instrumental variable that is relevant and excludable for all outcomes covered in this book.

Methods for Studying the Effects of Conditionality So far, we have discussed how scholars have tried to identify the impact of IMF intervention on a country’s social, political, and economic performance using a measure that is based on whether or not a country is participating in a program on a given year. We also noted that this is a crude aggregated measure that treats all programs as if they are identical. We described two main concerns with the dummy variable approach. First, such work obscures countries’ diverging experiences with IMF programs, which—​ as discussed in Chapter 2—​ are designed ad hoc and thereby entail heterogeneous policy content not accurately captured by a binary indicator. The empirical approach therefore implicitly sides with critics accusing the IMF of one-​size-​fits-​all policies (Stiglitz, 2002), despite the IMF’s strong rejection of this claim (Dawson, 2002). Second, these studies cannot isolate the effects of conditionality from alternative channels of program influence. These include, inter alia, scaled-​up technical assistance and policy advice (Broome & Seabrooke, 2015; IMF, 2016a), aid catalysis (IMF, 2004d; Stubbs et al., 2016), and moral hazard (Dreher & Walter, 2010). It is possible, for instance, that the impact of conditionality diverges from the impact of other aspects of IMF operations. Why did scholars persist with the dummy variable approach to measure IMF intervention for so long? The reason is, quite simply, because until recently they lacked systematic, transparent, and replicable data on the actual policy content of IMF programs. Indeed, we know of only 10 studies examining the effects of IMF conditionality per se (Beazer & Woo, 2016; Bulír & Moon, 2004; Casper,

54  40 Years of Structural Adjustment 2017; Chapman et al., 2017; Crivelli & Gupta, 2016; Dreher & Vaubel, 2004b; Ivanova et al., 2006; Rickard & Caraway, 2019; Wei & Zhang, 2010; Woo, 2013). Having overcome the initial hurdle in obtaining the data, how can one now identify the effect of conditionality, as distinct from other aspects of IMF operations? It turns out that there are analogous pitfalls from the previous generation of studies using the dummy variable approach, and therefore some of the same lessons are relevant. Remember our extension of James Vreeland’s doctor analogy in this chapter’s introduction: a patient suffering from a common cold might be recommended a course of painkillers and some rest, whereas a patient with a serious illness might be prescribed many powerful pills (and to extend the analogy further, some of these pills will have serious side effects of their own!). In essence, the circumstances of countries receiving many severe IMF conditions may be systematically different from those receiving fewer and more lenient conditions, and these underlying differences may also affect the outcome of interest (in the same way that countries entering into programs are systematically different from countries that do not). For example, low-​income countries like Burundi and Yemen may receive fewer and different kinds of conditions than middle-​income countries like Argentina and Turkey, perhaps because the IMF thinks low-​income countries lack the capacity to radically alter their underlying economies. Past studies already find that domestic political circumstances in the borrowing country, professional ties and shared beliefs between IMF staff and borrowing-​country officials, and international strategic factors all influence the number and types of conditions that countries receive (Caraway et al., 2012; Chwieroth, 2015; Dreher et al., 2009, 2015; Dreher & Jensen, 2007; Gould, 2003; Nelson, 2014a; Stone, 2008). It may also be the case that some factors that systematically differ between countries that receive more and different kinds of conditionality are inherently unobservable. While it has been established that countries with higher political will select into IMF participation, what is less apparent is whether countries select into conditions in the same way (Vreeland, 2006). There is some evidence to suggest that certain conditions are sought by governments to gain leverage over domestic opposition to policy change (Caraway et al., 2012; Rickard & Caraway, 2014; Vreeland, 2006), even if conditionality is primarily used as a coercive instrument to compel countries into implementing reforms they may not otherwise wish to undertake (Grabel, 2011; Simmons et al., 2008; Stiglitz, 2002). Regardless, failure to account for factors that codetermine conditionality and the outcome of interest would result in a biased estimate of the effect of conditionality on the outcome of interest. Furthermore, the few studies examining the effects of IMF conditionality are yet to converge on a single method to address these concerns—​and some of them do not bother with it at all! Given this lack of consensus, it is therefore necessary to

How to Evaluate the Effects  55 adapt similar kinds of complex multivariate statistical methods described above to account for these potential sources of bias. A distinct statistical quandary introduced by trying to identify the effects of conditionality is separating its effect from alternative channels of program influence. An obvious solution is to restrict the analyses only to those observations (i.e., country-​years) where a country actually participated in an IMF program. This would capture the conditioned effect—​or average treatment effect on the treated—​of IMF conditionality. Six of the 10 studies focusing on the effect of conditionality follow this approach (Beazer & Woo, 2016; Casper, 2017; Dreher & Vaubel, 2004b; Ivanova et al., 2006; Rickard & Caraway, 2019; Woo, 2013). This option may seem intuitive because it yields findings that are easy to interpret: the analyst need only consider the effect of whatever variables are used to measure conditionality. They can then make claims about the kind of socioeconomic outcomes that a country under an IMF program but with different conditionality design—​for instance, one with fewer conditions, or one with or without a particular type of condition—​might experience. However, adopting this strategy also means that results can only be interpreted within the context of country-​ years with an IMF program. For example, the analyst cannot make claims about what the socioeconomic impact might be if a country accepts a certain condition package versus choosing not to participate in a program at all. This is therefore not very helpful for policy practitioners, civil society, or governments deciding on whether or not entering into a program is worth it. In effect, it offers only a limited set of policy implications surrounding IMF program design. Below, we endorse an alternative approach where both an IMF participation dummy and measures of conditionality are included in the statistical model. This allows the analyst to distinguish effects of conditionality from other aspects of IMF intervention, while also enabling counterfactual comparisons to cases where countries choose not to participate in a program. Four of the 10 studies on the impact of IMF conditionality employ this strategy (Bulír & Moon, 2004; Chapman et al., 2017; Crivelli & Gupta, 2016; Wei & Zhang, 2010). An additional statistical challenge relates to the question of which measure of conditionality to use in analyses. IMF conditionality is typically measured on three dimensions: magnitude, or the number of conditions applicable either in total or within a given policy area; scope, or the total number of policy areas subject to conditionality; and depth, or the relative stringency of each of the conditions (IEO, 2007a; Stone, 2008). We do not use the scope measure because it would not enable us to disaggregate analyses into policy areas to identify the particular mechanisms by which conditionality affects the outcome of interest. Using the magnitude of conditionality as a characterization of program stringency and intrusiveness has become best practice among analysts (Beazer & Woo, 2016; Caraway et al., 2012; Chwieroth, 2015; Copelovitch, 2010b; Dreher

56  40 Years of Structural Adjustment et al., 2009, 2015; Dreher & Jensen, 2007; Dreher & Vaubel, 2004b; Gould, 2003; Rickard & Caraway, 2014), and we follow this approach throughout this book. Even so, focusing on conditionality’s magnitude involves analytical limitations, since the measure does not capture the difficulty—​ or depth—​ in implementing any individual condition (Dreher et al., 2015). We believe this to be an unavoidable sacrifice because it would be logistically impossible to measure individual condition depth given the vastly different characteristics of IMF borrowers: the same condition will be easier (or harder) to implement depending on the pre-​existing domestic institutional environment, which will vary across country (Copelovitch, 2010b). Although the IMF’s Independent Evaluation Office (2007b) tried to measure the difficulty of implementation for conditions based on whether parliamentary approval was required, this criterion is insensitive and arbitrary in its own right. Ultimately, coding condition depth entails, in our view, an unacceptable level of subjectivity.

Statistical Bias in Multiple IMF Variables The method we propose is adaptable to the inclusion of either the number of all conditions, the number of different types of conditions (i.e., quantitative or structural), or the number of conditions in particular policy areas. Our unit of analysis is the country-​year, where multiple characteristics for each country are observed across multiple years (i.e., panel data). We also include both an IMF participation dummy and conditionality variables in our statistical models, so that the effects of conditions can be isolated from other aspects of IMF operations. We proceed under the assumption that both IMF participation and conditionality are subject to the sources of statistical bias described above: countries select into programs based on pre-​existing country characteristics, some of which cannot be directly observed or measured, which in turn influence the outcome of interest; and pre-​existing country characteristics also determine the type and amount of conditions that countries receive. Our solution is thus to combine instrumental variable approaches for both IMF participation and conditionality. How is this achieved? Recall in the discussion on instrumentation techniques that we referred to in passing of a computationally more complex estimation technique known as 3SLS. Here we adopt a technique similar to 3SLS called maximum likelihood statistical estimation (MLE) to estimate a system of three simultaneous equations. MLE is similar in the sense that it incorporates information from cross-​correlations of the error terms in an equation system, but it is more flexible than 3SLS because it can incorporate nonlinear relationships if need be—​for example, if the outcome of interest is dichotomous. Recall that 3SLS still requires a valid instrument to produce unbiased estimates, and this

How to Evaluate the Effects  57 is also the case for MLE. Yet now we require at least two plausibly excludable instruments: one for IMF participation and another for conditionality. We deliberate on this hurdle subsequently. Our system of three equations reads as follows:

 = γ X + γ Z + µ + δ (1) IMFPROG it 1 it 2 it i t



 = α X + α Y + µ + δ (2) IMFCOND it 1 it 2 it i t



 + β IMFCOND  + β X + µ + δ + ε (3) Wit = β1 IMFPROG it it 2 3 it i t it

Here, i is country and t is year. Equation (3) is the outcome equation, where W  is the predicted value for IMF participation is the outcome of interest; IMFPROG  is the predicted value for the number of derived from Equation (1);  IMFCOND conditions derived from Equation (2); X denotes a vector of control variables; μ is a set of country dummy variables to account for time-​invariant country-​level characteristics (i.e., country fixed effects); δ is a set of year dummy variables to account for common external shocks across all countries (i.e., year fixed effects); and ε is the error term. Equation (1) is a linear model to obtain predicted values  . It is assumed to be a function of X, a list of of IMF participation, IMFPROG covariates from the outcome equation; Z, an excludable instrument; μ, a set of country dummies; and δ, a set of year dummies. Equation (2) obtains the  . It is also a function of predicted values for IMF conditionality, IMFCOND X, μ, and δ; as well as, Y, another excludable instrument. The system of three simultaneous equations is estimated through MLE using the cmp module for Stata statistical software version 15, which allows us to  , and IMFCOND  . The procejointly estimate the covariates of W,  IMFPROG dure also enables us to compute robust country-​clustered standard errors using the Sandwich estimator (Wooldridge, 2010). Standard errors are a measure of the statistical accuracy of an estimate of the effect. The clustered Sandwich estimator tends to produce larger standard errors than the unadjusted version because it adjusts for potential heteroscedasticity and serial correlation in our dataset.2 This technique can accommodate different lag structures for variables on the right-​hand side of the equations. For instance, the values of these variables could be the same year as the outcome of interest (no lag), one year earlier than that of the outcome of interest (lagged one year), or two years earlier (lagged two years). It is common for analysts to include an IMF participation variable lagged 2 Further details on how the model is jointly estimated, including the theoretical properties of the estimator and its distributional assumptions, are available in statistical literature by David Roodman (2011) and by the authors of this book (Stubbs et al., 2020).

58  40 Years of Structural Adjustment one year to correspond with the budget cycle (Crivelli & Gupta, 2016), since governments typically plan their spending a year in advance. But lags of either zero, one, or two years may be appropriate for multiple variables depending on the effect pathways one purports to measure (Oberdabernig, 2013). There may be instances where researchers wish to test on even deeper lags where effects are expected to unfold only after a substantive period of time has elapsed. Ultimately, the appropriateness of the lag structure will depend on theoretical expectations and the outcome of interest. Thus far, we have described how our method might capture the effect of each condition using a measure of conditionality that aggregates all conditions in a country-​year. This kind of analysis is useful initially to establish the blanket, overall effect of conditionality on the outcome of interest. For example, we could say that—​taken as a whole—​conditionality either increases, decreases, or has no effect on government social expenditure. But we are also interested in establishing the impact of different types of conditions (i.e., structural and quantitative) and conditions in given policy areas. This information might be useful even if there is no blanket effect of conditionality. Perhaps structural conditions increase social spending but quantitative conditions decrease it; these two effects may cancel each other out so that there is no effect on aggregate. Yet the fact that certain conditions affect outcomes in different ways is of substantive policy interest. The IMF could manipulate the types of conditions included in a program based on their expected effects, in order to achieve desired outcomes. Alternatively, a recipient government could decide to accept or reject a reform program by looking at the expected effects of the types of conditions included within. To this end, our method can be extended to accommodate disaggregated counts of conditions. We do this by incorporating two IMF conditionality variables in the outcome equation. One is a count of the condition type or policy area of substantive interest, and the other is a count of all remaining conditions. For example, if we wanted to know the effect of structural conditions on social spending, then we would include a variable counting the number of structural conditions as well as a variable counting the number of quantitative conditions. Similarly, if we were interested in the impact of a specific policy area condition—​ say, poverty reduction policies—​then we would include a count of conditions requiring adoption of poverty reduction policies and a count of conditions outside of that policy area. All conditions are therefore still jointly included in the outcome equation, so as to ensure that the effect of the “residual” conditions is not attributed to the condition type or policy area of substantive interest. We do not ever include more than two conditionality variables per outcome equation because the analysis begins to confront computational limits—​specifically, MLE estimates cannot reliably converge on an expected pattern at that point.

How to Evaluate the Effects  59 For disaggregated analyses, we therefore have a system of four equations estimated by MLE:

 = γ X + γ Z + µ + δ (4) IMFPROG it 1 it 2 it i t



 1 = λ X + λ Y 1 + µ + δ (5) IMFCOND it 1 it 2 it i t



 2 = ζ X + ζ Y 2 + µ + δ (6) IMFCOND it 1 it 2 it i t

 + β IMFCOND  1 + β IMFCOND  2 + β X + µ + δ + ε (7) Wit = β1 IMFPROG it 2 it 2 it 3 it i t it Here, Equation (7) is the outcome equation, and is identical to Equation (3) except that there are now two predicted values for the number of conditions,  2 . These could correspond with, for instance, the  1 and IMFCOND IMFCOND predicted number of structural and quantitative conditions; alternatively, they could be the predicted number of poverty reduction policies and all remaining conditions. Equation (4) is the linear model to obtain predicted values of IMF  , and is unchanged from Equation (1). Equations participation, IMFPROG (5) and (6) obtain the predicted values for the respective IMF conditionality variables included in Equation (7), where Y1 is an excludable instrument for  1  and Y2 is a different excludable instrument for IMFCOND  2. We IMFCOND turn to the issue of identifying excludable instruments next.

An Excludable Instrument for IMF Program Participation The perennial challenge of instrumental variable approaches is finding measurable variables that affect the variables the analyst wishes to instrument but not the outcome variables, except via the impact on the instrumented variables. For our purposes, we need at least two different instrumental variables. One instrument must affect the decision to participate in a program in the first place but not the outcome of interest, except via the impact on participation. The other instrument must affect the number of conditions applicable but not the outcome of interest, except via the impact on conditionality. Here we assess potential instruments for IMF participation and conditionality respectively. We noted earlier that past studies using the IMF participation dummy usually rely on UNGA voting similarity with the United States (Dreher & Gassebner, 2012; Steinwand & Stone, 2008; Woo, 2013); however, the local average treatment effect of the UNGA instrument might not be generalizable since politically motivated programs could be less (or more) effective (Dreher et al., 2018), and for some of our outcomes of interest it may not fulfill the exclusion criterion.

60  40 Years of Structural Adjustment Our preferred approach is therefore to adopt an instrument originally proposed by Valentin Lang (2021): the interaction of the within-​country average of program participation across the period of interest with the year-​on-​year budget constraint of the IMF. Statistical proofs for the general statistical strategy of using interactions of nonexcludable with excludable variables to generate instrumental variables are available elsewhere (Bun & Harrison, 2019; Nizalova & Murtazashvili, 2016), and the strategy is already being applied in cogent fields of inquiry such as the aid effectiveness literature (Dreher & Langlotz, 2020; Nunn & Qian, 2014). Formally specified, the predicted values for IMF participation in Equations (1) and (4) above are thus calculated as follows:

(

)

 = y X + y IMFPROG × IMFBUDG + µ + δ (8) IMFPROG it 1 it 2 i t i t

 is the predicted value of IMF Here, i is country and t is year. IMFPROG partici­pation; IMFPROG  is the average value of participation in country i, which will be different for each country but will be constant across years within any given country; IMFBUDG is the budget constraint of the IMF in year t, which will be different for each year but will be constant across countries in any particular year; X is a list of covariates from the outcome equation and the current account on the balance of payments as a share of GDP, a key determinant of IMF program participation (Lang, 2021; Steinwand & Stone, 2008); μ is a set of country dummies; and δ is a set of year dummies. The key identifying assumption—​an expectation about the data generating process that allows us to draw causal inferences—​is that the outcome of interest in countries with different exposure to program participation will not be affected differently by changes in the IMF’s budget constraint except through the impact of program participation. Does Lang’s instrument satisfy standard criteria of an instrumental variable? Before considering this, we need a suitable measure of the budget constraint. Lang proposes the natural logarithm of the IMF’s liquidity ratio (i.e., the ratio of liquid resources to liquid liabilities). In this formulation, liquid resources comprise usable currencies plus Special Drawing Rights contributed, and liquid liabilities incorporates total members’ reserve tranche positions plus outstanding IMF borrowing from members. The natural logarithm of the liquidity ratio is taken due to its skewed distribution. Figure 3.1 charts temporal variation of the IMF liquidity ratio between 1995 and 2019. We can see that liquidity fluctuated at relatively low levels between 1995 and 2005, before increasing sharply to its peak in 2008. Since then, liquidity has tapered off despite remaining high.

How to Evaluate the Effects  61

Natural log of IMF liquidity ratio

7 6.5 6 5.5 5 4.5 1995

2005

2000

2010

2015

2020

Year

Figure 3.1  IMF liquidity ratio, 1995–​2019 Note: Data is sourced from Lang (2021) for 1995–​1999 and IMF financial statements for 2000–​2019.

Recall that an instrument must be partially correlated with IMF participation to fulfill the relevance criterion. Lang ventures that the interaction of a country’s average program participation with the IMF budget constraint should fulfill this criterion because the cross-​country average of participation approximates the general propensity of a country to enter into a program in any given year. This relationship is attributed to IMF recidivism, the tendency for countries with a lengthier history of IMF program participation to be more likely to enter into subsequent programs (Bird et al., 2004; Conway, 2007; Vreeland, 2007). The reasons for recidivism vary. One possibility is that the IMF does not successfully promote economic development, and so countries repeatedly need more financial assistance. Alternatively, interpersonal connections established in an initial program between Fund staff and country officials could lower transaction and negotiation costs of future agreements, making a repeat program more attractive (Vreeland, 2007). Or, it could be that there is initially a high political stigma or “sovereignty cost”—​opposition may accuse a government of “selling out” to Western powers—​attached to IMF program participation that diminishes the more a country participates (Vreeland, 2007). In addition, the IMF is likely to lend more credit when it has high liquidity, all else considered. Underpinning this claim is the idea that the IMF is, at least to some degree, a self-​serving international bureaucracy that seeks to maximize revenues from interest payments, safeguard future resources, and maintain a position of global power (Babb & Buira, 2005; Barnett & Finnemore, 2004; Dreher & .

62  40 Years of Structural Adjustment Vaubel, 2004b; Vaubel, 1996). Thus, in years where the IMF has higher liquidity, the IMF has greater capacity and organizational incentive to be more generous and less risk averse. It serves as an opportunity for the IMF to expand clientele beyond the usual pool of recidivists, achieved by granting loans to countries not otherwise considered in times of lower liquidity, since potential financial losses from such countries defaulting on loan repayments are relatively dampened when resources are abundant. In any case, the assumption of instrument relevance can be tested. In analyses conducted in subsequent chapters, the instrument consistently satisfies diagnostic benchmarks indicating strong instruments—​in statistical parlance, a Kleibergen-​Paap F-​statistic above 10 (Staiger & Stock, 1997). Another requirement for the instrument to be valid is that it must not affect the outcome of interest—​in this book, health expenditure, income inequality, or health outcomes—​except via IMF participation. That is to say, it must fulfill the exclusion criterion. Remember that the exclusion criterion is inherently untestable. We can only put forward what we believe is a convincing case for the excludability of the interaction of a country’s average program participation with IMF liquidity. Our reasoning is that changes in a country’s probability of participating in programs that deviate from its long-​run average are brought about only by IMF decisions that apply to all countries. Even so, one potential concern is that a country’s average program participation might impact the outcome of interest directly (i.e., not just via IMF program participation in a particular country-​year), thereby breaching the exclusion criterion. The inclusion of country fixed effects in both IMF participation and outcome equations, as we do here, accounts for this possibility. This is because the instrument needs only to be excludable conditional on the baseline controls (Dreher & Langlotz, 2020; Nunn & Qian, 2014). And since the average propensity of a country to participate is ipso facto a time-​invariant country-​level characteristic, it is perfectly correlated with country fixed effects—​so controlling for them is tantamount to controlling for a country’s average program participation. Similarly, one might be worried that IMF liquidity could be linked to the outcome of interest other than through program participation (or conditionality, described further below). While we can think of no direct pathway from IMF liquidity to outcomes in a given country, perhaps there is an indirect pathway, “donor concern,” where poor socioeconomic outcomes lead donors to replenish IMF resources and step-​up bilateral support to enable greater country assistance. This would mean that if we generate a statistical model to predict, say, government social spending, then the error terms will not be random because they will be correlated with donor concern, an effect we may erroneously attribute to IMF participation. In defense, we find it unlikely that there are unobserved channels of influence—​such as donor concern influencing IMF liquidity and

How to Evaluate the Effects  63 country-​specific outcomes—​because the IMF’s main budget line, the General Resource Account, is predetermined by organizational factors divorced entirely from the characteristics of borrowing countries (Lang, 2021). The financial resources that members commit to the IMF is set by the Board of Governors in quota reviews conducted every five years (IMF, 2017b). Nonetheless, it could be that the IMF’s concessional lending budget, which, unlike its General Resource Account, is replenished through voluntary contributions rather than quota subscriptions, may undermine excludability of the instrument (IMF, 2016e). Donors may be less willing to spend in times of global financial crisis, resulting in both a reduction of the IMF’s concessional budget and deteriorating socioeconomic outcomes in struggling countries. We maintain that this does not represent a legitimate threat to the excludability of the instrument because the inclusion of year fixed effects in both participation and outcome equations captures common external shocks across all countries, ensuring that IMF liquidity is not correlated with the error terms in the outcome equation (Nunn & Qian, 2014). We are therefore confident that conditional on the baseline controls, the IMF budget constraint is an exogenous source of variation in our statistical models. So far, we have established that the link between the average propensity of a country to participate in an IMF program and its participation decision in any given year depends on the IMF’s liquidity (Lang, 2021). In years where IMF liquidity is relatively low, IMF programs go to recidivists. But this link is weaker in years when liquidity is relatively high because the IMF promotes participation by seeking new borrowers, which dilutes the strength of the relationship between recidivism and program participation. Statistically speaking, the instrumentation strategy is based on the equivalent of a continuous difference-​in-​differences logic: differences in IMF liquidity result in differences to the link between IMF recidivism and program participation. As a difference-​in-​differences statistical design, additional assumptions must be fulfilled. Specifically, nonparallel trends across groups with different exposure to the country-​varying component of the interacted instrument can introduce statistical bias (Christian & Barrett, 2017). In our case, trends over time in the average probability of IMF program participation—​ our exposure variable—​ should be similar across groups of countries with above-​mean program exposure and groups of countries with below-​mean program exposure. In addition, statistical bias would also be introduced if there was a nonlinear trend in the time-​varying component of the interacted instrument that is similar to the respective trends in the exposure and outcome variables in the high-​exposure group of countries (Christian & Barrett, 2017). In our application, we would be concerned about similarly shaped nonlinear trends in IMF liquidity, the average probability of IMF program participation, and our outcomes of interest if these trends only occur among high-​exposure countries but not low-​exposure countries. This is

64  40 Years of Structural Adjustment because such nonlinear parallel trends would create spurious correlation not controlled for by year fixed effects unless the same nonlinear trends are also present among the low-​exposure group of countries. In Figure 3.2, we graphically assess these two additional assumptions. The set of graphs allows us to compare trending behavior across exposure groups for the average probability of IMF program participation (top left panel) and the outcome variables of interest: health expenditure (top right), income inequality (bottom left), and health outcomes represented by infant mortality (bottom right). The above-​mean exposure group is portrayed by the solid line and the below-​mean exposure group by the dashed line. For all outcomes, we find both exposure groups to be similar in terms of their trending patterns. Furthermore, there is no trend similarity between IMF liquidity, as shown by the dash-​dotted line, and the average probability of IMF program participation, or between IMF liquidity and any of the outcomes, among countries exposed to above-​average participation. Consequently, there is no apparent violation of the statistical assumptions of the difference-​in-​difference approach.

An Excludable Instrument for IMF Conditionality An excludable instrument is also needed to account for statistical bias linked to IMF conditionality. Our solution is to deploy the same statistical insights regarding interactions of nonexcludable with excludable variables and exploit the same source of exogenous variation as we did for the program participation instrument. Thus, our instrument for IMF conditionality is the interaction of the within-​country average of the number of conditions across the period of interest with the year-​on-​year budget constraint of the IMF. The predicted values for IMF conditionality in Equation (2) are derived as follows:

(

)

 IMFCOND = α1 Xit + α 2 IMFCONDi × IMFBUDGt + µi + δt (9) it

 is the predicted number of IMF con­ Here, i is country and t is year. IMFCOND ditions; IMFCOND  is the average number of conditions in country i, which will vary across countries but remain the same over time within each country; IMFBUDG is the budget constraint of the IMF in year t, which will vary across years but remain the same across countries for each year; X is a list of covariates from the outcome equation and the current account on the balance of payments as a share of GDP (Lang, 2021); μ are country fixed effects; and δ are year fixed effects. Again, our measure of the budget constraint is the natural logarithm of

Probability of IMF participation

Year

2010

2015

2020

Year

2010

2015

2020

Above-mean exposure

4.5

2005

34

2000

5

36

1995

5.5

6

6.5

7

38

40

42

44

Income inequality

4.5

2005

0

2000

5

.2

6

6.5

7

5.5

1995

IMF participation

.4

.6

.8

1 Natural log of IMF liquidity ratio Natural log of IMF liquidity ratio

Year

2010

Year

2015

2020

2020

5.5

6

6.5

7

4.5

IMF liquidity

4.5 2010

2015

20 2005

Health outcomes

2005

5

5.5

6

6.5

7

5

2000

2000

Health expenditure

30

40

50

60

1995

1995

70

0

1

2

3

4

Below-mean exposure

Government health spending (% of GDP) Infant mortality (per 1,000 live births)

Figure 3.2  Parallel trends in IMF program participation instrument

Gini in disposable income

Natural log of IMF liquidity ratio Natural log of IMF liquidity ratio

66  40 Years of Structural Adjustment the IMF’s liquidity ratio. And our identifying assumption is that the outcome of interest in countries with different exposure to conditionality will not be affected differently by changes in the IMF’s budget constraint other than through the impact of conditions. In terms of instrument relevance, the cross-​sectional average of conditionality approximates the general propensity of a country to obtain a specific amount of conditions in any given year. Furthermore, research shows that on average, the IMF raises the number of conditions per program when country demand for loans is strong, and reduces the number of conditions when country demand for loans is weak (Chapman et al., 2017; Dreher & Vaubel, 2004b). The rationale for this relationship is that as the IMF assists more countries, resource scarcity—​ measured here as IMF liquidity—​prompts them to assign more conditions to any given country as a safeguard measure for loan repayments (Dreher & Vaubel, 2004b; Vreeland, 2003). The inverse also holds: we have described already how the IMF is more generous with its loans when it has high liquidity in order to maximize revenues from interest payments and maintain a position of global power; this implies less conditions in times of resource abundance as the IMF tries to entice borrowers into programs. Indeed, James Vreeland (2003) demonstrates in a conditionality game that a greater number of conditions decreases country demand for IMF loans, so it would be rational for the IMF to reduce the number of conditions if it wishes to entice countries to borrow. In subsequent chapters, we conduct additional tests checking for instrument relevance. In terms of instrument excludability, our explanation follows a similar logic as Lang’s (2021) IMF participation instrument vis-​à-​vis the exogenous variation of the budget constraint. The instrument fulfils the exclusion criterion—​that is, it does not affect health expenditure, income inequality, or health outcomes—​ because country-​specific changes in conditionality that deviate from its long-​ run average are brought about only by decisions of the IMF that do not pertain to any given country. Some of these decisions include the introduction of social spending floors in the late 1990s or the conditionality streamlining initiative of the early 2000s (IMF, 2001c). One might be concerned about potential direct effects of the general propensity of a country to obtain a specific amount of conditions in any given year on the outcome of interest (i.e., not just via IMF conditionality in a particular country-​year). As with the IMF program participation instrument above, we control for this effect through the inclusion of country fixed effects in both conditionality and outcome equations. There could also be a question on the excludability of the liquidity variable insofar as wealthy member countries can replenish IMF resources in response to a greater number of countries participating in programs, which would diminish the Fund’s risk aversion such that the organization is willing to agree to fewer conditions when bargaining a new program with a recipient country (Dreher & Vaubel, 2004b). As

How to Evaluate the Effects  67 pointed out earlier, this logic is flawed because the amount of financial resources that members commit to the IMF’s General Resource Account is predetermined via five-​yearly quota reviews, and fluctuations in voluntary contributions to the IMF’s concessional lending budget are accounted for by the inclusion of year fixed effects in both conditionality and outcome equations. Since our conditionality instrument is again akin to a continuous difference-​ in-​ differences statistical design, trends over time in the number of IMF conditions and the outcome variable should be similar across above-​mean conditionality exposure and below-​mean conditionality exposure groups of countries (i.e., parallel trends). Inference would also be threatened if similarly shaped nonlinear trends in the IMF liquidity ratio, the number of IMF conditions, and our outcomes of interest if these trends only occur among high-​exposure countries (i.e., nonoverlapping trends). In Figure 3.3, we find both exposure groups to be similar in terms of their trending patterns with respect to the average number of conditions (top left panel), health expenditure (top right), income inequality (bottom left), and health outcomes (bottom right). Furthermore, there is no trend similarity between IMF liquidity and the average number of conditions, or between IMF liquidity and any of the outcomes, among high-​exposure countries. We adapt the same instrument for disaggregated counts of conditions. Instruments are constructed for each condition type within a single model based on the interaction of the within-​country average of that condition type with IMF liquidity. For example, if we wanted to know the effect of structural conditions on government social spending, then our instrument would be the interaction of the within-​country average of the number of structural conditions with IMF liquidity. Recall that all conditions are jointly included in the outcome equation, so we would also include a variable counting the number of quantitative conditions, instrumented by the interaction of the within-​country average of the number of quantitative conditions with IMF liquidity. The predicted values for disaggregated counts of conditions in Equations (5) and (6) are thus derived as follows:

(

)

(

)

 1 = λ X + λ IMFCOND1 × IMFBUDG + µ + δ (10)   IMFCOND it 1 it 2 i t i t  2 = ζ X + ζ IMFCOND2 × IMFBUDG + µ + δ (11)   IMFCOND it 1 it 2 i t i t  1 and IMFCOND  2 are the predicted values for the number Here, IMFCOND of disaggregated conditions (e.g., structural and quantitative conditions), and IMFCOND1 and IMFCOND2 are their respective averages in country i over the period of interest. We test for instrument relevance in subsequent chapters; in select instances, these instruments do not reach the statistical threshold of

Mean number of conditions

Year

2010 2015

2020

Year

2010

2015

2020

Above-mean exposure

4.5

2005

34

2000

5

36

1995

5.5

6

6.5

7

38

40

42

44

Income inequality

4.5

2005

0

2000

5

5

6

6.5

7

5.5

1995

IMF conditionality

10

15

20

25

Figure 3.3  Parallel trends in IMF conditionality instrument

Gini in disposable income

Year

2010

Year

2010

2015

2015

2020

2020

5.5

6

6.5

7

4.5

Government health spending (% GDP) Infant mortality (per 1,000 live births)

Natural log of IMF liquidity ratio Natural log of IMF liquidity ratio Below-mean exposure

IMF liquidity

4.5 2005

Health outcomes

2005

20 2000

2000

5

5.5

6

6.5

7

5

1995

1995

Health expenditure

30

40

50

60

70

0

1

2

3

4 Natural log of IMF liquidity ratio Natural log of IMF liquidity ratio

How to Evaluate the Effects  69 a Kleibergen-​Paap F-​statistic above 10. We acknowledge where this is the case and discount the results. Checks for parallel and nonoverlapping trends for each disaggregated condition count on the outcomes of interest showed no issues throughout.3 Overall, our approach addresses the main sources of statistical bias that one can expect when examining the effects of IMF intervention: first, countries select into programs based on unobservable characteristics that also influence the outcome of interest, and second, unobservable characteristics also determine the type and amount of conditions that countries receive. Our approach enables scholars to account for program heterogeneity and to distinguish effects of conditionality from other aspects of IMF operations. Elsewhere, we demonstrate via Monte Carlo simulations that this strategy is unbiased and performs better than popular alternatives on standard diagnostics across a range of scenarios (Stubbs et al., 2020).

Roadmap for Subsequent Analyses In order to investigate the impact of IMF intervention on each of our outcomes of interest, we apply a standard methodological template to the analyses undertaken in Part II of this book. We test several features of IMF conditionality. First, we examine the aggregate effect of the total number of conditions. We only count binding conditions (known as “prior actions” or “performance criteria”), following established procedures in this field of study (Copelovitch, 2010b; Rickard & Caraway, 2014; Woo, 2013). Binding conditions directly determine scheduled disbursements of loans and must be implemented for the program to continue, whereas nonbinding conditions serve as markers for broader progress assessment and nonimplementation does not automatically suspend the loan (IMF, 2001d). Our chosen measure also allows us to empirically isolate a conditionality effect from an effect—​or lack thereof—​due to country noncompliance, since the binding character of these conditions precludes the possibility of the latter (Dreher, 2006; Vreeland, 2003, 2006). Second, we disaggregate our total count to test the impact of different forms of conditions. Hypothesized pathways covered in each chapter suggest both quantitative and structural conditions could influence the outcome of interest in some way. Quantitative conditions are measured as a count of the binding quantifiable macroeconomic targets that countries must meet and maintain during the program, typically monitored at quarterly intervals and composing the majority of 3 Results of these tests are available online in the supplementary statistical code (www.imf​moni​ tor.org).

70  40 Years of Structural Adjustment conditionality. Structural conditions are measured as a count of the binding microeconomic reforms that alter the underlying structure of an economy. Recall that the IMF formally designates conditions under these categories (Bird, 2009; IMF, 2015a). Third, we evaluate eight mutually exclusive condition policy areas, again counting the number of binding conditions in each area. This is so that we can identify specific policy pathways linking IMF conditionality to the outcome of interest. Four are core IMF policy areas. External debt issues refer to conditions on external debt management and external arrears. Financial sector, monetary policy, and Central Bank includes conditions on financial institution regulation, financial state-​owned enterprise privatization, treasury bills, interest rates, Central Bank regulation, money supply, and domestic credit. Fiscal issues, revenues, and taxation incorporates conditions on expenditure administration, fiscal transparency, audits, budget preparation, domestic arrears, fiscal balance, customs administration, tax policy, tax administration, and audits of private enterprises. External sector (trade and exchange) indicates conditions on foreign reserves, trade liberalization, exchange rate policy, capital account liberalization, and foreign direct investment. Another four are in non–​core IMF policy areas. State-​owned enterprise privatization, reform, and pricing pertains to conditions related to state-​owned enterprise restructuring, subsidies, price liberalization, audits, marketing boards, corporatization and rationalization, and nonfinancial state-​owned enterprise privatization (including liquidation and bankruptcy proceedings). Labor (public and private sector) is conditions on wage and employment limits, pensions, and social security institutions. Institutional reforms include conditions on judicial system reforms, anti-​corruption measures, competition enhancement, private sector development, devolution, sectoral policies, social policies (excluding poverty reduction policies), price increases for basic needs goods (food, water, public transport, and so on), land registries, granting of property rights, environmental regulations, and access to commons. Last, poverty reduction policies refer to Poverty Reduction Strategy Paper development, increases in social sector spending, and implementation of social safety nets. By examining disaggregated policy areas, we aim to identify the particular mechanisms by which conditionality affects the outcome of interest. Beyond the specific effects of conditionality, alternative channels of IMF program influence are also captured throughout the analyses. These include, inter alia, credit injections (Dreher, 2006), aid and investment catalysis (IMF, 2004d; Stubbs et al., 2016), scaled-​up technical assistance (Broome & Seabrooke, 2015; IMF, 2016a), and moral hazard (Dreher & Walter, 2010). In statistical parlance, we collapse these aspects of the IMF’s operations into a binary indicator coded 1 if a country had an active IMF program in a given calendar year for at least five

How to Evaluate the Effects  71 months of the year, and 0 otherwise (Dreher, 2006). All IMF variables enter the model one year earlier than the outcome of interest, or lagged, to allow for some delay in the realization of effects. Outside of IMF intervention, we also account for the impact of confounding factors in all analyses. For each outcome of interest, we include a set of control variables in our statistical models. Since the appropriate controls will vary based on the outcome, we describe these in their respective chapters and provide a full list of data sources. Notwithstanding these controls, we include country fixed effects (i.e., a set of binary indicators for each country in the sample) to account for all time-​invariant country-​level characteristics, such as legal origin, geography, or ethnolinguistic fractionalization. We also include year fixed effects (i.e., a set of binary indicators for each year in the sample) to account for time-​ variant confounds that all countries are subject to, such as fluctuations in world commodity prices or the overall level of IMF lending. Inclusion of both country and year fixed effects makes for a stringent test as to whether IMF conditionality affects the outcome. Our maximum sample of countries for all multivariate statistical analyses performed in this book are the 189 IMF member-​states as of 2019 (IMF, 2022a). We therefore exclude countries that became members after 2019 (e.g., Andorra), are no longer or never were IMF members (e.g., Cuba, Liechtenstein, Monaco, and North Korea), or are dependent territories (e.g., Cook Islands, Greenland, Hong Kong, Macau, and New Caledonia). In practice our analyses cover less than the full complement of IMF member-​states because of missing data on outcome and control variables—​a common challenge to research using country-​ year observations. Following standard practice, we confront missing data by employing casewise deletion: the observation (i.e., country-​year) is excluded from the analysis if at least one variable contains a missing value. Shortcomings of this strategy are that it ignores information carried by nonmissing values for deleted observations and can introduce bias if data is not missing at random. An increasingly popular alternative to casewise deletion is multiple imputation—​ where missing values are replaced with substituted values that are estimated based on other available information (Honaker & King, 2010; Lall, 2016). We eschew this approach because it yields results that are frequently more biased (i.e., deviate further from the true value of the parameter being estimated) and less efficient (i.e., produce larger standard errors) than casewise deletion (Arel-​ Bundock & Pelc, 2018; Pepinsky, 2018). Like most social scientists before us, we are simply unable to completely absolve our analyses of whatever missing data bias may exist. Finally, our threshold for reporting results is when we can say with at least 90 percent statistical certainly that the estimated effect of IMF conditionality is distinguishable from zero, or what is known in statistical parlance as a cutoff

72  40 Years of Structural Adjustment p-​value of 10 percent (p