## wp1893 - References

## Source details

**Canonical URL:** [wp1893 - References](https://www.imf.org/-/media/files/publications/wp/2018/wp1893.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp1893.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp1893.pdf.json)

---

### Introduction and research objective
- Research objective: empirically investigate the effect of financial crises on the government balance sheet, separating impacts on financial assets and liabilities, and explore mechanisms (fiscal stabilization, bank bailouts).
- Sample and data:
  - Quarterly Government Finance Statistics (GFS) stock data for 27 countries (mostly advanced economies), period 1999Q1-2017Q1.
- Econometric approaches:
  - Static fixed effects models.
  - Dynamic fixed effects / ARDL.
  - Panel vector autoregressive (VAR) with impulse response functions (IRFs).
- High-level findings (summary):
  - Financial crises deteriorate government net financial worth, driven by increases in liabilities.
  - No statistically significant contemporaneous decrease in total financial assets — even highly liquid assets — in crisis periods.
  - When financial crises coincide with banking crises, both liabilities and assets increase (consistent with bank bailout processes).
  - Both liabilities and assets are countercyclical; the GDP shock effect is about twice as large on liabilities as on assets.
  - Short-term interest rate and global volatility shocks tend to inflate both assets and liabilities.

### Data, crisis identification, and variables
- Government balance sheet measures:
  - Dependent variable: government net financial worth = financial assets less liabilities (GFSM 2014 definition); nonfinancial assets excluded.
  - GFS asset categories used: (1) monetary gold and SDRs, (2) currency and deposits, (3) debt securities, (4) loans, (5) equity and investment fund shares, (6) insurance, pension, and standardized guarantees (IPSG), (7) financial derivatives, (8) other accounts receivable.
- Asset aggregations by liquidity:
  - Total financial assets: all GFS financial assets.
  - Assets held in debt instruments: exclude equity and derivatives.
  - Liquid financial assets: currency and deposits, and debt securities.
  - Highly liquid assets: currency and deposits.
- Data limitations and exclusions:
  - GFS financial assets exclude central bank international reserves.
  - Total liabilities do not include contingent liabilities.
  - No breakdown by currency, maturity, or counterparty sector available.
- Financial crisis construction:
  - Daily MSCI indices normalized into CMAXidt; crisis dummy when CMAXidt < CMAXit − 2σidt.
  - Crisis intensity: CRISISit = count of days in quarter with crisisidt = 1.
  - Identified episodes: 2008–2009 GFC, 2001–2002 dot-com, European debt crisis.
- Control variables used (sources: WEO, OECD MEI, IFS, FRED):
  - Real GDP growth (annualized % change), inflation (annualized % CPI change), net operating balance (% of GDP), short-term (3-month) and long-term (10-year) government interest rates, nominal effective exchange rate changes, VIX (quarterly average).

### Stylized facts and descriptive statistics (selected)
- Sample observations:
  - Observations: 1999–2007 = 1025; 2008–2017 = 1145; Total = 2170.
- Balance sheet levels (percentage of quarterly GDP; mean (standard deviation)):
  - Net financial worth: 1999-2007 = -0.607 (1.657); 2008-2017 = -1.017 (2.304); Total = -0.823 (2.034).
  - Liabilities: 1999-2007 = 2.182 (1.095); 2008-2017 = 2.906 (1.361); Total = 2.564 (1.294).
  - Financial assets: 1999-2007 = 1.575 (1.024); 2008-2017 = 1.889 (1.621); Total = 1.741 (1.381).
  - Liquid assets*: 1999-2007 = 0.397 (0.349); 2008-2017 = 0.528 (0.510); Total = 0.466 (0.446).
  - Highly liquid assets**: 1999-2007 = 0.257 (0.203); 2008-2017 = 0.323 (0.169); Total = 0.292 (0.189).
  - * ‘Liquid financial assets’ comprise currency and deposits, and debt securities.
  - ** ‘Highly liquid assets’ are currency and deposits.
- First-difference statistics (percentage of quarterly GDP; mean (standard deviation)):
  - Net financial worth: 1999-2007 = 0.0116 (0.0941); 2008-2017 = -0.0213 (0.124); Total = -0.00599 (0.113).
  - Liabilities: 1999-2007 = -0.0118 (0.0903); 2008-2017 = 0.03610 (0.149); Total = 0.0139 (0.128).
  - Financial assets: 1999-2007 = -0.000149 (0.0929); 2008-2017 = 0.01490 (0.123); Total = 0.00788 (0.110).
- Pairwise correlations (selected; significance as in source):
  - First-difference correlation assets–liabilities = 0.56; elasticity controlling for GDP growth ≈ 0.49 (standard error 0.04).
  - First-difference correlation Real GDP growth – liabilities = -0.34; Real GDP growth – assets = -0.23; Real GDP growth – net financial worth = 0.17.
  - Level correlation Net financial worth – Liabilities = -0.742***; Net financial worth – Financial assets = 0.778***.

### Methodology and identification
- Fixed effects baseline:
  - govit = αi + β×crisit + X′it γ + Z′t δ + εit where govit = total liabilities, financial assets, or net financial worth (all % of GDP); crisisit defined via MSCI-based rule.
  - Country fixed effects αi; controls X′it; Z′t = VIX; cluster standard errors at country level; include country-specific quarter dummies.
  - Variables nonstationary in levels → used in first differences.
- Dynamic / ARDL:
  - govit = αi + Σn=0..q βn×crisit−n + Σn=0..q X′it−n γn + Z′t δ + εit to capture dynamics and reduce simultaneity concerns.
- Panel VAR:
  - Yit = B0 + Σn=1..p Bn Yit−n + αi + eit with Yit = [git πit govit iit vixit]′; govit in first difference; GDP and prices in log differences.
  - Lag length p=1 (BIC); Cholesky ordering: output growth, inflation, government balance sheet variable, short-term interest rate, VIX.
  - IRFs to 16 quarters; shocks orthogonalized via Cholesky.

### Estimation results — fixed effects, ARDL, and robustness (selected)
- Baseline fixed effects (selected coefficients; standard errors in parentheses; significance notation preserved in tables):
  - Liabilities (example): Financial crisis = 0.038*** (0.007).
  - Financial assets (example): Financial crisis = 0.012** (0.006).
  - Net financial worth (example): Financial crisis = 0.008 (0.007).
  - Notes: many specifications reported with varying sample sizes (e.g., Observations reported per specification: 1793, 1494; Countries: 27, 23).
- Crisis intensity (CRISISit):
  - Liabilities increase with crisis intensity: significant at 1% without controls; 5% with controls.
  - Financial assets: no significant impact with controls.
- Dynamic fixed effects / ARDL (examples with q=1):
  - Liabilities q=1: Financial crisis = 0.021** (0.009).
  - Net financial worth q=1: Financial crisis = -0.018** (0.007).
  - VIX effect: VIX = 0.002*** on Liabilities (standard errors reported).
- Transmission channels — fiscal stabilization:
  - Interaction crisis × surplusit not statistically significant for balance sheet variables.
  - When surplus positive, governments prefer to decrease liabilities rather than accumulate assets.
- Transmission channels — banking crises and bailouts:
  - Interaction crisis × bankit: when financial crises coincide with banking crises, governments increase both liabilities and financial assets (significant at 1% with country fixed effects; 5% with controls).
  - Interpreted as balance-sheet impact of bailouts (equity injections, takeover of loans).
- Exchange rate depreciation channel:
  - Interaction crisis × depreciationit not statistically significant in this mostly advanced-economy sample.
- Robustness checks:
  - Flow variables: Net incurrence of liabilities (col 1 Table 14): Financial crisis = 0.011** (0.005) in simple regression; significance declines with controls but CRISISit yields 10% significance in some specifications.
  - Debt haircuts: none in sample (per Cruces and Trebesch 2013).
  - Outlier exclusion (Norway and Hungary) yields similar or stronger results.
  - Seemingly unrelated regressions (SUR) confirm baseline fixed effects findings.
  - Banking crisis interactions survive time fixed effects.

### Dynamics — Panel VAR impulse responses (selected)
- Model specifics:
  - Panel VAR p=1; IRFs horizon = 16 quarters; error bands 5% on each side via Monte-Carlo with 1000 replications.
- Response to one-standard-deviation GDP growth shock (Figure 3):
  - Liabilities decline ≈ 4% at impact.
  - Assets decline ≈ 2% at impact.
  - Net financial worth increases ≈ 1.5% after the growth shock.
  - Interpretation: liabilities reaction ≈ twice assets reaction.
- Effect of government balance sheet shocks on GDP (Figure 4):
  - Positive shock in liabilities lowers output growth.
  - Positive shock in financial assets has a positive short-term impact on output growth.
  - Net financial worth improvements associated with higher GDP growth (Granger causality tests confirm).
- Short-term interest rate shock (Figure 6):
  - One-standard-deviation rise in short-term interest rate increases both liabilities and assets by about 0.1%–0.2% per quarter, with effects emerging after ~4 quarters and persisting.
- VIX (global volatility) shock (Figure 7):
  - Global volatility shock increases liabilities by ≈ 1% and assets by ≈ 0.5%; net financial worth unchanged.
- Inflation shock (Figure 5):
  - No significant responses of assets or liabilities within 5% confidence interval.
- General dynamics:
  - Assets and liabilities co-move (symmetry) and exhibit asymmetries in magnitude and stickiness of responses.

### Key quantitative table highlights (selected entries preserved exactly)
- Table 2 (levels; % of quarterly GDP; mean (sd)):
  - Net financial worth: 1999-2007 = -0.607 (1.657); 2008-2017 = -1.017 (2.304); Total = -0.823 (2.034).
  - Liabilities: 1999-2007 = 2.182 (1.095); 2008-2017 = 2.906 (1.361); Total = 2.564 (1.294).
  - Financial assets: 1999-2007 = 1.575 (1.024); 2008-2017 = 1.889 (1.621); Total = 1.741 (1.381).
- Table 3 (first differences; % of quarterly GDP; mean (sd)):
  - Net financial worth: 1999-2007 = 0.0116 (0.0941); 2008-2017 = -0.0213 (0.124); Total = -0.00599 (0.113).
  - Liabilities: 1999-2007 = -0.0118 (0.0903); 2008-2017 = 0.03610 (0.149); Total = 0.0139 (0.128).
  - Financial assets: 1999-2007 = -0.000149 (0.0929); 2008-2017 = 0.01490 (0.123); Total = 0.00788 (0.110).
- Table 6 (example fixed effects coefficients reported in source):
  - Liabilities (col 1): Financial crisis = 0.038*** (0.007).
  - Financial assets (col 2): Financial crisis = 0.012** (0.006).
  - Net financial worth (col 3): Financial crisis = 0.008 (0.007).
- Table 9 (dynamic fixed effects, q=1 examples):
  - Liabilities q=1: Financial crisis = 0.021** (0.009).
  - Net financial worth q=1: Financial crisis = -0.018** (0.007).
- Table 14 (flows):
  - Net incurrence of liabilities (col 1): Financial crisis = 0.011** (0.005).
  - Net acquisition of financial assets (col 3): Financial crisis = 0.007 (0.005).
- Notes across tables:
  - Significance notation: *Significant at 10%. **Significant at 5%. ***Significant at 1%.
  - ‘Financial crisis’ typically defined as national MSCI index decline > two standard deviations below quarterly mean in at least one day of the quarter.

### Conclusions, policy implications, and suggested research
- Conclusions (summary):
  - Financial crises deteriorate net financial worth predominantly via increased liabilities; governments do not appear to liquidate financial assets contemporaneously in crises.
  - Banking crises induce government asset and liability increases consistent with bailout operations; the fiscal value of acquired assets is ambiguous.
  - Liabilities respond about twice as much as assets to GDP shocks; assets are “stickier.”
  - Interest rate and global volatility shocks tend to raise both assets and liabilities.
- Policy implications:
  - Incorporating financial assets into Debt Sustainability Analysis (DSA) is complex because assets and liabilities follow different determinants and dynamics; treating them symmetrically in DSA may mislead projections and policy recommendations.
  - Governments, when improving fiscal balances, tend to reduce liabilities rather than build asset buffers — relevant for debt policy design.
  - Bank bailouts mechanically change balance sheets; the quality and liquidity of acquired assets matter for true fiscal resilience.
  - Data improvements needed: breakdowns by currency, maturity, counterparty sector, and broader inclusion of nonfinancial assets and contingent liabilities would enhance fiscal/financial vulnerability analysis.
- Suggested avenues for future research:
  - Better modeling of government asset dynamics and portfolio allocation decisions.
  - Empirical work leveraging exogenous variation (resource discoveries, migration) to explain asset accumulation.
  - Application of sectoral balance sheet approaches (BSA) and cross-country financial linkage analyses to capture transmission and distributional effects.
- Caveats:
  - Sample mostly advanced economies and period 1999Q1-2017Q1 — results may differ for EMEs.
  - Potential endogeneity and simultaneity addressed via fixed effects, ARDL, and panel VAR/GMM, but identification limitations remain.

*Source: wp1893 - References (excerpt of working paper content, including Sections I–VII, Tables and Figures list).*

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

### wp1893 - References .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### Introduction
- Research objective: empirically investigate the effect of financial crises on the government balance sheet, separating impacts on financial assets and liabilities, and explore mechanisms (fiscal stabilization, bank bailouts).
- Data and sample:
  - Quarterly Government Finance Statistics (GFS) stock data.
  - Sample: 27 countries (mostly advanced economies), period 1999Q1-2017Q1.
- Econometric approaches:
  - Static fixed effects models.
  - Dynamic fixed effects / ARDL.
  - Panel vector autoregressive (VAR) with impulse response functions (IRFs).
- Key high-level findings:
  - Financial crises deteriorate government net financial worth, driven by increases in liabilities.
  - No statistically significant contemporaneous decrease in total financial assets — even highly liquid assets — in crisis periods.
  - When financial crises coincide with banking crises, both liabilities and assets increase (consistent with bank bailout processes).
  - Both liabilities and assets are countercyclical; the GDP shock effect is about twice as large on liabilities as on assets.
  - Short-term interest rate and global volatility shocks tend to inflate both assets and liabilities.

### Literature Review
- Historical context:
  - Debate on gross vs net debt since the 1980s (Boskin 1982; Eisner and Pieper 1984).
  - Early empirical work on net debt (Corsetti and Roubini 1991).
- Key empirical points from literature:
  - Milesi-Ferretti and Moriyama (2006): strong correlation between changes in liabilities and assets; asset decumulation in EU runup to Maastricht.
  - Mixed findings on net debt’s additional explanatory power for borrowing costs (Gruber and Kamin 2012; Ichiue and Shimizu 2015).
  - Emerging market evidence: financial assets reduce sovereign spreads and crisis probability when liquid (Henao-Arbelaez and Sobrinho 2017; Hadzi-Vaskov and Ricci 2016).

### Data Description
- Government balance sheet focus:
  - Analyzed variable: government net financial worth = financial assets less liabilities (GFSM 2014 definition).
  - Excludes nonfinancial assets due to limited, inconsistent coverage and liquidity concerns.
- GFS asset categories used: (1) monetary gold and SDRs, (2) currency and deposits, (3) debt securities, (4) loans, (5) equity and investment fund shares, (6) insurance, pension, and standardized guarantees (IPSG), (7) financial derivatives, (8) other accounts receivable.
- Aggregations by liquidity for analysis:
  - Total financial assets: all GFS financial assets.
  - Assets held in debt instruments: exclude equity and derivatives.
  - Liquid financial assets: currency and deposits, and debt securities.
  - Highly liquid assets: currency and deposits.
- Exclusions and data limitations:
  - GFS financial assets exclude central bank international reserves.
  - Total liabilities do not include contingent liabilities.
  - No breakdown by currency, maturity, or counterparty sector available.
- Financial crisis identification:
  - Daily MSCI indices normalized into CMAXidt (ratio of daily MSCI to quarterly max).
  - Crisis dummy: CMAXidt < CMAXit − 2σidt (two standard deviations below quarterly mean).
  - Crisis intensity: CRISISit = count of days in quarter with crisisidt = 1.
  - Method identifies 2008-2009 GFC, 2001-2002 dot-com bubble, and European debt crisis episodes.
- Control variables (from WEO, OECD MEI, IFS, FRED):
  - Real GDP growth (annualized % change), inflation (annualized % CPI change), net operating balance (% of GDP), short-term (3-month) and long-term (10-year) government interest rates, nominal effective exchange rate changes, VIX (quarterly average).

### Stylized Facts
- Descriptive statistics (Tables 2–3):
  - Average net financial worth is negative for most countries; only Bulgaria, Czech Republic, Estonia, Finland, Norway, and Sweden have a positive mean across the full sample.
  - Post-GFC deterioration: liabilities almost 50% higher in 2008-2017 than in 1999-2007; financial assets also increased but by less than liabilities.
  - Growth rates: liabilities more volatile than assets; both had negative average growth 1999–2007 and positive 2008–2017.
- Pairwise correlations (Tables 4–5, Figure 2):
  - First-difference correlation between assets and liabilities = 0.56; elasticity controlling for GDP growth ≈ 0.49 (standard error 0.04).
  - GDP growth correlations: liabilities and assets are countercyclical; correlation with first difference of liabilities = -0.34, with assets = -0.23; net financial worth correlation with GDP growth = 0.17 (weakly procyclical).
  - Net operating balance correlates more with decreases in liabilities (-0.30) than with increases in assets (0.06).
  - Short-term interest rate correlation with net financial worth ≈ 0; long-term interest rate correlation with net financial worth = -0.08 (liabilities correlate 0.09).
  - Global volatility associated with decreases in net financial worth as increases in liabilities exceed increases in assets.

### Methodology
- Fixed effects baseline:
  - Equation (1): govit = αi + β×crisit + X′it γ + Z′t δ + εit where govit = total liabilities, financial assets, or net financial worth (all % of GDP); crisisit defined as above.
  - Country fixed effects αi; controls X′it (growth, inflation, net operating balance, short- and long-term rates, exchange rate changes); Z′t = VIX; cluster standard errors at country level; include country-specific quarter dummies.
  - Variables nonstationary in levels → used in first differences. Hausman test favors fixed effects.
- Dynamic / ARDL specification:
  - Equation (2): govit = αi + Σn=0..q βn×crisit−n + Σn=0..q X′it−n γn + Z′t δ + εit to capture dynamic responses and reduce simultaneity concerns.
- Panel VAR:
  - Equation (3): Yit = B0 + Σn=1..p Bn Yit−n + αi + eit with Yit = [git πit govit iit vixit]′; govit in first difference; GDP and prices in log differences.
  - Lag length p=1 (Bayesian information criterion).
  - Cholesky ordering: output growth, inflation, government balance sheet variable, short-term interest rate, VIX (VIX reacts contemporaneously to all).
  - IRFs computed to 16-quarter horizon; shocks orthogonalized via Cholesky.

### Estimation Results — Fixed Effects and Robustness
- Baseline fixed effects (Table 6):
  - Crisis dummy significantly deteriorates net financial worth:
    - without controls: significant at 1% level.
    - with controls: significant at 10% level.
  - Increase in liabilities in crisis periods:
    - significant at 5% level with controls.
  - No significant contemporaneous impact on total financial assets or on assets by liquidity class (Table 7).
  - Interpretation: governments tend to increase debt rather than liquidate assets in stock market crises; possible demand-side and supply-side constraints on asset sales; asset portfolio composition matters but is hard to assess in aggregate data.
- Intensive-margin crisis (Table 8):
  - Using CRISISit (number of crisis days in quarter), liabilities increase with crisis intensity (significant at 1% without controls; 5% with controls); no significant impact on financial assets with controls.
- Dynamic fixed effects / ARDL (Table 9):
  - Results hold with lags: crisis increases liabilities and worsens net financial worth (both significant at 5%); no significant effect on financial assets.
- Transmission channels — fiscal stabilization (Table 10):
  - Interaction of crisis with surplusit (net operating balance positive) not statistically significant for balance sheet variables.
  - When surplus positive, governments prefer to decrease liabilities rather than accumulate assets (consistent with preference to reduce higher-cost debt).
- Transmission channels — banking crises and bailouts (Tables 11–12):
  - Interaction crisis × bankit (banking crisis dummy) shows that when financial crises coincide with banking crises, governments increase both liabilities and financial assets (significant at 1% with country fixed effects; 5% with controls).
  - Interpretation: bank bailouts (equity injections, takeover of loans) inflate government assets and liabilities; effect on fiscal sustainability ambiguous.
- Transmission channels — exchange rate depreciation (Table 13):
  - Interaction crisis × depreciationit (nominal effective exchange rate fall larger than country average) not statistically significant; no evidence that exchange rate depreciation drives crisis effects on assets/liabilities in this mostly AE sample.
- Robustness checks:
  - Flow variables (net incurrence of liabilities, net purchase of financial assets): crisis effect on net incurrence of liabilities significant at 5% in simple regression; loses significance with controls (but significant at 10% using CRISISit).
  - Debt haircuts: none in sample (per Cruces and Trebesch 2013).
  - Outliers: excluding Norway and Hungary (Table 15) yields similar or stronger results.
  - Seemingly unrelated regressions (SUR, Table 16) confirm baseline fixed effects findings.
  - Fixed effects with time dummies not reported for crisis dummy because global crises overlap with time dummies; banking crisis interaction survives time fixed effects.

### Dynamics — Panel VAR / IRFs
- Model details:
  - Panel VAR with p=1 lag; eigenvalue stability condition satisfied; IRFs for 16 quarters (4 years).
- Responses to a one-standard-deviation GDP growth shock (Figure 3):
  - Both liabilities and financial assets decrease; liabilities decline ≈ 4% at impact, assets decline ≈ 2% at impact (liabilities effect ≈ twice assets effect).
  - Net financial worth increases ≈ 1.5% after the growth shock (weakly procyclical).
- Effect of government balance sheet shocks on GDP (Figure 4):
  - Positive shock in liabilities lowers output growth.
  - Positive shock in financial assets has a positive short-term impact on output growth.
  - Net financial worth improvements are associated with higher GDP growth (Granger causality tests confirm).
- Responses to inflation shock (Figure 5):
  - No significant responses of assets or liabilities within 5% confidence interval.
- Responses to short-term interest rate shock (Figure 6):
  - A one-standard-deviation rise in short-term interest rate increases both liabilities and assets by about 0.1%–0.2% per quarter, with effects emerging after ~4 quarters and persisting.
- Responses to VIX shock (Figure 7):
  - Global volatility shock increases liabilities by ≈ 1% and assets by ≈ 0.5%; net financial worth unchanged.
- General dynamic conclusion: assets and liabilities have both symmetries (co-movement, countercyclicality) and asymmetries (magnitude of responses, stickiness of assets).

### Conclusions and Policy Implications
- Summary conclusions:
  - Financial crises deteriorate net financial worth predominantly via increased liabilities; no evidence governments liquidate financial assets contemporaneously in crises.
  - Banking crises induce government asset and liability increases consistent with bailout operations, producing an “artificial” net worth improvement of ambiguous fiscal value.
  - Liabilities respond about twice as much as assets to GDP shocks; assets are “stickier.”
  - Interest rate and global volatility shocks tend to raise both assets and liabilities.
- Policy implications and cautions:
  - Incorporating financial assets into Debt Sustainability Analysis (DSA) is complex: assets and liabilities follow different determinants and dynamics; treating them symmetrically in DSA may mislead projections and policy recommendations.
  - Governments, when improving fiscal balances, tend to reduce liabilities rather than build asset buffers — relevant for debt policy design.
  - Bank bailouts change balance sheets mechanically; quality and liquidity of acquired assets matter for true fiscal resilience.
  - Further data improvements needed: breakdowns by currency, maturity, counterparty sector, and broader inclusion of nonfinancial assets and contingent liabilities would enhance fiscal/financial vulnerability analysis.
- Suggested avenues for future research:
  - Better modeling of government asset dynamics and portfolio allocation decisions.
  - Empirical work leveraging exogenous variation (resource discoveries, migration) to explain asset accumulation.
  - Application of sectoral balance sheet approaches (BSA) and cross-country financial linkage analyses to capture transmission and distributional effects.
- Caveats:
  - Sample mostly advanced economies and period 1999Q1-2017Q1 — results may differ for EMEs.
  - Potential endogeneity and simultaneity addressed via fixed effects, ARDL, and panel VAR/GMM, but identification limitations remain.

*Source: wp1893 - References (excerpt of working paper content, including Sections I–VII, Tables and Figures list).*

### References

### References

### Cited literature (selected entries from the list)
- Abrigo, M. R. and Love, I. (2015). Estimation of Panel Vector Autoregression in Stata: A Package of Programs. University of Hawaii Working Paper.
- Afonso, A., Baxa, J., and Slavík, M. (2017). Fiscal Developments and Financial Stress: A Threshold VAR Analysis. Empirical Economics, pages 1–29.
- Allen, M., Setser, B., Keller, C., Rosenberg, C. B., and Roubini, N. (2002). A Balance Sheet Approach to Financial Crisis. IMF Working Paper No. 13/95.
- Blanchard, O. and Perotti, R. (2002). An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output. The Quarterly Journal of Economics, 117(4):1329–1368.
- Bova, E., Dippelsman, R. J., Rideout, K., and Schaechter, A. (2013). Another Look at Governments’ Balance Sheets: The Role of Nonfinancial Assets. IMF Working Paper No. 13/95.
- Dippelsman, R., Dziobek, C. H., and Gutiérrez Mangas, C. (2012). What Lies Beneath: The Statistical Definition of Public Sector Debt. IMF Staff Discussion Notes No. 12/09.
- Laeven, L. and Valencia, F. (2012). Systemic Banking Crises Database: An Update. IMF Working Paper No. 12/163.
- Milesi-Ferretti, G. M. (2004). Good, Bad or Ugly? On the Effects of Fiscal Rules with Creative Accounting. Journal of Public Economics, 88(1-2):377–394.
- Reinhart, C. M., Rogoff, K. S., and Savastano, M. A. (2003). Debt Intolerance. NBER Working Paper 9908.
- Roodman, D. (2009). How to Do xtabond2: An Introduction to Difference and System GMM in Stata. Stata Journal, 9(1):86–136.
- International Monetary Fund (2014). Government Financial Statistics Manual 2014. Washington, D.C.
- International Monetary Fund (2016). Analyzing and Managing Fiscal Risks – Best Practices. Washington, D.C.

### Types of sources represented
- IMF Working Papers and Staff Discussion Notes.
- NBER and University working papers.
- Peer-reviewed journal articles (e.g., The Quarterly Journal of Economics, Empirical Economics, Journal of Public Economics).
- Statistical manuals and methodological guides (e.g., European System of Accounts – ESA 2010; System of National Accounts).
- Conference papers, memos, and mimeos.

### Appendix — Tables (selected quantitative highlights)
- Table 1: List of countries in the sample (includes Austria, Belgium, Brazil*, Bulgaria, Canada, Czech Republic, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, Romania, Slovak Republic, Spain, Sweden, United Kingdom, United States*). Note: * Government balance sheet (quarterly) data for these countries start in 2001Q1.
- Table 2: Descriptive statistics – Balance sheet variables in level (all values are in percentage of quarterly GDP; mean coefficients with standard deviations in parentheses).
  - Net financial worth: 1999-2007 = -0.607 (1.657); 2008-2017 = -1.017 (2.304); Total = -0.823 (2.034)
  - Liabilities: 1999-2007 = 2.182 (1.095); 2008-2017 = 2.906 (1.361); Total = 2.564 (1.294)
  - Financial assets: 1999-2007 = 1.575 (1.024); 2008-2017 = 1.889 (1.621); Total = 1.741 (1.381)
  - Financial assets held in debt instruments: 1999-2007 = 1.572 (1.023); 2008-2017 = 1.878 (1.618); Total = 1.733 (1.378)
  - Liquid assets*: 1999-2007 = 0.397 (0.349); 2008-2017 = 0.528 (0.510); Total = 0.466 (0.446)
  - Highly liquid assets**: 1999-2007 = 0.257 (0.203); 2008-2017 = 0.323 (0.169); Total = 0.292 (0.189)
  - Observations: 1999-2007 = 1025; 2008-2017 = 1145; Total = 2170
  - * ‘Liquid financial assets’ comprise currency and deposits, and debt securities.
  - ** ‘Highly liquid assets’ are currency and deposits.
- Table 3: Descriptive statistics – Balance sheet variables in first difference (all values are in percentage of quarterly GDP; mean coefficients with standard deviations in parentheses).
  - Net financial worth: 1999-2007 = 0.0116 (0.0941); 2008-2017 = -0.0213 (0.124); Total = -0.00599 (0.113)
  - Liabilities: 1999-2007 = -0.0118 (0.0903); 2008-2017 = 0.03610 (0.149); Total = 0.0139 (0.128)
  - Financial assets: 1999-2007 = -0.000149 (0.0929); 2008-2017 = 0.01490 (0.123); Total = 0.00788 (0.110)
  - Financial assets held in debt instruments: 1999-2007 = -0.000228 (0.0923); 2008-2017 = 0.01460 (0.122); Total = 0.00768 (0.110)
  - Liquid assets*: 1999-2007 = 0.00303 (0.0574); 2008-2017 = 0.00404 (0.0817); Total = 0.00357 (0.0714)
  - Highly liquid assets**: 1999-2007 = 0.000393 (0.0487); 2008-2017 = 0.00292 (0.0721); Total = 0.00175 (0.0623)
  - Observations: 1999-2007 = 1025; 2008-2017 = 1145; Total = 2170
- Table 4 and Table 5: Correlation matrices (level and first difference). Selected entries (significance noted as in source):
  - Level: Correlation Net financial worth – Liabilities = -0.742***; Net financial worth – Financial assets = 0.778***; Real GDP growth – Net financial worth = 0.0623***.
  - First difference: Correlation Net financial worth – Liabilities = -0.584***; Net financial worth – Financial assets = 0.345***; Real GDP growth – Net financial worth = 0.169***.
- Tables 6–16: Regression results (fixed effects, dynamic fixed effects, interaction specifications, robustness checks).
  - Table 6 (Fixed effects regressions): Financial crisis coefficients across dependent variables (examples):
    - Liabilities (col 1): Financial crisis = 0.038*** (0.007)
    - Financial assets (col 2): Financial crisis = 0.012** (0.006)
    - Net financial worth (col 3): Financial crisis = 0.008 (0.007)
    - Leverage (col 4): Financial crisis = -0.000 (0.007)
    - Observations reported per specification: 1793, 1494, 1793, 1494, 1793, 1494, 1793, 1494
    - Countries: 27, 23, 27, 23, 27, 23, 27, 23
    - R2 adjusted examples: 0.364, 0.518, 0.144, 0.175, 0.222, 0.296, 0.249, 0.312
  - Table 9 (Dynamic fixed effects regressions): Number of lags q = 1, 2, 4 reported; Financial crisis short-run coefficients (examples):
    - Liabilities q=1: Financial crisis = 0.021** (0.009)
    - Net financial worth q=1: Financial crisis = -0.018** (0.007)
    - VIX effects: VIX = 0.002*** on Liabilities (standard errors reported)
    - Observations: 1484, 1462, 1419 depending on lag; Countries = 23
  - Table 14 (Government flows): Financial crisis coefficients:
    - Net incurrence of liabilities (col 1): Financial crisis = 0.011** (0.005)
    - Net acquisition of financial assets (col 3): Financial crisis = 0.007 (0.005)
    - Observations: 1814, 1507; Countries: 27, 23, 27, 23
  - Robustness specifications include interactions with fiscal balance, banking crises (short and long lengths), large exchange rate depreciation, exclusion of outlier countries (Norway and Hungary), seemingly unrelated regressions, and panel clustering of standard errors.
- Notes across tables:
  - Significance notation: *Significant at 10%. **Significant at 5%. ***Significant at 1%.
  - ‘Financial crisis’ is typically defined as a dummy equal to one when the national MSCI index experienced a decline in prices of more than two standard deviations below the quarterly mean in at least one day of the quarter, except where an alternative (e.g., crisis intensity) is specified.
  - Many regressions use country fixed effects, control for seasonality, and cluster standard errors at the country level.

### Appendix — Figures (methodological notes on IRFs)
- Figures 3–7 (Impulse response functions and GDP responses) share estimation details:
  - IRFs are estimated using GMM on the system of equations, which includes real GDP, price level, government balance sheet variables, short-term interest rate, and the VIX index.
  - One lag is used in the reported IRFs (except where panel lags are varied in dynamic regressions).
  - Controls: country fixed effects, seasonal dummies, and episodes of financial crisis.
  - Standard errors are clustered at the country level.
  - Shocks: One standard deviation shocks are orthogonalized recursively according to Cholesky decomposition (for GDP shock and other shocks as specified).
  - Error bands are 5% on each side and are generated by Monte-Carlo simulations with 1000 replications.

*Content derived from “wp1893 - References” (appendix tables and figures notes) as provided.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp1893.pdf_