## c2 - Introduction

## Source details

**Canonical URL:** [c2 - Introduction](https://www.imf.org/-/media/files/publications/gfsr/2018/april/chapter-2/doc/c2.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2018/april/chapter-2/doc/c2.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2018/april/chapter-2/doc/c2.pdf.json)

---

### Overview and motivation
- After years of accommodative monetary policy, financial conditions remain loose in most advanced and emerging market economies.
- Withdrawal of monetary policy stimulus has begun in several advanced economies and is expected to keep proceeding at a gradual pace in the United States.
- Spreads (including corporate spreads) have remained compressed by historical standards in both advanced and emerging market economies.
- Corporate credit-to-GDP ratios remain at or near their historical highs in both advanced economies and emerging markets.
- The share of bond issuance by nonfinancial corporations with low ratings (high-yield and BBB-rated bonds) has rebounded from its crisis trough in the United States and is at or near an all-time high in the euro area and the United Kingdom.
- Policy concern: nonfinancial corporate credit may have been excessively allocated to risky firms, potentially jeopardizing financial stability.
- October 2017 GFSR flagged that some indicators of nonfinancial corporate vulnerability had picked up in several major economies.

### Key research questions
- How has the riskiness of credit allocation evolved across advanced economies and emerging markets?
- How does the riskiness of credit allocation relate to financial conditions over time and to episodes of high credit growth and strong risk appetite?
- To what extent does the riskiness of credit allocation predict financial sector stress and downside risks to GDP growth, and at what horizons?
- How do regulatory, supervisory, and legal environments affect the dynamics and cyclicality of the riskiness of credit allocation?

### Empirical approach and data
- Sample: 55 economies (26 advanced economies and 29 emerging market economies) over the 1991–2016 period.
- Core data source: Worldscope database (annual financial variables for listed firms); robustness checks with Orbis (50 economies, start date 2000) and WIND for China (about 37,000 firm-year pairs, 1995–2016).
- Post-cleaning Worldscope observations: about 500,000 nonfinancial firm-year observations.
- Four firm-level vulnerability indicators:
  - Leverage ratio (total debt to total assets).
  - Interest coverage ratio (ICR; EBITDA to interest expenses).
  - Debt-to-profit ratio (debt overhang; total debt to EBITDA).
  - Expected default frequency (EDF; Black-Scholes-Merton based).
- Construction of riskiness-of-credit-allocation measure:
  - Firms sorted by changes in net debt to lagged total assets into five equal-size bins; “top issuers” = firms with largest increases in debt; “bottom issuers” = firms with largest decreases in debt.
  - For each vulnerability indicator, compute decile ranking (1 to 10) of firms in country-year; measure = average decile of top issuers minus average decile of bottom issuers.
  - Country-specific mean subtracted to ensure comparability and remove sectoral composition effects.
  - Interpretation: positive (negative) values indicate allocation riskiness above (below) country sample average.

### Conceptual drivers and ambiguity of implications
- Main drivers:
  - Financial accelerator (procyclicality via higher firm net worth and eased financing frictions).
  - Time-varying investor beliefs and risk appetite.
  - Search-for-yield by intermediaries with maturity mismatch.
  - Deterioration of banks’ screening capacity during prolonged expansions.
  - Bank capital effects with theoretical ambiguity (higher capital can expand credit to poorer-fundamentals firms in some models).
- Ambiguity: increased lending to riskier firms can be rational and profitable or reflect poorer screening/excessive risk taking that creates vulnerabilities.

### Main empirical findings — predictive power and cyclical properties
- The riskiness of credit allocation helps predict:
  - Full-blown banking crises, financial sector stress, and downside risks to growth at horizons up to three years.
  - A one standard deviation increase in the riskiness measure increases the odds of a systemic banking crisis by a factor of about 4.
  - Gain in explanatory power from adding riskiness variable = between 11 and 25 percentage points.
- Banking sector equity stress:
  - The riskiness measure adds predictive power for banking sector equity stress for horizons t to t + 3.
  - A one standard deviation increase in the riskiness measure increases the odds of bank equity stress by a factor of 1.3 to 2 across horizons.
  - Banking sector equity stress defined as annual excess equity return of banking sector below country-specific mean by at least one standard deviation.
- Downside growth risks:
  - The riskiness of credit allocation is strongly related to the median and left tail (20th percentile) of the distribution of cumulative real GDP growth one to three years ahead.
  - Credit booms accompanied by a rise in allocation riskiness signal elevated risks to growth two and three years ahead.
  - When credit is stagnant or falling, the riskiness of credit allocation has a negligible effect on downside risks to GDP growth.
- Global dynamics:
  - Global riskiness followed a cyclical pattern over the past 25 years, rebounded since its post-global-financial-crisis trough, and was slightly below its historical average at end-2016.
  - Historical pattern: elevated late 1990s; fell 2000–04; historical low in 2004; rose steeply 2004–08 and peaked at crisis onset.
- Country heterogeneity:
  - Association between credit growth and allocation riskiness is stronger when lending standards are easier, domestic financial conditions are looser, credit spreads are lower, and global risk appetite is higher.
  - Macroprudential tightening, greater supervisory independence, smaller government footprint in nonfinancial corporate sector, and stronger minority shareholder protection reduce the likelihood that credit expansion is associated with riskier allocation.

### Quantified effects and key numeric facts (preserved)
- Sample period for country figures: 1995–2016.
- Shaded threshold for global real GDP growth in figures: 2.5 percent.
- Shaded threshold for country growth in figures: 15th percentile of the growth distribution.
- One standard deviation increase in change of credit-to-GDP ratio = 5.5 percentage points.
- Associated increase in riskiness of credit allocation = 0.12–0.25 standard deviation (range across measures).
- One standard deviation increase in riskiness measure multiplies odds of systemic banking crisis by about 4 (contextual baseline probability of crisis in sample ≈ 5 percent; fourfold increase → 21 percent).
- Gain in explanatory power when adding riskiness variable = between 11 and 25 percentage points.
- Forecast horizon for banking sector stress predictive power: horizons 0, 1, 2, 3 years.
- Predictive multiplier for banking sector stress: one standard deviation increase in riskiness increases odds by factor 1.3 to 2 across horizons.
- Panel logit estimates for systemic banking crisis (selected coefficients):
  - Change in Credit-to-GDP Ratio: 0.202*** (0.0699) in one specification.
  - Riskiness_Leverage: 1.924*** (0.674); Riskiness_ICR: 2.533*** (0.861); Riskiness_Debt Overhang: 2.087*** (0.461); Riskiness_EDF: 2.113*** (0.734).
- Quantile regression results for cumulative real GDP growth (t to t + 3) (selected coefficients):
  - Change in Credit-to-GDP Ratio: values include –0.232*** (0.0335), –0.268*** (0.0358), –0.239*** (0.0364) across specifications.
  - Riskiness_Leverage at 20th percentile: –0.468*** (0.144); at 50th percentile: –0.480*** (0.107).
  - Selected interaction: Change in Credit-to-GDP Ratio × Riskiness_Leverage = –0.0820*** (0.0288).
  - Riskiness_ICR interactions: Change in Credit-to-GDP Ratio × Riskiness_ICR = –0.237*** (0.0467) in one specification.

### Measurement properties and robustness
- Measures constructed from firm decile rankings minimize influence of outliers, abstract from changes in mean levels, and normalize across countries.
- Decile approach: higher decile = larger value of underlying vulnerability.
- Minimum sample-size thresholds typically used: 40 firms for Worldscope indicators; Orbis and WIND use 50 firms for country-year retention.
- EDF requires firm equity market data and cannot be computed for unlisted firms; Orbis robustness uses EBIT instead of EBITDA for debt overhang and omits ICR where interest expense data are poor.
- Robustness:
  - Similar results when using Orbis (50 economies, start 2000) and WIND (China) samples.
  - Results robust to alternative credit measures (BIS series), additional controls (corporate spreads, median firm leverage), and different estimators (two-way clustered standard errors).

### Policy implications and recommendations
- Monitor both the volume and allocation of credit for financial stability surveillance.
- Supervisory actions:
  - Monitor credit origination standards and the riskiness of credit allocation continuously.
  - Intensify supervisory scrutiny during episodes of large credit expansion and loose financial conditions; require corrective action if needed.
- Data and surveillance:
  - Collect granular firm-level financial statement data and accelerate data availability for timely measurement.
- Macroprudential and institutional tools:
  - Tighten macroprudential policies to reduce cyclicality of allocation riskiness (evidence that net tightening of capital conservation buffers, ceilings/penalties on credit growth, and minimum leverage ratio tightening reduce the credit-growth–allocation-risk link in several specifications).
  - Strengthen supervisory independence and corporate governance (minority shareholder protection) to reduce cyclicality.
  - Consider increased provisioning requirements and thicker countercyclical capital buffers when allocation riskiness rises during strong credit growth; calibrate buffers with allocation riskiness in mind.
  - Discourage directed-credit policies that ignore underlying credit risk during strong credit growth.
- Caveats and research needs:
  - Evidence on macroprudential policy effects is tentative; further research needed on calibration, timing, and GDP growth trade-offs of macroprudential actions, and on the role of thicker capital buffers in improving macro-financial outcomes following a rise in allocation riskiness.

*Source: c2 - Introduction (PDF chapter).*

### Introduction

### c2 - Introduction

### Overview of recent financial conditions and concerns
- After years of accommodative monetary policy, financial conditions remain loose in most advanced and emerging market economies.
- Withdrawal of monetary policy stimulus has begun in several advanced economies and is expected to keep proceeding at a gradual pace in the United States.
- Despite a recent rebound in financial market volatility, financial conditions have remained loose and spreads (including corporate spreads) have remained compressed by historical standards in both advanced and emerging market economies.
- Corporate credit-to-GDP ratios remain at or near their historical highs in both advanced economies and emerging markets.
- The share of bond issuance by nonfinancial corporations with low ratings (high-yield and BBB-rated bonds) has rebounded from its crisis trough in the United States and is at or near an all-time high in the euro area and the United Kingdom.

### Policy concern and motivation
- Concern: nonfinancial corporate credit may have been excessively allocated to risky firms, especially in advanced economies, potentially jeopardizing financial stability.
- Persistently easy financial conditions can lead to a search for yield that pushes investors beyond traditional risk tolerance into riskier investments.
- October 2017 GFSR flagged that some indicators of nonfinancial corporate vulnerability had picked up in several major economies.
- Empirical and country-level studies document that the composition of corporate credit flows changes with financial conditions and that the riskiness of corporate credit allocation is procyclical.

### Key research questions addressed in the chapter
- How has the riskiness of credit allocation evolved in recent years across a broad spectrum of advanced economies and emerging markets?
- How does the riskiness of credit allocation relate to measures of financial conditions over time? Does it generally rise during periods of high credit growth? Is it more likely to increase when high credit growth is associated with strong risk appetite?
- To what extent does the riskiness of credit allocation help predict financial sector stress and downside risks to GDP growth? How far in advance can it predict these occurrences? Do its predictive properties reinforce those of credit growth documented in the literature?
- How is the dynamic of the riskiness of credit allocation affected by the regulatory, supervisory, and legal environments? What is the link between the cyclicality of the riskiness of credit allocation and common indicators of banking sector soundness?

### Empirical approach and data
- No cross-country measures readily capture the riskiness of total credit flows across firms; the chapter constructs several measures mapping the flow of credit across firms to the distribution of firm-level vulnerability indicators for 55 economies since 1991.
- Sample: 55 economies comprising 26 advanced economies and 29 emerging market economies over the 1991–2016 period.
- Data source: Worldscope database (annual financial variables for listed firms); Annex 2.1 provides sample details and data cleaning explanations.
- Four firm-level vulnerability indicators are used:
  - Leverage ratio
  - Interest coverage ratio (ICR)
  - Debt-to-profit ratio (debt overhang)
  - Expected default frequency (EDF) — a market-based indicator of credit risk
- Construction of the riskiness-of-credit-allocation measure:
  - Compute raw measure as the average vulnerability indicator among firms whose debt (sum of loans and bonds) increases the most minus the average among firms whose debt increases the least—or declines the most.
  - Transform raw measure by subtracting its country-specific mean to remove influence of country-specific sectoral composition and ensure cross-country and cross-measure comparability.
  - Interpretation: an increase in the measure signals that the vulnerability of firms getting relatively more credit has risen relative to the vulnerability of firms getting relatively less credit. A positive (negative) value indicates the riskiness is above (below) its country sample average.
  - Box 2.1 provides a detailed explanation of construction and interpretation.

### Conceptual framework: drivers and channels
- Main drivers of variation in the riskiness of credit allocation:
  - Procyclicality via financial accelerator: positive macro shocks or lower interest rates raise firm net worth, easing financing frictions and increasing access for higher-leverage firms.
  - Time-varying investor beliefs, risk appetite, and perceptions of economic uncertainty affect credit spreads and expected volatility; optimism can push credit to riskier firms.
  - Search-for-yield by intermediaries with long-term liabilities and short-term assets when monetary conditions are loose.
  - Deterioration of banks’ screening capacity during sustained credit expansions (loss of institutional memory, profitability pressures from intermediating larger volumes).
  - Bank capital effects: higher bank capital can expand credit to poorer-fundamentals firms (Holmstrom and Tirole 1997); theoretical ambiguity exists on how short-term rates, bank leverage, and risk taking interact.
- Ambiguity in implications:
  - Increased lending to riskier firms can be rational and profitable (healthy financial functioning) or reflect poorer screening/excessive risk taking leading to vulnerabilities and potential harm to future growth.

### Main empirical findings
- Taking the riskiness of credit allocation into account helps better predict full-blown banking crises, financial sector stress, and downside risks to growth at horizons up to three years. Thus, the riskiness of credit allocation is an indicator of financial vulnerability.
- A period of high credit growth is more likely to be followed by a severe downturn over the medium term if it is accompanied by an increase in the riskiness of credit allocation.
- When credit is stagnant or falling, the riskiness of credit allocation has a negligible effect on downside risks to GDP growth.
- Global dynamics:
  - The riskiness of credit allocation at the global level followed a cyclical pattern over the past 25 years.
  - It rebounded since its post-global-financial-crisis trough and was slightly below its historical average at the end of 2016 (the latest data point).
  - The global dynamic is broadly the same across the four borrower vulnerability indicators used.
  - Historical pattern: elevated levels in the late 1990s; fell in 2000–04 after the Asian and Russian crises and dot.com burst; historical low in 2004; rose steeply during 2004–08 and peaked at the onset of the global financial crisis.
- Cross-country and conditional associations:
  - At the country level, the riskiness of credit allocation is more strongly associated with credit growth when lending standards are easier, when domestic financial conditions are looser, when credit spreads are lower, and when global risk appetite is higher.
  - A period of credit expansion is less likely to be associated with a riskier credit allocation when:
    - Macroprudential policy has been tightened,
    - Banking supervisor is more independent,
    - Government has a smaller footprint in the nonfinancial corporate sector,
    - Minority shareholder protection is greater.

### Chapter structure and next steps
- The chapter:
  - Lays out a stylized conceptual framework for macro-financial shocks and the riskiness of credit allocation.
  - Describes construction of the new measures, their evolution at the global level and in selected economies, their cyclical properties, and relationship to various indicators of financial conditions.
  - Presents empirical analysis of the relationship between the new indicators and future financial instability as well as downside risks to GDP growth.
  - Explores determinants of the riskiness of credit allocation and its cyclicality, including macroprudential policies and aspects of the supervisory, legal, and institutional frameworks.
  - Concludes with policy implications.

*Source: c2 - Introduction (PDF chapter)*

### Annex 2.1 for a precise definition of the firm-level indicators used in

### Annex 2.1 for a precise definition of the firm-level indicators used in

### Overview and measurement
- The chapter constructs four firm-level measures to build composite indicators of the "riskiness of credit allocation":
  - Leverage-based measure
  - Interest coverage ratio–based (ICR-based) measure
  - Expected default frequency–based (EDF-based) measure
  - Debt overhang–based measure
- Measures are presented as indexes (global median or country medians) and are often shown as a simple two-year moving average.
- Shaded periods in figures indicate episodes when global real GDP growth was less than 2.5 percent or country growth was below the 15th percentile of the growth distribution, depending on the figure.

### Country-level patterns and complementarity across measures
- The four measures generally display similar dynamics, but there are notable differences across countries and periods; correlation is generally high but smallest between the leverage-based and the EDF-based measures.
- United States and Japan:
  - Dynamics in the United States and Japan are very similar in both cyclicality and magnitude across much of the sample.
  - For 2014–16, the riskiness decreased in the United States to a relatively low level, while in Japan it remained relatively high in historical perspective.
  - Corporate leverage increased across the board in the United States during 2010–16; since increases were similar across groups of firms, relative comparisons between groups used to track distributional changes may not rise over this period.
- Spain and Germany:
  - Spain experienced a credit boom from the late 1990s to the mid-2000s, followed by a deep recession during the global financial crisis and the euro area sovereign debt crisis; measures reflected a steep rise in riskiness until 2008 and a sudden large fall thereafter.
  - Germany did not have a credit boom during the 20-year period and exhibited narrower variation; the measure moved into positive territory in recent years, suggesting a higher level of risk taking.
- India and China:
  - India's evolution broadly followed global patterns and was at a relatively low level in 2016.
  - China shows weaker synchronization with global developments, with peaks and troughs occurring with a two- to three-year lag; a peak in 2009–10 is consistent with the 2008 stimulus plan leading to misallocation of credit.
- Korea and United Kingdom:
  - In Korea, only accounting-based measures indicated high riskiness before the late 1990s crisis; EDF-based measures did not signal problems at that time, suggesting market optimism and that accounting measures better reflected fundamentals.
  - Disconnects between measures appear for the United Kingdom during the 1990s and 2010s; volatility in firm-level equity prices may affect the EDF-based measure.

### Cyclicality, drivers, and amplification by financial conditions
- Regression analysis indicates the riskiness of credit allocation is procyclical:
  - It 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 all four measures.
- Quantified effect of credit expansion:
  - 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 measure.
- Amplification by loose financial conditions or lending standards:
  - Credit expansions accompanied by loose financial conditions or loose lending standards are more likely driven by credit supply shifts and higher risk appetite, producing riskier allocations.
  - Specific components that matter:
    - Low corporate credit spreads (or high global risk appetite proxied by the Chicago Board Options Exchange Volatility Index [VIX]) during credit expansions lead to riskier allocations than expansions with high credit spreads (or low global risk appetite).
    - A higher stock market price-to-book ratio is associated with a higher level of the riskiness of credit allocation.
- Robustness:
  - Trends and properties remain when using a different sample covering both listed and unlisted firms (Orbis database) for a smaller set of countries (50 economies) from 2000; similarity across datasets is noted despite differences in cross-sectional coverage.

### Predictive power for financial crises, banking stress, and growth risks
- Dynamics around systemic banking crises:
  - The riskiness of credit allocation follows a clear inverted-U shape around crisis episodes: it rises gradually during the five years preceding a crisis, reaches a relatively high level at crisis onset (time 0), and falls following the onset.
  - This pattern holds across all four firm-level measures and signals forthcoming crises better than conventional individual corporate vulnerability indicators, which tend to pick up only after crises begin.
- Effect on odds of systemic banking crisis:
  - Regression analysis (controlling for change in credit volumes and financial conditions) finds that a greater riskiness of credit allocation increases the odds of a future systemic banking crisis.
  - A one standard deviation increase in the riskiness measure increases the odds of a crisis by a factor of about four.
  - The gain in explanatory power when adding the riskiness variable is between 11 and 25 percentage points.
  - (Contextual example from sample: the probability of observing a crisis is about 5 percent; odds are about 5.3 percent. A fourfold increase would raise the odds to 21 percent.)
- Forecasting banking sector equity stress:
  - The riskiness of credit allocation adds predictive power for banking sector equity stress for horizons from zero to three years.
  - A one standard deviation increase in the riskiness measure increases the odds of bank equity stress by a factor of 1.3 to 2 across horizons.
  - Banking sector equity stress is defined as the annual excess equity return of the banking sector being lower than the country-specific mean by at least one standard deviation.
- Predicting low realizations of future GDP growth:
  - The chapter extends cross-country regressions to test whether the riskiness of credit allocation predicts downside risks to growth; econometric framework details are provided in Annex 2.3.

### Key quantitative findings (numeric fidelity preserved)
- Sample period for country figures: 1995–2016.
- Shaded threshold for global real GDP growth in some figures: 2.5 percent.
- Shaded threshold for country growth in some figures: 15th percentile of the growth distribution.
- Effect size of credit-to-GDP change:
  - One standard deviation increase in the change of the credit-to-GDP ratio = 5.5 percentage points.
  - Associated increase in riskiness of credit allocation = 0.12–0.25 standard deviation (range across measures).
- Effect on crisis odds:
  - One standard deviation increase in riskiness measure multiplies odds of systemic banking crisis by about 4.
  - Gain in explanatory power when adding riskiness variable = between 11 and 25 percentage points.
- Forecast horizon for banking sector stress predictive power: t to t + 3 (horizons 0, 1, 2, 3 years).
- Predictive multiplier for banking sector stress:
  - One standard deviation increase in riskiness increases odds by a factor of 1.3 to 2 across horizons.

*International Monetary Fund | April 2018*

### CHAPTER 2 ThE RISkINESS OF CREdIT ALLOCATION: A SOuRCE OF FINANCIAL VuLNERABILITY?

### CHAPTER 2 ThE RISkINESS OF CREdIT ALLOCATION: A SOuRCE OF FINANCIAL VuLNERABILITY?

### Riskiness of credit allocation — key empirical findings
- A riskier credit allocation signals downside risks to growth in the short to medium term. The analysis examines the predictive power of the riskiness of credit allocation on two percentiles (20th and 50th) of cumulative real GDP growth one to three years into the future.
- The riskiness of credit allocation is strongly related to the median and left tail of the growth distribution over all horizons.
- These effects are in addition to those of changes in the credit-to-GDP ratio and financial conditions.
- The effect on the downside risks to growth is significant when measures of the riskiness of credit allocation are constructed based on a sample that covers unlisted as well as listed firms.
- The effects of a riskier credit allocation complement those of credit expansions on growth-at-risk over the medium term: credit booms accompanied by a rise in the riskiness of credit allocation signal elevated risks to growth two and three years ahead.
- Conversely, during credit contractions or relatively soft credit expansions, a higher riskiness of credit allocation does not increase downside risks to future GDP growth; at a three-year horizon, when the change in the credit-to-GDP ratio is low by historical standards, an increase in risk taking has no significant impact on downside risks to growth.
- Solid colored bars in the chapter’s figures indicate that effects are statistically significant at the 10 percent level or higher; empty bars indicate absence of statistical significance.

### Quantitative relationships and scenarios
- Predictive targets: 20th percentile and 50th percentile of cumulative real GDP growth from year t to year t + h, with h = 1, 2, 3.
- The analysis shows the impact of a one unit increase in the riskiness of credit allocation on the 20th and 50th percentiles of the distribution of future cumulative GDP growth.
- The association between riskiness and downside growth risks depends on the size of credit expansion:
  - High credit growth (defined as one standard deviation above mean credit growth) together with rising allocation riskiness implies larger negative impacts on the 20th and 50th percentiles of future cumulative GDP growth two and three years ahead.
  - Low credit growth (one standard deviation below mean credit growth) can weaken or reverse the association between higher riskiness of allocation and downside growth risks.

### Structural determinants and policy settings
- Bank capital:
  - Bank capital appears to have little significant effect on the cyclicality of the riskiness of credit allocation. Greater buffers are generally associated with greater cyclicality, but evidence is not robust.
- Macroprudential policies:
  - Macroprudential policy tightening reduces the cyclicality of the new vulnerability measure.
  - Tightening of minimum leverage ratio and changes in ceilings and penalties related to credit growth dampen the increase in the riskiness of credit allocation associated with faster credit growth.
  - Increases in capital conservation buffers reduce the level of the riskiness of credit allocation.
  - Tightening of minimum capital requirements is found to be associated with a nonrobust increase in the riskiness of credit allocation, suggesting reverse causality. Loan provisioning requirements are not found to have any significant effects, either in level or when interacted with the change in the credit-to-GDP ratio.
- Supervisory, legal, and institutional factors:
  - Greater supervisory independence is associated with reduced cyclicality of the riskiness of credit allocation.
  - A smaller government footprint in the nonfinancial corporate sector reduces cyclicality of the measure.
  - Greater protection of minority shareholders reduces cyclicality and highlights the role of corporate governance for financial stability.
- Interaction effects summarized graphically: the range of impact of a contemporaneous increase in the change in the credit-to-GDP ratio by one standard deviation on the four measures depends on policy and institutional settings (leverage ratio constraint, ceilings and penalties on bank credit growth, independence of supervisory authority from banks, rareness of state-owned enterprises, and minority shareholder protection).

### Measurement methodology (construction of the riskiness-of-allocation measure)
- Four firm-level vulnerability indicators used: leverage (total debt to total assets), debt overhang (total debt to earnings before interest, taxes, depreciation, and amortization [EBITDA]), interest coverage ratio (ICR; EBITDA to interest expenses), and expected default frequency.
- For each firm-year and each vulnerability indicator:
  - Each firm is assigned the value (from 1 to 10) of its decile in the distribution of the indicator in the country where it is located. A higher decile represents a larger value of the underlying vulnerability.
  - Firms are sorted by changes in net debt to lagged total assets into five equal-size bins. Firms in the bin with the largest increases in debt are “top issuers”; firms in the bin with the largest decreases in debt are “bottom issuers.”
  - The measure equals the difference between the average vulnerability decile for the top issuers and the average for the bottom issuers.
- Interpretation:
  - The measure answers: what is the evolution of the vulnerability profile of firms that are accumulating debt the fastest relative to that of firms reducing debt the fastest?
  - Sign adjustments ensure the measure rises when the vulnerability of firms whose total debt issuance is the largest is increasing (for debt overhang, deciles of EBITDA to debt are used to avoid classifying firms with negative earnings as low-vulnerability firms).
- Properties and rationale:
  - The measure abstracts from changes in the mean and shape of the distribution; only firm rankings matter.
  - Using deciles minimizes influence of outliers, avoids picking up secular trends, normalizes across countries, and simplifies comparisons across indicators. A downside is loss of information about cross-sectional dispersion.
  - The measure reflects broad debt (loans and bond financing) and the continuous nature of firm vulnerability and default risk.
- Example:
  - If firm leverage increases by 5 percentage points for all firms and issuance increases equally, mean leverage increases by 5 percentage points but the measure of allocation riskiness does not change.
  - If leverage increases by 5 percentage points for top issuers, decreases by 5 percentage points for bottom issuers and remains unchanged for others, mean leverage does not change but the measure of allocation riskiness rises.

### Policy implications and recommendations
- Both the volume and allocation of credit matter for financial stability; policymakers and supervisors should monitor both.
- Supervisory actions:
  - Monitor credit origination standards and the riskiness of credit allocation on a continuous basis.
  - Intensify supervisory scrutiny during episodes of large credit expansion and loose financial conditions.
  - Require corrective action if needed.
- Data and surveillance:
  - The riskiness of credit allocation can be measured using firm-level financial statement data available in many countries; usefulness depends on the speed of data availability.
  - Policymakers should engage in efforts to collect granular data as swiftly as possible.
- Policy tools to tame allocation riskiness during credit expansions:
  - Strengthen supervisory independence to better control lending and origination standards in good times.
  - Promote sound corporate governance (to reduce “gamble for resurrection” behavior by vulnerable firms’ managers).
  - Use macroprudential policies (including tightening some regulatory capital requirements) to reduce banks’ ability or willingness to lend to vulnerable firms.
  - Consider increased provisioning requirements and thicker countercyclical capital buffers when allocation riskiness rises during strong credit growth; the calibration of capital buffers should consider the riskiness of credit allocation.
  - Discourage policies that direct credit to certain firms or sectors without due consideration of underlying credit risk during strong credit growth.
- Caveats and research needs:
  - Evidence on macroprudential policy effects is tentative; further research is needed to understand how macroprudential policy affects the riskiness of credit allocation.
  - Exploring calibration and timing of macroprudential actions and associated GDP growth trade-offs is an essential next step; in particular, examining the role thicker capital buffers could play in improving macro-financial outcomes following a rise in allocation riskiness is warranted.

### Recent global pattern and monitoring needs
- The riskiness of credit allocation at the global level rebounded since its post-global-financial-crisis trough and was back to its historical average at the end of 2016.
- The relatively mild credit expansion in recent years, combined with postcrisis regulatory tightening, contributed to a softer rebound in the riskiness of credit allocation than might be expected given very loose financial conditions.
- Country-level heterogeneity is important; rises in allocation riskiness have been more pronounced in certain countries.
- As financial conditions loosened further in 2017, allocation riskiness might have continued to rise, warranting close monitoring and heightened vigilance.
- Riskiness of credit allocation to households may also be relevant and may not follow the same patterns as corporate allocation; monitoring household allocation riskiness is difficult for many countries, but selected household surveys suggest increased indebtedness of lower-income, more vulnerable households in various countries.

*Source: IMF staff estimates and chapter text.*

### Box 2.1. Measuring the Riskiness of Credit Allocation

### Box 2.1. Measuring the Riskiness of Credit Allocation

### Overview
- Presents firm-level measures of the "riskiness of credit allocation" constructed from Worldscope firm data to analyze cross-country distributions and dynamics.
- The panel covers 55 economies for the period 1991–2016.
- Data are demeaned at the country level for histogram construction and analysis.

### Data and Sample
- Worldscope database: core sample after cleaning contains about 500,000 nonfinancial firm-year observations from 55 economies during 1991 to 2016.
- Orbis database: used for robustness analysis; after cleaning and selection, Orbis sample covers 50 economies. Start date for Orbis-based analysis is 2000 due to data availability considerations.
- WIND database: used for China analysis (Box 2.2); about 37,000 firm-year pairs from 1995 to 2016 are used in that analysis. Ownership information is available for most firms only from 2004.
- Filtering rules applied across databases:
  - Drop financial sector firms (except real estate).
  - Drop observations with incompatible values (negative market capitalization, total assets, total debt, total liability, interest expenses; operating profit margin or short-term debt/total debt exceeding 100 percent).
  - Keep observations only with full information on net debt issuance, leverage, EBITDA (or EBIT in Orbis robustness), and market capitalization (and additional variables as required per indicator).
  - Economy-year pairs retained only if they meet minimum-firm thresholds: typically no fewer than 40 firms (Orbis uses at least 50 nonfinancial firms; WIND uses at least 50 nonfinancial firms for China).
  - Specific indicator minimums: for interest coverage ratio–based indicator, minimum of 40 observations for interest expenses (exception made for Ireland with 38–39 in some years); for debt overhang–based and EDF-based indicators, minimum of 40 observations for non-zero debt and EDF respectively.

### Measures and Definitions
- Leverage ratio: defined as the ratio of total debt to total assets.
- Interest coverage ratio (ICR): defined as the ratio of interest expenses to EBITDA.
- Debt overhang measure: defined as the ratio of total debt to EBITDA.
- Expected default frequency (EDF): computed using the Black-Scholes-Merton model as in Vassalou and Xing (2004). Ingredients: value of equity, sum of short-term debt and half of long-term debt and interest payments, expected returns, the risk-free rate, and volatility of the price of equity.
- Return on assets (used in Box 2.2): defined as the ratio of EBITDA to total assets.
- In Orbis robustness analysis, EBIT is used instead of EBITDA for the debt overhang indicator due to EBITDA availability; ICR is not used in Orbis robustness due to poor availability of interest expense data.
- EDF computation requires firm-level equity market information and therefore cannot be done for unlisted firms.

### Distributional Properties and Measurement Notes
- Country-level indicators are demeaned before constructing cross-country histograms.
- The distributions of the indicator values across country-year observations have the shape of a bell curve and have a standard deviation of about one.
- Differences in the average value of the indicator across countries may reflect differences in the industrial composition of their corporate sectors; such cross-country differences cannot be interpreted to mean that some countries have inherently riskier credit allocations.
- The long-term average of the measure in each country could be interpreted as representing the neutral allocation of credit in the absence of cyclical fluctuations.

### Usage and Interpretation
- The measures are intended to capture dimensions of credit allocation quality (leverage, ICR, debt overhang, EDF) at the firm level and aggregate them to analyze cross-country and time-series patterns.
- Minimum sample-size thresholds and data-cleaning rules are critical to ensure robustness of country-year estimates.
- EDF cannot be computed for unlisted firms; this limits coverage of the EDF-based indicator relative to others when unlisted firms are prevalent.
- Where data availability is limited in a database (for example, Orbis in the 1990s), start dates and sample balance are adjusted (e.g., Orbis analysis begins in 2000).

### Key Quantitative Facts
- Panel coverage: 55 economies, 1991–2016 (Worldscope-based analysis).
- Post-cleaning Worldscope observations: about 500,000 nonfinancial firm-year observations.
- Orbis sample covers 50 economies (start date for Orbis analysis: 2000).
- WIND (China) sample: about 37,000 firm-year pairs from 1995 to 2016; ownership data mainly from 2004 onward.
- Standard deviation of country-level demeaned indicator distributions: about one.
- Minimum firm-count thresholds typically used: 40 firms (Worldscope indicators), 50 firms (Orbis and WIND country-year retention).

*International Monetary Fund | April 2018 — Box 2.1. Measuring the Riskiness of Credit Allocation*

### Annex 2.2. The Determinants of the Riskiness

### Annex 2.2. The Determinants of the Riskiness of Credit Allocation

### Overview and empirical approach
- Purpose: Provide a general overview of empirical methodologies used to analyze cyclical determinants of the riskiness of credit allocation and its relationship to institutional and policy variables.
- Robustness criterion: A finding is defined as robust across measures when the regression coefficient is significant for at least two of the four measures and when the sign is identical across all four measures. Consistency of the signs of the effects in level and in interaction is also required.
- Robustness checks reported: alternative data sources for credit (BIS total credit to the nonfinancial private sector and credit to the nonfinancial corporate sector), different ways to capture the business cycle (output gap) and credit cycle (real credit growth), and two-way clustered standard errors at country and year levels.
- Fixed effects and standard errors: country (α_i^X) and year (γ_t^X) fixed effects are included. Standard errors are clustered at the country level for all specifications. Instrumenting GDP growth and change in credit-to-GDP by lagged values yields quantitatively very similar coefficients.

### Cyclicality of the riskiness of credit allocation (method and key results)
- Empirical specification (A2.2.1):  
  Riskiness_{i,t}^X = α_i^X + γ_t^X + β_1^X ∆Credit_{i,t} + β_2^X ∆GDP_{i,t} + β_3^X Appreciation_{i,t} + ε_{i,t}^X  
  where X ∈ {leverage, interest coverage ratio, debt overhang, expected default frequency}; ∆Credit = change in bank credit to nonfinancial private sector to nominal GDP; ∆GDP = real GDP growth; Appreciation = domestic currency appreciation against the US dollar.
- Main reported estimates (Annex Table 2.2.1) — Dependent variable: riskiness based on leverage:
  - Change in Credit-to-GDP Ratio: 0.05*** (column 1, standard error (0.01)); 0.04*** (column 2, (0.01)); 0.06*** (column 3, (0.01)). Robustness count reported as 44 in columns (4)-(5).
  - Real GDP Growth: 0.08*** ((0.02)); 0.12*** ((0.02)); 0.05* ((0.03)).
  - Appreciation against the US Dollar: –0.04*** ((0.01)); –0.02* ((0.01)); –0.05*** ((0.01)).
  - Observations: 986, 563, 423 (columns 1–3). Number of Countries: 55, 26, 29. R^2: 0.31, 0.34, 0.37.
  - Note: The number of measures (out of four) that have the same sign and are significant at the 10 percent level or higher is reported in columns (4) and (5).

### Financial conditions, lending standards, and the riskiness of credit allocation
- Empirical specification (A2.2.2):  
  Riskiness_{i,t}^X = α_i^X + γ_t^X + β^X Controls_{i,t} + δ^X FC_{i,t} + θ^X × FC_{i,t} × ∆Credit_{i,t} + ε_{i,t}^X  
  where Controls_{i,t} includes ∆Credit, ∆GDP, and Appreciation. FC_{i,t} is the financial conditions index (FCI), financial variables representing components of the FCI, or lending standards. Both ∆Credit and FC are demeaned at country level. δ̂^X is the level effect; θ̂^X is the marginal effect on credit cyclicality.
- Summary of reported coefficient estimates (Annex Table 2.2.2) — Dependent variable: riskiness based on leverage:
  - Change in Credit-to-GDP Ratio (baseline in each column): 0.05*** ((0.02) in column 1; (0.01) in columns 2–5).
  - Bank Lending Standards: level effect –0.10 (standard error (0.07)); interaction: Change in Credit-to-GDP Ratio × Bank Lending Standards = –0.03* ((0.02)). Robustness counts: level effect 41; interaction 44.
  - Financial Conditions Index (FCI): level effect –0.05 ((0.07)); interaction: Change in Credit-to-GDP Ratio × FCI = –0.01** ((0.00)). Robustness counts: level 30; interaction 44.
  - Corporate Credit Spreads: level effect –0.07 ((0.06)); interaction: Change in Credit-to-GDP Ratio × Corporate Credit Spreads = –0.02** ((0.01)). Robustness counts: level 40; interaction 42.
  - Stock Price-to-Book Ratio: level effect 0.20*** ((0.06)); interaction: Change in Credit-to-GDP Ratio × Stock Price-to-Book Ratio = 0.01 ((0.01)). Robustness counts: level 44; interaction 40.
  - Change in Credit-to-GDP Ratio × Log(VIX): interaction = –0.04** ((0.02)). Robustness count 43.
  - Controls: real GDP growth and domestic currency appreciation are controlled in all regressions. Observations and country counts vary by column: Observations 2668, 2466, 3949, 869, 986 (columns 1–5); Number of Countries 214, 137, 51, 55, 155. R^2 ranges 0.31–0.39.
  - Note: Other investigated variables (stock market volatility, credit boom dummy, length and phase dummies of credit boom, cross-border bank-flows-to-GDP ratio, housing price inflation) did not have robust significant impacts and are not included in the table.

### Policy and institutional settings and the riskiness of credit allocation
- Empirical specification (A2.2.3):  
  Riskiness_{i,t}^X = α_i^X + γ_t^X + β^X Controls_{i,t} + ρ^X Z_{i,t} + φ^X × Z_{i,t} × ∆Credit_{i,t} + ε_{i,t}^X  
  where Controls_{i,t} as before. Z_{i,t} represents measures of financial market depth, banking system soundness, macroprudential policy, legal and institutional framework, and supervision quality. Financial development and soundness variables enter with a one-year lag. Due to limited time-series variation, legal/institutional and supervisory quality variables are averaged at country level and enter only as interaction terms.
- Key reported results (Annex Table 2.2.3) — Dependent variable: riskiness based on leverage. Columns present individual-variable regressions and a macroprudential “horse race”:
  - Change in Credit-to-GDP Ratio (baseline across columns): 0.05*** in columns (1)–(5); 0.12*** in column (6); –0.00 in column (7); 0.15*** in column (8). Standard errors: (0.01) for columns (1)–(5); (0.02) for column (6); (0.02) for column (7); (0.03) for column (8).
  - Lag Buffers from Banking Default: level effect 0.01 ((0.01)); interaction Change in Credit-to-GDP Ratio × Lag Buffers from Banking Default = 0.005** ((0.002)). Robustness counts: level 3; interaction 3.
  - Net Tightening of Capital Conservation Buffers: level effect –0.45** ((0.21)); interaction Change in Credit-to-GDP Ratio × Net Tightening of Capital Conservation Buffers = –0.09*** ((0.03)). Robustness counts: level 4; interaction 3. (Column (4) is the horse race between macroprudential measures.)
  - Net Tightening of Minimum Leverage Ratio: level effects –0.29 ((0.20)) and –0.30 ((0.20)) in two specifications; interaction Change in Credit-to-GDP Ratio × Net Tightening of Minimum Leverage Ratio = –0.09* ((0.05)) and –0.09* ((0.05)). Robustness counts: level 4; interaction 0 (level) and 2 (interaction) as reported.
  - Net Tightening on Ceilings and Penalties on Bank Credit Growth: level effect –0.57 ((0.54)) in two specifications; interaction Change in Credit-to-GDP Ratio × Net Tightening on Ceilings and Penalties on Bank Credit Growth = –0.07** ((0.03)) in both specifications. Robustness counts: level 4; interaction 2 (level) and 4 (interaction).
  - Change in Credit-to-GDP Ratio × Independence of Supervisory Authority from Banks = –0.09*** ((0.02)). Robustness counts: 4 and 3 reported.
  - Change in Credit-to-GDP Ratio × Rareness of State-Owned Enterprises = –0.01* ((0.01)). Robustness counts: 4 and 4.
  - Change in Credit-to-GDP Ratio × Minority Shareholder Protection Index = –0.02*** ((0.01)). Robustness counts: 4 and 3.
  - Observations vary by column: 861, 976, 976, 976, 976, 929, 739, 898. Number of Countries: 55, 54, 54, 54, 54, 52, 37, 46. R^2 ranges 0.31–0.34.
  - Note: Column (4) is a horse race between different macroprudential policies. The number of measures (out of four) that have the same sign and are significant at the 10 percent level or higher is reported in columns (9) and (10). For macroprudential policies, robustness information is based on the horse race.

_Source: IMF staff (Annex 2.2 content provided)._

### Annex 2.3. The Riskiness of Credit Allocation

### Annex 2.3. The Riskiness of Credit Allocation

### Methodology and scope
- Empirical frameworks analyze how the riskiness of credit allocation affects:
  - the occurrence of systemic banking crises,
  - banking sector equity-price stress events,
  - downside risks to GDP growth.
- Credit data robustness: results hold using alternative credit data sources, including Bank for International Settlements series (total credit to the nonfinancial private sector and credit to the nonfinancial corporate sector).
- Robustness to additional controls: corporate spreads, median firm leverage (or median interest coverage ratio), and share of high-yield bond issuance.
- Additional robustness checks considered (and found not to have a robust significant impact on riskiness of credit allocation): measures of financial depth (private credit to GDP, bank assets to GDP, bank credit to deposits, external loans and deposits to domestic deposits), capital account openness; measures of banking sector soundness (bank concentration; probability of default of the banking sector; ratios of bank capital to total assets, bank regulatory capital to risk-weighted assets, and bank return on equity); 11 additional macroprudential instruments (including countercyclical capital buffers and minimum capital requirements); supervisory quality indicators (dummy for high supervisory quality based on Basel Core Principles assessments, restructuring power of supervisory authority, degree of independence from political influence); legal and institutional indicators (anti-self-dealing; burden of proof and disclosure index; corruption index; corporate governance opacity index).

- Data and variable treatment:
  - ∆ Credit is the change in the ratio of bank credit to the nonfinancial private sector to nominal GDP.
  - Riskiness measures: leverage indicator; interest coverage ratio indicator; debt overhang indicator; expected default frequency indicator.
  - All explanatory variables enter as the lag of their simple three-year moving average and are demeaned at the country level.
  - The change in the credit-to-GDP ratio is winsorized at the 1 percent level to reduce influence of outliers.
  - Country fixed effects are included in the panel logit specifications.
  - Standard error robustness: results robust to alternative estimators including two-way-clustered standard errors.
  - For downside GDP risk analysis: quantile regressions with nonadditive fixed effects are used to examine the 20th and 50th percentiles of future growth distribution; also a logit regression with a low-growth outturn dummy (low-growth = cumulative real GDP growth over future h years below the 20th percentile of country-specific distribution).

### Impact on systemic banking crisis risk (panel logit framework)
- Model form: panel logit for log odds of start of systemic banking crisis with country fixed effects and controls (change in current-account-balance-to-GDP ratio, real GDP growth, financial conditions index).
- Key coefficient estimates (Annex Table 2.3.1):
  - Change in Credit-to-GDP Ratio:
    - (1) 0.202*** (0.0699)
    - (2) 0.141* (0.0737)
    - (3) 0.05650 (0.0849)
    - (4) 0.07450 (0.100)
    - (5) 0.0808 (0.108)
    - (6) –0.0902 (0.131)
  - Riskiness measures (each reported in separate specifications):
    - Riskiness_Leverage 1.924*** (0.674)
    - Riskiness_Interest Coverage Ratio 2.533*** (0.861)
    - Riskiness_Debt Overhang 2.087*** (0.461)
    - Riskiness_Expected Default Frequency 2.113*** (0.734)
  - Observations: 443, 443, 443, 443, 431, 361 (columns 1–6 respectively)
  - Number of Countries: 21, 21, 21, 21, 20, 17
  - Country Cluster: Yes (all)
  - Country Fixed Effects: Yes (all)
  - Pseudo R2:
    - (1) 0.243
    - (2) 0.353
    - (3) 0.465
    - (4) 0.487
    - (5) 0.515
    - (6) 0.606
- Interpretation (based on presented coefficients): higher measured riskiness of credit allocation (across the four indicators) is associated with significantly higher odds of the start of a systemic banking crisis in these specifications.

### Effect on banking sector equity stress risk (panel logit framework)
- Model form: panel logit for probability of a stress event of banking sector equity prices in window t to t + h (h = 0,...,3). A stress event: annual excess equity return on banking sector (relative to a short-maturity zero-coupon government bond yield) below country-specific mean by more than one standard deviation.
- Controls: financial conditions index included in each estimation.
- Key coefficient estimates (Annex Table 2.3.2):
  - Change in Credit-to-GDP Ratio (columns vary by horizon and specification; selected values shown):
    - –0.000975 (0.0381)
    - 0.01290 (0.0433)
    - 0.02940 (0.0309)
    - 0.03060 (0.0357)
    - 0.03160 (0.0365)
    - 0.03170 (0.0427)
    - 0.03450 (0.0437)
    - 0.0253 (0.0464)
  - Riskiness measures (selected significant estimates):
    - Riskiness_Leverage 0.898*** (0.246) and 0.727*** (0.246)
    - Riskiness_Interest Coverage Ratio 0.690*** (0.256) and 0.717** (0.320)
    - Riskiness_Debt Overhang 0.569** (0.223) and 0.440 (0.271)
    - Riskiness_Expected Default Frequency 0.451* (0.274) and 0.321 (0.296)
  - Observations by column: 573, 573, 573, 573, 552, 552, 505, 505
  - Number of Countries by column: 36, 36, 36, 36, 34, 34, 33, 33
  - Country Cluster: Yes (all)
  - Country Fixed Effects: Yes (all)
  - Pseudo R2 by column: 0.0882, 0.130, 0.0495, 0.115, 0.0517, 0.102, 0.0388, 0.0950
- Interpretation: riskiness indicators (especially leverage and interest coverage ratio measures) are positively and significantly associated with the probability of a banking sector equity-price stress event; the change in credit-to-GDP ratio shows no systematic positive association in these specifications.

### Impact on downside risks to GDP growth (quantile regression and interactions)
- Model form: regress cumulative real GDP growth over future h years (h = 1,...,3) on lagged change in credit-to-GDP ratio, lagged riskiness measure, their interaction, and controls (real GDP growth, financial conditions index). Estimation via quantile regressions with nonadditive fixed effects to examine the 20th and 50th percentiles of the future growth distribution. Also complementary logit for low-growth outturns (below 20th percentile).
- Main results for cumulative real GDP growth rate over future three years (t, t + 3) (Annex Table 2.3.3):
  - Change in Credit-to-GDP Ratio (columns 1–16; selected reported values, all with standard errors in parentheses):
    - –0.232*** (0.0335)
    - –0.268*** (0.0358)
    - –0.239*** (0.0364)
    - –0.254*** (0.0471)
    - –0.228*** (0.0278)
    - –0.291*** (0.0260)
    - –0.172*** (0.0274)
    - –0.231*** (0.0380)
    - –0.224*** (0.0349)
    - –0.268*** (0.0367)
    - –0.192*** (0.0233)
    - –0.207*** (0.0472)
    - –0.219*** (0.0294)
    - –0.272*** (0.0315)
    - –0.213*** (0.0369)
    - –0.128*** (0.0328)
  - Riskiness_Leverage:
    - –0.468*** (0.144) at 20 pt
    - –0.480*** (0.107) at 50 pt
    - Change in Credit-to-GDP Ratio × Riskiness_Leverage:
      - –0.0549** (0.0253)
      - –0.0820*** (0.0288)
  - Riskiness_Interest Coverage Ratio (ICR):
    - –0.927*** (0.207) and –0.421*** (0.118); other columns show –1.306*** (0.237) and –0.391*** (0.0948)
    - Change in Credit-to-GDP Ratio × Riskiness_ICR:
      - –0.237*** (0.0467)
      - –0.217*** (0.0360)
  - Riskiness_Debt Overhang:
    - –0.406* (0.237), –0.328** (0.140), –0.522*** (0.161), –0.229* (0.132)
    - Change in Credit-to-GDP Ratio × Riskiness_Debt Overhang:
      - –0.146*** (0.0297)
      - –0.204*** (0.0277)
  - Riskiness_Expected Default Frequency:
    - –0.879*** (0.243), –0.383** (0.161), –0.942*** (0.233), –0.397** (0.190)
    - Change in Credit-to-GDP Ratio × Riskiness_Expected Default Frequency:
      - –0.0798 (0.0749)
      - –0.171*** (0.0199)
  - Observations and countries (selected):
    - Observations vary by column: 602 (many columns), 592 (some), 532 (other columns)
    - Number of Countries: 41 (many columns), 39 (some columns)
- Interpretation:
  - The change in the credit-to-GDP ratio is consistently negative and statistically significant at the 20th and 50th percentiles: higher recent credit growth is associated with worse future lower-tail and median growth outcomes in these specifications.
  - Higher measured riskiness of credit allocation (across leverage, ICR, debt overhang, expected default frequency) is associated with lower future growth at the 20th and 50th percentiles.
  - Interaction terms: negative and in many cases significant interactions between change in credit-to-GDP ratio and riskiness measures indicate that credit growth combined with higher riskiness of allocation is associated with larger adverse effects on future growth (stronger negative impact on downside growth outcomes).

### Estimation notes, robustness, and complementary findings
- Crisis model specification follows Jordà, Schularick, and Taylor (2016a) with one difference: uses real GDP growth instead of real GDP growth per capita; results robust to using real GDP growth per capita.
- Banking equity stress model follows framework in Baron and Xiong (2017).
- Downside growth analysis also confirmed by logit regressions for low-growth outturns, consistent with quantile regression findings.
- Results robust to alternative estimators for panel logit models (including two-way-clustered standard errors).
- Fiscal policies: the financial conditions index includes the sovereign spread, partially capturing impact of fiscal policies. (IMF (2016) finds fiscal policies affect economic recoveries in a different empirical framework.)

*Source: Annex 2.3. The Riskiness of Credit Allocation and Macro-Financial Outcomes (IMF staff estimates, as presented in the provided content).*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2018/april/chapter-2/doc/c2.pdf_