## wpiea2021229-print-pdf

## Source details

**Canonical URL:** [wpiea2021229-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf.json)

---

### I. Introduction, objectives, and dataset
- Fiscal context:
  - public debt rose by close to 10 percentage points of GDP in emerging market economies (EMEs) and by 5 percent in low-income developing countries (LIDCs) in the space of one year amid the pandemic.
  - current debt levels in EMDEs are now some 20–25 percentage points of GDP higher than before the global financial crisis.
- Research questions:
  - Do low interest-growth differentials (r − g) matter for debt dynamics in EMDEs?
  - How important is r − g across phases of fiscal crises?
  - Are other factors more important in explaining debt dynamics?
- Dataset:
  - unbalanced panel of 150 EMDEs for 1971–2018: 92 EMEs and 58 LIDCs.
  - dataset supplemented with external debt composition by creditor type, external debt guarantees, and an index of financial account liberalization.

### II. Measurement and stylized facts on interest-growth differential (r − g)
- Measurement:
  - r − g in period t measured as (i_t − γ_t) / (1 + γ_t), where i_t = effective interest rate on debt (interest bill in year t divided by average debt stock in year t and t−1) and γ_t = nominal GDP growth rate.
  - valuation effects from exchange rate fluctuations are not accounted for in i_t and may underestimate r − g in stress periods.
- Stylized facts:
  - average r − g over full sample: -8½ percent in EMEs and -9½ in LIDCs.
  - about 80 percent of EMEs and 90 percent of LIDCs had negative r − g over the sample.
  - notable exceptions: 2009 positive r − g in EMEs; 2015 uptick in EMEs; WEO estimates indicate r − g turned positive in 2020 (outside core sample).
  - over last fifty years average r − g increased by about 3½ percentage points in EMEs and by 1 percentage point in LIDCs.
  - within-country standard deviation of r − g ~10 percentage points on average for both groups.
  - persistence (AR(1)) low, ranging from 0.3 to 0.4 on average.

### III. Determinants of r − g: empirical patterns and regression results
- Dominant drivers:
  - correlation between real interest rates and r − g ≈ 0.9 for both country groups; correlation between growth and r − g ≈ 0.1 for full sample.
  - real interest rates are the dominant explanatory variable for r − g; real growth has much smaller explanatory value.
- Financial openness and creditor composition:
  - financial openness improved more markedly in EMEs than in LIDCs.
  - official sector share of external public debt: EMEs = 56 percent; LIDCs = 85 percent.
  - share of concessional borrowing in external public debt (2018): EMEs = 13 percent; LIDCs = 33 percent.
  - empirical model finds:
    - Financial openness positively associated with r − g only for EMEs.
    - Higher shares of official and concessional debt associated with lower r − g in EMEs (impact of concessional debt much higher).
    - For LIDCs, changes in public debt (increases) are more important drivers of higher r − g.
- Selected regression coefficients (fixed effects; significance levels preserved):
  - Lagged Real Effective Interest Rate: EMEs = 0.241*** (0.062) in Column (1); LIDCs = 0.167*** (0.049) in Column (1).
  - Lagged Real GDP Growth: EMEs = -0.179*** (0.055) in Column (1); LIDCs = 0.037 (0.068) in Column (1).
  - Financial Openness Index: EMEs = 5.404*** (1.270) in Column (1); LIDCs = 0.813 (2.347) in Column (1).
  - Change in Public Debt: EMEs = 0.178*** (0.062) in Column (1); LIDCs = 0.107*** (0.024) in Column (1).
  - Lagged VIX: EMEs = 0.194*** (0.023); LIDCs = 0.089*** (0.021).
  - Lagged Exchange Rate Depreciation: EMEs = -0.074*** (0.009); LIDCs = -0.057*** (0.013).
  - Share Official Debt (selected): EMEs = -0.064** (0.025).
  - Share Concessional Debt (selected): EMEs = -0.088* (0.049).
  - Example model fit (Column (1)): Observations EMEs = 1,670; LIDCs = 1,304; R-squared = 0.382 (EMEs), 0.259 (LIDCs); Number of countries = 66 (EMEs), 52 (LIDCs).

### IV. Debt dynamics: decomposition and average contributions
- Debt decomposition identity:
  - change in debt-to-GDP = cumulative lagged debt × (i_t − γ_t)/(1+γ_t) + cumulative primary deficit p_t + cumulative stock-flow adjustment sfa_t.
- Average contributions (full sample; percent of GDP, annual averages):
  - r − g contributed to reducing debt:
    - EMEs: 2¾ percentage points of GDP per year.
    - LIDCs: about 5½ percentage points of GDP per year.
  - Since 2010, increases in primary deficits, less favorable r − g, and positive stock-flow adjustments contributed to average debt increases in the range of 1.2 to 1.4 percent of GDP per annum.
- Table highlights (selected numbers from decomposition):
  - Full Sample (1970-2018): Change in Debt = -0.21; Primary Deficit = 0.69; r − g = -3.74; SFA = -5.73; (final component) = 2.85.
  - EMEs (full sample): Change in Debt = 0.15; Primary Deficit = 0.41; r − g = -2.68; SFA = -5.03; (final component) = 2.42.
  - LIDCs (full sample): Change in Debt = -0.83; Primary Deficit = 1.18; r − g = -5.61; SFA = -6.97; (final component) = 3.60.
- Tail events and extreme debt increases:
  - 75th percentile annual debt increases:
    - EMEs: 7½ percentage points of GDP.
    - LIDCs: 10¾ percentage points of GDP.
  - In these extreme events, stock-flow adjustments dominate:
    - Stock-flow adjustments: EMEs = 7 percent of GDP; LIDCs = 11¾ percent of GDP.
  - r − g contribution is less than half the SFA contribution in these spikes.

### V. Fiscal crises: incidence, dynamics, and phases
- Incidence and duration (1970–2018):
  - Number of crisis starts: EMEs = 220; LIDCs = 214.
  - Average duration: EMEs = 5.3 years; LIDCs = 6.1 years.
  - Majority of episodes associated with credit events: EMEs = 59.6 percent; LIDCs = 82.0 percent.
  - Persistence: EMEs = 17.7 percent; LIDCs = 29.4 percent.
  - Long haulers (percent of crises): EMEs = 33.6 percent; LIDCs = 35.5 percent.
- Event-study (7-year windows, t=0 crisis start):
  - Primary balances:
    - EMEs: above normal before crisis; deteriorate to a trough about 1 percentage point of GDP below normal at crisis hit; rebound to normal within three years after crisis start.
    - LIDCs: primary balance relatively flat around crises.
  - r − g: no statistically significant differences around crisis episodes for both EMEs and LIDCs.
  - Debt service-to-revenue (EMEs): peaks 4¼ percentage points above normal two years after crisis start.
  - Stock-flow adjustments: upward trend in run-up to crisis in EMEs and at onset in LIDCs.
- Entry vs exit and duration effects (definitions: entry = three years before start to peak; exit = peak to end):
  - Entry phase cumulative contributions:
    - EMEs: primary deficits cumulative effect = 19 percent of GDP; stock-flow adjustments material but smaller than in LIDCs; r − g large but insufficient to offset other factors.
    - LIDCs: stock-flow adjustments cumulative impact ≈ 18½ percent of GDP; r − g large and in LIDCs enough to offset negative effects during entry.
  - Exit phase:
    - All components significantly smaller; key drivers of exit are reversal of primary deficits (turning into surpluses) and reduction in stock-flow adjustments; some normalization of interest rates helps.
  - Long-haulers vs short-haulers:
    - Entry-phase debt accumulation much larger in long-haulers.
    - Long-haulers: EMEs driven more by primary deficits; LIDCs driven more by SFA.
    - Short-haulers: SFA is the overriding force in entry for both groups.

### VI. Stock-flow adjustments (SFA): composition, drivers, and regressions
- Decomposition of SFA:
  - sfa_t = ε_t (1 − α_t) (1 + i_t) d_{t−1} + bbt_t
    - first term: valuation effect from nominal exchange rate depreciation ε_t and share of domestic-currency debt α_t.
    - bbt_t: residual capturing below-the-line operations (bank recapitalizations, bailouts, called guarantees) and errors and omissions.
- Relative importance during crisis entry (percent of GDP):
  - Valuation effects:
    - EMEs: about 8 percent of GDP of the debt increase.
    - LIDCs: about 10¾ percent of GDP of the debt increase.
  - Below-the-line operations: even larger in EMEs and non-negligible in LIDCs.
- Regression evidence (fixed-effect panel; selected findings):
  - Inflation and exchange rate depreciation positively associated with SFA for both groups.
  - Impact of exchange rate depreciation increases with the share of external debt.
  - Private debt has statistically significant positive effect only in EMEs (when not interacting with crisis status).
  - Changes in private guaranteed external debt positively associated with SFA; association during fiscal crisis matters only for EMEs.
  - Other public sector external debt not significant.
  - Selected coefficients (examples):
    - Lag External public debt * depreciation: EMEs = 0.004*** (0.001); LIDCs = 0.001* (0.001).
    - 5y change of private external guaranteed debt: 1.352*** (0.311) in one specification.
    - Lag private debt * fiscal crisis: negative and significant in some specifications (e.g., -0.037** (0.016)).
  - Interpretation: valuation effects are robust determinants of SFA; below-the-line operations particularly important for EMEs and associated with contingent liabilities from the private sector.

### VII. Conclusions and policy implications
- Empirical summary:
  - Both EMEs and LIDCs have benefited historically from highly negative r − g, but r − g has displayed an upward trend over recent decades (more so in EMEs), with high volatility and low persistence.
  - Low r − g has played a debt-stabilizing role over five decades (more in LIDCs than EMEs) but has been insufficient to prevent recent accumulation of debt driven by large primary deficits and stock-flow adjustments.
  - In fiscal crises, r − g plays a stabilizing role but primary deficits and stock-flow adjustments are the critical determinants of debt run-ups and persistence.
- Two policy takeaways:
  - To the extent that foreign currency exposures are large, inflation and exchange rate stability are paramount.
  - Greater attention should be given to hidden debts because materialization of contingent liabilities is often what gets EMDEs into trouble.

*Source: wpiea2021229-print-pdf - APPENDIX TABLES (IMF).*

### REFERENCES _____________________________________________________________________________________________25

### REFERENCES _____________________________________________________________________________________________25

### FIGURES
- 1. Debt and Interest Expense in Emerging and Developing Economies ______________________________ 5
- 2. Interest-Growth Differentials in Emerging and Developing Economies___________________________ 11
- 3. Share of Countries with Negative Interest-Growth Differentials___________________________________ 11
- 4. Persistence and Volatility of Interest-Growth Dif ferentials _________________________________________ 12
- 5. Real Interest Rates in Emerging and Developing Economies ______________________________________ 13
- 6. Real Growth in Emerging and Developing Economies _____________________________________________ 13
- 7. Financial Openness in Emerging and Developing Economies _____________________________________ 14
- 8. Debt Decomposition in Emerging and Developing Economies ___________________________________ 17
- 9. Crisis Incidence in Emerging and Developing Economies__________________________________________ 18
- 10. Macroeconomic Dynamics during Crisis Episodes ________________________________________________ 19
- 11. Debt Decomposition in EMDEs around Fiscal Crisis Episodes ____________________________________ 21
- 12. Debt Dynamics around Fiscal Crisis by Duration __________________________________________________ 21
- 13. Debt Decomposition in EMEs around Fiscal Crisis Episodes______________________________________ 23
- A.1.1. Country Coverage: Interest-Growth Differential ________________________________________________ 35

### TABLES
- 1. Correlation between Real Interest Rate, Real Growth, and Real R-G ______________________________ 29
- 2. Determinants of the Interest-Growth Dif ferential___________________________________________________ 30
- 3. Debt Decomposition by Income Groups and Decades _____________________________________________ 31
- 4. Fiscal Crises in EMDEs, 1970–2018 ___________________________________________________________________ 32
- 5.    Determinants of Stock-Flow Adjustment ____________________________________________________________ 33

### APPENDIX
- APPENDIX. DATA: DEFINITIONS AND SOURCES ___________________________________________________34

*https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf*

### APPENDIX TABLES

### APPENDIX TABLES

### I. Introduction — context and research questions
- Fiscal deficits in emerging and developing economies (EMDEs) have ballooned amid the pandemic:
  - public debt rose by close to 10 percentage points of GDP in emerging market economies (EMEs) and by 5 percent in low-income developing countries (LIDCs) in the space of one year.
  - current debt levels in EMDEs are now some 20–25 percentage points of GDP higher than before the global financial crisis.
- Debate framed around interest-growth differentials (r-g):
  - one view: debt stabilizes to pre-pandemic levels over the next five years but remains historically high.
  - counterpoint: historically low interest payments and negative r-g (interest rates below growth) could buoy sustainability.
- Paper’s objective and dataset:
  - examines whether low r-g matters for debt dynamics in EMDEs using a dataset covering 150 EMDEs for 1971–2018.
  - three questions:
    1. empirical regularities on r-g and differences between EMEs and LIDCs;
    2. whether r-g plays a debt-stabilizing role and its importance across phases of fiscal crises;
    3. whether other factors are more important in explaining debt dynamics.

### Key findings (summary)
- Long-run r-g trends and drivers:
  - average r-g has been negative over the last five decades but has shown an upward drift as real interest rates increased—more so in EMEs than LIDCs.
  - empirical model suggests financial liberalization and changing creditor profile (less reliance on official and concessional financing) may partly explain r-g trends in EMEs.
  - in LIDCs, no robust relationship found with those factors; increases in public debt are particularly important for raising r-g.
- Role of r-g in debt dynamics:
  - r-g has played a debt-stabilizing role over five decades, though less so in EMEs than in LIDCs.
  - low r-g has not prevented recent accumulation of debt in EMDEs due to large primary deficits and stock-flow adjustments (valuation effects and materialization of contingent liabilities).
  - during run-up to crises, r-g remains relatively low, but persistent primary deficits lead to abrupt rises in debt service in EMEs and fiscal crises.
- Stock-flow adjustments and crises:
  - accumulation of stock-flow adjustments is a key trademark of fiscal crises in EMDEs, especially in the crisis entry phase.
  - valuation effects from foreign-exchange debt and currency depreciation are the most important drivers of stock-flow adjustments.
  - evidence that materialization of contingent liabilities associated with private sector may play a role in EMEs.
  - exit from crises generally entails primary deficits turning into surpluses and stock-flow adjustments becoming negligible; r-g does not change significantly in and out of crisis.

### II. Related literature — positioning and gaps
- Two strands connected:
  - literature on dynamics of r-g (mostly advanced economies): r-g usually negative; low inflation volatility and accommodative monetary policy important.
  - literature on determinants of interest rates and debt sustainability in EMEs: fiscal indicators correlate with sovereign spreads; composition of debt (foreign currency share) matters; vulnerability to global financial conditions persists.
- Contributions of this paper:
  - first systematic comparison of r-g trends and differences between EMEs and LIDCs using a comprehensive dataset.
  - expands determinants of r-g to include creditor profile and changing financing landscape.
  - analyzes debt dynamics not only in normal times but during crisis episodes, across debt distribution, and distinguishes phases/typologies of fiscal crises.
  - examines nature of below-the-line operations (stock-flow adjustments, valuation effects, contingent liabilities).

### III. Data — scope, coverage, and methodological notes
- Dataset:
  - unbalanced panel of 150 EMDEs for 1971–2018 (updated from Moreno Badia et al. 2020).
  - includes 92 EMEs and 58 LIDCs.
  - data coverage improves over time and accounts for 90 percent of GDP in EMDEs by the mid-1990s.
  - supplemented with data on external debt composition by creditor type, external debt guarantees, and an index of financial account liberalization.
- Methodological issues and pragmatic choices:
  - Valuation effects:
    - currency composition of debt data are patchy; IMF WEO includes foreign-currency denominated debt but time series often short.
    - World Bank IDS reports external general government debt going back to 1970s but:
      - IDS definition based on residency may over-estimate foreign-currency debt if external creditors hold domestic-currency debt.
      - IDS excludes short-term maturities, underestimating foreign claims.
      - IDS perimeter of government may not coincide with Global Debt Database public debt perimeter.
    - Pragmatic approach: for each country choose dataset with better coverage (in two thirds of cases IDS); cap share of public external debt at 100 percent when government perimeter of external debt is broader than public debt (following Panizza and Teddei 2020).
  - Contingent liabilities:
    - databases on materialization of contingent liabilities are scant (e.g., Bova et al. 2019 covers 44 EMDEs for 1990–2014, only one LIDC).
    - proxies used for realization of contingent liabilities:
      - private credit and publicly guaranteed external private debt to capture private sector debt overhang spillovers.
      - external debt of the other public sector (outside the general government) to proxy state-owned enterprises’ liabilities.
    - these proxies target common, large sources of contingent liabilities (recapitalizations and debt assumptions).

### IV. Anatomy of interest-growth differentials (r-g) — measurement and stylized facts
- Measurement:
  - r-g in period t measured as (i_t − γ_t) / (1 + γ_t), where:
    - i_t = effective interest rate on debt = interest bill in year t divided by average debt stock in year t and t-1;
    - γ_t = nominal GDP growth rate.
  - note: due to data constraints, cannot disaggregate interest bill between domestic and foreign currency debt; valuation effects from exchange rate fluctuations are not accounted for in i_t and may underestimate r-g in stress periods.
- Stylized facts:
  - On average, r-g negative for both EMEs and LIDCs since early 1970s except:
    - 2009: EMEs positive r-g due to output contraction during global financial crisis.
    - 2015: EMEs uptick driven by oil producers’ slowdown.
    - World Economic Outlook estimates indicate average r-g also turned positive for both EMEs and LIDCs in 2020 amid COVID-19 output collapse (note: mentioned but outside core sample).
  - Levels:
    - average r-g over full sample: -8½ percent in EMEs and -9½ in LIDCs.
    - gap between groups peaked in 2009 when r-g increased by more than 15 percentage points in EMEs (into positive territory); LIDCs remained well into negative territory.
    - by 2018 average r-g in LIDCs was more than 4 percentage points below EMEs.
  - Trend and synchronization:
    - about 80 percent of EMEs and 90 percent of LIDCs had negative r-g over sample.
    - big swings in mid-1980s, 1990s, early 2000s, and 2009.
    - post-global financial crisis r-g fluctuations became synchronized across country groups: correlation between EMEs and LIDCs is 0.54 for full sample and 0.98 from 2010.
    - over last fifty years average r-g increased by about 3½ percentage points in EMEs and by 1 percentage point in LIDCs.
    - compression in interquartile range from about 11 percentage points pre-global financial crisis to about 6½ thereafter for both groups.
  - Volatility and persistence:
    - within-country standard deviation of r-g ~10 percentage points on average for both groups over the full sample.
    - decline in volatility post-global financial crisis more pronounced in EMEs than LIDCs.
    - persistence (first-order autocorrelation AR(1)) is low, ranging from 0.3 to 0.4 on average.
    - implication: favorable r-g episodes are volatile and not highly persistent, consistent with boom-bust cycles in debt crises.

### V. Determinants of r-g — empirical patterns
- Role of real interest rates and growth:
  - real interest rates have been negative for much of the sample, but inflation has fallen faster than nominal effective interest rates over fifty years, pushing real interest rates up in both groups.
  - increase in real rates more pronounced among EMEs; on average real rates close to zero for EMEs after the global financial crisis.
  - real growth rates marginally higher in EMEs than LIDCs over full sample:
    - EMEs: 3¾ percent average real growth.
    - LIDCs: 3½ percent average real growth.
  - long-term trends in real growth differ:
    - since the 1970s, real growth fell by 3¼ percentage points in EMEs and by about ½ percentage point in LIDCs.
- Correlations and changing drivers:
  - correlation between real interest rates and r-g ≈ 0.9 for both country groups, implying real interest rate is a dominant driver.
  - correlation between growth and r-g about 0.1 for full sample.
  - post-global financial crisis, correlation of real interest rate with r-g reduced while correlation of real growth with r-g increased, especially for EMEs—possibly reflecting reduced financial repression and increasing importance of income convergence.

*Source: wpiea2021229-print-pdf - APPENDIX TABLES (IMF).*

### 1970. Although,  an imperfect measure of financial repression, this index gives an idea of the

### wpiea2021229-print-pdf

### Financial openness, creditor composition, and r-g
- Financial openness has improved more markedly in EMEs than in LIDCs; financial openness index is used as an imperfect measure of capital controls and financial repression.
- Official sector share of external public debt:
  - EMEs: 56 percent.
  - LIDCs: 85 percent.
- Share of concessional borrowing in external public debt (declined by half from its peak):
  - EMEs (2018): 13 percent.
  - LIDCs (2018): 33 percent.
- Interpretation: graduation from financial repression and shifts in debt composition (less official and concessional lending) may have increased borrowing costs and thus raised r-g in EMEs; effect is muted or absent in LIDCs where liberalization has been limited.

### Empirical model and main statistical findings
- Baseline fixed effects specification estimated (r − g)t = β1 FFINOP_it + X_it + α_i + ε_it, where FFINOP is the continuous financial openness indicator (Chinn and Ito 2006); X includes lagged real interest rate, lagged real growth, change in public debt, initial public debt, shares of official and concessional external public debt, expected inflation (lagged), exchange rate depreciation, private debt, commodity prices, VIX index, and a HIPC dummy.
- Key empirical results:
  - Financial openness is positively associated with r − g only for EMEs.
  - For EMEs, higher shares of official and concessional debt are associated with lower r − g; the impact of concessional debt is much higher.
  - For LIDCs, debt composition does not drive changes in r − g, though nominal effective interest rates show a negative association with the share of official debt.
  - Changes in public debt (increases) are more important drivers of higher r − g in LIDCs.
  - Dominance analysis confirms real interest rates have much higher explanatory value than real growth rates for both EMEs and LIDCs, consistent with previous studies.

### Debt dynamics: decomposition and average contributions
- Debt decomposition identity: change in debt-to-GDP over period equals cumulative (i) lagged debt × (i_t − γ_t)/(1+γ_t), (ii) cumulative primary deficit p_t, and (iii) cumulative stock-flow adjustment sfa_t.
- Average contributions to debt changes (full sample):
  - r − g contributed to reducing debt:
    - EMEs: 2¾ percentage points of GDP per year.
    - LIDCs: about 5½ percentage points of GDP per year.
  - Average primary deficits and stock-flow adjustments are significantly higher in LIDCs than EMEs.
  - Over time:
    - In EMEs, negative contribution of r − g has been declining by decade.
    - In LIDCs, negative contribution widened in the 1990s and 2000s but started to decline after the global financial crisis.
  - Since 2010, increases in primary deficits, less favorable r − g, and positive stock-flow adjustments contributed to average debt increases in the range of 1.2 to 1.4 percent of GDP per annum.
- Conclusion: negative r − g played a debt-stabilizing role over the last five decades, more in LIDCs than EMEs, but this dampening effect has declined and recently has not offset rising primary deficits and positive SFA.

### Tail events and extreme debt increases
- 75th percentile of the distribution shows annual debt increases above the average of:
  - EMEs: 7½ percentage points of GDP.
  - LIDCs: 10¾ percentage points of GDP.
- In these extreme events, stock-flow adjustments dominate:
  - Stock-flow adjustments:
    - EMEs: 7 percent of GDP.
    - LIDCs: 11¾ percent of GDP.
  - r − g contribution is less than half the SFA contribution in these spikes.
- Interpretation: extreme debt spikes are largely driven by factors other than r − g, consistent with previous literature.

### Fiscal crises: incidence and dynamics
- Since 1970:
  - Number of fiscal crisis episodes:
    - EMEs: 220 episodes.
    - LIDCs: 214 episodes.
  - Average duration: 5–6 years.
  - Majority of episodes associated with credit events.
  - Crises more persistent in LIDCs: in close to one third of cases a crisis starts within three years of the end of the previous one.
  - Crisis waves peaked in the mid-1990s when about 50 percent of EMEs and 80 percent of LIDCs were in fiscal crisis.
- Event-study findings (7-year windows, t=0 start of fiscal crisis):
  - Primary balances:
    - EMEs: above normal before crisis, deteriorate to a trough about 1 percentage point of GDP below normal when crisis hits; rebound to normal within three years after crisis start.
    - LIDCs: primary balance relatively flat and similar to normal times around crises.
  - r − g: no statistically significant differences around crisis episodes for both EMEs and LIDCs.
  - Debt service-to-revenue (EMEs): significant upward trend, peaking 4¼ percentage points above normal two years after crisis start.
  - Stock-flow adjustments: upward trend in the run-up to crisis in EMEs and at onset in LIDCs, though differences are negligible relative to tranquil times.
- Interpretation: r − g shows a lagged response and limited value as a leading indicator of impending fiscal crises.

### Debt dynamics over the course of a crisis: entry vs exit and duration effects
- Definitions:
  - Peak: time when debt reaches maximum after crisis onset.
  - Entry (into the crisis): period from three years before start of episode until its peak.
  - Exit (out of the crisis): period from peak to end of episode.
- Average cumulative contributions (entry phase):
  - EMEs:
    - Primary deficits: cumulative effect of 19 percent of GDP (staggering driver of debt accumulation).
    - Stock-flow adjustments: material but smaller than in LIDCs.
    - r − g: large but insufficient to offset other factors; cannot prevent sizable debt increases.
  - LIDCs:
    - Stock-flow adjustments: cumulative impact close to 18½ percent of GDP.
    - r − g: large and, in LIDCs, enough to offset negative effects of other factors during entry phase.
- Exit phase:
  - All components significantly smaller for both groups.
  - Key drivers of exit: reversal of primary deficits (turning into surpluses) and reduction in stock-flow adjustments; some normalization of interest rates helps.
- Long-haulers (crises longer than average) vs short-haulers:
  - Entry phase debt accumulation much larger in long-haulers irrespective of income group.
  - Long-haulers: primary deficits and SFA sizable in entry; primary deficits more important in EMEs, SFA more important in LIDCs.
  - Exit phase: long-haulers show reduction in adverse dynamics; EMEs run primary surpluses in exit while LIDCs may still run primary deficits of much smaller scale.
  - Short-haulers: more benign entry-phase dynamics; SFA is the overriding force in both income groups for short episodes.

*Source: Authors’ calculations and analysis from wpiea2021229-print-pdf*

### 1. EMEs 2. LIDCs

### 1. EMEs 2. LIDCs

### Determinants of stock-flow adjustments
- Stock-flow adjustments arise from:
  - Valuation effects through the impact of exchange depreciation on foreign currency denominated debt.
  - Below-the-line operations such as transactions in financial assets (for example, the injection of capital in banks), bailouts of state-owned enterprises, or government guarantees that have been called.
- Decomposition used:
  - sfa_t = ε_t (1 − α_t) (1 + i_t) d_{t−1} + bbt_t
  - Where ε_t is nominal exchange rate depreciation, α_t is the share of domestic-currency debt, and bbt_t is a residual capturing below-the-line operations and errors and omissions.
  - First term: valuation effect. Second term: materialization of contingent liabilities.

### Relative importance of valuation effects and below-the-line operations
- During the entry phase of fiscal crises:
  - Valuation effects account for about 8 percent of GDP of the debt increase in EMEs.
  - Valuation effects account for about 10¾ percent of GDP of the debt increase in LIDCs.
  - Below-the-line operations are even larger in EMEs and non-negligible in LIDCs.
- In the way out of the crisis:
  - Valuation effects remain present.
  - Below-the-line operations do not play much of a role.
- Sample/note:
  - “Into the crisis” refers to the period three years before the start of the crisis to its peak.
  - “Out of the crisis” is the period between the peak and the end of the crisis.
  - Peak is defined on the basis of debt levels.
  - Because of data constraints on external debt, the sample is restricted to 196 crisis episodes.

### Regression evidence on correlates of contingent liabilities
- Approach:
  - Fixed-effect panel regression on determinants of stock-flow adjustments.
  - Controls include pre-existing level of debt, inflation, currency depreciation, and the share of external debt.
  - For LIDCs, a dummy for the years between decision and completion points to capture HIPC-related debt relief is included.
  - Close correlates examined: private debt, private external guaranteed debt, and other public sector external debt.
- Key regression findings:
  - Inflation and exchange rate depreciation are positively associated with stock-flow adjustments for both country groups.
  - The impact of exchange rate depreciation increases with the share of external debt.
  - Private debt has a statistically significant positive effect only in EMEs (when not interacting with crisis status).
  - When interacting with fiscal crisis indicators:
    - The valuation coefficient becomes larger.
    - Private debt has a negative effect (possible explanation: private debt may migrate into the public sector balance sheet with a long lag during crises).
  - Changes in private guaranteed external debt are positively associated with stock-flow adjustments; when in fiscal crisis this association only matters for EMEs.
  - Other public sector external debt does not have a significant impact regardless of country group.
- Summary of regression implications:
  - Valuation effects are robust determinants of stock-flow adjustments for both EMEs and LIDCs.
  - Higher foreign exchange exposures lead to larger stock-flow adjustments in the face of exchange rate depreciation.
  - Below-the-line operations are particularly important for EMEs and appear associated with contingent liabilities originating in the private sector.

### Conclusions and policy implications
- Empirical context:
  - Both EMEs and LIDCs have benefited historically from highly negative r-g, but r-g has displayed an upward trend over recent decades (more so in EMEs), with high volatility and low persistence.
  - Even with low r-g, EMDEs have accumulated debt in recent years, indicating limits to what favorable r-g can accomplish amid large primary deficits or stock-flow adjustments.
  - In fiscal crises, r-g always plays a debt-stabilizing role both entering and exiting crises; the critical differences arise from primary deficits and stock-flow adjustments.
- Two policy takeaways emphasized:
  - To the extent that foreign currency exposures are large, inflation and exchange rate stability are paramount.
  - Greater attention should be given to hidden debts because the materialization of contingent liabilities is often what gets EMDEs into trouble.

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf*

### REFERENCES

### REFERENCES

### Key bibliographic sources
- Comprehensive list of working papers, journal articles, and IMF publications cited, including but not limited to works by Abbas et al. (2011); Acosta-Ormaechea (2020); Alexopoulou, Bunda, and Ferrando (2010); Alves, De Clerck, and Gamboa-Arbelaez (2020); Baldacci and Kumar (2010); Barrett (2018); Barro (2020); Blanchard (2019); Blanchard, Felman, and Subramanian (2021); Bohn (1998); Bova et al. (2019); Burger et al. (2018); Buysse, De Sloover, and Essers (2021); Campos, Jaimovich, and Panizza (2006); Chinn and Ito (2006); Ciarlone, Piselli and Trebeschi (2009); Csontó (2014); Csontó and Ivaschenko (2013); Dell’Erba, Hausmann and Panizza (2013); Di Serio, Fragetta, and Melina (2021); Escolano (2010); Escolano, Shabunina, and Woo (2017); Evans (2020); Ghosh et al. (2013); Gourinchas and Obstfeld (2012); Graf von Luckner et al. (2021); Grömping (2007); Hartelius, Kashiwase and Kodres (2008); Hasanhodzic (2020); IMF publications (2016, 2019, 2020, 2021a, 2021b); Jaramillo et al. (2017); Jaramillo and Weber (2013); Jordà, Schularick, and Taylor (2016); Kumar and Baldacci (2010); Lian, Presbitero, and Wiriadinat (2020); Mauro and Zhou (2021); Mauro et al. (2015); Mbaye, Moreno-Badia, and Chae (2018a, 2018b); Medas et al. (2018); Mehrotra and Sergeyev (2020); Miyayima, Mohanty, and Chan (2015); Moreno-Badia et al. (2020); Panizza and Taddei (2020); Turner and Spinelli (2011); Reinhart and Rogoff (2009, 2011, 2013); Reinhart, Rogoff, and Savastano (2003); Sabbadini (2018).

### Major empirical results and datasets cited
- Historical public debt database: Abbas, A.N. Belhocine, A. ElGanainy, and M. Horton (2011).
- Global Debt Database and methodology: Mbaye, Moreno-Badia, and Chae (2018b).
- Public Sector Balance Sheet Database: Alves, De Clerck, and Gamboa-Arbelaez (2020).
- Fiscal risk and debt sustainability guidance: IMF (2016), Fiscal Monitor chapters (IMF 2020 Chapter 3; IMF 2021a).

### Tables: Key numerical findings from empirical analysis

- Table 1 — Correlation between Real Interest Rate, Real Growth, and Real r-g (Full sample and subperiods):
  - Full Sample (EMEs pooling / average correlation): r-g, g = -0.27 / -0.41; r-g, r = 0.92 / 0.85; r, g = 0.09 / 0.02; N = 3041.
  - Full Sample (LIDCs pooling / average correlation): r-g, g = -0.32 / -0.38; r-g, r = 0.89 / 0.83; r, g = 0.14 / 0.11; N = 1782.
  - Pre GFC (EMEs pooling / average): r-g, g = -0.21 / -0.38; r-g, r = 0.94 / 0.85; r, g = 0.11 / 0.04; N = 2145.
  - Post GFC (EMEs pooling / average): r-g, g = -0.54 / -0.56; r-g, r = 0.71 / 0.79; r, g = 0.05 / -0.07; N = 896.
  - Notes: r-g defined in nominal terms. Real variables use GDP deflator. Second and fourth columns report average of within-country correlation.

- Table 2 — Determinants of the Interest-Growth Differential (selected coefficients; fixed effects panel regressions; robust standard errors clustered by country):
  - Lagged Real Effective Interest Rate:
    - Column (1): EMEs = 0.241*** (0.062); LIDCs = 0.167*** (0.049)
    - Column (3): EMEs = 0.215*** (0.059); LIDCs = 0.165*** (0.049)
    - Column (5): EMEs = 0.354*** (0.093); LIDCs = 0.166*** (0.049)
  - Lagged Real GDP Growth:
    - Column (1): EMEs = -0.179*** (0.055); LIDCs = 0.037 (0.068)
    - Column (5): EMEs = -0.243*** (0.077); LIDCs = 0.037 (0.068)
  - Financial Openness Index:
    - Column (1): EMEs = 5.404*** (1.270); LIDCs = 0.813 (2.347)
  - Lagged Public Debt:
    - Columns report EMEs ≈ 0.008 (0.013) and LIDCs = -0.027** (0.013)
  - Change in Public Debt:
    - Column (1): EMEs = 0.178*** (0.062); LIDCs = 0.107*** (0.024)
  - Lagged VIX:
    - Column (1): EMEs = 0.194*** (0.023); LIDCs = 0.089*** (0.021)
  - Lagged Exchange Rate Depreciation:
    - Column (1): EMEs = -0.074*** (0.009); LIDCs = -0.057*** (0.013)
  - Share Official Debt (selected): EMEs = -0.064** (0.025); Share Concessional Debt (selected): EMEs = -0.088* (0.049).
  - Observations and model fit (examples):
    - Column (1): Observations EMEs = 1,670; LIDCs = 1,304; R-squared = 0.382 (EMEs), 0.259 (LIDCs); Number of countries = 66 (EMEs), 52 (LIDCs).
  - General dominance statistics (examples):
    - Lag Real Effective Interest Rate: 0.215 (EMEs col1) / 0.185 (LIDCs col1)
    - Lag Real GDP Growth: 0.013 (EMEs col1) / 0.002 (LIDCs col1)
  - Notes: Given large outliers, r-g, inflation, and public debt are winsorized to top and bottom 1 per cent. Significance: *** p<0.01, ** p<0.05, * p<0.1.

- Table 3 — Decomposition of Debt Changes (Percent of GDP; averages):
  - Full Sample (1970-2018): Change in Debt = -0.21; Primary Deficit = 0.69; r-g = -3.74; r g = 1.99; SFA = -5.73; (final component) = 2.85.
  - EMEs (full sample): Change in Debt = 0.15; Primary Deficit = 0.41; r-g = -2.68; r g = 2.35; SFA = -5.03; (final component) = 2.42.
  - LIDCs (full sample): Change in Debt = -0.83; Primary Deficit = 1.18; r-g = -5.61; r g = 1.36; SFA = -6.97; (final component) = 3.60.
  - EMEs by decade:
    - 1980s: Change in Debt = 1.60; Primary Deficit = 2.29; r-g = -6.67; r g = 2.63; SFA = -9.30; (final) = 5.98.
    - 1990s: Change in Debt = -0.07; Primary Deficit = 1.77; r-g = -3.66; r g = 2.84; SFA = -6.50; (final) = 1.82.
    - 2000s: Change in Debt = -1.30; Primary Deficit = -1.36; r-g = -2.22; r g = 2.46; SFA = -4.68; (final) = 2.29.
    - 2010s: Change in Debt = 1.20; Primary Deficit = 0.62; r-g = -1.13; r g = 1.99; SFA = -3.12; (final) = 1.72.
  - LIDCs by decade:
    - 1980s: Change in Debt = 3.14; Primary Deficit = 1.24; r-g = -3.89; r g = 1.63; SFA = -5.52; (final) = 5.79.
    - 1990s: Change in Debt = 0.01; Primary Deficit = 0.74; r-g = -6.93; r g = 1.87; SFA = -8.81; (final) = 6.21.
    - 2000s: Change in Debt = -4.92; Primary Deficit = 0.51; r-g = -7.74; r g = 1.34; SFA = -9.08; (final) = 2.32.
    - 2010s: Change in Debt = 1.39; Primary Deficit = 2.01; r-g = -3.36; r g = 1.05; SFA = -4.41; (final) = 2.74.
  - Notes: Reported numbers are the average annual changes of debt and contributions. r term refers to the component 푖푖푡 − 훾훾푡 over 1+훾훾푡 times previous debt stock; SFA refers to stock-flow adjustment.

- Table 4 — Fiscal Crises in EMDEs, 1970–2018:
  - Number of crisis starts: EMEs = 220; LIDCs = 214.
  - Of which (percent):
    - Credit event: EMEs = 59.6; LIDCs = 82.0
    - Exceptionally large official financing: EMEs = 21.1; LIDCs = 18.0
    - Implicit domestic default: EMEs = 9.4; LIDCs = 4.7
    - Loss of market confidence: EMEs = 21.5; LIDCs = 2.8
  - Average per country: EMEs = 2.4; LIDCs = 3.7
  - Average duration: EMEs = 5.3; LIDCs = 6.1
  - Persistence (percent of crises): EMEs = 17.7; LIDCs = 29.4
  - Long haulers (percent of crises): EMEs = 33.6; LIDCs = 35.5
  - Notes: Crisis year classification uses four criteria (credit event, exceptionally large official financing, implicit domestic default, loss of market confidence). To separate events, require at least two years of no fiscal crisis between events.

- Table 5 — Determinants of Stock-Flow Adjustment (selected coefficients; fixed effects panel regressions):
  - Lag public debt:
    - Column (1) EMEs = -0.037** (0.019)
    - Column (2) LIDCs = 0.036** (0.013)
    - Other columns report mixed signs and significance across specifications.
  - Inflation:
    - Column (1) EMEs = 0.065*** (0.010); Column (2) LIDCs = 0.025 (0.065)
  - Lag External public debt * depreciation:
    - Column (1) EMEs = 0.004*** (0.001); Column (2) LIDCs = 0.001* (0.001)
  - HIPC dummy coefficients reported in some specifications (e.g., -1.649, -1.974, -3.452, -4.314**).
  - Lag private debt * fiscal crisis: negative and significant in some specifications (e.g., -0.037** (0.016); -0.090* (0.047)).
  - Lag External public debt * depreciation * fiscal crisis:
    - Examples: 0.009*** (0.002); 0.002*** (0.001); 0.003*** (0.001); 0.002*** (0.000)
  - 5y change of private external guaranteed debt:
    - Examples: 1.352*** (0.311); 2.178* (1.139)
  - Observations and fit examples:
    - Column (1): Observations = 1,838 (EMEs); Column (2): Observations = 1,417 (LIDCs).
    - R-squared examples: 0.294 (EMEs col1); 0.157 (LIDCs col2).
  - Notes: Stock-flow adjustment, inflation, exchange rate variables and HIPC dummy defined as in notes. Robust standard errors in parentheses clustered at country level. Significance: *** p<0.01, ** p<0.05, * p<0.1.

### Appendix — Data: definitions, sample, and sources
- Sample coverage:
  - Unbalanced panel of 150 countries — 92 EMEs and 58 LIDCs (IMF World Economic Outlook classification) over 1971–2018.
  - Country classification provided in Table A.1.1.
- Coverage evolution:
  - Interest expense coverage grows from 19 countries accounting for about 38 percent of EMDEs’ output in 1971 to 99 percent of GDP by 2018.
- Key data sources and variable definitions (Table A.1.2 highlights):
  - Public debt in percent of GDP: IMF, Global Debt Database (Mbaye, Moreno Badia, and Chae 2018b). Public debt includes total debt liabilities of the government with domestic and foreign creditors; debt category chosen for longest time series (non-financial public sector, general government, or central government).
  - General government interest expenses in percent of GDP: IMF, World Economic Outlook; Medas et al. (2018); Abbas et al. (2011).
  - Public debt service in percent of revenue: IMF, WEO; World Bank, World Development Indicators. Public debt service includes interest expense and external debt amortization.
  - General government primary balance, percent of GDP: IMF WEO; Medas et al. (2018); Abbas et al. (2011); Mauro et al. (2015).
  - Stock and flow adjustment: Authors’ calculations based on IMF WEO (see equation (2) in main text).
  - Interest-growth differential: Authors’ calculations based on IMF WEO and Global Debt Database; calculated as 푖푖푡 − 훾훾푡 over 1+훾훾푡 where 푖푖푡 is nominal effective interest rate and 훾훾푡 is nominal GDP growth rate. Average effective interest rate on debt computed as ratio of interest bill in year t to the stock of government debt (average of year-end t and t−1).
  - Inflation and deflator: IMF WEO (percent change of CPI; percent change of GDP deflator).
  - External exchange rate depreciation: IMF WEO (percent change of exchange rate national units per U.S. dollar, end of period).
  - External debt stocks (other public sector, private guaranteed by public sector, general government): Authors’ calculations based on IMF WEO and International Debt Statistics (IDS). IDS codes and concepts noted for each series.
  - Share of concessional debt and share of official debt: Authors’ calculations based on IMF WEO and IDS (IDS codes specified).
  - Global oil price and non-fuel commodity price: IMF WEO; IMF Primary Commodity Prices.
  - VIX index: Bloomberg Finance L.P. (period end).
  - HIPC dummy: IMF (2019) — dummy equal to 1 if country is in a year between HIPC decision and completion.
  - Fiscal crisis dummy: Medas et al. (2018) and authors’ calculations — year classified as fiscal crisis if at least one of four criteria met (credit event, exceptionally large official financing, implicit domestic default, loss of market confidence).
  - Private debt: IMF Global Debt Database (total debt, loans and securities, percent of GDP).
  - Financial Openness Index: Chinn and Ito (2006) updated to 2018.
- Notes on data construction:
  - Strategy maximizes time coverage while keeping consistent definitions at country level; original series adjusted for differences in definition and coverage when feasible.
  - Biggest data constraint often from interest expense series, affecting r-g and primary balance calculations.

*Italic: Source — wpiea2021229-print-pdf - REFERENCES (authors’ calculations and cited works as listed in the PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021229-print-pdf.pdf_