## wpiea2021125-print-pdf — Excerpted Chapters: 1.1; 2.2; 4.2; Appendix C–G

## Source details

**Canonical URL:** [wpiea2021125-print-pdf — Excerpted Chapters: 1.1; 2.2; 4.2; Appendix C–G](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021125-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021125-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021125-print-pdf.pdf.json)

---

### Related Literature
- Key overview
  - Nearly a hundred episodes of debt distress since 1990; almost forty of these ended in default.
  - Modern debt crises can occur at debt-to-output levels ranging from 89% to 180% (examples given: Argentina in 2019 at 89%, Greece in 2011 at 180%) and can occur even when prior debt levels had been sustainable (example: Lebanon in 2020).
  - Examples of sustained deficits without high market stress: Japan ran consolidated fiscal deficits over 5 percent of GDP on average for the past 20 years and reached gross debt-to-output levels of over 200%, while bond demand remained strong.
  - Balance sheet data availability has improved but remains imperfect, motivating complementary use of qualitative case discussion alongside quantitative analysis.

- Three relevant strands of literature (framework for this paper)
  - Government operating balances and propensity to default
    - Competing narratives: Conesa & Kehoe (2017) emphasize revenue drops during recessions and “gamble for redemption” behavior; Reinhart & Rogoff (2009) emphasize failure to adjust due to “this time is different” beliefs.
    - Political economy and tax/spending composition matter: tax progressivity (Ferriere, 2015; Andreasen, Sandleris & Van der Ghote, 2016) and the tendency to cut public investment (Joo, 2014) are highlighted.
    - Austerity can reduce aggregate demand and erode fiscal revenues (Standard & Poor’s, 2012), complicating the effectiveness of fiscal adjustment.
  - Linkages between government and private sector balance sheets
    - Importance of stocks (balance sheets) not just flows; Allen et al. (2002) provide a theoretical framework aggregating government, household, bank and nonbank commercial sectors into a country’s consolidated balance sheet.
    - Banking sector distress often precedes or coincides with sovereign debt crises (Reinhart & Rogoff, 2010); causality between banks and government is ambiguous (contingent liabilities vs. coerced bank purchases of sovereign debt).
    - Financial intermediaries reduce credit supply in proportion to government securities they hold during defaults, magnifying output contractions (Gennaioli et al., 2014).
    - Bank overexposure to sovereign debt is strongly associated with subsequent banking crises (Balteanu & Erce, 2017).
    - Research is scarcer on household and corporate balance sheets. Studies note corporate foreign-currency financing increases sovereign risk (Du & Schreger, 2017; Wu, 2020). Yousefi (2019) constructs a public sector balance sheet strength measure and finds markets price government assets as well as liabilities.
  - Market expectations of default reacting to macro fundamentals and political economy constraints
    - Traditional view: default events arise from growth, interest rates, and fiscal imbalances, expanded to include banking–sovereign linkages (Reinhart & Rogoff, 2011).
    - Political economy constraints to reallocating private wealth to the sovereign during crisis include reputational costs, international trade exclusion costs, domestic financial system costs, and political costs (Borensztein & Panizza, 2008)—the first three are quantifiable; political costs are harder to assess and context-specific.
    - Markets form expectations based on macroeconomic conditions, institutional and political constraints, and availability of extra-governmental resources.

- Empirical and methodological points motivating the paper
  - Paper aims to document empirical regularities for public and private sector balance sheets during sovereign distress and assess how changes in sectoral net worth relate to sovereign default likelihood.
  - Authors combine quantitative logistic-probability modeling with qualitative country-case discussion to address data limitations and causal ambiguity.
  - Causality between balance sheets and fiscal stress can run in both directions:
    - Private-sector balance sheet distress (notably banking liabilities) can trigger or accelerate sovereign crises.
    - Resilient private balance sheets can serve as additional revenue sources for the sovereign, enabling higher public debt sustainability.
  - Some empirical regularities presented are largely unexplored in recent research (e.g., tax rate patterns before default); others corroborate known findings (e.g., countries with recent default history are perceived as inherently riskier).
  - Estimates suggest that, after controlling for public debt, public sector net worth may not be the central driver of default likelihood.

- Fiscal stress and distress definitions and sample statistics used in the paper
  - “Sovereign distress” defined as compromised liquidity or solvency of the public sector; episodes end as either “defaults” (credit event, payments rescheduled or suspended) or “survivors” (no restructuring).
  - Default identification relies on Asonuma & Trebesch (2016) panel dataset (extended to 2020 by the authors).
  - Fiscal stress (Baldacci et al., 2011) criteria:
    - i. Sovereign spreads greater than 1,000 basis points,
    - ii. Inflation rates exceeding 500% for emerging markets (or 35% for advanced economies),
    - iii. Access to a large IMF-supported program, or
    - iv. A public debt default or restructuring.
  - A “survivor” is an episode with any of i)–iii) but not iv); the year of peak tension (t = 0) is the year of restructuring onset for defaults or the year with the highest bond spread for survivors.
  - Sample counts:
    - For 1990–2019: 57 episodes of fiscal stress without default (survivors) and 40 instances of default/restructuring.
    - Balance sheet assessment sample (2000–2019): 38 events total, of which 12 are defaults and 26 are survivors.
  - Summary statistic: sectoral net wealth is generally not markedly different between survivors and defaults, except Depository Corporations’ net worth at end-2019 is lower in defaulting countries than in survivors.

### Fiscal Reactions to Sovereign Distress: Stylized Facts
- Macroeconomic dynamics around distress episodes
  - Episodes driven by a sharp deterioration of the government’s net operating balance on both the revenue and expenditure sides.
  - Public saving decreases in the years leading up to the event, and private investment retracts markedly.
  - Public consumption rises before the event peak for survivors; public investment is generally not slashed for survivors. When debt distress ends in default, public investment shows strong contractions before the event.
  - Magnitude of fall in net operating balances is similar for defaults and survivors: close to one within-country standard deviation from peak to trough.
  - Timing differences:
    - For the average default episode, operating balances decrease two years before the credit event.
    - For the average survivor episode, operating balances decrease one year before the crisis peak.

- Composition of government finances
  - Sovereign distress features falls in net operating balances and tax revenues.
  - Defaults exhibit a pronounced increase in interest expenses in the years leading up to the credit event.
  - Survivors show higher interest expense after the crisis, partly reflecting increases in public debt during the episode.
  - Net financing needs:
    - Survivors: deterioration at crisis peak (a one-standard deviation change in net incurrence of liabilities) and matching shift in net borrowing.
    - Defaults: increase in net borrowing is more commonly seen two years before the credit event (not featuring the trough at t = 0).
  - Interpretation: In the five to two years prior to default, governments often finance more of overall expenses through net borrowing, with parallel increases in interest expense, potentially overstretching debt-servicing capacity and leading to default.

- Tax rates and fiscal effort
  - Tax rates data sourced from the IMF Fiscal Affairs Department Tax Policy Rates Database (sample smaller for defaults).
  - Defaults generally do not raise taxes during the crisis: income, VAT, indirect tax rates, and social contributions show no strong movements until and including the year of cessation in debt payments.
  - Evidence from constructed indices:
    - Average fiscal effort and tax effort index values: 1.25 for defaults, 1.1 for survivors, implying defaults tend to have a higher overall tax burden given macroeconomic and demographic characteristics.
  - Post-default tax behavior:
    - Tax rates (VAT, personal income tax rates, employee and employer social security contributions) show increases between 0.5 and 1.5 standard deviations after a sovereign default.
  - Caveat: tax rate sample size for defaults is much smaller than for survivors; further research needed.

- Summary findings on fiscal reactions
  - Rapid decline in government net operating balances for both defaults and survivors.
  - Defaults are more likely to feature sharp reductions in public investment in the runup to the credit event; survivors are less likely to cut public investment sharply.
  - Private investment and public saving deteriorate markedly in all cases.
  - Countries that default may enter crises in a more fiscally constrained position than survivors (supported by higher fiscal effort and tax effort indices and cuts in public investment).
  - Defaulting countries tend to face a sharper rise in interest expenditures, reflecting higher risk premia as crisis unfolds.

### Sectoral wealth and balance-sheet dynamics
- Data scope and sources
  - Government: IMF Public Sector Balance Sheet (PSBS) dataset — sample covers 43 countries for 2000-2016, with 13 survivors and 3 defaults (discussion limited to survivors due to small default count).
  - Households: Credit Suisse Global Wealth reports — covers 172 countries, with 29 survivors and 12 defaults during 2000-2019.
  - Banking sector: Monetary and Financial Statistics (MFS) database — covers 67 countries, with 32 survivors and 14 defaults across 2001-2019.
  - Stocks reported as shares of GDP; charts also available in Appendix 4 in constant 2014 US dollars.

- Government balance sheets
  - Public net worth and public net financial worth decline by close to 1 standard deviation from the year before the crisis peak to the year after, and continue to deteriorate.
  - Government assets as a share of output increase (due to denominator decline in GDP) but do not jump as liabilities do.
  - Government accumulation of financial and nonfinancial assets grinds to a halt on the crisis-peak year; liabilities (debt securities and loans) continue to increase over a ten-year window.

- Household balance sheets
  - Survivors: net household wealth accumulation stops around one year before the crisis peak and resumes between two to three years after.
  - Defaults: household wealth accumulation for the average country appears to stall up to five years before the credit event and then actively decreases (falling about 0.5 standard deviations) one year before default.
  - In most default cases in the sample, household net wealth begins an uninterrupted increase one year after the credit event.

- Banking sector balance sheets
  - Defaults:
    - Collapse in assets and net worth of commercial banks starting one year before the credit event.
    - Assets as a share of output decrease by 1 standard deviation from peak to trough.
    - Net worth falls close to 1 standard deviation in the two years before default and continues to decrease during and after the event.
    - Loans to the private sector (as share of total assets) peak three years before default; loans to households build up up to three years before then contract rapidly by around 1.5 standard deviations.
    - Central bank purchases of government securities (as share of total assets) increase up to three years before default with an uptick of 1 standard deviation from trough to peak, then reverse after the credit event.
    - Depository corporations’ net claims on central government (share of output) increase rapidly up to the year preceding default and reverse thereafter.
  - Survivors:
    - Deterioration is less pronounced: little movement in bank net worth and assets decrease by less than 1 standard deviation around crisis peak.
    - Loans to non-residents tend to build up between five to two years before the crisis peak with an increase of 1 standard deviation, then revert with a fall of the same magnitude as crisis nears.

- Summary findings on sectoral wealth
  - Distress episodes, even without default, are damaging to government wealth: net worth decline begins at crisis peak and continues for many years.
  - Survivors are not associated with severe declines in household wealth; defaults show strong impacts with household wealth accumulation stopping years before the event and resuming quickly after restructuring/default.
  - Commercial banking sector net worth stops growing a year before the peak of a survivor’s crisis episode and then falls continuously afterward; for defaults, net worth declines sharply the year before and continues down over a five-year post-crisis window.
  - Defaults feature sharp contractions in loans to the private sector; survivors’ loan contractions are concentrated in non-resident exposures.
  - Central bank purchases of government securities display an inverse V-shape in defaults; survivors do not show such movements. Net claims on government by the banking sector increase for both crisis types.

### Predicting default using sectoral wealth — baseline model and evidence
- Model framework
  - Random effects logit model predicting P(Default=1|X) = 1 / (1 + e^(−(βX))).
  - Baseline regressors drawn from an equation of motion for debt: interest payments, real output growth, the primary balance, and public debt stock.
  - Alternative specifications include revenues and expenditures split, global risk appetite (VIX), nominal exchange rate changes, and a binary for “serial defaulters” (Default in last 20yrs).
  - Preferred specification: specification (4) where probability of default is a function of equation-of-motion components plus control for serial default behavior.
  - Robustness: restricting sample to 2000-2019 does not substantially change coefficient significance or economic interpretation.
  - Endogeneity tests and collinearity diagnostics: most regressors have VIF close to 1, all VIF < 2.

- Comparison with market measures
  - Predicted default probabilities from the logit model closely track 5-year CDS spreads most of the time (CDS data obtained from Bloomberg since the 2000s).
  - Normalization for visualization: both predicted probabilities and CDS spreads are normalized by subtracting sample mean and dividing by sample standard deviation.
  - Notable mismatches:
    - Japan 2008: logit model estimates default probability over 10%, but average yearly spreads remained below 100 basis points.
    - Ukraine 2015: CDS spreads rose well above 2000 basis points, while logit model probability estimates were below 5%.
  - Interpretation: mismatches can reflect non-macroeconomic uncertainty (e.g., geopolitical risk) or additional liquid resources not captured by baseline fundamentals.

- Selected baseline model coefficient highlights (Table 1: Probability of Default, 1990-2019)
  - (L) Debt/GDP: 0.0140 ∗ (t-stat 1.86); 0.0213 ∗∗∗ (t-stat 3.24); 0.0174 ∗∗ (t-stat 1.98); 0.0113 ∗ (t-stat 1.76) across specifications.
  - (L) GDP Growth: -0.162 ∗∗∗ (t-stat -3.17); -0.163 ∗∗∗ (t-stat -3.39); -0.180 ∗∗∗ (t-stat -3.21); -0.156 ∗∗∗ (t-stat -3.28).
  - (L) Interest Expense/GDP: 0.227 ∗∗ (t-stat 1.96); 0.168 ∗ (t-stat 1.66); 0.170 ∗ (t-stat 1.78).
  - (L) Primary Balance/GDP: coefficients reported (not statistically significant in all specifications).
  - (L) Default in last 20yrs: 1.756 ∗ (t-stat 1.72); 1.092 ∗ (t-stat 1.83) in some specifications.
  - Model constants and F-statistics shown per specification; sample sizes: Countries 113, 117, 109, 113; N 2783, 2932, 3026, 2969 as reported in the table.

- Implications for analysis and policy
  - Macroeconomic fundamentals (debt/GDP, GDP growth, interest expenses, primary balance) meaningfully explain default probability.
  - Private sector balance-sheet health and sectoral wealth dynamics add predictive information and can help explain mismatches between model-implied and market-implied default risk.
  - Market measures (CDS spreads) often internalize balance-sheet information but may diverge when non-macroeconomic risks or unobserved liquid resources are relevant.

### Sectoral Wealth and Default Probabilities (Methodological extension and results)
- Methodological extension
  - Extended baseline logit model by incorporating balance sheet resilience measures as additional regressors: net worth-to-GDP for households, banks (depository corporations), and general government.
  - Variables normalized by the entire sample and standard deviation (common baseline of zero risk).
  - Time trend inclusion does not alter results.
  - Net government worth is computed net of public debt because Debt/GDP is already included as a regressor.

- Key regression results (Table 2: Probability of Default, with Sectoral Wealth (2000-2019))
  - Common regressors across specifications (coefficients and t-statistics in parentheses)
    - (L) Debt/GDP:
      - Column (1): 0.0210 *** (2.93)
      - Column (2): 0.00557 (0.32)
      - Column (3): 0.0182 ** (2.24)
      - Column (4): 0.0279 *** (3.14)
    - (L) Interest Expense/GDP:
      - Column (1): 0.190 ** (2.23)
      - Column (2): 0.448 (1.10)
      - Column (3): 0.237 ** (2.20)
      - Column (4): 0.216 ** (2.06)
    - (L) GDP Growth:
      - Column (1): -0.192 *** (-3.28)
      - Column (2): -0.313 ** (-2.26)
      - Column (3): -0.227 *** (-3.06)
      - Column (4): -0.211 *** (-2.93)
    - (L) Primary Balance/GDP:
      - Column (1): -0.0827 (-1.25)
      - Column (2): -0.0478 (-0.21)
      - Column (3): -0.0838 (-1.08)
      - Column (4): -0.0936 (-1.15)
    - (L) Default in last 20yrs:
      - Column (1): 0.415 (0.61)
      - Column (2): 1.676 (1.09)
      - Column (3): 0.470 (0.63)
      - Column (4): 0.231 (0.30)
  - Sectoral net worth regressors (coefficients and t-statistics)
    - (L) Net HH Wealth/GDP (Column 1): -0.0107 *** (-3.05)
    - (L) Gov. Net Worth/GDP (Column 2): -0.00211 (-0.17)
    - (L) Banking Sector (DC) Net Worth/GDP (Column 3): -0.115 ** (-2.24)
    - (L) Private Sector (HH+DC) Net Worth/GDP (Column 4): -0.0114 *** (-2.60)
  - Constants and model statistics
    - Constants:
      - Column (1): -5.908 *** (-4.49)
      - Column (2): -21.63 (-1.38)
      - Column (3): -6.373 *** (-4.93)
      - Column (4): -6.455 *** (-4.29)
    - F-Statistic: Column (1) 30.10; Column (2) 8.26; Column (3) 25.99; Column (4) 23.91
    - Countries: Column (1) 104; Column (2) 55; Column (3) 83; Column (4) 75
    - N: Column (1) 1951; Column (2) 760; Column (3) 1402; Column (4) 1266

- Main empirical findings
  - Higher household net wealth and higher banking sector net worth are negatively associated with the probability of experiencing default; relationships are statistically and economically significant where indicated above.
  - Deterioration in the banking sector’s net worth has a much stronger association with default probability than a similar deterioration in household balance sheets: "The effect of a 1 percentage point decline in depository corporations’ net worth on the probability of default is almost ten times that of a decrease in net household wealth."
  - Government net worth (net of public debt) is not found to be significantly associated with default likelihood after controlling for public debt, though this result is suggestive due to limited default cases in the PSBS database.
    - Possible interpretations: limited investor access to government balance sheets; non-immediacy of other government liabilities; public assets may not be short-term revenue sources.
  - Countries with a recent history of default are more likely to undergo another credit event (consistent with literature on serial defaulters).

- Model predictions and case illustrations
  - Figure 9 compares predicted default probabilities:
    - Black line: base logit model.
    - Blue line: base model + household net worth.
    - Green line: base model + banking sector (depository corporations) net worth.
    - Shaded grey areas: default years.
  - Case examples:
    - Uruguay 2002: Extended model with banking sector net worth provided a stronger signal, reflecting massive run on deposits linked to Argentina’s crisis.
    - Japan 2008: Baseline logit showed a large shift, but probability of default appears much lower when household wealth is included—consistent with high household wealth and perceived fiscal space.
    - Hungary 2008: Extended model with banking sector net worth produced especially high default probability predictions; context: financing problems on government securities market and a 2008 SBA that committed resources to the banking sector.
    - Episodes with CDS spread overreactions include Argentina 2013, Ecuador 2013, and Pakistan 2011—consistent with "serial defaulters" literature.

- Interpretations and policy implications
  - Market expectations of default tend to track deteriorations in macroeconomic fundamentals (debt-to-output ratios, sudden declines in economic growth, deterioration in interest burden or in primary expenditures).
  - Balance sheet stocks, particularly sectoral net worth, are important complements to flow measures (operating balances) for assessing debt sustainability.
    - A rapid deterioration in private sector net worth can make a country more vulnerable to default than indicated by macro fundamentals alone, especially when implicit public guarantees exist during a banking crisis.
    - Robust household wealth and a resilient banking sector can bolster a government's ability to weather fiscal stress and may explain muted market reactions in some episodes.
  - Assessments of debt sustainability could benefit from inclusion of balance sheet robustness measures to capture vulnerabilities not revealed by standard debt dynamics models.
  - In scenarios ending in default, adjustment is more likely implemented through cuts to public investment, highlighting limited short-term resource-raising options and the relevance of private balance sheet conditions.

### Appendix C–G highlights: figures, summary statistics, t-tests, and diagnostics
- Figures referenced include:
  - Figure 11a: Government Finances during Debt Distress (Defaults)
  - Figure 11b: Government Finances during Debt Distress (Survivors)
  - Figure 12: Public Sector Balance Sheets during times of Distress, in Constant US Dollars (Survivors only)
  - Figure 13a: Household Wealth, in Constant US Dollars (Defaults)
  - Figure 13b: Household Wealth, in Constant US Dollars (Survivors)
  - Appendix D note: stock data denominated in constant 2014 USD; variables normalized by subtracting country mean and dividing by country standard deviation for visual comparability.

- Summary statistics for debt crises (Table 3) — comparison period 1990-2019 (Defaults vs Survivors)
  - Number of Episodes3957
  - Debt-to-GDP at event peak, average94% (Defaults), 55% (Survivors)
  - Debt-to-GDP at event peak, maximum in sample181% (Defaults), 114% (Survivors)
  - Debt-to-GDP at event peak, minimum in sample24% (Defaults), 7% (Survivors)
  - Primary Balance on year prior to event peak, average0% (Defaults), -1.2% (Survivors)

- Mean-difference (t-test) approach (selected outputs)
  - Table 4: T-test for Household Net Wealth as %GDP (Year 2019)
    - Diff = mean(survivor) - mean(default): 52.74 37.83 -23.99 129.48
    - Diff t=1.39; P(T < t)=0.91 (one-sided), P(T < t)=0.17 (two-sided), P(T < t)=0.08 (opposite one-sided)
  - Table 5: T-test for Household Net Wealth as %GDP (Year 2000)
    - Diff = 38.50 32.93 -28.28 105.29
    - Diff t=1.16; P(T < t)=0.87 (one-sided), P(T < t)=0.25 (two-sided), P(T < t)=0.12 (opposite one-sided)
  - Table 6: T-test for Depository Corporations’ Assets as %GDP (Year 2019)
    - Default mean7.56 8.62 28.59 68.34 106.77
    - Survivor mean8.94 26.17 28.29 76.54 102.30
    - Diff1.86 10.56 -19.71 23.45; Diff t=-0.176; P(T < t)=0.56, P(T < t)=0.86, P(T < t)=0.43
  - Table 8: T-test for Depository Corporations’ Net Worth as %GDP (Year 2019)
    - Default mean1.60 1.41 4.67 4.45 10.74
    - Survivor mean11.26 1.02 4.71 10.47 14.76
    - Diff5.01 11.75 1.44 8.59; Diff t=2.865; P(T < t)=0.99, P(T < t)=0.007, P(T < t)=0.003
  - Note: Government net worth t-tests may include countries with available government net worth outside default years.

- List of defaults and survivors (selected years, Table 12: 1990–2019)
  - 1990: Defaults — Algeria, Jamaica, Chile, Bulgaria, Philippines; Survivors — Iceland, Peru
  - 1991: Defaults — Albania, Russia; Survivors — Czech Republic, Sri Lanka, Hungary
  - 1992: Survivors — Slovenia, South Africa, North Macedonia, Bosnia and Herzegovina, Croatia, Serbia, Montenegro
  - 1997: Defaults — Korea, Thailand, Croatia
  - 1998: Defaults — Russia, Ukraine, Pakistan; Survivors — Albania, Philippines
  - 2002: Defaults — Argentina; Survivors — Brazil, Serbia, Turkey
  - 2009: Defaults — (list includes) Colombia, Serbia, Poland, El Salvador, Ukraine, Sri Lanka, Bosnia and Herzegovina, Mexico, Georgia, Guatemala, Belize, Argentina, Pakistan, Angola
  - 2019: Defaults — Argentina

- Logit model results and diagnostics (Appendix F)
  - Key explanatory variables: Debt/GDP, Real GDP growth, and Interest Expense/GDP.
  - Model coefficients stable across 1990-2019 and 2000-2019 samples.
  - Extending the base logit model to incorporate sectoral net wealth: coefficients for net worth of the banking sector and net household wealth are statistically significant and have the expected sign.
  - Endogeneity checks: regressing each sectoral net wealth variable on the probability of default next period and other logit regressors finds the probability of default is not a statistically significant explanatory variable for net wealth of any of the three sectors considered.
  - Collinearity diagnostics: mean VIFs low; two variables exceed a VIF of 5 are primary expenditures and revenue (removing either does not lead to important changes in remaining coefficients).

- Selected regression coefficients (representative specifications)
  - Table 13: Probability of Default (1990-2019), specification (2)
    - (L) Debt/GDP 0.0213*** (t=3.24)
    - (L) GDP Growth -0.163*** (t=-3.39)
    - (L) Interest Expense/GDP 0.168* (t=1.66)
    - Constant -6.142** (t=-2.43)
    - Countries117; N2783; F-Statistic30.68
  - Table 14: Probability of Default (2000-2019), specification (2)
    - (L) Debt/GDP 0.0215*** (t=3.10)
    - (L) GDP Growth -0.226*** (t=-3.80)
    - (L) Interest Expense/GDP 0.228** (t=2.11)
    - Constant -6.521** (t=-2.30)
    - Countries117; N2188; F-Statistic29.83
  - Table 15: Probability of Survivor-type Crisis Peak (1990-2019), specification (1)
    - (L) Debt/GDP -0.0121* (t=-1.72)
    - (L) Primary Balance/GDP -0.141*** (t=-3.01)
    - (L) VIX 0.124*** (t=4.83)
    - Constant -4.182*** (t=-10.90)
    - Countries113; N2783; F-Statistic11.14

- Endogeneity test (Table 16) — selected coefficients
  - Dependent: (L)HH Wealth — Default coefficient -13.66 (t=-1.39)
  - Dependent: (L)Banking Sector Net Worth — Default coefficient -1.725 (t=-1.35)
  - Dependent: (L)Gov. Net Worth (Excl. Debt) — Default coefficient 13.80 (t=0.69)
  - Other controls: (L) Debt/GDP 0.639*** (t=8.93) in (1); (L) Interest Expense/GDP -8.688*** (t=-8.51) in (1)
  - Countries: 104, 83, 55 across specifications; N1951, 1402, 760; F-Statistic173.91 and 150.86 reported for subsets.

- Collinearity diagnostics (mean VIFs) — representative specifications
  - Net Household Wealth (2000-2019): Mean VIF 1.21; example VIFs: Debt/GDP 1.49; Net Household Wealth/GDP 1.04
  - Government Net Worth (2000-2019): Mean VIF 1.37; example VIFs: Debt/GDP 1.80; Government Net Worth/GDP 1.17
  - Banking Sector Net Worth (2000-2019): Mean VIF 1.20; example VIFs: Debt/GDP 1.45; Banking Sector (DC) Net Worth/GDP 1.03

*Italic: IMF Working Paper — content excerpt from wpiea2021125-print-pdf (sections 1.1; 2.2; 4.2; Appendix C–G).*

### 1.1    Related Literature    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### 1.1    Related Literature

### Key overview
- Nearly a hundred episodes of debt distress since 1990; almost forty of these ended in default.
- Modern debt crises can occur at debt-to-output levels ranging from 89% to 180% (examples given: Argentina in 2019 at 89%, Greece in 2011 at 180%) and can occur even when prior debt levels had been sustainable (example: Lebanon in 2020).
- Examples of sustained deficits without high market stress: Japan ran consolidated fiscal deficits over 5 percent of GDP on average for the past 20 years and reached gross debt-to-output levels of over 200%, while bond demand remained strong.
- Balance sheet data availability has improved but remains imperfect, motivating complementary use of qualitative case discussion alongside quantitative analysis.

### Three relevant strands of literature (framework for this paper)
- Studies on the relationship between government operating balances and the propensity to default.
  - Competing narratives: Conesa & Kehoe (2017) emphasize revenue drops during recessions and “gamble for redemption” behavior; Reinhart & Rogoff (2009) emphasize failure to adjust due to “this time is different” beliefs.
  - Political economy and tax/spending composition matter: tax progressivity (Ferriere, 2015; Andreasen, Sandleris & Van der Ghote, 2016) and the tendency to cut public investment (Joo, 2014) are highlighted.
  - Austerity can reduce aggregate demand and erode fiscal revenues (Standard & Poor’s, 2012), complicating the effectiveness of fiscal adjustment.
- Literature on linkages between government and private sector balance sheets.
  - Importance of stocks (balance sheets) not just flows; Allen et al. (2002) provide a theoretical framework aggregating government, household, bank and nonbank commercial sectors into a country’s consolidated balance sheet.
  - Banking sector distress often precedes or coincides with sovereign debt crises (Reinhart & Rogoff, 2010); causality between banks and government is ambiguous (contingent liabilities vs. coerced bank purchases of sovereign debt).
  - Financial intermediaries reduce credit supply in proportion to government securities they hold during defaults, magnifying output contractions (Gennaioli et al., 2014).
  - Bank overexposure to sovereign debt is strongly associated with subsequent banking crises (Balteanu & Erce, 2017).
  - Research is scarcer on household and corporate balance sheets. Studies note corporate foreign-currency financing increases sovereign risk (Du & Schreger, 2017; Wu, 2020). Yousefi (2019) constructs a public sector balance sheet strength measure and finds markets price government assets as well as liabilities.
- Literature on market expectations of default reacting to macro fundamentals and political economy constraints.
  - Traditional view: default events arise from growth, interest rates, and fiscal imbalances, expanded to include banking–sovereign linkages (Reinhart & Rogoff, 2011).
  - Political economy constraints to reallocating private wealth to the sovereign during crisis include reputational costs, international trade exclusion costs, domestic financial system costs, and political costs (Borensztein & Panizza, 2008)—the first three are quantifiable; political costs are harder to assess and context-specific.
  - Markets form expectations based on macroeconomic conditions, institutional and political constraints, and availability of extra-governmental resources.

### Empirical and methodological points motivating the paper
- The paper aims to document empirical regularities for public and private sector balance sheets during sovereign distress and assess how changes in sectoral net worth relate to sovereign default likelihood.
- The authors combine quantitative logistic-probability modeling with qualitative country-case discussion to address data limitations and causal ambiguity.
- Causality between balance sheets and fiscal stress can run in both directions:
  - Private-sector balance sheet distress (notably banking liabilities) can trigger or accelerate sovereign crises.
  - Resilient private balance sheets can serve as additional revenue sources for the sovereign, enabling higher public debt sustainability.
- Some empirical regularities presented are largely unexplored in recent research (e.g., tax rate patterns before default); others corroborate known findings (e.g., countries with recent default history are perceived as inherently riskier).
- Estimates suggest that, after controlling for public debt, public sector net worth may not be the central driver of default likelihood.

### Fiscal stress and distress definitions and sample statistics (used in the paper)
- “Sovereign distress” defined as compromised liquidity or solvency of the public sector; episodes end as either “defaults” (credit event, payments rescheduled or suspended) or “survivors” (no restructuring).
- Default identification relies on Asonuma & Trebesch (2016) panel dataset (extended to 2020 by the authors).
- Fiscal stress (Baldacci et al., 2011) criteria used to identify episodes:
  - i. Sovereign spreads greater than 1,000 basis points,
  - ii. Inflation rates exceeding 500% for emerging markets (or 35% for advanced economies),
  - iii. Access to a large IMF-supported program, or
  - iv. A public debt default or restructuring.
- A “survivor” is an episode with any of i)–iii) but not iv); the year of peak tension (t = 0) is the year of restructuring onset for defaults or the year with the highest bond spread for survivors.
- Sample counts:
  - For 1990–2019: 57 episodes of fiscal stress without default (survivors) and 40 instances of default/restructuring.
  - Balance sheet assessment sample (2000–2019): 38 events total, of which 12 are defaults and 26 are survivors.
- Summary statistic noted: sectoral net wealth is generally not markedly different between survivors and defaults, except Depository Corporations’ net worth at end-2019 is lower in defaulting countries than in survivors.

_Italic: IMF Working Paper — content excerpt titled “1.1    Related Literature” from the source PDF._

### 2.2    Fiscal Reactions to Sovereign Distress:  Stylized Facts

### 2.2    Fiscal Reactions to Sovereign Distress:  Stylized Facts

### Macroeconomic dynamics around distress episodes
- Episodes of sovereign distress are driven by a sharp deterioration of the government’s net operating balance on both the revenue and expenditure sides.
- Public saving decreases in the years leading up to the event, and private investment retracts markedly.
- Public consumption rises before the event peak for survivors, and public investment is generally not slashed for survivors. When debt distress ends in default, public investment shows strong contractions before the event.
- The magnitude of the fall in net operating balances is similar for defaults and survivors: close to one within-country standard deviation from peak to trough.
- Timing differences:
  - For the average default episode, operating balances decrease two years before the credit event.
  - For the average survivor episode, operating balances decrease one year before the crisis peak.

### Composition of government finances
- Sovereign distress features falls in net operating balances and tax revenues.
- Defaults exhibit a pronounced increase in interest expenses in the years leading up to the credit event.
- Survivors show higher interest expense after the crisis, partly reflecting increases in public debt during the episode.
- Net financing needs:
  - Survivors: deterioration at crisis peak (a one-standard deviation change in net incurrence of liabilities) and matching shift in net borrowing.
  - Defaults: increase in net borrowing is more commonly seen two years before the credit event (not featuring the trough at t = 0).
- Interpretation: In the five to two years prior to default, governments often finance more of overall expenses through net borrowing, with parallel increases in interest expense, potentially overstretching debt-servicing capacity and leading to default.

### Tax rates and fiscal effort
- Tax rates data sourced from the IMF Fiscal Affairs Department Tax Policy Rates Database (sample smaller for defaults).
- Defaults generally do not raise taxes during the crisis: income, VAT, indirect tax rates, and social contributions show no strong movements until and including the year of cessation in debt payments.
- Evidence from constructed indices:
  - Average fiscal effort and tax effort index values: 1.25 for defaults, 1.1 for survivors, implying defaults tend to have a higher overall tax burden given macroeconomic and demographic characteristics.
- Post-default tax behavior:
  - Tax rates (VAT, personal income tax rates, employee and employer social security contributions) show increases between 0.5 and 1.5 standard deviations after a sovereign default.
- Caveat: tax rate sample size for defaults is much smaller than for survivors; further research needed.

### Summary findings on fiscal reactions
- Sovereign distress is characterized by a rapid decline in government net operating balances for both defaults and survivors.
- Defaults are more likely to feature sharp reductions in public investment in the runup to the credit event; survivors are less likely to cut public investment sharply.
- Private investment and public saving deteriorate markedly in all cases.
- Countries that default may enter crises in a more fiscally constrained position than survivors (supported by higher fiscal effort and tax effort indices and cuts in public investment).
- Defaulting countries tend to face a sharper rise in interest expenditures, reflecting higher risk premia as crisis unfolds.

*Italicized source attribution: 2.2 Fiscal Reactions to Sovereign Distress: Stylized Facts — wpiea2021125-print-pdf*

### Sectoral wealth and balance-sheet dynamics

- Data scope and sources:
  - Government: IMF Public Sector Balance Sheet (PSBS) dataset — sample covers 43 countries for 2000-2016, with 13 survivors and 3 defaults (discussion limited to survivors due to small default count).
  - Households: Credit Suisse Global Wealth reports — covers 172 countries, with 29 survivors and 12 defaults during 2000-2019.
  - Banking sector: Monetary and Financial Statistics (MFS) database — covers 67 countries, with 32 survivors and 14 defaults across 2001-2019.
- Stocks reported as shares of GDP; charts also available in Appendix 4 in constant 2014 US dollars.

### Government balance sheets
- Public net worth and public net financial worth decline by close to 1 standard deviation from the year before the crisis peak to the year after, and continue to deteriorate.
- Government assets as a share of output increase (due to denominator decline in GDP) but do not jump as liabilities do.
- Government accumulation of financial and nonfinancial assets grinds to a halt on the crisis-peak year; liabilities (debt securities and loans) continue to increase over a ten-year window.

### Household balance sheets
- Survivors: net household wealth accumulation stops around one year before the crisis peak and resumes between two to three years after.
- Defaults: household wealth accumulation for the average country appears to stall up to five years before the credit event and then actively decreases (falling about 0.5 standard deviations) one year before default.
- In most default cases in the sample, household net wealth begins an uninterrupted increase one year after the credit event.

### Banking sector balance sheets
- Defaults:
  - Collapse in assets and net worth of commercial banks starting one year before the credit event.
  - Assets as a share of output decrease by 1 standard deviation from peak to trough.
  - Net worth falls close to 1 standard deviation in the two years before default and continues to decrease during and after the event.
  - Loans to the private sector (as share of total assets) peak three years before default; loans to households build up up to three years before then contract rapidly by around 1.5 standard deviations.
  - Central bank purchases of government securities (as share of total assets) increase up to three years before default with an uptick of 1 standard deviation from trough to peak, then reverse after the credit event.
  - Depository corporations’ net claims on central government (share of output) increase rapidly up to the year preceding default and reverse thereafter.
- Survivors:
  - Deterioration is less pronounced: little movement in bank net worth and assets decrease by less than 1 standard deviation around crisis peak.
  - Loans to non-residents tend to build up between five to two years before the crisis peak with an increase of 1 standard deviation, then revert with a fall of the same magnitude as crisis nears.

### Summary findings on sectoral wealth
- Distress episodes, even without default, are damaging to government wealth: net worth decline begins at crisis peak and continues for many years.
- Survivors are not associated with severe declines in household wealth; defaults show strong impacts with household wealth accumulation stopping years before the event and resuming quickly after restructuring/default.
- Commercial banking sector net worth stops growing a year before the peak of a survivor’s crisis episode and then falls continuously afterward; for defaults, net worth declines sharply the year before and continues down over a five-year post-crisis window.
- Defaults feature sharp contractions in loans to the private sector; survivors’ loan contractions are concentrated in non-resident exposures.
- Central bank purchases of government securities display an inverse V-shape in defaults; survivors do not show such movements. Net claims on government by the banking sector increase for both crisis types.

*Italicized source attribution: 2.2 Fiscal Reactions to Sovereign Distress: Stylized Facts — wpiea2021125-print-pdf*

### Predicting default using sectoral wealth — baseline model and evidence

- Model framework:
  - Random effects logit model predicting P(Default=1|X) = 1 / (1 + e^(−(βX))).
  - Baseline regressors drawn from an equation of motion for debt: interest payments, real output growth, the primary balance, and public debt stock.
  - Alternative specifications include revenues and expenditures split, global risk appetite (VIX), nominal exchange rate changes, and a binary for “serial defaulters” (Default in last 20yrs).
- Preferred specification: specification (4) where probability of default is a function of equation-of-motion components plus control for serial default behavior.
- Robustness:
  - Restricting sample to 2000-2019 does not substantially change coefficient significance or economic interpretation.
  - Endogeneity tests and collinearity diagnostics: most regressors have VIF close to 1, all VIF < 2.
- Comparison with market measures:
  - Predicted default probabilities from the logit model closely track 5-year CDS spreads most of the time (CDS data obtained from Bloomberg since the 2000s).
  - Normalization for visualization: both predicted probabilities and CDS spreads are normalized by subtracting sample mean and dividing by sample standard deviation.
  - Notable mismatches:
    - Japan 2008: logit model estimates default probability over 10%, but average yearly spreads remained below 100 basis points.
    - Ukraine 2015: CDS spreads rose well above 2000 basis points, while logit model probability estimates were below 5%.
  - Interpretation: mismatches can reflect non-macroeconomic uncertainty (e.g., geopolitical risk) or additional liquid resources not captured by baseline fundamentals.

### Selected baseline model coefficient highlights (Table 1: Probability of Default, 1990-2019)
- (L) Debt/GDP: 0.0140 ∗ (t-stat 1.86); 0.0213 ∗∗∗ (t-stat 3.24); 0.0174 ∗∗ (t-stat 1.98); 0.0113 ∗ (t-stat 1.76) across specifications.
- (L) GDP Growth: -0.162 ∗∗∗ (t-stat -3.17); -0.163 ∗∗∗ (t-stat -3.39); -0.180 ∗∗∗ (t-stat -3.21); -0.156 ∗∗∗ (t-stat -3.28).
- (L) Interest Expense/GDP: 0.227 ∗∗ (t-stat 1.96); 0.168 ∗ (t-stat 1.66); 0.170 ∗ (t-stat 1.78).
- (L) Primary Balance/GDP: coefficients reported (not statistically significant in all specifications).
- (L) Default in last 20yrs: 1.756 ∗ (t-stat 1.72); 1.092 ∗ (t-stat 1.83) in some specifications.
- Model constants and F-statistics shown per specification; sample sizes: Countries 113, 117, 109, 113; N 2783, 2932, 3026, 2969 as reported in the table.

### Implications for analysis and policy
- Macroeconomic fundamentals (debt/GDP, GDP growth, interest expenses, primary balance) meaningfully explain default probability.
- Private sector balance-sheet health and sectoral wealth dynamics add predictive information and can help explain mismatches between model-implied and market-implied default risk.
- Market measures (CDS spreads) often internalize balance-sheet information but may diverge when non-macroeconomic risks or unobserved liquid resources are relevant.

*Italicized source attribution: 2.2 Fiscal Reactions to Sovereign Distress: Stylized Facts — wpiea2021125-print-pdf*

### 4.2    Sectoral Wealth and Default Probabilities

### 4.2    Sectoral Wealth and Default Probabilities

### Methodological extension
- Extended baseline logit model by incorporating balance sheet resilience measures as additional regressors: net worth-to-GDP for households, banks (depository corporations), and general government.
- Variables normalized by the entire sample and standard deviation (common baseline of zero risk).
- Time trend inclusion does not alter results.
- Net government worth is computed net of public debt because Debt/GDP is already included as a regressor.

### Key regression results (Table 2: Probability of Default, with Sectoral Wealth (2000-2019))
- Coefficients and t-statistics (in parentheses). Significance: * p <0.10, ** p <0.05, *** p <0.01.
- Common regressors across specifications:
  - (L) Debt/GDP:
    - Column (1): 0.0210 *** (2.93)
    - Column (2): 0.00557 (0.32)
    - Column (3): 0.0182 ** (2.24)
    - Column (4): 0.0279 *** (3.14)
  - (L) Interest Expense/GDP:
    - Column (1): 0.190 ** (2.23)
    - Column (2): 0.448 (1.10)
    - Column (3): 0.237 ** (2.20)
    - Column (4): 0.216 ** (2.06)
  - (L) GDP Growth:
    - Column (1): -0.192 *** (-3.28)
    - Column (2): -0.313 ** (-2.26)
    - Column (3): -0.227 *** (-3.06)
    - Column (4): -0.211 *** (-2.93)
  - (L) Primary Balance/GDP:
    - Column (1): -0.0827 (-1.25)
    - Column (2): -0.0478 (-0.21)
    - Column (3): -0.0838 (-1.08)
    - Column (4): -0.0936 (-1.15)
  - (L) Default in last 20yrs:
    - Column (1): 0.415 (0.61)
    - Column (2): 1.676 (1.09)
    - Column (3): 0.470 (0.63)
    - Column (4): 0.231 (0.30)
- Sectoral net worth regressors:
  - (L) Net HH Wealth/GDP (Column 1): -0.0107 *** (-3.05)
  - (L) Gov. Net Worth/GDP (Column 2): -0.00211 (-0.17)
  - (L) Banking Sector (DC) Net Worth/GDP (Column 3): -0.115 ** (-2.24)
  - (L) Private Sector (HH+DC) Net Worth/GDP (Column 4): -0.0114 *** (-2.60)
- Constants:
  - Column (1): -5.908 *** (-4.49)
  - Column (2): -21.63 (-1.38)
  - Column (3): -6.373 *** (-4.93)
  - Column (4): -6.455 *** (-4.29)
- Model statistics:
  - F-Statistic: Column (1) 30.10; Column (2) 8.26; Column (3) 25.99; Column (4) 23.91
  - Countries: Column (1) 104; Column (2) 55; Column (3) 83; Column (4) 75
  - N: Column (1) 1951; Column (2) 760; Column (3) 1402; Column (4) 1266

### Main empirical findings
- Higher household net wealth and higher banking sector net worth are negatively associated with the probability of experiencing default; relationships are statistically and economically significant where indicated above.
- Deterioration in the banking sector’s net worth has a much stronger association with default probability than a similar deterioration in household balance sheets: "The effect of a 1 percentage point decline in depository corporations’ net worth on the probability of default is almost ten times that of a decrease in net household wealth."
- Government net worth (net of public debt) is not found to be significantly associated with default likelihood after controlling for public debt, though this result is suggestive due to limited default cases in the PSBS database.
  - Possible interpretations: limited investor access to government balance sheets; non-immediacy of other government liabilities; public assets may not be short-term revenue sources.
- Countries with a recent history of default are more likely to undergo another credit event (consistent with literature on serial defaulters).

### Model predictions and case illustrations
- Figure 9 compares predicted default probabilities:
  - Black line: base logit model.
  - Blue line: base model + household net worth.
  - Green line: base model + banking sector (depository corporations) net worth.
  - Shaded grey areas: default years.
- Case examples:
  - Uruguay 2002: Extended model with banking sector net worth provided a stronger signal, reflecting massive run on deposits linked to Argentina’s crisis.
  - Japan 2008: Baseline logit showed a large shift, but probability of default appears much lower when household wealth is included—consistent with high household wealth and perceived fiscal space.
  - Hungary 2008: Extended model with banking sector net worth produced especially high default probability predictions; context: financing problems on government securities market and a 2008 SBA that committed resources to the banking sector.
  - Episodes with CDS spread overreactions include Argentina 2013, Ecuador 2013, and Pakistan 2011—consistent with "serial defaulters" literature.

### Interpretations and policy implications
- Market expectations of default tend to track deteriorations in macroeconomic fundamentals (debt-to-output ratios, sudden declines in economic growth, deterioration in interest burden or in primary expenditures).
- Balance sheet stocks, particularly sectoral net worth, are important complements to flow measures (operating balances) for assessing debt sustainability.
  - A rapid deterioration in private sector net worth can make a country more vulnerable to default than indicated by macro fundamentals alone, especially when implicit public guarantees exist during a banking crisis.
  - Robust household wealth and a resilient banking sector can bolster a government's ability to weather fiscal stress and may explain muted market reactions in some episodes.
- Assessments of debt sustainability could benefit from inclusion of balance sheet robustness measures to capture vulnerabilities not revealed by standard debt dynamics models.
- In scenarios ending in default, adjustment is more likely implemented through cuts to public investment, highlighting limited short-term resource-raising options and the relevance of private balance sheet conditions.

*Source: IMF Working Paper chapter — 4.2 Sectoral Wealth and Default Probabilities (2000–2019).*

### Appendix  C.  Additional  Figures:  Government  Finances  and  Ag-

### Appendix C. Additional Figures: Government Finances and Aggregate Macroeconomic Variables

### Figures and balance-sheet presentation
- Figures referenced:
  - Figure 11a: Government Finances during Debt Distress (Defaults)
  - Figure 11b: Government Finances during Debt Distress (Survivors)
  - Figure 12: Public Sector Balance Sheets during times of Distress, in Constant US Dollars (Survivors only)
  - Figure 13a: Household Wealth, in Constant US Dollars (Defaults)
  - Figure 13b: Household Wealth, in Constant US Dollars (Survivors)
- Appendix D note: stock data for government and private sector balance sheets are denominated in constant 2014 USD; variables normalized by subtracting country mean and dividing by country standard deviation for visual comparability.

### Summary statistics for debt crises (Table 3)
- Comparison period: 1990-2019
- Defaults vs Survivors:
  - Number of Episodes3957
  - Debt-to-GDP at event peak, average94% (Defaults), 55% (Survivors)
  - Debt-to-GDP at event peak, maximum in sample181% (Defaults), 114% (Survivors)
  - Debt-to-GDP at event peak, minimum in sample24% (Defaults), 7% (Survivors)
  - Primary Balance on year prior to event peak, average0% (Defaults), -1.2% (Survivors)

### Mean-difference (t-test) approach (Appendix E)
- Method: t-tests conducted separately on the first and last years of the sample (usually 2000 and 2019) because standard independent-sample t-tests cannot be performed on panel data.
- Comparison groups: countries that featured a default at any point in 2000-2019 (categorized as default if experienced both types) vs countries with a “survivor” distress episode.
- Notable t-test outputs (selected):
  - Table 4: T-test for Household Net Wealth as %GDP (Year 2019)
    - Diff = mean(survivor) - mean(default): 52.7437.83-23.99    129.48
    - Diff t=1.39; P(T < t)=0.91 (one-sided), P(T < t)=0.17 (two-sided), P(T < t)=0.08 (opposite one-sided)
  - Table 5: T-test for Household Net Wealth as %GDP (Year 2000)
    - Diff = 38.5032.93-28.28    105.29
    - Diff t=1.16; P(T < t)=0.87 (one-sided), P(T < t)=0.25 (two-sided), P(T < t)=0.12 (opposite one-sided)
  - Table 6: T-test for Depository Corporations’ Assets as %GDP (Year 2019)
    - Default mean7.568.6228.5968.34    106.77
    - Survivor mean8.9426.1728.2976.54    102.30
    - Diff1.8610.56-19.71    23.45; Diff t=-0.176; P(T < t)=0.56, P(T < t)=0.86, P(T < t)=0.43
  - Table 8: T-test for Depository Corporations’ Net Worth as %GDP (Year 2019)
    - Default mean1.601.414.674.45    10.74
    - Survivor mean11.261.024.7110.47    14.76
    - Diff5.011.751.44    8.59; Diff t=2.865; P(T < t)=0.99, P(T < t)=0.007, P(T < t)=0.003
- Note on government net worth data: ‘default’ group may include countries for which no wealth data exist for the exact year of default; government net worth t-tests include countries with available government net worth outside default years.

### List of defaults and survivors (selected years)
- Table 12 (1990–2019): examples by year (Crisis Peak)
  - 1990: Defaults — Algeria, Jamaica, Chile, Bulgaria, Philippines; Survivors — Iceland, Peru
  - 1991: Defaults — Albania, Russia; Survivors — Czech Republic, Sri Lanka, Hungary
  - 1992: Survivors — Slovenia, South Africa, North Macedonia, Bosnia and Herzegovina, Croatia, Serbia, Montenegro
  - 1997: Defaults — Korea, Thailand, Croatia
  - 1998: Defaults — Russia, Ukraine, Pakistan; Survivors — Albania, Philippines
  - 2002: Defaults — Argentina; Survivors — Brazil, Serbia, Turkey
  - 2009: Defaults — (list includes) Colombia, Serbia, Poland, El Salvador, Ukraine, Sri Lanka, Bosnia and Herzegovina, Mexico, Georgia, Guatemala, Belize, Argentina, Pakistan, Angola
  - 2019: Defaults — Argentina

### Logit model results and diagnostics (Appendix F)
- Main findings:
  - Key explanatory variables: Debt/GDP, Real GDP growth, and Interest Expense/GDP.
  - Model coefficients are fairly stable across the 1990-2019 and 2000-2019 samples.
  - Extending the base logit model to incorporate sectoral net wealth: coefficients for net worth of the banking sector and net household wealth are statistically significant and have the expected sign.
- Endogeneity checks:
  - Regressed each sectoral net wealth variable on the probability of default next period and other logit regressors.
  - Finding: the probability of default is not a statistically significant explanatory variable for net wealth of any of the three sectors considered.
- Collinearity diagnostics and robustness:
  - Collinearity is present, but variance inflator factors (VIFs) for statistically significant variables are not high enough to cast serious doubts on results.
  - The two variables that exceed a VIF of 5 are primary expenditures and revenue (noting that removing either of these does not lead to important changes in remaining coefficients).

### Selected regression coefficients (Tables 13–15, representative specifications)
- Table 13: Probability of Default (1990-2019), specification (2)
  - (L) Debt/GDP 0.0213*** (t=3.24)
  - (L) GDP Growth -0.163*** (t=-3.39)
  - (L) Interest Expense/GDP 0.168* (t=1.66)
  - Constant -6.142** (t=-2.43)
  - Countries117; N2783; F-Statistic30.68
- Table 14: Probability of Default (2000-2019), specification (2)
  - (L) Debt/GDP 0.0215*** (t=3.10)
  - (L) GDP Growth -0.226*** (t=-3.80)
  - (L) Interest Expense/GDP 0.228** (t=2.11)
  - Constant -6.521** (t=-2.30)
  - Countries117; N2188; F-Statistic29.83
- Table 15: Probability of Survivor-type Crisis Peak (1990-2019), specification (1)
  - (L) Debt/GDP -0.0121* (t=-1.72)
  - (L) Primary Balance/GDP -0.141*** (t=-3.01)
  - (L) VIX 0.124*** (t=4.83)
  - Constant -4.182*** (t=-10.90)
  - Countries113; N2783; F-Statistic11.14

### Endogeneity test (Table 16)
- GLS estimation of lagged wealth variables on default probability (2000-2019), selected coefficients:
  - Dependent: (L)HH Wealth — Default coefficient -13.66 (t=-1.39)
  - Dependent: (L)Banking Sector Net Worth — Default coefficient -1.725 (t=-1.35)
  - Dependent: (L)Gov. Net Worth (Excl. Debt) — Default coefficient 13.80 (t=0.69)
  - Other significant controls: (L) Debt/GDP 0.639*** (t=8.93) in (1); (L) Interest Expense/GDP -8.688*** (t=-8.51) in (1)
  - Countries: 104, 83, 55 across specifications; N1951, 1402, 760; F-Statistic173.91 and 150.86 reported for subsets.

### Collinearity diagnostics (Tables 17–19), mean VIFs
- Specification using Net Household Wealth (2000-2019)
  - Mean VIF 1.21
  - Example VIFs: Debt/GDP 1.49; Net Household Wealth/GDP 1.04
- Specification using Government Net Worth (2000-2019)
  - Mean VIF 1.37
  - Example VIFs: Debt/GDP 1.80; Government Net Worth/GDP 1.17
- Specification using Banking Sector Net Worth (2000-2019)
  - Mean VIF 1.20
  - Example VIFs: Debt/GDP 1.45; Banking Sector (DC) Net Worth/GDP 1.03

*Source: Appendix C–G, “Additional Figures: Government Finances and Aggregate Macroeconomic Variables” (selected tables and text).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021125-print-pdf.pdf_