## ppea2019017

## Source details

**Canonical URL:** [ppea2019017](https://www.imf.org/-/media/files/publications/pp/2019/ppea2019017.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/pp/2019/ppea2019017.pdf.md)
- [Structured JSON version](/-/media/files/publications/pp/2019/ppea2019017.pdf.json)

---

### Overview
- Date on document: June 14, 2019
- Purpose: Account of consultations and analysis supporting “A Strategy for IMF Engagement on Social Spending.”
- Background papers summarized:
  - Paper I: consultation with CSOs, unions, academics, economists, social spending experts, and IDIs.
  - Paper II: cross-country empirical analysis of whether IMF-supported programs protected social spending.
  - Paper III: survey of IMF Area Department mission chiefs.
  - Paper IV: issues in advising on targeting of transfers (trade-offs of targeting approaches).
  - Paper V: text mining analysis of IMF surveillance and program staff reports since the late 1970s.

### Paper I — Consultation with Third Parties: Key findings
- Stakeholder engagement:
  - Consultative group of 12 representatives convened April 2018 and consulted at all stages.
  - Online consultation ran May to July 2018 in English, French, Arabic, Bahasa, and Spanish.
  - Events included IMF Spring Meetings 2018 townhall, CSO policy forum, ITUC–WB/IMF meetings (March 2018 and March 2019), ITUC-Asia Pacific Symposium (October 2018), ITUC conference on Financing Social Protection (September 2018), IMF Annual Meetings panel (October 2018), one-day LSE workshop (November 2018), bilateral consultations with ILO, UNICEF, World Bank.
- Aggregate views:
  - Broad definition of social spending preferred (social protection, education, health; go beyond “basic”).
  - Strategy should be consistent with providing social protection to all, not just the poor.
  - Stakeholders wanted specific policy guidance in the Board paper.
  - Social spending viewed as macro-critical for tackling inequality, promoting social cohesion, and stabilizing the economy.
  - Fund urged to work closely with IDIs beyond the World Bank (ILO, UN bodies, WHO) and with local CSOs.
  - Social spending framed as long-term investment; Fund’s value-added is helping countries create fiscal space.
  - Many participants advised the Fund to be less biased toward targeted transfers and more supportive of universal transfers.
- Outcome: Comments were considered in developing the Board paper.

### Paper II — Empirical evidence on IMF programs and social spending (Scope, data, main results)
- Time period analyzed: 2000–2016 (starting year chosen due to WHO Global Health Expenditure Database).
- Program sample (2000–2016): Total 283 programs approved.
  - 123 GRA programs.
  - 160 PRGT programs.
  - 13 blended arrangements (treated as PRGT in analyses).
  - 18 PSI arrangements.
- Three evaluation measures used:
  - Spending as a share of GDP.
  - Real per capita spending.
  - Spending in percent of total government spending.
- Main empirical result:
  - On average, no difference between spending trends in program countries and similar non-program countries.
  - Spending declined in over one-quarter of program years for both education and health across all three measures.
  - For real per capita education and health spending, on average spending continued to increase over program life; regressions with country fixed effects suggest increases were higher in later program years relative to approval year.
  - Exception: education spending as a share of GDP and as a share of spending fell in countries with GRA programs; real per capita education spending on average protected across facilities.
- Selected extrema and ranges (as reported in figures and Appendix Table 2):
  - Change in health spending (% of GDP): Max = 1.8, Min = -1.9.
  - % change in real per capita health spending: Max = 125.9, Min = -58.9; other panel Max = 568.2, Min = -55.6.
  - Change in health spending (% of total expenditure): Max = 7.9, Max = 12.5, Min = -15.8, Min = -4.9.
  - Change in education spending (% of GDP): Max = 1.7, Min = -1.1; other panel Max = 4.1, Min = -3.9.
  - % change in real per capita education spending: Max = 43.1, Min = -31.3; other panel Max = 276.2, Min = -43.7.
  - Change in education spending (% of total expenditure): Min = -4.1, Min = -13.2, Max = 4.4, Max = 13.1.
- Figure 3 selected average marginal effects (changes from 25th to 75th percentile of regressors; significance markers preserved):
  - Health spending (percent of GDP) panel: -0.033, -0.019, 0.045**, -0.038**, 0.029.
  - Education spending (percent of GDP) panel: 0.062***, -0.041, 0.059***, -0.05***, 0.138***.
  - Health spending (real per capita percent change) panel: 0.02, 0.01, 0.076***, -0.063***, 0.2***.
  - Education spending (real per capita percent change) panel: -0.083***, 0.04***, 0.065***, -0.057***, 0.052***.
  - Regressors (in order shown): Real GDP Growth; Inflation; Fiscal Balance Change / GDP; Revenue / GDP; Init. Education Spending / GDP (or Init. Health Spending / GDP).
- Robustness: similar results when including debt-to-GDP ratio among controls.

### Paper II — Key empirical interpretations and methodological notes
- Higher real GDP growth:
  - Positively correlated to the likelihood of large drops in education spending when expressed as a share of GDP (mechanical ratio effects).
  - Associated with a lower probability of a sharp decline in real social spending per capita.
- Inflation:
  - Higher inflation associated with higher probability of a sharp decline in social spending (via higher nominal GDP growth).
- Fiscal consolidation:
  - Larger short-term fiscal consolidation increases probability of declines in spending.
  - Revenue-based consolidation attenuates the probability of social spending declines relative to expenditure-based consolidation (revenue increases reduce the probability).
- Initial spending levels:
  - Declines more likely where initial spending is high (may reflect spending inefficiencies).
- Estimation approach:
  - IPWRA (Inverse Probability-Weighted Regression Adjustment) doubly robust estimator used to estimate ATT for 101 lending arrangements with at least two years duration; point estimates statistically insignificant across three years after approval in Appendix Tables 6 and 7.
  - Program definition treats each lending arrangement as a separate program; sample constructed using rolling three-year windows (t-1 is year prior to program approval).
  - Pre-treatment participation variables include per capita GDP (PPP), real per capita GDP growth, cash balance and government debt (% of GDP), reserves (months of imports), external debt (% of GDP), trade and capital account balances, PRGT dummy.

### Paper II — Policy-relevant implications
- Strengthen measures that reinforce growth and revenue mobilization to help avoid short-term declines in social spending.
- Revenue-based consolidation preferable to expenditure-based consolidation for protecting social spending.
- Address spending inefficiencies (e.g., high wages, high pharmaceutical costs) to reduce likelihood of cuts in efficient spending components.
- Program design should consider initial spending levels; spending better protected in countries with relatively low initial spending.
- Program indicators constructed at the lending-arrangement level improve insights into heterogeneity and duration effects.

### Paper III — Mission Chiefs’ Survey: main results (fieldwork July–December 2018)
- Coverage:
  - Survey targeted mission chiefs for IMF membership plus select jurisdictions; overall response rate: about 80 percent.
  - Response rates by income group: EMEs 73 percent; LIDCs 81 percent; AEs 89 percent.
  - Fragile states sub-sample response rate: 83 percent.
- Macro-criticality of social spending:
  - Almost 80 percent of mission chiefs (MCs) see social spending as macro-critical.
  - LIDCs: almost 90 percent; EMEs: 80 percent; AEs: almost 60 percent; fragile states: 85 percent.
  - By region: about 90 percent in emerging and developing Europe, MENAP and Sub-Saharan Africa.
- Drivers cited for macro-criticality (multiple responses allowed): Social/political stability; distributional objectives; large human capital gaps (education, health); large social protection gaps; inefficiency of social spending; future/current social spending pressures; demographic factors; reforms; conflict/refugees; shocks.
- Sources of analysis/expertise used by country teams:
  - Team’s own resources: 80 percent.
  - IMF departmental analysis/TA: about 50 percent.
  - External IDIs: almost 80 percent.
  - World Bank: important source for EMEs (60 percent of EMEs MCs) and LIDCs (53 percent of LIDCs MCs); low use for AEs (6 percent).
- Interaction modalities:
  - Mainly bilateral discussions with sectoral experts (HQ or missions); collaboration on analytical projects less frequent (18 percent).
  - Mapping of interactions by topic (percent of total interactions): Grand total: World Bank 45.8; OECD 1.5; Other regional development banks 11.0; ILO 1.2; UN Agencies 10.4; Local Development Partners 10.2; NGO 8.2; Academics 9.2; Other 2.5; Total 100.0.
- Constraints limiting engagement: competing priorities, poor data quality/availability, lack of expertise, limited authority interest, country capacity constraints, inability to draw on outside expertise.
- Policy advice prevalence:
  - Surveillance countries: 67 percent received recommendations on social spending reforms.
  - Program countries: 69 percent.
  - Introduction/expansion of means-tested schemes recommended in 64 percent of cases overall; expansion of non-means-tested programs recommended in about 18 percent.
- Conditionality and outcomes in programs:
  - Nearly all IMF programs include conditionality aimed at protecting/strengthening social spending.
  - MCs view conditionality effectiveness: 63 percent effective; 22 percent neutral.
  - Mechanisms used: Indicative targets (floors): 70 percent; MEFP commitments: about 40 percent; Structural benchmarks: 24 percent.
  - MCs reported conditions are met most of the time, but 75 percent acknowledged at least one conditionality was missed during program life (reasons: shortfalls in external donor financing, government revenue shortfalls, lack of ownership).

### Paper IV — The debate on universal and targeted transfers (issues, trade-offs, financing)
- Purpose: Set out issues to consider when advising on targeting of transfers and trade-offs across targeting approaches.
- Conceptual points:
  - Universal and targeted approaches are complementary; the Fund does not have a bias in favor of one approach.
  - Targeting implementation difficulties exist; greater reliance on universal-type transfers typically involves fiscal costs that need financing via efficient and progressive taxation.
  - Choice depends on country economic, political, and social circumstances and constraints.
- Targeting designs discussed:
  - Means-tested transfers: theoretical efficiency under perfect targeting; practical limits due to administrative capacity, informal sector, volatile incomes.
  - Categorical targeting: eligibility by characteristics (children, elderly, location); trade-offs between excluding poor without characteristic vs. including non-poor with characteristic.
  - Proxy-means testing (PMT): continuous welfare score; advantages in poverty impact for a given budget; challenges include leakage, undercoverage, need for regular updating, and horizontal inequity around cutoffs.
  - Tiered PMT (differentiated universal benefits): universal coverage with differentiated benefit levels (tiering). Example outcomes preserved:
    - Under a UBI, the poorest 30 percent receive 30 percent of the fixed transfer budget (by design).
    - Tiered PMT with benefit ratios 4:2:1 across bottom three PMT deciles / next four deciles / highest three deciles: bottom three deciles receive over 40 percent of benefits; richest three deciles receive just above 15 percent of benefits.
    - A PMT targeting 50 percent of the population has a slightly higher share to bottom three deciles than the tiered PMT but comes with significant undercoverage of lower welfare deciles.
- Financing coverage expansion: recommended instruments and considerations:
  - Strengthening personal income taxes (PITs), corporate income taxation, and broadening PIT coverage of wages/salaries.
  - Strengthening consumption taxes (broad-based, minimize rate differentiation); setting tax registration thresholds at reasonable levels to enhance progressivity.
  - Expanding efficient excises (fossil fuels, alcohol, tobacco, sugar) to raise revenues and address externalities.
  - Caveats: broadened consumption taxes can increase burdens on vulnerable groups; safety nets must protect these groups; strengthening revenue administration is typically required.

### Paper V — Trends and patterns in Fund engagement: text mining analysis
- Method:
  - Term frequency normalized by document word count and scaled by 10,000; averaged across staff reports per country-year.
  - List of social spending-related terms pre-defined; included general and specific social spending concepts and distribution analysis concepts.
- Main results:
  - Discussion of social spending and inequality increased significantly over past decades; peak in 1999 (PRGF introduction).
  - Streamlining during 2009–2011 and global financial crisis temporarily crowded out attention.
  - By 2018, Fund documents had about 20 (normalized) occurrences of social spending related terms on average (equivalent to about 40 unnormalized occurrences per report), with variation across countries.
  - Surveillance reports show somewhat higher normalized frequency than program reports particularly after 2000.
- Cross-country variation:
  - Wide variation in intensity across countries; examples of repeated high-attention surveillance countries: Austria, Belgium, Finland, Netherlands, Norway, Luxembourg.
  - Program report outliers include Argentina, Bolivia, Kazakhstan, Malawi, Peru.
- Topic emphasis by country group:
  - Pensions, health, and education most-discussed.
  - Pensions prominent in advanced economies; health and education more prevalent in low-income countries.
- Distributional analysis:
  - Distributional discussion followed similar temporal pattern; increased sharply in program reports between 1999 and 2005, dipped in 2009, and more prominent in PRGT programs than GRA programs.

### Appendix I — Selected tables and numeric highlights
- IMF Lending Arrangements Approved, 2000–16 (Appendix Table 1):
  - Total: 283
  - GRA: 123
  - EFF: 17
  - FCL: 17
  - PCL: 1
  - PLL: 3
  - SBA: 85
  - PRGT: 160
  - ECF: 51
  - ECF-EFF: 4
  - ESF: 11
  - PRGT: 62
  - PRGT-EFF: 3
  - PSI: 18
  - SBA-ESF: 1
  - SBA-SCF: 5
  - SCF: 5
- Binary variable for large declines (Appendix Table 2) — dummy construction ranges:
  - Change in Health Spending (% of GDP): (-0.2, 3.1) / [-3.0, -0.2]
  - % Change in Health Spending (real per capita): (-10.1, 568.2) / [-64.2, -10.7]
  - Change in Education Spending (% of GDP): (-0.4, 5.1) / [-3.9, -0.4]
  - % Change in Education Spending (real per capita): (-7.1, 276.2) / [-43.7, -7.5]
- Probit/Panel analysis (selected coefficients and marginal effects; significance markers preserved):
  - Fiscal Balance / GDP: positive and often significant (examples: 0.053***; average marginal effects 0.014***).
  - Revenue / GDP: negative and often significant (examples: -0.062***; average marginal effects -0.016***).
  - Initial Health/Education Spending / GDP: positive and significant in many specifications (examples: 0.249***; average marginal effects 0.065***).
  - Observations in selected columns: 309, 203, 106; number of observations varies by specification.
- IPWRA outcome examples (Appendix Tables 6–9):
  - Government Expenditure on Health — Percent of GDP: Impact of IMF program on outcome equation for treated countries (examples): 2nd: -0.009 (0.05); 3rd: -0.023 (0.08); 4th: -0.096 (0.14).
  - Government Expenditure on Education — Percent of GDP: Impact of IMF program on outcome equation for treated countries (examples): 2nd: -0.125 (0.09); 3rd: -0.086 (0.14); 4th: -0.016 (0.24).
  - Observations and program cases vary by table (e.g., Health tables show Observations: 997, 997, 859, 859, 729, 729; Program cases: 136, 136, 101, 101, 60, 60).
- Replications of past methodologies (Appendix Table 5) — selected program coefficients:
  - IMF program coefficient examples:
    - Fixed effect (Education % of GDP): 0.26* (0.133).
    - System GMM (Education % of GDP): 0.22** (0.101).
    - Fixed effect (Health % of GDP): 0.17** (0.070).
    - System GMM (Health % of GDP): 0.27*** (0.094).
  - Lagged dependent variables commonly high and significant across specifications (examples: 0.71***, 0.85***, 0.61***).
- Mission Chiefs’ Survey summary:
  - Almost 80 percent of MCs see social spending as macro-critical.
  - Own resources used by 80 percent of MCs; IDIs used in almost 80 percent of cases.
  - Introduction/expansion of means-tested schemes recommended in 64 percent of cases overall.
  - Program conditionality mechanisms used: Indicative targets (floors) 70 percent; MEFP commitments ~40 percent; Structural benchmarks 24 percent.
  - 75 percent of MCs acknowledged at least one missed conditionality during program life.

*IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS (EXECUTIVE SUMMARY), June 14, 2019*

### EXECUTIVE SUMMARY

### EXECUTIVE SUMMARY

### Overview
- This Supplement presents an account of the extensive consultations and the results of analysis that supported the definition of “A Strategy for IMF Engagement on Social Spending.”
- Papers summarized:
  - Paper I: summarizes comments received from civil society organizations (CSOs), unions, academics, economists, social spending experts, and international development institutions (IDIs).
  - Paper II: cross-country empirical analysis assessing whether IMF-supported programs adequately protected social spending.
  - Paper III: results of a survey of IMF Area Department mission chiefs.
  - Paper IV: issues to consider when providing policy advice on the targeting of transfers and trade-offs involved.
  - Paper V: documents how discussion of social spending issues evolved in IMF surveillance and program staff reports since the late 1970s.
- Date on document: June 14, 2019

### Paper I — Consultation with Third Parties: Key Findings
- Stakeholder composition and process:
  - Consultation included CSOs, unions, academics, economists, social spending experts, and IDIs.
  - A consultative group consisting of 12 representatives from NGOs, unions, academics, economists, and social protection experts was created in April 2018 and consulted at all stages.
  - An online consultation ran from May to July 2018 in English, French, Arabic, Bahasa, and Spanish.
  - Events and engagements included IMF Spring Meetings 2018 townhall and CSO policy forum, ITUC–WB/IMF meetings (March 2018 and March 2019), ITUC-Asia Pacific Symposium (October 2018), ITUC conference on Financing Social Protection (September 2018), a high-level panel at the IMF Annual Meetings (October 2018), a one-day workshop at LSE (November 2018), and bilateral consultations with IDIs including the ILO, UNICEF, and the World Bank.
- Aggregate views from participants:
  - Most participants agreed with using a broad definition of social spending.
  - Most participants requested that the strategy be consistent with providing social protection to all, not just the poor.
  - Most participants would have preferred if the Board paper provided specific policy guidance.
  - Participants suggested social spending is always macro-critical, highlighting its role in tackling inequality, promoting social cohesion, and stabilizing the economy.
  - Participants asked that the Fund work closely with other IDIs beyond the World Bank, such as the ILO, UN bodies, and WHO, and with local CSOs.
  - Participants encouraged the Fund to consider social spending as a long-term investment, not a cost to be contained, and saw the Fund’s value-added in helping countries create fiscal space for social spending.
  - Most participants advised the Fund to be less biased toward targeted transfers and more supportive of universal transfers.
- Specific consultations and inputs referenced:
  - Letter by 53 economists to the IMF Managing Director and statement from the Global Coalition on Social Protection Floors were noted examples of stakeholder interest.
  - Staff began regularly attending the UN Social Protection Inter-Agency Cooperation Board (SPIAC-B) in early 2018.
- Outcome:
  - The comments received were considered and addressed in developing the Board paper.

### Paper II — Empirical Evidence on IMF Programs and Social Spending
- Scope and main result:
  - The cross-country empirical analysis assesses whether IMF-supported programs adequately protected social spending.
  - The analysis confirms that, on average, there is no difference between spending trends in program countries compared to similar countries without a program.
  - However, the analysis finds that in a significant number of instances spending decreased in program countries.
  - Consequently, the paper examines factors affecting the probability of a decline in social spending.
- Contents and methodological notes (as summarized in contents):
  - Sections include Data and Measurement; Social Spending Trends During Programs; The Impact of IMF Programs on Social Spending; Main Implications and Conclusions.
  - Box: Inverse Probability-Weighted Regression Adjustment (IPWRA) Estimation.
  - Figures highlight trends and probabilities: Figures 1–5 cover trends in public spending on health and education during and outside IMF programs, average marginal effects on probability of large declines, and sample construction.

### Paper III — Mission Chiefs’ Survey: Main Results
- Purpose:
  - Presents results of a survey of IMF Area Department mission chiefs on the nature and extent of engagement on social spending and challenges faced.
- Key findings:
  - A vast majority of mission chiefs regard social spending as macro-critical for their country.
- Contents:
  - Sections include Macro-criticality of Social Spending Issues; Resources for Addressing Social Spending Issues and Interaction with Other Institutions; Policy Advice; Programs: Objectives and Conditionality.
  - Figures cover response rates by region and income group, perceptions of macro-criticality, reasons why social spending is macro-critical, factors affecting IMF country teams’ engagement, and program conditionality on social spending.
  - Table maps interaction with other institutions by topics and counterparts.
  - Annex includes the Questionnaire.

### Paper IV — The Debate on Universal and Targeted Transfers
- Focus:
  - Sets out issues to consider when providing policy advice on targeting of transfers and trade-offs involved in different targeting approaches.
- Key points:
  - Highlights challenges to achieving greater coverage and financing it.
  - Points out need to consider both tax and transfer sides when designing redistributive fiscal policy.
- Specific topics addressed:
  - Means-tested transfers; Categorical targeting; Proxy-means testing; Financing transfers.
  - Figures illustrate coverage under alternative categorical programs, benefit share/level/coverage, and distributional impact of tax and transfer programs.

### Paper V — Trends and Patterns in Fund Engagement: Text Mining Analysis
- Objective:
  - Documents evolution of discussion of social spending issues in IMF surveillance and program staff reports since the late 1970s using text mining.
- Key findings:
  - Discussions of social spending and inequality issues increased significantly over past decades.
  - There is wide variation in the intensity of the discussion of social spending issues across both surveillance and program staff reports.
- Methodology and outputs:
  - Sections include Description of the Database Used for Text Mining; Methodology; Results.
  - Box lists social spending related terms used in the text mining analysis.
  - Figures present overview of Fund documents by type and normalized frequency counts of social spending concepts overall, by surveillance vs. program, and by topic.

### Administrative and Production Details
- Approved By: Michael Keen and Kristina Kostial
- Prepared by an inter-departmental team led by David Coady (FAD) and Zuzana Murgasova (SPR) and consisting of Maura Francese, Dominique Guillaume, Nikhil Brahmankar, Wendell Daal, Brooks Evans, Csaba Feher, Emine Hanedar, Emmanouil Kitsios, Jorge Martinez, Delphine Prady, Baoping Shang (all FAD); Fei Liu, Gohar Minasyan, Ke Wang, Irene Yackovlev (all SPR); and Nicolas Mombrial (COM).
- Research assistance: Nghia-Piotr Le.
- Production assistance: Liza Prado.

*IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS (EXECUTIVE SUMMARY), June 14, 2019*

### 3.      The following sections provide an overview of the inputs received, organized into

### 3.      The following sections provide an overview of the inputs received, organized into 

### B. Scope of the Framework and Definition of Social Spending
- General reception:
  - All participants welcomed development of a strategy to guide more effective IMF engagement on social spending issues.
  - Several participants regretted the paper focused on process and preferred guidance on Fund policy in specific areas (pensions reform, minimum wages, unemployment insurance); they hoped these would be addressed in the upcoming Guidance Note.
- Definition and coverage:
  - Most participants agreed with focusing on a broader definition of social spending, including social protection, education and health spending, though they suggested going beyond “basic” education and health.
  - Participants urged the Fund to stress interdependence of social protection, education and health (e.g., “there can be no income security without health security and proper education and vice versa”) and to ensure social protection remains central, not second-order.
  - Recommendation to discuss specific social insurance benefits (old age pensions, disability, maternity) and analyze impact of IMF-supported programs on social protection spending, not just health and education spending.
- Paper’s response:
  - Acknowledges inter-dependent nature of social spending components and that the definition of “basic” differs by country and expands with development.
  - IMF policy advice in social protection, education and health is based on existing Fund Board policy papers listed in Box 7 of the main Board paper; this advice will be summarized in a Guidance Note if and when the IMF’s Executive Board endorses the strategy.
  - Issue of sector-specific policy advice and resources for country-team engagement will be summarized in the Guidance Note.
- Scope suggestions and limits:
  - Some participants suggested including spending on water, sanitation, childcare, housing, minimum wages, physical infrastructure, and disaster prevention/mitigation as relevant to the SDGs.
  - Paper clarifies it focuses on social spending (social protection, education, health). Countries can define broader priority spending areas complementary to social spending; some items (childcare subsidies, social housing) may sometimes fall under social protection.
  - IMF can help achieve these objectives where they are macro-critical, including by creating fiscal space.
- Targeting and approach:
  - Many participants advised against viewing social protection as charity for the poor only; instead view social protection as addressing lifecycle risks affecting everyone and emphasize social pooling of risk.
  - Paper recognizes difference between social assistance (protect households from poverty) and social insurance (protect broader population across lifecycle).
  - Clarifies that “targeting” in the paper refers specifically to social assistance transfers; policy advice in both areas will be taken up in the Guidance Note.
- Footnotes and references noted in source:
  - IEO 2017 report focused on the narrower concept of social protection.
  - For definition of “macro-critical” see the Main Paper.

### C. Rationale and Timing for IMF Engagement
- Views on when to engage:
  - Several participants questioned anchoring engagement only in macro-criticality; argued importance of social spending for growth, inequality, social stability, and international agreements (SDGs) justify routine engagement.
  - Suggested the Fund examine social sector spending trends and performance in all Article IV consultations and be mindful of policy advice impacts through ex-ante impact analysis.
- Paper’s position:
  - IMF committed to supporting members in achieving social objectives consistent with mandate to support macroeconomic and financial stability (see third Section of the Main Paper).
  - Country preferences on social objectives differ; mission chiefs’ survey and text mining confirm IMF engagement on social spending in surveillance and program activities.
  - Strategy highlights existing good practice to make engagement more systematic and recognizes need for sound analytical work drawing on other IDIs where warranted.
- Macro-criticality operationalization:
  - Many participants viewed Fund’s definition of macro-criticality as offering little operational guidance and requested a clear, operationally meaningful definition and technical dialogue at country level with governments and development partners.
  - Paper indicates the Guidance Note will provide illustrative examples to guide teams in evaluating macro-criticality and staff will consider fora for sectoral policy dialogue with external stakeholders.
- Suggested breadth of macro-criticality:
  - Most participants recommended a wide definition of macro-criticality, including roles for reducing inequality, promoting inclusive growth, acting as shock absorber, smoothing incomes, increasing household incomes and demand, formalizing labor markets, coping with climate change, reducing gender inequality, improving productivity, strengthening social/political stability, and meeting the SDGs.
  - LSE workshop participants emphasized addressing persistent poverty and long-term global trends: ageing populations, rapid technological change, globalization.
  - Paper recognizes these various roles and anticipates regular updates to the Guidance Note as Fund experience evolves.

### D. Cooperation with Development Partners and Civil Society
- Role of IDIs and expertise:
  - Consensus that, given limited resources and expertise, the Fund should refrain from a leading role in designing social protection schemes and should increase engagement with and rely more on IDIs (beyond World Bank: UNICEF, UNESCO, WHO, ILO).
  - Strategy should provide guidance on role of these IDIs, their expertise, and how to engage them effectively.
  - IDIs called for more intense and regular inter-agency cooperation on institutions’ overall policy stance and technical advice in specific country cases.
  - IDIs encouraged the Fund to adhere to the ILO Convention 102 and Recommendation 202 and to routinely attend SPIAC-B or Universal Social Protection 2030.
  - Only some participants recommended the Fund develop in-house expertise on social spending issues.
- Paper’s response:
  - Envisages strengthening collaboration with a broad set of stakeholders with social spending expertise; consistent with IEO evaluation views.
  - Emphasizes need for sufficient in-house expertise to appropriately engage with development partners.
- Engagement with CSOs and national stakeholders:
  - Fund was strongly encouraged to work with CSOs, local experts, unions, faith-based groups to improve understanding of country context and policy advice; Fund could play catalytic role nurturing national dialogue on social spending.
  - Paper recognizes importance of two-way communication with national and international stakeholders and the IMF’s catalytic role through analytical work.

### E. The IMF’s Role and Approach
- Shift in perspective:
  - Most participants urged moving from short-term view of social spending as a cost to a focus on long-term economic value; fiscal sustainability assessments should pay attention to long-term impacts and benefits.
  - Inclusive social protection may appear costlier short-term but can be more fiscally sustainable long-term due to broader political support and maintained budgets.
  - IMF should take careful stance on pension reforms or cutting employer social security contributions where advice that focuses on immediate fiscal costs risks long-term unsustainability.
- Paper’s position:
  - Discusses long-term investment nature of social spending, role in promoting growth and equity, and complementarity of social spending components.
  - When significant short-term fiscal adjustment is needed, measures should focus on raising revenue and increasing spending efficiency and progressivity where possible.
  - Where large social spending gaps exist (e.g., in LIDCs), IMF-supported programs typically designed to protect initial levels of social spending and create fiscal space.
  - Paper notes Fund’s Technical assistance (TA) on medium-term revenue strategies.
- Early engagement and continuous objectives:
  - Participants encouraged earlier IMF engagement and viewing income security as continuous objective, not only crisis response.
  - Paper highlights need for early engagement on social spending issues.
- Fiscal space and financing:
  - Most participants saw IMF value-added in advising on creating fiscal space for implementing internationally agreed social protection goals; creating fiscal space should not imply reducing benefits or coverage but ensuring appropriate benefits and coverage financed through additional revenue (e.g., more effective and progressive tax systems).
  - One participant suggested working with other agencies to establish minimum expenditure targets and mobilize resources via progressive taxes.
  - Paper emphasizes IMF role in creating fiscal space and notes Fund is a primary provider of TA in tax policy, revenue administration, and public financial management with significantly increased TA over the last decade.
- Efficiency, equity, and tradeoffs:
  - Fund’s role in making social spending more efficient and equitable and analyzing tradeoffs between policy options seen as important.
  - Paper reflects this via discussion of channels through which social spending can be macro-critical: fiscal sustainability, spending adequacy, and spending efficiency.
- Program engagement and conditionality:
  - Most participants thought IMF should refrain from making specific design of social protection systems a condition for support—these are government decisions and areas of greater expertise for others.
    - Paper clarifies conditionality needs to be critical for program success and social spending conditionality may need development in collaboration with other IDIs.
  - Participants welcomed use of social spending floors but regretted current floors are vague, limited in scope, and vary across countries.
    - Suggested ring-fencing all social spending through social spending floors and expanding them during crises or austerity, while defining content more accurately and keeping flexibility to tailor to government needs.
    - For social protection, floors could be based on the ILO Social Protection Floors Recommendation, 2012 (No. 202), incorporated into the SDGs (target 1.3).
    - Paper highlights importance of effective use and documentation of conditionality, including spending floors tailored to program objectives, critical to program success, country-specific, and based on data availability.
    - Acknowledges quality, disaggregated data are crucial and that GFSM2014 and COFOG standards provide useful frameworks for enhancing data quality.
    - Social spending floors can protect existing spending or increase social spending.
  - Participants felt social spending floors are not sufficiently enforced and suggested turning floors into binding performance criteria.
    - Paper notes IMF Guidelines on Conditionality: Executive Board considers observance of performance criteria (PCs), indicative targets (ITs), and structural benchmarks (SBs) when assessing program reviews.
    - Social spending targets can be established as a PC, but not if there are concerns about data quality.
    - Paper discusses importance of improving quality and timeliness of social spending data for policy analysis and program monitoring; where data quality is an issue, conditionality can address this during the program.

*Italic: Source — Excerpt from the IMF background papers (pp. 11–16) provided in the content unit.*

### 18.      Participants saw the Fund as having, like the World Bank, a bias in favor of targeted

### ppea2019017 - 18.      Participants saw the Fund as having, like the World Bank, a bias in favor of targeted

### IMF position on universal vs. targeted social assistance transfers
- The paper clarifies the views of the IMF on the appropriate use of universal and targeted social assistance transfers: the Fund does not have any bias in favor of one approach.
- Universal and targeted approaches are seen as complementary tools for achieving social objectives.
- The paper and a background note acknowledge implementation difficulties with targeting in some countries and emphasize that greater reliance on universal-type transfers typically involves fiscal costs which need to be financed through efficient and progressive taxation.
- The paper describes advantages and disadvantages of universal versus targeted approaches and highlights trade-offs for policy makers. The appropriate choice depends on country economic, political, and social circumstances and constraints.

### Participants' perspectives and criticisms of targeting
- Participants perceived the Fund (like the World Bank) as having a bias in favor of targeted social benefits based on means testing; this perceived bias was seen as at odds with the SDGs and other agencies’ views (e.g., ILO, UNICEF). (footnote marker 12 present)
- Arguments raised by participants preferring greater emphasis on universal access to social protection:
  - Targeting using means testing has flaws including large exclusion errors and high implementation cost for recipients and countries with limited administrative capacity.
  - Targeted programs only for the poor are likely to get limited political support and therefore may be unsustainable in the long term.
  - Social protection should focus on getting people to participate in economic activity rather than “drip feeding” cash transfers that may keep the poor in unproductive employment.
  - When “all paid in, all gained,” universal schemes create broad public support and strengthen the social contract.
  - Concerns about “leakage to the rich” in the absence of targeting were seen as potentially unwarranted; governments could address leakage through the tax system if chosen for political reasons.
- Several participants (including academics at the LSE workshop) argued the universal vs. targeted dichotomy is overly simplistic; focus should be on national systems that can include both targeted and universal transfers. Targeting should not be limited to proxy-means testing; most social protection schemes are targeted in one way or another (either by income or to vulnerable groups such as children).
- The paper reflects these nuances, in particular in Background Paper IV.

### Background: empirical analysis of IMF programs’ impact on social spending
- Paper title: BACKGROUND PAPER II. IMPACT OF IMF PROGRAMS ON SOCIAL SPENDING: EMPIRICAL EVIDENCE.
- Scope and intent:
  - Analyze whether levels of public education and health spending have been safeguarded in IMF-supported programs.
  - Address methodological and data challenges in existing studies.
  - Answer two questions:
    - Have IMF-supported programs safeguarded the level of public education and health spending?
    - What is the effect of IMF-supported programs on public education and health spending, relative to the counterfactual under which they had not engaged in a program?
- Time period analyzed: 2000–2016.
- Starting year rationale: The year 2000 is the starting year in the current Global Health Expenditure Database provided by the WHO.

### Data definitions and measurement choices
- Government spending on education: refers to all levels of education; data from UNESCO Institute for Statistics; mapped to ISCED using the 2011 UNESCO method; includes current and capital spending and transfers from international sources to government.
- Public health spending: defined as the sum of domestic general government health spending and external health spending channeled through government; refers to current health spending given WHO database limitations; adding external transfers channeled via government captures whether IMF programs catalyze donor assistance.
- Three measures used to evaluate program impact on spending:
  - Spending as a share of GDP.
  - Real per capita spending.
  - Spending in percent of total government spending.
- Pros and cons of measures are detailed (e.g., share of GDP can reflect GDP changes rather than spending changes; real per capita can suffer measurement error; percent of total government spending evaluates protection relative to other spending but can mask overall spending declines).

### Program sample and classification (2000–2016)
- Number of programs overviewed: 283 programs approved over the period 2000–2016.
- Program categorization in sample:
  - 123 GRA programs.
  - 160 PRGT programs.
  - 13 blended arrangements (PRGT with GRA resources; treated as PRGT in analyses).
  - 18 Policy Support Instrument (PSI) arrangements.
- Blended programs: program length taken as the longer duration among both program types.
- LIC facilities prior to PRGT (SAF, ESAF, PRGF) labeled as PRGT-supported programs for study purposes.

### Key empirical findings on social spending trends during programs
- On average, both health and education spending are protected during program years.
- Exception noted: reduction in education spending as a share of GDP and as a share of spending observed in countries with GRA programs; however, real per capita education spending is on average protected across facilities.
- Using the measure percent of total government spending:
  - On average, health spending has been increasing in programs.
  - Some evidence that education spending as a share of total spending fell in GRA programs.
- Spending declined in over a quarter of program years for both education and health, across all three spending measures.
- For real per capita education and health spending, evidence shows on average spending continued to increase over the course of the program; regressions controlling for country fixed effects suggest real per capita health and education spending increases were higher in later years of the program relative to the approval year.
- Spending expressed as share of GDP or total government spending did not show statistically significant differences relative to the approval year, except programs that lasted five years which had significant positive changes in health spending as a share of GDP.

### Selected numeric details from figures and analyses (preserved exactly)
- Figure 1 captioned ranges and extrema for changes in health spending (percent of GDP): Max = 1.8, Min = -1.9.
- For percent change in real per capita health spending: Max = 125.9, Min = -58.9; other panel Max = 568.2, Min = -55.6.
- For change in health spending (percent of total expenditure): Max = 7.9, Max = 12.5, Min = -15.8, Min = -4.9 (values appear as displayed in figure).
- Change in education spending (percent of GDP): Max = 1.7, Min = -1.1; other panel Max = 4.1, Min = -3.9.
- Percent change in real per capita education spending: Max = 43.1, Min = -31.3; other panel Max = 276.2, Min = -43.7.
- Change in education spending (percent of total expenditure): Min = -4.1, Min = -13.2, Max = 4.4, Max = 13.1 (values as shown).
- Figure 2 regression results presented as year-by-year point estimates with 90 percent confidence intervals for:
  - 2nd Program Year, 3rd Program Year, 4th Program Year, 5th Program Year (no numeric point estimates listed in text excerpt except the graphical axis labels).
- Figure 3 average marginal effects on probability of large declines in social spending (changes from 25th to 75th percentile of regressors) — selected coefficients shown exactly as in figure panels:
  - Health spending (percent of GDP) panel: -0.033, -0.019, 0.045**, -0.038**, 0.029.
  - Education spending (percent of GDP) panel: 0.062***, -0.041, 0.059***, -0.05***, 0.138***.
  - Health spending (real per capita percent change) panel: 0.02, 0.01, 0.076***, -0.063***, 0.2***.
  - Education spending (real per capita percent change) panel: -0.083***, 0.04***, 0.065***, -0.057***, 0.052***.
- Regressors represented in Figure 3 (in order shown): Real GDP Growth; Inflation; Fiscal Balance Change / GDP; Revenue / GDP; Init. Education Spending / GDP (or Init. Health Spending / GDP as appropriate).
- Robustness checks: similar results when including the debt-to-GDP ratio among control variables (statement present in text).

### Policy-relevant implications drawn in the text
- Strengthening measures that reinforce growth and revenue mobilization can help to avoid short-term declines in social spending.
- Declines are much more likely where initial spending is high, which may reflect spending inefficiencies.
- Program design could be further strengthened, where warranted, to avoid declines in social spending; empirical exploration identifies macroeconomic and fiscal conditions affecting the probability of large declines.

*IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS (excerpt provided in source content)*

### 11.      Results suggest that higher real GDP growth is positively correlated to the likelihood

### 11.      Results suggest that higher real GDP growth is positively correlated to the likelihood 

### Key empirical findings
- Higher real GDP growth is positively correlated to the likelihood of large drops in education spending when the latter is expressed as a share of GDP (Figure 3).
- Controlling for inflation, higher real GDP growth translates into higher nominal GDP growth that is (mechanically) negatively correlated to the ratio of spending over GDP.
- Higher real GDP growth is associated with a lower probability of a sharp decline in real social spending per capita.
- The probability of a sharp decline in social spending is higher for countries with higher inflation and therefore also higher growth in nominal GDP.
- The probability of a decline in spending (both as a share of GDP and in real per capita terms) is greater where the magnitude of short-term fiscal consolidation is high.
- Revenue-based consolidation attenuates the probability of social spending declines relative to expenditure-based consolidation; revenue increases have a negative impact on the probability of a decline (i.e., they reduce that probability).
- Declines in spending are much more likely where initial spending is high (Figure 4); addressing spending inefficiencies (e.g., high wages, high pharmaceutical costs) can reduce the likelihood of decreases in efficient spending components.
- Most previous empirical studies find social spending trends on average similar in program and non-program countries, though two studies find a negative impact for health.
- Using updated estimates for the period 2000–2016 and PRGT-eligible countries, education spending shows sign differences across Heckman and system-GMM estimates (significant positive in Heckman; insignificant negative in system-GMM). For health spending, both approaches yield negative—albeit statistically insignificant—estimates of program impact.
- IPWRA (Inverse Probability-Weighted Regression Adjustment) estimates for 101 lending arrangements (duration of at least two years, 2000–2016) show point estimates that are statistically insignificant across the three years considered after program approval (Appendix Tables 6 and 7).
- Weak evidence suggests the impact on health spending as a share of GDP may be greater in program countries where initial spending is relatively low (Columns 2 and 4; Table 7).
- For education spending (sample with 81 program cases), point estimates are negative in early years of programs and turn positive for longer-duration programs when initial spending is considered (Appendix Tables 8 and 9).
- On average, Fund-supported programs do not appear to have a statistically significant effect on social spending trends; it is hard to reject that spending trends have been similar in both program and non-program countries.
- Spending declined in over one-quarter of countries, with reductions being large in some cases.
- When spending is measured in real per capita terms, there is evidence that on average spending continues to increase over the life of the program.

### Methodology and data
- Time period analyzed: 2000–2016.
- Program definition: each lending arrangement is considered a separate program even when arrangements overlap in a given year; this increases the sample of program cases and captures heterogeneity within arrangements typically ranging between 2–5 years.
- Sample construction: rolling three-year periods where t-1 is the year prior to program approval; treated group includes programs with at least one completed program year following approval; control group includes countries without a program during the three-year window (Figure 5).
- Estimation approach introduced: IPWRA (doubly robust estimator) combining inverse-probability weighting (propensity score) with regression adjustment to estimate the average treatment effect on the treated (ATT).
  - Pre-treatment characteristics used in the participation equation include:
    - per capita GDP in purchasing power parity terms;
    - real per capita GDP growth;
    - cash balance and government debt in percent of GDP;
    - reserves in months of imports;
    - external debt in percent of GDP;
    - trade and capital account balances;
    - PRGT dummy (eligibility for concessional lending).
- First-stage (participation) probit model shows expected signs and statistical significance for most macroeconomic variables:
  - More developed and faster growing countries are less likely to engage in a program.
  - Countries with weaker fiscal balances, lower foreign reserves, weaker trade positions and higher external debt are more likely to request an IMF program.
- The null hypothesis that the first stage model balances the covariates cannot be rejected, supporting the specification.
- IPWRA relies on the unconfoundedness assumption conditional on observed pretreatment regressors.

### Policy implications and recommendations
- Measures that reinforce real GDP growth can help protect or even enhance real social spending.
- Program measures that strengthen revenue mobilization (e.g., stronger revenue administration and higher taxes) can help avoid short-term declines in social spending; revenue-based consolidation is less likely to reduce social spending than expenditure-based consolidation.
- Addressing spending inefficiencies (for example, lowering the cost of pharmaceuticals and containing wages) can reduce the likelihood of decreases in efficient spending components.
- Program design should consider initial spending levels; spending appears better protected in countries with relatively low initial spending levels.
- Constructing program indicators to reflect distinct lending arrangements (rather than aggregating overlapping arrangements) provides better insight into heterogeneity of outcomes and typical program-duration effects.

### Conclusions
- Consistent with previous studies, education and health spending has on average been protected in programs, but sizeable decreases occur in a large share of program countries.
- On average, education and health spending in program countries is similar to that in otherwise comparable non-program countries.
- New econometric analysis using a doubly robust IPWRA approach and a program definition that treats each lending arrangement separately confirms that, on average, spending trends are similar in program and non-program countries.
- Evidence indicates spending is better protected in countries with relatively low initial spending levels; yet over one-quarter of countries experienced declines, some of them large, highlighting opportunities to strengthen program design to prevent unwarranted reductions.

*IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS, INTERNATIONAL MONETARY FUND*

### Appendix I. Selected Tables

### Appendix I. Selected Tables

### IMF Lending Arrangements Approved, 2000–16 (Appendix Table 1)
- Total: 283
- GRA: 123
- EFF: 17
- FCL: 17
- PCL: 1
- PLL: 3
- SBA: 85
- PRGT: 160
- ECF: 51
- ECF-EFF: 4
- ESF: 11
- PRGT: 62
- PRGT-EFF: 3
- PSI: 18
- SBA-ESF: 1
- SBA-SCF: 5
- SCF: 5
- Sources: IMF Monitoring of Fund Arrangements database and IMF Financial database.
- Note: Data refer to the arrangements approved over the period 2000–16.

### Binary Variable for Large Declines in Health and Education Spending (Appendix Table 2)
- Dummy variable construction: country-years with declines in spending greater than the median negative changes observed.
- Dummy Variable equals: 0 / 1
- Change in Health Spending (% of GDP): (-0.2, 3.1) / [-3.0, -0.2]
- % Change in Health Spending (real per capita): (-10.1, 568.2) / [-64.2, -10.7]
- Change in Education Spending (% of GDP): (-0.4, 5.1) / [-3.9, -0.4]
- % Change in Education Spending (real per capita): (-7.1, 276.2) / [-43.7, -7.5]
- Sources: WHO; UNESCO; and IMF staff calculations.

### Probit Analysis of Large Reductions in Social Spending (Appendix Table 3) — Main Elements
- Regression framework:
  - Panel A: Probit regression coefficients for dummy of a large reduction in social spending over the period (t-h) to (t). Regressors include real GDP growth, inflation, initial social spending, and changes in revenue and fiscal balances (shares of GDP).
  - Panel B: Average marginal effects of increasing regressors from 25th to 75th percentile.
- Selected Panel A probit coefficients (Education spending (% of GDP) / Education spending (NCU per Capita) / Health spending (% of GDP) / Health spending (NCU per Capita)):
  - Real GDP Growth: 0.021*** 0.032*** -0.029* -0.010 0.006 -0.091*** 0.006 -0.005 0.013 -0.029*** -0.044*** -0.037*** (standard errors shown in parentheses)
  - Inflation: -0.007 -0.011* -0.002 -0.003 -0.011* 0.003 0.002 -0.002 0.007 0.008*** 0.005 0.015*** 
  - Fiscal Balance / GDP: 0.053*** 0.053*** 0.133*** 0.039** 0.060*** 0.092* 0.066*** 0.056*** 0.102*** 0.064*** 0.060*** 0.089***
  - Revenue / GDP: -0.062*** -0.065*** -0.035 -0.046** -0.067*** -0.066 -0.062*** -0.053*** -0.047 -0.067*** -0.059*** -0.068*
  - Initial Health/Education Spending / GDP: 0.249*** 0.264*** 0.129 0.050 0.050 -0.098 (selected columns)
  - Constant (examples): -1.995*** -2.216*** -1.250* -0.842*** -0.965*** 0.089 -1.896*** -1.699*** -2.986*** -1.138*** -0.935*** -2.021***
- Pseudo R-squared (selected columns): 0.120 0.172 0.116 0.036 0.055 0.190 0.186 0.200 0.265 0.141 0.175 0.244
- Number of observations (selected columns): 309 203 106 306 202 104 644 428 216 644 428 216
- Panel B: Average marginal effects (examples, corresponding order as above):
  - Real GDP Growth: 0.005*** 0.008*** -0.008* -0.003 0.001 -0.022*** 0.002 -0.001 0.003 -0.007*** -0.01*** -0.007***
  - Inflation: -0.002 -0.003* -0.001 -0.001 -0.003* 0.001 0.000 0.002 0.002*** 0.001 0.003***
  - Fiscal Balance / GDP: 0.014*** 0.013*** 0.034*** 0.011** 0.015*** 0.023* 0.017*** 0.014*** 0.025*** 0.015*** 0.014*** 0.016***
  - Revenue / GDP: -0.016*** -0.016*** -0.009 -0.012** -0.017*** -0.016 -0.016*** -0.013*** -0.011 -0.015*** -0.013*** -0.012*
  - Initial Health/Education Spending / GDP: 0.065*** 0.066*** 0.033 0.014 0.013 -0.024 0.094*** 0.113*** 0.119*** 0.026*** 0.052*** 0.035**
- Source: IMF staff calculations.
- Note: Standard errors shown in parentheses. Significance: ***, **, * indicate 99 percent, 95 percent, and 90 percent levels, respectively.

### Summary of Studies on the Impact of IMF Programs on Social Spending, 2010–2018 (Appendix Table 4) — Selected Papers, Results, and Limitations
- Clements, Gupta, and Nozaki (2013)
  - Period: 1985–2009; Countries: 59 LICs (eligible for concessional lending), 61 Non-LICs
  - Methodology: Dynamic model; Heckman two-step estimator; System-GMM with IVs (international reserves, exchange rate to USD, index of exchange rate regime).
  - Results:
    - System-GMM estimates suggest programs raise education and health spending in the first year by 0.22 and 0.27 percentage points of GDP, respectively, for LICs.
    - For LICs, education and health spending as a share of total government spending increases by about 1 percentage point and 0.5 percentage point, respectively, in the first year of the program.
    - Program effects on social spending were found to be insignificant for non-LICs.
  - Limitations:
    - Heckman requires joint normality and correct probit specification; requires exclusion restrictions that may not hold.
    - IVs in system-GMM may be invalid because external indicators can correlate with health and education spending.
- IMF (2017) Social Safeguards and Program Design in PRGT and PSI-Supported Programs
  - Period: 1988–2014; Countries: 59 LICs
  - Methodology: Replicates Heckman two-step of Clements, Gupta, and Nozaki.
  - Results:
    - Heckit analysis on 48 LICs over 1988–2014 shows that education spending increases by 0.32 percentage of GDP on average during IMF programs.
    - Evidence from 59 LICs over 1995–2014 suggests Fund programs do not have a significant impact on health spending.
  - Limitations: Subject to limitations identified in Clements, Gupta, and Nozaki (2013).
- Stubbs and Kentikelenis (2017)
  - Period: 1988–2014; Countries: 59 LICs
  - Methodology: Variant of Heckman two-stage; excludes PSI arrangements, includes GRA-funded programs.
  - Results:
    - Program participation associated with annual reductions in health spending of about 1.7 percentage points of GDP; no statistically significant effect on education spending.
  - Limitations:
    - Variables used for exclusion restrictions (total number of countries under IMF programs and UNGA voting similarity with the United States) may correlate with determinants of spending.
- Stubbs et al. (2017)
  - Period: 1995–2014; Countries: 16 West African countries (selected analysis)
  - Methodology: Heckman plus two-stage least squares instrumenting program participation and conditionality with UNGA voting similarity and total number of countries under IMF programs.
  - Results:
    - An additional binding condition reduces government health spending per capita by 0.248 percent; joint effect of program and number of conditionalities statistically insignificant.
  - Limitations: Subject to limitations identified in Stubbs and Kentikelenis (2017).
- Gupta, Schena, and Yousefi (2018)
  - Countries: 50 EMs, 42 LICs
  - Methodology: Variants of autoregressive distributed lag (ARDL) specifications.
  - Results:
    - Structural expenditure conditionality increases education and health spending by about 0.5-2 percent of GDP in the long run.
    - Public investment-related conditionalities reduced the budget share of health spending by between 1.5–2.8 percent.
    - Benefit of IMF conditionality found mainly in LICs.
  - Limitations:
    - ARDL approach is essentially OLS and does not address endogeneity concerns.

### Summary of Regression Results Replicating Past Methodologies (Appendix Table 5) — Selected Coefficients
- Note: Standard errors in parentheses. Significance: ***, **, * for 99 percent, 95 percent, 90 percent.
- Dependent variables: Education spending (% of GDP) and Health spending (% of GDP)
- Selected results (columns labeled with estimator type):
  - IMF program:
    - Column (1) Fixed effect (Education % of GDP): 0.26* (0.133)
    - Column (2) System GMM (Education % of GDP): 0.22** (0.101)
    - Column (3) Fixed effect (Health % of GDP): 0.17** (0.070)
    - Column (4) System GMM (Health % of GDP): 0.27*** (0.094)
    - Column (5) Fixed effect (Education % of GDP): 0.32** (0.134)
    - Column (6) Fixed effect (Health % of GDP): 0.03 (0.067)
    - Column (7) Fixed effect (Education % of GDP): 0.27*** (0.094)
    - Column (8) System GMM (Education % of GDP): -0.30 (0.231)
    - Column (9) Fixed effect (Health % of GDP): -0.03 (0.044)
    - Column (10) System GMM (Health % of GDP): 0.03 (0.149)
  - Lagged dependent variable (examples): 0.71***, 0.85***, 0.61***, 0.84***, 0.69***, 0.76***, 0.71***, 0.84***, 0.74***, 0.90*** (standard errors shown)
  - Government balance (examples): 0.02** (0.010); other columns around 0.00
  - Inverse Mills ratio (selected columns): -0.17** (0.080); -0.06* (0.036); -0.14* (0.074); -0.01 (0.043); -0.16*** (0.052); 0.02 (0.032)
- Number of observations and countries (selected):
  - Observations: 580, 580, 687, 687, 366, 809, 388, 388, 813, 813
  - Number of countries: 54, 54, 59, 59, 45, 46, 45, 65, 65
- Sources: Clements, Gupta and Nozaki (2013); 2017 Board paper on Social Safeguards; 2019 Board paper on Social Spending. (Table reproduces IMF staff calculations and replications.)

### Government Expenditure on Health — Percent of GDP (Appendix Table 6)
- Columns report year in program (2nd, 2nd, 3rd, 3rd, 4th, 4th) with treatment and outcome equations.
- Impact of IMF program on spending (outcome equation for countries with an IMF program):
  - 2nd: -0.009 (0.05)
  - 2nd: -0.003 (0.05)
  - 3rd: -0.023 (0.08)
  - 3rd: -0.012 (0.08)
  - 4th: -0.096 (0.14)
  - 4th: -0.085 (0.14)
- Public health spending (%GDP) coefficients (treatment/outcome interplay): -0.069** (0.03); -0.164*** (0.05); -0.199*** (0.07)
- Control variables (selected coefficients and significance):
  - Cash balance (%GDP): -0.052*** (0.02) across columns
  - Reserves in months of imports: -0.060*** (0.01) across columns
  - GDP per capita (PPP): -0.000*** (0.00) across columns
  - GDP per capita growth: -0.032** (0.01); -0.043*** (0.02); -0.071*** (0.02) depending on column
  - External debt (%GDP): 0.002*** (0.00) across columns
  - Trade balance (%GDP): -0.009*** (0.00); -0.007* (0.00); -0.005 (0.00)
  - Capital account balance (%GDP): 0.096*** (0.04) to 0.087** (0.04)
  - PRGT eligible country: 0.110 (0.16); 0.313* (0.18); 1.017*** (0.25) across columns
- Observations: 997, 997, 859, 859, 729, 729
- Program cases (number of treated): 136, 136, 101, 101, 60, 60
- Non-program (number of controls): 861, 861, 758, 758, 669, 669
- P-value for covariate balance: 0.291, 0.291, 0.115, 0.115, 0.514, 0.514
- Robust standard errors in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1.

### Government Expenditure on Health — Constant NCU per capita (Appendix Table 7)
- Impact of IMF program on spending (outcome equation for countries with an IMF program):
  - 2nd: -0.213 (2.57)
  - 2nd: 1.094 (2.44)
  - 3rd: -0.059 (4.80)
  - 3rd: 2.325 (4.23)
  - 4th: -5.431 (8.68)
  - 4th: -2.046 (7.76)
- Public health spending (%GDP): -4.279*** (1.22); -8.902*** (2.05); -9.866*** (2.92)
- Control variables match Appendix Table 6 (coefficients and significance identical for listed controls).
- Observations: 997, 997, 859, 859, 729, 729
- Program cases (number of treated): 136, 136, 101, 101, 60, 60
- Non-program (number of controls): 861, 861, 758, 758, 669, 669
- P-value for covariate balance: 0.291, 0.291, 0.115, 0.115, 0.514, 0.514

### Government Expenditure on Education — Percent of GDP (Appendix Table 8)
- Impact of IMF program on spending (outcome equation for countries with an IMF program):
  - 2nd: -0.125 (0.09)
  - 2nd: -0.132 (0.08)
  - 3rd: -0.086 (0.14)
  - 3rd: -0.104 (0.14)
  - 4th: -0.016 (0.24)
  - 4th: 0.108 (0.22)
- Public education spending (%GDP): -0.114** (0.05); -0.258** (0.10); -0.159** (0.08)
- Control variables (selected coefficients and significance):
  - Cash balance (%GDP): -0.078*** (0.03) in early columns
  - Reserves in months of imports: -0.044* (0.02); -0.035 (0.03); -0.107** (0.05)
  - GDP per capita (PPP): -0.000*** (0.00)
  - GDP per capita growth: -0.054** (0.02) in some columns
  - External debt (%GDP): 0.003*** (0.00)
  - Trade balance (%GDP): -0.020*** (0.00); -0.016*** (0.01)
  - PRGT eligible country: 0.215 (0.22); 0.513* (0.27); 1.546*** (0.40)
- Observations: 542, 542, 442, 442, 367, 367
- Program cases (number of treated): 81, 81, 47, 47, 29, 29
- Non-program (number of controls): 461, 461, 395, 395, 338, 338
- P-value for covariate balance: 0.400, 0.400, 0.162, 0.162, 0.189, 0.189

### Government Expenditure on Education — Constant NCU per capita (Appendix Table 9)
- Impact of IMF program on spending (outcome equation for countries with an IMF program):
  - 2nd: -3.616 (3.99)
  - 2nd: -4.178 (3.91)
  - 3rd: -0.382 (5.62)
  - 3rd: -1.395 (5.59)
  - 4th: -3.268 (11.83)
  - 4th: 6.056 (10.19)
- Public education spending (%GDP): -3.624*** (1.36); -6.797*** (1.84); -7.884*** (2.15)
- Control variables (selected coefficients and significance mirror Appendix Table 8):
  - Cash balance (%GDP): -0.078*** (0.03)
  - Reserves in months of imports: -0.037 (0.02); -0.032 (0.03); -0.106** (0.05)
  - GDP per capita (PPP): -0.000*** (0.00)
  - GDP per capita growth: -0.053** (0.02) in some columns
  - External debt (%GDP): 0.003*** (0.00)
  - Trade balance (%GDP): -0.020*** (0.00); -0.016*** (0.01)
  - PRGT eligible country: 0.245 (0.22); 0.498* (0.27); 1.540*** (0.40)
- Observations: 537, 537, 439, 439, 365, 365
- Program cases (number of treated): 79, 79, 47, 47, 29, 29
- Non-program (number of controls): 458, 458, 392, 392, 336, 336
- P-value for covariate balance: 0.366, 0.366, 0.151, 0.151, 0.185, 0.185

### Mission Chiefs’ Survey — Overview of Results (Background Paper III excerpt)
- IMF Area Department mission chiefs surveyed between July and December 2018.
- Survey indicated:
  - social spending is widely considered to be macro-critical;
  - when engaging on social spending IMF country teams rely on their own resources and also extensively leverage internal and external expertise;
  - program objectives are typically in line with countries’ social spending priorities;
  - IMF policy advice to member countries commonly involves social spending reforms.

*Appendix I. Selected Tables — IMF staff compilation from the source PDF.*

### 1.      A survey of IMF mission chiefs (MCs) provides insights into staff views on the Fund’s

### 1.      A survey of IMF mission chiefs (MCs) provides insights into staff views on the Fund’s engagement on social spending

### Survey design and response rates
- Survey scope: mission chiefs leading country work for the 189 IMF member countries, plus Aruba, Curação and St. Maarten, Hong Kong, Macao SAR, and West Bank and Gaza.
- Fieldwork period: July through December 2018.
- Purpose: capture MC views on IMF work on social spending, including analysis, policy advice, interaction with other stakeholders, and specific questions on program context, design, and conditionality.
- Prepared by: Maura Francese and Nghia Piotr Le (FAD).
- Overall response rate: (about 80 percent; Figure 1).
- Response rates by income group: 73 percent for emerging economies (EMEs); 81 percent for low-income and developing countries (LIDCs); 89 percent for advanced countries (AEs).
- Fragile states sub-sample response rate: 83 percent.
- Regions and country grouping follow the WEO country classification (AE; CIS; EDA; EDE; LAC; MENAP; SSA).

### B. Macro-criticality of social spending issues — prevalence and drivers
- Prevalence:
  - Almost 80 percent of MCs see social spending as macro-critical.
  - LIDCs: almost 90 percent.
  - EMEs: 80 percent.
  - AEs: almost 60 percent.
  - Fragile states: 85 percent.
  - By region: about 90 percent in emerging and developing Europe, MENAP and Sub-Saharan Africa.
  - Social spending is less often considered macro-critical in CIS countries.
- Most-cited drivers for deeming social spending macro-critical (multiple responses allowed):
  - Social And/Or Political Stability.
  - Distributional Objectives.
  - Large Human Capital Gaps – Education.
  - Large Human Capital Gaps – Health.
  - Large Social Protection Gaps.
  - Social Spending is Inefficient.
  - Future Social Spending Pressures.
  - Current Social Spending Pressures.
  - High Population Growth.
  - Population Ageing.
  - Country is Implementing Major Reforms.
  - Conflict And/Or Refugees.
  - To Adapt to Technological Change.
  - Risk of Crowding Out Other Important Public Spending.
  - Significant Adverse Impacts on Incentives.
  - Natural Disasters.
  - Other.
- Heterogeneity by country group:
  - AEs: spending pressures, population ageing, and distributional objectives most commonly cited.
  - LIDCs: development gaps in education, health, and social protection, and social/political stability most commonly cited.
  - Countries under IMF-supported programs: need to close development gaps and risks to social/political stability are most often cited, particularly in fragile states.

### C. Resources for addressing social spending issues and interaction with other institutions
- Main sources of analysis and expertise:
  - Own resources and analysis: 80 percent of MCs indicated this is the main basis.
  - Analysis, tools, and technical assistance provided by other IMF departments: about 50 percent of MCs.
  - Fiscal Affairs Department (FAD) is the main IMF provider of analysis and expertise; tools and TA from FAD are widely used.
    - Examples of FAD tool usage (reported in footnote): almost 30 percent use FAD’s tool for assessing spending; almost 20 percent use the Department’s long-term pension and health projections; about 16 percent use the inequality database; a similar share uses tools and templates for energy subsidy reform.
    - FAD TA used by 15 percent of surveyed teams when engaging on social spending issues.
- External contributors:
  - International Development Institutions (IDIs): almost 80 percent of cases.
  - Country authorities: 31 percent.
  - Academics: 20 percent.
- World Bank role:
  - Low use for AEs: Only 6 percent of AE MCs selected World Bank as an important source of information.
  - Major partner for EMEs and LIDCs: World Bank was indicated as an important source of analysis and resources by 60 percent of EMEs MCs and 53 percent of LIDCs MCs; just above 60 percent for fragile states.
  - For AEs, shares flagging other international institutions and authorities as important information sources are 53 and 47 percent respectively.
- Interaction modalities with development partners:
  - Most often bilateral discussions between IMF country teams and sectoral experts (either at headquarters or during missions).
  - Collaboration on analytical projects less frequent: 18 percent of MCs indicated collaboration on analytical projects as a modality.
  - World Bank accounts for almost half of total interactions and is the most frequent counterpart across all topics.
  - Interaction with OECD and ILO is rare: 1.3 percent and 1.2 percent of interactions respectively.
- Mapping of interactions by topic and counterpart (percent of total interactions):
  - Social Assistance: World Bank 11.0; OECD 0.8; Other regional development banks 2.8; ILO 0.5; Un Agencies 3.1; Local Development Partners 1.8; NGO 2.5; Academics 2.8; Other 1.0; Total 26.4.
  - Social Insurance: World Bank 5.9; OECD 0.2; Other regional development banks 1.0; ILO 0.3; Un Agencies 0.8; Local Development Partners 1.5; NGO 1.5; Academics 2.0; Other 0.5; Total 13.7.
  - Health: World Bank 9.7; OECD 0.2; Other regional development banks 2.3; ILO 0.2; Un Agencies 2.3; Local Development Partners 2.5; NGO 1.3; Academics 1.3; Other 0.5; Total 20.3.
  - Education: World Bank 10.2; OECD 0.3; Other regional development banks 2.6; ILO 0.0; Un Agencies 3.0; Local Development Partners 2.8; NGO 2.0; Academics 1.8; Other 0.3; Total 23.1.
  - Other: World Bank 8.9; OECD 0.0; Other regional development banks 2.3; ILO 0.2; Un Agencies 1.2; Local Development Partners 1.6; NGO 1.0; Academics 1.3; Other 0.2; Total 16.6.
  - Grand total: World Bank 45.8; OECD 1.5; Other regional development banks 11.0; ILO 1.2; Un Agencies 10.4; Local Development Partners 10.2; NGO 8.2; Academics 9.2; Other 2.5; Total 100.0.
- Factors limiting engagement on social spending (percent reporting; key constraints):
  - Competing priorities for country analysis.
  - Poor data quality/availability.
  - Lack of expertise within the team.
  - No need for the Fund’s engagement in this area.
  - The authorities are sensitive to the topic.
  - The authorities are not interested.
  - Country capacity constraints.
  - Other.
  - Not within the tasks mandated to the team.
  - Inability to draw on outside expertise.
- Obstacles to external cooperation:
  - Lack of information on other institutions’ organizational set up, work plans, and country engagement.
  - Differences in institutional focus (e.g., improving social outcomes versus fiscal sustainability).
- Steps identified to enhance cooperation:
  - Establishing cooperation processes and data sharing.
  - More extensive use of informal channels.

### D. Policy advice
- Prevalence of IMF recommendations on social spending reforms:
  - Surveillance countries: 67 percent.
  - Program countries: 69 percent.
  - Variation by income group/region:
    - AEs and EMEs: recommended in three-quarters of the cases.
    - LIDCs and fragile states: recommended in 1 of 2 cases (about 50 percent), with less frequent recommendations in Sub-Saharan Africa.
- Nature of policy advice on program design:
  - Introduction or expansion of means-tested schemes recommended in the majority of cases: 64 percent.
  - Downsizing of schemes that do not require a means test suggested in 18 percent of the cases.
  - Expansion of social programs not based on a means test recommended in about 18 percent of cases.
  - In program contexts: introduction or expansion of a means-tested program is recommended in 4 of 5 cases.
  - In fragile states: introduction or expansion of means-tested programs recommended in one half of cases.
- Controversy around IMF advice:
  - Policy advice on social spending reported controversial with authorities or other stakeholders in 15 percent of cases overall.
  - Higher reported controversy in EDE, LAC and CIS: 42 percent, 31 percent, and 30 percent of cases respectively.

### E. Programs: objectives and conditionality
- Program objectives and social priorities:
  - Nearly all MCs indicated programs entailed either fiscal consolidation or a neutral fiscal stance; only a few cases of fiscal expansion.
  - 70 percent of MCs view program objectives as consistent with country authorities’ social priorities.
  - Additional 19 percent were neutral on the consistency question.
- Social spending outcomes in programs:
  - Key social spending items: maintained in 54 percent of cases.
  - Key social spending items: increased in 37 percent of cases.
  - Program objective is typically to maintain or increase spending as a share of GDP or in nominal terms.
- Conditionality on social spending:
  - Almost all IMF programs include conditionality aimed at protecting and strengthening social spending.
  - Effectiveness as viewed by MCs:
    - 63 percent indicated program conditionality is an effective way to protect spending.
    - 22 percent were neutral.
  - Mechanisms used (percent of MCs indicating usage):
    - Indicative targets (floors): 70 percent.
    - Commitments in Memoranda of Economic and Financial Policies (MEFPs): about 40 percent.
    - Structural benchmarks (SBs): 24 percent.
  - Conditions have been applied across the spectrum of spending categories (education, health, social assistance, unemployment benefits, pensions, etc.) with no clear regional or income-group pattern.
  - Compliance and shortfalls:
    - MCs reported that conditions are met most of the time.
    - 75 percent of MCs acknowledged that during the life of the program at least one conditionality (either quantitative or structural) was missed at some point.
    - Common reasons for targets being missed: shortfalls in external donor financing flows, government revenue shortfalls, or lack of ownership.
- Design and implementation improvements suggested:
  - Narrow the definition of spending floors (e.g., target the most critical programs selectively).
  - Broaden consultation with IDIs.
  - Improve data quality.
  - Strengthen implementation capacity in program countries.
  - Early engagement with IDIs to ensure conditionality reflects country-specific factors.

### Annex I: Questionnaire (high-level)
- Questionnaire sent to all country mission chiefs.
- Questions 1–14 for all countries; questions 15–24 only for program countries.
- Purpose: input to the Board paper on IMF Engagement on Social Spending: A Strategic Framework, central to management response to the IEO Report on The IMF and Social Protection.
- Definition for the survey/Board paper: social spending is spending on 1) basic health, 2) basic education, and 3) social protection (which consists of social insurance and social assistance).
  - Social insurance policies: e.g., unemployment insurance and pensions; typically financed by contributions and payroll taxes.
  - Social assistance policies: e.g., universal and targeted transfers; typically financed by general government revenues.

*Prepared by Maura Francese and Nghia Piotr Le (FAD), Mission Chiefs’ Survey on IMF Engagement on Social Spending.*

### 1. Your department: ______________________________________________________

### 1. Your department: ______________________________________________________

### Administrative and assignment information
- Mission chief for [country]: __________________________________________________
- Start date of assignment _________________________________
- Type of Fund engagement during your assignment:
  - a) Surveillance
  - b) Program/near program. Please specify program type, Start Year and End Year (if applicable)

### Assessment of macro-criticality of social spending (Q5–Q6)
- Q5: Do you assess social spending issues to be macro-critical for your country?
  - a) Yes
  - b) No
- Q6: Reasons social spending may be macro-critical (check all that apply)
  - a) Current social spending pressures are putting fiscal sustainability at risk
  - b) Future social spending pressures are expected to put fiscal sustainability at risk
  - c) Social spending is crowding out other important public spending thus creating risks for internal or external stability or growth.
  - d) Lack of adequate social spending is a risk to social and/or political stability.
  - e) The country is facing significant challenges due to population ageing.
  - f) Significant/increasing social spending needs due to high population growth.
  - g) Significant/increasing social spending needs due to conflict and/or refugees.
  - h) Significant/increasing social spending needs due to natural disasters.
  - i) Social spending is inefficient (too much spending, but very little social outcomes).
  - j) Social spending has significant adverse impacts on incentives e.g. affecting labor market participation.
  - k) Social spending is key for achieving the authorities’ distributional objectives.
  - l) The country has large human capital gaps (may include SDG commitments)–   scaled up education spending is needed.
  - m) The country has large human capital gaps (may include SDG commitments)–   scaled up health spending is needed.
  - n) The country has large social protection gaps (may include SDG commitments)–   scaled up social protection is needed.
  - o) Scaled up or significantly reformed social protection is needed to protect against the shocks of technological change (e.g., gig economy).
  - p) Scaled up or significantly reformed education spending is needed to adapt to technological change.
  - q) The country is implementing major reforms of social spending (e.g. move towards universal basic income, pension reform, healthcare reform, etc.) that may have macroeconomic consequences.
  - r) Other (please specify)

### Team recommendations on social spending reforms (Q7–Q9)
- Q7: Has the team recommended reforms in social spending (including social protection, health, education)?
  - a) Yes
  - b) No
- Q8: If yes, types of recommendations (check all that apply)
  - a) Introduction/expansion of targeted schemes that require some type of means test
  - b) Reduction of targeted schemes that require some type of means test
  - c) Introduction/expansion of schemes that do not require some type of means test (including universal schemes)
  - d) Reduction of schemes that do not require some type of means test (including universal schemes)
- Q9: For any option (a)-(d) checked: Has this recommendation been controversial with authorities or other stakeholders?
  - a) Yes
  - b) No

### Sources of analysis and external cooperation (Q10–Q14)
- Q10: Whose analysis/resources relied on as main source (check all that apply)
  - a) Team’s own resources
  - b) Analysis conducted by another department. (please specify which below)
  - c) Technical Assistance. (please specify the department(s) below)
  - d) Support and tools provided by another department. (please specify the department(s) below)
  - e) Analysis conducted by the World Bank. (please specify below)
  - f) Analysis conducted by international institutions (e.g. AfDB, ADB, IADB, ILO, OECD, Unicef, EU). (please specify which institution(s) below)
  - g) Authorities’ analysis
  - h) Academics’ analysis
  - i) Other (please specify)
- Q11: Explicit discussions with listed institutions on social spending (matrix to indicate whether discussions “Worked well” or “Did not work well” across areas: Social assistance, Social insurance, Health, Education, Other such as energy subsidies and/or food subsidies)
  - Institutions listed include: World Bank; OECD; AfDB; IADB; ADB; EBRD; IsDB; NDB; AIIB; other regional development banks; ILO; UN agencies (e.g.,UNDP, UNICEF); Local development partners (e.g., EU, DfID); NGOs or CSOs; Academics; Other (please specify)
- Q12: How do you interact with other institutions/organization and leverage external expertise?
  - a) Bilateral discussions with experts (e.g. WB managers and economists) at HQ
  - b) Bilateral discussions with experts (e.g. WB managers and economists) during missions
  - c) Bilateral discussions with experts (e.g. WB managers and economists) by ResRep
  - d) Collaborating in analytical projects
  - e) Other (please specify)
- Q13: Obstacles faced when cooperating with other institutions (check all that apply)
  - a) Lack of country-level involvement (country presence) by the other institutions/organizations
  - b) Lack of information on who does what (information on work plans and engagement by other institutions)
  - c) Lack of interest from the other institutions to cooperate with the Fund
  - d) Conceptual differences in understanding/approaches to social spending issues
  - e) Differences in institutional focus (e.g. improving social outcomes versus fiscal sustainability)
  - f) Other (please specify)
- Q14: What would improve cooperation with other external institutions? (check all that apply)
  - a) Established discussion channels/cooperation processes
  - b) More extensively using informal contact/discussion channels
  - c) Data sharing
  - d) Other (please specify)
  - e) Current set up works quite well/well enough

### Tools used and their usefulness (Q15–Q16)
- Q15: Tools the team has used (check all that apply)
  - a) Expenditure Assessment Tool (EAT)
  - b) FAD’s long-term pension and health expenditure projections
  - c) FAD’s Pension Reform Template
  - d) FAD’s Income Inequality (Gini) Database
  - e) FAD’s Energy Subsidies tools and templates
  - f) Macroeconomic and Distributional Implications of Fiscal Policies model developed by SPR
  - g) CEQ (Commitment to Equity project) incidence analysis methodology (in cooperation with CEQ)
  - h) WB ASPIRE database
  - i) WB PovcalNet analysis
  - j) WB Poverty and Social Impact Analysis (PSIA)
  - k) Interagency Social Protection Assessments (ISPA) tools
  - l) The SDG Indicators Global Database
  - m) The ILO’s social protection platform
  - n) Other (please specify)
- Q16: Have you found the tool(s) selected above have been helpful?
  - a) Yes
  - b) No

### Challenges and capacity constraints (Q17–Q19)
- Q17: Challenges that prevented fuller coverage of social spending issues (check all that apply)
  - a) Covering social spending is not within the tasks mandated to the team
  - b) Lack of expertise within the team
  - c) Inability to draw on outside expertise
  - d) Cannot do adequate analysis because of data quality/availability
  - e) Competing priorities
  - f) The authorities are not interested
  - g) The authorities are sensitive to the topic
  - h) Achieving progress is unlikely anyway because of capacity constraints in the country’s public administration
  - i) Another institution (please specify) is taking the lead on social spending issues and there is no need for the Fund’s engagement in this area
  - j) Other (please specify)
- Q18: Areas where more tools would be helpful (check all that apply)
  - a) Health
  - b) Education
  - c) Pensions
  - d) Unemployment insurance
  - e) Social assistance transfers to households (such as unconditional and conditional cash transfers)
  - f) Other (please specify)
- Q19: For each option a–f checked in Q18, list any tools you would find useful to conduct analysis (e.g., expenditure benchmarking tool)

### Questions for program countries (Q20–Q30)
- Q20: Were program objectives consistent with authorities’ social priorities?
  - a) Strongly agree
  - b) Agree
  - c) Neutral
  - d) Disagree
  - e) Strongly disagree
  - f) Not applicable
- Q21: Does the program entail fiscal consolidation?
  - a) The program entails fiscal consolidation.
  - b) The program entails fiscal expansion.
  - c) The program is fiscally neutral.
- Q22: Does the program seek to protect or expand social spending?
  - a) Yes, key social spending is maintained. Then choose from (in nominal terms/ in real per capita terms/as a share of GDP/as a share of total public spending).
  - b) Yes, key social spending is increased. Then choose from (in nominal terms/ in real per capita terms/as a share of GDP/as a share of total public spending).
  - c) No. If no, please specify reasons______________
- Q23: Does the program have quantitative/structural conditionality on social spending? (check all that apply)
  - a) No
  - b) Yes, there are indicative targets (ITs) on a social spending floor
  - c) Yes, there are quantitative performance criteria (floors) (PCs) on a social spending floor
  - d) Yes, there is quantitative conditionality on specific areas, (please specify)
  - e) Yes, there are measures as commitments in the Memorandum of Economic and Financial Policies (MEFP)
  - f) Yes, there are measures as structural benchmarks (SBs)
- Q24: If the program includes conditionality, what is included in ‘key social spending’?
  - a) Pension benefits
  - b) Unemployment benefits
  - c) Disability benefits
  - d) Social assistance benefits
  - e) Health benefit
  - f) Education spending
  - g) Other (please specify)
- Q25: How effective has conditionality on social spending been?
  - a) The measures are met most of the time.
  - b) The measures are not met most of the time.
- Q26: If social spending conditionality was missed, main reasons (select all that apply)
  - a) Shortfall in external donor financing flows
  - b) Shortfall in government revenue (unrelated to external financing)
  - c) Lack of country ownership
  - d) Social spending conditionality defined too broadly
  - e) Social spending target became irrelevant
  - f) Other. Please specify: _________________________________________________. 
  - g) Not applicable
- Q27: Program conditionality on social spending is an effective way to protect such spending during an IMF program in your country
  - a) Strongly agree
  - b) Agree
  - c) Neutral
  - d) Disagree
  - e) Strongly disagree
  - f) Don’t know
  - g) Not applicable
- Q28: Could the design/implementation of the social spending target be improved to increase compliance rate?
  - a) Yes, the design could be improved.
  - b) Yes, the implementation could be improved.
  - c) No
- Q29: If a) to Q28, how could design be improved? (select all that apply)
  - a) More targeted specification of spending floors (e.g. targeting few or most critical sectors and line ministries)
  - b) Revisiting and revising these targets more frequently
  - c) Seeking expertise from the World Bank and other development partners
  - d) Adopting contingency plans to preserve spending from fiscal shocks
  - e) Including adjustors in the design of target to account for external shocks (e.g. shortfall in external assistance)
  - f) Actively seeking and incorporate authorities’ inputs in designing of the targets
  - g) Improving the quality of fiscal data
  - h) Making the target a binding conditionality (e.g. performance criteria)
  - i) Other (please specify)
- Q30: If b) to Q28, how could implementation be improved? (select all that apply)
  - a) More detailed specification of spending floors by type of spending.
  - b) Seeking feedback from the World Bank and other development partners on implementation challenges in the country before conditionality is defined
  - c) Seeking expertise from the World Bank and other development who have more expertise on implementation issues during the program
  - d) Strengthening capacity building focused on improving administrative capacity for social spending to accompany the delivery of the reform process
  - e) Other (please specify)

### Background Paper IV — The debate on universal and targeted transfers (A–C)
- Paper purpose:
  - Sets out issues to consider when providing policy advice on targeting of transfers and trade-offs in different targeting approaches.
  - Emphasizes that broader population coverage may be desirable due to administrative constraints or social and political preferences, and that large population coverage and fiscal cost needs to be accompanied by progressive and efficient taxation.
  - Notes importance of considering both tax and transfer sides to ensure taxes do not significantly offset redistributive impact of transfers and that lower-income groups’ share in transfers must be sufficiently higher than their share in taxes.
- A. Introduction
  - 1. There is a growing debate on the relative merits of universal and targeted social assistance transfers, especially in low-income contexts.
  - 2. Universal vs targeted: universal benefit available to everyone without eligibility conditions (e.g., UBI); targeted benefit uses eligibility criteria based on income or characteristics correlated with poverty.
- B. Means-tested Transfers
  - 3. Theoretical case: perfect means-tested targeting maximizes social welfare under a budget constraint by directing transfers to poor households and avoiding transfers to non-poor.
  - 4. Practical limits: many countries lack capacity for perfect means tests due to low administrative capacity, large informal sectors, multiple and volatile income sources, in-kind income, poor bookkeeping, and social/political reluctance to means testing.
- C. Categorical Targeting
  - 5. Categorical targeting bases eligibility on characteristics (e.g., presence of children, elderly, location, disability) that are highly—but imperfectly—correlated with poverty; can also differentiate transfer levels across categories.
  - Figure notes (from empirical example based on India’s 2011–12 National Sample Survey):
    - In the survey, 33 percent of households have children aged 0–5 years, 50 percent have children aged 0–10 years, and 62 percent have either children aged 0–10 years or elderly 65 years or above.
    - Over 35(60) percent of children aged up to 10 years are in the bottom 2(4) income deciles, with very little variation across age levels.
    - Compared to child and elderly transfers, the share of benefits accruing to the bottom five deciles is always higher under PMT targeting at the 50th percentile.

*Prepared by David Coady and Nghia Piotr Le (FAD); source content: IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS (excerpt).*

### 6.      The imperfect nature of categorical targeting gives rise to a trade-off between poverty

### 6.      The imperfect nature of categorical targeting gives rise to a trade-off between poverty

### Categorical targeting: trade-offs and coverage
- Restricting transfers to households with children (versus a UBI) can channel a larger share of the poverty budget to the poor and thus have a larger poverty impact, but:
  - Poor households without children are excluded.
  - Non-poor households with children are included.
- Coverage can be increased by expanding eligibility (for example, to older children or the elderly).
- Simulation findings (household survey data):
  - Uniform benefits for children up to 5 years are very progressive: coverage of the bottom quintile is around 50 percent, falling to around 15 percent for the top quintile.
  - Expanding eligibility to children up to 10 years or to include the elderly increases overall household coverage, including coverage of lower-income groups.
  - Universal benefits ensure universal coverage but, under a fixed budget, require lower transfer levels per household across all income groups.
- Policy implication:
  - The choice between universal and categorical transfers involves a trade-off among poverty impact, coverage of the poor, and the size of the transfer budget (and therefore required tax levels).

### Proxy-means testing (PMT): performance and challenges
- PMT attaches a continuous score to households based on household characteristics correlated with welfare (often from regression coefficients of income or consumption).
- Leakage and undercoverage occur under PMT; it is argued to be prone to significant leakage and undercoverage of the poorest.
- Simulation findings (same survey data):
  - Under PMT schemes, coverage is substantially higher for lower-income groups than higher-income groups.
  - As program coverage expands from 10 percent to 100 percent of the population, coverage of lower-income groups increases markedly.
  - Coverage of the bottom quintile reaches around 80–90 percent at 40 percent population coverage.
  - To ensure almost universal coverage (say above 80 percent) of each of the bottom three deciles, the program would need to expand to 50 percent of the population.
- Advantages of PMT:
  - For a given budget, PMT typically yields a larger poverty reduction impact and better coverage of lower-income groups than categorical child or elderly targeting.
  - Differentiating transfers by household size and composition (e.g., using PMT to target child transfers) can increase poverty impact.
- Challenges and risks:
  - Random exclusion and exclusion around the eligibility cut-off score generate lack of transparency and community discontent (horizontal inequity).
  - PMT scoring systems require regular updating due to the structural/statistical nature of the approach.

### PMT as a basis for differentiated universal benefits (tiered PMT)
- Alternative design: use PMT only to differentiate benefit levels while providing universal coverage.
  - This eliminates undercoverage of poor beneficiaries (though benefit amounts still vary).
- Simulation outcomes (benefit shares and coverage):
  - Under a UBI, the poorest 30 percent receive 30 percent of the fixed transfer budget (by design).
  - Under a “tiered PMT” with benefit ratios of 4:2:1 across the lowest three PMT deciles, the next four deciles, and the highest three deciles:
    - The bottom three deciles receive over 40 percent of benefits.
    - The richest three deciles receive just above 15 percent of benefits.
  - A PMT that targets 50 percent of the population has a slightly higher share of benefits accruing to the bottom three deciles than the tiered PMT, but it comes with significant undercoverage of lower welfare deciles.
- Equity considerations:
  - Tiering increases the share of benefits accruing to the poorest deciles and eliminates eligibility undercoverage.
  - Tiering eases, but does not completely eliminate, horizontal equity concerns.

### Financing coverage expansion: tax instruments and policy design
- Coverage expansion needs to be financed through progressive and efficient taxation. Recommended components:
  - Strengthening personal income taxes (PITs):
    - Where administrative capacity is low, initially focus on broadening coverage of taxes on wages and salaries and on taxation of higher incomes.
    - Strong PITs allow clawback of universal transfers from higher-income groups and reduce reliance on less progressive tax instruments.
    - Reinforce PITs with effective taxation of corporate income.
  - Strengthening consumption taxes:
    - Broad-based consumption taxes play a key role in increasing tax capacity in developing economies.
    - Efficiency requires minimizing differentiation of consumption tax rates across goods.
    - Expanding social safety net coverage dilutes the case for preferential consumption tax rates on distributional grounds.
    - Setting the tax registration threshold at a reasonably high level can enhance progressivity because smaller businesses typically have lower incomes and lower-income groups often buy from small-scale retailers.
    - In high-inequality settings, significant redistribution can be achieved through simple tax-and-transfer systems; for instance, a UBI financed by higher consumption taxes can be a feasible and efficient redistributive approach.
  - Expanding use of efficient excises:
    - Taxing consumption that generates negative externalities can raise significant revenues in an efficient and equitable manner.
    - Increasing taxes on fossil fuel energy can reduce pollution and health damage while ensuring a progressive distribution of the tax burden.
    - Other candidates for excise taxes on efficiency grounds include alcohol, tobacco, and possibly sugar.
- Important caveats:
  - Broadening the consumption tax base can increase the tax burden on vulnerable groups; safety nets must be capable of protecting these groups via coverage and increased transfers.
  - These tax policies typically require significant investment in strengthening revenue administration systems to fight tax evasion and avoidance, both domestic and cross-border.

*Source: IMF staff paper section on targeting, proxy-means testing, tiered PMT, and financing transfers.*

### 4.      The intensity of the discussion of social spending in documents is measured by the

### 4.      The intensity of the discussion of social spending in documents is measured by the term frequency

### Methodology: term frequency definition and implementation
- Term frequency is defined as the number of occurrences in a document of specific terms, normalized by the number of words in that document to account for possible variations of document length over time (and scaled by a factor 10,000 for presentational purposes).
- To allow comparisons between countries with a different number of staff reports per year, the term frequency is averaged across all staff reports in a given year for each country.
- The list of social spending-related terms was pre-defined by IMF subject matter experts who examined a representative number of staff reports.
- The term frequency approach was preferred over topic modeling because:
  - Topic modeling is unsupervised with no benchmarks to measure against and no direct method to adjust parameters.
  - The number of possible topics in topic modeling is limited, whereas staff reports cover a very broad range of topics over time.
- Sensitivity analysis findings:
  - The trend of social issues discussed in IMF programs is broadly uncorrelated to the length of IMF documents.
  - Staff report lengths changed: dropped from 2,015 words on average in 1982 to 974 words in 1984, and from 1,950 words on average in 2005 to 1,224 words in 2007.
  - The “normalized average frequency count per document” has a flatter shape and a less pronounced dip between 2007 and 2009 than the unnormalized indicator (“average frequency count per document”); overall both indicators have broadly similar trends.
- To ensure comparability between Article IV and program staff reports (different counts per year), the frequency count is the average of the normalized term frequency for all the staff reports for a country in a year.
- To capture both general and specific discussions, both general and specific terms were included. Overall social spending frequency is the sum of the frequencies of general and specific social spending terms.
- The list of terms uses only root words (for example, a search for “pension” captures pension, pensions, pensioner).

### Box 1 — Social spending related terms used in the text mining analysis
- Overall social spending concepts is the sum of the general social spending concepts and specific social spending concepts.
- General social spending concepts:
  - social spending, social expenditure, social policy, social protection, social program, safety net, social transfer, social assistance, social benefit, social package.
- Specific social spending concepts:
  - Pension: pension, retirement, retiree, old-age benefit
  - Health: health, health expenditure, health spending
  - Education: education, education expenditure, education spending
  - Income support: income support, guaranteed minimum income, meal program, food stamp, ration card, voucher
  - Energy subsidy: energy subsidy
  - Other subsidies: food subsidy, agriculture subsidy, consumer subsidy, price subsidy, fertilizer subsidy
  - Other benefits: disability benefit, maternity benefit, child benefit, child allowance, unemployment benefit
- Distribution analysis concepts:
  - inequality, distribution, redistribution, distributional, poverty, vulnerable, Gini, income decile, income quantile, regressive, progressive

### Main results — trends and intensity of discussion
- The discussion of social spending issues increased over time:
  - Frequency of discussions rose steadily for decades and peaked in 1999 when the Poverty Reduction Growth Facility (PRGF) was introduced.
  - Streamlining following the overhaul of the IMF’s lending and conditionality frameworks during 2009–2011, in tandem with the global financial crisis, appears to have temporarily crowded out attention on social spending issues.
  - Subsequent recovery of social spending issues in all staff reports coincides with renewed IMF research on social spending issues and its link with inclusive growth.
- Current intensity:
  - On average, Fund documents had about 20 (normalized) occurrences of social spending related terms in 2018, although with variation across countries.
  - Equivalent to about 40 unnormalized occurrences per report.
- Surveillance versus program reports:
  - The discussion in surveillance and program staff reports followed a broadly similar pattern.
  - Particularly after 2000, the normalized frequency count for surveillance staff reports is somewhat higher than for program staff reports.

### Cross-country and within-document variation; outliers
- Wide variation in intensity across countries and documents; some countries are repeated outliers with large references to social spending.
- Surveillance documents:
  - Number of staff reports without any discussion of social spending issues fell from an average of 33 documents in 1979 to zero after 2012.
  - After 1993, few instances of reports where social spending issues are not dealt with.
  - Countries with repeatedly relatively high attention in surveillance: Austria, Belgium, Finland, Netherlands, Norway, Luxembourg.
  - The discussion of social spending issues in SIPs follows a similar pattern to Article IV documents.
- Program documents:
  - Number of staff reports without any discussion of social spending issues fell to zero after 1999.
  - In recent years, discussion diverged between PRGT and GRA program cases, with PRGT programs falling slightly.
  - Countries with substantial repeated attention in program reports: Argentina, Bolivia, Kazakhstan, Malawi, Peru.
- Examples of countries with zero frequency counts and outlier years are listed in the source figures and tables.

### Variation by income group and region
- Surveillance documents:
  - Social spending issues are discussed more in high-income countries.
  - By region, surveillance reports discuss social spending more for European countries followed by the Western Hemisphere.
- Program documents:
  - Differences across income groups and regions are less pronounced than in surveillance documents.

### Topic-level emphasis across country groups
- Specific social spending topics receiving most attention:
  - Pensions, health, and education.
- Variation by country group:
  - Pension-related issues figure most prominently in advanced economies.
  - Health and education issues are more prevalent in low-income countries.
  - Energy subsidies, social assistance, income support, and other subsidies are less discussed overall, except for Middle-Eastern countries where subsidy issues were raised more regularly.

### Distributional analysis
- Discussion of distributional issues follows broadly similar temporal pattern as social spending discussion.
- Program staff reports:
  - Discussion of distributional issues increased sharply in program staff reports between 1999 and 2005, declining thereafter with a dip in 2009.
  - Distributional analysis is more prominent in program staff reports than in surveillance staff reports.
  - Within program documents, distributional issues are more discussed in PRGT programs than in GRA programs.
- Surveillance documents:
  - Discussion of distributional issues in surveillance documents has continued to increase in recent years, with a similar pattern shown in SIPs and many repeated outliers.
  - Positive outliers tend to be staff reports for low and middle-income countries for both surveillance and program staff reports.

*IMF ENGAGEMENT ON SOCIAL SPENDING—BACKGROUND PAPERS (excerpt from source content)*

---


_Source: https://www.imf.org/-/media/files/publications/pp/2019/ppea2019017.pdf_