## wpiea2025097-print-pdf - References

## Source details

**Canonical URL:** [wpiea2025097-print-pdf - References](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025097-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025097-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025097-print-pdf.pdf.json)

---

### Major findings and contributions
- IMF-supported programs with macro-financial policy commitments, relative to those without them, are followed by periods of lower non-performing loans (NPLs) and lower credit-to-GDP levels.
- Effects are relatively stronger in countries that have positive credit gaps and lower credit-to-GDP ratios in the years preceding program approval.
- Results are stronger in the pre-Covid period (2010-2019), which coincided with greater prevalence of macro-financial commitments, particularly on strengthening financial regulation and supervision.
- Ex-post policies (those implemented after a crisis) are associated with stronger and more abrupt declines in credit-to-GDP following macro-financial policies.
- Ex-ante policies (those implemented before risks materialize, e.g., regulatory requirements) are followed by milder and more gradual declines in credit-to-GDP.
- Macro-financial policy commitments in MEFPs are stronger when the program is preceded by a positive credit gap.

### Data sources and scope
- Panel dataset of 137 IMF-supported programs during the 2010-2023 period.
- MONA database used to identify structural benchmarks (SBs):
  - Extracted SBs cover 148 programs corresponding to 76 countries.
  - Average number of macro-financial structural benchmarks per program-country pair: 2.5.
  - Programs in the study have an average duration of 2.2 years, and an average of 5.3 reviews.
- MEFP textual analysis dataset constructed via manual selection of text excerpts from MEFPs describing future-tense commitments to macro-financial policy actions; MEFP commitments exclude SBs extracted from MONA.
- NPLs (percent over total gross loans) sourced from IMF Financial Soundness Indicators (FSI) database; gaps filled with manual data collection from IMF staff reports when available.
- Credit gap estimated as deviation of credit-to-GDP relative to a one-sided Hodrick–Prescott (HP) filter using a high smoothing parameter; smoothing parameter—400,000 for quarterly data.
- GDP, inflation, primary fiscal balance, and exchange rates obtained from the World Economic Outlook (WEO) database (July-2024 vintage).
- iMaPP database used to confirm enactment/approval of macroprudential reforms referenced in MEFPs.

### Methodology
- Identification strategy: staggered difference-in-differences approach allowing for dynamic responses and reliance on control group selection.
- Complementary robustness checks: regressions using textual analysis method of Hassan et al. (2019) and comparison with iMaPP policy action timing.
- MEFP textual coding: commitments classified by category (e.g., bank restructuring, resolution of NPLs, liquidity regulation) and by institution type (banks and non-bank financial institutions (NBFIs)); commitments counted per review and summed across reviews.
- Classification rules:
  - Positive (negative) credit gap countries: median positive (negative) credit gap over the pre-program period (pre-program period defined as the 3-year period prior to the first treatment year).
  - High or low credit-to-GDP classification: a country’s median credit-to-GDP during the 3-year pre-program period compared to the median of all 3-year pre-program observations for countries with at least one IMF-supported program between 2007 and 2023.
- Winsorization: NPLs-to-total gross loans, credit-to-GDP, inflation and exchange rate depreciation winsorized at the 1 percent level for both tails; GDP growth winsorized at the 10 percent level.

### Descriptive trends and stylized facts
- Sharp rise in share of macro-financial SB conditionality relative to other policy areas in GRA programs following the GFC.
- During 2009-2015 most commitments fell under legal reform category (SB14); SB31 was significant early but diminished later.
- Around 2016-2018 a new wave of macro-financial conditionality emerged following a global decline in commodity prices affecting oil-exporting countries.
- After post-GFC regulatory reforms, macro-financial conditionality became more muted relative to other conditionality types, including during and after COVID-19.
- MEFPs show increasing importance of macro-financial conditionality following the GFC up to 2016 and a decline in emphasis during 2021-2023.
- Macro-financial conditionality is front-loaded across program reviews; MEFP commitments dominated by financial legislation, then NPL resolution and bank supervision.

### Empirical strategy — identification and sample
- Staggered DiD with country fixed effects (α_i) and year fixed effects (β_t); relative time dummies K_it = t − E_i capture dynamics.
- Controls typically include fiscal balance, nominal bilateral exchange rate depreciation vis-à-vis U.S. dollar, inflation, and dummy for FSAP within three years prior to program.
- Treatment definitions:
  - Treated once a country’s program review contains macro-financial SB conditionality or MEFP commitment; first treatment year is year of introduction; treated is absorbing.
  - Control groups: (1) “Never-treated” controls (ongoing program with no macro-financial SB or MEFP commitments in treatment years), (2) “Not-yet-treated” controls (not in program in year t but will have one with macro-financial conditionality in subsequent years).
- Estimator: staggered DiD estimator by De Chaisemartin and d’Haultfoeuille (2020) to address negative weighting and related issues.

### Main empirical results — NPLs and credit-to-GDP (pre-Covid, 2010-2019)
- MEFP macro-financial commitments (pre-Covid, 2010-2019):
  - Credit-to-GDP decreases by 1.5 percentage points (0.07 SD) in the second year after introduction of MEFP commitments relative to control.
  - NPLs-to-gross-loans decrease by 1.6 percentage points (0.23 SD) in the second year (10 percent significance level).
  - NPLs-to-gross-loans decrease by 2.2 percentage points (0.31 SD) in the third year (5 percent significance level).
- SB13 (general structural benchmark, pre-Covid):
  - No statistically significant effects on credit-to-GDP.
  - NPLs decrease by 5.4 percentage points by the third year (0.78 SD).
- Structural benchmarks implemented at 100 percent:
  - 2.3 percentage point decline in NPLs by the fourth year following commitments; parallel trends assumption holds.
- When structural benchmarks are not fully implemented (<100 percent), results are not statistically significantly different from zero.
- Both SB and MEFP regressions show parallel pre-treatment trends.

### Post-Covid period results and constraints (2020-2023)
- Lower sample size post-Covid limits estimation to first and second treatment years with full controls; third-year estimates require fewer controls.
- In the post-Covid period, neither macro-financial SBs nor MEFP policies are associated with declines in NPLs in any treatment year (no robust decline detected).
- Ex-post macro-financial policies (post-Covid) associated with large declines in credit-to-GDP:
  - 5.3 (0.21 SD) percentage points in the first year.
  - 10.9 (0.42 SD) percentage points in the second year (column 7).
- Post-Covid ex-post policies coincided with higher share of MEFP policies targeted at bank restructuring.

### Policies targeting NPLs and bank resolution — magnitudes (pre-Covid)
- NPL policies constitute 19 percent of total MEFP commitments in 2010-2023.
- Bank resolution policies constitute 6 percent of total MEFP commitments in 2010-2023.
- MEFP commitments on bank resolution (pre-Covid):
  - Credit-to-GDP declines by 4.46 percentage points (0.2 SD) in the second year relative to control.
  - This is almost 3 percentage points larger than pooled MEFP macro-financial commitments (1.5 percent decline).
- SB31 (bank restructuring structural benchmark):
  - Credit-to-GDP declines by 4.09 percentage points (0.18 SD) by the third year.
- MEFP bank resolution third-year decline reported as 6.7 percentage points in Table 4 column 1.
- MEFP commitments on NPL policies:
  - NPL-to-loan ratios decline by 4.43 percentage points (0.63 SD) by the third year.

### Ex-post versus ex-ante distinctions and effects
- Definitions:
  - Ex-ante: build resilience before crises (capital requirements, stress-tests, liquidity, ex-ante NPL policies, financial legislation, bank supervision, macroprudential measures).
  - Ex-post: tackle realized shocks (bank restructuring, bank resolution, ex-post NPL policies).
- Pre-Covid:
  - Ex-post MEFP commitments: credit-to-GDP contracts by -0.9 percentage points (0.04 SD) in the first year and by 3.25 percentage points (-0.15 SD) by the third year.
  - Ex-ante policies: NPL declines similar in magnitude to ex-post in pre-Covid (e.g., coefficient of -2.49 in the 3rd year).
  - Pre-Covid comparisons: ex-ante -2.7 percentage points by third year; ex-post -3.3 percentage points by third year.
- Post-Covid:
  - NPL coefficients from ex-ante policies are non-significant.
  - Ex-post policies show large, abrupt deleveraging.

### Credit growth and denominator concerns
- Concern: declining NPL-to-loan ratios could be driven by credit expansions (denominator effects / evergreening).
- Credit growth regressions (Appendix Table A7):
  - Ex-ante policies: no statistically significant coefficients for credit growth.
  - Ex-post policies: credit growth rises only in the fourth year following MEFP commitments — increase of 4.2 percentage points (significant at 10 percent).
- Despite this, credit-to-GDP consistently declines each year following ex-post policies, implying GDP growth rebounds more strongly than credit growth in these countries.

### Heterogeneity by credit-cycle position and initial financial deepening
- Positive versus negative credit gap (Table 6, 2010-2019):
  - Positive credit-gap countries:
    - NPLs contract by 3.7 percentage points (-0.5 SD) in the third year relative to control.
    - Credit-to-GDP negative in first year; decreases by 1.16 (0.06 SD) percentage points relative to control in third year.
  - Negative credit-gap countries:
    - Insignificant changes in NPLs and credit-to-GDP.
- High versus low credit-to-GDP (Table 7, 2010-2019):
  - SB14 (financial legislation) — Low credit-to-GDP countries:
    - Credit-to-GDP decreases by 0.5 (0.08 SD) percentage points in year of treatment start.
    - Credit-to-GDP decreases to -1.2 (0.2 SD) percentage points by third year.
    - NPL/Gross loans ratio decreases by -7.42 (1.04 SD) percentage points by third treatment year.
  - SB14 — High credit-to-GDP countries: null results for credit-to-GDP and NPLs.
  - SB13 (general macro-financial): followed by increase in credit-to-GDP in third year by 3.9 percentage points (0.58 SD), significant at 5 percent.

### Implementation and MEFP-to-policy translation
- Panel regressions of number of policy reforms from iMaPP on contemporaneous MEFP macroprudential commitments show positive, statistically significant coefficients with and without controls and fixed effects.
- Point estimates:
  - For every MEFP commitment, between 0.09 to 0.12 macroprudential policies are implemented on average in the same year, depending on specification.
- Robustness: staggered DiD with dynamic effects shows each additional macroprudential commitment is associated with implemented macroprudential policies (details in appendix).
- Aggregate implementation estimate highlighted: 0.66 more iMaPP policy implementations in the same year as the commitments (Tables 2 and A6).

### Robustness, sample construction, and caveats
- Robustness checks restrict control group to exclude any macro-financial commitment and compare treated countries to a very small subset with no macro-financial conditionality.
- Winsorization and sample exclusions noted:
  - Winsorize NPLs-to-total gross loans, credit-to-GDP, inflation and exchange rate depreciation at 1 percent both tails; GDP growth at 10 percent tail.
  - Excluded countries despite NPL data availability: Haiti, St. Kitts and Nevis, Niger, Sao Tome and Principe, and Togo for inconsistent NPL series; Mongolia NPL series not included.
- Estimation trade-offs: broader control samples improve size but may reduce comparability; more restrictive controls improve similarity but reduce sample.

### Policy implications and recommendations
- Incorporating macro-financial policy commitments into IMF-supported programs is associated with reductions in NPLs and credit-to-GDP, suggesting program design should account for macro-financial linkages.
- Program design should consider a country’s position in the credit cycle (credit gap) and initial financial deepening, as commitments appear more effective in taming credit-related vulnerabilities in countries with positive credit gaps and lower financial deepening.
- Distinguish between ex-ante and ex-post macro-financial policies in program design given different magnitudes and temporal patterns of impact on credit outcomes.
- Emphasize ex-ante macro-financial policies to achieve milder, gradual adjustments before risks materialize and to avoid costly ex-post corrections.
- Timely introduction and implementation of ex-ante macro-financial policies can improve asset quality (NPL ratios) and tame excessive credit growth when above trend.

*Source: wpiea2025097-print-pdf - References (Macro-Financial Policies and Vulnerabilities in IMF-Supported Programs, Working Paper No. WP/25/097).*

### References .............................................................................................................

### wpiea2025097-print-pdf - References .............................................................................................................

### Major findings and contributions
- IMF-supported programs with macro-financial policy commitments, relative to those without them, are followed by periods of lower non-performing loans and lower credit-to-GDP levels.
- Effects are relatively stronger in countries that have positive credit gaps and lower credit-to-GDP ratios in the years preceding program approval.
- Results are stronger in the pre-Covid period (2010-2019), which coincided with greater prevalence of macro-financial commitments, particularly on strengthening financial regulation and supervision.
- Ex-post policies (those implemented after a crisis) are associated with stronger and more abrupt declines in credit-to-GDP following macro-financial policies.
- Ex-ante policies (those implemented before risks materialize, e.g., regulatory requirements) are followed by milder and more gradual declines in credit-to-GDP.
- Macro-financial policy commitments in MEFPs are stronger when the program is preceded by a positive credit gap.

### Data sources and scope
- Panel dataset of 137 IMF-supported programs during the 2010-2023 period.
- MONA (Monitoring of Fund Arrangements) database used to identify structural benchmarks (SBs).
  - Extracted SBs cover 148 programs corresponding to 76 countries.
  - Average number of macro-financial structural benchmarks per program-country pair: 2.5.
  - Programs in the study have an average duration of 2.2 years, and an average of 5.3 reviews.
- MEFP textual analysis dataset constructed via manual selection of text excerpts from MEFPs describing future-tense commitments to macro-financial policy actions; MEFP commitments exclude SBs extracted from MONA.
- Non-performing loans (NPLs, percent over total gross loans) sourced from IMF Financial Soundness Indicators (FSI) database (IMF, 2019b); gaps filled with manual data collection from IMF staff reports when available.
- Credit gap estimated as deviation of credit-to-GDP relative to a one-sided Hodrick–Prescott (HP) filter using a high smoothing parameter; smoothing parameter—400,000 for quarterly data.
- GDP, inflation, primary fiscal balance, and exchange rates obtained from the World Economic Outlook (WEO) database (July-2024 vintage).
- iMaPP (Integrated Macroprudential Policy) database used to confirm enactment/approval of macroprudential reforms referenced in MEFPs.

### Methodology
- Identification strategy: staggered difference-in-differences approach allowing for dynamic responses and reliance on control group selection.
- Complementary robustness checks: regressions using textual analysis method of Hassan et al. (2019) and comparison with iMaPP policy action timing.
- MEFP textual coding: commitments classified by category (e.g., bank restructuring, resolution of NPLs, liquidity regulation) and by institution type (banks and non-bank financial institutions (NBFIs)); commitments counted per review and summed across reviews.
- Classification rules:
  - Positive (negative) credit gap countries: median positive (negative) credit gap over the pre-program period (pre-program period defined as the 3-year period prior to the first treatment year).
  - High or low credit-to-GDP classification: a country’s median credit-to-GDP during the 3-year pre-program period compared to the median of all 3-year pre-program observations for countries with at least one IMF-supported program between 2007 and 2023.
- Winsorization: NPLs-to-total gross loans, credit-to-GDP, inflation and exchange rate depreciation winsorized at the 1 percent level for both tails; GDP growth winsorized at the 10 percent level.

### Key statistics and sample details
- Number of IMF-supported programs analyzed: 137 (2010-2023).
- Extracted structural-benchmark coverage: 148 programs, 76 countries.
- Average number of macro-financial structural benchmarks per program-country pair: 2.5.
- Average program duration: 2.2 years.
- Average number of reviews per program: 5.3.
- NPL ratio in the pre-Covid sample: average 8.9 percent; standard deviation 7.6 percent.
- HP filter smoothing parameter used for quarterly data: 400,000.
- Pre-Covid sample years used for pre-trend estimation: include 2007-2009 for treatments starting in 2010; pre-Covid programs span the years 2010-2019.
- Post-Covid pre-trend estimation years: include 2017-2019 for treatments starting in 2020; post-Covid programs span the years 2020-2023.
- Structural benchmark categories in MONA analyzed: SB31 (restructuring and privatization of financial institutions), SB14 (legal reforms and regulations), SB13 (other macro-financial reform).

### Conceptual definitions and scope
- Macro-financial policies in this paper include: macroprudential policies, bank restructuring measures, resolution of NPLs, measures to strengthen financial supervision, bank capital and liquidity regulations, and in general any other policy or legislative action that aims at strengthening financial stability.
- Macro-financial policies do not include anti money laundering (AML-CFT) measures.
- Programs covered in the sample: Extended Credit Facility (ECF), Extended Fund Facility (EFF), Exogenous Shock Facility (ESF), Policy Coordination Instrument (PCI), Precautionary Credit Line (PCL), Precautionary Liquidity Line (PLL), Poverty Reduction and Growth Facility (PRGF), Policy Support Instrument (PSI), SBA (Stand-By Arrangement), Standby Credit Facility (SCF). Precautionary programs with no ex-post conditionalities (e.g., FCL) are excluded.

### Policy implications (as presented)
- Incorporating macro-financial policy commitments into IMF-supported programs is associated with reductions in NPLs and credit-to-GDP, suggesting program design should account for macro-financial linkages.
- Program design should consider a country’s position in the credit cycle (credit gap) and initial financial deepening, as commitments appear more effective in taming credit-related vulnerabilities in countries with positive credit gaps and lower financial deepening.
- Distinguish between ex-ante and ex-post macro-financial policies in program design given different magnitudes and temporal patterns of impact on credit outcomes.

*Source: wpiea2025097-print-pdf - References .............................................................................................................*

### 3. Descriptive Trends

### 3. Descriptive Trends

### 3.1 Stylized Facts — evolution of macro‑financial conditionality
- There was a sharp rise in the share of macro‑financial SB conditionality relative to other policy areas in General Resources Account (GRA) programs following the GFC.
- During 2009-2015:
  - Most commitments fell under the legal reform category (SB14).
  - The restructuring and privatization category (SB31) held significant importance at the beginning of this period but diminished towards the end.
- Around 2016-2018 a new wave of macro‑financial conditionality emerged following a global decline in commodity prices that negatively affected oil‑exporting countries.
- After post‑GFC financial regulatory reforms and strengthened banking sector resilience, macro‑financial conditionality became more muted relative to other conditionality types, including during and after the COVID‑19 pandemic.
  - The decline in the share of SB macro‑financial conditionality is attributed to the nature of the COVID‑19 shock: it did not originate in the financial sector; the financial sector confronted it from a position of strength owing to post‑GFC reforms; and unprecedented and timely policy support helped avert a financial meltdown.
- The Memorandum of Economic and Financial Policies (MEFP) shows a different pattern:
  - Two noteworthy periods: (1) increasing importance of MEFP macro‑financial conditionality following the GFC and up to 2016, and (2) a decline in emphasis on macro‑financial conditionality during the Covid period of 2021-2023.
- Distributional and categorical patterns:
  - Macro‑financial conditionality (both SBs and MEFP commitments) has been front‑loaded across program reviews throughout the sample period.
  - Most SB conditionality is under legal reforms across the sample years.
  - MEFP commitments are dominated by financial legislation, followed by commitments to resolve non‑performing loans and bank supervision.

### Macro‑financial vulnerabilities and correlations with conditionality
- Key vulnerability indicators varied between 2010 and 2022: NPLs (to‑gross loans) and credit‑to‑GDP.
  - For GRA program countries:
    - NPLs and credit‑to‑GDP ratios rose sharply following the GFC before declining in the second half of the 2010s.
  - For countries with no IMF‑supported programs:
    - Relatively stable NPLs during the same period and slightly rising credit‑to‑GDP ratios post‑2014.
  - For PRGT countries:
    - Average NPLs surged in 2015-17, coinciding with higher IMF program demand by low‑income countries and falling oil prices.
    - Credit‑to‑GDP ratios remained at a lower baseline relative to GRA countries throughout the sample period.
- Correlations documented:
  - The number of macro‑financial SBs correlates strongly with the percentage of NPLs to total loans.
  - The average number of macro‑financial benchmarks per program responds almost immediately to average NPLs; this holds for both GRA and PRGT program countries.
  - The decline in NPLs for GRA countries after peaking in 2013 is followed by a period of lower emphasis on macro‑financial SB conditionality (measured by percentage of total SBs in active programs).
  - A similar pattern is observed for PRGT countries after NPLs peaked in 2018.
  - The correlation between credit‑to‑GDP ratios and the number of structural benchmarks per program appears weaker, especially for PRGT program countries.

### 4. Empirical Strategy — identification and sample
- Estimation approach:
  - A staggered difference‑in‑differences (DiD) methodology is used to estimate the response of non‑performing loans and credit‑to‑GDP to macro‑financial conditionality, accounting for different timing of IMF‑supported programs across countries.
  - Model components:
    - Y_it denotes the main outcomes of interest.
    - α_i denotes country fixed effects.
    - β_t denotes year fixed effects.
    - K_it = t − E_i denotes dummies for relative time to the event of the first year of a macro‑financial SB conditionality or MEFP commitment introduction in an IMF‑supported program in the country (E_i).
    - X_it denotes observable country characteristics (controls).
    - γ_k are coefficients of interest corresponding to pre‑trends for k<0 and dynamic effects for k≥0; they measure changes in outcomes for treated countries relative to pre‑program reference year and the control group.
  - Controls included, unless otherwise indicated:
    - fiscal balance,
    - nominal bilateral exchange rate depreciation vis‑à‑vis the U.S. dollar,
    - inflation,
    - a dummy equal to 1 if a Financial Sector Assessment Program (FSAP) stability assessment was completed within the three‑year period prior to the program.
- Treatment and control group definitions:
  - Sample includes only countries with an IMF‑supported program anytime during 2010-2023.
  - Treatment group (for any year t):
    - Countries with IMF‑supported programs with structural benchmarks or MEFPs containing macro‑financial policy commitments in at least one program review during the treatment years.
    - A country is considered “treated” for year t and all subsequent program years once macro‑financial conditionality appears in a review; the first treatment year is the year of introduction of a macro‑financial commitment.
    - For countries with multiple programs, the first program is treated as an absorbing state; no multiple treatments are considered.
  - Main control group:
    - (1) “Never‑treated” control: countries in an ongoing program with no macro‑financial SB conditionality or macro‑financial MEFP commitments in the corresponding treatment years.
    - (2) “Not‑yet‑treated” control: countries not in a program in year t but that will have one with macro‑financial conditionality in subsequent years.
  - Robustness and estimator choice:
    - Plain two‑way fixed effects specifications can suffer from issues (e.g., negative weighting); the paper uses the staggered DiD estimator by De Chaisemartin and d’Haultfoeuille (2020) to address such issues.
- Sample selection nuances:
  - Table A3 (appendix) compares treatment and control groups for 2007-2023:
    - In Table A3 column (3), all variables except credit‑to‑GDP, NPLs and change in fiscal balance are statistically identical between control and treatment countries.
    - In Table A3 column (2), inflation is the only variable statistically different between control and treatment countries.
  - These observed differences motivate the inclusion of controls like inflation and motivate potential sample splits by pre‑treatment credit‑to‑GDP levels.
  - Tradeoff noted: larger control samples include advanced economies that may not trend similarly to treated countries; more restrictive controls improve similarity but reduce sample size.

### 5. Results — overview and transmission
- Research objectives:
  - Demonstrate the transmission channel: MEFP commitments → policy implementations → impact on outcomes such as NPLs and credit‑to‑GDP.
  - Using the full sample, analyze how macro‑financial conditionalities and commitments affect asset quality (NPLs) and deleveraging (credit‑to‑GDP).
  - Explore heterogeneity: ex‑ante (pre‑crisis, e.g., regulatory requirements) versus ex‑post (during/after crisis, e.g., restructuring) policies; subsamples by credit‑cycle position and level of credit‑to‑GDP.
- Do MEFP commitments translate into policies?
  - Panel regressions of the number of policy reforms from the iMaPP database on contemporaneous MEFP macroprudential commitments show positive and statistically significant coefficients, both with and without controls and fixed effects.
  - Point estimates:
    - For every MEFP commitment, between 0.09 to 0.12 macroprudential policies are implemented on average in the same year, depending on the specification used.
  - A staggered DiD with dynamic effects (Table A5) also shows each additional macroprudential commitment is associated with implemented macroprudential policies (details in appendix tables).

*Italic: Source — wpiea2025097-print-pdf, Section 3: Descriptive Trends (IMF Working Paper).*

### 0.66 more policy implementations (iMaPP policies) in the same year as the commitments. Tables 2 and A6

### wpiea2025097-print-pdf - 0.66 more policy implementations (iMaPP policies) in the same year as the commitments. Tables 2 and A6

### Robustness and sample construction
- A robustness analysis restricts the control group to exclude any other macro-financial commitment and compares treated countries with specific macro-financial commitments (such as SB13 or SB31) to a very small subset of countries with no macro-financial conditionality whatsoever.
- Panel regressions (Table 2) use the dependent variable: number of macroprudential policies implemented in a given year; independent variable: number of macroprudential MEFP commitments.
- Controls in these regressions include: real GDP growth, fiscal balance change, exchange rate depreciation and a dummy equaling one if there was an FSAP in the last three years and zero otherwise.
- Sample for Table 2 runs from 2010-2019.

### Full-sample staggered diff-in-diff analysis (Section 5.2)
- Main empirical focus: estimating changes in two outcomes following macro-financial commitments/conditionality: the NPL-to-gross loans ratio and the credit-to-GDP ratio.
- Analysis covers full sample of countries and years, with granular results for specific categories of macro-financial SBs conditionalities and MEFP commitments.

### NPLs and Credit-to-GDP following MEFP commitments and SB conditionality (Section 5.2.1; Table 3)
- Pre-Covid period defined as 2010-2019; Post-Covid period defined as 2020-2023.
- MEFP macro-financial commitments (pre-Covid, 2010-2019):
  - Credit-to-GDP decreases by 1.5 percentage points (0.07 standard deviation, SD) in the second year after introduction of MEFP commitments relative to control.
  - NPLs-to-gross-loans decrease by 1.6 percentage points (0.23 SD) in the second year (10 percent significance level).
  - NPLs-to-gross-loans decrease by 2.2 percentage points (0.31 SD) in the third year (5 percent significance level).
- For general structural benchmark SB13 (pre-Covid):
  - No statistically significant effects on Credit-to-GDP.
  - NPLs decrease by 5.4 percentage points by the third year (0.78 SD).
- When restricting to structural benchmarks implemented at 100 percent:
  - A 2.3 percentage point decline in non-performing loans by the fourth year following commitments, with the parallel trends assumption holding.
- When structural benchmarks are not fully implemented (<100 percent), results are not statistically significantly different from zero; observed increase in NPLs before implementation followed by a decline after implementation (not statistically significant).
- Both SB and MEFP regressions show parallel trends in the pre-conditionality/commitment period (no significant pre-treatment differences).

### Post-Covid period estimation constraints and results
- Lower sample size in post-Covid period limits estimation to first and second treatment years when including all pre-Covid control variables; only three control variables included when estimating third treatment year.
- In the post-Covid period, both macro-financial SBs and MEFP policies are not associated with declines in NPLs in any treatment year (no robust decline detected).

### Case studies: Grenada, Burkina Faso, Suriname (Section 5.2.2; Box 1)
- Grenada (ECF 2014-2017):
  - Macro-financial MEFP commitments introduced in 2014 (vertical dashed line).
  - Context: high NPLs due to prolonged GDP decline between 2008-2012; reforms included NPL policies such as reforming loan classifications and provisioning requirements.
  - Source: Staff Report First Review for the Extended Credit Facility (ECF) for Grenada (2014).
- Burkina Faso (ECF 2013-2016):
  - SB13 introduced in 2013 (vertical dashed line); SB13 aim to increase access to finance and are associated with increases in credit-to-GDP through increased access to finance (see appendix figure A7).
  - Source: Staff Report Seventh Review ECF Burkina Faso (2014).
- Suriname (EFF 2021-2023):
  - Macro-financial MEFP commitments introduced in 2021 (vertical dashed line).
  - Program approved December 2021 for 36 months (EFF); included roadmaps for bank resolution plans which could lead to continued decline in credit-to-GDP in immediate post-Covid period (2022-2023).
  - Source: Staff Report Request for EFF Suriname (2021).

### Policies targeting NPLs and credit-to-GDP (Section 5.2.3; Table 4)
- Rationale for focusing on NPL-related and bank resolution policies:
  - NPL policies constitute 19 percent of total MEFP commitments in 2010-2023.
  - Bank resolution policies constitute 6 percent of total MEFP commitments in 2010-2023.
  - NPL policies directly affect NPL outcomes; bank resolution policies can affect the stock of credit.
- MEFP commitments on bank resolution (pre-Covid, Table 4, column 1):
  - Associated with larger declines in credit-to-GDP relative to pooled MEFP results.
  - Credit-to-GDP ratio declines by 4.46 percentage points (0.2 SD) in the second year relative to control.
  - This magnitude is almost 3 percentage points larger than pooled MEFP macro-financial commitments (1.5 percent decline in Table 3, column 1).
- SB31 (bank restructuring structural benchmark, Table 4, column 5):
  - Credit-to-GDP declines by 4.09 percentage points (0.18 SD) by the third year of treatment.
  - MEFP bank resolution third-year decline reported as 6.7 percentage points in column 1 of Table 4.
  - Interpretation: MEFP bank resolution commitments tend to decrease credit-to-GDP (often involve bank closure), while SB restructuring (SB31) can include recapitalization which may bolster credit-to-GDP.
- MEFP commitments on NPL policies (Table 4, column 2):
  - NPL-to-loan ratios decline by 4.43 percentage points (0.63 SD) by the third year (pre-Covid average).
- SB14 (financial legislation) as treatment (Table 4, column 6; appendix Table A6):
  - Pre-Covid: decline in NPL-to-loan ratios by the third year with coefficient magnitude -2.8 percentage points (not significant at 10 percent).
  - Appendix Table A6 / Table A7 restricting control sample produces coefficient -2.7 percentage points (-0.38 SD) for third-year decline with 10 percent significance level.

### Ex-post versus ex-ante MEFP commitments (Section 5.3; Table 5)
- Definitions:
  - Ex-ante MEFP commitments: policies to build resilience before crises (bank capital requirements, stress-tests, liquidity, ex-ante NPL policies, financial legislation, bank supervision, macroprudential measures).
  - Ex-post MEFP commitments: policies to tackle realized shocks (bank restructuring, bank resolution, ex-post NPL policies).
- Pre-Covid results (Table 5, columns 5 and 7):
  - Ex-post MEFP commitments show earlier and stronger contractions in credit-to-GDP.
  - Credit-to-GDP contracts by -0.9 percentage points (0.04 SD) in the first year and by 3.25 percentage points (-0.15 SD) by the third year of the commitment in the pre-Covid period.

*Source: wpiea2025097-print-pdf - 0.66 more policy implementations (iMaPP policies) in the same year as the commitments. Tables 2 and A6*

### 5. In the post-Covid period, these coefficients are larger in magnitude and statistically significant, showing

### 5. In the post-Covid period, these coefficients are larger in magnitude and statistically significant, showing

### Effects of Ex-Post Versus Ex-Ante Macro-Financial Policies
- Ex-post macro-financial policies (post-Covid) are associated with large declines in credit-to-GDP:
  - 5.3 (0.21 SD) percentage points in the first year.
  - 10.9 (0.42 SD) percentage points in the second year (column 7).
- Ex-post policies have therefore led to large deleveraging, especially in the post-Covid period, which coincided with a larger percentage of MEFP policies targeted at bank restructuring.
- Ex-ante policies are associated with NPL declines similar to ex-post policies in the pre-Covid period:
  - Coefficient of -2.49 in the 3rd year of the treatment.
- Pre-Covid comparisons (ex-ante versus ex-post):
  - Ex-ante: -2.7 percentage points by the third year of the treatment.
  - Ex-post: -3.3 percentage points by the third year of the treatment.
  - Interpretation: ex-post policies imply abrupt and substantial changes (larger corrections) and are associated with larger deleveraging; ex-ante policies are more gradual and preemptive.
- Post-Covid: coefficients for NPL ratios from ex-ante policies are non-significant (column 4).

### Credit Growth and Potential Denominator Effects on NPL Ratios
- Concern: declining NPL-to-loan ratios may be driven by credit expansions affecting the denominator (evergreening mechanism).
- Separate regressions for credit growth (Appendix Table A7):
  - Ex-ante policies: no statistically significant coefficients for credit growth.
  - Ex-post policies: credit growth rises only in the fourth year following MEFP commitments:
    - Increase of 4.2 percentage points (statistically significant at the 10 percent level).
  - Despite this, credit-to-GDP consistently declines each year following ex-post policies, implying GDP growth rebounds more strongly than credit growth in these countries.

### Heterogeneity by Position in the Credit Cycle (Positive Versus Negative Credit Gap)
- Classification: countries split by median positive or negative credit-to-GDP gap using the pre-treatment period (three years before treatment), with the credit gap measured by deviations of credit-to-GDP using a one-sided HP filter.
- Key findings (Table 6, 2010-2019):
  - Positive credit-gap countries:
    - NPLs contract by 3.7 percentage points (-0.5 SD) in the third year relative to control.
    - Credit-to-GDP is negative in the first year after treatment; decreases by 1.16 (0.06 SD) percentage points relative to control in the third year.
    - Interpretation: deleveraging for positive credit-gap countries.
  - Negative credit-gap countries:
    - Insignificant changes in NPLs.
    - Null-results for credit-to-GDP as outcome variable.

### Heterogeneity by Initial Level of Credit-to-GDP (High Versus Low Financial Deepening)
- Method: split countries above/below the median of credit-to-GDP in the 3 years prior to program start.
- Financial legislation SBs (SB14) effects (Table 7, 2010-2019):
  - Low credit-to-GDP countries:
    - Credit-to-GDP decreases by 0.5 (0.08 SD) percentage points in the year of treatment start.
    - Credit-to-GDP decreases further to –1.2 (0.2 SD) percentage points by the third year after treatment start.
    - NPL/Gross loans ratio decreases by –7.42 (1.04 SD) percentage points by the third treatment year relative to control.
    - Interpretation: financial legislation associated with deleveraging and large NPL improvements in low credit-to-GDP countries.
  - High credit-to-GDP countries:
    - Null results for credit-to-GDP and NPLs relative to control.
- General macro-financial commitments (SB13) effects:
  - SB13 is followed by an increase in credit-to-GDP in the third year after treatment start by 3.9 percentage points (0.58 SD), statistically significant at the 5 percent level.
  - Explanation: SB13 includes policies such as “improving access to financial services,” consistent with observed credit increases.

### Conclusions and Policy Recommendations
- Broad results:
  - Macro-financial conditionality in IMF-supported programs tends to decrease NPL-to-loan ratios and, in some cases, credit-to-GDP ratios (relative to control groups), especially prior to the COVID-19 pandemic.
  - NPL-specific policies and bank resolution policies have larger responses for NPLs and credit-to-GDP respectively.
- Targeting and timing matter:
  - Countries entering programs with a positive credit gap benefit most from deleveraging and reductions in NPL-to-loan ratios.
  - Countries with low initial credit-to-GDP (lower financial development) tend to receive policies aimed at financial inclusion and may experience higher credit-to-GDP following a program.
- Post-Covid pattern:
  - Only sharp declines in credit-to-GDP hold for ex-post macro-financial policies aimed at tackling realized shocks.
- Policy implications:
  - Emphasize ex-ante macro-financial policies to achieve milder, gradual adjustments before risks materialize and to avoid costly ex-post corrections.
  - Program design should consider the country’s position in the credit cycle and the initial level of financial deepening.
  - Timely introduction and implementation of ex-ante macro-financial policies can improve asset quality (NPL ratios) and tame excessive credit growth when above trend.

*Source: IMF Working Paper — Macro-Financial Policies and Vulnerabilities in IMF-Supported Programs (selected section).*

### References

### References and Appendix (wpiea2025097-print-pdf - References)

### Key bibliographic references
- Acharya, V. V., Engle, R., and Pierret, D. (2014). Testing macroprudential stress tests: The risk of regulatory risk weights. Journal of Monetary Economics, 65, 36-53.
- Alam, Z., Alter, M. A., Eiseman, J., Gelos, M. R., Kang, M. H., Narita, M. M., Nier, E. and Wang, N. (2019), “Digging deeper–Evidence on the effects of macroprudential policies from a new database”, International Monetary Fund.
- Ari, A., Chen, S., and Ratnovski, L. (2020). "The Dynamics of Non-Performing Loans During Banking Crises: A New Database." IMF Working Paper No. 20/119.
- Balima, H. W. and Sokolova, A. (2021), “IMF programs and economic growth: A meta-analysis”, Journal of Development Economics 153, 102741.
- Berger, A. N., and Bouwman, C. H. S. (2013). "How Does Capital Affect Bank Performance During Financial Crises?" Journal of Financial Economics, 109(1), 146-176.
- (Additional references listed in source covering macroprudential policy, bank regulation, program conditionality, systemic banking crises, text-analytic methods, and IMF publications.)

### Appendix — Summary statistics and data handling notes
- Winsorization:
  - NPLs-to-total gross loans, credit-to-GDP, inflation and exchange rate Depreciation are winsorized at the 1 percent level for both tails.
  - GDP growth is winsorized at the 10 percent level.
- Pre-trend sample inclusion:
  - The number of observations include the years 2007-2009 to estimate pre-trends.
- Credit gap definition and sample selection:
  - The credit-to-GDP gap is sourced from the IMF's Systemic Risk Tracker, estimated using a one-sided HP filter with a high smoothing parameter, as proposed by Borio and Lowe (2002).
  - A country belongs to the positive credit gap sample if in the 3 years prior to the first treatment year of SB14 the median credit-to-GDP gap of the country is positive and vice-versa for the negative credit gap sample.
  - The majority i.e. 61 percent of our pre-Covid full sample belong to the positive credit gap sample.
- Pre-Covid program sample timing:
  - The programs included in the pre-Covid sample span the years 2020-2023.
- Country exclusions and data gaps:
  - Excluded despite NPL data availability in staff reports: Haiti, St. Kitts and Nevis, Niger, Sao Tome and Principe, and Togo, due to lack of consistency of the NPL series across time.
  - Dataset does not include NPLs series of Mongolia.
  - Remaining differences in sample size between NPLs and credit-to-GDP are due to countries missing NPLs in specific series.

### Appendix — Alternative sample definitions and grouping
- Credit level sample:
  - High credit-to-GDP sample: countries whose median credit-to-GDP level in the 3 years prior to first treatment year of SB14 is lower than the median credit-to-GDP of a pooled sample based on their income group; vice-versa for low credit-to-GDP sample.
- Time coverage for program consideration:
  - Only programs starting in 2010 are considered in some figures and pre-trend tests.
- Cross-section size for treatment/control distributions:
  - The cross-section of countries equals 68 units for figures plotting switching into treatment by year (sample years 2007-2019).

### Textual analysis (MEFP commitments collection and robustness)
- Pre-processing steps (following Hassan et al. (2019)):
  - Remove punctuation, capitals and stopwords; split text into bigrams.
  - Create macro-financial treatment intensity: count macro-financial bigrams found in the Basel Framework and divide by total bigrams in the MEFP.
- Recategorization:
  - Continuous treatment-intensity variable is recategorized into five percentiles of 20 percent each to obtain treatment intensity coefficients.
- Notes on computational granularity:
  - The continuous variable has fine steps (examples: 0.2 vs 0.21 macro-financial bigrams over total bigrams), motivating percentile recategorization.

### MEFP commitments — classification and examples (Ex-Ante and Ex-Post)
- Ex-Ante Policies categories and examples:
  - Bank capital requirements: increases in Tier 1 Capital (e.g. Honduras 2019, Jordan 2012); increases in Tier 2 Capital (e.g. Mongolia 2018).
  - Stress-tests: implement stress test on credit risk (e.g. Albania 2014); interest rate risk and foreign exchange risk (e.g. Egypt 2016).
  - Liquidity policies: deposit insurance scheme (e.g. Mozambique 2013); Net Stable Funding Ratio (e.g. Armenia 2019).
  - NPL Policies (Ex-Ante): set up new credit registry (e.g. Niger 2017); wider loan coverage and strengthen provisioning (e.g. Moldova 2017, Seychelles 2015); Asset Quality Review (e.g. Grenada 2014).
  - Financial legislation (Ex-Ante): corporate governance (e.g. Ukraine 2018); bolster supervisory powers for resolution and restructuring (e.g. Pakistan 2013); strengthen bank licensing (e.g. Armenia 2014).
  - Bank supervision (Ex-Ante): increase on-site and off-site visits (e.g. Bangladesh 2013); implement risk-based supervision (e.g. Sierra Leone 2017).
  - Macroprudential (Ex-Ante): implement Countercyclical Capital Buffer (CCyB) (e.g. Georgia 2012); Capital Conservation Buffer (CCB) (e.g. Georgia 2016); DSTI limits (e.g. Kyrgyz Republic 2015); LTV limits (e.g. Sri Lanka 2016).
- Ex-Post Policies categories and examples:
  - Bank restructuring (Ex-Post): private sector recapitalization of weak (non-failed) banks (e.g. Tunisia 2013); remove obstacles to asset and collateral valuation to aid sell-off (e.g. Romania 2011).
  - Bank resolution (Ex-Post): winding down banks via M&A (e.g. Serbia 2015); set up Bad Bank structures (e.g. Kyrgyz Republic 2011).
  - NPL (Ex-Post): loan recovery commitments and expedite private debt restructuring (e.g. Ghana 2015).

### Figures and descriptive diagnostics (selected themes)
- Macro-Financial Balancing figures:
  - Figure A1 and A2 plot averages for treatment vs control groups by year for variables including NPL, Credit-to-GDP, Credit growth, Inflation, GDP growth, Probability of FSAP in the last 3 years.
  - Control group definition: IMF-supported program countries with no or not yet a financial reform MEFP commitment (or no general financial SB conditionality) in corresponding years.
  - Treatment group definition: IMF-supported program countries with any macro-financial MEFP commitment (or general financial SB conditionality) in at least one program review during that treatment year; once treated, countries are considered treated for that year and following years of the program.
  - Inclusion of 2007-2009 to test for pre-trends; programs starting in 2010 are considered.
- Frontloading and timing of conditionality:
  - Figures (A10–A13) address average macro-financial MEFP commitments per program, breakdown by GRA and PRGT, frontloading of macro-financial commitments across reviews (2010-2023), and average ex-ante vs ex-post MEFP commitments per year.
  - GRA refers to general resource programs for non-low-income countries; PRGT refers to low-income countries.
- Correlations and time-series:
  - Figures A18–A21 explore correlations of macro-financial SBs with NPLs, and evolution of NPLs and macro-financial SBs for GRA and PRGT countries.
  - Figure A21 plots evolution of Credit-to-GDP and macro-financial SBs (PRGT).
- Program counts and distributions:
  - Figures A22–A23 show distribution by year of countries switching into treatment and control for sample years 2007-2019 for any macro-financial MEFP commitment and for general macro-financial SB conditionality (SB13).

### Regression analyses and tables (overview)
- Textual analysis regressions:
  - Table A4: Textual Analysis Robustness – Credit-to-GDP and NPLs following higher prevalence of macro-financial content in MEFPs.
- Implementation regressions:
  - Table A5: Implementation of Macroprudential Policies Following Macroprudential MEFP Commitments — treatment indicator restricted to macroprudential commitments (LTVs, CCyBs, and DSTIs) and dependent variable from iMaPP data.
- NPLs and credit-to-GDP conditionality regressions:
  - Table A6: NPLs and Credit-to-GDP Following Financial Legislation Conditionality (SB14) in Pre- and Post-Covid Periods.
- Credit growth regressions:
  - Table A7: Credit Growth Following Financial SB14, Ex-Ante and Ex-Post Conditionality for 2010-2019 (positive and negative credit gap samples).
  - Table A8: Credit Growth and Credit-to-GDP Following Financial SB13 Conditionality Using High and Low Credit-to-GDP Samples for 2010-2019.
- NPL implementation splits:
  - Table A9: NPLs following financial SB13 conditionality using 100 percent Implemented and at least 1 not implemented SB samples for 2010-2019.
- Country lists and crisis notes:
  - Table A10 A and B: Lists of countries in High and Low Credit Gap samples including note of the year the country experienced a crisis according to Laeven and Valencia (2020).

*Source: wpiea2025097-print-pdf - References (Macro-Financial Policies and Vulnerabilities in IMF-Supported Programs, Working Paper No. WP/25/097).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025097-print-pdf.pdf_