## _wp15190

## Source details

**Canonical URL:** [_wp15190](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15190.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15190.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15190.pdf.json)

---

### I. Background and Objectives
- Banking crises are often preceded by asset and credit booms that turn into busts.
- The credit-to-GDP gap (difference between the credit-to-GDP ratio and its long-term trend) is a leading indicator of systemic banking crises and is endorsed for setting the countercyclical capital buffer in Basel III (BCBS, 2010).
- Low income countries (LICs) received scant attention in prior work because of the origin of the recent global financial crisis and data limitations in LICs.
- In LICs, rapid credit growth may reflect financial deepening rather than systemic risk buildup, complicating the nexus between credit and financial stability.
- Main goal: develop an analytical toolkit to assess whether credit growth in LICs indicates financial deepening or poses risks to financial stability.
- Approach overview:
  - Assess whether the BCBS credit-to-GDP gap is a good leading indicator of banking crises across income levels and regions.
  - Evaluate an alternative measure of excess credit that considers structural characteristics (financial possibility frontier — FPF).
  - Focus on Sub-Saharan Africa (SSA) because of heterogeneity and rapid change.
  - Combine quantitative gap measures with qualitative analysis due to LIC data limitations.

### II. Measures, Theory, and Identification
- Theoretical causal chains:
  - Credit booms → excessive risk taking (Stiglitz and Weiss, 1981).
  - Financial accelerator: credit supply increases and credit standards loosen with rising collateral values (Bernanke and Gertler, 1989; Kiyotaki and Moore, 1997; Gilchrist and Zakrajsek, 2008).
  - Accommodative monetary policies, especially prolonged, can stimulate excessive risk taking.
- Identification of booms:
  - General approach: capture boom as deviation of a credit measure from its historical trend (a gap).
  - Methodological variants cited: HP-filtered real credit per capita, credit-to-GDP HP gap with high smoothing, backward-looking rolling cubic trend over 10 years.
  - Prominent indicator: BIS gap — percentage deviation of the credit-to-GDP ratio from its long-term trend estimated with a backward-looking HP filter with a high smoothing parameter.
- Structural benchmarking (Financial Possibility Frontier — FPF):
  - FPF: structural upper limit to financial deepening given invariant structural factors (Beck et al., 2008; Al Hussainy et al., 2011).
  - Empirical form: ܻ_{i,t} = X_{i,t} β + η_{i,t} + ε_{i,t} where ܻ is financial development and X structural indicators.
  - Frontier gap = actual level of financial depth minus the frontier; negative gap → below capacity; positive gap → possible overheating/overstepping fundamentals.

### III. Dataset, Crisis Definition, and Methodology
- Sample composition:
  - 81 countries covering all income levels and regions (Appendix Table 2).
  - For 24 countries use BIS total credit to the private sector series (quarterly averaged to annual); for remaining countries use IMF IFS domestic credit to the private non-financial sector (row 32d).
  - Data requirements: series at least 25 years long and without gaps; data must be available for at least ten years prior to a crisis.
  - Note: exclusion of shorter series may induce sample selection bias.
- Systemic banking crisis definition:
  - Primary: Laeven and Valencia (2013) — significant banking distress and significant banking policy intervention.
  - Extended with Reinhart and Rogoff (2010) broader criteria.
  - Laeven and Valencia note necessary conditions: non-performing loans above 20 percent and fiscal recapitalization costs exceeding 5 percent of GDP.
- Predictive evaluation (signal detection; Kaminsky and Reinhart, 1998):
  - Indicator exceeding threshold → signal that a crisis will materialize within forecast period.
  - Forecast window assumed as one-to-three years.
  - Classification of errors: missed crisis (Type I) and false alarm (Type II).
  - Performance summarized with Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC).
  - Note: credit-to-GDP ratios do not always fall during crises; GDP drops can increase ratios and complicate interpretation.

### IV. Predictive Performance of the BIS Gap (method details and results)
- BIS gap estimation specifics:
  - BIS gap = percentage deviation of credit-to-GDP from its long-term trend.
  - Long-term trend estimated using a one-sided (backward-looking) HP filter.
  - The smoothing parameter λ is set equal to 1600.
  - “Real-time” estimates obtained by running the HP filter recursively on each data point (filter uses only information up to that year).
  - Computation starts 10 years after the beginning of each series (HP filter then has a minimum of 10 observations for each trend estimate), implying the first nine years of observations are lost.
  - Signals issued during crisis periods are not considered to reduce noise from GDP contractions that could spur false alarms.
- AUC interpretation reminders:
  - 0.5 indicates the indicator is not informative (ROC = 45-degree diagonal).
  - The higher the distance of the AUC from 0.5 the more informative the indicator.
- Empirical AUCs by income level:
  - High Income 0.78
  - Upper Middle Income 0.61
  - Lower Middle Income 0.55
  - Low Income 0.47
- Empirical AUCs by region:
  - North America 0.83
  - South East Asia and Pacific 0.79
  - Europe and Central Asia 0.71
  - Latin America and Caribbean 0.68
  - Middle East and North Africa 0.51
  - Sub-Saharan Africa 0.39
- Key interpretive findings:
  - BIS gap performs well in high income countries and to a lesser extent upper middle income countries and most regions; exceptions: MENA and SSA.
  - In lower income countries and MENA the credit-to-GDP gap is uninformative (AUC between 0.55 and 0.47), suggesting rapid credit growth often reflects benign financial deepening.
  - SSA result (AUC = 0.39) suggests the indicator sometimes has predictive power only when credit-to-GDP is below trend because crises in SSA frequently originated from sources other than credit booms and often occurred when the BIS gap was negative.

### V. Frontier (Structural) Indicator — Construction and Comparative Performance
- Frontier construction:
  - Frontiers estimated using FinStats (World Bank tool implementing Beck et al. (2008) methodology), producing structural benchmarks for 177 countries.
  - Frontier gap = actual level of financial depth minus the frontier.
- Comparative performance by financial development stage (measured by credit-to-GDP at time of crisis):
  - For countries at an early stage (credit-to-GDP up to 60 percent): frontier gap, on average, outperforms BIS gap.
  - For intermediate and advanced stages (credit-to-GDP between 60 and 200 percent): BIS gap clearly outperforms frontier gap.
  - For oversized financial sectors (credit-to-GDP above 200 percent): frontier gap has highest predictive performance.
  - Observations with AUC < 0.5 for both gaps are concentrated at low levels of financial development (credit-to-GDP up to 60 percent), indicating excess credit was not the driver of financial stress in those countries.

### VI. Policy-Relevant Conclusions and Early Warning Recommendations for LICs
- Core conclusions:
  - Excessive credit should be measured with different indicators at different stages of financial development.
    - BIS gap shows an inverted U-shape performance across levels of financial development.
    - Frontier gap shows a U-shape performance.
    - For early-stage systems, frontier gap is more informative; as financial sectors develop, BIS gap becomes more informative; for oversized sectors both do well but frontier gap may be most powerful.
  - Systemic banking crises in early-stage systems frequently originate from sources other than excess credit (weak supervision, governance issues, composition of credit, external shocks).
- EWS recommendations for LICs:
  - Primary indicator: frontier gap (more informative on average for early-stage systems).
  - Complement BIS gap with frontier gap because:
    1. In some LICs BIS gap outperformed frontier gap (rapid catch-up relative to the structural benchmark can pose systemic risk).
    2. When both gaps are positive, financial stress could be higher.
    3. Frontier gap is static (snapshot); BIS gap is dynamic (shows whether imbalances are rising or correcting).
  - Risk assessment matrix (BIS gap × Frontier gap) — four zones and interpretations:
    - Upper right (both gaps positive): possible overheating; risk of boom-bust episode might be high; risk heightened with bank balance sheet problems and weak supervisory capacity.
    - Lower right (BIS positive, Frontier negative): financial deepening / catch-up process; can be healthy but may be too fast in some countries.
    - Lower left (both negative): inefficient financial sector working below potential; systemic risk from excess credit expected to be low, but crises may stem from other sources.
    - Upper left (BIS negative, Frontier positive): correction of overheating is ongoing.
  - Complement gap-based assessments with in-depth analysis of:
    - Bank balance sheets (concentration, currency/maturity mismatches).
    - Regulatory and supervisory environments.
    - Macro-financial vulnerabilities (asset bubbles, commodity prices).
- Caveats and data limitations:
  - Credit-to-GDP relation is less stable in LICs due to large output volatility; drops in GDP may increase gaps and falsely signal overheating.
  - Data limitations in LICs:
    - Quarterly credit and GDP series available for very few LICs.
    - Limited number of systemic financial crises in LICs → wide confidence bands for ROC curves.
    - Credit and GDP series may have data gaps and structural breaks (civil wars, compilation changes, large exchange rate swings).
  - Rebased GDP examples and effects:
    - Nigeria’s 2013 GDP nearly doubled.
    - Tanzania’s grew by a third.
    - Kenya’s and Zambia’s increased by a quarter.
    - Uganda’s rose by 13 percent.
  - Persistence in trend estimates: a jump in the credit-to-GDP ratio of 10 percentage points may take more than 20 years to dissipate in long-term trend estimates (Drehmann and Tsatsaronis (2014)).
  - FinStats frontier estimates are based on a limited set of structural variables; adding variables could improve country-specific precision but reduces sample size of the frontier regression.

### VII. Systemic Risk Buildup in Sub-Saharan Africa (stylized facts, cross-country assessment, and classifications)
- Historical patterns:
  - 1970s: banking crises rare due to highly-regulated environments and limited banking sector development.
  - 1980s–1990s: numerous banking failures driven by governance problems, weak supervision, large government involvement, poor lending practices, amplified by macro shocks and limited diversification.
  - Since 2000: banking systems became more resilient; since 2000 only Nigeria recorded a systemic event (Reinhart and Rogoff (2010); Laeven and Valencia, 2013).
- Link between credit booms and SSA crises:
  - Laeven and Valencia (2013): 37 systemic banking crises in SSA for 1980-1999, only three preceded by a credit boom.
  - Mendoza and Terrones (2012) criteria identify three additional occurrences where SSA crises were preceded by a credit boom.
  - Many crises had stagnant or decreasing credit ahead of crisis; some cases had credit levels above the frontier even with a negative BIS gap (examples: Cameroon, Senegal).
  - Implication: BIS gap performed poorly as a leading indicator in SSA; frontier gap often provided more reliable signal in low-development contexts.
- Nigeria case (drivers and chronology):
  - Consolidation spurred large credit expansion, fueled by oil-related inflows and loose monetary policy.
  - Regulatory and supervisory frameworks were not upgraded.
  - Both the BIS and the frontier gaps increased considerably, reaching more than 15 percentage points in 2009.
  - Large share of expanding credit financed equities and un-hedged oil imports; global financial crisis and oil price collapse produced widespread nonperforming loans and a banking crisis.
  - Regulatory change noted: The Central Bank of Nigeria announced on July 2, 2004 that banks would be required to achieve a minimum capital level of Naira 25 billion up from Naira 2 billion, by December 31, 2005.
- Cross-country assessment in SSA (end-2013 snapshot and changes 2005–2013):
  - At end-2013 the majority of SSA countries had a credit-to-GDP ratio above historical trends; often reflecting financial deepening and catch-up relative to the frontier.
  - Classification groups (based on BIS gap and frontier gap) and changes 2005–2013 (Table 2, % of total observations):
    - Frontier Gap (+) BIS gap (+): 2005 = 20.0% ; 2013 = 37.5%
    - Frontier Gap (+) BIS gap (-): 2005 = 10.0% ; 2013 = 9.4%
    - Frontier Gap (-) BIS gap (+): 2005 = 66.7% ; 2013 = 40.6%
    - Frontier Gap (-) BIS gap (-): 2005 = 3.3% ; 2013 = 12.5%
    - Frontier Gap (+): 2005 = 30.0% ; 2013 = 46.9%
    - BIS Gap (+): 2005 = 86.7% ; 2013 = 78.1%
  - Interpretations:
    - BIS gap flagged too rapid increase in credit; frontier gap signaled that financial deepening went too far relative to structural characteristics.
    - Rapid financial deepening can outpace real economy absorption, leading to flows to non-priority sectors and asset bubbles.
    - Risk from excess credit is heightened with bank balance sheet problems, weak supervisory capacity, simultaneous asset bubbles, and macroeconomic vulnerabilities.
    - Country-specific factors matter: frontier gap might be overstated for diversified economies (e.g., Kenya).
- Conclusions for SSA:
  - Linkage between financial crises and credit booms in SSA likely to strengthen as banking sectors deepen.
  - Measures comparing credit with historical trend better proxy systemic risk in advanced systems; structural benchmark indicators work better in LICs.
  - Need to monitor rapid changes in banking sector structure and accompany with regulatory/supervisory upgrades and corporate governance improvements.
  - Development of LIC-specific leading indicators focusing on vulnerabilities beyond excess credit is needed.
  - Determination of threshold levels for BIS and frontier gaps in LICs remains an open issue.

### VIII. Appendix — Signaling Detection Approach and Sample Composition (key points)
- Key methodology points:
  - Signaling variable S_t = 1 when indicator x_t exceeds threshold θ in signaling window (t,…,t+2); signaling window set to three years.
  - Outcomes: True positive (S=1 & C=1), False negative / missed crisis (S=0 & C=1) — Type I error, False positive / false alarm (S=1 & C=0) — Type II error, True negative (S=0 & C=0).
  - Trade-off across thresholds: low thresholds reduce missed crises but increase false alarms; high thresholds do the opposite.
  - Discriminatory capacity assessed with ROC curve and AUC: AUC of 0.5 = not informative.
  - The credit-to-GDP gap is tested recursively on a grid of thresholds to assess predictive performance across possible thresholds.
- Sample composition highlights (selected entries from Appendix Table 2):
  - Australia HIEAP BIS 1960 2013
  - China UMI EAP BIS 1977 2013
  - India LMI EAP IFS 1960 2013
  - Nigeria LMI SSA IFS 1960 2012
  - South Africa LMI SSA IFS 1965 2013
  - United States HI MNA BIS 1960 2013
  - Note: LI, LMI, UMI, HI denote income categories; region codes include SSA, MENA, LAC, EAP, SA, NA, and ECA. Income classification refers to 2013.

*Source: _wp15190 - REFERENCES*

### REFERENCES .............................................................................................................

### _wp15190 - REFERENCES

### I. BACKGROUND
- Banking crises are often preceded by asset and credit booms that turn into busts.
- The credit-to-GDP gap (difference between the credit-to-GDP ratio and its long-term trend) is a leading indicator of systemic banking crises and is endorsed for setting the countercyclical capital buffer in Basel III (BCBS, 2010).
- Most prior results are based on advanced economies and large emerging markets; LICs received scant attention due to the origin of the recent global financial crisis and data limitations in LICs.
- In LICs, rapid credit growth may reflect financial deepening rather than systemic risk buildup, complicating the nexus between credit and financial stability.

### OBJECTIVES AND APPROACH
- Main goal: develop an analytical toolkit to assess whether credit growth in LICs indicates financial deepening or poses risks to financial stability.
- Steps:
  - Assess whether the BCBS credit-to-GDP gap is a good leading indicator of banking crises across income levels and regions.
  - Evaluate whether an alternative measure of excess credit that considers structural characteristics of an economy may be a better leading indicator in early-stage financial development.
  - Focus on Sub-Saharan Africa (SSA) because of its heterogeneity and rapidly changing financial landscape.
- Emphasis on combining quantitative gap measures with qualitative analysis due to data limitations in LICs.

### KEY FINDINGS
- Using the gap with respect to the long-term trend in LICs shows no apparent link between credit booms and banking crises in MENA and SSA regions; rapid credit expansions in early-stage financial systems are mostly connected to financial deepening rather than vulnerability buildup.
- Complementing the historical-trend gap with an indicator that links financial development to structural characteristics enhances the capacity to predict banking crises and to distinguish “bad” from “good” booms in LICs.
- Both measures of excess credit have caveats heightened by large data limitations in LICs; therefore, it is necessary to:
  - Combine the two examined approaches.
  - Perform in-depth qualitative analysis to judge whether observed expansions are healthy deepening or risky.

### APPLICATION TO SSA
- The framework provides a preliminary assessment of systemic risk buildup in individual SSA countries.
- Two contrasting country groups in SSA:
  - A number of SSA countries—mainly the most advanced and/or politically stable—where the linkage between financial crises and credit boom-bust is likely to become stronger; rapid credit increases are associated with financial depth higher than what is compatible with fundamentals.
  - Other SSA countries with less developed financial systems where brisk credit growth appears compatible with fundamentals and healthy financial deepening.
- Caveat: financial vulnerabilities and systemic risk buildup in SSA may still originate from sources other than excessive credit growth.

### RELATION TO LITERATURE
- Connects to literature on boom-bust episodes and financial crises (Reinhart and Rogoff, 2009; Dell’Ariccia et al., 2013) and to empirical research on early warning systems and indicators of financial crises (Kaminsky and Reinhart, 1999; Borio and Drehmann, 2009).
- Differs by focusing on the relationship between financial stability and financial development and by proposing different leading indicators depending on the stage of financial development.
- Also linked to benchmarking financial systems literature (Beck et al., 2008; Barajas et al., 2012) through use of the financial possibility frontier concept to distinguish financial deepening from systemic risk buildup in LICs.

### ORGANIZATION OF THE PAPER
- Section II: reviews literature on excess credit, financial sector stability, and financial deepening.
- Section III: tests whether excess credit is a good leading indicator of financial crises across income levels and regions.
- Section IV: proposes a simple toolkit to assess systemic risk buildup in LICs.
- Section V: presents stylized facts on SSA financial crises and provides a preliminary assessment of systemic risk buildup in individual SSA countries using the introduced framework.

*Source: _wp15190 - REFERENCES*

### Section IV. Finally, Section VI discusses the findings and suggests future research.

### _wp15190 - Section IV. Finally, Section VI discusses the findings and suggests future research.

### II. BOOMS, CRISES AND GAPS — A. On booms and crises
- The early warning literature links excess credit to financial cycles and asserts “peaks in the financial cycle (i.e. booms) are closely associated with systemic banking crises” (Borio, 2012).
- Theoretical chains of causation:
  - Credit booms → excessive risk taking during financial cycle upswing (Stiglitz and Weiss, 1981).
  - Financial accelerator: supply of credit increases and credit standards loosen pari passu with improvement in collateral values (Bernanke and Gertler, 1989; Kiyotaki and Moore, 1997; Gilchrist and Zakrajsek, 2008).
  - Accommodative monetary policies, especially prolonged, can stimulate excessive risk taking.
- Empirical evidence:
  - Positive link between rapid credit expansion and financial stress (Gourinchas et al., 2001; Mendoza and Terrones, 2004; Dell’Ariccia et al., 2012; Crowe et al., 2013).
  - Booms are more likely to end in crises in advanced economies or emerging markets than in LICs:
    - Barajas et al. (2007), sample of 100 countries for the period 1980-2004: only 11 percent of booms in LICs were followed by a crisis vs. 36 percent in emerging markets.
    - Dell’Ariccia et al. (2012): booms that start at a lower level of financial depth are less likely to end badly.
    - Arena et al. (2015), sample of 135 developing countries: boom episodes associated with a financial crisis are more frequent in upper-middle income countries (50 percent) than in LICs (10 percent).
- Interpretation:
  - Relationship between credit and financial stability may be non-linear.
  - In countries with developed financial systems, rapid credit expansion frequently reflects systemic risk buildup.
  - In LICs, fast credit growth is mainly connected to healthy financial deepening and should be associated with a permanent take-off in financial depth rather than cyclical upswings.

### II. BOOMS, CRISES AND GAPS — B. On the Identification of Credit Booms
- General empirical approach: capture boom phase as deviation of a credit measure from its historical trend (a gap).
- Methodological variants:
  - Mendoza and Terrones (2008): split real credit per capita into cyclical and trend components; identify booms when credit exceeds long-run trend computed using Hodrick-Prescott (HP) filter.
  - Borio and Drehmann (2009): credit gap based on credit-to-GDP ratio; trend estimated with HP filter using a high smoothing parameter.
  - Dell’Ariccia et al. (2012): credit gap as percentage deviation of credit-to-GDP ratio from a backward-looking, rolling, cubic trend estimated over a 10-year period.
- Prominent indicator in early warning literature: gap between credit-to-GDP ratio and its long-term trend computed with a backward looking HP filter with a high smoothing parameter (referred to as the BIS gap).
  - Basel Committee on Banking Supervision and Basel III endorse this gap as guidance for countercyclical capital buffer (BCBS, 2010).
- Limitations in LICs:
  - Positive credit-to-GDP gap may reflect improved fundamentals and financial deepening rather than systemic risk (Geršl and Seidler, 2012; IMF, 2014b).
  - If financial deepening is steady it is embedded in the long-term trend; if sudden and rapid, it may not be captured by the trend and could be signaled as vulnerability.
- Empirical performance:
  - Most predictive performance tests of the credit-to-GDP gap focus on advanced economies (Appendix Table 1).
  - IMF (2011b), sample of 169 countries (advanced, emerging, LICs): credit-to-GDP gap did not perform well as a signaling variable for emerging markets and low income countries.
  - Drehmann and Tsatsaronis (2014), sample of advanced and emerging markets: credit-to-GDP gap remains a good indicator in emerging markets, though performance is weaker than in advanced economies.
  - Note: emerging markets in that sample mainly have developed financial sectors; only SSA country included is South Africa.

### II. BOOMS, CRISES AND GAPS — C. On Cycles versus Structure
- Key issue: distinguishing when credit growth is unsustainable and signals imbalance versus when it signals beneficial financial deepening.
- Cyclical indicator limitation: gap relative to long-term trend may not distinguish good booms from bad in LICs.
- Structural benchmarking approach: financial possibility frontier (FPF).
  - Concept: structural factors invariant in the short term impose an upper limit to financial deepening — the financial possibility frontier (Beck et al., 2008; Al Hussainy et al., 2011).
  - Empirical implementation: panel regression estimating relationship between financial development measure ܻ and structural indicators ܺ:
    - ܻ_{i,t} = X_{i,t} β + η_{i,t} + ε_{i,t}
  - Interpretation of gap from frontier:
    - Negative gap → inefficient financial sector operating below capacity.
    - Positive gap (overshooting predicted level) → overheating pressures and financial stress (Barajas et al., 2013).
  - The frontier-gap is structural; trend-gap is cyclical.
- Research question posed: can the frontier-based indicator better measure excess credit in LICs by flagging when financial deepening has gone too far?
- Next section will address this comparison.

### III. SYSTEMIC RISK ASSESSMENT ACROSS LEVELS OF FINANCIAL DEVELOPMENT — A. Scope of the Analysis and Policy Relevance
- Objectives:
  - Should excess credit be measured with different indicators depending on the level of financial development?
  - What is the role of credit growth in systemic banking crises across levels of financial development?
- Policy relevance:
  - Activation and calibration of macroprudential tools hinge on indicator performance.
  - Basel Committee suggests using BIS gap for countercyclical capital buffer; mechanical application may be inappropriate if BIS gap is a poor leading indicator in low financial development countries (could hinder financial development by mis-signaling).
  - If systemic risk sources differ by financial development level, alternative early warning indicators must be sought — especially relevant for LICs where vulnerabilities might not manifest as credit booms.
- Analysis plan:
  - Step 1: analyze BIS gap performance across income levels and regions.
  - Step 2: complement with a structural benchmark indicator (FPF gap) and assess comparative predictive performance across levels of financial development.
  - Step 3: propose a simple toolkit for first assessment of systemic risk buildup across levels of financial development, focusing on LICs.

### III. SYSTEMIC RISK ASSESSMENT ACROSS LEVELS OF FINANCIAL DEVELOPMENT — B. Dataset and Definition of Systemic Banking Crisis
- Sample:
  - 81 countries covering all income levels and regions (Appendix Table 2).
  - For 24 countries (mainly advanced economies and large emerging markets) use new BIS series of total credit to the private sector, adjusted for structural breaks; quarterly data averaged to annual.
    - The 24 countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, India, Indonesia, Ireland, Italy, Japan, Korea, Netherlands, Norway, Portugal, Singapore, South Africa, Sweden, Thailand, U.K., and the U.S.
    - BIS data cover total credit including cross-border finance to the private non-financial sector.
  - For the remaining countries, use annual series of domestic credit to the private non-financial sector from IMF’s International Financial Statistics (row 32d in the old IFS classification — domestic credit to private sector by banks).
    - Caveat: this captures only bank credit; nonbank lending-driven booms may be missed. In LICs bank credit may be a reliable proxy for total credit.
- Data requirements:
  - Series for each country must be at least 25 years long and without data gaps.
  - Data must be available for at least ten years prior to a crisis.
  - Note: exclusion of shorter series or series with gaps may induce sample selection bias, under-representing fragile countries.
- Definition and timing of banking crises:
  - Primary source: Laeven and Valencia (2013) systemic banking crises database.
    - Banking crisis defined by two conditions: i) significant signs of financial distress in the banking system (bank runs, losses, bank liquidations) and ii) significant banking policy intervention measures in response.
    - Laeven and Valencia note necessary conditions for systemic crisis include non-performing loans above 20 percent and fiscal recapitalization costs exceeding 5 percent of GDP.
  - Extended with Reinhart and Rogoff (2010) broader database: crisis if i) bank runs lead to closure/merger/takeover or takeover by public sector of one or more institutions; or ii) if no runs, closure/merger/takeover or large-scale government support of the banking sector.
    - Reinhart’s database covers 70 countries in the period 1800-2010.

### III. SYSTEMIC RISK ASSESSMENT ACROSS LEVELS OF FINANCIAL DEVELOPMENT — C. Methodology
- Predictive evaluation approach: signal detection method (Kaminsky and Reinhart, 1998).
  - When indicator exceeds threshold → signal that a crisis will materialize within forecast period.
  - Forecast window assumed as one-to-three years.
- Classification of forecast errors:
  - Missed crisis (Type I error): indicator did not signal a crisis that occurred in the forecast horizon.
  - False alarm (Type II error): indicator signaled a crisis that did not materialize in the forecast horizon.
- Performance evaluation:
  - Test indicator over a fine grid of thresholds.
  - Use Receiver Operating Characteristic (ROC) curve to illustrate tradeoff between sensitivity (true positive rate) and accuracy (false positive rate).
  - Area Under the Curve (AUC) synthesizes signaling power of the indicator.
- Empirical note:
  - Credit-to-GDP ratios do not always fall during crises; they sometimes increase due to GDP decline being larger than lending slowdown or drawing on preexisting credit lines. This “statistical” increase can complicate interpretation of indicators based on credit-to-GDP.

*Source: _wp15190 - Section IV. Finally, Section VI discusses the findings and suggests future research.*

### 0.5 indicates that the indicator is not informative, as for any positive signal the probability

### 0.5 indicates that the indicator is not informative, as for any positive signal the probability that the event of interest will materialize in the forecast horizon is equal to the probability of a false alarm. This case corresponds to an ROC curve coincident with the 45-degree diagonal.

### Predictive performance of the BIS gap
- Method and estimation details
  - BIS gap: percentage deviation of the credit-to-GDP ratio from its long-term trend.
  - Long-term trend estimated using a one-sided (backward-looking) HP filter.
  - The smoothing parameter λ is set equal to 1600.
  - “Real-time” estimates of the gap are obtained by running the HP filter recursively on each data point (the filter uses only information up to that year).
  - To obtain more robust trend estimates, computation starts 10 years after the beginning of each series (the HP filter then has a minimum of 10 observations for each trend estimate), implying the first nine years of observations are lost.
  - Signals issued during crisis periods are not considered to reduce noise from GDP contractions that could spur false alarms.
- AUC and ROC interpretation
  - 0.5 indicates the indicator is not informative (ROC = 45-degree diagonal).
  - The higher the distance of the AUC from 0.5 the more informative the indicator.
  - Indicators expected to increase ahead of crisis (e.g., BIS gap) have maximum predictive power when AUC = 1.
  - Indicators expected to decrease ahead of crisis (e.g., risk aversion) have maximum predictive power when AUM = 0.
- Empirical findings (Figure 1, Table 1)
  - Income-level AUCs:
    - High Income 0.78
    - Upper Middle Income 0.61
    - Lower Middle Income 0.55
    - Low Income 0.47
  - Regional AUCs:
    - North America 0.83
    - South East Asia and Pacific 0.79
    - Europe and Central Asia 0.71
    - Latin America and Caribbean 0.68
    - Middle East and North Africa 0.51
    - Sub-Saharan Africa 0.39
  - Key interpretive findings:
    - BIS gap performs well as a leading indicator in high income countries and, to a lesser extent, upper middle income countries and most regions (exceptions: MENA and SSA).
    - In lower income countries and the MENA region the credit-to-GDP gap is uninformative (AUC between 0.55 and 0.47), possibly because rapid credit growth often reflects benign financial deepening rather than crises.
    - SSA result (AUC = 0.39) suggests the indicator sometimes has predictive power only when credit-to-GDP is below trend; this reflects that crises in SSA frequently originated from sources other than credit booms and often occurred when the BIS gap was negative.

### Complementing the BIS gap: the frontier (structural) indicator
- Rationale and construction
  - The frontier (financial possibility frontier) is the upper limit to sound and efficient financial deepening; beyond it risks to financial stability may arise.
  - Frontiers estimated using FinStats (World Bank tool implementing Beck et al. (2008) methodology), producing structural benchmarks for 177 countries.
  - Frontier gap = actual level of financial depth minus the frontier.
- Comparative performance vs BIS gap (Figures 2 and 3)
  - Performance by level of financial development (measured by credit-to-GDP at time of crisis):
    - For countries at an early stage of financial development (credit-to-GDP up to 60 percent): frontier gap, on average, outperforms BIS gap.
    - For intermediate and advanced stages (credit-to-GDP between 60 and 200 percent): BIS gap clearly outperforms frontier gap.
    - For oversized financial sectors (credit-to-GDP above 200 percent): frontier gap has highest predictive performance.
  - Observations with AUC < 0.5 for both gaps are concentrated at low levels of financial development (credit-to-GDP up to 60 percent), indicating excess credit was not the driver of financial stress in those countries.

### Putting the results together — policy-relevant conclusions and early warning for LICs
- Main conclusions
  - Excessive credit should be measured with different indicators at different stages of financial development.
    - BIS gap shows an inverted U-shape performance across levels of financial development.
    - Frontier gap shows a U-shape performance.
    - For early-stage financial systems, frontier gap is more informative; as financial sectors develop, BIS gap becomes more informative; for oversized sectors both do well but frontier gap may be most powerful.
  - Systemic banking crises in early-stage financial systems frequently originate from sources other than excess credit (weak supervision, governance issues, composition of credit, external shocks).
- Early Warning System (EWS) recommendations for LICs
  - Primary indicator: frontier gap (more informative on average for early-stage systems).
  - Complement BIS gap with frontier gap for three reasons:
    1. In some LICs BIS gap outperformed frontier gap (rapid catch-up relative to the structural benchmark can pose systemic risk).
    2. When both gaps are positive, financial stress could be higher (Barajas et al. (2013) — when financial depth exceeds the frontier as result of rapid credit expansion, instability likelihood increases further).
    3. Frontier gap is static (snapshot); BIS gap is dynamic (shows whether imbalances are rising or correcting).
  - Risk assessment matrix (BIS gap × Frontier gap) — four zones and interpretations:
    - Upper right (both gaps positive): possible overheating; risk of a boom-bust episode might be high; risk heightened with bank balance sheet problems and weak supervisory capacity.
    - Lower right (BIS positive, Frontier negative): financial deepening / catch-up process; can be healthy but may be too fast in some countries.
    - Lower left (both negative): inefficient financial sector working below potential; systemic risk from excess credit expected to be low, but crises may stem from other sources.
    - Upper left (BIS negative, Frontier positive): correction of overheating is ongoing.
  - Complement gap-based assessments with in-depth analysis of:
    - Bank balance sheets (concentration, currency/maturity mismatches).
    - Regulatory and supervisory environments.
    - Macro-financial vulnerabilities (asset bubbles, commodity prices).
- Caveats and data limitations
  - Credit-to-GDP relation is less stable in LICs due to large output volatility; drops in GDP may increase gaps and falsely signal overheating.
  - Data limitations in LICs:
    - Quarterly credit and GDP series available for very few LICs.
    - Limited number of systemic financial crises in LICs → wide confidence bands for ROC curves.
    - Credit and GDP series may have data gaps and structural breaks (civil wars, changes in compilation, large exchange rate swings).
  - Rebased GDP examples (impact on measurement):
    - Nigeria’s 2013 GDP nearly doubled.
    - Tanzania’s grew by a third.
    - Kenya’s and Zambia’s increased by a quarter.
    - Uganda’s rose by 13 percent.
  - Measurement persistence: a jump in the credit-to-GDP ratio of 10 percentage points may take more than 20 years to dissipate in long-term trend estimates (Drehmann and Tsatsaronis (2014)).
  - FinStats frontier estimates are based on a limited set of structural variables; including additional structural variables could improve country-specific precision but reduces sample size of the frontier regression.

### Systemic risk buildup in SSA — stylized facts and implications
- Historical patterns and drivers
  - 1970s: banking crises rare due to highly-regulated environment and limited banking sector development.
  - 1980s–1990s: numerous banking failures driven by governance problems at bank level, weak supervision, large government involvement, poor lending practices, often amplified by macro shocks and limited economic diversification.
  - Since 2000: banking systems became more resilient; since 2000 only Nigeria recorded a systemic event (Reinhart and Rogoff (2010); Laeven and Valencia, 2013).
- Link between credit booms and SSA crises
  - Historically, credit booms rarely preceded SSA banking crises:
    - Laeven and Valencia (2013): 37 systemic banking crises in SSA for 1980-1999, only three preceded by a credit boom.
    - Mendoza and Terrones (2012) criteria identify three additional occurrences where SSA crises were preceded by a credit boom.
  - Many crises had stagnant or decreasing credit ahead of crisis; in some cases credit levels were above frontier even if BIS gap was negative (examples: Cameroon, Senegal).
  - Implication: BIS gap performed poorly as a leading indicator in SSA because many crises occurred when BIS gap was negative; frontier gap often provided more reliable signal in low-development contexts.
- Recent trends and risks
  - Linkage between financial crises and credit booms in SSA likely to strengthen as banking sectors deepen and sophistication rises.
  - Example: Nigeria experienced a 1991–1995 crisis typical of earlier African crises (weak governance, poor practices, excessive credit not a factor) and a 2009 crisis that combined weak supervision with a boom-bust episode following consolidation in 2005–06.

*Source: IMF IFS, FinStats, and Authors’ Calculations.*

### introduction of stricter capital requirements.

### introduction of stricter capital requirements.

### Background: Nigeria case and drivers of crisis
- Consolidation in the banking sector spurred a large credit expansion, further fueled by large oil-related inflows and a loose monetary policy stance.
- Regulatory and supervisory frameworks were not upgraded to sustain and monitor bank growth and ensure proper enforcement of prudential standards.
- Both the BIS and the frontier gaps increased considerably, reaching more than 15 percentage points in 2009 (Figure 8).
- A large share of the expanding credit was used to purchase equities (often with commercial banks extending the credit) and another significant share financed un-hedged oil imports.
- When the global financial crisis burst the equity bubble and oil prices collapsed, many stock-backed and oil-related loans became nonperforming and the credit boom ended in a full-fledged banking crisis (Sanusi, 2010).
- Regulatory change cited: The Central Bank of Nigeria announced on July 2, 2004 that banks would be required to achieve a minimum capital level of Naira 25 billion up from Naira 2 billion, by December 31, 2005.

### Cross-country assessment in SSA (scope and patterns)
- Focus: countries at an early stage of financial development (credit-to-GDP ratio up to 60 percent) using the EWS for LICs introduced in Section IV.
- At end-2013:
  - The majority of SSA countries had a credit-to-GDP ratio above historical trends.
  - In many cases, this reflects financial deepening as countries start from a very low level of financial depth and are catching up with respect to the financial possibility frontier.
  - In some countries, financial depth is well above its benchmark level and simultaneous acceleration of credit may signal systemic risk buildup.

### Classification and changes 2005–2013 (gaps and risk groups)
- Based on the BIS gap and the frontier gap, SSA countries are classified into four groups: “possible overheating”, “correction of overheating is ongoing”, “financial deepening”, “inefficient financial sector”.
- Comparative changes between 2005 and 2013:
  - The share of countries with a positive frontier gap grew from 30.0 percent to 46.9 percent (Table 2).
  - The share of countries with both gaps positive rose from 20.0 percent in 2005 to 37.5 percent in 2013.
  - Table 2 (Gaps Sign: Comparison 2005-2013 (% of total observations)):
    - Frontier Gap (+) BIS gap (+): 2005 = 20.0% ; 2013 = 37.5%
    - Frontier Gap (+) BIS gap (-): 2005 = 10.0% ; 2013 = 9.4%
    - Frontier Gap (-) BIS gap (+): 2005 = 66.7% ; 2013 = 40.6%
    - Frontier Gap (-) BIS gap (-): 2005 = 3.3% ; 2013 = 12.5%
    - Frontier Gap (+): 2005 = 30.0% ; 2013 = 46.9%
    - BIS Gap (+): 2005 = 86.7% ; 2013 = 78.1%

### Key interpretations and risks
- The BIS gap flagged a too rapid increase in credit; the frontier gap signaled that financial deepening went too far given the country’s structural characteristics.
- Rapid financial deepening can outpace the real economy's absorption capacity, resulting in flows to non-priority sectors and asset market bubbles.
- Risk from excess credit is heightened when combined with:
  - bank balance sheet problems,
  - weak supervisory capacity,
  - simultaneous asset bubbles (stock market, real estate, commodity prices),
  - macroeconomic vulnerabilities.
- Country-specific factors matter when interpreting the two gaps; e.g., the frontier gap might be overstated for countries with a diversified economy (such as Kenya).

### Conclusions on indicators and financial development stages
- Measures that compare credit level with its historical trend are better proxies for systemic risk buildup in countries at an advanced stage of financial development.
- Indicators that contrast actual credit to a benchmark based on structural characteristics work better for low income countries (LICs).
- Empirical implications:
  - In countries at an early stage of financial development, credit expansion is mainly positively associated with economic growth (financial deepening); systemic risk may arise when credit expansion exceeds the economy’s absorption capacity.
  - In countries with a developed financial system, rapid credit growth is frequently associated with excessive risk-taking, weaker lending standards, and a higher likelihood of systemic banking crisis; acceleration above secular trend likely reflects systemic risk buildup rather than additional deepening.
  - For countries with an oversized financial sector, the size itself is a source of systemic risk, independent of credit acceleration, reflecting riskier business models, larger interconnectedness, and higher vulnerability to global shocks.

### Policy implications and monitoring priorities
- Rapid changes in banking sector structure should be carefully monitored and accompanied by necessary regulatory and supervisory upgrades, and proper corporate governance.
- Assessments based on the BIS and frontier gaps should be complemented with comprehensive analysis of:
  - bank balance sheets,
  - regulatory and supervisory frameworks,
  - macro-financial vulnerabilities.
- There is a need to develop LIC-specific leading indicators and early warning models that focus on vulnerabilities beyond excess credit, because banking crises in LICs may occur even without excessive credit by any measure.
- Determination of threshold levels for BIS and frontier gaps in LICs remains an open issue:
  - Financial deepening could justify higher BIS gap thresholds than in advanced economies.
  - Fragile financial systems and supervisory frameworks in LICs could imply that even small deviations of credit-to-GDP from trend or frontier can spur a banking crisis.

### Appendix summary: Signaling detection approach (key points)
- Methodology: signal detection approach (Kaminsky and Reinhart, 1998).
- Signal definition:
  - Signaling variable S_t = 1 when the indicator x_t exceeds threshold θ in the signaling window (t,…,t+2); signaling window set to three years.
- Outcomes for each threshold and data point:
  - True positive (S=1 & C=1), False negative / missed crisis (S=0 & C=1) — Type I error, False positive / false alarm (S=1 & C=0) — Type II error, True negative (S=0 & C=0).
- Trade-off between Type I and Type II errors across thresholds: low thresholds reduce missed crises but increase false alarms; high thresholds do the opposite.
- Discriminatory capacity assessed with ROC curve and area under the curve (AUM/AUC):
  - AUM of 0.5 = not informative; higher distance from 0.5 indicates more informative indicator.
  - Indicators expected to increase ahead of crisis (e.g., credit-to-GDP gap) have maximum predictive power when AUM = 1.
- The credit-to-GDP gap is tested recursively on a grid of thresholds to assess predictive performance across possible thresholds.

*Source: _wp15190 - introduction of stricter capital requirements.*

### Appendix Table 2. Sample Composition

### Appendix Table 2. Sample Composition

### Sample composition — part 1
- AlgeriaUMIMENA IFS 19641991
- ArgentinaUMILAC IFS 19602013
- AustraliaHIEAP BIS 19602013
- AustriaHIECA BIS 19602013
- BangladeshLIEAP IFS 19742013
- BelgiumHIECA BIS 19712013
- BeninLISSA IFS 19602013
- BoliviaLMILAC IFS 19602013
- Burkina Faso LISSA IFS 19622013
- BurundiLISSA IFS 19642013
- CameroonHIMNA IFS 19602013
- CanadaLMISSA BIS 19602013
- Centr. Afric. Rep. LISSA IFS 19602013
- ChadLISSA IFS 19602013
- ChileHILAC IFS 19602013
- ChinaUMIEAP BIS 19772013
- ColombiaUMILAC IFS 19602013
- Congo, Dem. Rep. LISSA IFS 19632013
- Congo, Rep. LMISSA IFS 19602013
- Costa Rica UMILAC IFS 19602013
- Cote d'Ivoire LMISSA IFS 19622013
- CyprusHIECA BIS 19752013
- DenmarkHIECA BIS 19602013
- Dominican Republic UMILAC IFS 19602013
- EcuadorUMILAC IFS 19602013
- Egypt, Arab Rep. LMIMENA IFS 19652013
- El Salvador LMILAC IFS 19652013
- FinlandHIECA BIS 19702013
- FranceHIECA BIS 19692013
- GermanyHIECA BIS 19702013
- GhanaLMISSA IFS 19602013
- GreeceHIECA BIS 19602013
- GuatemalaLMILAC IFS 19602013
- GuyanaLMILAC IFS 19602013
- HondurasLMILAC IFS 19602013
- IcelandHIECA BIS 19602013
- IndiaLMIEAP IFS 19602013
- IndonesiaLMIEAP IFS 19762013
- IrelandHIECA BIS 19712013
- IsraelHIMENA IFS 19602011
- ItalyHIECA BIS 19602013
- JamaicaUMILAC IFS 19602013

### Sample composition — part 2 (continued)
- JapanHIEAP BIS 19642013
- JordanUMIMENA IFS 19652013
- KenyaLISSA IFS 19612013
- Korea, Rep. HIEAP BIS 19622013
- KuwaitHIMENA IFS 19622012
- MadagascarLISSA IFS 19622013
- MalaysiaUMIEAP IFS 19642013
- MaliLISSA IFS 19672013
- MauritaniaLIMENA IFS 19622012
- MexicoUMILAC IFS 19802013
- MoroccoLMIMENA IFS 19602013
- MyanmarLIEAP IFS 19602004
- NetherlandsHIECA BIS 19612013
- New Zealand HIEAP BIS 19602010
- NigerLISSA IFS 19622013
- NigeriaLMISSA IFS 19602012
- NorwayHIECA BIS 19602013
- ParaguayLMILAC IFS 19602013
- PeruUMILAC IFS 19602013
- PhilippinesLMIEAP IFS 19602013
- PortugalHIECA BIS 19602013
- SenegalLMISSA IFS 19602013
- Sierra Leone LISSA IFS 19602013
- SingaporeHIEAP BIS 19632013
- South Africa LMISSA IFS 19652013
- SpainHIECA BIS 19602013
- Sri Lanka LMIEAP IFS 19602012
- SwazilandLMISSA IFS 19732013
- SwedenHIECA BIS 19612013
- SwitzerlandHIECA IFS 19602013
- ThailandUMIEAP BIS 19602013
- TogoLISSA IFS 19622013
- TunisiaUMIMENA IFS 19652013
- TurkeyUMIECA IFS 19862013
- United Kingdom HIECA BIS 19632013
- United States HIMNA BIS 19602013
- UruguayHILAC IFS 19602013
- Venezuela, RBUMILAC IFS 19622013
- ZimbabweLISSA IFS 19792006

### Notes on classifications and data sources
- LI, LMI, UMI, HI stand for low income, lower middle income, upper middle income, and high income, respectively.
- SSA, MENA, LAC, EAP, SA, NA, and ECA stand for Sub-Saharan Africa, Middle-East and North Africa, Latina America and Caribbean, East Asia and Pacific, South Asia, North America, and Europe and Central Asia, respectively.
- Income level classification is from the World Bank and refers to 2013.

*Source: Appendix Table 2. Sample Composition (source PDF content).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15190.pdf_