## Sovereign Debt Repatriation During Crises — Working Paper No. WP/2022/077

## Source details

**Canonical URL:** [Sovereign Debt Repatriation During Crises — Working Paper No. WP/2022/077](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022077-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022077-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022077-print-pdf.pdf.json)

---

### Data description
- Data set: sovereign debt investor database recently extended to all advanced and emerging markets following the approach of Arslanalp and Tsuda (2014a; 2014b).
- Sample coverage: 189 countries: All 120 advanced and emerging market countries that are classified as having market access according to the IMF as of end-2019, plus 69 low-income countries.
- Time span: Annual data compiled from 1989 until 2020 (where available).
- External debt data availability: External debt data are available for 180 countries.
- Primary international sources used: BIS, Eurostat (supplemented with AMECO), IMF/World Bank Quarterly Public Sector Debt (QPSD) database, IMF Global Debt Database (GDD), ECB, World Bank International Debt Statistics (IDS), and national sources to extend series where necessary.
- Debt concept: Gross general government debt (or central government debt if general government data unavailable).
- Debt instruments included: currency and deposits; debt securities; and loans.
- Exclusions: accounts payable, insurance and pension reserves, social security obligations, financial derivatives, contingent liabilities such as guaranteed debt.
- Valuation and basis: All debt figures are expressed in face value and on a gross basis.
- Measurement currency: Debt is measured in local currency units.
- For European Union countries: Definition matches the Maastricht debt definition.
- Investor residence principle: Investor types are defined according to the residence principle — classified as external if held by nonresidents from the perspective of the country of issuance.
- Treatment of foreign official loans: For most analyses foreign official loans (bilateral and multilateral) are netted out from external debt to focus on private external investors; data sources include Arslanalp and Tsuda with additional IDS and national sources.
- Figures referenced:
  - Figure 1: Unconditional distributions — LHS: External private plus domestic debt as a percent of GDP; RHS: External private debt as a percent of external private plus domestic debt.
  - Figure 2: Unweighted country-average over time (1990–2020) — LHS: Total debt (% of GDP); RHS: External debt share (% of external private plus domestic debt).

### Summary statistics (key numbers)
- Table 1: Summary statistics (Mean, SD, min, max, N)
  - Total debt % of GDP: Mean 56, SD 40, min 0, max 292, N 5,255
  - Total debt % of GDP net of foreign official loans: Mean 34, SD 31, min 0, max 254, N 4,926
  - External debt % of GDP: Mean 31, SD 30, min 0, max 286, N 4,710
  - External debt % of GDP net of foreign official loans: Mean 8, SD 12, min 0, max 354, N 4,673
  - External debt % of total: Mean 54, SD 28, min 0, max 100, N 4,757
  - External debt % of total net of foreign official loans: Mean 21, SD 25, min 0, max 100, N 4,536
- Interpreting Table 1:
  - "Total debt relative to GDP stands at 57 percent on average across all countries and years."
  - "External debt to GDP is just over 30 percent and the external debt share therefore a little over 50 percent."
  - Net of foreign official loans:
    - "Total debt to GDP is just 34 percent on average, and only a fifth of this is externally held."
  - The paper focuses on debt measures net of foreign official loans for analyses of private sovereign debt investor behavior.
- Distributional notes:
  - Both total debt to GDP and external debt share (net of official loans) are skewed to the left.
  - Low-debt examples: Australia and Chile.
  - High-debt examples: Japan and Lebanon.
  - Some countries have mainly external debt during crises; others have little external debt for most of the sample (e.g., India and Saudi Arabia).

### Time trends
- Unweighted country averages (net of foreign official loans) over 1990–2020:
  - Debt as a share of GDP trended down through most of the 1990s until the 2008 financial crisis; picked up thereafter rising to around 40 percent at the end of the sample (an all-time high).
  - External debt share increased from a trough of around 15 percent in the early 1990s to 25 percent currently; largely flat in the 2000s with increases in the 1990s and 2010s.

### Country group differences (key numbers)
- Table 2: Summary statistics by country groups (Mean)
  - Advanced countries:
    - Total debt % of GDP: 61
    - Total debt % of GDP net of foreign official loans: 58
    - External debt % of GDP: 25
    - External debt % of GDP net of foreign official loans: 22
    - External debt % of total: 40
    - External debt % of total net of foreign official loans: 38
  - Emerging countries:
    - Total debt % of GDP: 47
    - Total debt % of GDP net of foreign official loans: 34
    - External debt % of GDP: 21
    - External debt % of GDP net of foreign official loans: 8
    - External debt % of total: 55
    - External debt % of total net of foreign official loans: 24
  - Low-income countries:
    - Total debt % of GDP: 66
    - Total debt % of GDP net of foreign official loans: 24
    - External debt % of GDP: 44
    - External debt % of GDP net of foreign official loans: 2
    - External debt % of total: 66
    - External debt % of total net of foreign official loans: 10
- Interpretation:
  - Advanced countries have virtually no foreign official loans on average.
  - Emerging market and especially low-income countries: foreign official loans make up a large fraction of their external debt.

### Group dynamics by crisis and region
- Advanced countries:
  - Sharp increase in debt to GDP after the 2008 financial crisis: average debt to GDP increases from below 50 percent before the crisis to nearly 70 percent by 2012, then levels off before COVID-19.
  - External share increased since mid-1990s from 30 percent to 40 percent by the start of the financial crisis and stayed roughly constant since.
- Emerging market and low-income countries:
  - Total debt trended down during the 2000s, rising again in the 2010s.
  - Low-income trend partly driven by debt relief initiatives (incl. HIPC/MDRI) and commodity booms.
  - Emerging markets: external debt share increased from 20 to 30 percent over the sample.
  - Low-income countries: external share stayed below ten percent for the last two decades.
- Regional patterns for emerging markets:
  - Latin America: consistently higher external debt share; debt to GDP rose around the turn of the millennium.
  - Emerging Asia: spike around the late 1990s Asian crisis.
  - EMEA: debt to GDP fell around 2000s but external share has been trending up.

### Baseline findings: Repatriation during crises (external debt share)
- Main outcome: external debt share (external private debt as percent of external private plus domestic debt).
- Event-study baseline results (selected dynamics for External debt share):
  - Sovereign defaults: elevated external share prior to default and a decline around the event; evidence of repatriation in the run-up to sovereign default episodes.
    - D(-4) 10.95 * (0.05)
    - D(-3) 9.83 * (0.07)
    - D(-2) 11.92 ** (0.03)
    - D(-1) 9.80 * (0.08)
    - D(0) 6.95 (0.18)
    - D(1) 7.80 (0.11)
    - D(2) 3.35 (0.47)
    - D(3) 1.59 (0.70)
    - D(4) −0.09 (0.98)
  - Banking crises: decline in external share observed but not always statistically significant; baseline estimates show persistent positive deviations before and after crisis onset.
    - D(-4) 9.33 *** (0.00)
    - D(-3) 5.94 ** (0.02)
    - D(-2) 8.46 *** (0.00)
    - D(-1) 6.70 *** (0.01)
    - D(0) 6.96 ** (0.02)
    - D(1) 4.49 * (0.07)
    - D(2) 5.56 ** (0.04)
    - D(3) 4.87 ** (0.04)
    - D(4) 3.82 ** (0.03)
  - Currency crises: external share rises until start year and then declines at crisis start; repatriation more tied to immediate crisis onset.
    - D(-4) 1.83 (0.48)
    - D(-3) −0.50 (0.84)
    - D(-2) 0.85 (0.72)
    - D(-1) 1.19 (0.60)
    - D(0) 6.52 ** (0.02)
    - D(1) 2.91 (0.20)
    - D(2) 2.67 (0.22)
    - D(3) 3.18 (0.12)
    - D(4) 3.46 * (0.07)
- Note: Robust SE in parentheses. *p<0.1, **p<0.05, ***p<0.01.

### Alternative measure: external debt flows and exchange rate effects
- Flow variable constructed as exchange-rate adjusted change in external private debt, reported as x_t/GDP_t with
  - x_t = [f$_,t$ e_t − f$_,t−1$ e_{t−1}] ẽ_t + (f_t − f_{t−1})
  - Definitions provided for f$_,t$, f_t, e_t, ẽ_t, and adjustments for debt write-offs using Cruces and Trebesch (2013).
- Assumptions on currency issuance:
  - For 24 “major” emerging market countries actual currency split available from 2004; before 2004 assume exclusively foreign currency issuance; in 2004 average local currency share = six percent.
- Results (flows):
  - Point estimates consistent with baseline: external debt flows fall below the non-crisis mean over the course of sovereign defaults, though uncertainty is larger for flows than for the stock measure.
  - Banking crises: flows not significantly different from baseline, though point estimates suggest a decline following crisis onset and significantly higher flows in the year of crisis start.
  - Currency crises: point estimates suggest expatriation before and reversal after crisis start, but most estimates statistically indistinguishable from non-crisis mean over event window (except higher flows at very end).

### Repatriation and preemptive debt restructuring
- Default classification per Asonuma and Trebesch (2016): strictly preemptive, weakly preemptive, post-default (hard/unilateral).
- Key findings from regressions (External debt share and ExtFlows):
  - Hard (post-default/unilateral) and weakly preemptive restructurings are associated with a significant decline in the external debt share.
  - Strictly preemptive restructurings are not associated with significant repatriation in the external debt share.
  - Using external debt flows: significantly negative flows only for weakly preemptive restructurings; downward trend for post-default restructurings; quick rebound after strictly preemptive restructurings.
- Selected exact coefficients from Table 5 (External debt share, ExtShare column for key leads):
  - Hard default: D(-2) = 17.04 ** (robust SE 0.03); D(-1) = 9.59 (0.18); D(0) = 6.45 (0.33).
  - Weakly preemptive: D(-4) = 21.21 ** (0.02); D(-3) = 19.54 ** (0.01); D(0) = 12.35 (0.14).
  - Strictly preemptive: D(-2) = −9.14 (0.21); D(0) = −3.47 (0.64).
  - Table (5) observations: 4,605 / 4,570. Adj. R^2: 0.02 / 0.02.

### Market structure, secondary markets, and financial repression
- Hypotheses tested:
  - Free-market mechanism: larger, more liquid secondary markets facilitate repatriation.
  - Financial repression: domestic policy constraints could force repatriation.
- Classification dimensions (averaged over three years prior to default):
  - Market size: log of total outstanding debt in USD deflated by US GDP deflator.
  - Capital account openness: Chinn-Ito index.
- Sample: classification feasible for 38 of 65 default episodes; cutoffs based on mean across episodes.
- Results:
  - Repatriation pronounced among defaults in large and open markets.
  - No repatriation observed in large and closed markets.
  - Small and closed markets: estimates imprecise.
  - Interpretation: financial controls (closed markets) do not reinforce repatriation; market size facilitates repatriation; openness matters conditional on market size.
- Caveat: relatively few sovereign defaults per group; results may not generalize.

### Other channels and limitations
- Domestic capital flight:
  - Literature highlights domestic capital flight and reserve losses; study focuses on domestic financial system as counterparty to foreigners selling sovereign bonds — both phenomena can co-occur.
- Maturity:
  - Debt maturity tends to shorten during defaults; ownership breakdown by maturity not available in current data — repatriation measure captures net outcomes across maturities.
  - Authors note maturity-dimension extension as future research.

### Repatriation and recessions
- Two questions examined:
  1. Does GDP fall more in crises with strong repatriation?
  2. Do external creditors leave during normal cyclical downturns or only during full-blown crises?
- Definitions:
  - Strong (weak) repatriation: above (below) average cumulative change in external debt share over event window (crisis-type-specific average). Alternative thresholds yield similar results.
  - GDP: WDI (or WEO if longer coverage), real logs, HP-filtered with λ = 100 for annual data.
- Findings:
  - Crises with strong repatriation show larger drops in external debt share by definition.
  - Differences in GDP outcomes between strong- and weak-repatriation crises are not statistically significant.
  - Point estimates suggest strong repatriation defaults may be associated with a less sharp but more protracted output decline, but no strong evidence of worse GDP implications for crises with disproportionate external investor departures.
- Cyclical repatriation regressions (selected coefficients and samples from Table 6):
  - Column (1) Ext. share: GDP (cycle) coefficient = −0.15 (SE 0.14), sample = Full, Obs = 4,157, Adj. R^2 = 0.01.
  - Column (2) Ext. flows/GDP: GDP (cycle) coefficient = 0.02 (SE 0.11), sample = Full, Obs = 4,121, Adj. R^2 = 0.01.
  - Column (5) Ext. share (LIC sample): GDP (cycle) coefficient = −0.35 ** (SE 0.02), sample = LIC, Obs = 1,649, Adj. R^2 = 0.04.
- Interpretation:
  - Repatriation is generally acyclical: external investors do not tend to leave during normal cyclical downturns.
  - Exception: for low-income countries, external investors provide relatively more lending during recessions (significantly negative coefficient).

### Other variables during crises: Debt, GDP, and current account
- GDP and current account dynamics:
  - All crises associated with recessions; peak to trough GDP declines by around three percent in all cases.
  - Banking crises tend to occur when GDP is above the non-crisis baseline.
  - Sovereign default and currency crises start with GDP not significantly different from their non-crisis means.
  - Current account improves clearly during defaults.
  - Current account displays an upward trend during currency crises (less pronounced).
  - Banking crises accompanied by current account deterioration in years after crisis.
- Indebtedness and external debt share dynamics:
  - Total private debt to GDP is countercyclical in all country groups; least pronounced for low-income countries.
  - Sovereign defaults:
    - Indebtedness rises in the run-up and falls substantially after defaults.
    - Magnitude: indebtedness rises to around 20 percentage points above the non-crisis baseline and then falls back down to zero in the five years after the start of a default.
  - Currency crises: similar pattern to defaults with a rise prior to event and smaller fall post onset.
  - Banking crises: associated with a rise in debt levels and no pronounced post-crisis fall (consistent with government borrowing to smooth banking crises).

### Robustness checks and sensitivity analyses
- Alternative sovereign default measure ("big defaults only") confirms baseline: external share falls over the event window; its first difference is significantly negative, as are external flows.
- Adding market size and openness to baseline specification leaves main results robust.
- Crisis consolidation: alternative consolidation assumptions between zero and three years for defaults and currency crises produce results within baseline confidence intervals and qualitatively similar.
- Granger causality tests:
  - Two sets of country-by-country Granger causality tests reported; counts by country type presented in source (printed rows appear as concatenated numbers).
  - Summary: "Overall there are substantially more countries in the first two columns than the last two (about 70 percent of the total number of countries), that is, we find more evidence in favor of crises causing repatriation than the other way around."
  - Evidence that crises Granger-cause repatriation is stronger for advanced and emerging markets than for low-income countries.
- Foreign official loans and alternate debt measures:
  - Including foreign official loans offsets private repatriation during sovereign defaults and at the onset of currency crises: the external share including foreign official loans is large and flat over the event window for sovereign defaults and does not decline at the start of currency crises.
  - Total debt to GDP including foreign official loans increases post currency crisis and to a lesser extent after sovereign defaults.
  - Interpretation: governments use foreign official loans predominantly when hit by crises, especially sovereign default and currency crises.

### Robustness, subsample, and quarterly evidence
- Fixed effects and sample restrictions: alternative specifications include (1) Baseline; (2) No country or year fixed effects; (3) No year fixed effects; (4) Emerging market countries only; (5) Advanced economies only; (6) Market access countries only. Adjusted R2 reported across models include 0.03, 0.05, 0.03, 0.05, 0.21, 0.06.
- Event horizon sensitivity: shorter (2 year) and longer (10 year) event windows preserve qualitative patterns.
- Severe currency crises (top quartile of nominal exchange rate depreciations in EM/LIC sample; total 47 crises): point estimates suggest a drop in external share between one year before and one year after crisis onset; confidence intervals wider than for all currency crises.
- Granger-test restricted sample yields qualitatively similar repatriation dynamics; in this subset repatriation may start later for banking crises and defaults.
- Quarterly data:
  - Quarterly investor data available from 2004 for 48 countries; only three sovereign defaults have full 9 year crisis event window coverage in quarterly series: Greece 2012Q3, Ukraine 2015Q1, Argentina 2019Q4.
  - Quarterly results provide higher-frequency checks for a much smaller set of events.

### Policy implications and interpretation
- Empirical findings:
  - Systematic repatriation—shifts of sovereign debt from external private to domestic investors—occurs during some types of crises (notably sovereign defaults) but not others.
  - Timing and extent of repatriation depend on crisis type and market structure.
  - Pattern consistent with a differential default risk hypothesis: debt moves toward domestic investors who may be treated less harshly.
  - Pattern less consistent with a pure informational asymmetry explanation, which would likely operate across crisis types.
- Policy options suggested:
  - Careful management of debt renegotiations seems conducive towards avoiding repatriation flows.
  - Large and open debt markets facilitate repatriation flows.

*Italic: Source: Sections 2.1, 2.2, 2.3, 4.3, 4.9, and related excerpts from wpiea2022077-print-pdf.*

### 2.1  Data description

### 2.1  Data description

### Data coverage and sources
- Data set: sovereign debt investor database recently extended to all advanced and emerging markets following the approach of Arslanalp and Tsuda (2014a; 2014b).
- Sample coverage: 189 countries: All 120 advanced and emerging market countries that are classified as having market access according to the IMF as of end-2019, plus 69 low-income countries.
- Time span: Annual data compiled from 1989 until 2020 (where available).
- External debt data availability: External debt data are available for 180 countries.
- Primary international sources used: BIS, Eurostat (supplemented with AMECO), IMF/World Bank Quarterly Public Sector Debt (QPSD) database, IMF Global Debt Database (GDD), ECB, World Bank International Debt Statistics (IDS), and national sources to extend series where necessary.
- The Arslanalp and Tsuda external debt series follows original methodology with additional source data from the ECB, QPSD, IDS, and national sources.
- The sovereign debt investor data set is available at the URLs provided in the original data documentation (Arslanalp and Tsuda).

### Definition and measurement of debt
- Debt concept: Gross general government debt (or central government debt if general government data unavailable).
- Debt instruments included: currency and deposits; debt securities; and loans.
- Exclusions: accounts payable, insurance and pension reserves, social security obligations, financial derivatives, contingent liabilities such as guaranteed debt.
- Valuation and basis: All debt figures are expressed in face value and on a gross basis.
- Measurement currency: Debt is measured in local currency units.
- For European Union countries: Definition matches the Maastricht debt definition.
- Note on currency and deposits: "Currency and deposits mainly represent saving certificates and retail bonds that can be redeemed before maturity, and are generally a relatively small share of total debt for most economies."

### Investor classification and external debt concept
- Investor residence principle: Investor types are defined according to the residence principle — classified as external if held by nonresidents from the perspective of the country of issuance.
- Contrast to older definitions: This residence-based external debt differs from definitions based on currency of debt, country of jurisdiction, or market of issuance.
- For most emerging market and low-income countries: general government debt consists predominantly of central government debt.

### Treatment of foreign official loans
- For most analyses in the study: foreign official loans (bilateral loans from other countries and multilateral loans from international financial institutions) are netted out from external debt to focus on private external investors.
- Data sources for foreign official loans: updated Arslanalp and Tsuda data set with additional sources from the World Bank’s IDS database and national sources.

### Figures referenced
- Figure 1: Unconditional distributions of observations in the panel.
  - LHS: External private plus domestic debt as a percent of GDP.
  - RHS: External private debt as a percent of external private plus domestic debt.
- Figure 2: Unweighted country-average over time (1990–2020).
  - LHS: External private plus domestic debt as a percent of GDP (Total debt).
  - RHS: External private debt as a percent of external private plus domestic debt (External debt share).

---

### 2.2  Summary statistics (key numbers)
- Table 1: Summary statistics (Mean, SD, min, max, N)
  - Total debt % of GDP: Mean 56, SD 40, min 0, max 292, N 5,255
  - Total debt % of GDP net of foreign official loans: Mean 34, SD 31, min 0, max 254, N 4,926
  - External debt % of GDP: Mean 31, SD 30, min 0, max 286, N 4,710
  - External debt % of GDP net of foreign official loans: Mean 8, SD 12, min 0, max 354, N 4,673
  - External debt % of total: Mean 54, SD 28, min 0, max 100, N 4,757
  - External debt % of total net of foreign official loans: Mean 21, SD 25, min 0, max 100, N 4,536
- Interpreting Table 1:
  - "Total debt relative to GDP stands at 57 percent on average across all countries and years."
  - "External debt to GDP is just over 30 percent and the external debt share therefore a little over 50 percent."
  - Net of foreign official loans:
    - "Total debt to GDP is just 34 percent on average, and only a fifth of this is externally held."
  - The paper focuses on debt measures net of foreign official loans for analyses of private sovereign debt investor behavior.
- Distributional notes:
  - Both total debt to GDP and external debt share (net of official loans) are skewed to the left.
  - Examples of low debt observations: Australia and Chile.
  - Examples of high debt observations: Japan and Lebanon.
  - Some countries have mainly external debt during crises; others have little external debt for most of the sample (e.g., India and Saudi Arabia).

### Time trends (from Figure 2 summary)
- Unweighted country averages (net of foreign official loans) over three decades:
  - Debt as a share of GDP trended down through most of the 1990s until the 2008 financial crisis; picked up thereafter rising to around 40 percent at the end of the sample (an all-time high).
  - External debt share increased from a trough of around 15 percent in the early 1990s to 25 percent currently; largely flat in the 2000s with increases in the 1990s and 2010s.

### 2.3  Country group differences (key numbers)
- Table 2: Summary statistics by country groups (Mean)
  - Advanced countries:
    - Total debt % of GDP: 61
    - Total debt % of GDP net of foreign official loans: 58
    - External debt % of GDP: 25
    - External debt % of GDP net of foreign official loans: 22
    - External debt % of total: 40
    - External debt % of total net of foreign official loans: 38
  - Emerging countries:
    - Total debt % of GDP: 47
    - Total debt % of GDP net of foreign official loans: 34
    - External debt % of GDP: 21
    - External debt % of GDP net of foreign official loans: 8
    - External debt % of total: 55
    - External debt % of total net of foreign official loans: 24
  - Low-income countries:
    - Total debt % of GDP: 66
    - Total debt % of GDP net of foreign official loans: 24
    - External debt % of GDP: 44
    - External debt % of GDP net of foreign official loans: 2
    - External debt % of total: 66
    - External debt % of total net of foreign official loans: 10
- Interpretation:
  - Advanced countries have virtually no foreign official loans on average.
  - Emerging market and especially low-income countries: foreign official loans make up a large fraction of their external debt.

### Debt and external share dynamics by group (Figure 3 summary)
- Advanced countries:
  - Sharp increase in debt to GDP after the 2008 financial crisis: average debt to GDP increases from below 50 percent before the crisis to nearly 70 percent by 2012, then levels off before COVID-19.
  - External share increased since mid-1990s from 30 percent to 40 percent by the start of the financial crisis and stayed roughly constant since.
- Emerging market and low-income countries:
  - Total debt trended down during the 2000s, rising again in the 2010s.
  - Low-income country trend partly driven by debt relief initiatives (incl. HIPC/MDRI) and commodity booms.
  - Emerging markets: external debt share increased from 20 to 30 percent over the sample.
  - Low-income countries: external share stayed below ten percent for the last two decades.
- Regional patterns for emerging markets:
  - Latin America: higher external debt share consistently; debt to GDP rose around the turn of the millennium.
  - Emerging Asia: spike around the late 1990s Asian crisis.
  - EMEA (Europe, Middle East and Africa): debt to GDP fell around 2000s but external share has been trending up.

---

*Italic: Source: 2.1 Data description; 2.2 Summary statistics; 2.3 Country group differences (extracted from wpiea2022077-print-pdf).*

### 4.3  Repatriation during crises

### 4.3  Repatriation during crises

### Baseline findings (external debt share)
- Main outcome: external debt share (external private debt as percent of external private plus domestic debt).
- Results (column 1 of Table (4) and Figure (5)):
  - External share is significantly above non-crisis levels prior to sovereign defaults and falls back down to non-crisis levels by the time the crisis starts.
  - The decline over the event window for sovereign defaults is statistically significant — evidence of debt repatriation in the run-up to sovereign default episodes.
  - Banking crises show a decline in external share, but it is not statistically significant.
  - Currency crises: the external share rises until the start year of the crisis (above the non-crisis mean) and then declines significantly at crisis start — repatriation appears limited to immediate crisis onset.
- Note: Both external and domestic debt to GDP levels are high before sovereign defaults, but not banking or currency crises in the sample (see appendix figures).

### Alternative measure: external debt flows and exchange rate effects
- Two concerns with external debt share:
  - Repatriation is a flow concept; external debt share is a stock measure.
  - Exchange rate movements may cause valuation changes in external debt share rather than actual sales.
- Flow variable construction (exchange-rate adjusted change in external private debt, reported as x_t/GDP_t):
  - x_t = [f$_,t$ e_t − f$_,t−1$ e_{t−1}] ẽ_t + (f_t − f_{t−1})
  - Definitions: f$_,t$ = stock of foreign-denominated private external debt (LCU) at t; f_t = stock of local-currency denominated private external debt (LCU) at t; e_t = end-of-period exchange rate (LCU per USD) at t; ẽ_t = average exchange rate between t−1 and t.
  - Adjusted for debt write-offs using Cruces and Trebesch (2013) face-value estimates (dataset does not include write-offs in f_t and f$_,t$).
  - Assumptions on currency issuance by country group and treatment before 2004 (see text): for 24 “major” emerging market countries actual currency split available from 2004; before 2004 assume exclusively foreign currency issuance; in 2004 average local currency share = six percent.
- Results (Figure (6) and column 2 of Table (4)):
  - Point estimates consistent with baseline: external debt flows fall below the non-crisis mean over course of sovereign defaults.
  - Uncertainty is larger for flows than for stock measure.
  - Banking crises: flows not significantly different from baseline, though point estimates suggest a decline following crisis onset and significantly higher flows in the year of crisis start.
  - Currency crises: point estimates suggest expatriation before and reversal after crisis start, but cannot reject no significant difference in flows from non-crisis mean over event window (except for higher flows at very end of window).
  - Overall: significant repatriation during sovereign defaults remains after controlling for exchange rate effects; repatriation during banking and currency crises is more tentative once exchange rates are accounted for.

### Repatriation and preemptive debt restructuring
- Classification of defaults per Asonuma and Trebesch (2016): strictly preemptive, weakly preemptive, post-default (hard/unilateral).
- Regressions using three crisis dummies (Figure (7) and Table (5)):
  - Hard (post-default/unilateral) and weakly preemptive restructurings are associated with a significant decline in the external debt share.
  - Strictly preemptive restructurings are not associated with significant repatriation in the external debt share.
  - Using external debt flows (alternative measure): significantly negative flows only for weakly preemptive restructurings; downward trend for post-default restructurings; quick rebound after strictly preemptive restructurings.
- Selected exact coefficients from Table 5 (External debt share, ExtShare column for key leads):
  - Hard default: D(-2) = 17.04 ** (robust SE 0.03); D(-1) = 9.59 (0.18); D(0) = 6.45 (0.33).
  - Weakly preemptive: D(-4) = 21.21 ** (0.02); D(-3) = 19.54 ** (0.01); D(0) = 12.35 (0.14).
  - Strictly preemptive: D(-2) = −9.14 (0.21); D(0) = −3.47 (0.64).
  - Note: Table (5) also reports ExtFlows column and significance stars as shown in source. Obs: 4,605 / 4,570. Adj. R^2: 0.02 / 0.02.

### The role of secondary markets and financial repression
- Hypotheses:
  - Free-market mechanism: larger, more liquid secondary markets facilitate repatriation (Broner et al., 2010; 2014).
  - Financial repression (government moral suasion): domestic pressure could force repatriation (Reinhart and Sbrancia, 2015).
- Empirical approach:
  - Classify defaults by two dimensions averaged over three years prior to default: market size (log of total outstanding debt in USD deflated by US GDP deflator) and capital account openness (Chinn-Ito index).
  - Data allow classification for 38 of 65 default episodes.
  - Cutoffs based on mean across episodes.
- Results (Figure (9) and Figure (10)):
  - Repatriation is pronounced among defaults occurring in large and open markets.
  - No repatriation observed in large and closed markets.
  - Small and closed markets: estimates are imprecise (large uncertainty).
  - Interpretation: financial controls (closed markets) do not reinforce repatriation; defaults in tightly controlled markets exhibit less repatriation. Market size appears to facilitate repatriation; openness matters conditional on market size.
- Caveat: analysis relies on relatively few sovereign defaults per group; results may not generalize.

### Other channels
- Domestic capital flight
  - Literature emphasizes loss of reserves and domestic capital flight by households/firms (Frankel and Schmukler (1996); Choe et al. (2001)).
  - Possible reconciliation: informational advantages of domestic investors may have eroded; literature focuses on households/firms while this study focuses on domestic financial system as counterparty to foreigners selling sovereign bonds — both domestic capital flight and repatriation of sovereign debt can occur simultaneously.
- Maturity
  - Debt maturity tends to shorten during defaults (Arellano and Ramanarayanan, 2012; Sanchez et al., 2018).
  - Repatriation interactions with maturity are possible (e.g., foreigners selling long-term bonds, rolling over short-term holdings).
  - Data limitations: cannot analyze ownership breakdown by maturity; repatriation measure captures net outcomes across maturities. Authors note extension along maturity dimension as future research avenue.

### Repatriation and recessions
- Two questions addressed:
  1. Does GDP fall more in crises with strong repatriation?
  2. Do external creditors leave during normal cyclical downturns or only during full-blown crises?
- Definitions and methods:
  - Crises with strong (weak) repatriation: above (below) average cumulative change in external debt share over event window (crisis-type-specific average). Alternative thresholds (unconditional average, median, upper quartile, maximum) yield similar results.
  - GDP series: WDI (or WEO if longer coverage), real logs, HP-filtered with λ = 100 for annual data.
- Findings on GDP implications (Figure (11)):
  - Crises with strong repatriation are associated with larger drops in external debt share (by definition).
  - Differences in GDP outcomes between strong- and weak-repatriation crises are not statistically significant.
  - Point estimates: strong repatriation defaults may be associated with a less sharp but more protracted output decline, but no strong evidence of worse GDP implications for crises with disproportionate external investor departures.
- Cyclical repatriation (Table 6):
  - Regress cyclical component of GDP on repatriation measures, controlling for country and year fixed effects and crisis dummies.
  - Results (selected coefficients and samples):
    - Column (1) Ext. share: GDP (cycle) coefficient = −0.15 (SE 0.14), sample = Full, Obs = 4,157, Adj. R^2 = 0.01.
    - Column (2) Ext. flows/GDP: GDP (cycle) coefficient = 0.02 (SE 0.11), sample = Full, Obs = 4,121, Adj. R^2 = 0.01.
    - Column (5) Ext. share (LIC sample): GDP (cycle) coefficient = −0.35 ** (SE 0.02), sample = LIC, Obs = 1,649, Adj. R^2 = 0.04.
  - Interpretation:
    - Repatriation is generally acyclical: external investors do not tend to leave during normal cyclical downturns.
    - Exception: for low-income countries, coefficient is significantly negative, suggesting external investors provide relatively more lending during recessions in these countries.
- Additional context (Figure (12)):
  - Plots dynamics of total debt (external private plus domestic), output, and current account around crises by crisis type.
  - Overall, repatriation appears linked to elevated sovereign default risk in crises rather than to ordinary output downturns.

*Source: 4.3 Repatriation during crises (excerpt from provided IMF chapter PDF).*

### 4.9  Other variables during crises: Debt, GDP, and current account

### 4.9  Other variables during crises: Debt, GDP, and current account

### GDP and current account dynamics
- All crises are associated with recessions; peak to trough GDP declines by around three percent in all cases.  
- Onset timing:
  - Banking crises tend to occur when GDP is above the non-crisis baseline.
  - Sovereign default and currency crises start with GDP not significantly different from their non-crisis means.
- Current account behavior:
  - The current account improves very clearly in the course of defaults.
  - The current account displays an upward trend during currency crises, although this is less pronounced and smaller.
  - Banking crises are accompanied by a current account deterioration in the years after the crisis.

### Indebtedness and external debt share dynamics
- Total private debt to GDP is countercyclical in all country groups; this is least pronounced for low-income countries.  
- Sovereign defaults:
  - Indebtedness, measured as total debt to GDP, rises in the run-up to sovereign defaults and then falls substantially in the wake of defaults.
  - Magnitude: indebtedness rises to around 20 percentage points above the non-crisis baseline and then falls back down to zero in the five years after the start of a default.
- Currency crises:
  - Follow a similar pattern to defaults: a rise in debt prior to the event and a somewhat smaller fall post crisis onset.
- Banking crises:
  - Associated with a rise in debt levels (no pronounced post-crisis fall), consistent with governments borrowing to smooth negative effects of banking crises.

### Robustness checks and sensitivity analyses
- Alternative sovereign default measure ("big defaults only") confirms baseline: the external share falls over the event window; its first difference is significantly negative, as are external flows.
- Controls:
  - Adding market size and openness to the baseline specification leaves main results robust (column 3 of Table (7) shows robustness).
- Crisis consolidation:
  - Baseline consolidates currency and default crises that start within three years as the same episode (first year used as crisis start).
  - Alternative consolidation assumptions between zero and three years for defaults and currency crises produce results that lie within baseline confidence intervals and are qualitatively insignificantly different from the baseline in most cases.
- Granger causality (reverse causality) tests:
  - Two sets of country-by-country Granger causality tests: (1) external share dependent, crisis dummy independent; (2) reversed.
  - Table (8) reports counts by country type for which non-causality is rejected at the 95% level. The printed table rows/columns appear as:
    - Advanced8131123
    - Emerging14156742
    - Low-income81410739
    - Total30421715104
  - Summary statement preserved from the source: "Overall there are substantially more countries in the first two columns than the last two (about 70 percent of the total number of countries), that is, we find more evidence in favor of crises causing repatriation than the other way around."
  - The evidence that crises Granger-cause repatriation is stronger for advanced and emerging markets than for low-income countries.

### Foreign official loans and alternate debt measures
- Baseline analysis nets out foreign official loans to focus on private lending.
- Including foreign official loans:
  - Foreign official loans offset private repatriation during sovereign defaults and at the onset of currency crises: the external share including foreign official loans is large and flat over the event window for sovereign defaults and does not decline at the start of currency crises.
  - There is no marked shift in dynamics surrounding banking crises when foreign official loans are included.
  - Total debt to GDP including foreign official loans:
    - Increases post currency crisis, and to a lesser extent after sovereign defaults.
    - Point estimates suggest debt levels including foreign official loans are more elevated than private debt throughout the event window.
  - Interpretation: governments use foreign official loans predominantly when hit by crises, especially sovereign default and currency crises.

### Policy implications and interpretation
- Findings:
  - There is systematic repatriation—shifts of sovereign debt from external private to domestic investors—during some types of crises but not others.
  - Timing and extent of repatriation depend on crisis type and market structure.
  - The pattern of repatriation being crisis-type dependent is consistent with a differential default risk hypothesis (debt moves toward domestic investors who will be treated less harshly).
  - The pattern is less consistent with a pure informational asymmetry explanation, since an informational advantage would likely operate across crisis types.
- Policy options suggested:
  - Careful management of debt renegotiations seems conducive towards avoiding repatriation flows.
  - Large and open debt markets facilitate repatriation flows.

*Source: 4.9 Other variables during crises: Debt, GDP, and current account — wpiea2022077-print-pdf*

### 8073. National Bureau of Economic Research, Inc (cit. on p. 21).

### Sovereign Debt Repatriation During Crises — Working Paper No. WP/2022/077

### Key empirical patterns: event-study dynamics of external debt share
- Sovereign defaults: Point estimates from the baseline event-study show elevated external debt share in the years before default and a decline around the event:
  - D(-4) 10.95 * (0.05)
  - D(-3) 9.83 * (0.07)
  - D(-2) 11.92 ** (0.03)
  - D(-1) 9.80 * (0.08)
  - D(0) 6.95 (0.18)
  - D(1) 7.80 (0.11)
  - D(2) 3.35 (0.47)
  - D(3) 1.59 (0.70)
  - D(4) −0.09 (0.98)
- Banking crises: Baseline estimates indicate persistent positive deviations in external debt share before and after crisis onset:
  - D(-4) 9.33 *** (0.00)
  - D(-3) 5.94 ** (0.02)
  - D(-2) 8.46 *** (0.00)
  - D(-1) 6.70 *** (0.01)
  - D(0) 6.96 ** (0.02)
  - D(1) 4.49 * (0.07)
  - D(2) 5.56 ** (0.04)
  - D(3) 4.87 ** (0.04)
  - D(4) 3.82 ** (0.03)
- Currency crises: Baseline estimates show a pronounced contemporaneous increase in external debt share at crisis onset and positive deviations in subsequent years:
  - D(-4) 1.83 (0.48)
  - D(-3) −0.50 (0.84)
  - D(-2) 0.85 (0.72)
  - D(-1) 1.19 (0.60)
  - D(0) 6.52 ** (0.02)
  - D(1) 2.91 (0.20)
  - D(2) 2.67 (0.22)
  - D(3) 3.18 (0.12)
  - D(4) 3.46 * (0.07)
- Table note: Dependent variable in all models is External debt share. Robust SE in parentheses. *p<0.1, **p<0.05, ***p<0.01.

### Robustness and subsample evidence
- Fixed effects and sample restrictions:
  - Results reported for alternative specifications: (1) Baseline; (2) No country or year fixed effects; (3) No year fixed effects; (4) Emerging market countries only; (5) Advanced economies only; (6) Market access countries only.
  - Table (9) presents these variants; adjusted R2 ranges reported across models include 0.03, 0.05, 0.03, 0.05, 0.21, 0.06.
- Event horizon sensitivity:
  - Figure 18 compares shorter (2 year) and longer (10 year) event windows to baseline and shows persistence of the qualitative patterns across horizons.
- Severe currency crises:
  - Severe currency crises are defined as the top quartile of nominal exchange rate depreciations in the sample of emerging market or low-income countries and include a total of 47 crises.
  - Point estimates suggest a drop in the external share between one year before and one year after the crisis onset for severe crises; confidence intervals for severe crises are wider than for all currency crises.
- Granger-test restricted sample:
  - Estimation restricting to countries where Granger tests suggest causality runs only in the intended direction yields qualitatively similar repatriation dynamics to the baseline.
  - In this subset, for defaults the level is lower, and for both banking crises and defaults repatriation may start later.

### Quarterly data: limitations and supplementary results
- Quarterly investor data availability and coverage:
  - Quarterly investor data are available from 2004 for 48 countries (major emerging market countries and advanced countries) from Arslanalp and Tsuda (2014a) and Arslanalp and Tsuda (2014b).
  - The quarterly series cover the full 9 year crisis event window for just three sovereign defaults: Greece 2012Q3, Ukraine 2015Q1, and Argentina 2019Q4.
  - The quarterly series cover part of the event window for three additional defaults: Argentina 2001Q4, Ukraine 2000Q1, and Uruguay 2003Q1.
  - Using quarterly data and the Frankel and Rose (1996) definition yields 22 currency crises. The authors know of no quarterly banking crisis measure.
- Quarterly results:
  - Figure (21) shows quarterly event-study dynamics of external debt share for sovereign defaults and currency crises using the available quarterly data; these provide higher-frequency checks though for a much smaller set of events.

### Additional empirical indicators and figures (as presented)
- Figures and diagnostics included in the source material:
  - Figure 16: Repatriation, market size, and openness.
  - Figure 17: External and domestic debt to GDP ratios during crises (External debt, pct of GDP (net of foreign official loans); Domestic debt, % of GDP (net of foreign official loans)).
  - Figure 19: Repatriation during severe currency crises (External debt share and External debt flows).
  - Figure 20: Repatriation for subset of countries from Granger tests (External debt share and External debt flows).
  - Figure 21: Quarterly data: Sovereign defaults and currency crises (External debt, % of total (net of foreign official loans) and individual default series for Greece 2012, Ukraine 2015, Argentina 2019).

### Data and sample composition notes
- The source provides a country classification and investor-base dataset covering:
  - Advanced countries, Emerging market countries (with “major” starred), and Low-income countries (full lists appear in the source).
- The empirical analysis uses an annual investor database to cover a much larger set of events than available in quarterly series.

*Source: Sovereign Debt Repatriation During Crises — Working Paper No. WP/2022/077*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022077-print-pdf.pdf_