## 1. Riskiness of Credit Allocation Histograms (wpiea2019207-print-pdf)

## Source details

**Canonical URL:** [1. Riskiness of Credit Allocation Histograms (wpiea2019207-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019207-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019207-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019207-print-pdf.pdf.json)

---

### I. Key contributions and high-level findings
- Introduces the riskiness of credit allocation (ISS): extent to which distribution of credit is tilted toward riskier borrowers.
- Sample: 55 countries (26 advanced economies and 29 emerging markets) over 1991–2016 using firm-level data (about 500,000 firm-year observations).
- Main empirical results:
  - ISS has followed a procyclical pattern at the global level over the 25-year sample.
  - ISS is positively associated with contemporaneous GDP growth and change in the credit-to-GDP ratio; association is stronger when domestic financial conditions are looser and when bank lending standards are easier.
  - ISS is not significantly associated with future mean GDP growth, but improves prediction of downside risks to GDP growth and banking crises at horizons of up to three years.
  - Predictive power of ISS is additional to that of changes in credit-to-GDP and of the price-of-risk proxied by a financial conditions index.

### II. Interpretation and proposed mechanisms
- Mechanisms supported:
  - Firm vulnerability channel: higher ISS captures a larger weak tail of firms conditional on credit expansion and financial conditions, implying greater amplification of negative shocks.
  - Risk-sentiment / reversal channel: ISS helps predict reversals of financial conditions and corporate spreads; acts as a risk-sentiment indicator.
- Theoretical context referenced:
  - Financial-frictions-based theories: costly state verification; deteriorating bank screening during expansions (Bernanke and Gertler 1989; Berger and Udell 2004; Dell’Ariccia and Marquez 2006).
  - Belief-based theories: shifts in investor beliefs/risk appetite causing compositional changes (Minsky 1977; Kindleberger 1978; Bordalo et al. 2018).

### III. Construction of the ISS indicator (methodology)
- Data source: Worldscope; country-years used only when at least 40 firms are observed.
- Core ISS measures: ISS_EDF and ISS_Leverage. Additional variants: ICR and debt-to-EBITDA.
- Steps (for each firm-level indicator X):
  - Step 1: Each firm assigned a decile value 1–10 in indicator X distribution in its country (higher decile = larger value).
  - Step 2: Firms sorted by change in net debt to lagged total assets into five equally-sized buckets; bucket with largest increases labeled “top issuers,” largest decreases labeled “bottom issuers.”
  - Raw ISS_{c,t} = average vulnerability decile of top issuers − average vulnerability decile of bottom issuers.
  - Raw measure normalized by subtracting its country-specific mean (focus on within-country dynamics).
- Rationale and technical choices:
  - Use of deciles abstracts from changes in mean/shape and minimizes outlier influence; deciles run from 1 to 10.
  - EDF (market-based) and leverage (accounting-based) address market liquidity concerns.
  - Resulting ISS indicators have bell-curve shape and a standard deviation of about one in their distribution.

### IV. Data and auxiliary series
- Macroeconomic and credit series: IMF IFS, World Economic Outlook, BIS, Haver Analytics.
- Financial variables: Bloomberg and Thomson Reuters; lending standards from Haver Analytics.
- Financial conditions indices constructed for 43 countries over 1990–2016.
- Crisis dating: Laeven and Valencia (2018).
- Change of credit-to-GDP ratio winsorized at the 1 percent level.
- Country coverage and data definitions summarized in Appendix Table A1 and Appendix Table A2.

### V. Dynamics and cross-country evolution (global and country patterns)
- Global median two-year moving average dynamics for ISS_EDF and ISS_Leverage:
  - Elevated levels in the late 1990s.
  - Fell in 2000–04 after Asian and Russian crises and dot-com burst.
  - Historical low in 2002 for ISS_EDF and in 2004 for ISS_Leverage.
  - Rose steeply thereafter, peaked at onset of global financial crisis, declined sharply after crisis, slightly below pre-crisis level at end-2016.
- Country-level examples (Figure 3 and Figure 4 panels):
  - United States and Japan: similar cyclicality and magnitudes; divergence in 2014–16 — riskiness decreased in United States to a relatively low level while in Japan it remained relatively high.
  - Spain: riskiness rose steeply mid-to-late 1990s, remained high until 2008 crisis, then fell suddenly and largely.
  - Germany: no credit boom over 20 years; variations within narrower range; moved into positive territory in recent years.
  - India: broadly followed global patterns; measure relatively low in 2016.
  - China: weaker synchronization with global developments; peaks/troughs with two-to-three-year lag; peak in 2009–10 consistent with end-2008 stimulus misallocation evidence.

### VI. Cyclicality — econometric framework and core estimates
- Empirical setup:
  - Equation (2): cross-country panel regressions with country fixed effects (α_i^V) and year fixed effects (γ_t^V); standard errors clustered at country level.
  - Explanatory variables include ∆GGDP (real GDP growth), ∆Credit-to-GDP, and domestic currency appreciation against the U.S. dollar (CITQTRT D).
- Main cyclicality results (Table 1):
  - Both EDF-based and leverage-based ISS increase when GDP growth or change in credit-to-GDP ratio are stronger.
  - A one standard deviation increase in change of credit-to-GDP (equivalent to 5.5 percentage points) is associated with an increase in ISS of 0.12–0.25 standard deviation, depending on ISS measure.
  - Adding credit boom dummy, credit boom length, or phase dummies yields none significant — absence of nonlinearities; relationship not driven solely by extreme credit expansions.
- Enriched specification (Equation (3)): interaction of credit expansion with financial conditions variable E (FCI, lending standards, or corporate spread); interaction coefficient θ_V captures modification of credit cyclicality by financial conditions.

### VII. Financial conditions, lending standards, and search-for-yield (interaction results)
- Table 2 interaction findings:
  - Association between larger credit expansions and riskier allocations is stronger when:
    - price of risk is low,
    - lending standards are loose,
    - corporate credit spreads are low.
  - Long-term rates: signs consistent with search-for-yield (lower long-term rates associated with higher riskiness) but statistical significance weak.
- Selected coefficient examples (from summary table):
  - Real GDP Growth: 0.08*** (0.02).
  - Δ(Credit-to-GDP Ratio): 0.02*** (0.01) and 0.05*** (0.01) in two specs.
  - Δ(Credit-to-GDP Ratio) x FCI: -0.01** (0.01).
  - Δ(Credit-to-GDP Ratio) x Bank Lending Standards: -0.04*** (0.01).

### VIII. ISS and future GDP growth — predictive evidence
- Baseline predictive regressions (Equation (4), Table 3):
  - Dependent: cumulative real GDP growth from t to t+h, h=1,2,3; explanatory variables enter as first lag of their simple three-year moving average.
  - Results: credit expansions and tight financial conditions tend to forecast future GDP declines; relationship between ISS and future mean GDP growth is never positive and rarely significant.
- Quantile regressions (Table 4):
  - Replacing dependent by deciles of 3-year cumulative GDP growth distribution:
    - Greater ISS shifts left tail and median (bottom five deciles) to the left.
    - Top deciles move to the right (generally not significant).
  - Conclusion: ISS significantly impacts downside risks to growth.
- Magnitudes (Table 5; left-tail effects):
  - Change in credit-to-GDP always has negative and significant effect.
  - FCI coefficient positive and significant mostly in first year.
  - EDF-based ISS:
    - Negative and significant for left-tail deciles over two- and three-year horizons; one standard deviation increase in EDF-based ISS shifts left tail of 3-year cumulative growth distribution left by 1–1.3 percentage point.
  - Leverage-based ISS:
    - Always negative and significant for left-tail deciles; one standard deviation increase shifts left tail of 3-year cumulative growth distribution left by 0.6–0.7 percentage point.

### IX. ISS as an early warning for banking crises
- Stylized pattern (Figure 4 Panel 1): inverted-U around systemic crisis — rises during five years preceding crisis, peaks, then falls after crisis onset; large pre-crisis credit expansions and low corporate spreads.
- Conditional fixed-effects logistic regressions (Equation (5), Table 6):
  - Dependent: start of systemic banking crisis (Crisisstart dummy).
  - Explanatory: change in credit-to-GDP (mv3), FCI (mv3), ISS (mv3), controls (current-account-balance-to-GDP, real GDP growth).
  - Key results:
    - Change in credit-to-GDP predicts crises when alone; price of risk predicts crises when alone.
    - When entered jointly, change in credit-to-GDP ceases to be significant.
    - ISS is significant when entered alone and jointly with change in credit-to-GDP and price of risk.
    - For given size of credit expansion and price of risk, greater ISS implies higher crisis probability.
    - Quantification: a one standard deviation rise in ISS increases odds of a crisis by a factor of about four.
    - Pseudo-R2 shows sizable improvement when ISS is added.
- Estimator remarks: conditional fixed effects logit provides consistent estimates but not individual fixed effects; unconditional-with-dummies estimator yields similar coefficients and addition of ISS boosts AUROC.

### X. Robustness and alternative constructions
- Robustness exercises:
  - Alternative firm vulnerability indicators (EDF, leverage, ICR, debt-to-EBITDA): core results hold.
  - Results robust to perturbations of ISS definition, alternative aggregate credit series, and large set of controls.
- Motivation for preferring ISS over high-yield share (HYS):
  - ISS covers loans and bonds and less affected by secular shifts across financing types.
  - Bond market development limited: in 55-country sample only 6 have at least one firm issuing a high-yield bond and one firm issuing an investment-grade bond every year between 1995 and 2016.
- Alternative aggregate credit series tested (Tables 7a, 7b, 8a):
  - Baseline: domestic bank credit to private nonfinancial sector (IFS).
  - Alternatives: total private nonfinancial credit (BIS), splits by domestic bank vs cross-border, cross-border claims on banks, credit gap (IFS-based; one-sided HP filter).
  - Results: ISS and credit quantity variable always significant in downside-risks-to-growth models; ISS and FCI always significant in crisis models.

### XI. Additional robustness checks and alternative ISS constructions
- Alternative constructions/perturbations (Table 9a, 9b):
  - Five perturbations: raw vulnerability average, debt-weighted deciles, sorting by net issuance in USD, reverse ISS, combination variants.
  - Results: EDF-based alternatives all have right sign and are significant; four of five leverage-based alternatives have right sign and are significant (one leverage alternative not significant).
- Other checks: individual lags of ISS, excluding three years post-crisis, using GDP per capita, controlling for other early warning indicators — results hold.

### XII. Mechanisms explored (Section VIII — regression evidence)
- Mechanism 1: ISS captures lending standards and concentration of vulnerable firms
  - Regression (Equation (6)): change in share of assets in high vulnerability firms (Share_HV) on change in credit-to-GDP, FCI, and ISS.
  - High vulnerability definitions: leverage >75th percentile; ICR <25th percentile; debt-to-EBITDA >75th percentile.
  - Table 10: ISS coefficient very significant across definitions and indicators — changes in ISS capture changes in distribution of corporate vulnerabilities.
  - Example coefficients: ISS Leverage examples 2.50*** (0.31), 2.81*** (0.51), 2.24*** (0.46).
- Mechanism 2: ISS as risk-sentiment indicator predicting reversals in financial conditions
  - Regression (Equation (7)): ΔF on lagged ΔCredit-to-GDP mv3, lagged F mv3, and lagged ISS mv3 (F = FCI or corporate spread).
  - Table 11: ISS helps predict reversals in financial conditions and corporate spreads; effect on reversal more than twice stronger when financial conditions are loose.
  - Example coefficients: ISS EDF 0.11** (0.05) and 0.27*** (0.09) in restricted samples.
- Forecasting/perception channel: professional forecasters (Table 12)
  - Regressions using IMF WEO forecasts:
    - When ISS is high, professional forecasters anticipate higher future GDP growth.
    - Forecast errors: one standard deviation increase in ISS associated with a forecast error greater than 0.5 percentage points (positive and significant).
    - Suggests forecasters fail to internalize negative effects of increases in ISS.

### XIII. Policy implications and conclusions
- Main conclusions:
  - ISS helps predict shifts in left tail of GDP growth distribution and systemic banking crises 2–3 years ahead in 55-country sample (1991–2016).
  - Predictive power of ISS is additional to aggregate credit quantities and price of risk.
  - Variations in ISS associated with thicker weak tails of corporate vulnerability distributions and with future reversals of financial conditions.
  - Economic forecasters mistakenly associate increases in ISS with increases in future GDP growth.
- Macroprudential policy implications:
  - Calibration of countercyclical capital buffers currently emphasizes aggregate credit quantities and credit gap.
  - Findings suggest policy-makers also need to take the riskiness of credit allocation into account to prevent financial instability and to distinguish good from bad credit booms.
- Theoretical implications:
  - Results favor models emphasizing credit supply shocks and behavioral biases in explaining credit expansions and macrofinancial consequences.
- Limitations and future research:
  - Predictive performance established in-sample; out-of-sample performance left for future research due to need for longer series or higher-frequency data.
  - Appendix Table B8 shows ISS predictive power for downside risks to growth in pre-2008 sample as well.

### XIV. Sample, data definitions, and country coverage (Appendix A)
- Firm-level sample and cleaning:
  - Worldscope database; drop financial sector (except real estate); drop negative or implausible values; require full info on net debt issuance, leverage, EBITDA, market capitalization.
  - Keep only country-year pairs with ≥40 firms and available aggregate credit to private sector.
  - Final sample: about 500,000 nonfinancial firm-year observations from 55 countries during 1990–2016.
  - Minimum observation notes for ICR, debt/EBITDA, EDF indicators; exception for Ireland in one case.
- Firm vulnerability indicators:
  - Leverage = total debt / total assets.
  - EDF = Black‑Scholes‑Merton model.
  - ICR = interest expenses / EBITDA (sign adjusted so ISS rises with vulnerability).
  - Debt/EBITDA = total debt / EBITDA (deciles of EBITDA/debt used when needed).
- Country coverage (selected):
  - Advanced Economies: Australia, Austria, Belgium, Canada, Czech Republic, Denmark, Finland, France, Germany, Greece, Hong Kong SAR, Ireland, Israel, Italy, Japan, Korea, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, United Kingdom, United States (ISS series from years listed in Appendix Table A.1).
  - Emerging Markets: Argentina, Brazil, Bulgaria, Chile, China, Croatia, Egypt, India, Indonesia, Jordan, Kuwait, Malaysia, Mexico, Morocco, Oman, Pakistan, Peru, Philippines, Poland, Romania, Russia, Saudi Arabia, Serbia, South Africa, Sri Lanka, Thailand, Turkey, Ukraine, Vietnam (ISS series from years listed in Appendix Table A.1).
- Country-level data sources (selected variable definitions and sources listed in Appendix Table A.2):
  - Real GDP growth: IMF, World Economic Outlook database.
  - Financial Conditions Index: authors’ estimates (methodology in Online Appendix).
  - Credit to Private Sector (baseline): IMF, International Financial Statistics (bank credit to private nonfinancial sector).
  - Cross-border credit series: BIS LBS A6.1-F; BIS CRE tables for sectoral credit.

*Source: wpiea2019207-print-pdf (IMF Working Paper).*

### 1. Riskiness of Credit Allocation Histograms..........................................................................25

### 1. Riskiness of Credit Allocation Histograms

### I. Key contributions and high-level findings
- Introduces a new dimension of credit to the private sector: the extent to which the distribution of credit is tilted towards riskier borrowers, labeled the riskiness of credit allocation (ISS).
- Sample: 55 countries (26 advanced economies and 29 emerging markets) over 1991–2016 using firm-level data.
- Main empirical results:
  - ISS has followed a procyclical pattern at the global level over the 25-year sample.
  - ISS is positively associated with contemporaneous GDP growth and change in the credit-to-GDP ratio; this association is stronger when domestic financial conditions are looser and when bank lending standards are easier.
  - ISS is not significantly associated with future mean GDP growth, but improves prediction of downside risks to GDP growth and banking crises at horizons of up to three years.
  - The predictive power of ISS is additional to that of changes in credit-to-GDP and of the price-of-risk proxied by a financial conditions index.

### II. Interpretation and proposed mechanisms
- Two mechanisms supported by the analysis:
  - Firm vulnerability channel: conditional on size of credit expansion and level of financial conditions, higher ISS captures a larger weak tail of firms, implying greater amplification of negative financial shocks.
  - Risk-sentiment / reversal channel: ISS helps predict reversals of financial conditions and corporate spreads, indicating a role as a risk sentiment indicator.
- Theoretical context:
  - Financial-frictions-based theories: costly state verification and deteriorating bank screening capacity during credit expansions can make lending to high-risk firms procyclical (Bernanke and Gertler 1989; Berger and Udell 2004; Dell’Ariccia and Marquez 2006).
  - Belief-based theories: shifts in investor beliefs and risk appetite can lead to more credit supplied to riskier firms during optimism, producing compositional changes not implied solely by aggregate credit volumes (Minsky 1977; Kindleberger 1978; Bordalo et al. 2018).

### III. Construction of the ISS indicator (methodology)
- Data source: firm-level data from the Worldscope database; country-years used only when at least 40 firms are observed.
- Two core ISS measures constructed: ISS_EDF and ISS_Leverage. Two additional ISS variants for robustness: interest coverage ratio (ICR) and debt-to-EBITDA.
- Construction steps (for each firm-level indicator X):
  - Step 1: In each year, assign each firm a decile value from 1 to 10 in the distribution of indicator X in its country (higher decile = larger value of indicator).
  - Step 2: Sort firms by change in net debt to lagged total assets into five equally-sized buckets; label the bucket with largest increases in debt as “top issuers” and the bucket with largest decreases as “bottom issuers.”
  - Raw ISS for country c and year t equals the average vulnerability decile of top issuers minus average vulnerability decile of bottom issuers (formula (1) in the source).
  - Because focus is on within-country dynamics, normalize the raw measure by subtracting its country-specific mean.
- Rationale and technical choices:
  - Use of deciles abstracts from changes in mean and shape of the vulnerability distribution and minimizes outlier influence; deciles run from 1 to 10.
  - Two main ISS constructions to address market liquidity concerns: EDF (market-based) and leverage (accounting-based); EDF construction explained in Online Appendix B.
  - The resulting ISS indicators have the shape of a bell curve and a standard deviation of about one in their distribution.

### IV. Data and auxiliary series
- Macroeconomic and credit series sourced from IMF’s International Financial Statistics (IFS), World Economic Outlook, BIS, and Haver Analytics.
- Financial variables from Bloomberg and Thomson Reuters; lending standards from Haver Analytics.
- Financial conditions indices constructed for 43 countries over 1990-2016 (see Online Appendix B).
- Crisis dating from Laeven and Valencia (2018).
- The change of the credit-to-GDP ratio is winsorized at the 1 percent level to reduce outlier influence.
- Country coverage and data definitions summarized in Appendix Table A1 and Appendix Table A2.

### V. Dynamics and cross-country evolution
- Global median two-year moving average dynamics for ISS_EDF and ISS_Leverage:
  - Elevated levels in the late 1990s.
  - Fell in 2000–04 after Asian and Russian crises and the burst of the dot-com bubble.
  - Reached historical low in 2002 for ISS_EDF and in 2004 for ISS_Leverage.
  - Rose steeply thereafter and hit a peak at the onset of the global financial crisis.
  - Declined sharply after the crisis and was slightly below its pre-crisis level at end-2016 (latest data point).
- Country-level evolution generally mirrors the global pattern with country-specific nuances.

### VI. Robustness and alternative constructions
- Robustness exercises:
  - ISS constructed using alternative firm vulnerability indicators (EDF, leverage, ICR, debt-to-EBITDA); core results hold across these measures.
  - Results robust to perturbations of the ISS definition, to alternative aggregate credit series, and to a large set of controls used in the financial crisis literature.
- Motivation for preferring ISS over high-yield share (HYS):
  - ISS covers both loans and bonds and is less affected by secular shifts across financing types.
  - Bond market development is limited in many sample countries, making HYS noisy and volatile; in the sample of 55 countries only 6 have at least one firm issuing a high-yield bond and one firm issuing an investment-grade bond every year between 1995 and 2016.

### VII. Implications for monitoring and policy
- Policy-relevant points (inferred from empirical findings and mechanisms discussed in the source):
  - Monitoring ISS provides additional early-warning information on downside risks to GDP growth and banking crises beyond aggregate credit growth and financial conditions measures.
  - Elevated ISS during credit expansions and loose financial conditions signals greater vulnerability due to both a larger weak tail of firms and heightened probability of reversals in financial conditions and spreads.
  - Using multiple firm-level vulnerability indicators (EDF and accounting ratios) improves robustness of surveillance, especially for countries with illiquid equity markets.

*Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019207-print-pdf.pdf*

### 2016. The two measures display similar patterns in the six countries.

### wpiea2019207-print-pdf - 2016. The two measures display similar patterns in the six countries.

### Country-level patterns of the riskiness of credit allocation (Figure 4 panels)
- United States (Figure 4, panel 1) and Japan (Figure 4, panel 2)
  - Dynamics very similar in both cyclicality and magnitudes over most of the sample.
  - The most recent period (2014–16) suggests divergence: riskiness decreased in the United States to a relatively low level while in Japan it remained at a level that is relatively high in historical perspective.
  - Note: the pattern in the United States closely resembles that shown on Figure 1 in GH. The decline in Japan in the first half of the 2000s is consistent with Fukuda and Nakamura (2011) on zombie lending.
  - In the United States during 2010–16 corporate leverage increased across the board, but differences across groups of firms imply the riskiness of credit allocation decreased during 2014-16 (average leverage decile of firms issuing the most debt decreased, while the average leverage decile of firms issuing the least debt increased).
- Spain (Figure 4, panel 3)
  - Credit boom from late 1990s to mid-2000s followed by deep recession during the global financial crisis and the euro area sovereign debt crisis.
  - Riskiness rose steeply in the mid-to-late 1990s, remained very high until 2008 crisis, then fell suddenly and largely—consistent with Banco de España (2017).
- Germany (Figure 4, panel 4)
  - No credit boom over the 20-year period; variations in riskiness remained within a narrower range similar to the United States and Japan.
  - The measure moved into positive territory in recent years, suggesting a higher level of risk-taking.
- India (Figure 4, panel 5)
  - Evolution broadly followed global patterns; measure was at a relatively low level in 2016.
- China (Figure 4, panel 6)
  - Synchronization with global developments is weaker; peaks and troughs appear to occur with a two-to-three-year lag.
  - A peak in 2009–10 is consistent with evidence that the stimulus plan beginning at end-2008 led to misallocation of credit (Cong and others 2017).

### Cyclicality: econometric framework and core findings
- Empirical setup
  - Equation (2) estimated in cross-country panel regressions with country fixed effects (α_i^V) and year fixed effects (γ_t^V); standard errors clustered at the country level.
  - 퐼퐼퐼퐼퐼퐼_{i,t}^V denotes the riskiness of credit allocation based on firm-level vulnerability indicator V for country i at time t.
  - Explanatory variables include ∆GGDP (real GDP growth), ∆Credit-to-GDP (change in ratio of bank credit to nonfinancial private sector to nominal GDP), and domestic currency appreciation against the U.S. dollar (CITQTRT D) to control for valuation effects on ISS.
- Main cyclicality results (Table 1)
  - Whether EDF-based (column (1)) or leverage-based (column (2)), the riskiness of credit allocation increases when GDP growth or changes in the domestic credit-to-GDP ratio are stronger.
  - The association of credit expansion with greater riskiness is statistically significant for both measures.
  - Quantification: a one standard deviation increase in the change of the credit-to-GDP ratio (equivalent to an increase of 5.5 percentage points) is associated with an increase in the riskiness of credit allocation of 0.12–0.25 standard deviation, depending on the ISS measure.
  - Adding a credit boom dummy, a variable capturing credit boom length, or phase dummies yields none significant—pointing to absence of nonlinearities and that relationship is not driven solely by extreme credit expansions.
- Enriched specification (Equation (3))
  - Includes interaction of credit expansion with a financial conditions variable E (either a financial conditions index (FCI), survey-based lending standards, or corporate credit spread).
  - Change in credit-to-GDP demeaned at country level; lending standards and corporate spreads transformed into z-scores.
  - The interaction coefficient θ_V captures how financial conditions modify credit cyclicality of riskiness.

### Financial conditions, lending standards, and search-for-yield
- Table 2 findings: interaction effects
  - The association between larger credit expansions and riskier allocations is stronger when:
    - the price of risk is low (columns 1 and 4),
    - lending standards are loose (columns 2 and 5),
    - corporate credit spreads are low (columns 3 and 6).
  - This indicates outward credit supply shifts are associated with riskier allocations.
  - Long-term rates: signs consistent with search-for-yield motives (lower long-term rates associated with higher riskiness), but statistical significance is weak.

### Riskiness of credit allocation and future GDP growth (Equations (4))
- Baseline predictive regressions (Table 3)
  - Dependent variable: cumulative real GDP growth from t to t+h, where h=1,2,3.
  - Explanatory variables enter as first lag of their simple three-year moving average.
  - Results: credit expansions and tight financial conditions tend to forecast future GDP declines; the relationship between ISS and future GDP growth is never positive and rarely significant—i.e., a riskier credit allocation does not provide an extra kick to mean future GDP growth.
- Quantile regression analysis (Table 4; Powell (2016) estimator)
  - Replacing dependent variable by deciles of the 3-year cumulative GDP growth distribution reveals heterogeneity across distribution.
  - Findings:
    - Greater riskiness of credit allocation shifts the whole left tail and the median (bottom five deciles) of the growth distribution to the left.
    - It moves top deciles to the right, although generally not significantly so.
    - Conclusion: riskiness of credit allocation has a significant impact on downside risks to growth.
- Magnitudes for left-tail effects (Table 5)
  - Focus on bottom two deciles for 1-year, 2-year, and 3-year ahead horizons.
  - Change in credit-to-GDP always has a negative and significant effect.
  - FCI coefficient positive and significant mostly in the first year.
  - EDF-based ISS:
    - Negative and significant for both deciles over two- and three-year horizons, but only for the second decile during the first year.
    - Quantitative effect: a one standard deviation increase in EDF-based ISS shifts left tail of 3-year cumulative growth distribution left by 1-1.3 percentage point.
  - Leverage-based ISS:
    - Always negative and significant for the left-tail deciles.
    - Quantitative effect: a one standard deviation increase in leverage-based ISS shifts left tail of 3-year cumulative growth distribution left by 0.6-0.7 percentage point.

### Riskiness of credit allocation as an early warning for banking crises (Equation (5))
- Stylized pattern (Figure 4 Panel 1)
  - ISS shows inverted-U shape around systemic crisis episodes: rises gradually during five years preceding crisis, reaches high level, then falls after crisis onset.
  - Pre-crisis: large credit expansions and low corporate spreads observed.
  - Conventional corporate vulnerability indicators pick up only post-crisis.
- Conditional fixed-effects logistic regressions (Table 6)
  - Dependent variable: start of a systemic banking crisis (Crisisstart dummy).
  - Explanatory variables: change in credit-to-GDP (mv3), FCI (mv3), ISS (mv3), controls for macro environment including change in current-account-balance-to-GDP and real GDP growth.
  - Key results:
    - Change in credit-to-GDP predicts crises (column (1)).
    - Price of risk predicts crises (column (2)).
    - When change in credit-to-GDP and price of risk enter jointly (column (3)), the change in credit-to-GDP ceases to be significant.
    - ISS is significant when entered alone (column (4)), together with change in credit-to-GDP (column (5)), and together with change in credit-to-GDP and price of risk (column (6)).
    - For a given size of credit expansion and given price of risk, a greater level of ISS implies a higher probability of financial crisis.
    - Quantification: a one standard deviation rise in the ISS measure increases the odds of a crisis by a factor of about four.
    - Pseudo-R2 shows sizable improvement when ISS is added.
- Estimator remarks
  - Conditional fixed effects logit provides consistent estimates but not individual fixed effects; unconditional-with-dummies estimator yields similar coefficients and addition of ISS boosts AUROC in that model (Online Appendix Table B2).

### Robustness checks (Section VII)
- Alternative aggregate credit series (Table 7a, 7b, 8a)
  - Baseline: credit to private nonfinancial sector by domestic banks (IFS).
  - Alternatives tested: total credit to private nonfinancial sector from BIS (domestic and foreign, bank and nonbank), splits by domestic bank credit and cross-border sources, cross-border claims on banks, cross-border claims on banks plus non-banks, and the credit gap (IFS-based; one-sided HP filter).
  - Results: ISS and credit quantity variable are always significant in downside-risks-to-growth models; ISS and FCI are always significant in crisis models; change in credit-to-GDP is rarely significant in crisis models.
- Alternative firm-level vulnerability indicators (Table 8a, 8b)
  - Alternatives: interest coverage ratio (ICR), debt-to-EBITDA.
  - Results: similar to baseline EDF and leverage indicators—ISS always significant in crisis model and associated with left-tail shifts in future GDP growth distribution (more significant for ICR).
- Alternative constructions/perturbations of ISS (Table 9a, 9b)
  - Five perturbations explored, including raw vulnerability instead of decile, debt-weighted deciles, sorting by net issuance in US dollars, reverse ISS (sort by vulnerability and use deciles of net issuance), and combination of sorting by net issuance in US dollars with raw vulnerability.
  - Results: all five EDF-based alternatives have right sign and are significant; four of five leverage-based alternatives have right sign and are significant (second alternative not significant).
- Other checks (Online Appendix B)
  - Results hold when using individual lags of ISS, excluding three years post-crisis, using GDP per capita, and controlling for other early warning indicators (real effective exchange rate, foreign exchange reserves, aggregate corporate vulnerability indicators).

### Mechanisms explored (Section VIII)
- Mechanism 1: ISS captures variations in lending standards and concentration of vulnerable firms
  - Regression (Equation (6)) relating change in share of assets in high vulnerability firms (Share_HV) to change in credit-to-GDP, FCI, and ISS.
  - High vulnerability defined as:
    - leverage above 75th percentile,
    - ICR below 25th percentile,
    - debt-to-EBITDA above 75th percentile.
  - Table 10: ISS coefficient very significant across vulnerability definitions and ISS indicators—changes in ISS capture changes in distribution of corporate vulnerabilities.
- Mechanism 2: ISS as a risk-sentiment indicator predicting reversals in financial conditions
  - Regression (Equation (7)): ΔF on lagged ΔCredit-to-GDP mv3, lagged F mv3, and lagged ISS mv3, where F is FCI or corporate spread.
  - Table 11: ISS helps predict reversals in financial conditions and corporate spreads.
    - Effect on reversal in financial conditions more than twice stronger when financial conditions are loose.
  - ISS exhibits features of a risk-sentiment indicator.
- Forecasting/perception channel: professional forecasters
  - Regressions using IMF WEO forecasts (Table 12)
    - When ISS is high, professional forecasters anticipate higher future GDP growth.
    - Forecast errors: one standard deviation increase in ISS associated with a forecast error greater than 0.5 percentage points (positive and significant).
    - Suggests forecasters fail to internalize negative effects of increases in ISS—difficult to reconcile with purely rational-expectations models.

### Policy implications and conclusions (Sections IX)
- Main conclusions
  - The ISS measure of debt issuer quality (riskiness of credit allocation) helps predict shifts in the left tail of GDP growth distribution and systemic banking crises 2 to 3 years ahead in a sample of 55 countries covering 1991–2016.
  - Predictive power of ISS is additional to changes in aggregate credit quantities and the price of risk.
  - Shifts in credit supply play a role in explaining variations in ISS; variations in ISS are associated with thicker weak tails of corporate vulnerability distributions and with future reversals of financial conditions.
  - Economic forecasters mistakenly associate increases in ISS with increases in future GDP growth.
- Macroprudential policy implications
  - Calibration of countercyclical capital buffers currently emphasizes aggregate credit quantities and the credit gap.
  - Findings suggest policy-makers also need to take the riskiness of credit allocation into account to prevent financial instability and to distinguish good from bad credit booms.
- Theoretical implications
  - Results favor models emphasizing credit supply shocks and behavioral biases in explaining credit expansions and their macrofinancial consequences.
- Limitations and future research
  - Predictive performance established in-sample; out-of-sample performance left for future research due to need for longer series or higher-frequency data.
  - Appendix Table B8 shows ISS predictive power for downside risks to growth in pre-2008 sample as well, reducing concern that ISS only captures Great Recession developments.

*Source: IMF Working Paper (wpiea2019207-print-pdf).*

### Appendix A for country coverage.

### Appendix A for country coverage

### 1. Overview of Figures and Measures
- Figures present two primary riskiness-of-credit-allocation measures:
  - "Expected Default Frequency–Based Measure" (EDF–based measure).
  - "Leverage–Based Measure".
- Time coverage in panels and country charts: 1995 through 2016 (individual panels show year ticks including 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016).
- Visual scales and axes presented include:
  - Percent of Total with tick labels 0, 5, 10, 15, 20, 25 and negative ticks -5 -4 -3 -2 -1 0 1 2 3 4 5.
  - Index scales with ticks from -1.0 to 1.0, and other panels ranging from -2.5 to 2.5 and -1.5 to 2.0.
  - EDF graphs showing vertical axis values from -0.8 to 0.8 in some panels and -1.0 to 1.0 in others.

### 2. Selected Economies: Riskiness of Credit Allocation, 1995–2016 (Figure 3)
- Data sources listed: Worldscope; and authors’ estimates.
- Note: Panels show the simple two-year moving average.
- Note: Shaded areas indicate periods of growth below the 15th percentile of the country-specific growth distribution.
- Country-specific panels (each shows both measures over 1995–2016):
  - 1. United States — labeled series: "Leverage–based measure" and "Expected default frequency–based measure".
  - 2. Japan — labeled series: "Leverage–based measure" and "Expected default frequency–based measure".
  - 3. Spain — labeled series: "Leverage–based" and "Expected default frequency–based measure".
  - 4. Germany — labeled series: "Leverage–based measure" and "Expected default frequency–based measure".
  - 5. India — labeled series: "Leverage–based measure" and "Expected default frequency–based measure".
  - 6. China — labeled series: "Leverage–based measure" and "Expected default frequency–based measure".
- Individual panel vertical axis ranges explicitly shown in the source for some countries:
  - Example ranges: -1.0 to 1.0; -2.5 to 2.5; -1.2 to 1.2; -1.0 to 1.0; -1.2 to 1.2.

### 3. Dynamics of the Riskiness of Credit Allocation Around a Crisis Year (Figure 4)
- Index: median across all crisis episodes; 11-year window.
- Data sources listed: Laeven and Valencia (2018); Worldscope; and IMF staff estimates.
- Notes:
  - Systemic banking crises are defined as in Laeven and Valencia (2018).
  - The crisis occurs at time 0.
  - Data series are de-meaned at the country level.
  - Panels show the median across all crisis countries in a balanced panel.
  - In panel 4, "median EDF (resp. leverage) refers to the median of the firm-level EDF (resp. leverage) indicator."
- Panels and labeled series in Figure 4:
  - 1. Riskiness of Credit Allocation — with "EDF–based measure" and "Leverage–based measure".
  - 2. Change in Credit-to-GDP Ratio.
  - 3. Price of Risk — labeled series "Corporate Spread" and "FCI".
- Index/time-axis annotated from -5 to 5 around crisis year (time 0) with ticks -5 -4 -3 -2 -1 0 1 2 3 4 5 shown in source.
- Additional numeric axis ticks visible in source panels:
  - For some panels: -1.5, -1.0, -0.5, 0.0, 0.5, 1.0, 1.5, 2.0.
  - For other panels: -8, -6, -4, -2, 0, 2, 4, 6.
  - A panel showing continuous small decimals: -0.02, -0.01, 0, 0.01, 0.02, 0.03, 0.04.
  - A panel with axis ticks: -1.0, 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0.

### 4. Presentation and Estimation Notes
- Figures use simple two-year moving averages for country panels (Figure 3).
- Median statistics in Figure 4 are computed across crisis episodes in a balanced panel after de-meaning series at the country level.
- Crisis timing: crisis defined per Laeven and Valencia (2018), with crisis at time 0 in the 11-year window.

*Source: wpiea2019207-print-pdf - Appendix A for country coverage.*

### 4.  Corporate Vulnerability Measures

### 4.  Corporate Vulnerability Measures

### Cyclicality of Riskiness of Credit Allocation
- Regressions: OLS with country and time fixed effects; standard errors clustered at country level.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Credit Expansion, Financial Conditions, and Riskiness of Credit Allocation
- Dependent variables reported include ISS Leverage and ISS EDF.
- Selected coefficients and sample statistics:
  - Real GDP Growth: 0.08*** (0.02) in both reported specifications.
  - Δ(Credit-to-GDP Ratio): 0.02*** (0.01) and 0.05*** (0.01) across two specs.
  - Appreciation against the US dollar: -0.01 (0.01) and -0.04*** (0.01).
  - Number of Observations: 936 and 986.
  - Number of Countries: 53 and 55.
  - R2: 0.15 and 0.31.
- Interaction results (selected):
  - Δ(Credit-to-GDP Ratio) x FCI: -0.01** (0.01) in two specifications.
  - Δ(Credit-to-GDP Ratio) x Bank Lending Standards: -0.04*** (0.01) and -0.03* (0.02).
  - Δ(Credit-to-GDP Ratio) x Corporate Credit Spread: -0.02** (0.01) in two specifications.
  - Financial Conditions Index coefficients reported (examples): -0.11 (0.12), -0.12 (0.12), -0.08 (0.12).
  - Bank Lending Standards coefficients reported (examples): -0.11 (0.07), -0.10 (0.07).
  - Corporate Credit Spread coefficients reported (examples): -0.05 (0.06), -0.07 (0.06).
  - Number of Observations range: 812 to 849 across panels; Number of Countries examples: 41, 42, 37.
  - R2 values reported: 0.16, 0.31, 0.23, 0.34, 0.39, 0.33.

### Riskiness of Credit Allocation and Cumulative GDP Growth
- Regressions: OLS with country and time fixed effects; explanatory variables enter as lag of simple three-year moving average; Δ(Credit-to-GDP Ratio) winsorized at 1 percent.
- Selected coefficients (panel examples):
  - Real GDP Growth: coefficients include -0.00413 (0.0635), -0.0947 (0.101), -0.2010 (0.130), 0.0347 (0.0575), -0.0657 (0.0912), -0.339*** (0.121).
  - Δ(Credit-to-GDP Ratio): -0.102*** (0.0261), -0.198*** (0.0416), -0.292*** (0.0536), -0.106*** (0.0250), -0.238*** (0.0396), -0.411*** (0.0524).
  - Financial Conditions Index: -0.524** (0.238), -0.810** (0.379), -0.889* (0.488), -0.353 (0.232), -0.641* (0.368), -0.978** (0.487).
  - ISS EDF and ISS Leverage reported with coefficients and SEs (examples): ISS EDF -0.0763 (0.133), -0.195 (0.213), -0.250 (0.274); ISS Leverage -0.260* (0.141), -0.319 (0.224), -0.231 (0.296).
  - Observations: 586 and 658 in respective panels; Number of Countries: 40 and 42.

### Riskiness of Credit Allocation and Risks to GDP Growth (All Deciles)
- Method: quantile regressions with nonadditive fixed effects.
- Panel A (ISS based on EDF), Δ(Credit-to-GDP Ratio) coefficients by decile (Decile 1 to Decile 9):
  - -0.137** (0.0626), -0.257*** (0.0293), -0.261*** (0.0342), -0.265*** (0.0224), -0.237*** (0.0191), -0.274*** (0.0229), -0.317*** (0.0348), -0.321*** (0.0225), -0.339*** (0.0731).
  - Financial Conditions Index by decile: 1.172 (0.940), -0.641 (0.587), 0.077 (0.374), -0.371 (0.365), -0.351 (0.365), -0.390 (0.434), -1.150** (0.490), -1.120** (0.496), -1.661** (0.759).
  - ISS EDF across deciles: -0.993*** (0.254), -1.321*** (0.164), -0.853*** (0.145), -0.639*** (0.109), -0.565*** (0.134), -0.146 (0.142), 0.129 (0.267), 0.306 (0.310), 0.519 (0.502).
  - Observations: 586 for each decile; Countries: 40.
- Panel B (ISS based on Leverage), Δ(Credit-to-GDP Ratio) coefficients by decile:
  - -0.259*** (0.0462), -0.264*** (0.0459), -0.270*** (0.0234), -0.267*** (0.0164), -0.244*** (0.0309), -0.307*** (0.0266), -0.395*** (0.0304), -0.462*** (0.0442), -0.600*** (0.0455).
  - Financial Conditions Index by decile: 2.222*** (0.704), 0.209 (0.758), -0.0118 (0.452), -0.487** (0.234), -0.422 (0.299), -0.828* (0.467), -1.506*** (0.426), -1.906*** (0.616), -2.519*** (0.801).
  - ISS Leverage across deciles: -0.650** (0.323), -0.614*** (0.200), -0.743*** (0.230), -0.689*** (0.0748), -0.559*** (0.170), -0.163 (0.152), 0.169 (0.173), 0.531 (0.356), 0.832*** (0.294).
  - Observations: 658 for each decile; Countries: 42.

### Downside Risks to Growth (1st and 2nd Decile) and Crisis Prediction Model
- Quantile regressions used for downside risks; conditional fixed effects logit for crisis prediction.
- Panel A (ISS based on EDF), Decile coefficients examples:
  - Δ(Credit-to-GDP Ratio): -0.0158*** (0.00454) for Decile 1; -0.0718*** (0.0188) for Decile 2 in column (1)-(2).
  - Financial Conditions Index examples: 1.517*** (0.0309).
  - ISS EDF examples: -0.0342 (0.0487), -0.181** (0.0717), -0.656*** (0.207), -0.708*** (0.135), -0.993*** (0.254), -1.321*** (0.164).
  - Observations: 549 for reported panels; Countries: 40.
- Panel B (ISS based on Leverage), Decile coefficients examples:
  - Δ(Credit-to-GDP Ratio): -0.0441*** (0.0137), -0.0586*** (0.0156).
  - Financial Conditions Index examples: 0.781** (0.349), 0.433*** (0.143).
  - ISS Leverage examples: -0.239** (0.114), -0.220*** (0.0537), -0.625** (0.306), -0.621*** (0.157), -0.650** (0.323), -0.614*** (0.200).
  - Observations: 658; Countries: 42.
- Crisis prediction (conditional fixed effects logit) — sample crisis panel:
  - Panel A (Riskiness of Credit Allocation Based on EDF), selected coefficients:
    - Δ(Credit-to-GDP Ratio): 0.243** (0.111), 0.0284 (0.164), 0.200* (0.0983).
    - Financial Conditions Index: -5.290*** (1.432), -5.230*** (1.491), -8.072*** (2.247).
    - ISS EDF: 1.264** (0.497), 1.105** (0.534), 3.594*** (1.042).
  - Panel B (Based on Leverage), selected coefficients:
    - Δ(Credit-to-GDP Ratio): 0.202*** (0.0699), 0.1200 (0.0750), 0.171** (0.0693).
    - Financial Conditions Index: -2.468** (0.963), -2.234** (0.947), -3.993*** (1.055).
    - ISS Leverage: 1.161*** (0.362), 1.120** (0.362), 2.560*** (0.746).
  - Observations and counts:
    - Panel A: Number of Observations 361; Number of Countries 17; Number of Crisis Episodes 17; Pseudo R2 examples: 0.328, 0.629, 0.629, 0.327, 0.373, 0.749.
    - Panel B: Number of Observations 443; Number of Countries 21; Number of Crisis Episodes 21; Pseudo R2 examples: 0.243, 0.383, 0.401, 0.264, 0.307, 0.55.

### Alternative Credit Series and Crisis Prediction (Tables 7a & 7b)
- Downside risks (Panel A: EDF-based) — Δ coefficients for alternative credit measures (examples):
  - Δ(Domestic Bank Private Nonfinancial Credit-to-GDP) (IFS): -0.257*** (0.0293).
  - Δ(Total Private Nonfinancial Credit-to-GDP) (BIS): -0.260*** (0.0221).
  - Delta Total Private Nonfinancial Credit-to-GDP: to Corporate: -0.275*** (0.0472).
  - Delta Total Private Nonfinancial Credit-to-GDP: to Household: -0.835*** (0.0768).
  - Δ(Cross-Border Credit to Non-Banks to GDP) (BIS): -0.197*** (0.0603).
  - Δ(Cross-Border Credit to Banks to GDP) (BIS): -0.175*** (0.0380).
  - Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS): -0.145*** (0.0343).
  - Credit-to-GDP Gap (based on IFS): -0.0778*** (0.00983).
  - ISS EDF across these specifications: coefficients (examples) -1.321*** (0.164), -1.426*** (0.202), -0.983*** (0.206), -0.662*** (0.231), -1.349*** (0.156).
  - Observations range: 490 to 592; Countries range: 36 to 40.
- Crisis prediction with alternative credit series (Panel A: EDF-based), selected coefficients:
  - Financial Conditions Index consistently large negative and significant across specs: -8.072*** (2.247), -7.780*** (1.902), -9.223*** (2.340), etc.
  - ISS EDF positive and significant: 3.594*** (1.042), 3.519*** (0.882), 4.439*** (1.242), etc.
  - Observations: examples 361, 311, 362; Countries: examples 17, 15; Number of Crisis Episodes: examples 17, 15.
- Panel B (Leverage-based crisis prediction) — selected coefficients:
  - Δ(Cross-Border Credit to Banks to GDP) (BIS): 0.196* (0.101).
  - Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS): 0.102** (0.0483).
  - Financial Conditions Index: -3.993*** (1.055), -3.980*** (1.036), -4.408*** (1.220), etc.
  - ISS Leverage: 2.560*** (0.746), 2.497*** (0.775), 2.548*** (0.849), etc.
  - Observations: 443 and variants; Countries: 21 and variants; Crisis episodes: 21 and variants; Pseudo R2 range: 0.549 to 0.578 across columns.

### Alternative Firm Vulnerability Indicators (Tables 8a & 8b)
- Quantile regressions and conditional logit used with alternative firm vulnerability measures (ICR, Debt/EBITDA).
- Panel A (ISS based on ICR) — Decile 1 and Decile 2 examples:
  - Real GDP Growth: 0.0995* (0.0574) and 0.102*** (0.0359) in columns (1)-(2).
  - Δ(Credit-to-GDP Ratio): -0.0568*** (0.0186) and -0.0704*** (0.0143) in columns (1)-(2).
  - Financial Conditions Index examples: 0.526* (0.277), 0.115 (0.168), 1.764*** (0.198).
  - ISS ICR coefficients: -0.160 (0.219), -0.186** (0.0810), -1.004*** (0.377).
  - Observations: 651; Countries: 42.
- Panel B (ISS based on Debt/EBITDA) — Decile examples:
  - Real GDP Growth examples: 0.106* (0.0554), 0.124*** (0.0397).
  - Δ(Credit-to-GDP Ratio): -0.0601*** (0.0165), -0.0748*** (0.0159).
  - ISS Debt/EBITDA examples: -0.0987 (0.166), -0.0866 (0.0839), -0.609*** (0.172).
  - Observations: 648; Countries: 42.
- Crisis prediction (alternative firm indicators):
  - Selected coefficients for crisis models (examples):
    - Δ(Credit-to-GDP Ratio): 0.202*** (0.0699), 0.1200 (0.0750), 0.185** (0.0839).
    - Financial Conditions Index: -2.468** (0.963), -2.234** (0.947), -4.581*** (1.615).
    - ISS Debt/EBITDA: 1.393*** (0.404), 1.401*** (0.494), 2.749*** (0.672).
    - ISS ICR: 1.557** (0.624), 1.488* (0.781), 3.082*** (1.132).
  - Observations: 443 and variants; Countries: 21 and variants; Pseudo R2 examples: 0.243, 0.383, 0.353, 0.298, 0.342, 0.611, 0.262, 0.304, 0.548.

### Alternative Constructions of Riskiness of Credit Allocation (Tables 9a & 9b)
- Table 9a: Estimates of riskiness coefficient in downside-risks-to-growth model (baseline = row (0)).
  - Baseline (0) — EDF-based: -1.321*** (0.164), Obs. 586.
  - Alternatives (selected):
    - (1) Simple average of raw vulnerability: -0.120*** (0.0229), Obs. 586.
    - (2) Debt-weighted deciles: -0.856*** (std not shown in excerpt), Obs. 586.
    - (3) By net debt issuance in USD: -0.962*** (0.193), Obs. 586.
    - (4) By vulnerability: -1.339*** (0.132), Obs. 586.
    - (5) = (1) & (3): -0.0563*** (0.0241), Obs. 586.
  - Leverage-based baseline: -0.614*** (0.200), Obs. 658; alternatives given with coefficients and SEs as provided.
- Table 9b: Crisis prediction — alternative ISS constructions (selected):
  - Baseline EDF-based: 3.594*** (1.042), Pseudo R2 0.749, Obs. 361.
  - Alternative (1) Simple average of raw vulnerability: 0.381*** (0.111), Pseudo R2 0.736, Obs. 140.
  - Alternative (2) Debt-weighted deciles: 1.761*** (0.624), Pseudo R2 0.701, Obs. 361.
  - Alternative (3) By net debt issuance in USD: 3.159*** (1.075), Pseudo R2 0.704, Obs. 361.
  - Alternative (4) By vulnerability: 3.007** (1.305), Pseudo R2 0.739, Obs. 361.
  - Alternative (5) = (1) & (3): 0.393*** (0.122), Pseudo R2 0.685, Obs. 361.
  - Leverage-based crisis results also reported with baseline 2.560*** (0.746), Pseudo R2 0.554, Obs. 443 and alternative constructions with coefficients and Pseudo R2.

### Riskiness of Credit Allocation and Change in Share of Vulnerable Assets (Table 10)
- Regressions: OLS with country fixed effects; clustered standard errors.
- Examples of coefficients:
  - ∆Credit-to-GDP: 0.18*** (0.05), 0.32*** (0.09), 0.36*** (0.09), 0.10* (0.06), 0.21** (0.09), 0.24*** (0.08) across columns with different vulnerability definitions.
  - FCI: 0.85*** (0.23), 1.49*** (0.23), 1.46*** (0.26), 0.89*** (0.21), 1.42*** (0.21), 1.45*** (0.26).
  - ISS EDF examples: 1.11*** (0.30), 1.08*** (0.31), 0.65** (0.32).
  - ISS Leverage examples: 2.50*** (0.31), 2.81*** (0.51), 2.24*** (0.46).
  - Observations: 812, 805, 842, 842, 833 in various columns; Countries: 41–42; R2 examples: 0.04, 0.07, 0.08, 0.11, 0.14, 0.12.

### Riskiness of Credit Allocation and Reversal in Financial Conditions (Table 11)
- Dependent variables: change in FCI (columns (1)–(4)) and change in corporate credit spread (columns (5)–(8)).
- Sample restrictions in some columns: only observations with lag of FCI's simple three-year moving average negative or lag of spread's simple three-year moving average negative.
- Selected coefficients:
  - FCI → ∆FCI examples: FCI -0.68*** (0.03), -1.45*** (0.08) in restricted samples.
  - Corporate Credit Spread → ∆spread examples: -0.35*** (0.04), -0.41*** (0.08).
  - ISS EDF: 0.11** (0.05), 0.27*** (0.09), 0.12** (0.04), 0.12** (0.06).
  - ISS Leverage: 0.08** (0.04), 0.25*** (0.08), 0.05* (0.03), 0.13** (0.05).
  - Sample sizes and countries vary by specification; Observations examples: 707, 348, 780, 377, 552, 328, 580, 345; Countries range: 37–42.
  - R2 examples for ∆FCI: 0.28, 0.30, 0.29, 0.31; for ∆spread: 0.15, 0.06, 0.14, 0.08.

### Riskiness of Credit Allocation, GDP Growth Forecast, and Forecast Error (Table 12)
- OLS with time and country fixed effects; explanatory variables enter as lag of simple three-year moving average and are demeaned at country level; Δ(Credit-to-GDP Ratio) winsorized at 1 percent.
- Panel A (ISS based on EDF), selected coefficients:
  - Δ(Credit-to-GDP Ratio) on Actual: -0.29*** (0.05) and -0.28*** (0.06) in reported columns.
  - Δ(Credit-to-GDP Ratio) on ForecastError: -0.01 (0.02) and 0.000 (0.02).
  - Financial Conditions Index examples: -0.89* (0.49), -0.13 (0.21), 0.76 (0.51), -0.98** (0.49), -0.01 (0.21), 0.97* (0.51).
  - ISS EDF and ISS Leverage reported in some columns with coefficients and SEs (examples): ISS EDF -0.25 (0.27), 0.28** (0.12), 0.53* (0.29); ISS Leverage -0.23 (0.3), 0.31** (0.13), 0.55* (0.31).
  - Observations: 586 and 658 in reported panels; Countries: 40 and 42.

*Source: Authors’ estimates (from the supplied chapter content).*

### Appendix A. Data Sources and Definitions

### Appendix A. Data Sources and Definitions

### Firm level data — Sample
- Data source: Worldscope database (universe of listed firms).
- Cleaning steps:
  - Drop financial sector firms (except those in the real estate sector).
  - Drop observations when market capitalization, total assets, total debt, total liability, or interest expenses are strictly negative; or when the operating profit margin or the ratio of short-term debt to total debt is larger than 100 percent.
  - Keep observations only if full information on net debt issuance; leverage; earnings before interest, taxes, depreciation, and amortization (EBITDA); and market capitalization is available.
  - Keep only country-year pairs with no fewer than 40 firms and available information on aggregate credit to the private sector.
- Final sample: about 500,000 nonfinancial firm-year observations from 55 countries during 1990 to 2016.
- Note (footnote 25):
  - For the construction of the interest coverage ratio–based indicator, a minimum of 40 observations for interest expenses is also required. An exception is made for one borderline case (Ireland), for which some years only have 38 or 39 observations.
  - For the construction of the ISS indicator based on debt/EBITDA, a minimum of 40 observations for non-zero debt is also required.
  - For the construction of the ISS indicator based on expected default frequency, a minimum of 40 observations for expected default frequency is also required.

### Firm vulnerability indicators
- Leverage ratio: ratio of total debt to total assets.
- Expected default frequency (EDF): computed using the Black-Scholes-Merton model; details and data series used for the risk-free rate are provided in the Online Appendix.
- Interest coverage ratio (ICR): ratio of interest expenses to EBITDA.
- Debt/EBITDA ratio: ratio of total debt to EBITDA.
- ISS construction notes:
  - The sign of the ISS for the ICR is adjusted so that it rises when the vulnerability of top issuers is increasing.
  - For debt overhang, the deciles of EBITDA to debt (instead of debt to EBITDA) are used to avoid classifying firms with negative earnings as low-vulnerability firms.

### Macrofinancial data — Overview
- Macrofinancial data sources, definitions, and transformations are summarized in Appendix Table A2.
- Details of the construction of the financial conditions index are provided in the Online Appendix B.

### Appendix Table A.1. Country Coverage
- Advanced Economies (ISS series from / Banking crisis start year where indicated):
  - Australia* 1991
  - Austria* 1991 2008
  - Belgium* 1991 2008
  - Canada* 1991 2000
  - Czech Republic* 1997
  - Denmark* 1991 2008
  - Finland* 1991
  - France* 1991 2008
  - Germany* 1991 2000
  - Greece* 1994 2008
  - Hong Kong SAR 1991
  - Ireland* 1999 2008
  - Israel* 2000
  - Italy* 1991 2008
  - Japan* 1991 1997
  - Korea* 1993 1997
  - Netherlands* 1991 2008
  - New Zealand* 1999
  - Norway* 1991
  - Portugal* 1996 2008
  - Singapore 1991
  - Spain* 1991 2008
  - Sweden* 1991 2008
  - Switzerland* 1991 2008
  - United Kingdom* 1991 2007
  - United States* 1991 2007

- Emerging Market Economies (ISS series from / Banking crisis start year where indicated):
  - Argentina* 2000 2001
  - Brazil* 1992 1994
  - Bulgaria* 2006
  - Chile* 1995
  - China* 2000
  - Croatia 2006
  - Egypt 2006
  - India* 1993
  - Indonesia* 1992 1997
  - Jordan 2006
  - Kuwait 2006
  - Malaysia* 1991 1997
  - Mexico* 1995
  - Morocco 2009
  - Oman 2006
  - Pakistan 1995
  - Peru* 2001
  - Philippines* 1996
  - Poland* 2000
  - Romania 2006
  - Russia* 2005 2008
  - Saudi Arabia 2006
  - Serbia 2010
  - South Africa* 1991
  - Sri Lanka 2006
  - Thailand* 1993 1997
  - Turkey* 1997 2000
  - Ukraine 2008 2014
  - Vietnam* 2007

### Appendix Table A.2. Country-Level Data Sources (variables, descriptions, sources)
- Real GDP growth
  - Description: Annual percentage change in the gross domestic product, constant prices in national currency.
  - Source: IMF, World Economic Outlook database
- Real GDP growth forecast
  - Description: Annual percentage change in the gross domestic product forecast, constant prices in national currency.
  - Source: IMF, World Economic Outlook database
- Nominal GDP
  - Description: Gross domestic product, current prices in national currency.
  - Source: IMF, World Economic Outlook database
- Inflation
  - Description: Annual percentage change in the consumer price index.
  - Source: Haver Analytics; IMF, International Finance Statistics
- Current account
  - Description: Current account balance, in US dollars.
  - Source: IMF, World Economic Outlook database
- Exchange rate
  - Description: National currency per US dollar.
  - Source: IMF, International Financial Statistics and World Economic Outlook databases
- Real Effective Exchange Rate
  - Description: Real effective exchange rate, based on the consumer price index.
  - Source: IMF, International Financial Statistics
- International Reserves
  - Description: Total reserves excluding gold in US dollars
  - Source: IMF, International Financial Statistics
- Lending Standards
  - Description: Cumulative net percentage balance (or diffusion index) of the weighted percentage of surveyed financial institutions reporting tightened credit standards minus the weighted percentage reporting eased credit standards. The variable is transformed into a z-score at the country level. An increase of this index implies a net tightening.
  - Source: Haver Analytics; IMF staff estimates
- Financial Conditions Index
  - Description: For methodology and variables included in the FCI, see the Online Appendix. Positive values of the FCI indicate tighter-than-average financial conditions.
  - Source: Authors' estimates
- Long-term Interest Rate
  - Description: 10-year government bond yield
  - Source: Bloomberg Finance L.P
- Corporate Spreads
  - Description: Corporate yield of the country minus sovereign yield of the benchmark country; JPMorgan Corporate Emerging Markets Bond Index Broad is used for emerging market economies where available. The variable is transformed into a z-score at the country level.
  - Source: Bloomberg Finance L.P.; Thomson Reuters Datastream
- Credit to Private Sector (baseline)
  - Description: Credit provided to the private sector by domestic money banks.
  - Source: IMF, International Financial Statistics
- Credit to Private Sector
  - Description: Total credit to the private non-financial sector in billions of domestic currency
  - Source: BIS CRE Table F2.3
- Credit to Corporate Sector
  - Description: Total credit to non‑financial corporations in billions of domestic currency
  - Source: BIS CRE Table F4.3
- Credit to Household Sector
  - Description: Total credit to households in billions of domestic currency
  - Source: BIS CRE Table F3.3
- Credit to Private Sector from Domestic Banks
  - Description: Bank credit to the private non-financial sector in billions of domestic currency
  - Source: BIS CRE Table F2.6
- Cross-Border Credit to Private Sector
  - Description: Cross-border claims of BIS-reporting banks on banks and non-banks in billions of US dollars
  - Source: BIS LBS A6.1-F
- Cross-Border Credit to Banks
  - Description: Cross-border claims of BIS-reporting banks on banks in billions of US dollars
  - Source: BIS LBS A6.1-F
- Cross-Border Credit to Non-Banks
  - Description: Cross-border claims of BIS-reporting banks on non-banks in billions of US dollars
  - Source: BIS LBS A6.1-F
- Systemic Banking Crisis
  - Description: Dummy for systemic banking crisis start year
  - Source: Laeven and Valencia (2018)

*Italicized line denoting the source provided separately by the pipeline.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019207-print-pdf.pdf_