## _wp09141

## Source details

**Canonical URL:** [_wp09141](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09141.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09141.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09141.pdf.json)

---

### Major findings and framing (Introduction and Summary)
- Collapse of the subprime mortgage market in the U.S. in 2007 and ensuing financial instability renewed interest in banking crises; literature documents both similarities (e.g. Reinhart and Rogoff, 2008a) and differences (e.g. Bordo, 2008; Gorton, 2008).
- Empirical literature on bank fragility has relied on crisis-dating schemes that primarily reflect government actions (regulators/central banks) rather than the contemporaneous realization of adverse shocks to the banking industry.
- Key implication: BC indicators are likely to date banking crisis onsets too late on average because government responses can lag due to uncertainty and political economy considerations.
- Main recommendation for empirical work: distinguish realizations of systemic bank shocks (SBS) from government interventions (BC) when studying causes and consequences of banking crises.

### Conceptual critique and identification
- Problem: equating dating of government response (BC) with dating of systemic bank shock (SBS) confounds the realization of adverse bank shocks with restorative policy responses.
- Using BC-type indicators as proxies for systemic risk realizations makes measurement “very sensitive to the classification criterion used” (De Nicolò et al., 2004).
- Paper’s main contribution: separate adverse shocks and policy responses to allow identification of theory-based measures of SBS.

### Theoretical model (Section IV)
- Setup:
  - Banks are Cournot-Nash competitors raising insured deposits, making risky loans, and holding risk free government bonds; deposit insurer bails out banks when they fail.
  - Agents: government, entrepreneurs, depositors, banks. All agents risk-neutral; time discrete.
  - Entrepreneurs: continuum indexed by reservation income on [0,1]; projects require investment standardized to 1 and yield Y with probability P1t+.
  - Deposits fully insured; inverse supply curve R = Z α_t.
  - Government issues fixed amount of bonds B; guarantees deposits and issues additional bonds to pay depositors when banks fail; intervenes when realized bank profits at 1t+ are negative.
  - Sequence: at t banks collect deposits and decide on lending based on expected p ≡ E_t+1[P_{t+1}]; at 1t+ realizations occur; negative profits imply SBS and government intervention at 2t+.
- Comparative statics (linear specification) — observable responses to three adverse shocks (p decreases, α increases, Y decreases):
  - Aggregate loans: decreases when p decreases; decreases when α increases; decreases when Y decreases.
  - Aggregate deposits: decreases when p decreases; decreases when α increases; decreases when Y decreases.
  - Bond interest rate: ↓ when p decreases; ↑ when α increases; ↓ when Y decreases.
  - Loan rate: ↑ when p decreases; ↑ when α increases; ↑ when Y decreases.
  - Deposit rate: ↓ when p decreases; ↑ when α increases; ↑ when Y decreases.
  - Loan rate − Deposit rate (interest margin): ↑ in response to all three adverse shocks.
  - Realized profits: ↓ in response to all three adverse shocks.
- Identification implication: an SBS can be identified empirically by severe declines in loans, deposits, bank profits, and significant increases in interest rate margins.

### Empirical strategy and datasets (Sections V, VIII)
- Country-level panel: up to 91 countries for the 1980-2002 period; merges and updates datasets used in DD (2005) and Beck et al. (2006).
- Bank-level panel: accounting data for about 120 emerging and developing countries for 1993-2004.
- Empirical steps:
  1. Benchmark Logit regressions with BC indicators (four crisis classifications) as dependent variables using contemporaneous macro variables.
  2. Construct theory-based SBS indicators:
     - Country: SBSL25, SBSL10 (real domestic lending growth below 25% and 10% percentiles across full cross-country distribution); SBSD25, SBSD10 (deposit-to-GDP growth below 25% and 10% percentiles).
     - Bank: FAIL5, FAIL10 (5th and 10th percentiles of (profits + equity capital)/assets distribution).
     - Include lagged SBS in regressions predicting BC to test lead-lag.
  3. Estimate Logit regressions with SBS indicators as dependent variables to compare determinants against BC regressions.

### Crisis classifications and dating discrepancies (Sections II–III)
- Four BC binary indicators: DD (Demirgüç-Kunt and Detragiache), CEA (Caprio et al.), RR (Reinhart and Rogoff), LV (Laeven and Valencia).
- Two versions: "Start-date only" and "All crisis years" (for RR and LV durations borrowed from CEA or DD where needed).
- Dating discrepancies:
  - Pairwise disagreements in starting dates and durations frequent; ratio of total crisis ranking discrepancies divided by total crisis rankings varies between 24.5% and 49.5%.
  - All four classifications agree on 41 dates of crisis onset.
- Conclusion: BC classifications predominantly index government responses and produce pervasive dating discrepancies.

### Country-level SBS findings (Section V and related tables)
- Predictive relationships:
  - Lagged SBS lending indicators (L.SBSL25, L.SBSL10) predict BC indicators in all specifications.
    - Examples: L.SBSL25 = 0.412*** [0.00388] (DD), 0.576*** [0.000126] (CEA), 0.519*** [0.000126] (RR), 0.428*** [0.00733] (LV).
    - L.SBSL10 = 0.365** [0.0469] (DD), 0.785*** [0.0000272] (CEA), 0.771*** [0.0000261] (RR), 0.632*** [0.000901] (LV).
  - Lagged SBS deposit indicators (L.SBSD25, L.SBSD10) are positively associated with BC but show weaker/mixed significance (e.g., L.SBSD10 = 0.340* [0.0922] for CEA; 0.338* [0.0949] for LV).
- Determinants of SBS (selected robust results):
  - rgdpgr (GDP growth rate): negative and statistically significant predictors of SBS lending indicators (examples: -0.119*** [0.000000706] for SBSL25; -0.0948*** [0.00119] for SBSL10).
  - depr (exchange rate depreciation): positive and significant (examples: depr = 1.238*** [0.00274] for SBSL25; depr = 1.615*** [0.000224] for SBSL10).
  - totch (terms of trade changes): often a strong determinant of SBS (negative sign on totch significant in multiple regressions).
  - Bank concentration (concen_mean, avgherf): positively associated with SBS (examples: concen_mean = 1.656*** [0.00437], avgherf = 1.460*** [0.0000475] for SBSL25).
  - Financial openness: some independent impact on SBS (significant in 3 out of 8 regressions).
- External shocks and currency/twin crises:
  - Currency crises and twin crises positively and significantly predict probability of SBS in most regressions.
  - SBS indicators in turn predict currency crises strongly; BC indicators do not predict currency crises when used as explanatory variables.

### Market structure, deposit insurance, and institutions (Section VI and Tables 9–11)
- Bank concentration:
  - In country-level SBS regressions, concentration significantly increases probability of an SBS.
  - Concentration does not robustly affect probability of government response (BC) across specifications.
- Deposit insurance (di):
  - Positively and significantly associated with BC indicators (government response more likely when formal deposit insurance exists). Examples: di = 1.325*** [0.00185] for CEA; di = 1.105*** [0.00423] for LV (in Table 9).
  - Does not significantly affect probability of SBS in most specifications.
- Institutional quality (kk_compo):
  - Often negative and significant for probability of government response (example: kk_compo = -1.295*** [0.000482] for CEA).
  - Less consistent effect on SBS.

### Bank-level evidence (Section VIII and Tables 14–15)
- Bank-level SBS indicators (FAIL5, FAIL10) strongly predict BC indicators in all specifications:
  - Examples: L.FAIL5 = 0.819***; L.FAIL10 = 0.772*** in Table 14.
- Key bank-level determinants:
  - L.growth: negative and significant for BC indicators (e.g., -0.164***; -0.200*** to -0.153*** across specifications).
  - L.infl: positive and significant for BC indicators (e.g., 0.045***).
  - L.lasset (log assets): positive and strongly significant in many specifications (examples: 0.349*** to 0.491***).
  - L.hhib (bank concentration/highest bank share): positive and significant for both SBS and BC in many specifications (examples: 1.388***, 1.157**, 3.467***, 2.367***).
  - L.kk_compo (quality of institutions): negative and highly significant for BC indicators (examples: -1.919***, -2.830***, -1.720***, -1.960***).
  - L.finopen (financial openness): negative and significant for BC indicators (examples: -1.059***, -1.029***, -1.026***, -0.942***).
  - L.crisis25: negative association with BC in some specs (e.g., -0.239** [0.043], -0.271** [0.032], -0.247* [0.052]) but positive association with FAIL5 and FAIL10 (0.522***, 0.475***), indicating currency crises increase SBS probability while sometimes associated with lower government response probability in bank-level regressions.

### Main conclusions and policy implications (Conclusion and synthesis)
- BC indicators largely measure lagged government responses to banking distress (restorative policy actions), not contemporaneous realizations of systemic bank shocks.
- SBS indicators — constructed from sharp declines in loans, deposits, profits, and interest margins or bank-level profit-capital drops — better capture realizations of systemic bank shocks and yield more consistent, economically interpretable determinants.
- Reinterpretations implied by the evidence:
  - Bank concentration: more concentrated systems are more likely to experience SBS; concentration does not robustly affect government response probability.
  - Deposit insurance: presence of formal deposit insurance raises likelihood/intensity of government response (BC) but does not increase probability of SBS.
  - External shocks and currency/twin crises: significant determinants of SBS and bi-directionally linked with SBS; these relationships are obscured when using BC indicators.
- Policy recommendation: empirical studies and policy analyses should distinguish SBS realizations from BC interventions; using BC indicators as proxies for banking shocks can bias inference about determinants and consequences of banking distress and misinform policy evaluation of regulatory features (e.g., deposit insurance) or market structure.

### Key summary statistics of crisis classifications (Table 1, selected figures)
- Panel A: Summary statistics
  - DD: Total country years = 2350; Total country years excluding crisis years = 2070; as % of total = 88.1; Total crises = 363; Average duration of systemic crises in years = 15.4; Total crisis years = 83; Average duration excluding crisis years after the first crisis = 4.4.
  - CEA: Total country years = 2143; Total country years excluding crisis years = 1833; as % of total = 85.5; Total crises = 382; Average duration of systemic crises in years = 17.8; Total crisis years = 78; Average duration excluding crisis years after the first crisis = 4.9.
  - RR: Total country years = 2375; Total country years excluding crisis years = 2171; as % of total = 91.4; Total crises = 300; Average duration of systemic crises in years = 12.6; Total crisis years = 69; Average duration excluding crisis years after the first crisis = 4.3.
  - LV: Total country years = 2275; Total country years excluding crisis years = 2021; as % of total = 88.8; Total crises = 339; Average duration of systemic crises in years = 14.9; Total crisis years = 84; Average duration excluding crisis years after the first crisis = 4.0.
- Panel B (pairwise comparisons, selected):
  - DD vs CEA: country years in common = 1720; Total agreed = 55; All crisis country years common = 2118; agreed = 263; discrepancies = 202; as % of common = 9.5; as % of agreed crisis = 43.4.
  - DD vs RR: country years in common = 1986; Only first crisis country year: as % of common = 2.3; as % of agreed crisis = 49.5; All crisis country years common = 2187; as % of common = 7.5; as % of agreed crisis = 39.7.
  - CEA vs LV: country years in common = 1769; Only first crisis country year: as % of common = 1.1; as % of agreed crisis = 23.0; All crisis country years common = 2089; as % of common = 4.0; as % of agreed crisis = 24.5.

*Source: _wp09141 - References (Introduction, Model, Empirical results, and References sections).*

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

### _wp09141 - References..............................................................................................................

### Major findings and framing (Introduction and Summary)
- The collapse of the subprime mortgage market in the U.S. in 2007 and ensuing financial instability renewed interest in banking crises; literature documents both similarities (e.g. Reinhart and Rogoff, 2008a) and differences (e.g. Bordo, 2008; Gorton, 2008).
- Empirical literature on bank fragility has focused on documenting empirical regularities while relying on loosely theory-derived definitions of "banking crisis" (BC): what it is, when it occurs, and how long it lasts.
- Most BC indicators are based on crisis dating schemes that identify crisis beginning dates, ending dates, and whether the crisis was “systemic” or not; these schemes are primarily based on information about government actions undertaken in response to banking distress.
- A detailed review shows virtually all BC classifications depend on information obtained from bank regulators and/or central banks and thus effectively measure a government response to perceived crises rather than the onset or duration of adverse shocks to the banking industry.
- Key implication: BC indicators are likely to date banking crisis onsets too late, at least on average, because government responses can lag due to uncertainty about the extent of problems and political economy considerations.

### Conceptual critique and identification problem
- Equating the dating of a government response to the dating of a systemic bank shock is problematic: it confounds the realization of an adverse bank shock with the restorative policy response.
- As noted by De Nicolò et al. (2004), using BC-type indicators as proxies for systemic risk realizations makes measurement “very sensitive to the classification criterion used.”
- Disentangling adverse shocks from policy responses is essential to understanding the mechanics of bank fragility; the paper’s main contribution is to separate these two elements.

### Theoretical contribution: Systemic Bank Shocks (SBS)
- Authors construct a simple model of a banking industry where an adverse shock to the banking system and a government response are explicitly defined and modeled.
- The model’s objective is to derive measures of systemic bank shocks (SBS) that are “empirically relevant” — i.e., measures that can be obtained from available data for a large number of countries and years.
- The model is intended as an identification tool for theory-based measures of SBS and is not presented as a contribution to banking theory per se.

### Empirical strategy and datasets
- Two samples are used for empirical analysis: a country-level dataset and a firm-level dataset (the latter includes a large number of banks).
- The empirical contribution is to separately identify binary indicators of SBS shocks and BC indicators.
- For BC indicators, four different data series from the literature are employed.
- The existing literature typically interprets an SBS event and a BC event as the same; the paper emphasizes two problems with that approach:
  - The two events occur on different dates.
  - One event is adverse for the industry (SBS shock) while the other is beneficial (government intervention).

### Empirical findings on determinants
- Causal variables studied include bank market structure, presence or lack of deposit insurance, and the occurrence of an external shock (e.g. a currency crisis).
- Each explanatory variable has a different effect on the probability of an adverse shock to the banking industry (SBS indicators) and on the probability of government intervention (BC indicators).
- This differentiation helps explain contradictory findings in the literature and suggests many prior interpretations may be erroneous.

### Structure of subsequent analysis (paper roadmap)
- Section II: Discusses criteria used in the literature to date beginnings, severity, and endings of banking crises; examines four well-known crisis dating studies and shows dating information is obtained from bank regulators and/or central banks, hence recording government intervention more than realizations of adverse shocks.
- Section III: Constructs BC indicators based on the four major crisis classifications used later in the empirical work and documents significant discrepancies among the four BC indicators in dating beginnings and endings of banking crises, indicating researcher disagreement in dating the same episodes of financial distress.

### Key implications for empirical work and policy analysis
- Researchers should distinguish between realizations of systemic bank shocks (SBS) and government interventions (BC) when studying causes and consequences of banking crises.
- Using BC indicators as proxies for bank shocks can bias inference about determinants and consequences of banking distress because interventions may alter subsequent outcomes.
- Policy analysis that aims to evaluate the impact of regulatory features (e.g., deposit insurance) or market structure on banking fragility should account for differing effects on SBS and BC probabilities.

*Source: _wp09141 - References (Introduction and Summary sections)*

### Section IV presents a theoretical model in which banking problems are produced by the

### _wp09141 - Section IV presents a theoretical model in which banking problems are produced by the

### Theoretical model (Section IV)
- Banks are Cournot-Nash competitors that raise insured deposits, make risky loans, and hold risk free government bonds; the deposit insurer bails out banks when they fail.
- Agents: government, entrepreneurs, depositors, banks. All agents are risk-neutral; time is discrete.
- Entrepreneurs: continuum indexed by reservation income distributed uniformly on [0,1]. Projects require investment standardized to 1 and yield Y with probability P1t+ (a random variable observed at date 1t+). Entrepreneur undertakes project if condition (1) holds: L R * EP_{t+1} * Y >= a − (as presented in the source).
- Deposits: fully insured; inverse supply curve R = Z α_t (notation preserved).
- Government: issues fixed amount of bonds B; guarantees deposits and issues additional bonds to pay depositors when banks fail; intervenes when realized bank profits at 1t+ are negative.
- Sequence: at t banks collect deposits and make decisions based on expected p ≡ E_t+1[P_{t+1}]; at 1t+ realizations occur, negative profits imply a systemic bank shock (SBS) and government intervention at 2t+.
- Comparative statics (linear specification): adverse shocks (p decreases, α increases, Y decreases) produce predictable observable effects:
  - Aggregate loans: decreases when p decreases; decreases when α increases; decreases when Y decreases.
  - Aggregate deposits: decreases when p decreases; decreases when α increases; decreases when Y decreases.
  - Bond interest rate: ↓ when p decreases; ↑ when α increases; ↓ when Y decreases.
  - Loan rate: ↑ when p decreases; ↑ when α increases; ↑ when Y decreases.
  - Deposit rate: ↓ when p decreases; ↑ when α increases; ↑ when Y decreases.
  - Loan rate − Deposit rate (interest margin): ↑ in response to all three adverse shocks.
  - Realized profits: ↓ in response to all three adverse shocks.
- Model implication: identify systemic bank shock with severe declines in loans, deposits, bank profits, and significant increases in interest rate margins; empirical measures will depend on data availability and timing.

### Empirical strategy and datasets (Sections V, VIII)
- Country-level panel: up to 91 countries for the 1980-2002 period; merges and updates datasets used in DD (2005) and Beck et al. (2006).
- Bank-level panel: accounting data for about 120 emerging and developing countries for 1993-2004 (used to construct SBS indicators based on bank profitability and capitalization).
- Empirical steps:
  1. Benchmark Logit regressions with BC indicators (four crisis classifications) as dependent variables using contemporaneous macro variables commonly used in literature.
  2. Construct theory-based SBS indicators (based on extreme drops in real lending and deposits for country dataset; based on sharp declines in profits standardized by assets and capital buffers for bank-level dataset). Include lagged SBS in regressions predicting BC indicators to test lead-lag.
  3. Estimate Logit regressions with SBS indicators as dependent variables to compare explanatory power and determinants versus BC regressions.

### Crisis classifications and BC (banking crisis) indicators (Sections II–III)
- Four BC binary indicators (1 if country-year is crisis year, 0 otherwise): DD (Demirgüç-Kunt and Detragiache), CEA (Caprio et al.), RR (Reinhart and Rogoff), LV (Laeven and Valencia).
- Two versions of each indicator:
  - "Start-date only": excludes country-years after first crisis year (identifies starting dates).
  - "All crisis years": includes all crisis country-years beginning with and beyond the starting date. For RR and LV (which do not report durations), durations from CEA or DD are used where available.
- Dating discrepancies:
  - Pairwise disagreements in starting dates and durations are frequent; ratio of total crisis ranking discrepancies divided by total crisis rankings varies between 24.5% and 49.5%.
  - All four classifications agree on 41 dates of crisis onset.
- Conclusion: BC classifications predominantly index government responses to banking distress (often derived from reported government measures), producing pervasive dating discrepancies.

### SBS indicators and predictive tests (Section V)
- Country-sample SBS indicators constructed due to data availability:
  - SBSL25 and SBSL10: equal 1 if real domestic lending growth is lower than the 25% and 10% percentiles of the entire cross-country distribution of real domestic bank credit growth.
  - SBSD25 and SBSD10: equal 1 if growth rate of deposit-to-GDP ratio is lower than the 25% and 10% percentiles of its distribution.
- Key empirical findings (country-level):
  - Lagged SBS lending indicators predict BC indicators in all specifications (both 25th and 10th percentile cut-offs).
  - Lagged SBS deposit indicators are positively associated with BC indicators but coefficients are weakly significant in only two of eight specifications.
  - Regressions with SBS indicators as dependent variables show stronger and more intuitively consistent explanatory power: more explanatory variables have expected signs and statistical significance compared with BC regressions.
  - External variables (terms of trade changes, exchange rate depreciation, currency and twin crises) positively and significantly impact probability of SBS; these external factors have weaker or no significant impact on BC indicators.
  - Conclusion: BC indicators appear to represent lagged government responses to SBS realizations rather than contemporaneous realizations of systemic shocks.

### Market structure and deposit insurance (Section VI)
- Bank concentration measures used: C3 concentration ratio (concenmean) and inter-temporal average Hirschman-Herfindhal index (avgherf / hhib depending on table).
- Country-level results:
  - Bank concentration significantly increases probability of a systemic bank shock when SBS indicators are the dependent variable.
  - Bank concentration does not significantly affect probability of government response when BC indicators are dependent variables (no robust effect across specifications).
  - Deposit insurance indicator (di = 1 if deposit insurance system in place) is positively and significantly associated with BC indicators (government response more likely when formal deposit insurance exists).
  - Deposit insurance does not significantly affect the probability of a systemic bank shock (SBS dependent regressions).
  - Moral hazard index (princom) is not statistically significant for probability of systemic bank shock; quality of institutions index (kk_compo) is negative and significant for probability of government response to distress but not consistently for SBS.
- Interpretation: presence of deposit insurance appears to raise likelihood or intensity of government intervention, not the incidence of systemic shocks; more concentrated systems are more prone to SBS.

### Currency and “twin” crises (Section VII)
- Currency crisis indicators constructed using Frankel and Wei (2004) algorithm based on monthly data: crisis35, crisis25, crisis15 thresholds equal to 35%, 25%, and 15% sums of exchange rate depreciation plus loss of international reserves. Twin crisis indicator equals 1 if sum passes 25% threshold and real credit growth is lower than the 25th percentile.
- Lagged-variable Logit specification to reduce simultaneity/endogeneity.
- Findings:
  - BC indicators: lagged real GDP growth is consistently negative and significant; lagged real interest rates positive and significant in many regressions; external variables (terms of trade, financial openness, exchange rate flexibility) generally do not significantly predict BC indicators; currency crises weakly predict BC indicators in some specifications; twin crisis indicators do not predict BC indicators.
  - SBS indicators: currency crises and twin crises positively and significantly predict the probability of a systemic bank shock in most regressions (exceptions noted in Equations 4-5). Worsening terms of trade is a strong determinant of SBS (negative sign on totch significant in 6 out of 8 regressions). Financial openness has some independent impact on SBS (significant in 3 out of 8 regressions). Exchange rate flexibility largely unimportant for SBS in country-level tests.
  - Currency crises as dependent variables: SBS indicators significantly predict currency crises in most specifications (SBS → currency crises linkage strong), whereas BC indicators do not predict currency crises when used as explanatory variables.
- Overall: external shocks and currency crises have significant two-way relationships with systemic bank shocks when SBS measures are used; these relationships are obscured when BC indicators are used.

### Bank-level evidence (Section VIII)
- Bank-level SBS indicators constructed using sharp declines in bank profitability adjusted for capitalization:
  - FAIL5 and FAIL10 correspond to 5th and 10th percentiles of distribution of (profits + equity capital)/assets across time and countries.
- Random effect Logit regressions with lagged explanatory variables; controls include gdppc, growth, infl, depr, and log assets (lasset).
- Findings:
  - Lagged SBS indicators strongly predict BC indicators in all specifications (high significance).
  - BC regressions at bank-level: higher GDP growth predicts lower probability of government response; inflation positively and significantly predicts BC indicators (stronger than in country data).
  - Bank concentration (hhib) positively and significantly increases probability of both government response (BC) and systemic bank shock (SBS).
  - Deposit insurance increases probability of government response (BC) but does not significantly affect probability of SBS.
  - External environment: financial openness associated with lower probability of government response (BC) but no effect on SBS; more flexible exchange rate arrangements associated with lower probability of both government response and SBS (a result not seen in country-level data); currency crises have little or negative impact on probability of government response but increase probability of systemic bank shock.
- Conclusion: bank-level data confirm country-level findings and strengthen evidence that SBS and BC indicators capture different phenomena; bank-level heterogeneity improves power and consistency of results.

### Main conclusions (Section VI [Conclusion] and across analysis)
- BC indicators largely measure lagged government responses to banking distress (restorative policy actions), not the contemporaneous realization of systemic bank shocks.
- SBS indicators (theory-based measures derived from declines in loans, deposits, profits, and interest margins or bank-level profit-capital drops) more directly capture realizations of systemic bank shocks and yield more consistent, economically interpretable determinants.
- Key reinterpretations implied by findings:
  - Bank concentration: more concentrated systems are more likely to experience systemic bank shocks (SBS), while concentration does not robustly affect probability of government response (BC).
  - Deposit insurance: presence of formal deposit insurance raises likelihood/intensity of government response (BC) but does not increase the probability of systemic bank shocks (SBS).
  - External shocks and currency/twin crises: significant determinants of SBS but weak predictors of BC indicators; SBS indicators predict currency crises strongly.
- Policy implication: empirical studies using BC indicators as proxies for banking crises may have conflated the incidence of banking shocks with subsequent policy responses; many cross-country determinants of bank fragility need to be reassessed using SBS-type measures.

*Source: _wp09141 - Section IV presents a theoretical model in which banking problems are produced by the (IMF PDF content provided).*

### REFERENCES

### _wp09141 - REFERENCES

### References cited
- Allen, Franklin and Douglas Gale, 2007, Understanding Financial Crises, Oxford University Press, Oxford, U.K.
- Arteta, Carlos, 2003. “Are Financially Dollarized Countries More prone to Costly Crises?”, International Finance Discussion Paper #763, March, Board of Governors of the Federal Reserve System, Washington D.C..
- Arteta, Carlos and Barry Eichengreen, “Banking Crises in Emerging Markets: Presumptions and Evidence”, in Blejer, M. and Skreb, M. (eds) Financial Policies in Emerging Markets, Cambridge, Massachusetts, M.I.T. Press.
- Barth, James, Gerard Caprio and Ross Levine, 2004, “Bank Regulation and Supervision: What Works Best?”, Journal of Financial Intermediation, Vol. 13: 205-248.
- Beck, Thorsten, Asli Demirgüç-Kunt and Ross Levine, 2006, “Bank Competition, Concentration and Crises: First Results”, Journal of Banking and Finance, Volume 30, 1581-1603.
- Bordo, Michael D., 2008, “Growing up to Financial Stability”, Economics, Vol. 2, :1-16; http://www.economics-ejournal.org/economics/journalarticles/2008-12
- Bordo, Michael, Barry Eichengreen, Daniela Klingebiel and Maria Soledad Martinez-Peria, 2001, “Is The Crisis Problem Growing More Severe?”, Economic Policy, volume 16, issue 32.
- Boyd, John H., Chun Chang and Bruce D. Smith, 2002, “Deposit Insurance: A Reconsideration”, Journal of Monetary Economics, Vol, 49, 6:1235-1260.
- Boyd, John H., Chun Chang and Bruce D. Smith, 2004, “Deposit Insurance and Bank Regulation in a Monetary Economy: A General Equilibrium Exposition”, Economic Theory, Vol. 24: 741-767.
- Boyd, John H., and Gianni De Nicolò, 2005, “The Theory of Bank Risk Taking and Competition Revisited,” Journal of Finance, Volume 60, Issue 3, 1329-1343.
- Boyd, John H., Gianni De Nicolò, and Abu M. Jalal, 2006, “Bank Risk Taking and Competition Revisited: New Theory and New Evidence,” IMF Working Paper 06/297.
- Boyd, John H., Gianni De Nicolò, and Abu M. Jalal, 2009, “Bank Competition, Risk and Asset Allocations”, forthcoming IMF Working Paper.
- Boyd, John H., Sungkyu Kwak and Bruce Smith, 2005, “The Real output Losses Associated with Modern Banking Crises”, Journal of Money, Credit and Banking, Vol. 37, no.6: 977-999.
- Caprio, Gerard and Daniela Klingebiel, 1996, “Bank Insolvencies: Cross-Country Experience,” Policy Research Working Paper PRWP1620, Washington, D.C.: World Bank.
- Caprio, Gerard and Daniela Klingebiel, 1999, “Episodes of Systemic and Borderline Financial Crises,” Washington, D.C.: World Bank Mimeo.
- Caprio, Gerard, Daniela Klingebiel, Luc Laeven and Guillermo Noguera, 2005, “Banking Crises Database”, in Systemic Financial Crises, P. Honahan and L. Laeven eds., Cambridge University Press, Cambridge, U.K.
- De Nicolò, Gianni, P. Bartholomew, J. Zaman and M. Zephirin, 2004, “Bank Consolidation, Internationalization and Conglomeration: Trends and Implications for Financial Risk”, Financial Markets, Institutions & Instruments, Vol. 13, No. 4, 173-217.
- De Nicolò, Gianni, Patrick Honohan and Alain Ize, 2005, “Dollarization of Bank Deposits: Causes and Consequences”, Journal of Banking and Finance, Vol. 29: 1697-1727.
- De Nicolò, Gianni and Elena Loukoianova, 2007, “Bank Ownership, Market Structure and Risk” IMF Working Paper 07/215.
- Demirgüç-Kunt, Asli and Enrica Detragiache, 1997, “The Determinants of Banking Crises: Evidence from Developing and Developed Countries”, IMF Working Paper 97/106.
- Demirgüç-Kunt, Asli and Enrica Detragiache, 2002, “Does Deposit Insurance Increase Banking System Stability? An Empirical Investigation,” Journal of Monetary Economics, Vol. 49, 1373-1406.
- Demirgüç-Kunt, Asli and Enrica Detragiache, 2005, “Cross-Country Empirical Studies of Systemic Bank Distress: A Survey”, National Institute of Economic Review, no.192, April.
- Eichengreen, Barry and Andy Rose, 1998 “Staying Afloat When the Wind Shifts: External Factors and Emerging-Market Banking Crises” NBER Working Paper No. 6370.
- Frankel, Jeffrey A. and Shang-Jin Wei, 2004, ”Managing Macroeconomic Crises”, NBER Working Paper # 10907, November.
- Gorton, Gary, 2008, ”The Subprime Crisis”, NBER Working Paper # 14398, October.
- International Monetary Fund, 1998, “Financial Crises: Characteristics and Indicators of Vulnerability”, Chapter 4 in the World Economic Outlook, May, Washington D.C.; 74-97.
- Kaminsky, Graciela. and Carmen Reinhart, 1999, “The Twin Crises: The Causes of Banking and Balance-of-Payment Problems,” American Economic Review, Vol. 89, No. 3, 473-500.
- Laeven, Luc and Fabian Valencia, 2008, “Systemic Banking Crises: A New Database”, IMF Working Paper 08/224, September.
- Lane, Philip R. and Gian Maria Milesi-Ferretti, 2007, “The External Wealth of Nations mark II: Revised and Extended Estimates of Foreign Assets and Liabilities”, Journal of International Economics, Vol. 73, Issue 2, 223-250.
- Lindgren, Carl-Johan, Gillian Garcia and Matthew Saal, 1996, Bank Soundness and Macroeconomic Policy, International Monetary Fund, Washington D.C.
- Martinez-Miera, David and Rafael Repullo, 2008, ”Does Competition Reduce the Risk of Bank Failure?”, CEPR Discussion Paper #6669, January.
- Rancière, Romain, Aaron Tornell and Frank Westermann, 2008, “Systemic Crises and Growth”, Quarterly Journal of Economics, Vol 34, Issue 1: 359-406.
- Reinhart, Carmen and Kenneth Rogoff, 2004, “The Modern History of Exchange Rate Arrangements: A Reinterpretation”, Quarterly Journal of Economics, Vol. 64, no.1: 1-48.
- Reinhart, Carmen and Kenneth Rogoff, 2008a, “Is the 2007 U.S. Subprime Financial Crisis So Different? An International Historical Comparison,” American Economic Review, Papers and Proceedings, 98:2, 339-334.
- Reinhart, Carmen and Kenneth Rogoff, 2008b, “This Time Is Different: A Panoramic View of Eight Centuries of Financial Crises” NBER Working Paper # 13882, March.
- Von Hagen, Jürgen, and Tai-Kuang Ho, 2007, “Money Market Pressure and the Determinants of Banking Crises”, Journal of Money, Credit and Banking, Vol. 39, no. 5, 1037-1066.

### Summary statistics of crisis classifications (Table 1)
- Panel A: Summary Statistics of Classifications of Systemic Banking Crises
  - DD: Total country years = 2350; Total country years excluding crisis years = 2070; as % of total = 88.1; Total crises = 363; Average duration of systemic crises in years = 15.4; Total crisis years = 83; Average duration excluding crisis years after the first crisis = 4.4
  - CEA: Total country years = 2143; Total country years excluding crisis years = 1833; as % of total = 85.5; Total crises = 382; Average duration of systemic crises in years = 17.8; Total crisis years = 78; Average duration excluding crisis years after the first crisis = 4.9
  - RR: Total country years = 2375; Total country years excluding crisis years = 2171; as % of total = 91.4; Total crises = 300; Average duration of systemic crises in years = 12.6; Total crisis years = 69; Average duration excluding crisis years after the first crisis = 4.3
  - LV: Total country years = 2275; Total country years excluding crisis years = 2021; as % of total = 88.8; Total crises = 339; Average duration of systemic crises in years = 14.9; Total crisis years = 84; Average duration excluding crisis years after the first crisis = 4.0
- Panel B: Pairwise comparisons (selected figures)
  - DD vs CEA: country years in common = 1720; Total agreed = 55; Only first crisis country year: as % of common = 2.0; as % of agreed crisis = 38.2; All crisis country years: country years in common = 2118; agreed = 263; discrepancies = 202; as % of common = 9.5; as % of agreed crisis = 43.4
  - DD vs RR: country years in common = 1986; Only first crisis country year: as % of common = 2.3; as % of agreed crisis = 49.5; All crisis country years: country years in common = 2187; as % of common = 7.5; as % of agreed crisis = 39.7
  - CEA vs LV: country years in common = 1769; Only first crisis country year: as % of common = 1.1; as % of agreed crisis = 23.0; All crisis country years: country years in common = 2089; as % of common = 4.0; as % of agreed crisis = 24.5

### Key cross-country logit regression findings (Tables 2–5, 7–14)
- Consistent significant coefficients across many specifications (preserve exact reported signs and significance):
  - rgdpgr (GDP growth rate)
    - Negative and statistically significant in many specifications: examples include coefficients -0.109***, -0.121***, -0.130***, -0.102*** (Table 2, full sample), and -0.0674***, -0.0867***, -0.0840***, -0.0839*** (Table 3, full sample). In common-sample specifications coefficients reach -0.139***, -0.139***, -0.150***, -0.144*** (Table 2, columns (5)-(8)).
  - rint (real interest rate)
    - Small positive coefficients often significant: examples 0.000417** [0.0116], 0.000353** [0.0284], 0.000646** [0.0158], 0.000301** [0.0361] (Table 2). In other tables e.g., 0.000295* [0.0700] (Table 3).
  - m2res (ratio of M2 to foreign exchange reserves)
    - Generally positive and sometimes significant: examples 0.00117 [0.103], 0.00114* [0.0975], 0.00125** [0.0138] (Table 2); 0.00206* [0.0508], 0.00191** [0.0401] (Table 3).
  - privcrd_gdp (bank credit to private sector to GDP)
    - Mixed results: significant positive in some specifications for BC indicators (e.g., 0.00129*** [0.0000312] in Table 2 column (1); 0.00111*** [0.000801] in Table 3 column (1)); negative or insignificant in other specifications for some crisis definitions.
  - L2.rdomcredgr (real domestic bank credit growth lagged twice)
    - Positive and significant in some start-date regressions: 0.0127** [0.0453], 0.0124** [0.0405], 0.0137** [0.0144] (Table 2).
- SBS (supervisory/bank soundness) indicators predict BC indicators (Table 4)
  - L.SBSL25 and L.SBSL10 (lagged SBS lending indicators) have positive and statistically significant coefficients:
    - L.SBSL25 = 0.412*** [0.00388] (DD), 0.576*** [0.000126] (CEA), 0.519*** [0.000126] (RR), 0.428*** [0.00733] (LV).
    - L.SBSL10 = 0.365** [0.0469] (DD), 0.785*** [0.0000272] (CEA), 0.771*** [0.0000261] (RR), 0.632*** [0.000901] (LV).
  - L.SBSD25 and L.SBSD10 (lagged SBS deposit indicators) show weaker or mixed significance (Table 5), with L.SBSD10 = 0.340* [0.0922] for CEA and 0.338* [0.0949] for LV in some specifications.
- SBS indicators determinants (Table 6, Table 8)
  - SBS lending indicators (SBSL25, SBSL10) are negatively related to rgdpgr: e.g., -0.119*** [0.000000706] (SBSL25) and -0.0948*** [0.00119] (SBSL10) in Table 6; -0.130*** and -0.135*** in Table 8.
  - Terms of trade changes (totch) and exchange rate depreciation (depr) are often positive and significant predictors of SBS indicators:
    - Example: depr = 1.238*** [0.00274] (SBSL25) and depr = 1.615*** [0.000224] (SBSL10) in Table 6; depr = 2.798*** [0.0000635] and 3.415*** [0.000000164] in Table 8.
  - Bank concentration (concen_mean, avgherf) positively associated with SBS indicators in several specifications (Table 8): concen_mean = 1.656*** [0.00437], avgherf = 1.460*** [0.0000475] (SBSL25).
- Deposit insurance, institutional quality, and concentration (Tables 9–11)
  - di (deposit insurance indicator) shows positive and significant associations in some regressions:
    - Example: di = 1.325*** [0.00185] for CEA (Table 9 column (2)); di = 1.105*** [0.00423] for LV (Table 9 column (4)). In other specifications di coefficients vary in sign and significance.
  - kk_compo (quality of institutions) often negative and significant for BC indicators:
    - Example: kk_compo = -1.295*** [0.000482] for CEA (Table 11).
  - avgherf (average Herfindhal index) often positively associated with SBS indicators and significant: e.g., avgherf = 1.416*** [0.000249] (SBSL25, Table 9).
- Currency and twin crises interactions (Tables 11–13)
  - L.crisis25 (currency crisis indicator lagged) often positive and in some cases significant for BC indicators (e.g., 0.501* [0.0685] for CEA in Table 11).
  - SBS indicators predict currency crises in some specifications:
    - L.SBSL25 = 0.420* [0.053], 0.414** [0.036], 0.329* [0.076] (Table 13).
    - L.SBSD25 = 0.258 [0.249], 0.435** [0.037], 0.607*** [0.009] (Table 13).

### Bank-level random-effect logit regression highlights (Tables 14–15)
- Table 14: SBS indicators predict BC indicators (bank-level random effect logit)
  - L.gdppc = 0.000*** (significant across specifications)
  - L.growth = -0.164*** (consistent negative effect on BC indicators)
  - L.infl = 0.045*** (positive)
  - L.lasset = 0.349*** to 0.410*** (positive and significant)
  - L.FAIL5 and L.FAIL10 (bank failure proxies) positive and significant: e.g., L.FAIL5 = 0.819***, L.FAIL10 = 0.772***.
- Table 15: Determinants of BC and SBS indicators (bank-level)
  - L.gdppc = 0.000*** (BC indicators), L.growth = -0.200*** to -0.153*** (BC indicators)
  - L.lasset positive and strongly significant: 0.370***, 0.357***, 0.447***, 0.491*** across BC indicators.
  - L.hhib (bank-level concentration/highest bank share) positive and significant across many SBS and BC specifications: e.g., 1.388***, 1.157**, 3.467***, 2.367***.
  - L.kk_compo (quality of institutions) negative and highly significant for BC indicators: -1.919***, -2.830***, -1.720***, -1.960***.
  - L.finopen negative and significant for BC indicators: -1.059***, -1.029***, -1.026***, -0.942***.
  - L.crisis25 negatively associated with BC indicators in some specifications: -0.239** [0.043], -0.271** [0.032], -0.247* [0.052]; but positively associated with FAIL5 and FAIL10: 0.522***, 0.475***.

### Crisis dating and counts (Table A1 and continuation)
- Number of crises reported across classifications: Number of crises = 83, 33, 78, 69, 85 (table shows totals across columns)
- Number of crisis/years as % of total years = 15.3, 7.6, 16.1 (as reported in the continued summary)
- Average duration of crisis across classifications = 4.4, 5.6, 4.9 (as reported in the continued summary)
- Country-level crisis start dates and durations are tabulated extensively for many countries (Table A1 and continuations), showing differing start dates and durations across DD, CEA, RR, LV, and Caprio et al. (2005) classifications.

*Source: _wp09141 - REFERENCES (IMF working paper PDF content provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09141.pdf_