## Box 2.2.1. Nonbanks Financial Institutions and the Sovereign-Bank Nexus in Emerging Markets

## Source details

**Canonical URL:** [Box 2.2.1. Nonbanks Financial Institutions and the Sovereign-Bank Nexus in Emerging Markets](https://www.imf.org/-/media/files/publications/gfsr/2022/april/english/ch2onlineannex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2022/april/english/ch2onlineannex.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2022/april/english/ch2onlineannex.pdf.json)

---

### Role and emerging patterns of NBFIs in EM government bond markets
- Domestic NBFIs hold almost one-third of total government debt as of end-2020.
- In some countries the share of domestic NBFI holdings exceeds 40 percent: Chile, India and Mexico.
- Foreign NBFIs’ share in EM government debt has almost doubled over the last decade and can amount to over half of government debt in some cases (example cited: Peru).
- The stock of external financing provided by NBFIs in EMs grew from around 12 percent of GDP in 2010 to 20 percent of GDP in 2020.
- NBFIs can diversify investor base and improve liquidity but structural vulnerabilities and procyclical behavior can amplify vulnerabilities and shock transmission through the sovereign-bank nexus.

### Channels of vulnerability and observed behavior during stress
- Large foreign nonbank holdings increase EM vulnerability to global capital market gyrations because foreign nonbank investors tend to be particularly fickle during global financial market stress.
- Sudden retrenchment by nonbank investors can trigger sharp price corrections of government debt and large currency depreciations, imposing substantial losses on banks’ trading books.
- Mutual funds and ETFs investing in EM assets acted highly pro-cyclically during the March 2020 turmoil, experiencing large outflows, especially those invested in local currency government bonds.
- Some NBFIs may sell government securities in stress to bolster liquidity or meet margin calls (example: “dash for cash” by U.S. fixed-income funds at the onset of the COVID-19 pandemic).
- Use of FX swaps to hedge currency mismatches is an additional risk transmission channel; widening FX swap basis during market stress increases hedging costs and risks from currency mismatches.

### Quantitative evidence on spillovers and NBFI influence
- A sample of 11 EMs is used to examine spillovers; during periods of strained financial conditions, movements in the default risk of nonbanks spill over to other sectors of the economy (spillover index constructed via Diebold and Yilmaz (2014) VAR-based forecast error variance decomposition; index ranges from 0 to 100).
- Spillovers from NBFIs to other sectors, especially banking and corporate sectors, increased during the pandemic.

### Macroeconomic drivers of banks’ sovereign holdings (country-level analysis)
- Empirical sample: 23 EMs over 2000-2020; estimation via ordinary least squares with country and time fixed effects and clustered standard errors.
- Key empirical findings:
  - An increase by around 6 percentage points (corresponding to 1 standard deviation) in the interest rate leads to about 1.2 percentage points increase in bank holdings of sovereign debt relative to their assets.
  - Banks hold more government debt when there are fewer private sector lending opportunities (lower private sector credit-to-GDP ratio and lower stock market capitalization-to-GDP).
  - An increase of around 21 percentage points in the gross public debt-to-GDP ratio (about 1 standard deviation) is associated with about a 5 percentage points increase in bank holdings of sovereign debt.

### Moral suasion and risk shifting (bank-level analysis)
- Bank-level sample: about 4,000 banks from 38 countries over 2011-2020; regression excludes small and extreme outliers.
- Identification:
  - Moral suasion: state-owned banks (banks with >25 percent government ownership) as control group; high fiscal need defined as expected maturing debt to total public debt ratio above the country-specific 75th percentile.
  - Risk shifting: interaction of state-ownership with bank capital ratio (total equity-to-total assets), focusing on less-capitalized banks and periods of sovereign distress.
  - Sovereign distress defined as CDS > 500 bps, or S&P long-term rating for sovereign FX debt is CCC- or lower, or government in external or domestic default.
- Key empirical results (selected estimates preserved exactly):
  - State Owned: 10.06 (7.10), 9.99 (6.85), 26.80*** (8.85), 25.88* (14.20)
  - High Need x State Owned: 9.81* (5.85), 9.88* (5.85), 26.34* (14.02), 25.75* (13.09)
  - Capital Ratio x State Owned: -6.63 (5.45), -4.95 (5.34), -21.64*** (8.21), -24.28*** (8.93)
  - Capital Ratio: 8.92*** (2.55), 5.31* (2.72), 17.80** (7.53), 4.35 (7.73)
- Magnitudes and patterns:
  - State-owned banks’ net purchases of domestic sovereign bonds are around 10 percentage points higher than private banks during times of high fiscal need.
  - During periods of high fiscal need when the sovereign is in distress, state-owned banks’ net purchases are 26 percentage points higher than private banks.
  - During sovereign distress, net purchases of less-capitalized state-owned banks are over 20 percentage points higher (capital ratio transformed to its standard deviation).
- Non-linear thresholds:
  - Additional purchases by state-owned banks during high fiscal need increase from 26 to 50 percentage points when the sovereign CDS threshold increases from 500 bps to 1000 bps.
  - Less-capitalized state-owned banks’ additional purchases increase to around 40 percentage points when the CDS threshold increases from 500 bps to 1000 bps.

### Policy implications and recommendations (focused)
- Broadening the reach of macroprudential tools to cover NBFIs involved in credit intermediation activities that might pose bank-like financial stability risks and/or regulatory arbitrage.
- Strengthening the resilience of NBFIs to reduce procyclicality and amplification of shocks.
- Enhancing transparency and data provision on NBFIs’ activities in government bond markets.
- Appropriate regulation of NBFIs in both advanced and emerging market economies to help lessen cross-border contagion risks that are increasingly material to financial stability in EMs and developing economies.

---

### Online Annex 2.5. Measuring the Strength of the Nexus (SVAR and panel local projections)

### Two-step empirical approach
- Step 1: Country-specific SVAR models estimate risk transmission between sovereign, banking, and corporate sectors. Endogenous vector y_t includes sovereign risk, bank credit risk, non-financial corporate risk, term spread and equity prices. Sovereign, bank, and corporate credit risks captured by sectoral expected default frequency (EDF). Model estimated in first differences.
- Step 2: Panel analysis (panel local projection methodology) assesses how fiscal and financial vulnerabilities amplify the nexus after an adverse shock (e.g., tightening in global financial conditions).

### Identification and estimation details
- Structural identification exploits heteroskedasticity following Rigobon (2003); five volatility regimes used (regimes classified per Rigobon and Sack (2003) using an 8-day rolling standard deviation threshold of 1 for residuals).
- Robustness: main results tested with four regimes and different lag structures (two to five lags); results broadly consistent.
- Diagnostics: Portmanteau tests for lags 1 to 5 do not reject no autocorrelation at the 10 percent level for all variables except stock returns.

### Volatility-regime days (selected country entries preserved exactly)
- ARG: Δ Sovereign EDF 22; Δ Bank EDF 99; Δ NFCs EDF 75; Term spread 0; Δ equity prices 146; No volatility change 2,814
- BRA: 67; 136; 230; 0; 294; 2,957
- CHN: 13; 10; 13; 0; 32; 4,019
- IND: 5; 70; 21; 87; 87; 2,789
- TUR: 55; 53; 223; 15; 64; 3,712
- ZAF: 0; 65; 54; 365; 243; 3,395
- Note: EDF = expected default frequency.

### Panel local projection model and vulnerability interactions
- Model projects endogenous variables at horizons h and includes interactions between an index of global financial conditions (GFC) and pre-shock vulnerabilities (e.g., high public debt-to-GDP, high banks’ government bond holdings-to-total assets).
- High vulnerability defined as one standard deviation above the sample mean for the metric used.
- Model estimated at quarterly frequency.

### Quantitative result on global financial shocks
- A tightening in global financial conditions corresponding to half of the magnitude observed during the March 2020 financial market turmoil is associated with, on average, a 0.4 percentage point increase in corporate default risk.
- The effect is persistent, lasting at least four quarters after the shock.
- Baseline model without interaction effects shows banks and corporates are the most affected following such a tightening.

---

### Change in Credit Risks and Bank Exposure Exercises (SVAR, bank-level panels, and exposure channel)

### Data and scope highlights
- SVAR analysis: daily data for 15 major EMs over 2006-2020.
- Panel model: quarterly data from 2006 to 2020.
- Banks panel exercise: final data includes 525 domestic banks from 18 EMs covering 1998 to 2020; after cleaning about 94 percent of total assets relative to the raw dataset is preserved.
- Sovereign EDF is CDS-implied; banking and corporate EDFs are simple averages of individual institutions in a country.

### Empirical methodology (exposure channel)
- Baseline panel regressions estimate changes in bank EDF, pre-tax profits divided by lagged total equity, log change in total equity, change in gross loans-to-total assets, log change in total gross loans, log change in total government debt holdings.
- Key regressors include sovereign exposure (government bond holdings divided by total assets in t-1), sovereign distress indicator (sCDS > 500 bps or default or S&P FX rating CCC- or lower), capital ratio, and bank controls.
- Fixed effects: country-year and bank fixed effects; OLS with bank-clustered standard errors.

### Empirical results — exposure channel (key magnitudes preserved)
- Bank EDF:
  - An ex-ante 10-percentage-points higher government debt holdings-to-total assets ratio implies a greater than 0.4 percentage points increase in EDF when sovereign CDS premium is above 500 bps.
  - Effects are twice larger for less capitalized banks.
  - Results more robust for more severe distress thresholds (e.g., 1000 bps).
- Bank profits:
  - Banks with ex-ante 10-percentage-points-higher government debt holdings-to-total assets ratio experience a 5.5 percentage points decline in profits-to-lagged-equity ratio following a severe sovereign distress (sCDS > 1000 bps).
  - Less capitalized banks experience an 8.5 percentage points decline.
- Bank equity:
  - Banks with 10-percentage-points higher government debt holdings experience about a 6 percent decline in total equity following a severe sovereign distress (sCDS > 1000 bps).
  - Using a stricter distress definition (annual average sCDS above threshold) yields stronger effects (6-10 percent decline in equity).
- Bank lending:
  - A 10-percentage-points higher government debt holdings is associated with 2.5 percentage-points-lower loans-to-total-assets ratio.
  - Effect for less capitalized banks: 4 percentage points lower loans-to-total-assets ratio.
  - Total loan growth declines by 3.5 to 5 percent for average and less-capitalized banks, respectively, during severe sovereign distress events.
- Bank government debt holdings during distress:
  - Banks with ex-ante higher government debt holdings tend to increase their holdings during sovereign distress, especially when sCDS rises above 600 bps.
  - Increase in holdings is similar or lower for less capitalized banks—no strong evidence of risk-shifting by weakly capitalized banks.
- International reserves adequacy:
  - A 10-percentage-point higher share of government debt holdings in banks domiciled in countries with low reserve adequacy leads to a nearly 5.5 percent additional decline in bank equity after a severe sovereign distress event (sCDS > 1000 bps).
  - Reserves adequacy indicator: short-term-external-debt-to-international-reserves ratio above the sample average in the year prior to distress (on average close to 0.5).

### Robustness and endogeneity checks (selected)
- Alternative sovereign distress proxies used, including exchange-rate-induced fiscal debt increase and maturing debt volume indicators.
- Findings under robustness checks:
  - Following exchange-rate-induced fiscal debt increase, exposed banks experience higher perceived EDF and lower profitability, equity, and lending (not always statistically significant).
  - When global risk aversion is high (higher VIX), exposed banks in high rollover-risk countries experience lower profits, lower equity, and lower loans-to-total-assets ratio; effects on bank default probabilities increase but are not significant.
- Results robust to controls for expected real GDP growth (WEO vintages), exchange rate depreciation interacted with sovereign distress, and sensitivity of bank stock returns to sovereign yields.
- Data limitations: foreign-currency fiscal debt assumed denominated in US dollars for exchange-rate shock exercise; lending to government, receivables from government, and state guarantees not directly captured due to data limitations.

### Policy-relevant implications (exposure channel)
- Sovereign distress can materially increase bank default risk (EDF), reduce bank profitability and equity, and constrain lending — effects amplified for banks with higher sovereign bond holdings and lower capital ratios.
- Strengthening bank capital buffers reduces amplification of sovereign-bank feedbacks; less-capitalized banks face substantially larger negative impacts.
- International reserve buffers matter: countries with lower reserves adequacy experience larger equity losses in banks holding more sovereign debt.
- Monitoring bank sovereign exposures and rollover risks is critical, particularly during episodes of heightened global risk aversion (VIX spikes).
- Policy tools to mitigate transmission include enhancing bank capital, limiting concentrated sovereign exposures, and maintaining adequate international reserves.

---

### Sovereign downgrades, bound firms, and banking-sector spillovers

### Identification and data (firm-level analysis)
- Difference-in-differences comparing “bound” firms (firms with a rating equal to or above their sovereign rating prior to the downgrade) to “unbound” firms (firms rated below their sovereign) around sovereign downgrades.
- Data: consolidated firm data from S&P Capital IQ; 84 unique sovereign downgrade events in 29 EMs; sample includes 717 firms; bound firms constitute 20.4 percent of the sample.

### Key findings on firms’ investment and debt issuance
- Bound firms cut investment and debt issuance more than unbound peers after sovereign downgrades.
- Cumulative change in investment ratio remains statistically significant at 3 years after downgrade.
- A stricter bound definition increases estimated impact by 4.5 percentage points relative to baseline.
- Matching and difference-in-differences estimation confirm bound firms cut investment and debt issuance by around 17 percentage points more than their peers two years after a downgrade.

### Indirect impact on banks via firms and NPLs
- Empirical exercise (sample: 25 EM countries from 1995 to 2021) shows that in countries with a more dominant presence of bound firms in the corporate sector, banks’ NPL ratio increases after the sovereign downgrade and the effect persists at least two years after the downgrade.
- Results robust to controls including country-median debt affordability of small and mid-size firms and lagged NPL ratio.
- Small and mid-size companies defined as firms with number of employees equal or below 250.

### Policy relevance
- Sovereign downgrades transmit to corporate-sector distress, reducing investment and debt issuance among bound firms and raising bank credit risk via higher NPLs.
- Monitoring the share of bound firms in corporate-sector assets and policies that support corporate balance-sheet resilience can help limit banking-sector spillovers following sovereign downgrades.

*Authors of the box: Andrea Deghi and Mustafa Yasin Yenice. Source: Global Financial Stability Report Chapter 2 (April 2022).*

### Box 2.2.1. Nonbanks Financial Institutions and the Sovereign-Bank Nexus in Emerging Markets

### Box 2.2.1. Nonbanks Financial Institutions and the Sovereign-Bank Nexus in Emerging Markets

### Role and emerging patterns of NBFIs in EM government bond markets
- Nonbank financial institutions (NBFIs) can diversify the investor base, improve liquidity, and weaken banks’ sovereign exposure, but structural vulnerabilities and procyclical behavior can amplify vulnerabilities and shock transmission through the sovereign-bank nexus.
- As of end-2020, domestic NBFIs hold almost one-third of total government debt (Figure 2.2.1, panel 1).
- In some countries the share of domestic NBFI holdings exceeds 40 percent: Chile, India and Mexico (Figure 2.2.1, panel 2).
- Foreign NBFIs’ share in EM government debt has almost doubled over the last decade and can amount to over half of government debt in some cases (example cited: Peru).
- The stock of external financing provided by NBFIs in EMs grew from around 12 percent of GDP in 2010 to 20 percent of GDP in 2020.

### Channels of vulnerability and observed behavior during stress
- Large foreign nonbank holdings can increase EM vulnerability to global capital market gyrations because foreign nonbank investors tend to be particularly fickle during global financial market stress.
- Sudden retrenchment by nonbank investors can trigger sharp price corrections of government debt and large currency depreciations, imposing substantial losses on banks’ trading books.
- Mutual funds and ETFs investing in EM assets acted highly pro-cyclically during the March 2020 turmoil, experiencing large outflows, especially those invested in local currency government bonds (Figure 2.2.1, panel 4).
- Some NBFIs may sell government securities in stress to bolster liquidity or meet margin calls, contributing to pressure in government bond markets (e.g., “dash for cash” by U.S. fixed-income funds at the onset of the COVID-19 pandemic).
- Use of FX swaps to hedge currency mismatches can be an additional risk transmission channel; widening FX swap basis during market stress increases hedging costs and risks from currency mismatches (noted occurrence during the outbreak of the Covid-19 crisis).

### Quantitative evidence on spillovers and NBFI influence
- A sample of 11 EMs is used to examine spillovers; during periods of strained financial conditions, movements in the default risk of nonbanks spill over to other sectors of the economy (spillover index constructed via Diebold and Yilmaz (2014) VAR-based forecast error variance decomposition; index ranges from 0 to 100).
- Spillovers from NBFIs to other sectors, especially banking and corporate sectors, increased during the pandemic (Figure 2.2.1, panel 6).

### Macroeconomic drivers of banks’ sovereign holdings (country-level analysis)
- Empirical sample: 23 EMs over 2000-2020; estimation via ordinary least squares with country and time fixed effects and clustered standard errors.
- Key empirical findings:
  - Banks tend to hold more government debt when interest rates are high: an increase by around 6 percentage points (corresponding to 1 standard deviation) in the interest rate leads to about 1.2 percentage points increase in bank holdings of sovereign debt relative to their assets.
  - Banks hold more government debt when there are fewer private sector lending opportunities (lower private sector credit-to-GDP ratio and lower stock market capitalization-to-GDP).
  - Bank holdings increase when the sovereign is more indebted: an increase of around 21 percentage points in the gross public debt-to-GDP ratio (about 1 standard deviation) is associated with about a 5 percentage points increase in bank holdings of sovereign debt.

### Moral suasion and risk shifting (bank-level analysis)
- Bank-level sample: about 4,000 banks from 38 countries over 2011-2020; regression excludes small and extreme outliers (see sample selection rules).
- Identification strategy:
  - Moral suasion: state-owned banks (banks with >25 percent government ownership) are the control group; high fiscal need defined as expected maturing debt to total public debt ratio above the country-specific 75th percentile.
  - Risk shifting: interaction of state-ownership with bank capital ratio (total equity-to-total assets), focusing on less-capitalized banks and periods of sovereign distress.
  - Sovereign distress defined as CDS > 500 bps, or S&P long-term rating for sovereign FX debt is CCC- or lower, or government in external or domestic default.
- Empirical results (key coefficients):
  - State-owned banks’ net purchases of domestic sovereign bonds are around 10 percentage points higher than private banks during times of high fiscal need.
  - During periods of high fiscal need when the sovereign is in distress, state-owned banks’ net purchases are 26 percentage points higher than private banks.
  - During sovereign distress, net purchases of less-capitalized state-owned banks are over 20 percentage points higher (capital ratio is transformed to its standard deviation in regressions).
- Non-linear threshold effects:
  - State-owned banks’ additional purchases during high fiscal need increase from 26 to 50 percentage points when the sovereign CDS threshold increases from 500 bps to 1000 bps.
  - Less-capitalized state-owned banks’ additional purchases increase to around 40 percentage points when the CDS threshold increases from 500 bps to 1000 bps.

### Policy implications and recommendations
- Given the significant role of domestic NBFIs in some EMs, urgent policy actions are needed to address financial stability risks from their activity, including:
  - Broadening the reach of macroprudential tools to cover NBFIs involved in credit intermediation activities that might pose bank-like financial stability risks and/or regulatory arbitrage.
  - Strengthening the resilience of NBFIs to reduce procyclicality and amplification of shocks.
  - Enhancing transparency and data provision on NBFIs’ activities in government bond markets.
  - Appropriate regulation of NBFIs in both advanced and emerging market economies to help lessen cross-border contagion risks that are increasingly material to financial stability in EMs and developing economies.

*Authors of the box: Andrea Deghi and Mustafa Yasin Yenice. Source: Global Financial Stability Report Chapter 2 (April 2022).*

### 1. Net Purchases of State-Owned Banks During High Need Periods (Moral

### 1. Net Purchases of State-Owned Banks During High Need Periods (Moral Suasion)

### Regression evidence on moral suasion and risk-shifting
- Dependent variable (bank-level panel): net purchases of sovereign debt measured as the log change in bank holdings (one year ahead).
- “Moral suasion” line: effect for state-owned banks during episodes of high fiscal need, defined when expected maturing debt as a share of total debt is above the 75th percentile for the sample period.
- “Risk shifting” line: effect for state-owned banks with a capital ratio one standard deviation below the mean.
- Sovereign distress definition (used in figures): dummy = 1 when sovereign CDS > thresholds (300 bpts, 400 bpts, ..., or 1000 bpts), or S&P long-term rating for sovereign FX debt is CCC- or lower, or government in external/domestic default according to the Harvard Business School Global Crises Data by Country; 0 otherwise.
- All regressions include bank controls and bank and country-year fixed effects. Standard errors clustered at the bank level. Bands indicate 90% confidence intervals.

### Key coefficients and statistical results (Online Annex Table 2.4.1 excerpts)
- Policy rate: 0.24** (0.10), 0.20* (0.10)
- Gross public debt, % of GDP: 0.24*** (0.03), 0.25*** (0.03)
- Change in NER (LC/USD): -0.04** (0.02), -0.02 (0.02)
- CPI inflation (eop): -0.17*** (0.06), -0.24** (0.10)
- Real GDP per capita: -0.62*** (0.14), -0.64*** (0.14)
- Real GDP growth: -0.07 (0.07), -0.16* (0.09)
- Private sector credit, % of GDP: -0.11*** (0.02)
- Stock market cap, % of GDP: -0.03** (0.02)

Moral suasion and bank-level interactions (selected estimates):
- State Owned: 10.06 (7.10), 9.99 (6.85), 26.80*** (8.85), 25.88* (14.20)
- High Need x State Owned: 9.81* (5.85), 9.88* (5.85), 26.34* (14.02), 25.75* (13.09)
- Capital Ratio x State Owned: -6.63 (5.45), -4.95 (5.34), -21.64*** (8.21), -24.28*** (8.93)
- Capital Ratio: 8.92*** (2.55), 5.31* (2.72), 17.80** (7.53), 4.35 (7.73)

Other bank-level controls (selected; coefficients reported where available):
- Deposits/Assets: -0.11 (0.22), 0.14 (0.54)
- Loans/Deposits: 0.03 (0.02), -0.02 (0.07)
- Total Assets (interaction reported): -14.29** (5.73), -92.33*** (17.65)

Model fit and samples:
- Cross-country panel (Panel 1): sample period 2000-20; dependent variable: bank holdings of sovereign debt as a fraction of total banking sector assets; country and year fixed effects; standard errors clustered at the country level.
  - Observations: 391; 359
  - R-squared: 0.85; 0.87
  - No. of countries: 23; 21
- Bank-level panel (Panel 2): sample period 2011-20; dependent variable: bank net purchases (log change); bank and country-year fixed effects; standard errors clustered at the bank level.
  - Observations: 4,087; 4,087; 529; 529
  - R-squared: 0.28; 0.28; 0.31; 0.35
  - Number of banks: 665; 665; 111; 111
  - Number of countries: 38; 38; 8; 8

Figure-related summary points:
- The plotted figures show additional purchases during high fiscal need and additional purchases by less-capitalized banks across sovereign distress thresholds sCDS > 300 bps, > 400 bps, > 500 bps, > 600 bps, > 700 bps, > 800 bps, > 900 bps, > 1000 bps, with 90% confidence intervals.
- Visual ranges span from -20 to 100 on the vertical axis (percentage points, one year ahead) in the moral suasion panel and 0 to 70 in the risk-shifting panel as presented.

### Interpretation of findings
- State-owned banks exhibit materially larger net purchases of sovereign debt during episodes of high fiscal need (moral suasion effect).
- Less-capitalized state-owned banks show behavior consistent with risk-shifting, with significantly larger additional purchases when capital ratios are low (capital ratio interactions negative, indicating stronger purchases by lower-capital banks).
- Macroeconomic conditions (policy rate, gross public debt, exchange rate changes, inflation, and income levels) are significant drivers of bank sovereign exposure.

---

### Online Annex 2.5. Measuring the Strength of the Nexus

### Two-step empirical approach
- Step 1: Country-specific SVAR models estimate risk transmission between sovereign, banking, and corporate sectors.
  - Endogenous vector y_t includes sovereign risk, bank credit risk, non-financial corporate risk, term spread and equity prices.
  - Exogenous vector x_t includes global financial conditions (or U.S. monetary policy shocks) and return on a trade-weighted dollar index.
  - Sovereign, bank, and corporate credit risks captured by sectoral expected default frequency (EDF).
  - Model estimated in first differences of EDF; stock indices use log differences.
- Step 2: Panel analysis (panel local projection methodology) assesses how fiscal and financial vulnerabilities amplify the nexus after an adverse shock (e.g., tightening in global financial conditions).

### Identification and estimation details (SVAR)
- Structural form: A y_t = ... + epsilon_t (equation (1)), transformed to reduced form (equation (2)) for estimation.
- Identification exploits heteroskedasticity following Rigobon (2003); structural shocks treated as uncorrelated and uncorrelated over time.
- Five volatility regimes used (greater than the minimum four needed) to enhance identification; regimes classified per Rigobon and Sack (2003) using an 8-day rolling standard deviation threshold of 1 for residuals.
- Robustness: main results tested with four regimes and different lag structures (two to five lags); results broadly consistent.
- Diagnostic: Portmanteau tests for lags 1 to 5 do not reject no autocorrelation at the 10 percent level for all variables except stock returns.

### Volatility-regime days (Online Annex Table 2.5.1 highlights)
- Table reports number of days with unique changes in volatility for each endogenous variable across countries (Δ Sovereign EDF, Δ Bank EDF, Δ NFCs EDF, Term spread, Δ equity prices, No volatility change).
- Example country entries (rows as reported):
  - ARG: Δ Sovereign EDF 22; Δ Bank EDF 99; Δ NFCs EDF 75; Term spread 0; Δ equity prices 146; No volatility change 2,814
  - BRA: 67; 136; 230; 0; 294; 2,957
  - CHN: 13; 10; 13; 0; 32; 4,019
  - IND: 5; 70; 21; 87; 87; 2,789
  - TUR: 55; 53; 223; 15; 64; 3,712
  - ZAF: 0; 65; 54; 365; 243; 3,395
- Note: EDF = expected default frequency.

### Panel local projection model (equation (3)) and vulnerability interactions
- Model projects endogenous variables at horizons h and includes interactions between an index of global financial conditions (GFC) and pre-shock vulnerabilities (e.g., high public debt-to-GDP, high banks’ government bond holdings-to-total assets).
- High vulnerability defined as one standard deviation above the sample mean for the metric used.
- Model estimated at quarterly frequency to match observable macro/financial data frequencies.

### Quantitative result on global financial shocks (Online Annex Figure 2.5.1 summary)
- A tightening in global financial conditions corresponding to half of the magnitude observed during the March 2020 financial market turmoil is associated with, on average, a 0.4 percentage point increase in corporate default risk.
- The effect is persistent, lasting at least four quarters after the shock.
- The baseline model without interaction effects shows banks and corporates are the most affected following such a tightening in global financial conditions.

*Sources: Bloomberg, Fitch Connect, World Economic Outlook; Datastream; Haver; Moody’s; IHS Markit; IMF, Monetary and Financial Statistics and World Economic Outlook databases; Standard & Poor’s Capital IQ; Harvard Business School Global Crises Data by Country; and IMF staff calculations.*

### 2. Change in Credit Risks Following a Shock to VIX

### 2. Change in Credit Risks Following a Shock to VIX

### Data and scope
- SVAR analysis: daily data for 15 major EMs over 2006-2020.
- Panel model: quarterly data from 2006 to 2020.
- Sovereign EDF is CDS-implied; banking and corporate EDFs are simple averages of individual institutions in a country.
- Sample (SVAR/panel context): Argentina, Brazil, Chile, China, Colombia, India, Indonesia, Malaysia, Peru, Philippines, Poland, South Africa, Thailand, Turkey, and Russia.
- Banks panel exercise: final data includes 525 domestic banks from 18 EMs (Argentina, Brazil, China, Colombia, Egypt, India, Indonesia, Malaysia, Pakistan, Peru, Philippines, Poland, Romania, Russian Federation, South Africa, Thailand, Turkey, and Vietnam) covering 1998 to 2020.
- Data cleaning thresholds: exclude banks with total assets less than $100,000; government debt securities-to-total assets ratio greater than 1; loans-to-total assets ratio below 1 percent or above 100 percent; deposits-to-total assets less than 1 percent; negative total equity or Tier 1 capital; require at least 5 consecutive years of data. After cleaning, about 94 percent of total assets relative to the raw dataset is preserved.

### Empirical methodology (exposure channel)
- Baseline panel regression (bank i in country c, year t) estimates changes in:
  - bank EDF; pre-tax profits divided by lagged total equity; log change in total equity; change in gross loans-to-total assets; log change in total gross loans; log change in total government debt holdings.
- Key regressors:
  - Sovereign exposure: total government bond holdings divided by total assets in t-1.
  - Sovereign distress: indicator equal to one if sovereign CDS premium exceeds 500 bps (at least once on a monthly average basis during year t), or is in outright default, or S&P long-term FX rating CCC- or lower; zero otherwise. Alternative thresholds up to 1000 bps and lower thresholds also examined.
  - Capital ratio: total equity-to-total assets ratio in t-1.
  - Bank controls (lagged): size (log of total assets), liquidity (non-cash assets-to-total assets), profitability (return on assets), exposure to central bank-to-total assets, interbank balances-to-total assets, loans-to-total assets (and interaction with sovereign stress).
- Fixed effects: country-year and bank fixed effects. OLS estimation with bank-clustered standard errors.
- Focus: immediate impact up to one year after sovereign distress.

### Empirical results — exposure channel (key findings and magnitudes)
- Bank expected default frequency (EDF)
  - An ex-ante 10-percentage-points higher government debt holdings-to-total assets ratio (≈ 1 standard deviation) implies a greater than 0.4 percentage points increase in EDF when sovereign CDS premium is above 500 bps.
  - Effects are twice larger for less capitalized banks.
  - Results more robust for more severe distress thresholds (higher CDS cutoffs, e.g., 1000 bps).
  - Note: the bank-level EDF exercise uses a more limited sample of mostly larger banks (118 banks) relative to other exercises (525 banks).
- Bank profits
  - Banks with ex-ante 10-percentage-points-higher government debt holdings-to-total assets ratio experience a 5.5 percentage points decline in profits-to-lagged-equity ratio following a severe sovereign distress (sCDS > 1000 bps).
  - Less capitalized banks experience an 8.5 percentage points decline (close to ½ standard deviation of profit-to-equity ratio).
  - Differential effects driven by government bond holdings and are statistically significant for more severe stress events.
- Bank equity
  - Banks with 10-percentage-points higher government debt holdings experience about a 6 percent decline in total equity following a severe sovereign distress (sCDS > 1000 bps).
  - Using a stricter distress definition (annual average sCDS above threshold) yields stronger effects (6-10 percent decline in equity).
- Bank lending (loans-to-assets and loan growth)
  - A 10-percentage-points higher government debt holdings is associated with 2.5 percentage-points-lower loans-to-total-assets ratio (≈ 1/3 of a standard deviation).
  - Effect for less capitalized banks: 4 percentage points lower loans-to-total-assets ratio.
  - Total loan growth declines by 3.5 to 5 percent for average and less-capitalized banks, respectively, during severe sovereign distress events.
- Bank government debt holdings during distress
  - Banks with ex-ante higher government debt holdings tend to increase their holdings during sovereign distress, especially when sCDS rises above 600 bps.
  - Increase in holdings is similar or lower for less capitalized banks, providing no strong evidence of risk-shifting by weakly capitalized banks.
- International reserves adequacy
  - Adverse effects are stronger in countries with lower reserves adequacy.
  - A 10-percentage-point higher share of government debt holdings in banks domiciled in countries with low reserve adequacy leads to a nearly 5.5 percent additional decline in bank equity after a severe sovereign distress event (sCDS > 1000 bps).
  - Reserves adequacy indicator: short-term-external-debt-to-international-reserves ratio above the sample average in the year prior to distress (on average close to 0.5).

### Robustness and endogeneity checks
- Alternative sovereign distress proxies:
  - Increase in fiscal debt due to exchange rate depreciation: computed as volume of foreign-currency public debt (year t-1) × log change in exchange rate (t-1 to t), normalized by total public debt (t-1). One-year lag used.
  - Maturing debt volume: indicator equals 1 when expected maturing debt-to-total fiscal debt is high (upper 1/3); interacts with change in log(VIX) to capture exogenous global risk-aversion shocks reducing international demand for EM sovereign debt.
- Findings under robustness checks:
  - Following an exchange-rate-induced fiscal debt increase, banks with higher government bond holdings experience higher perceived EDF (not always statistically significant), and lower profitability, equity, and lending.
  - When global risk aversion is high (higher VIX), exposed banks in high rollover-risk countries experience lower profits, lower equity, and lower loans-to-total-assets ratio; effects on bank default probabilities increase but are not significant.
- Additional controls and alternative exposure measures:
  - Results robust to controlling for expected real GDP growth (WEO vintages), exchange rate depreciation interacted with sovereign distress, and sensitivity of bank stock returns to sovereign yields.
  - Lending to government, receivables from government, and state guarantees are noted channels but not directly captured due to data limitations.
- Data limitations and assumptions:
  - Foreign-currency fiscal debt is assumed to be denominated in US dollars for the exchange-rate shock exercise.
  - Certain exposures (lending to government, state guarantees, liability-side implications of government deposits) are not captured due to lack of data.

### Policy-relevant implications (drawn from empirical findings)
- Sovereign distress can materially increase bank default risk (EDF), reduce bank profitability and equity, and constrain lending — effects amplified for banks with higher sovereign bond holdings and lower capital ratios.
- Strengthening bank capital buffers reduces amplification of sovereign-bank feedbacks; less-capitalized banks face substantially larger negative impacts.
- International reserve buffers matter: countries with lower reserves adequacy experience larger equity losses in banks holding more sovereign debt, underscoring the value of strong reserves to limit transmission.
- Monitoring bank sovereign exposures and rollover risks (maturing debt) is critical, particularly under episodes of heightened global risk aversion (VIX spikes).
- Policy tools to mitigate transmission include enhancing bank capital, limiting concentrated sovereign exposures, and maintaining adequate international reserves.

*Sources: Haver Analytics; Moody’s; Refinitiv Datastream; Bloomberg Finance L.P.; Fitch Connect; IMF, World Economic Outlook database; IHS Markit; Standard & Poor’s Capital IQ; IMF staff calculations. IMF | April 2022.*

### 2.11 panel 2 of the chapter.

### 2.11 panel 2 of the chapter

### Econometric specification for bank risk-taking outcomes
- The econometric model underlying Figure 2.11 panel 3-4 (equation (2)) relates a bank-level risk-taking indicator (either cumulative growth in bank i’s gross loans from year t-1 to year t+h or the cumulative change in the non-performing loan ratio over the same period) to:
  - Bank-specific characteristics: Bank_Char_{i,t-1} (e.g., capital-to-asset ratio) lagged one year.
  - Support Rating Floor purge: SRF purge_{t,i,h} (and interactions below).
  - Sovereign Distress_{s,t}.
  - Controls, fixed effects, and error term: Controls_{i,t-1}, θ_{s,t-1,0}, λ_{t,0}, ε_{i,t-1,t,h}.
- The model is estimated with annual-frequency data using ordinary least squares, with standard errors clustered at the bank level.
- The dependent variable is either:
  - cumulative growth in bank i’s gross loans from year t-1 to year t+h, or
  - cumulative change in the non-performing loan ratio over the same period.
- Bank_Char_{i,t-1} represents bank i’s risk-taking related characteristics (i.e., capital-to-asset ratio), lagged one year.

### Additional analysis and robustness checks (bank returns and sovereign stress)
- Baseline sovereign distress threshold: CDS spreads greater than 500 bps.
- Alternative sovereign distress thresholds tested:
  - CDS spreads above 400 bps (green line).
  - CDS spreads above 700 bps (red line).
- Key empirical finding:
  - Ten months after sovereign distress, banks’ abnormal returns decrease twice more if the sovereign was under extreme stress (i.e., CDS above 700 bps) relative to the baseline (CDS above 500 bps).
- Tests to isolate recent sovereign stress from prior stress:
  - Equation (3) augments the abnormal-return specification with quarterly lags (up to 12 months) of Support Rating Floor and interactions with sovereign stress dummies.
  - The red line in Online Annex Figure 2.7.3 panel 1 shows cumulative impact of more recent sovereign stress, confirming baseline results are not driven by prolonged sovereign distress.
- Time-varying bank risk exposure:
  - Rolling-window CAPM (24-month rolling-window CAPM) is used to compute abnormal stock returns allowing banks’ market beta to vary (banks’ risk exposure estimated in each month based on the past 24-month excess returns).
  - Online Annex Figure 2.7.3 panel 2 shows baseline results hold and strengthen when allowing for time-varying risk exposure.
- Confidence intervals in figures: dashed lines represent 90 percent confidence intervals.

### Data, sample coverage, and measurement details
- Support Rating Floor (SRF) data:
  - Downloaded from Fitch.
  - Ratings converted to discrete numerical values from 1 to 17 (higher values represent higher ratings).
  - In some figures SRF is noted as a numerical scale from 0 to 17 or 1-17 depending on the panel labeling; conversion in text: 1-17 (higher values correspond to higher rating or higher likelihood of receiving government support during distress).
- Bank-level financial data:
  - Unconsolidated bank-level annual financial statements data are used in the exercises (robustness checks with consolidated data still hold).
- Sample for baseline Figure 2.11 results:
  - 54 banks from 10 EMs.
  - Monthly data available for monthly stock returns, bank-level Support Rating Floor and balance sheet data.
  - Sample period: September 2007-December 2020.
- Sovereign-stress criteria:
  - Same criteria as in the exposure channel analysis.
  - Alternative thresholds used in figure comparisons: 400 bps, 500 bps, 700 bps.
- Rolling-window CAPM specifics:
  - Abnormal returns re-estimated based on a 24-month rolling window CAPM model.
  - Banks’ risk exposure to market excess returns estimated in each month based on the past 24-month excess returns.

### Stylized facts on Support Rating Floor and bank characteristics
- Distribution and stability:
  - Since the Global Financial Crisis, the distribution of EM banks across the spectrum of SRF ratings has stayed broadly similar.
- SRF association with bank size and capitalization:
  - Strong positive relationship between bank size and government support ratings (implying large implicit subsidies for too-big-to-fail banks).
  - Banks with higher SRFs tend to have lower capital-to-asset ratios (pointing to potential moral hazard concerns).
- State ownership:
  - Government-controlled banks (government ownership greater than 50 percent) have notably higher SRFs.
- Funding cost patterns:
  - In “normal times” between the Global Financial Crisis and the COVID-19 Crisis, high-government-support banks featured remarkably lower total interest expense to total interest-bearing liabilities ratio compared with low-government-support banks.
  - During systemic crises (Global Financial Crisis and COVID-19 Crisis), the difference in funding costs between high- and low-support banks became negligible (Online Annex Figure 2.7.5).
- Weights and definitions in charts:
  - Weighted averages in figures are calculated based on bank total assets.
  - Support Rating Floor is considered high if it is higher than or equal to the median in the year; otherwise treated as low.

### Macroeconomic channel analysis (sovereign downgrades → firm outcomes)
- Objective:
  - Examine impact of sovereign rating downgrades on nonfinancial corporates’ investment and debt financing, exploiting rating-agency sovereign ceiling rules.
- Identification strategy:
  - Difference-in-differences comparing “bound” firms (firms with a rating equal to or above their sovereign rating prior to the downgrade) to “unbound” firms (firms rated below their sovereign) around sovereign downgrades.
  - Working hypothesis: bound firms cut investment and reduce debt issuance more than non-bound firms after sovereign downgrades (β3,c < 0).
- Empirical specification (equation (1) of Online Annex 2.8):
  - Δ_h y_{c,j,s,t} = β1,c Bound_{j,s,t-1} + β2,c SovereignDowngrade_{s,t} + β3,c Bound_{j,s,t-1} × SovereignDowngrade_{s,t} + β4,c Controls_{j,s,t-1} + λ_{s,c} + γ_{c,t} + η_{t,c} + ε_{j,s,t}
  - Subscripts: c (country), j (firm), s (sector), t (time).
  - Δ_h y_{c,j,s,t} denotes cumulative change in firms’ investment or debt issuance over the next h years relative to the pre-downgrade period.
  - Controls include: firm size, Tobin’s Q, cash flow, cash holdings, leverage, and government ownership.
  - Regressions include sector, country, and year fixed effects.
  - Observations coinciding with sovereign debt and banking crises are excluded from baseline sample (banking crisis indicator from Laeven and Valencia (2018)).
- Data and sample:
  - Consolidated firm data from S&P Capital IQ.
  - 84 unique sovereign downgrade events in 29 EMs.
  - Sample includes 717 firms.
  - Bound firms constitute 20.4 percent of the sample.
- Stylized facts and identification support:
  - Distribution of firm ratings by sovereign rating and distribution of difference between firm and sovereign ratings show rating discontinuity consistent with sovereign ceiling policies.
  - Panels tracking investment and net debt issuance two years before and after sovereign downgrades show parallel trends pre-downgrade and divergence post-downgrade, consistent with differential impacts on bound vs. unbound firms.

*Italic: Source: ch2onlineannex - 2.11 panel 2 of the chapter (ch2onlineannex - 2.11 panel 2 of the chapter). Canonical URL: https://www.imf.org/-/media/files/publications/gfsr/2022/april/english/ch2onlineannex.pdf*

### 4. Change in Firms-Level Debt Issuance after a Sovereign Downgrade

### 4. Change in Firms-Level Debt Issuance after a Sovereign Downgrade

### Robustness and additional statistical checks
- The baseline estimates pertain to the cumulative change in investment ratio two years after a sovereign downgrade; the difference across bound and unbound firms’ cumulative change in investment ratio 3 years after a sovereign downgrade remains statistically significant, indicating a protracted impact on bound firms.
- A formal test of the parallel trends assumption using lagged values of “bound” firms, “sovereign downgrade” and their interaction term shows the coefficient on the lagged interaction term is not statistically significant when the cumulative change to the investment ratio two years after a downgrade is considered—suggesting no significant pre-downgrade difference between groups.
- A stricter definition of bound firms (ratings higher than or equal to sovereign ratings and government-obligation outlook “negative” pre-downgrade) yields results that hold and are 4.5 percentage points larger compared to the baseline.
- Placebo and exclusion tests to rule out macroeconomic confounders:
  - Sovereign downgrades occurring in recession years (two consecutive quarters with negative quarter-over-quarter real GDP growth rate) are excluded without materially changing results.
  - Replacing the sovereign downgrade indicator with a recession indicator yields no statistically significant difference between bound and unbound firms in investment ratio before and after the recession.
  - Excluding sovereign downgrades during the global financial crisis period (2008-2010) does not alter the findings.
  - Currency crisis episodes not accompanied by a sovereign downgrade (indicator from the Harvard Business School Global Crises Data by Country) show no statistical differences between bound and unbound firms during these episodes.

### Matching and difference-in-differences evidence
- Because bound firms are relatively small in number, a matching technique (estimator proposed in Ho, Imai, King, and Stuart (2007)) is used to match bound firms (treated) to comparable unbound firms (controls) on covariates:
  - Covariates: firms’ size, cash holding, cash flow, leverage, Tobin’s Q, capital stock, investment ratio (ratio of capital expenditure to lagged capital stock), change in investment ratio, government ownership, year and country.
- Difference-in-differences estimation comparing treatment and control groups confirms that bound firms cut their investment and debt issuance by around 17 percentage points more than their peers two years after a downgrade.

### Indirect impact of sovereign downgrades on banks via firms
- The adverse impact of sovereign downgrades on non-financial firms’ investment and debt issuance can translate into higher credit risk for banks.
- Empirical exercise (sample: 25 EM countries from 1995 to 2021) estimates the change in aggregate banking-system NPL ratio over the next h years from the pre-downgrade period (ΔNPL_c,h) as a function of:
  - Sovereign Downgrade_c,t
  - Share of bound firms’ assets to total corporate-sector assets_c,t−1 and its interactions with Sovereign Downgrade_c,t
  - Controls including Financial Conditions Indicator (FCI), banks’ equity-to-asset ratios, real GDP growth, double interaction terms and individual terms of the triple interaction, country and time fixed effects
- The coefficient of interest (β_3) indicates that in countries with a more dominant presence of bound firms in the corporate sector, banks’ NPL ratio increases after the sovereign downgrade and the effect persists at least two years after the downgrade occurs.
- Robustness: controlling for country-median debt affordability (debt-to-EBITDA ratio) of small and mid-size firms and including lagged NPL ratio leaves results broadly consistent and statistically significant.
- Small and mid-size companies are defined as firms with a number of employees equal or below 250.

### Key quantitative findings and sample details
- Cumulative change in investment ratio remains statistically significant at 3 years after downgrade (compared to baseline 2-year window).
- Stricter bound definition increases estimated impact by 4.5 percentage points relative to baseline.
- Bound firms cut investment and debt issuance by around 17 percentage points more than peers two years after a downgrade.
- Banking-system NPL increases in countries where bound firms represent a larger share of corporate-sector assets; effect persists at least two years post-downgrade.
- Sample for bank-NPL analysis: 25 EM countries, 1995 to 2021.
- Global financial crisis period considered for exclusion: 2008-2010.
- Recession defined as two consecutive quarters with negative quarter-over-quarter real GDP growth rate.
- Small and mid-size firms threshold: number of employees equal or below 250.

*Global Financial Stability Report Chapter 2 — IMF | April 2022*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2022/april/english/ch2onlineannex.pdf_