## _wp14176

## Source details

**Canonical URL:** [_wp14176](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14176.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14176.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14176.pdf.json)

---

### I. Introduction — aim and scope
- Objective: assess effectiveness of the IMF’s debt limits policy (DLP) on borrowing behavior in countries eligible to IMF concessional lending (PRGT-eligible).
- Key features of DLP:
  - Distinguishes loans by concessionality (grant element).
  - Concessional loans typically defined as grant element of 35 percent or higher (threshold can be higher for high risk of debt distress).
  - Under DLP access to non-concessional borrowing is limited while highly concessional borrowing is generally unconstrained.
- Analytical approach:
  - Use IMF program participation to proxy presence of DLP (DLP applies only to countries under IMF programs).
  - Focus on countries eligible for concessional financing.
  - Main identification challenge: endogeneity / selection into IMF programs; solution: propensity score matching (PSM).

### II. Data and methodology
- Sample and timing:
  - 70 countries eligible to receive concessional financing throughout 1986–2011.
  - Panel arranged in five-year frequencies: 1986–91, 1992–96, 1997–2001, 2002–06, 2007–11.
  - 57 benefited from IDA-only lending; 13 received a mix of IDA and IBRD lending.
  - 39 countries received debt relief under HIPC and MDRI.
- Treatment definition:
  - DLP dummy = 1 if country is under an IMF program for at least three years in a five years window.
  - Qualifying programs: ECF (and predecessors PRGF, ESAF, SAF), SBA, ESF, SCF, and PSIs.
- Outcome variables:
  - Size of borrowing: total PPG loan commitments and disbursements as a share of GDP; breakdowns into concessional vs non-concessional; official (bilateral/multilateral) vs private; IDA loans treated separately.
  - Terms of borrowing: average grace period, average interest rate, average maturity, average grant element on new external debt commitments.
  - Data sources: World Bank’s World Development Indicator, IMF’s WEO, World Bank’s Debt Reporting System (DRS).
  - Note on concessionality measures: DRS defines concessional debt as original grant element of 25 percent using 10 percent discount rate; DLP uses 35 percent grant element calculated using CIRRs (until Oct 2013) and since then 5 percent fixed discount rate. Authors computed alternative grant element with 5 percent discount rate for robustness.
- Estimation strategy:
  - Propensity score estimated via probit with time dummies.
  - Matching: radius matching (R=0.05 and R=0.1) and kernel matching (Epanechnikov).
  - Standard errors obtained by bootstrapping (500 replications).
  - Identifying assumptions: Conditional Independence Assumption (CIA) and common support.
  - Robustness checks: alternative selection model, analysis of difference in outcomes, Rosenbaum sensitivity analysis.

### III. Participation equation and predictors of IMF program entry
- Method: probit estimation of probability of being under IMF program / DLP.
- Predictors and findings:
  - Initial reserves shortage: significantly increases likelihood of entering IMF program (Initial Reserves coefficient negative and significant).
  - Bilateral aid: no significant correlation in baseline sample (Aid-to-GDP mixed significance across specs).
  - Poorer countries (lower GNI per capita): higher probability of IMF program (GNI per capita coefficient negative and significant; e.g., -0.510*** (0.197) in Table 6 Column (1)).
  - Resource rents: higher resource rents associated with lower probability of IMF program (e.g., -0.037** (0.015)).
  - Inflation: CPI inflation negatively associated with program probability in this sample (e.g., -0.018** (0.008)).
  - Structural/institutional characteristics:
    - Landlocked countries: higher probability (landlocked 0.512** (0.228)).
    - World Bank CPIA positively and significantly correlated with program probability (CPIA 0.538*** (0.194)).
    - Democracy not significant.
  - Time dummies: later periods show higher program probability (e.g., 5-period dummy 1.274*** (0.398) in Table 6).

### IV. Average treatment effect (ATT) estimates — main results
- Matching approach: propensity scores with radius and kernel matching; common support enforced.
- Core ATT findings (selected, from Table 7 and related tables):
  - Disbursements PPG (as % of GDP):
    - Kernel r=0.05: 0.77** (0.387)
    - Kernel r=0.1: 0.60 (0.386)
    - Radius matching: 0.79** (0.389)
  - Disbursements PPG concessional external debt (as % of GDP):
    - Kernel r=0.05: 0.72** (0.33)
    - Kernel r=0.1: 0.59* (0.338)
    - Radius matching: 0.71** (0.33)
  - Disbursements PPG non-concessional external debt (as % of GDP): 0.05 (0.161); 0.00 (0.161); 0.07 (0.159) — not statistically significant.
  - Disbursements PPG IDA (as % of GDP): 0.32* (0.189); 0.31* (0.181); 0.31* (0.18)
  - Disbursements PPG bilateral (as % of GDP): 0.21* (0.128); 0.17 (0.119); 0.23* (0.125)
  - Disbursements PPG multilateral (as % of GDP): 0.60* (0.321); 0.48 (0.332); 0.59** (0.305)
  - Commitments PPG (as % of GDP):
    - Kernel r=0.05: 1.56*** (0.536)
    - Kernel r=0.1: 1.39*** (0.55)
    - Radius matching: 1.60*** (0.531)
  - Commitments PPG IDA (as % of GDP): 0.52** (0.256); 0.47* (0.262); 0.51** (0.252)
  - Terms of borrowing (average grant element, average interest, average maturity, average grace period): ATTs generally not statistically significant across matching specifications.
- Interpretation by authors:
  - Results consistent with a catalytic role of IMF programs in attracting concessional official financing to PRGT-eligible countries.
  - No evidence that DLP significantly increases non-concessional borrowing or private creditor lending over this period.

### V. Robustness checks
- Alternative selection model (parsimonious probit): matching results broadly in line with baseline.
- Outcome in differences (change vs level):
  - Significant change observed in Commitments PPG: 0.315*** (0.101) in kernel change specification (Table A3).
  - No significant change in total and concessional disbursements in differences specification.
  - Authors note capacity constraints may limit absorption (commitments > disbursements).
- Hidden bias (Rosenbaum sensitivity, Table A4):
  - Commitments PPG: Γ = 1.7; Probability = 0.032
  - Disbursements PPG concessional: Γ = 1.4; Probability = 0.042
  - Disbursements PPG: Γ = 1.3; Probability = 0.038
  - Authors caution interpreting sensitivity results.

### VI. Heterogeneity analysis — control function regressions
- Purpose: test whether average null effects mask heterogeneity across countries (sources tested: GNI per capita, infrastructure gap, growth prospects, total debt as percent of GDP).
- Approach: control function regression (OLS within common support) with interactions between DLP dummy and heterogeneity variables; propensity score included as control.
- Determinants of average grant element (Table 8 highlights):
  - Propensity score positively and significantly correlated with grant element (e.g., Propensity score 21.665*** (4.704)).
  - DLP dummy positive in some specifications (e.g., DLP 7.643*** (2.404) in Column (1)) but not robust across all specs.
  - Logarithm of GNI per capita negatively and significantly correlated with grant element (e.g., -8.387*** (1.844)): richer LICs receive smaller grant element.
  - Telephone lines per 100 people (av5) negative and significant in one spec: -0.488* (0.265); becomes insignificant when GNI per capita controlled.
  - Growth prospects proxy and total debt percent of GDP not significant.
- Determinants of non-concessional borrowing (Table 9 highlights):
  - Propensity score negative and significant in several specs (e.g., -1.975*** (0.519) in Column (1)).
  - DLP dummy mixed: -0.407* (0.221) in one spec; other specs show varied DLP estimates.
  - Logarithm of GNI per capita positively and significantly correlated with non-concessional borrowing in some specs (e.g., 0.578** (0.249)).
  - Telephone lines per 100 people (beginning of 5 years) positive and significant in one spec: 0.067** (0.026).
  - Interaction terms (DLP x covariates): most not statistically significant — limited evidence that DLP effects vary systematically by these heterogeneity dimensions.

### VII. Conclusions and policy implications
- Empirical conclusions:
  - Concessional borrowing (as % of GDP) is significantly higher in countries under DLP — consistent with catalytic effect of IMF programs in attracting concessional official financing or reflecting correlated determinants of both IMF programs and donor allocations.
  - No evidence that DLP significantly impacts:
    - Level of non-concessional borrowing.
    - Terms of new borrowing (interest rate, grace period, maturity, grant element).
    - Private creditor lending.
  - Heterogeneity analysis: non-concessional borrowing increases with country development (GNI per capita), suggesting development, not absence of DLP constraints, explains higher non-concessional borrowing in relatively richer LICs.
- Policy implications emphasized by authors:
  - Removing concessionality requirements under DLP is unlikely, by itself, to cause a major shift toward non-concessional financing in the poorest LICs.
  - The 2009 reform that tailors debt limits to country circumstances (capacity and debt vulnerability) could change DLP impact; limited post-2009 coverage in sample — warrants future work.
  - Further research suggested:
    - Examine debt dynamics since the 2009 reform to test whether DLP impact on non-concessional borrowing has increased.
    - Investigate potential overuse of concessionality requirements in LIC programs that could dilute impact or mask interaction with fiscal conditionality; study granular interactions of debt limits and fiscal conditionality.

*Source: _wp14176 - 3. Auxiliary Tables (IMF staff calculations).*

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

### _wp14176 - References .............................................................................................................

### References
- References ................................................................................................................................34

### Appendixes
- 1. Figures..................................................................................................................................19
- 2. Tables ...................................................................................................................................21

*Source: _wp14176 - References .............................................................................................................*

### 3. Auxiliary Tables.....................................................................................................

### 3. Auxiliary Tables

### I. Introduction — aim and scope
- Objective: assess effectiveness of the IMF’s debt limits policy (DLP) on borrowing behavior in countries eligible to IMF concessional lending (PRGT-eligible).
- Key features of DLP:
  - Distinguishes loans by concessionality (grant element).
  - Concessional loans typically defined as grant element of 35 percent or higher (threshold can be higher for high risk of debt distress).
  - Under DLP access to non-concessional borrowing is limited while highly concessional borrowing is generally unconstrained.
- Analytical approach: use IMF program participation to proxy presence of DLP (DLP applies only to countries under IMF programs); focus on countries eligible for concessional financing.
- Main identification challenge: endogeneity / selection into IMF programs; solution: propensity score matching (PSM).

### II. Data and methodology
- Sample and timing:
  - 70 countries eligible to receive concessional financing throughout 1986–2011.
  - Panel arranged in five-year frequencies: 1986–91, 1992–96, 1997–2001, 2002–06, 2007–11.
  - Among 70 countries: 57 benefited from IDA-only lending and 13 received a mix of IDA and IBRD lending.
  - 39 countries received debt relief under HIPC and MDRI.
- Treatment definition:
  - Dummy = 1 if country is under an IMF program for at least three years in a five years window.
  - Qualifying programs include ECF (and predecessors PRGF, ESAF, SAF), SBA, ESF, SCF, and PSIs.
- Outcome variables:
  - Size of borrowing: total PPG loan commitments and disbursements as a share of GDP; breakdowns into concessional vs non-concessional, official (bilateral/multilateral) vs private; IDA loans treated separately.
  - Terms of borrowing: average grace period, average interest rate, average maturity, average grant element on new external debt commitments.
  - Data sources: World Bank’s World Development Indicator, IMF’s WEO, World Bank’s Debt Reporting System (DRS).
  - Note: DRS defines concessional debt as original grant element of 25 percent using 10 percent discount rate; DLP uses 35 percent grant element calculated using CIRRs (until Oct 2013) and since then 5 percent fixed discount rate. Authors computed alternative grant element with 5 percent discount rate for robustness.
- Estimation strategy:
  - Propensity score estimated via probit; time dummies included.
  - Matching algorithms: radius matching (R=0.05 and R=0.1) and kernel matching.
  - Standard errors computed by bootstrapping.
  - Key identifying assumptions: Conditional Independence Assumption (CIA) and common support.
  - Robustness checks include alternative selection model, analysis of difference in outcomes, Rosenbaum sensitivity analysis for hidden bias.

### III. Participation equation and predictors of IMF program entry
- Method: probit estimation of probability of being under IMF program / DLP.
- Predictors and findings (consistent with literature where noted):
  - Official reserves shortage: significantly increases likelihood of entering IMF program.
  - Bilateral aid: no significant correlation in this sample.
  - Poorer countries: higher probability of IMF program.
  - Resource rents: higher resource rents associated with lower probability of IMF program.
  - Inflation: countries with high inflation were slightly less likely to sign an IMF program in this sample.
  - Structural characteristics: landlocked resource-scarce countries have higher probability of IMF program.
  - Institutional characteristics: World Bank CPIA positively and significantly correlated with program probability; democracy not significant.
  - Political/economic links with Fund: country size and IMF quota positive but not significant in this sample.

### IV. Average treatment effect (ATT) estimates — main results
- Matching approach: used propensity scores and implemented radius and kernel matching; common support enforced by minimum-maximum criterion.
- Core ATT findings (reported in Table 7 and discussed in text):
  - Total amount of borrowing (disbursement and commitment) is significantly higher in countries under the DLP than it would have been without the policy.
  - Concessional borrowing (disbursement and commitment) is significantly higher in countries under the DLP.
  - Non-concessional borrowing: ATT not significant (no evidence DLP significantly impacts level of non-concessional borrowing).
  - Creditor composition:
    - Multilateral and bilateral loans: higher amounts in DLP countries.
    - Private creditor lending: no statistically significant difference between DLP and control groups.
  - Terms of borrowing (average grant element, average interest rate, average maturity, average grace period): ATTs not significant (no detectable effect of DLP on terms of new commitments).
- Interpretation highlighted by authors:
  - Evidence consistent with a catalytic role of IMF programs in attracting concessional resources to PRGT-eligible countries.
  - Lack of effect on non-concessional and private lending may reflect limited private lender catalysis in countries with weak fundamentals and/or that LICs were not able to attract significant non-concessional financing over the period.

### V. Robustness checks
- Alternative selection model: parsimonious probit including only significant covariates; matching results broadly in line with baseline.
- Outcome in differences (change vs level): significant change in total borrowing commitment observed under DLP; no significant change in total and concessional disbursements.
  - Authors suggest capacity constraints in recipient countries may limit ability to absorb committed resources (commitments > disbursements).
- Hidden bias:
  - Rosenbaum sensitivity analysis indicates total commitment and concessional disbursement are least sensitive to hidden bias (Table A4).
  - Authors caution sensitivity results should be interpreted carefully.

### VI. Heterogeneity analysis — control function regressions
- Purpose: test whether average null effects mask heterogeneity across countries (sources of heterogeneity tested: GNI per capita, infrastructure gap, growth prospects, total debt as percent of GDP).
- Approach: control function regression (OLS within common support) with interaction terms between DLP dummy and heterogeneity variables; propensity score included as control.
- Key findings (Tables 8 and 9):
  - Average grant element (terms):
    - After controlling for propensity score, DLP not significantly associated with average grant element (Columns 1–2).
    - GNI per capita negatively correlated with grant element (Column 3): richer LICs receive less favorable financing terms (smaller grant element).
    - Infrastructure proxy (telephone lines per 100 people, 5-year average) negative and significant (Column 4): interpreted as infrastructure gap associated with higher grant element; once GNI per capita controlled, infrastructure gap becomes insignificant (Column 5).
    - Growth prospects proxy (difference in WEO vintages) not significant (Column 6).
    - Total debt percent of GDP not significantly correlated with grant element (Column 7).
    - Results robust when using values at beginning of 5-year period (Columns 8–9).
  - Average non-concessional borrowing (as percent of GDP):
    - After controlling for propensity score, no significant difference between DLP and control group (Columns 1–2).
    - GNI per capita positively and significantly correlated with non-concessional borrowing (Column 3): richer LICs borrow more on non-concessional terms.
    - Infrastructure gap significance (telephone lines per 100 people) reinforces that higher development attracts non-concessional lending (Column 8).
    - No evidence that DLP effect on non-concessional borrowing varies across population members.

### VII. Conclusions and policy implications
- Primary empirical conclusions:
  - Concessional borrowing (as percent of GDP) is significantly higher in countries under DLP — consistent with a catalytic effect of IMF programs in attracting concessional official financing or reflecting correlated determinants of both IMF programs and donor allocations.
  - No evidence that DLP significantly impacts:
    - Level of non-concessional borrowing.
    - Terms of new borrowing (interest rate, grace period, maturity, grant element).
    - Private creditor lending.
  - Heterogeneity analysis indicates non-concessional borrowing increases with country development (GNI per capita), not because of absence of DLP constraints.
- Policy implications emphasized by authors:
  - Removing concessionality requirements under DLP is unlikely, by itself, to cause a major shift toward non-concessional financing in the poorest LICs.
  - The 2009 reform that tailors debt limits to country circumstances (capacity and debt vulnerability) could change DLP impact; limited post-2009 coverage in sample — warrants future work.
  - Further research suggested:
    - Examine debt dynamics since the 2009 reform to test whether DLP impact on non-concessional borrowing has increased.
    - Investigate potential overuse of concessionality requirements in LIC programs that could dilute impact or mask interaction with fiscal conditionality; study granular interactions of debt limits and fiscal conditionality.

*Source: _wp14176 - 3. Auxiliary Tables (PDF), IMF working paper content provided.*

### Appendix 1. Figures

### Appendix 1. Figures

### Figures: PPG External Debt and Components
- Figure 1. PPG External Debt Disbursement (Percent of GDP)
  - Sources: World Development Indicators (WDI); and authors’ calculations.
- Figure 2.1. Concessional Debt Disbursement (Percent of GDP)
  - Sources: World Development Indicators (WDI); and authors’ calculations.
- Figure 2.2. Non Concessional Debt Disbursement (Percent of GDP)
  - Sources: World Development Indicators (WDI); and authors’ calculations.
- Charts include series labeled:
  - Concessional
  - Non Concessional
  - DLP
  - Non DLP
- Time periods shown on figures (categorical): 1986-91, 1992-96, 1997-01, 2002-06, 2007-11.

### Figures: Average Financial Terms (Figure 3)
- Sources: World Development Indicators (WDI); and authors’ calculations.
- Panels and measures shown:
  - Average Grant Element (5% discount rate), In percent: series for Non DLP and DLP across periods labeled 1–5.
  - Average Maturity (Years): series for Non DLP and DLP across periods labeled 1–5.
  - Average Grace Period (Years): series for DLP and Non DLP across periods labeled 1–5.
  - Average Interest (Percent): series for Non DLP and DLP across periods labeled 1–5.

---

### Appendix 2. Tables

### Table 1. List of Countries
- Country list with Region and IDA/HIPC indicators (values 0 or 1). Sample entries:
  - Afghanistan, I. S. of — South Asia — 1 1
  - Armenia — Europe & Central Asia — 0 0
  - Bangladesh — South Asia — 1 0
  - ... (full country list as presented)
- Final entries include:
  - Yemen, Republic Of — Middle East & North Africa — 1 0
  - Zambia — Sub-Saharan Africa — 1 1
  - Zimbabwe — Sub-Saharan Africa — 0 0

### Table 2. Description of Variables
- Size of borrowing (sources WDI):
  - COMMEXP_GDP — Commitments PPG (as % of GDP)
  - COMMIDA_GDP — Commitments PPG IDA (as % of GDP)
  - COMMPR_GDP — Commitments PPG Private Creditors (as % of GDP)
  - DPPG_GDP — Disbursements PPG (as % of GDP)
  - DPPGCON_GDP — Disbursements PPG concessional external debt (PPG) (as % of GDP)
  - DPPGNOCON_GDP — Disbursements PPG non concessional external debt(PPG) (as % of GDP)
  - DPPGIDA_GDP — Disbursements PPG IDA (as % of GDP)
  - DPPGBIL_GDP — Disbursements PPG bilateral (as % of GDP)
  - DPPGMUL_GDP — Disbursements PPG multilateral (as % of GDP)
  - DPRV_GDP — Disbursements PPG Private Creditors (as % of GDP)
- Term of borrowing (sources WDI / author calculations):
  - GRCPERCO — Average grace period on new external debt commitments (years)
  - GRCELCO — Average grant element on new external debt commitments (%)
  - GRCELCO_5 — Average grant element on new external debt commitments' based on 5% discount rate
  - GRCELC~10 — Average grant element on new external debt commitments' based on 10% discount rate (Author calculation)
  - INTCOM — Average interest on new external debt commitments (%)
  - MTR — Average maturity on new external debt commitments' private (years)
- Dependent variable definitions:
  - DLP — Countries subjected to Debt Limit Policy: dummy = 1 if the country is under IMF program for at least three years in a five years window (Bal Gunduz and others, (2013))
  - DLPNCBP — Country subjected to Debt Limit Policy and/or IDA Non Concessional Borrowing Policy (NCBP): dummy = 1 if under IMF program and/or IDA NCBP for at least three years in a five years window (IMF)
- Geographic and institutional characteristics (sources indicated)
  - politicalglobalization — Globalization index (KOF Institute)
  - landlocked — 1 if landlocked (CEPII)
  - democracy — Dummy = 1 if regime qualifies as democratic (Cheibub, Gandhi, and Vreeland (2010))
  - cpia — World Bank's CPIA Index (IMF)
  - grostar — Trading partner real GDP growth (WEO)
  - quota_gdp — Logarithm of the IMF quota (IMF)
  - SIZE — Logarithm of GDP (constant 2005 PPP) (WDI)
  - UNGA — Voting inline with G7 (Dreher and Sturm (2012))
- Initial macroeconomic buffer:
  - AIDGDPI — Initial aid/GDP (at the beginning of each five-years period) (WDI)
  - RESIN — Initial reserves in months of import (at the beginning of each five-years period) (VE-LIC database)
- Country income and macroeconomic conditions:
  - INFLCPI — Inflation, average consumer prices (annual percent change) (WEO)
  - GNIPC — Logarithm of GNI per capita (current US$) (WDI)
  - resource_rents — Resource Rent as a share of GDP (WDI)
  - CURRACC — Current account deficit (as % of GDP) (WDI)
  - TDPPG_GDP — Total Public Debt PPG (as % of GDP) (WDI)
  - GDPGR — Gross domestic product, constant prices (annual percent change) (WDI)
  - DEF_GDP — General Government fiscal deficit (as % of GDP) (WEO)
- Time-variant and averaged variables:
  - IDA — Dummy = 1 if country is IDA only (time variant) BEGINNING OF 5 YEARS PERIOD (IMF)
  - i_TLPPLP — Telephone lines per 100 people BEGINNING OF 5 YEARS PERIOD (WDI)
  - PROJGR — Growth rate of GDP projection BEGINNING OF 5 YEARS PERIOD (WEO)
  - av5_i_TLPPLP — Telephone lines per 100 people (five years average) (WDI)
  - av5_PROJGR — Growth rate of GDP projection (five years average) (WEO)
  - av5_TDPPG_GDP — Total Public Debt PPG (as % of GDP) - five years average (WDI)
- Outcome variables, selection model, heterogeneity analysis: noted as in the table headings.

### Table 3. Summary Statistics (All outcome variables are 5 years period average)
- Sample statistics (Mean, Std. Dev., Min, Max, Obs., Countries) for key variables (selected entries):
  - Commitments PPG (as % of GDP): Mean 5.319, Std. Dev. 3.866, Min 0.000, Max 25.926, Obs. 319, Countries 68
  - Commitments PPG IDA (as % of GDP): Mean 1.513, Std. Dev. 1.512, Min 0.000, Max 7.195, Obs. 296, Countries 68
  - Disbursements PPG (as % of GDP): Mean 4.279, Std. Dev. 2.693, Min 0.040, Max 13.066, Obs. 323, Countries 69
  - Disbursements PPG concessional external debt (PPG) (as % of GDP): Mean 3.324, Std. Dev. 2.230, Min 0.040, Max 10.924, Obs. 323, Countries 69
  - Disbursements PPG non concessional external debt(PPG) (as % of GDP): Mean 0.923, Std. Dev. 1.360, Min 0.000, Max 8.050, Obs. 323, Countries 69
  - Average grace period on new external debt commitments (years): Mean 7.654, Std. Dev. 1.799, Min 2.335, Max 10.892, Obs. 324, Countries 70
  - Average grant element on new external debt commitments (%): Mean 60.672, Std. Dev. 14.482, Min 17.040, Max 81.000, Obs. 324, Countries 70
  - Average grant element on new external debt commitments' based on 5% discount rate: Mean 36.855, Std. Dev. 15.286, Min -9.809, Max 60.800, Obs. 310, Countries 67
  - Average interest on new external debt commitments (%): Mean 1.991, Std. Dev. 1.317, Min 0.150, Max 7.156, Obs. 321, Countries 70
  - Average maturity on new external debt commitments' private (years): Mean 29.404, Std. Dev. 7.126, Min 10.070, Max 42.285, Obs. 325, Countries 70
- Selected macro variables (Mean, Std. Dev., Min, Max, Obs., Countries):
  - Political globalization: Mean 44.605, Std. Dev. 16.895, Min 13.804, Max 84.551, Obs. 323, Countries 68
  - Landlocked: Mean 0.353, Std. Dev. 0.479, Min 0.000, Max 1.000, Obs. 340, Countries 68
  - Democracy: Mean 0.342, Std. Dev. 0.475, Min 0.000, Max 1.000, Obs. 342, Countries 70
  - Trading partner real GDP growth: Mean 4.085, Std. Dev. 1.849, Min -0.655, Max 9.468, Obs. 337, Countries 68
  - Logarithm of GNI per capita (current US$): Mean 6.236, Std. Dev. 0.784, Min 4.868, Max 8.470, Obs. 307, Countries 69
  - Total Public Debt PPG (as % of GDP): Mean 88.051, Std. Dev. 68.558, Min 17.311, Max 492.711, Obs. 295, Countries 67
  - IDA: Mean 0.797, Std. Dev. 0.403, Min 0.000, Max 1.000, Obs. 350, Countries 70

### Table 4. Summary Statistics by Group (DLP = 0 vs DLP = 1)
- Outcome variable and covariate means and dispersion presented separately for DLP = 0 and DLP = 1 across the same variable set as Table 3.
- Selected comparisons (means):
  - Commitments PPG (as % of GDP): DLP = 0 mean 5.027, DLP = 1 mean 5.779
  - Disbursements PPG (as % of GDP): DLP = 0 mean 4.125, DLP = 1 mean 4.532
  - Disbursements PPG non concessional external debt (as % of GDP): DLP = 0 mean 1.092, DLP = 1 mean 0.702
  - Average grant element (5% discount rate): DLP = 0 mean 33.136, DLP = 1 mean 42.358
  - Average grace period (years): DLP = 0 mean 7.362, DLP = 1 mean 8.099
  - Total Public Debt PPG (as % of GDP) - 5 years average: DLP = 0 mean 93.200, DLP = 1 mean 81.349

---

### Correlations and Selection Model

### Table 5. Correlation of Variables (selected notes)
- Correlation table presented with significance notation: * p<0.05.
- Highlighted significant correlations include:
  - GNIPC correlated negatively with several variables (noted with * p<0.05).
  - SIZE shows multiple significant correlations (noted with * p<0.05).
  - Resource rents, CPIA, and TDPPG_GDP show significant correlations in the matrix.

### Table 6. Selection Model (Dependent Variable: IMF's Debt Limit Policy)
- Two model specifications:
  - Column (1): IMF's Debt Limit Policy
  - Column (2): IMF's Debt Limit Policy or IDA Non-Concessional Borrowing Policy
- Selected coefficients (standard errors in parentheses):
  - landlocked: 0.512** (0.228) in (1); 0.430* (0.231) in (2)
  - Aid-to-GDP: 0.024 (0.016) in (1); 0.028* (0.016) in (2)
  - Initial Reserves: -0.113** (0.050) in (1); -0.135** (0.052) in (2)
  - 3.period dummy: 0.669** (0.337) in (1); 0.697** (0.339) in (2)
  - 4.period dummy: 0.641* (0.358) in (1); 0.723** (0.361) in (2)
  - 5.period dummy: 1.274*** (0.398) in (1); 1.518*** (0.409) in (2)
  - CPI Inflation: -0.018** (0.008) in (1); -0.020** (0.009) in (2)
  - GNI per capita: -0.510*** (0.197) in (1); -0.555*** (0.202) in (2)
  - Resource rents: -0.037** (0.015) in both models
  - CPIA: 0.538*** (0.194) in (1); 0.560*** (0.197) in (2)
- Observations: 226; Pseudo R(squared): 0.248 in (1), 0.271 in (2)
- Significance: *** p<0.01, ** p<0.05, * p<0.1

---

### Propensity Score Matching and Treatment Effects

### Table 7. Propensity Score Matching (Radius, Kernel, Matching)
- Reported average treatment effects for DLP on size of borrowing and terms of borrowing, with bootstrapped standard errors (500 replications).
- Selected estimated effects (kernel r=0.05; r=0.1; radius matching):
  - Disbursements PPG (as % of GDP): 0.77** (0.387); 0.60 (0.386); 0.79** (0.389)
  - Disbursements PPG concessional external debt (as % of GDP): 0.72** (0.33); 0.59* (0.338); 0.71** (0.33)
  - Disbursements PPG non-concessional external debt (as % of GDP): 0.05 (0.161); 0.00 (0.161); 0.07 (0.159)
  - Disbursements PPG IDA (as % of GDP): 0.32* (0.189); 0.31* (0.181); 0.31* (0.18)
  - Disbursements PPG bilateral (as % of GDP): 0.21* (0.128); 0.17 (0.119); 0.23* (0.125)
  - Disbursements PPG multilateral (as % of GDP): 0.60* (0.321); 0.48 (0.332); 0.59** (0.305)
  - Commitments PPG (as % of GDP): 1.56*** (0.536); 1.39*** (0.55); 1.60*** (0.531)
  - Commitments PPG IDA (as % of GDP): 0.52** (0.256); 0.47* (0.262); 0.51** (0.252)
  - Term of borrowing results: coefficients reported but not statistically significant for grace period, grant element, interest, maturity in many specifications (see table for exact estimates and standard errors).
- Notes: Epanechnikov kernel used for kernel regression matching. Bootstrapped standard errors based on 500 replications.

---

### Heterogeneity Analysis: Determinants of Grant Element and Non-Concessional Borrowing

### Table 8. Heterogeneity Analysis - Determinants of Grant Element (Dependent Variable: Average Grant Element (with 5% discount rate))
- Selected coefficients (robust standard errors in parentheses):
  - Propensity score: 21.665*** (4.704) in (1); coefficients significant across specifications.
  - Debt Limit Policy Dummy (DLP): 7.643*** (2.404) in (1); varies across specifications (not always significant).
  - Logarithm of GNI per capita: -8.387*** (1.844) in models where included.
  - Telephone lines per 100 people - five years average: -0.488* (0.265) in one specification.
  - Interaction terms (DLP x covariates) included in some specs (e.g., DLP x Logarithm of GNI per capita: 1.718 (2.112) not significant).
- Observations range from 176 to 190 across specifications.
- R-squared ranges from 0.07 to 0.314 across columns.
- Robust standard errors reported. Significance: *** p<0.01, ** p<0.05, * p<0.1

### Table 9. Heterogeneity Analysis - Determinants of Non-Concessional Borrowing (Dependent Variable: Average Non-concessional Borrowing (as % of GDP))
- Selected coefficients (robust standard errors in parentheses):
  - Propensity score: -1.975*** (0.519) in (1) and significant in multiple specifications.
  - Debt Limit Policy Dummy (DLP): -0.407* (0.221) in (1); other specifications show varied DLP estimates (some positive, some not significant).
  - Logarithm of GNI per capita: 0.578** (0.249) in one specification.
  - Telephone lines per 100 people - beginning of 5 years period: 0.067** (0.026) in one specification.
  - Interaction terms (DLP x covariates) included; most interactions not statistically significant.
- Observations range from 176 to 190 across specifications.
- R-squared values range from 0.025 to 0.211 across columns.
- Robust standard errors reported. Significance: *** p<0.01, ** p<0.05, * p<0.1

---

*Italic: Source — _wp14176 - Appendix 1. Figures*

### Appendix 3. Auxiliary Tables

### Appendix 3. Auxiliary Tables

### Selection model — alternative specifications (Table A1)
- Dependent Variable: Debt Limit Policy; columns (1)–(11) report alternative specifications.
- Key estimated coefficients (coefficient (standard error)) and significance:
  - landlocked: 0.577** (0.240); 0.536** (0.241); 0.486** (0.233); 0.469** (0.233); 0.531** (0.233); 0.497** (0.247); 0.477** (0.235); 0.758*** (0.199); 0.652** (0.272); 0.469** (0.232); 0.439** (0.217)
  - politicalglobalization: -0.0020 (0.008); -0.002 (0.008); 0.001 (0.008); -0.004 (0.008); -0.003 (0.008); -0.0010 (0.009); 0.005 (0.008); 0.002 (0.008); -0.005 (0.01); (0.009)
  - democracy: 0.087 (0.238); 0.064 (0.239); 0.155 (0.221); 0.118 (0.219); 0.167 (0.221); 0.064 (0.233); 0.090 (0.228); 0.068 (0.221); 0.167 (0.269); 0.171 (0.224)
  - growth of trading partners: 0.030 (0.066); 0.028 (0.065); 0.036 (0.062); 0.006 (0.063); 0.041 (0.063); 0.058 (0.067); 0.034 (0.062); 0.011 (0.058); 0.006 (0.079); 0.034 (0.064)
  - IMF quota: 0.041 (0.052); 0.049 (0.053); 0.038 (0.051); 0.043 (0.052); 0.040 (0.052); 0.047 (0.056); 0.042 (0.052); 0.021 (0.049); -0.003 (0.074); 0.061 (0.054)
  - Aid-to-GDP: 0.025 (0.016); 0.023 (0.016); 0.022 (0.016); 0.018 (0.016); 0.024 (0.017); 0.026 (0.018); 0.023 (0.016); 0.046*** (0.013); 0.049** (0.023); 0.022 (0.016)
  - Initial Reserves: -0.111* (0.058); -0.103* (0.059); -0.122** (0.052); -0.101* (0.052); -0.120** (0.052); -0.091* (0.053); -0.110** (0.050); -0.072 (0.047); -0.082 (0.058); -0.119** (0.052); -0.105** (0.048)
  - 2.period dummy: -0.293 (0.366); -0.253 (0.366); -0.209 (0.358); -0.263 (0.361); -0.26 (0.362); -0.191 (0.358); -0.413 (0.339); -0.985* (0.593); -0.219 (0.359); -0.237 (0.345)
  - 3.period dummy: 0.631* (0.349); 0.619* (0.350); 0.682** (0.339); 0.549 (0.347); 0.638* (0.342); 0.920*** (0.342); 0.672** (0.336); 0.386 (0.310); 0.187 (0.532); 0.650* (0.35); 0.579* (0.332)
  - 4.period dummy: 0.758** (0.374); 0.801** (0.377); 0.656* (0.359); 0.455 (0.381); 0.612* (0.364); 0.858** (0.341); 0.636* (0.357); 0.460 (0.335); 0.302 (0.536); 0.698* (0.375); 0.552* (0.333)
  - 5.period dummy: 1.176*** (0.433); 1.178*** (0.430); 1.238*** (0.406); 1.090*** (0.414); 1.198*** (0.409); 1.510*** (0.418); 1.327*** (0.408); 0.850** (0.352); 0.809 (0.581); 1.320*** (0.418); 0.979*** (0.341)
  - CPI Inflation: -0.019 (0.011); -0.015 (0.011); -0.017** (0.009); -0.016* (0.009); -0.018** (0.009); -0.018** (0.009); -0.018** (0.008); -0.016** (0.007); 0.007 (0.012); -0.019** (0.009); -0.012 (0.008)
  - GNI per capita: -0.587*** (0.212); -0.539** (0.212); -0.564*** (0.203); -0.526*** (0.201); -0.514** (0.203); -0.434** (0.215); -0.575*** (0.222); -0.462* (0.24); -0.539*** (0.205); -0.668*** (0.148)
  - Resource rents: -0.037** (0.018); -0.040** (0.019); -0.038** (0.016); -0.035** (0.015); -0.038** (0.016); -0.040** (0.017); -0.039** (0.015); -0.039*** (0.015); -0.031* (0.017); -0.041** (0.016); -0.034*** (0.013)
  - Size: 0.182 (0.137); 0.142 (0.136); 0.124 (0.132); 0.083 (0.132); 0.147 (0.132); 0.208 (0.148); 0.138 (0.131); 0.276** (0.113); 0.087 (0.165); 0.183 (0.136)
  - CPIA: 0.543** (0.211); 0.575*** (0.212); 0.554*** (0.197); 0.495** (0.199); 0.550*** (0.198); 0.523** (0.245); 0.519*** (0.196); 0.653** (0.288); 0.538*** (0.199); 0.586*** (0.178)
- Additional single-variable rows reported in table:
  - Current acct. deficit (%of GDP)_5year-average: -0.026 (0.020)
  - Total Public Debt (% of GDP)_5-year average: 0.000 (0.003)
  - GDP Growth_5-year average: 0.091** (0.042); 0.079** (0.038)
  - Current account deficit (% of GDP): -0.009 (0.016)
  - Total Public Debt (% of GDP): 0.001 (0.002)
  - GDP Growth: -0.024 (0.026)
  - IDA: -0.226 (0.357); 0.056 (0.283)
  - Government fiscal deficit (% of GDP): 2.507 (3.509)
  - UNGA: 0.292 (0.88)
- Constant terms (selected columns): -2.639 (3.724); -2.202 (3.697); -1.399 (3.555); -0.891 (3.502); -2.225 (3.563); -4.229 (3.876); -1.421 (3.702); -6.932*** (2.408); -1.686 (4.311); -2.818 (3.581); 1.998* (1,058)
- Sample sizes and fit:
  - Observations / N: 210; 204; 223; 223; 221; 195; 226; 232; 162; 221; 236
  - Pseudo R(squared): 0.267; 0.255; 0.246; 0.267; 0.244; 0.238; 0.249; 0.202; 0.279; 0.259; 0.249
- Notes: Standard errors in parentheses. Significance legend: *** p<0.01, ** p<0.05, * p<0.1.

### Propensity score matching — kernel and radius matching (Table A2)
- Matching estimators use an Epanechnikov kernel; bootstrapped standard errors (500 replications) reported in parentheses.
- Kernel matching (columns: kernel; r=0.05; r=0.1; matching) — estimated average treatment effects (coefficient (standard error)):
  - Size of borrowing
    - Disbursements PPG (as % of GDP): 0.621 (0.424); 0.754* (0.392); 0.552 (0.414)
    - Disbursements PPG concessional external debt (as % of GDP): 0.602* (0.354); 0.715** (0.343); 0.566 (0.348)
    - Disbursements PPG non-concessional external debt (as % of GDP): 0.019 (0.186); 0.038 (0.17); -0.014 (0.191)
    - Disbursements PPG IDA (as % of GDP): 0.286 (0.197); 0.426** (0.183); 0.253 (0.182)
    - Disbursements PPG bilateral (as % of GDP): 0.083 (0.135); 0.079 (0.125); 0.079 (0.128)
    - Disbursements PPG multilateral (as % of GDP): 0.589* (0.321); 0.723** (0.294); 0.544 (0.334)
    - Disbursements PPG Private Creditors (as % of GDP): 0.047 (0.119); -0.045 (0.104); -0.066 (0.121)
    - Commitments PPG (as % of GDP): 1.085** (0.528); 1.267** (0.529); 0.974* (0.541)
    - Commitments PPG IDA (as % of GDP): 0.235 (0.287); 0.471* (0.254); 0.170 (0.32)
    - Commitments PPG Private Creditors (as % of GDP): 0.060 (0.148); 0.074 (0.124); 0.044 (0.16)
  - Term of borrowing (selected):
    - Average grace period on new external debt commitments (years): 0.000 (0.236); 0.173 (0.22); -0.004 (0.263)
    - Average grant element on new external debt commitments (%): 0.365 (1.585); 1.142 (1.584); 0.450 (1.683)
    - Average grant element on new external debt commitments (%) _10% discount rate: 0.940 (1.627); 1.401 (1.472); 1.031 (1.674)
    - Average grant element on new external debt commitments (%) _5% discount rate: 1.613 (2.045); 2.078 (1.848); 1.708 (1.975)
    - Average interest on new external debt commitments (%): -0.142 (0.168); -0.184 (0.138); -0.146 (0.166)
    - Average maturity on new external debt commitments' (years): 0.457 (0.903); 0.925 (0.895); 0.520 (0.888)

### Propensity score matching — parsimonious participation equation and change in outcomes (Table A3)
- Matching estimators using kernel and radius matching; Epanechnikov kernel; bootstrapped standard errors (500 replications).
- Kernel / r=0.05 / r=0.1 / matching — coefficients (standard errors):
  - Size of borrowing (change of outcome variables):
    - Disbursements PPG (as % of GDP): -0.090 (0.178); -0.130 (0.207); -0.053 (0.157)
    - Disbursements PPG concessional external debt (as % of GDP): -0.064 (0.177); -0.106 (0.197); -0.027 (0.147)
    - Disbursements PPG non-concessional external debt (as % of GDP): 1.981 (2.337); 2.154 (2.113); 2.052 (2.214)
    - Disbursements PPG IDA (as % of GDP): 0.177 (0.189); 0.128 (0.192); 0.183 (0.184)
    - Disbursements PPG bilateral (as % of GDP): 1558.547 (1559.996); 1558.598 (1566.375); 1558.588 (1621.875)
    - Disbursements PPG multilateral (as % of GDP): 0.192 (0.179); 0.188 (0.178); 0.203 (0.173)
    - Disbursements PPG Private Creditors (as % of GDP): -0.482 (1.048); -0.714 (1.146); -0.358 (0.893)
    - Commitments PPG (as % of GDP): 0.315*** (0.101); 0.309*** (0.104); 0.316*** (0.114)
    - Commitments PPG IDA (as % of GDP): 0.235** (0.111); 0.170 (0.127); 0.234** (0.117)
    - Commitments PPG Private Creditors (as % of GDP): 2.497 (3.194); 2.506 (2.841); 2.537 (3.217)
- Notes: Epanechnikov kernel; bootstrapped standard errors based on 500 replications.

### Rosenbaum sensitivity analysis for hidden selection bias (Table A4)
- Γ parameter (measure of how much hidden bias can be present) and associated probabilities:
  - Disbursements PPG (as % of GDP): Γ = 1.3; Probability = 0.038
  - Disbursements PPG concessional external debt (as % of GDP): Γ = 1.4; Probability = 0.042
  - Disbursements PPG IDA (as % of GDP): Γ = 1.2; Probability = 0.035
  - Disbursements PPG bilateral (as % of GDP): Γ = 1.0; Probability = 0.186
  - Disbursements PPG multilateral (as % of GDP): Γ = 1.3; Probability = 0.032
  - Commitments PPG (as % of GDP): Γ = 1.7; Probability = 0.032
  - Commitments PPG IDA (as % of GDP): Γ = 1.2; Probability = 0.032
- Note: Γ is a measure of how much hidden bias can be present, i.e., how much Γ can deviate from 1, before the results of the study begin to change.

*Source: IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14176.pdf_