## wp17173 — Section III and IV summary

## Source details

**Canonical URL:** [wp17173 — Section III and IV summary](https://www.imf.org/-/media/files/publications/wp/2017/wp17173.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17173.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17173.pdf.json)

---

### Literature review: determinants of sovereign spreads and crisis probability
- Two empirical strands inform the paper:
  - Determinants of sovereign bond spreads and market access:
    - Most studies focus on gross debt and find a positive correlation between gross debt and long-term interest rates or spreads.
    - Recent work highlights investor base composition: larger non-resident participation and larger domestic investor participation tend to lower borrowing costs even when debt levels are moderate to high.
    - A smaller literature examines assets’ role (often net debt or OECD/AM focus); some studies find net debt relevant for yields and forward rates.
    - Hadzi-Vaskov and Ricci (2016) find gross debt and assets have significant effects on spreads that roughly offset each other; this paper refines that view by distinguishing asset categories.
    - This paper: net debt matters for sovereign spreads and probability of default, but gross debt and assets may not offset depending on asset category; for more liquid assets, assets have a larger impact than gross debt on spreads and probability of default.
  - Determinants of probability of financial crises in developing and emerging economies:
    - Larger gross debt and international financial volatility raise crisis likelihood in EMs; stronger fundamentals (reserves, growth, lower current account deficits) reduce crisis probability.
    - Little prior attention to non-reserve asset holdings; assets and reserves are complementary self-insurance instruments and can serve as collateral and signaling devices to lower borrowing costs, smooth rollover risks and reduce debt distress risk.
- Contribution of the paper:
  - Controls for international reserves and shows government financial assets (beyond reserves) contain information for predicting sovereign spreads and default probability.
  - Emphasizes distinct effects across asset categories — liquidity matters.

### Methodology: economic rationale and empirical specifications
- Economic model and key relations:
  - Lender breakeven: p_it(θR_it) + (1−p_it)R_it = R_t^*
  - Spread relation (Equation (1)): s_it = h p_it / (1 − h p_it) R_t^*
  - Default probability specification (Equation (2)): p_it = p(A_it, X_it)
- Two empirical targets: country risk premium (spread) and default probability.
- Baseline spread panel specification (Equation (3)):
  - spread_it = α_i + β Assets_it−1 + Λ′ X_it + e_it
  - Λ′ X_it = Λ_1 b_it−1 + Λ′_2 Z_it−1 + Λ_3 W_t
  - Z includes: real GDP growth, reserves-to-GDP, current account balance-to-GDP, inflation, shocks to country credit ratings; W baseline is the VIX.
  - Expectations: Λ_1 > 0 (gross debt raises spreads); β < 0 (assets reduce spreads). Identification allows distinct impacts (Λ_1 ≠ −β).
- Heterogeneity for Emerging Markets (Equation (4)):
  - spread_it = α_i + β Assets_it−1 + γ Assets_it−1 EM + δ EM + Λ′ X_it + e_it
  - β and β + γ capture marginal impacts for AMs and EMs.
- Nonlinearity: pooled quantile regressions (Equations (5) and (6)) to let β(τ) and β(τ)+γ(τ) vary across spread distribution (e.g., 50th vs 90th percentile).
- Default probability: pooled probit (Equations (7) and (8)):
  - P(y=1 | Assets_it, X_it) = Φ(α + β Assets_it + Λ′ X_it)
  - y = 1 indicates debt distress (default, restructuring, or near-default proxied by IMF financing exceeding one hundred per cent of quota on a commitment basis); episodes not preceded by another episode ending in any of the two previous years.
- Estimation practices:
  - Panel fixed effects with clustered standard errors; OLS for spreads; pooled quantile regressions; pooled probit for distress probability.
  - Debt and assets used as ratios to nominal GDP.

### Data: sample, asset categories, measurement, and descriptive patterns
- Sample:
  - 110 market-access countries: 30 AMs and 80 EMs.
  - Asset data series span back to the late 1980s; descriptive stats use 1980–2015 where applicable.
- Financial asset categories (GFSM 2014 definitions; eight instruments):
  - A1 Monetary gold and SDRs
  - A2 Currency and deposits
  - A3 Debt securities
  - A4 Loans
  - A5 Equity and investment fund shares
  - A6 Insurance, pension and standardized guarantees (IPSG)
  - A7 Financial derivatives
  - A8 Other accounts receivable
- Aggregated asset groupings and sample averages (1980–2015, simple averages across countries):
  - Total financial assets (A1+A2+A3+A4+A5+A6+A7+A8): 43.8% GDP
  - Held in debt instruments (A1+A2+A3+A4+A6+A8): 24.3% GDP
  - Liquid (A2+A3): 11.4% GDP
  - Highly liquid (A2): 7.3% GDP
- Asset holdings by government units and measurement choices:
  - Central government typically holds the bulk; data do not identify sovereign wealth funds separately in many cases.
  - Assets exclude central bank international reserves and non-financial assets; assets may be domestic or abroad and denominated in domestic or foreign currency; data lump these together.
  - Data sources: OECD, Eurostat, IMF’s Government Finance Statistics Yearbook (GFSY), WEO, IFS; longest available series used per country.
- Descriptive statistics and crisis patterns:
  - Mean total assets (percent of GDP, obs 543 AM, 291 EM, 834 All): AM 47.4, EM 37.1, All 43.8.
  - Held in debt instruments mean: AM 27.2, EM 19.1, All 24.3.
  - Instrument medians (1980–2015, examples): A2 Currency & deposits median AM 7.1 / EM 7.6; A3 Debt securities median AM 5.6 / EM 1.1; A5 Equity & investment fund shares median AM 19.7 / EM 17.6; A8 Other accounts receivable median AM 7.4 / EM 7.2.
  - Distribution skew: top asset holder (Norway) about ten times larger than sample mean.
  - Cyclicality: assets and gross debt countercyclical in AMs and roughly acyclical in EMs overall; spreads countercyclical in both groups.
- Crisis frequency and duration (1980–2015, Table 5):
  - Number of episodes: AMs 9, EMs 112, All 121.
  - Average duration in years: AMs 3, EMs 5, All 5.
  - Crisis triggers include outright defaults, restructurings, IMF financing, and combinations.

### Main results — Financial assets and sovereign spreads
- Panel OLS key patterns (selected reported coefficients and signs):
  - Debt/GDP, lagged (Λ1) positive and significant (e.g., 2.302*** in one specification).
  - Reserves/GDP, lagged negative and significant (examples: −2.421**, −2.731***, −2.417***, −2.576*** across specifications).
  - Real GDP growth, lagged negative and significant (e.g., −13.68***).
  - Inflation positive and significant (e.g., 14.56***).
  - Country credit ratings negative and significant (e.g., −5.775***).
  - Assets/GDP (β) impact less systematic than reserves; total assets and assets held in debt instruments add information in some specifications.
  - F-tests sometimes indicate joint impact of gross debt and assets not statistically different from zero in some specifications.
- Quantile regressions:
  - Gross debt response to spreads positive and increasing in risk (Λ1(τ)).
  - AMs: marginal impact of assets on spreads β(τ) is near zero or around OLS average — assets less signaling-relevant.
  - EMs: sensitivity of spreads to assets (β(τ)+γ(τ)) larger and statistically significant for higher quantiles of spread risk and across asset categories.
  - Quantified example from the paper: a 10-percentage point (pp) increase in assets by an EM at the 90th percentile of risk would reduce sovereign spreads by 60–100bps, compared to 0–50bps for a country around the median of the distribution.
- Economic interpretation:
  - Very liquid assets better mitigate macro shocks, risk premia, exchange rates, and commodity-price shocks; deployable against liquidity pressures and debt service difficulties in EMs.
  - Asset-liability management (borrowing long-term to build liquid buffers) can be useful but is not universal — if borrowing costs exceed asset returns, fiscal deterioration may follow.

### Main results — Financial assets and probability of debt distress
- Pooled probit results (Table 7) — key patterns:
  - Gross debt is a key determinant of debt crisis likelihood across specifications (Debt/GDP, lagged positive and often significant; examples include coefficients like 2.125*** in some blocks).
  - Assets alone in full specifications typically not statistically significant for AMs.
  - Liquid and highly liquid assets have economically significant effects in EMs: ceteris paribus, a 10-percentage point increase in both gross debt and assets would reduce crisis probability by 2–4 pp depending on specification.
  - Some marginal effects reported for assets are large and significant (examples: Marginal effect of assets −0.387**, −0.448** in blocks).
- Liquidity vs solvency crises:
  - Liquidity episodes: 156 identified (123 in EMs) using criteria including spread spikes above 1000bps or deviations >2 standard deviations and large IMF programs.
  - Solvency episodes: 67 identified (65 in EMs) using deep debt treatments (face value reductions or NPV haircuts above median).
  - Highly liquid assets correlate more with liquidity crises; total assets more correlated with solvency crises. Marginal effects and significance vary by specification and asset category.
- Robustness to broader distress definitions and rare-events:
  - Broadening distress to include spread spikes yields 149 episodes (33 in AMs, 116 in EMs); under this definition, gross debt–crisis correlation weakens but liquid assets remain significant in EMs; estimates imply a 10-pp increase in debt and assets would reduce crisis probability by about 6 percentage points in EMs.
  - Using WEO asset data yields significant asset coefficients for the overall sample but less differentiation for EMs, reflecting WEO’s mixing of asset types.
  - Rare-event considerations: unconditional probability of debt distress in regression sample ~2 percent; alternative estimators and robustness checks applied.

### Robustness checks and sensitivity analyses
- Alternative global risk controls tested (international risk-free rates, term premia, US corporate high yield) — none dominated baseline VIX.
- Inclusion of output gap (HP-filter) alongside GDP growth: output gap statistically significant with positive sign in most quantiles; inclusion does not undermine GDP growth nor assets results.
- Pre-GFC sample (up to 2007): assets in debt instruments and highly liquid assets remain significant at top quantiles but smaller sample warrants caution.
- HVR parsimonious specification with WEO assets: gross debt matters at higher quantiles; assets matter across distribution for EMs.
- Probit robustness: adding output gap weakens some individual asset coefficient significance but marginal effects generally significant; EM-only regressions confirm liquidity importance.
- WEO-based regressions and vintage/revision cautions: WEO reporting heterogeneous; assets often inferred as residual (gross minus net debt) and subject to optimistic revisions.

### Illustrative simulations and counterfactuals
- Counterfactuals for an EM at sample median fundamentals with starting gross debt of 70 and 90 percent of GDP:
  - Scenario 1 (reduce spreads from 90th percentile 600bps to sample mean 300bps):
    - Required asset holdings depend on asset quality:
      - Highly liquid/liquid categories: around 10–15 percent of GDP.
      - All financial instruments (total assets): around 40–45 percent of GDP.
  - Scenario 2 (reduce likelihood of debt distress to sample mean 5 percent):
    - For initial gross debt 70 percent of GDP: required assets 3.5–7.5 percent of GDP depending on asset category.
    - For initial gross debt 90 percent of GDP: required assets between 7 percent of GDP (liquid categories) and about 25 percent of GDP (less liquid categories).
- Interpretation: asset quality and policy goals determine asset-liability management design; accumulating buffers complements fiscal adjustment and other measures; fiscal savings may be the feasible source for buffers.

### Policy implications and concluding remarks
- Main findings:
  - Assets reduce debt sustainability risks in EMs; asset liquidity matters — liquid and highly liquid assets more effective in reducing spreads and crisis probability for high-risk EMs.
  - Assets are less systematically relevant for AMs in reducing spreads or distress probability.
  - Nonlinear effects: assets more valuable at higher spread quantiles for EMs.
- Policy guidance:
  - Accumulating asset buffers can strengthen market access and debt sustainability prospects; size and profile of holdings should be country-specific.
  - Liquid assets are especially valuable for countries vulnerable to rollover risk or seeking market access; long-maturity assets may help meet future liabilities.
  - Asset accumulation strategies should be anchored by prudent fiscal policies and strong fiscal institutions; financing assets with debt is viable only with careful asset-liability management and favorable market conditions.
  - Asset accumulation should complement other insurance instruments: bilateral swap lines, regional financing arrangements, IMF precautionary facilities, and market-based hedging.
- Data and measurement caveats:
  - Paper excludes non-financial assets due to data limitations and limited utility for debt sustainability assessments (illiquid, hard to value).
  - Improving transparency, coverage, and reliability of asset data is needed to operationalize benchmarks for assets in debt sustainability analyses.

*Source: wp17173 (IMF Working Paper — Section III identification strategies; Section IV data and identification).*

### Section III discusses our identification strategies. Section IV describes the data and identify

### wp17173 - Section III discusses our identification strategies. Section IV describes the data and identify

### Literature review: determinants of sovereign spreads and crisis probability
- Two major strands of empirical literature inform the paper:
  - Determinants of sovereign bond spreads and market access:
    - Most studies focus on gross debt and typically find a positive correlation between gross debt and long-term interest rates or spreads (examples cited: Edwards (1986), Eichengreen and Mody (2000), Borensztein and Panizza (2008), Gelos and Sandleris (2011), Comelli (2012), Cruces and Trebesch (2013)).
    - More recent research finds that larger non-resident participation in local sovereign debt markets and larger participation of domestic investors tend to face lower borrowing costs even when debt levels are moderate to high (examples cited: Arslanalp and Poghosyan (2014), Ebeke and Lu (2014), Asonuma et al (2015)).
    - A handful of studies examine assets’ role but typically focus on net debt or OECD/advanced economies (examples: Ford and Laxton (1999), Conway and Orr (2003), Chinn and Frankel (2005), Gruber and Kamin (2012), Ichiue and Shimizu (2015)).
    - Gruber and Kamin (2012) find a robust and significantly positive impact of net debt on long-term bond yields of OECD countries.
    - Ichiue and Shimizu (2015) find net debt relevant for long-term forward rates for ten AMs but assets are not—consistent with the paper’s findings for AMs (Section V).
    - Hadzi-Vaskov and Ricci (2016), closest to this paper, allow distinct impacts of gross debt and assets for AMs and EMs and find that gross debt and assets have significant effects on spreads that roughly offset each other; they conclude net debt is appropriate for assessing indebtedness effects on spreads.
    - This paper: finds net debt matters for sovereign spreads and probability of default, but effects of gross debt and assets may not offset depending on asset category; for more liquid assets, assets have a larger impact than gross debt on spreads and probability of default.
  - Determinants of probability of financial crises in developing and emerging economies:
    - Studies typically find larger levels of gross debt and international financial volatility lead to higher likelihood of crises in EMs, while stronger fundamentals (adequate reserve coverage, robust growth, lower current account deficits) reduce crisis probability (examples: Manasse et al (2003), Kraay and Nehru (2006), Baldacci et al (2011), Catão and Milesi-Ferretti (2014)).
    - Little attention has been paid to the role of assets (beyond international reserves) in mitigating debt crisis likelihood.
    - Assets and reserves are complementary self-insurance instruments that can serve as collateral and signaling devices to lower borrowing costs, smooth rollover risks and reduce debt distress risk (examples: Aizenman and Marion (2004), Jeanne et al (2011), Bianchi et al (2012), Alfaro and Kanczuk (2013)).
- This paper’s contribution:
  - Controls for international reserves but shows that over and beyond reserves, government financial assets contain useful information for predicting sovereign spreads and default probability.
  - Emphasizes potential distinct effects across asset categories (liquidity matters).

### Methodology: economic rationale and empirical specifications
- Simple economic model (one-period sovereign bonds; risk-neutral international investors):
  - Lender breakeven condition:
    - p_it(θR_it) + (1−p_it)R_it = R_t^*
      - R is sovereign gross borrowing rate; θ is lender’s recovery rate (h ≡ 1−θ is the lender’s haircut).
  - Spread relation (non-linear increasing in default probability and haircut):
    - s_it = h p_it / (1 − h p_it) R_t^*   (Equation (1))
  - Default probability depends on assets and fundamentals:
    - p_it = p(A_it, X_it)   (Equation (2))
  - Two complementary empirical targets: country risk premium (spread) and default probability.
  - Spread equation assumed linear in assets and fundamentals; default probability assumed normal and estimated with probit.
- Financial assets and sovereign risk — baseline panel specification:
  - spread_it = α_i + β Assets_it−1 + Λ′ X_it + e_it   (Equation (3))
    - Λ′ X_it = Λ_1 b_it−1 + Λ′_2 Z_it−1 + Λ_3 W_t
    - b = gross debt-to-GDP ratio; Z = country fundamentals; W = global factors.
    - Z includes: real GDP growth, reserves-to-GDP ratio, current account balance-to-GDP ratio, inflation rate, unexpected shocks to country credit ratings (proxied by residuals from regressing credit ratings on macro fundamentals, U.S. interest rate, and country’s history of default/restructurings).
    - Baseline W is the VIX.
    - Domestic fundamentals are lagged to mitigate endogeneity/reverse causality.
  - Coefficient expectations:
    - Λ_1 expected positive (gross debt raises spreads).
    - β expected negative (assets reduce spreads).
  - Identification strategy allows distinct impacts of gross debt and assets (does not impose Λ_1 = −β).
- Heterogeneity for Emerging Markets:
  - spread_it = α_i + β Assets_it−1 + γ Assets_it−1 EM + δ EM + Λ′ X_it + e_it   (Equation (4))
    - Coefficients β and β + γ capture marginal impact of assets for AMs and EMs respectively.
    - Interaction of EM with gross debt was tested but estimated coefficient was statistically non-significant most of the time.
- Nonlinearity and quantile regression approach:
  - Pooled quantile regressions to allow parameters to vary across the conditional distribution of spreads:
    - Q_Spread_it(τ) = α(τ) + β(τ) Assets_it−1 + Λ(τ)′ X_it   (Equation (5))
    - Q_Spread_it(τ) = α(τ) + β(τ) Assets_it−1 + γ(τ)(Assets_it−1 EM) + δ(τ) EM + Λ(τ)′ X_it   (Equation (6))
    - Coefficients of interest: β(τ) and β(τ) + γ(τ); allows testing whether assets matter more for riskier countries (e.g., 90th percentile) than for safer ones (e.g., median).
- Financial assets and probability of debt distress — pooled probit:
  - P(y=1 | Assets_it, X_it) = Φ(α + β Assets_it + Λ′ X_it)   (Equation (7))
  - P(y=1 | Assets_it, X_it) = Φ(α + β Assets_it + γ (Assets_it EM) + δ EM + Λ′ X_it)   (Equation (8))
    - y = 1 indicates debt distress (default, restructuring, or near-default proxied by IMF financing exceeding one hundred per cent of quota on a commitment basis).
    - Episodes must not be preceded by another episode ending in any of the two previous years.
    - Pooled approach used to retain countries that never experienced distress and to mitigate incidental parameter problem with fixed effects.
- Estimation practices:
  - Panel data techniques with fixed effects and clustered standard errors; OLS for spread equations; pooled quantile regressions; pooled probit for default probability.
  - Assets and debt used as ratios to nominal GDP to avoid loss of observations.

### Data: sample, asset categories, and measurement
- Sample:
  - 110 market-access countries: 30 AMs and 80 EMs (aligned with IMF debt sustainability framework for market-access countries).
  - Asset data collected from several sources inside and outside the IMF; asset categories go back to the late 1980s.
- Financial asset categories (GFSM 2014 definitions; eight asset instruments):
  - A1 Monetary gold and SDRs
  - A2 Currency and deposits
  - A3 Debt securities
  - A4 Loans
  - A5 Equity and investment fund shares
  - A6 Insurance, pension and standardized guarantees (IPSG)
  - A7 Financial derivatives
  - A8 Other accounts receivable
- Aggregated asset groupings and coverage (simple averages across countries for 1980-2015):
  - Total financial assets (A1+A2+A3+A4+A5+A6+A7+A8): 43.8% GDP
  - Held in debt instruments (A1+A2+A3+A4+A6+A8): 24.3% GDP
  - Liquid (A2+A3): 11.4% GDP
  - Highly liquid (A2): 7.3%  (note 1/ indicates may differ from statistical definition because it excludes high-quality securities due to data availability)
- Asset holdings by government units:
  - Central government typically holds bulk of each asset category; remainder held by regional governments and social security funds.
  - Data do not identify sovereign wealth fund holdings separately; for some countries these may be recorded at central government level.
  - Due to concentration at central government level and data limitations, regressions do not control for institutional holdings.
  - Data limitations prevent controlling for encumbered assets or asset availability across government units (e.g., assets held by social security or sovereign wealth funds that may not be immediately available).
- Data sources and measurement choices:
  - Data obtained from OECD, Eurostat, and IMF’s Government Finance Statistics Yearbook (GFSY).
  - For each country, the available data series with the longest time coverage was used.
  - Assets exclude international reserves at the central bank and non-financial assets (buildings, land).
  - Assets may be held domestically or abroad and denominated in domestic or foreign currency; available data lump these together.
  - Additional asset data: IMF’s WEO as reported by IMF country desks; government deposits from monetary surveys reported to IMF’s International Financial Statistics.
  - Debt and assets are used as ratios to nominal GDP.

*Source: wp17173 - Section III discusses our identification strategies. Section IV describes the data and identify (IMF working paper content).*

### Section II provided an overview of overlapping issues with international reserves. For a discussion on non-

### wp17173 - Section II provided an overview of overlapping issues with international reserves. For a discussion on non-

### Data, variables, and methods
- Asset data sources: Eurostat, OECD, GFSY, WEO, IFS; balance-sheet based gross asset data used (i.e., without netting out cross-holdings). Reporting standards not uniform until 2001; more information available upon request.
- Sovereign spreads: JP Morgan EMBI spreads, complemented by long-term spreads (non-EMBI countries use difference with US bond yield for non-European countries or Germany for European countries).
- Signals of distress: Reinhart and Rogoff (2011) (domestic defaults), Das et al (2012b) (official creditor restructurings), Cruces and Trebesch (2013) (private foreign creditor restructurings), Catão and Milesi-Ferretti (2014) (external defaults).
- Main econometric approaches:
  - Panel fixed-effects OLS (Table 6).
  - Pooled quantile regressions (specifications 5 and 6) to capture nonlinearities across spread distribution (Figures 6, A2–A5).
  - Pooled probit regressions for probability of debt distress (Table 7).
- Control variables based on literature survey (Table A1, Table A2): debt/GDP (lagged Λ1), assets/GDP (lagged β), assets/GDP × EM (lagged γ), Real GDP growth (lagged), CAB/GDP (3 year avg., lagged), Reserves/GDP (lagged), Inflation (lagged), Country credit ratings (lagged), VIX, among others.
- Sample and coverage:
  - Country groups: Advanced markets (AMs) and Emerging markets (EMs) listed in Appendix A.
  - Sample period reported: 1980–2015 for descriptive statistics and correlations; regressions use available observations (e.g., Observations: 725 in Table 6; Observations: 712–714 in Table 7 depending on specification).

### Descriptive statistics and patterns
- Asset size and composition:
  - Median and mean asset statistics reported (Table 1): e.g., Mean total assets (AM) 47.4, (EM) 37.1, (All) 43.8 (percent of GDP, obs 543, 291, 834 respectively); Held in debt instruments mean 27.2 (AM), 19.1 (EM), 24.3 (All).
  - Instrument breakdown (median 1980–2015, Table 3): A2 Currency & deposits 7.1 (AM) / 7.6 (EM); A3 Debt securities 5.6 (AM) / 1.1 (EM); A5 Equity & investment fund shares 19.7 (AM) / 17.6 (EM); A8 Other accounts receivable 7.4 (AM) / 7.2 (EM).
  - Distribution skewed to the right: top asset holder (Norway) asset portfolio about ten times larger than the sample mean.
- Cyclicality and crisis dynamics:
  - Assets and gross debt tend to be countercyclical in AMs and roughly acyclical in EMs in the overall sample; spreads are clearly countercyclical in both groups (Table 4).
  - Since the Global Financial Crisis (GFC) assets look procyclical and gross debt countercyclical (Figure 3).
  - Around debt distress episodes (EMs, median, Figure 5): assets typically decline prior to the crisis and are rebuilt thereafter; gross debt increases from about 40 percent of GDP to more than 60 percent of GDP during crises.
- Crisis frequency and duration (Table 5):
  - Number of episodes (1980–2015): AMs 9, EMs 112, All 121.
  - Average duration in years: AMs 3, EMs 5, All 5.
  - Triggers: outright defaults, restructurings, IMF financing, combinations.

### Financial assets and sovereign spreads (main results)
- Panel OLS (Table 6) key patterns:
  - Debt/GDP, lagged (Λ1) has positive and significant coefficients across specifications (e.g., 2.302*** in one specification; see table for multiple exact values).
  - Reserves/GDP, lagged negative and significant (e.g., -2.421**, -2.731***, -2.417***, -2.576*** across specs).
  - Real GDP growth, lagged negative and significant (e.g., -13.68***).
  - Inflation positive and significant (e.g., 14.56***).
  - Country credit ratings negative and significant (e.g., -5.775***).
  - Assets/GDP (β) impact on spreads is less systematic than reserves; only broader categories (total assets and assets held in debt instruments) add information in some specifications.
  - F-test for joint impact of gross debt and assets often not statistically different from zero in some specifications (see F-test values in Table 6).
- Quantile regressions (Figure 6 and appendix figures):
  - Gross debt response to spreads is positive and increasing in country risk (Λ1(τ)).
  - For AMs: marginal impact of assets on spreads β(τ) is either zero or around OLS average—assets less signalling-relevant for AMs.
  - For EMs: sensitivity of spreads to assets (β(τ)+γ(τ)) is larger and statistically significant for countries at higher quantiles of risk and across all asset categories.
  - Quantified example: "a 10-percentage point (pp) increase in assets by an EM at the 90th percentile of risk would reduce sovereign spreads by 60–100bps, compared to 0-50bps for a country around the median of the distribution."
- Economic interpretation:
  - Very liquid assets better mitigate macro shocks, risk-premia, exchange rates, commodity price shocks; can be deployed against liquidity pressures and debt service difficulties in EMs.
  - Asset-liability management: possible to borrow long-term to build liquid buffers; but not a universal strategy—if borrowing costs exceed returns on assets, fiscal deterioration may follow.

### Financial assets and probability of debt distress
- Pooled probit results (Table 7):
  - Gross debt is a key determinant of likelihood of debt crisis across specifications and asset categories (coefficients on Debt/GDP, lagged positive and often significant; e.g., 2.125*** etc. depending on block).
  - Coefficient on assets alone in full specification typically not statistically significant for AMs—assets less important determinant of probability of debt distress in AMs.
  - Liquid and highly liquid assets have economically significant effects in EMs: ceteris paribus, a 10-percentage point increase in both gross debt and assets would reduce crisis probability by 2-4 pp depending on specification.
  - For some specifications and asset categories the marginal effect of assets is large and significant (see marginal effects in Table 7 blocks, e.g., Marginal effect of assets -0.387**, -0.448** in some blocks).
- Liquidity vs solvency crises (Tables 8 and related text):
  - Liquidity crises identified by light restructurings, large spikes in bond spreads (levels above 1000bps or deviations >2 standard deviations), and large IMF programs (above 100 percent of quota); produced 156 liquidity episodes (123 in EMs).
  - Solvency crises proxied by deep debt treatments (outright defaults, Paris Club with face value reduction, commercial debt restructurings with face value reduction/NPV haircuts above sample median), yielding 67 episodes (65 in EMs).
  - Table 8 results: highly liquid assets have larger correlation with liquidity crises than total assets; total assets more correlated with solvency crises. Marginal effects and statistical significance vary by specification (see table for exact coefficients and significance).
- Robustness and rare-events considerations:
  - Broadening definition of debt distress to include spread spikes identifies 149 episodes (33 in AMs, 116 in EMs). Under this broader definition, correlation between gross debt and crisis probability weakens; liquid assets remain statistically significant in EMs; estimates imply a 10-pp increase in debt and assets would reduce crisis probability by about 6 percentage point in EMs.
  - Using WEO asset data (longer span) yields statistically significant coefficients on assets for the overall sample but less differentiated impact for EMs—WEO mixes liquid and less liquid assets across countries.
  - Small-sample/rare-event bias acknowledged: unconditional probability of debt distress in regression sample ~2 percent; tests include alternative estimators (penalized MLE, logistic) and specification robustness checks.

### Robustness checks and sensitivity analysis
- Alternative global risk controls tested: international risk-free rates, term-premia, US corporate high yield—none dominated baseline VIX specification.
- Inclusion of output gap (HP-filter) alongside GDP growth: output gap statistically significant with positive sign in most quantiles; inclusion does not undermine GDP growth nor assets results—marginal impact of assets on spreads similar magnitude, variance somewhat lower (Figure A7).
- Pre-GFC sample (up to 2007): assets in debt instruments and highly liquid assets remain significant at top quantiles, but sample size reduced—caution advised (Figure A8).
- HVR parsimonious specification (Hadzi-Vaskov and Ricci 2016) with WEO assets: gross debt matters at higher quantiles; assets matter across distribution for EMs (Figure A9).
- Probit robustness: adding output gap weakens individual asset coefficient significance but marginal effects generally significant (Table A7); EM-only regressions confirm asset liquidity importance and stronger joint marginal impact of gross debt and assets for liquid categories (Table A8).
- WEO-based regressions and vintage/revision cautions: WEO asset reporting heterogeneous across countries and perimeters; assets often inferred as residual (gross debt minus net debt) and subject to optimistic revisions (Table A5, Figure A1).

### Illustrative simulations and counterfactuals
- Counterfactuals based on baseline quantile and probit regressions for a country with macro fundamentals at EM median, starting gross debt levels of 70 and 90 percent of GDP:
  - Scenario 1 (reduce spreads from 90th percentile 600bps to sample mean 300bps):
    - Required asset holdings depend on asset quality:
      - Highly liquid/liquid categories: around 10-15 percent of GDP.
      - All financial instruments (total assets): around 40-45 percent of GDP.
  - Scenario 2 (reduce likelihood of debt distress to sample mean 5 percent):
    - For initial gross debt 70 percent of GDP: required assets 3.5-7.5 percent of GDP depending on asset category.
    - For initial gross debt 90 percent of GDP: required assets between 7 percent of GDP (liquid categories) and about 25 percent of GDP (less liquid categories).
- Interpretation: asset quality and policy goals matter for asset-liability management design; accumulating buffers likely complements fiscal adjustment and other measures; fiscal savings may be only feasible source for accumulating needed buffers.

### Policy implications and concluding remarks
- Main findings:
  - Assets reduce risks to debt sustainability in EMs; asset liquidity matters—liquid and highly liquid assets more effective in reducing crisis probability and spreads for high-risk EMs.
  - Assets are less systematically relevant for AMs in reducing spreads or distress probability.
  - Nonlinear effects across spread distribution revealed by quantile regressions: assets more valuable at higher spread quantiles for EMs.
- Policy guidance:
  - Countries that accumulate asset buffers could strengthen market access and debt sustainability prospects; the size and profile of asset holdings should be country-specific.
  - Liquid assets are particularly valuable for countries vulnerable to rollover risk or trying to establish market access; long-maturity assets may be relevant for meeting future liabilities.
  - Asset accumulation strategies should be anchored by prudent fiscal policies and strong fiscal institutions; financing assets with debt can be viable only under careful asset-liability management and favorable market conditions.
  - Asset accumulation should complement other crisis insurance instruments: bilateral swap lines, regional financing arrangements, IMF precautionary facilities, and market-based hedging.
- Data and measurement note:
  - Paper excludes non-financial assets due to data limitations and likely limited utility for improving debt sustainability (illiquid, hard to value), referring readers to Bova et al (2013).
  - Existing data limitations (coverage, valuation, reporting standards) constrain definitive prescriptions; improving transparency, coverage and reliability of asset data would be needed to operationalize benchmarks for assets in debt sustainability analyses.

*Italicized source: IMF Working Paper wp17173 (content unit provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17173.pdf_