## wpiea2019094 - 2016. Following

## Source details

**Canonical URL:** [wpiea2019094 - 2016. Following](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019094.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019094.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019094.pdf.json)

---

### Literature and framing
- Three theoretical views summarized:
  - "supply-leading view" (Schumpeter (1934), Goldsmith (1969), McKinnon (1973) and Shaw (1973)).
  - "demand-following view" (Robinson (1952)).
  - Skepticism about overstated contribution (Lucas (1988)).
- Endogenous growth contributions cited: Greenwood and Jovanovic (1990), Bencivenga (1991), Saint-Paul (1992), Rousseau and Wachtel (2000).
- Empirical nonlinearities:
  - Inverted U-shaped relationship between financial development and growth with a turning point around "90 − 110% of GDP" (Cecchetti and Kharoubbi (2012), Law and Singh (2014), Arcand et al. (2015)).
- Country/income heterogeneity:
  - Growth effects vary by country/income level (Rioja and Valev (2004a, 2004b), Rousseau and Wachtel (2000)); growth not affected when inflation exceeds "13 percent".
- Banking-crisis EWS literature:
  - Two main approaches: signal approach and binary regression approach (logit/probit).
  - Common credit variables: credit-to-GDP ratio and credit growth rate; predictive power is mixed across studies (Demirguc-Kunt and Detragiache; Davis and Karim; Mathonnat and Minea; Jorda et al.; Schularick and Taylor; Borio and Lowe).

### Methodology: dynamic panel logit model
- Model structure and parameters:
  - Banking crisis binary variable y_t equals 1 during crisis periods and 0 otherwise.
  - Explanatory matrix x_t contains financial development indicators and macroeconomic controls.
  - Dynamic panel logit uses logistic c.d.f. F; country fixed effects η_i; parameters α (lagged binary crisis), π (lagged index), and β (one-step-ahead effect of explanatory variables).
  - α ≠ 0 or π ≠ 0 implies static logit models are biased; crisis persistence captured via α and π.
- Estimation approach:
  - Builds on Kauppi and Saikkonen (2008) and Candelon et al. (2014).
  - Exact maximum likelihood (EML) framework with fixed effects; no cross-sectional dependence assumed; correction à la Carro (2007) implemented.
  - Four model variants: static (α = π = 0), dynamic with lagged y_t−1, dynamic with lagged π_t−1, and combined dynamic with both lagged terms.
  - Model selection by Bayesian information criterion (BIC).

### Data and financial development measures
- Financial development (FD) decomposition:
  - Two aggregate sub-indices: Financial Institutions (FI) and Financial Markets (FM).
  - Each sub-index decomposed into three dimensions: Depth (D), Access (A), Efficiency (E).
  - Six sub-indices: FID, FIA, FIE, FMD, FMA, FME.
  - Global FD index built from FI and FM via PCA; FI and FM built from sub-indices via PCA; sub-indices built from component variables via PCA (example: FID includes Private-sector credit, Pension fund assets, Mutual fund assets, and Insurance premiums, life and non-life).
- Data sources and coverage:
  - Svirydzenka (2016) database: 183 countries covering the period "1980−2015" at annual frequency for FD indices.
  - Banking crisis dummy: Laeven and Valencia (2013) database (covers up to 2011) extended for "2012 − 2016" with Candelon et al. (2018) data; banking crisis database contains "100 countries from 1980 to 2016 on a yearly basis".
  - Macroeconomic controls: output growth rate and interest rate spread (difference between the "10 − year treasury rate" and a "3 − month monetary rate"), yearly, from IFS.
  - Final sample: "our sample comprises 98 countries for the period 1980 – 2016."

### Empirical findings — baseline and heterogeneity
- Model selection and dynamics:
  - Model (2), including lagged binary variable y_t−1, has the lowest BIC in each specification and is always selected, implying static logit models are inadequate and crisis persistence matters.
  - Confirms nonlinearity of causality in crisis prediction.
- Macroeconomic controls:
  - Output growth rate: an increase reduces significantly the probability of a banking crisis one year ahead.
  - Interest rate spread (long minus short): an inversion of the yield curve (negative spread) signals a higher risk of a banking crisis.
- Financial development effects (whole panel):
  - Financial development increases the probability of occurrence of a banking crisis in a one-year period.
  - Destabilizing effect operates primarily via financial institution development and to a lesser extent via financial market development.
- Heterogeneity by country group (three clusters: AM, EM, LIDC):
  - Advanced economies (AM):
    - Depth (FID) and access (FIA) cause banking crises.
    - Efficiency reduces the future occurrence of a banking crisis.
    - Macro variables: term spread and output growth important for future crises.
  - Emerging markets (EM):
    - Access to financial institutions is stabilizing (reduces crisis probability).
    - Depth/efficiency are not aggregate significant in increasing crisis probability; output growth matters.
  - Least income developed countries (LIDC):
    - Financial institution depth (FID) is a leading indicator for future crises.
    - Access to financial institutions increases financial stability and reduces probability of crises a year ahead (linked to financial inclusion).
    - Term spread not included because of data availability; most LIDC are small open economies and interest rate takers with unmature financial capital markets.
- Aggregate vs. sub-index significance:
  - Aggregate variables FD, FI and FM are only significant for the advanced countries’ cluster, but sub-indices are significant for the three clusters.
- Dimensions summary:
  - Financial institution depth is destabilizing across clusters.
  - Financial institution access reduces future occurrence of a crisis.
  - Financial institution efficiency typically has a negative sign but often not statistically significant.
  - Financial market depth shows a positive sign and access a negative sign; almost all market coefficients often not statistically different from 0.

### Robustness checks and nonlinear interactions
- Interaction framework:
  - Financial development interacts with regimes of nonperforming loans (NPL) and capital account openness (KAO, Chinn and Ito (2006)).
  - Interaction index z_j equals 1 when z_i,t > median(z_i) and 0 otherwise.
  - Models include exchange rate regime (Err) dummies for fixed and flexible regimes (Levy-Yeyati and Sturzenegger (2005), (2016)).
- Key robustness outcomes:
  - Interactive terms (KAO and NPL) do not affect the main FD–crisis relationship: interaction coefficients not significantly different from 0 at 99%.
  - Fixed exchange rates: results similar to baseline.
  - Flexible exchange rate regime:
    - Access to financial institutions becomes destabilizing.
    - Financial deepening becomes stabilizing.
    - Only efficiency of financial institutions appears to decrease occurrence of a banking crisis (possible role of capital movement on foreign exchange markets).
- Sample and horizon checks:
  - Pre-2008 estimation: results similar to baseline; impact not driven by the great crisis.
  - Two-year horizon (lag-two specification): results similar to baseline.
- Policy implication from robustness:
  - Link between financial development and probability of banking crisis is structural and holds across samples and horizons, supporting need for structural regulation policies independent of business cycle or temporary events.

### Policy implications and recommendations
- Regulatory measures supported:
  - Implement capital requirements and access control to loans and deposits for financial institutions.
- Calibration by country development level:
  - Advanced countries:
    - Impose strict access control for financial intermediaries (restricting access that may increase nonperforming loans).
    - Impose higher capital requirements on banks to smooth increases in financial access and depth.
  - Emerging markets:
    - Do not necessarily impose higher capital requirements in the same way; regulators should encourage higher efficiencies in financial institutions.
    - Access can be increased to promote stability.
  - Low- and middle-income countries (LIDC and EM):
    - Enhance access to financial services to promote financial inclusion and reduce inequalities.
    - Support fintech innovations (mobile application payments, etc.) to expand access.
- Macroprudential and supervisory adjustments:
  - Fintech explosion (mobile payment, cryptocurrency, and offshore banking) raises regulatory and supervisory concerns; macroprudential rules should be adjusted accordingly.
  - Financial stability assessments (e.g., FSAP) should include a shock associated with financial development and its components to better assess vulnerabilities amid financial innovation.
- Key message:
  - Financial development can be destabilizing; regulation should not be uniform across countries and must account for the country’s degree of development and differing roles of depth, access, and market vs. institution channels.
  - Structural regulation policies should be independent of the business cycle or specific temporary events.

*Italic source: wpiea2019094 - 2016. Following (IMF PDF content).*

### 2016. Following

### wpiea2019094 - 2016. Following

### Literature and framing
- Three theoretical views on finance and growth are summarized: the "supply-leading view" (Schumpeter (1934), Goldsmith (1969), McKinnon (1973) and Shaw (1973)), the "demand-following view" (Robinson (1952)), and skepticism about overstated contribution (Lucas (1988)). Endogenous growth contributions cited: Greenwood and Jovanovic (1990), Bencivenga (1991), Saint-Paul (1992), Rousseau and Wachtel (2000).
- Recent empirical work finds nonlinearities and an inverted U-shaped relationship between financial development and growth, with a turning point around "90 − 110% of GDP" (Cecchetti and Kharoubbi (2012), Law and Singh (2014), Arcand et al. (2015)).
- Evidence that growth effects vary by country/income level: Rioja and Valev (2004a, 2004b), Rousseau and Wachtel (2000) (note: growth not affected when inflation exceeds "13 percent").
- The literature on banking-crisis early warning systems (EWS) uses two main approaches: the signal approach and the binary regression approach (logit/probit). Credit variables often considered: credit-to-GDP ratio and credit growth rate; mixed evidence on predictive power (Demirguc-Kunt and Detragiache; Davis and Karim; Mathonnat and Minea; Jorda et al.; Schularick and Taylor; Borio and Lowe).

### Methodology: dynamic panel logit model
- Model structure:
  - Banking crisis binary variable {y_t} for country i takes value 1 during crisis periods and 0 otherwise.
  - Explanatory matrix {x_t} contains financial development indicators and macroeconomic controls.
  - Dynamic panel logit formulation uses logistic c.d.f. F; country fixed effects η_i; parameters α (lagged binary crisis), π (lagged index), and β (one-step-ahead effect of explanatory variables).
  - α ≠ 0 or π ≠ 0 implies static logit models are biased; persistence of crises captured by α and π.
- Estimation:
  - Builds on Kauppi and Saikkonen (2008) and Candelon et al. (2014).
  - Exact maximum likelihood (EML) framework with fixed effects; no cross-sectional dependence assumed; correction à la Carro (2007) implemented.
  - Four model variants considered: static (α = π = 0), dynamic with lagged y_t−1, dynamic with lagged π_t−1, and combined dynamic with both lagged terms.
  - Model selection by Bayesian information criterion (BIC).

### Data and financial development measures
- Financial development (FD) decomposed into:
  - Two aggregate sub-indices: Financial Institutions (FI) and Financial Markets (FM).
  - Each sub-index decomposed into three dimensions: Depth (D), Access (A), Efficiency (E).
  - Six sub-indices: FID, FIA, FIE, FMD, FMA, FME.
  - Global FD index built from FI and FM via PCA; FI and FM built from sub-indices via PCA; sub-indices built from component variables via PCA (example: FID includes Private-sector credit, Pension fund assets, Mutual fund assets, and Insurance premiums, life and non-life).
- Data sources and coverage:
  - Svirydzenka (2016) database: 183 countries covering the period "1980−2015" at annual frequency for FD indices.
  - Banking crisis dummy: Laeven and Valencia (2013) database (covers up to 2011) extended for "2012 − 2016" with Candelon et al. (2018) data; banking crisis database contains "100 countries from 1980 to 2016 on a yearly basis".
  - Macroeconomic controls: output growth rate and interest rate spread (difference between the "10 − year treasury rate" and a "3 − month monetary rate"), yearly, from IFS.
  - Final sample summary stated: "our sample comprises 98 countries for the period 1980 – 2016."

### Empirical results (summary)
- Model selection and dynamics:
  - Model (2), the one including the lagged binary variable y_t−1, presents the lowest BIC in each specification and is always selected, implying that static logit models are inadequate and crisis persistence matters.
  - The finding confirms nonlinearity of causality in crisis prediction.
- Macroeconomic controls:
  - Output growth rate: an increase reduces significantly the probability of a banking crisis one year ahead.
  - Interest rate spread (long minus short): an inversion of the yield curve (negative spread) signals a higher risk of a banking crisis.
- Financial development and crisis probability (whole panel):
  - Financial development increases the probability of occurrence of a banking crisis for the whole panel.
  - The destabilizing effect operates via financial institution development and to a lesser extent via financial market development.
- Heterogeneity by country group (three clusters by development level):
  - Advanced economies: depth (FID) and access (FIA) cause banking crises.
  - Emerging markets (EM) and least income developing countries (LIDC): only financial institution depth (FID) is a leading indicator for future crises.
  - For EM and LIDC, access to financial services enhances financial stability; in contrast, in developed countries access should be limited.

### Robustness and interpretation notes
- The paper performs robustness checks (Section VI referenced) and selects dynamic specification based on BIC.
- The discussion links findings to theoretical mechanisms from the literature: excess credit quantity, prolonged stability encouraging excess borrowing (Minsky), competition and lower lending standards after liberalization (Keeley; Dell’Ariccia and Marquez), and other channels.

### Policy implications and recommendations
- Regulatory measures to stabilize the system are supported:
  - Implement capital requirements and access control to loans and deposits for financial institutions.
- Regulation should be calibrated to country development level:
  - Advanced countries: impose strict access control for financial intermediaries (restricting access that may increase nonperforming loans).
  - Low- and middle-income countries: enhance access to financial services to promote financial inclusion and reduce inequalities; support fintech innovations (mobile application payments, etc.) to expand access.
- Key message: financial development can be destabilizing and regulation should not be uniform across countries; it must account for the country’s degree of development and the differing roles of depth, access, and market vs. institution channels.

*Italic source: wpiea2019094 - 2016. Following (IMF PDF content).*

### Appendix 2 provides a list of the countries as well as the date of banking crises. It also reports the country group

### wpiea2019094 - Appendix 2 provides a list of the countries as well as the date of banking crises. It also reports the country group

### Findings on financial development and banking crises
- Panel: 98 countries over the period 1980–2016.
- Aggregate result:
  - Financial development appears to increase financial instability, increasing the probability of occurrence of a crisis in a one-year period (see model (1)).
  - The destabilizing effect operates via either financial institutions (model (2)) or financial market development (model (3)).
- Dimensions of financial development:
  - Financial institution depth is destabilizing.
  - Financial institution access reduces the future occurrence of a crisis.
  - Financial institution efficiency typically has a negative sign, but a banking crisis is not significantly affected (coefficients often not statistically different from 0).
  - Financial market development shows a positive sign for depth and a negative sign for access; however, almost all coefficients are not statistically different from 0, implying a small impact of improvements in financial institution or market efficiency.

### Cluster analysis by country groups
- Country clusters (following Svirydzenka (2016)): LIDC (least income developed countries), EM (emerging markets), AM (advanced markets).
- Common finding across clusters:
  - Financial institution development has a destabilizing impact for all three clusters.
  - Aggregate variables FD, FI and FM are only significant for the advanced countries’ cluster, but sub-indices are significant for the three clusters.
- Differences across clusters:
  - Advanced economies (AM):
    - Depth (FID) and access (FIA) are destabilizing and increase the probability of a future occurrence of a crisis.
    - Efficiency reduces the future occurrence of a banking crisis.
    - Macro variables: term spread and output growth are important for occurrence of future banking crises.
  - Emerging markets (EM):
    - Access to financial institutions increases financial stability (stabilizing).
    - Depth/efficiency are not (aggregate) significant in increasing crisis probability; output growth matters for future crises.
  - Least income developed countries (LIDC):
    - Financial institution depth is a leading indicator for future crises.
    - Access to financial institutions increases financial stability and reduces the probability of financial crises a year ahead (linked to financial inclusion).
    - Term spread not included because of data availability; most LIDC are small open economies and interest rate takers with unmature financial capital markets.

### Robustness checks
- Nonlinear interaction framework:
  - Financial development interacts with regimes of nonperforming loans (NPL) and capital account openness (KAO, measured by the method of Chinn and Ito, 2006).
  - Interaction index z_j equals 1 when z_i,t > median(z_i) and 0 otherwise; models include exchange rate regime (Err) dummies for fixed and flexible regimes (Levy-Yeyati and Sturzenegger (2005), (2016)).
- Key outcomes:
  - Introduction of interactive terms (KAO and NPL) does not affect the relationship between financial development and the banking crisis: none of the coefficients associated with the interaction term are significantly different from 0 at 99%.
  - Fixed exchange rates: results similar to baseline.
  - Flexible exchange rate regime:
    - Access to financial institutions becomes destabilizing.
    - Financial deepening becomes stabilizing.
    - Only efficiency of financial institutions appears to decrease occurrence of a banking crisis (possible explanation: capital movement on foreign exchange markets).
- Sample and horizon checks:
  - Pre-2008 estimation: results reported in Table (6) do not show a major quantitative difference from Table (1); the impact of financial development on future banking crises is not driven by the great crisis and is stable over time.
  - Two-year horizon: model re-estimated with lag-two (Pr(y_it = 1) = F(η_i + α y_i,t−2 + x_i,t−2 β + δ π_i,t−2)); results (Table (7)) are similar to Table (1).
- Policy implication from robustness: link between financial development and probability of banking crisis is structural and holds across samples and horizons, supporting need for structural regulation policies independent of business cycle or specific temporary events.

### Conclusion and policy implications
- Paper scope and innovation:
  - Decomposes financial development into access, depth and efficiency across 98 countries.
  - Relies on a dynamic logit panel model including past crisis observations and fixed effects to address unobserved heterogeneity.
- Overall conclusion:
  - Financial development, particularly from an institutional dimension and to a lesser extent a market dimension, increases the probability of occurrence of a crisis within a one- to two-year horizon.
  - The fintech explosion (mobile payment, cryptocurrency, and offshore banking) raises regulatory and supervisory concerns; macroprudential rules should be adjusted accordingly.
- Heterogeneous effects and policy recommendations:
  - Impact of financial development on stability varies with component (access, depth, efficiency) and country group.
  - Financial stability assessments (e.g., FSAP) should include a shock associated with financial development and its components to better assess vulnerabilities amid financial innovation.
  - Financial regulation (including Basel agreements) should account for specificities of emerging markets versus advanced countries:
    - Advanced economies: impose higher capital requirements on banks to smooth increases in financial access and depth.
    - Emerging markets: higher capital requirements need not be imposed in the same way; regulators should encourage higher efficiencies in financial institutions.
  - Structural regulation policies should be independent of the business cycle or specific temporary events.

*Source: Appendix 2 content and related sections from the provided IMF working paper excerpt.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019094.pdf_