## _wp09198

## Source details

**Canonical URL:** [_wp09198](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09198.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09198.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09198.pdf.json)

---

### Overview
- Objective: Model the feedback loop from financial system stress to the real economy and empirically estimate its magnitude.
- Scope and sample:
  - Countries included: France, Germany, Italy, Spain, Sweden, Switzerland, and the United Kingdom.
  - Number of banks analyzed: 26 (largest banks in each country).
  - Time period: 1991–2007 (quarterly data).
- Key risk indicators used:
  - Merton-type Distance-to-Default (DD) measures and Moody’s KMV Expected Default Frequencies (EDF): one-year EDF (EDF1) and five-year EDF (EDF5).
  - Economy-wide/system indicators: Asset-weighted DD (DD-aw), Simple average DD (DD-av), DD-index (DataStream banking sector index), DD-system (hypothetical “superbank”), Average weighted EDF1, Average weighted EDF5.
- Primary outcomes modeled:
  - Real private sector credit growth (nominal variables deflated by the GDP deflator).
  - Real GDP growth.

### Methodology
- Modeling approaches:
  - Economy-wide panel regressions linking market-based financial fragility measures to private credit growth and to GDP.
  - Bank-specific panel regressions linking individual bank DDs and EDFs to bank loan growth.
- Time series treatment and estimation:
  - Preliminary vector error-correction model to test cointegration between credit and GDP.
  - Nonstationary macro variables differenced (four quarter differences used to account for seasonality); risk variables treated as stationary.
  - Variables expressed in natural logarithms; nominal variables deflated using the GDP deflator.
  - Country-specific simultaneous-equation system capturing endogeneity between GDP and credit (notation includes ΔuCt, Rt, It, ΔuGDPt, ΔuDJt, ΔuHt, ΔuSt).
- Controls:
  - House price growth (Global Insight indices), DJ Stoxx 600, V-DAX new, iTraxx Crossover CDS, relative size of the financial sector (total financial sector assets/GDP), interbank rates and marginal lending rates.
- Bank-level specifics:
  - Dependent variable ΔCjt: growth rate of total loans for bank j.
  - Bank interest rate proxy: total net interest income divided by total loans.
  - Competition control: average-weighted EDF1, EDF5, and DD for other banks in the same country.
- Preferred macro specification:
  - Controls for GDP growth and house prices; exclude DJ Stoxx 600.
  - Include relative size of the financial sector in EDF regressions where significant; exclude in DD regressions.
  - Prefer interbank interest rate (IBR) over long-term rate.

### Box 1 — Distance-to-Default (DD)
- The DD measure follows the Black and Scholes (1973) and Merton (1974) structural valuation model; equity is treated as a call option on assets with strike price equal to book value of total liabilities.
- Estimation:
  - Methodology described in Vassalou and Xing (2004) using daily equity data and annual accounting data.
  - DD formula as presented: (notation with Vt, Lt, μ, σ as in source).
  - Implied volatility σ proxied by rolling 12-month historical volatility.
- Variants: bank-specific DDs, DD-aw, DD-av, DD-index, DD-system.

### Key empirical findings
- Statistical significance:
  - Financial sector fragility indicators (DD and EDF measures) have significant short-run effects on private sector credit growth and GDP growth, mostly at the 1 percent uncertainty level.
  - Vector error-correction results show strong cointegration between credit and GDP (t-values in the 4–6 range) and rapid adjustment (t-values in the 9–11 range).
- Estimated regression coefficients (preferred specifications):
  - Regression coefficient used for credit-impact calculations (preferred specification): 0.00648.
  - Regression coefficient used for GDP-impact calculations (preferred specification): 0.00368.
- EDF results (private sector credit regressions):
  - Lags of EDF1 and EDF5 negative and significant; sample coefficients include L1edf1 around -0.00593*** (standard error 0.0014) and L1edf5 around -0.00900*** (standard error 0.0018) in selected tables.
  - Coefficient on EDF5 slightly larger in magnitude than EDF1.
- DD results (private sector credit regressions):
  - Lags of DD measures positive and significant; selected coefficients include L1dd examples: 0.00602*** (0.0021), 0.00610** (0.0025), 0.00648*** (0.0021).
  - System DD measures (L1sysdd, L1ddindex, L1awdd) also positive and significant in multiple specifications.
- Interest rates and controls:
  - Interest rate variables significant at the 5 percent level in most specifications, with negative coefficients (e.g., ibr -0.00369** (0.0016); lr -0.00705** (0.0028)).
  - Lagged GDP growth and house price growth are typically positive and often significant controls.
- Goodness-of-fit:
  - For macro regressions, R2 for EDF regressions: 0.40-0.45.
  - For macro regressions, R2 for DD regressions: 0.02–0.10.
  - Preferred macro regressions reported R2 between 0.14 and 0.21 depending on specification.

### Quantified impacts (model-implied)
- Credit-growth impacts (macro regressions):
  - Increase in financial sector risk between 6/29/2007 and 12/31/2007 implies an implied annual decrease in real credit growth (Jun.–Dec. 2007) of 0.4 percentage point on average across sample countries.
    - This 0.4 percentage point fall vs. average real credit growth of 4.4 percent (1991–2006) implies a decrease of some 10 percent.
  - Increase in financial sector risk between 6/29/2007 and 6/30/2008 (out-of-sample) implies an annual decrease in real credit growth (Jun. 2007–Jun. 2008) of 1.7 percentage point on average across banks (based on DD regression).
    - This translates into a 32 percent average model-implied reduction in credit growth across countries.
  - Country extremes (model-implied reductions in credit growth):
    - France and Sweden: around 60 percent reductions in credit growth.
    - Switzerland: 200 percent (i.e., a credit contraction equivalent to 100 percent of average annual growth, or some 1.9 percent).
- GDP-growth impacts (macro regressions):
  - Average decrease in DD between July and end-2007 would reduce real GDP growth by around 0.2 percentage points.
  - Decrease in DD between July 2007 and July 2008 would lower GDP growth by around 1.0 percentage point.
  - Switzerland: estimates predict financial stress to reduce GDP growth by some 2.2 percentage point.
- Bank-specific implied impacts (using Jul. 2007–Jul. 2008 DD decline):
  - Predicted decline in real bank credit growth: a little below 10 percentage points on average across countries.
  - Switzerland: decline in real bank credit of 21 percentage points.
  - Implied decline in GDP growth from bank-specific regressions: 1.1 percentage point on average; up to 2.5 percentage points for Switzerland.
- Comparison:
  - Bank-specific regressions imply larger effects on bank credit than macro regressions imply for private sector credit, consistent with partial replacement of bank credit by other credit during downturns.
  - Ultimate effects on real GDP growth are of similar order of magnitude across both modeling approaches.

### Bank-specific regression results (selected)
- EDF-based bank regressions (Tables 12–13):
  - edf1 and edf5 coefficients strongly negative for changes in total loans (D4tlo), e.g., edf1 -0.0459*** (0.0054); edf5 -0.0635*** (0.0065).
  - bankint (bank interest rate) typically negative and significant in many specifications.
  - L1D4gdp strongly positive and significant (e.g., 1.508*** (0.2300)).
- DD-based bank regressions (Table 14):
  - dd coefficients positive and significant in selected specifications (e.g., 0.0343*** (0.0130)).
  - bankint negative and significant in several specifications.
- Competition controls (Tables 15–17):
  - edf1 and edf5 remain negative and significant when competition controls included (e.g., edf1 -0.0399*** (0.0092); edf5 -0.0415*** (0.0100)).
  - Competition-related variables show mixed significance and smaller magnitudes in many specifications.
- Bank-specific GDP regressions (Table 18):
  - dd positively associated with D4gdp in several specifications (e.g., 0.00449*** (0.0010)).

### Caveats and limitations
- Sample limitations:
  - Estimations are calibrated primarily on noncrisis periods; the sample does not include a major crisis, so the model may understate effects during crises where loss of confidence or herd behavior amplifies impacts.
- Scope of effects:
  - Estimates capture short-term direct effects of increased financial sector risk on credit and GDP growth; multiplier and medium- to long-run effects (e.g., via lower investment) are not fully captured and may increase total impact.
- Nonlinearities and out-of-sample risks:
  - Some observed decreases in DD (from July 2007 onward) are steeper than in-sample variation, taking estimates outside validated ranges; non-linearities in natural-log estimations increase when DD estimates approach values close to zero.
  - For calculation purposes, DD below zero was set at 0.1.
- Measurement notes:
  - Off-balance-sheet credit extension is not taken into account directly; it may influence market-based risk variables via effects on market value of assets.

### Policy implications and applications
- Macroprudential supervision:
  - Incorporate feedback effects from financial risk to credit and output to internalize the social cost of systemic risk and to mitigate pro-cyclical tendencies of asset prices and risk.
  - Monitor co-movements in fragility across institutions even when individual institutions are far from default barriers.
  - Consider countercyclical capital ratios (shoring up capital in good times, relaxing in bad times) to reduce the social costs of financial instability.
- Stress testing:
  - Replace the usual assumption of unchanged financial-institution behavior with an explicit feedback effect from financial strain to credit and output to add realism to macroeconomic stress tests.
  - Use model estimates to translate financial sector losses into output losses for cost–benefit analyses of fiscal/financial rescue packages.
- Cross-border systemic risk:
  - Model supports examination of spillovers to other countries and richer cost–benefit analysis for cross-border systemic risk management.

### Data and appendices (selected)
- Data sources: DataStream, Moody’s KMV, IMF’s International Financial Statistics (IFS), Global Insight, Bloomberg, bank-specific sources, and IMF staff calculations.
- Appendix I — Data highlights:
  - Average credit-growth (by country): France 4.6, Germany 5.0, Italy 4.7, Spain 6.3, Sweden 4.4, Switzerland 5.5, U.K. 5.1.
  - Standard deviation (credit-growth): France 1.5, Germany 2.2, Italy 3.3, Spain 2.1, Sweden 2.3, Switzerland 3.5, U.K. 1.8.
  - Sample availability for credit-growth: 1991-2007 for all seven countries.
- Appendix B — Cointegration and VECM:
  - Strong support for cointegrating relationship between credit and GDP; risk measures show significant short- to medium-term effects with expected signs (positive for DD; negative for EDF).
  - Selected VECM adjustment coefficient examples: 0.0122 (t = 10.2), 0.0119 (t = 10.1), 0.0117 (t = 9.5).
  - Selected short-term DD/EDF coefficient examples with t-values reported in Table 2 (e.g., EDF5 -41908 (t = -7.4)).

_Italic: Source: Excerpt from _wp09198 (References and accompanying sections), IMF Staff paper (1991–2007 sample description, methodological overview, and Box 1 on Distance-to-Default)._

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

### _wp09198 - References..............................................................................................................

### Overview
- Objective: Model the feedback loop from financial system stress to the real economy and empirically estimate its magnitude.
- Scope: Sample of mature Western European economies; both economy-wide and bank-specific market-based risk indicators are used to predict short-run feedback effects on private sector credit and GDP.
- Sample details:
  - Countries included: France, Germany, Italy, Spain, Sweden, Switzerland, and the United Kingdom.
  - Number of banks analyzed: 26 (largest banks in each country).
  - Time period: 1991–2007 (quarterly data).

### Research motivation and contribution
- Motivation:
  - The recent financial crisis shows large and potentially prolonged real effects of financial sector turmoil, including international contagion and adverse feedback to real economies.
  - Existing financial stability work largely focuses on identifying risks and monitoring institutions; the subsequent feedback from financial risk to the macroeconomy through the credit channel is often unmodeled.
- Contribution:
  - Explicitly incorporates market-based financial-sector fragility variables (Merton-type distance-to-default and Moody’s KMV EDFs) into panel regressions to estimate their short-run effects on private credit growth and GDP.
  - Provides in-sample and out-of-sample predictions of declines in credit growth driven by observed increases in financial-system fragility.
  - Offers both economy-wide (macro) and bank-specific estimates, enabling assessment of system-wide and individual-bank feedback effects.
  - Positions the model as a tool for richer cross-border systemic risk cost-benefit analysis and for more realistic macroeconomic stress tests that allow financial institutions’ behavior to respond to strain.

### Data and variables
- Macroeconomic variables:
  - Real GDP and real private sector credit (nominal variables deflated by the GDP deflator).
  - Economy-wide interest rates: interbank rates and marginal rates of new lending.
- Bank-level variables:
  - Total loans from banks’ balance sheets as the bank-equivalent of private sector credit.
  - Total net interest income divided by total loans as a proxy for the average interest rate charged by each bank.
  - Note: Off-balance-sheet credit extension is not taken into account directly; it may influence market-based risk variables via effects on market value of assets.
- Risk indicators:
  - Merton-type Distance-to-Default (DD) measures and Moody’s KMV Expected Default Frequencies (EDF), using both one-year EDF and five-year EDF.
  - Economy-wide/system indicators constructed by averaging or aggregating bank-level DDs and EDFs:
    - Asset-weighted DD (DD-aw)
    - Simple average DD (DD-av)
    - System-wide DD based on DataStream banking sector index (DD-index)
    - Portfolio/system DD constructed from a hypothetical “superbank” (DD-system)
    - Average weighted DD, Average weighted EDF1, Average weighted EDF5
- Observed pattern: 2004 to mid-2007 characterized by low risk (high DD indicators and low EDFs).

### Box 1 — The Distance-to-Default Measure
- The DD measure used is grounded in the Black and Scholes (1973) and Merton (1974) structural valuation model (BSM), treating equity as a call option on assets with strike price equal to book value of total liabilities.
- Estimation details:
  - DD measures estimated with the methodology described in Vassalou and Xing (2004) using daily equity data and annual accounting data.
  - The formula for DD as shown in the source:
    
    
    
    
    tt
    t
    LV
    DD
    /ln
    , 
    where Vt and Lt are respectively the (market-based) value of assets and (accounting) value of liabilities, and μ and σ are the mean and variance of the company’s stock price respectively.
  - Proxy for implied volatility σ: rolling 12-month historical volatility.
- Variants constructed: bank-specific DDs, DD-aw, DD-av, DD-index, and DD-system (superbank aggregation).

### Methodology and models estimated
- Two modeling approaches:
  - Economy-wide (macro) panel regressions linking market-based financial fragility measures to real private sector credit growth and to GDP.
  - Bank-specific panel regressions linking individual banks’ DDs and EDFs to their loan growth (proxy for credit supply).
- Regression frameworks include:
  - Vector Error Correction Model regressions (Appendix Table 2).
  - Macroeconomic panel regressions for private sector credit using EDF1, EDF5, DD, system DD, DD Index, Average Weighted DD, Average Weighted EDF1, Average Weighted EDF5 (Appendix Tables 3–10).
  - Macroeconomic panel regressions for GDP using DD (Appendix Table 11).
  - Bank-specific panel regressions using EDF1, EDF5, DD and with competition controls (Appendix Tables 12–17).
  - Bank-specific panel regressions for GDP using DD (Appendix Table 18).

### Key analytical findings (as described)
- Both macro (economy-wide) and bank-specific estimates yield significant feedback effects of financial-sector fragility on credit growth and output, described as substantial in magnitude.
- The model finds meaningful short-run feedback from market-based risk indicators to the real economy through the credit channel.
- Comparative literature:
  - Cihak and Koeva Brooks (2008) found real effects from banking losses and stock price developments in the Euro area in the second half of 2007 in the order of magnitude of 0.2–0.3 percentage point of GDP.
  - Prior literature often focuses on stock market variables and the yield curve rather than explicit bank fragility measures.

### Applications and policy implications
- Macroprudential supervision:
  - Better internalize the cost of systemic risk arising from financial strains by accounting for the feedback from financial risk to credit and output.
  - Use model estimates to mitigate pro-cyclical tendencies of asset prices and risk and to account for spillovers.
- Stress testing:
  - Replace the usual assumption of unchanged financial-institution behavior with an explicit feedback effect from financial strain to credit and output, adding realism to macroeconomic stress tests.
- Cross-border systemic risk:
  - Model supports examination of spillovers to other countries and richer cost-benefit analysis for cross-border systemic risk management.

*Source: Excerpt from _wp09198 (References and accompanying sections), IMF Staff paper (1991–2007 sample description, methodological overview, and Box 1 on Distance-to-Default).*

### 6.3 percent per year) compared to the 1990s (2.8 percent), even when abstracting from the

### III. METHODOLOGY

### Data and controls
- Real private sector credit growth series and GDP growth are used; Figure 3 shows credit growth over 1991–2007, with credit growth on average considerably above GDP growth.
- House price indices from Global Insight are used to control for changes in house prices.
- Stock market and volatility controls: Dow Jones Stoxx 600 index, V-DAX new index (30-day implied volatility in the DAX), and the iTtraxx Crossover CDS index of European sub-investment grade names.
- A control variable for the relative size of the financial sector is constructed as total financial sector assets divided by GDP.
- Data frequencies and transformations:
  - Macroeconomic data: quarterly frequency.
  - Financial markets data: daily; collapsed to quarterly by taking the last observation of the quarter.
  - Accounting variables: annual; assumed to remain constant over the year when converted to quarterly frequency.
- Data sources: IMF’s International Financial Statistics (IFS), DataStream, and Bloomberg.
- A complete overview of the data, including summary statistics and sources, is reported in Appendix Table 1 (as referenced).

### Model specification and stationarity treatment
- Focus: the credit channel — credit as the main variable through which financial risk affects the broader economy.
- Preliminary step: estimate a vector error-correction model to test for a cointegrating relationship between credit and GDP (Appendix B and Table B.2).
- Main regression framework: two equations
  - Credit as a function of financial sector risk variables (and controls, including lagged GDP growth).
  - GDP growth as a function of (lagged) credit.
- Nonstationarity: stationarity is rejected for credit, GDP, the DJStoxx 600, housing prices, and the relative size of the financial sector; stationarity is not rejected for the risk variables and interest rates.
- Treatment of nonstationarity:
  - Non-stationary variables are differenced; four quarter differences are used to account for seasonality.
  - Lags of the differenced series are used to prevent simultaneity problems.
  - All variables are expressed in natural logarithms and nominal variables deflated using the GDP deflator.

### System of simultaneous equations (country-specific)
- The country-specific simultaneous-equation system (equation (1)) captures the impact of financial risk on bank lending while accounting for endogeneity between GDP and credit.
- Notation (as defined in the source):
  - ΔuCt = the growth rate of credit to the private sector at time t over the last u quarters (u omitted for u = 1).
  - Rt = a financial risk measure at time t.
  - It = interest rates at time t.
  - ΔuGDPt = the growth rate of real GDP at time t over the last u quarters.
  - ΔuDJt = the growth rate of the DJStoxx 600 index at time t over the last u quarters.
  - ΔuHt = the growth of the Global Insights housing price index at time t over the last u quarters.
  - ΔuSt = the growth of the relative size of the financial sector (total financial sector assets/GDP) over the last u quarters.
- Panel: seven countries in the sample.
- Various control variables considered (implied volatility indices and national stock indices); many were dropped due to insignificance or strong collinearity with primary risk measures (e.g., implied volatility is a major driver of DD).

### Individual bank model
- Panel of 26 large banks from the 7 countries.
- Same basic model specification as the macro model, indexed by bank j = 1,...,26.
- Dependent variable ΔCjt: growth rate of total loans on bank j’s balance sheet (proxy for private sector credit at bank level).
- Interest rate Ijt: average interest rate bank j charges on its loans (total interest income divided by total loans).
- Competition variable (in some regressions): captures how a bank reacts to changes in the risk profile of domestic peers; constructed as the average-weighted EDF1, EDF5, and DD for other banks in the same country.
- Bank-level regressions use bank-specific risk measures (DD and EDF) only (no weighted averages or system-wide indicators).

### Macroeconomic regression results — main findings
- Financial sector fragility has a significant effect on credit and GDP growth across specifications, mostly at the 1 percent uncertainty level (see Appendix Tables 3–11 as referenced).
- EDF findings:
  - Relation between EDFs (one-year EDF and five-year EDF) and credit growth is negative and significant at the 1 percent level.
  - Results hold for asset-weighted EDF and simple average EDF.
  - Coefficient on the five-year EDF is slightly higher than the one-year EDF.
- DD findings:
  - DD measures (simple average DD, asset-weighted DD, DD-index, DD-system) show significance at the 1 percent level.
  - Lower DD (banks closer to default barrier) associated with lower credit growth.
- These results are estimated for the pre-crisis period up to 2007; authors note they would likely be stronger including ongoing crisis data.
- Interest rates:
  - Interest rate variable significant at the 5 percent level in most specifications, with a negative coefficient (both interbank rate and IFS country-wide rate on new loans).
  - Distinguishes price effects from volume effects; the identified volume effect of financial fragility on credit remains after accounting for price effects.
- Control variables:
  - GDP growth in the credit equation is generally highly significant; significance varies when EDF is the risk variable.
  - Using DD as risk variable, GDP growth is always a significant control, mostly at the 1 percent level.
  - House prices generally significant at the 1 percent level; inclusion can reduce significance of GDP growth in some specifications.
  - Relative size of the financial sector (financial assets/GDP) significant at the 1 or 5 percent level when EDFs are used; generally not significant in DD regressions (not significant even at the 10 percent level).
  - DJ Stoxx 600, VDAX-New, and iTraxx crossover CDS index generally yield insignificant coefficients.
- Signs of coefficients as expected:
  - Credit growth positively associated with GDP growth.
  - House price growth positively associated with credit growth (mortgage lending channel).
  - Financial sector expansion relative to GDP positively associated with credit growth (partly tautological since loans are a component of financial sector assets).
- Robustness:
  - Results robust to changes in specification and lag structures; coefficient magnitudes stable and generally remain significant.
  - Alternative dependent variable: using domestic credit instead of private sector credit yields similar sign and magnitude but many coefficients lose significance; private sector credit preferred as more relevant.

### Preferred specifications and fit
- Preferred macro regression:
  - Include controls for GDP and house prices, exclude the DJ Stoxx 600.
  - Include relative size of the financial sector in EDF regressions where it is significant; exclude it in DD regressions where it is not significant.
  - Prefer interbank interest rate over long-term rate (interbank rate more responsive and more relevant for most loans).
- Fit of preferred regressions: R2 between 0.14 and [text ends here].

*Source: Excerpt from the provided PDF content (methodology and results sections).*

### 0.21 depending on the exact specification. For models in first difference, this is well within

### _wp09198 - 0.21 depending on the exact specification. For models in first difference, this is well within

### Key empirical findings
- Regression coefficient used for credit-impact calculations (preferred specification): 0.00648.
- Regression coefficient used for GDP-impact calculations (preferred specification): 0.00368.
- For macro regressions, R2 for EDF regressions: 0.40-0.45.
- For macro regressions, R2 for DD regressions: 0.02–0.10.

### Quantified impacts on credit growth (macro regressions)
- Increase in financial sector risk between 6/29/2007 and 12/31/2007 implies:
  - Implied annual decrease in real credit growth (Jun.-Dec. 2007): 0.4 percentage point on average across sample countries.
  - This 0.4 percentage point fall vs. average real credit growth of 4.4 percent (1991–2006) implies a decrease of some 10 percent.
- Increase in financial sector risk between 6/29/2007 and 6/30/2008 (out-of-sample):
  - Implied annual decrease in real credit growth (Jun. 2007–Jun. 2008): 1.7 percentage point on average across banks (based on DD regression).
  - This translates into a 32 percent average model-implied reduction in credit growth across countries.
- Country and bank extremes from Table 1 (model-implied reductions in credit growth):
  - France and Sweden: around 60 percent reductions in credit growth.
  - Switzerland: 200 percent (i.e., a credit contraction equivalent to 100 percent of average annual growth, or some 1.9 percent).

### Quantified impacts on GDP growth (macro regressions)
- Average decrease in DD between July and end-2007 would reduce real GDP growth by around 0.2 percentage points.
- Decrease in DD between July 2007 and July 2008 would lower GDP growth by around 1.0 percentage point.
- For Switzerland, estimates predict financial stress to reduce GDP growth by some 2.2 percentage point (enough to bring it into negative territory by most estimates).

### Bank-specific regression results and implications
- Bank-specific panel regressions yield coefficient estimates in the same range as macro regressions; some control variables differ.
- Preferred EDF1/EDF5 specification controls: DJ Stoxx index, financial sector size, and GDP (house price and competition variables excluded).
- Preferred DD specification controls: house price index and GDP only.
- R2 for bank-specific regressions:
  - EDF regressions: generally in the range of 0.40-0.45.
  - DD regressions: in the order of 0.02–0.10.
- Implied effects from bank-specific regressions (using decline in DD Jul. 2007–Jul. 2008):
  - Predicted decline in real bank credit growth: a little below 10 percentage points on average across countries.
  - Switzerland: decline in real bank credit of 21 percentage points.
  - Implied decline in GDP growth: 1.1 percentage point on average; up to 2.5 percentage points (for Switzerland).

### Comparison and interpretation
- Effects on bank credit from bank-specific regressions are larger than effects on private sector credit from macro regressions, suggesting partial replacement of bank credit by other credit during some downturns.
- Ultimate effects on real GDP growth are of similar order of magnitude across bank-specific and macroeconomic models.

### Caveats and limitations
- Estimates capture short-term direct effects of increased financial sector risk on credit and GDP growth; multiplier and medium- to long-run effects (e.g., via lower investment) are not fully captured and may increase total impact.
- Estimations are calibrated primarily on noncrisis periods; the sample does not contain a major crisis, so model may understate effects during crises where loss of confidence or herd behavior amplifies impacts.
- Some observed decreases in DD (from July 2007 onward) are steeper than in-sample variation, taking estimates outside validated ranges; non-linearities in natural-log estimations increase when DD estimates approach values close to zero.
- For calculation purposes, DD below zero was set at 0.1.

### Policy implications and uses
- Quantifying real economic impact of financial sector fragility:
  - Demonstrates that real economic variables depend on financial sector risk and that risk variables play an important role in the credit channel.
  - Can inform policymakers about costs and benefits of financial sector policies and provide benchmarks for fiscal policy decisions aimed at stabilizing the financial sector.
- Macroprudential and supervisory implications:
  - Prudential supervision could be enhanced by incorporating effects of financial instability on the real economy, focusing on co-movements in fragility across institutions even when institutions are far from default barriers.
  - Countercyclical capital ratios (shoring up capital in good times, relaxing in bad times) might be preferable from a social welfare perspective compared to waiting until firms approach minimum regulatory capital ratios.
- Stress-testing applications:
  - The feedback effect from financial sector stress to credit contraction can be quantified and incorporated into macro stress scenarios to improve realism and accuracy.
  - Results provide a metric to translate financial sector losses into output losses, aiding cost–benefit analyses of fiscal/financial rescue packages and informing macroprudential stress tests.

*Source: Datastream, author's calculations.*

### APPENDIX I. DATA AND TABLES

### APPENDIX I. DATA AND TABLES

### A. Data
- Table 1. Data Summary — country-level arithmetic summaries (Averages and Standard Deviations) and sample availability:
  - Average credit-growth (by country): France 4.6, Germany 5.0, Italy 4.7, Spain 6.3, Sweden 4.4, Switzerland 5.5, U.K. 5.1
  - Standard Deviation (credit-growth): France 1.5, Germany 2.2, Italy 3.3, Spain 2.1, Sweden 2.3, Switzerland 3.5, U.K. 1.8
  - Availability (credit-growth): 1991-2007 for all seven countries
  - Average (risk measure 1): 0.10, 0.22, 0.14, 0.15, 0.19, 0.15, 0.15
  - Standard Deviation (risk measure 1): 0.08, 0.22, 0.09, 0.14, 0.20, 0.13, 0.09
  - Availability (risk measure 1): 1992-2007 for most, Switzerland 1998-2007; overall 1992-2007
  - Average (risk measure 2): 0.19, 0.33, 0.23, 0.21, 0.27, 0.25, 0.22
  - Standard Deviation (risk measure 2): 0.10, 0.23, 0.12, 0.16, 0.21, 0.14, 0.10
  - Availability (risk measure 2): 1992-2007 for most, Switzerland 1998-2007; overall 1992-2007
  - Average (Total Loans / Dom Cred / PS Cred / EDF5 / GDP / ITraXX / Crossover / V-Dax New / DJ Stoxx 600 / House Prices) — selected numeric entries shown in table:
    - Example averages: 1214, 210, 207, 53, 3287, 866, 2985, 4146, 5652, 1206 (as shown in table)
    - Standard deviations for these series: 2699, 804, 574, 760, 3394, 3134, 0103, 5865, 534 (as shown in table)
    - Availability for many financial series: 1990-2007 or 1991-2007 as specified in the table
  - Average (additional country-level statistic): 119.5, 99.4, 100.8, 1018.3, 264.5, 360.2, 322.6
  - Standard Deviation (same): 40.5, 5.3, 12.6, 461.9, 87.1, 54.4, 145.9
  - Availability for these series: ranges including 1990-2007 and 1995-2007 as shown
- Data sources listed in the table: DataStream, Moody's KMV, IFS, Global Insight, bank-specific, and IMF staff calculations (as provided in table cells).
- Note in table footer: Source: DataStream, Moody'sKMV, IFS, Global Insight, and IMF staff calculations.

### B. Cointegrating Relationship
- Model specification:
  - Vector error-correction model explaining change in credit growth (ΔCred) with long-run equilibrium between non-stationary variables credit (“Cred”) and “GDP”, speed of adjustment coefficient α, and short-term variables (Risk, Interest, Controls) influencing deviations:
    - Functional form indicated: ΔCred_t = α (Cred_{t-1} − β GDP_{t-1} ) + γ Risk + γ Interest + γ Controls + ε_t (equation labeled (0.1) in source)
- Main empirical findings:
  - Strong support for existence of a cointegrating relationship between credit and GDP.
  - Long-term equilibrium relationship between credit and GDP: t-values in the 4–6 range.
  - Adjustment coefficient (speed of adjustment toward equilibrium): t-values in the 9–11 range.
  - These results hold for regressions with both private sector credit and total domestic credit as dependent variables.
- Financial risk indicators:
  - Risk measures (DD or EDF) show significant short- to medium-term effects.
  - For both dependent variables, t-values for risk variables range from 4 to 7.5.
  - Estimated coefficients have expected signs:
    - Positive for DD-based measures.
    - Negative for EDF-based measures.
- Price variable and control variables:
  - Interbank interest rates: coefficient estimates often not statistically different from 0 but generally with expected negative sign.
  - Lagged GDP growth: expected positive sign; coefficient significantly different from 0 at the 5 percent level for the domestic credit regressions.
  - Adding other control variables did not materially change results.
- Country and dummy variable findings:
  - Adding dummies illustrated differences between countries; level dummies and interaction dummies with risk variables showed significance in some cases.
  - Adding 6 country dummies for 7 countries often led to overidentified estimation and unreliable results.
  - A dummy for Euro area membership suggested risk variables play a stronger role in the Euro area; this result may be unduly influenced by Sweden’s data (which include direct aftermath of Swedish banking crisis).

- Vector Error Correction Model regression summary (selected reported estimates and t-values from Table 2):
  - Cointegrating relationship (GDP(-1) coefficients across regressions): examples include -1.917 (t = -4.8), -1.661 (t = -4.1), -2.480 (t = -5.8), -2.377 (t = -5.3), -2.497 (t = -5.2), etc. (coefficients and bracketed t-values as listed in Table 2).
  - Adjustment Coefficient examples: 0.0122 (t = 10.2), 0.0119 (t = 10.1), 0.0117 (t = 9.5), 0.0112 (t = 9.1), 0.0108 (t = 10.4), etc.
  - Short-term dynamics — DD-based and EDF-based coefficient examples (with t-values):
    - DD: 2006 (t = 5.4), 1873 (t = 4.9)
    - AWDD: 1978 (t = 5.4), 1895 (t = 5.1)
    - SYSDD: 1689 (t = 4.8), 1475 (t = 4.1)
    - DDINDEX: 1401 (t = 4.3), 1328 (t = 4.0)
    - EDF1: -37202 (t = -6.0), -39434 (t = -6.2)
    - AWEDF1: -35878 (t = -5.9), -38627 (t = -6.2)
    - EDF5: -41908 (t = -7.4), -43159 (t = -7.5)
    - AWEDF5: -39463 (t = -7.1), -40943 (t = -7.2)
  - Interbank rates (IBR) — examples of short-term coefficients and t-values indicate often weak statistical significance (examples reported with small absolute t-values or near zero t-statistics).
  - DGDP(t-1) examples: coefficients include 0.0358 (t = 0.7), 0.0194 (t = 0.4), 0.0388 (t = 0.7), 0.1197 (t = 2.4), 0.1229 (t = 2.3), etc.

### C. Regression Outcomes
- Macroeconomic panel regressions for Private Sector Credit (selected robust patterns across Tables 3–11):
  - EDF1 and EDF5 (probability-of-default based measures):
    - Lags of EDF measures (L1edf1, L1edf5) have statistically significant negative coefficients in multiple specifications:
      - L1edf1 examples: -0.00593*** (standard error 0.0014), -0.00572*** (0.0014), -0.00582*** (0.0014), -0.00571*** (0.0014), -0.00558*** (0.0015), etc. (Table 3)
      - L1edf5 examples: -0.00900*** (0.0018), -0.00878*** (0.0018), -0.00847*** (0.0018), -0.00896*** (0.0019), -0.00837*** (0.0019), etc. (Table 4)
    - Coefficients are significant at conventional levels (*** p<0.01, ** p<0.05) indicating higher EDFs are associated with subsequent reductions in private sector credit growth.
  - DD and DD-derived measures:
    - Lags of DD measures (L1dd, L1sysdd, L1ddindex, L1awdd) have statistically significant positive coefficients:
      - L1dd examples: 0.00602*** (0.0021), 0.00610** (0.0025), 0.00648*** (0.0021), etc. (Table 5)
      - L1sysdd examples: 0.00751*** (0.0024), 0.00809*** (0.0029), 0.00831*** (0.0024), etc. (Table 6)
      - L1ddindex examples: 0.00684*** (0.0022), 0.00729*** (0.0027), 0.00915*** (0.0023), etc. (Table 7)
      - L1awdd examples: 0.00593*** (0.0021), 0.00600** (0.0025), 0.00641*** (0.0021), etc. (Table 8)
    - Positive and significant coefficients indicate higher DD measures are associated with subsequent increases in private sector credit growth across specifications.
  - Controls and other covariates:
    - Interbank rate (ibr) and lending rate (lr) commonly show negative coefficients:
      - Examples: ibr -0.00369** (0.0016), lr -0.00705** (0.0028) in EDF1 regressions (Table 3).
      - In DD regressions ibr -0.00448*** (0.0014), lr -0.00670*** (0.0025) (Table 5).
    - Lagged GDP growth (L1D4gdp) typically positive and often highly significant:
      - Examples across tables: 0.294*** (0.0690), 0.263*** (0.0690), 0.333*** (0.0650), 0.329*** (0.0660), etc.
    - Lagged private sector credit growth (L1D4pscred) shows strong persistence:
      - Examples: 0.0598*** (0.0110), 0.0599*** (0.0110), 0.0672*** (0.0110), 0.0653*** (0.0110), etc.
    - House price growth (L1D4housepi) and tagdp (L1D4tagdp) often enter positively and significantly in many specifications:
      - L1D4housepi examples: 0.0957*** (0.0250), 0.0895*** (0.0250), 0.0809*** (0.0220), 0.0796*** (0.0220), etc.
      - L1D4tagdp examples: 0.0277*** (0.0093), 0.0282*** (0.0092), etc.
  - Goodness-of-fit and sample sizes (selected):
    - Observations vary across specifications (examples): 400, 391, 375, 349, 333, 344, 328, etc. (reported in Tables 3–10).
    - R-squared values typically in modest ranges (examples): 0.13, 0.07, 0.16, 0.06, 0.17, 0.06, 0.11, 0.04 (as reported).
- GDP regressions using DD (Table 11):
  - L4dd coefficients (effect of lagged DD on GDP growth) show some positive and significant entries:
    - Examples: 0.00297** (0.0015), 0.00372** (0.0015), 0.00369** (0.0015), etc.
  - L1D4pscred and L1D4gdp show strong positive persistence in GDP regressions:
    - L1D4pscred examples: 0.135*** (0.0150), 0.123*** (0.0140), 0.0799*** (0.0160), etc.
    - L1D4gdp examples: 1.685*** (0.1400), 1.676*** (0.1400), 1.569*** (0.1500), etc.
  - D4djstoxx and L1D4housepi often positive and significant in GDP regressions:
    - D4djstoxx examples: 0.0245*** (0.0037), 0.0283*** (0.0038)
    - L1D4housepi examples: 0.101*** (0.0140), 0.0973*** (0.0170)
- Bank-specific panel regressions (Tables 12–18) — selected patterns:
  - Using EDF1 and EDF5 (Tables 12 and 13):
    - edf1 and edf5 coefficients are negative and strongly significant for D4tlo (changes in total loans) in multiple specifications:
      - edf1 examples: -0.0459*** (0.0054), -0.0423*** (0.0054), -0.0456*** (0.0054) (Table 12).
      - edf5 examples: -0.0635*** (0.0065), -0.0590*** (0.0065), -0.0628*** (0.0066) (Table 13).
    - bankint (bank interest) negative and significant in many specifications: e.g., -0.0517*** (0.0120), -0.0555*** (0.0120).
    - L1D4gdp (lagged GDP growth) positive and significant: e.g., 1.508*** (0.2300), 1.388*** (0.2300).
  - Using DD measures (Table 14):
    - dd coefficient positive and significant in some specifications: e.g., 0.0343*** (0.0130), 0.0358*** (0.0130).
    - bankint negative and in several specifications statistically significant: e.g., -0.0348** (0.0170), -0.0375** (0.0170).
    - L1D4gdp positive and significant in many specifications (magnitudes vary by specification).
  - Adding competition controls (Tables 15–17):
    - EDF1/EDF5 regressions with competition controls: edf1 and edf5 remain negative and significant in many specifications (e.g., edf1 -0.0399*** (0.0092); edf5 -0.0415*** (0.0100)).
    - Competition-related variables (Rawedf1, Rawedf5, Rawdd) sometimes show smaller or mixed signs and statistical significance across specifications.
    - Bank lending growth (L1D4tlo) shows positive and significant coefficients where included (e.g., 0.0200*** (0.0047) in Table 15).
  - Bank-specific GDP regressions using DD (Table 18):
    - dd positively associated with D4gdp in several specifications (e.g., 0.00449*** (0.0010)).
    - L1D4gdp and other macro controls show strong persistence and significance in bank-level GDP specifications.

- Statistical notation and significance:
  - Standard errors are reported in parentheses in all tables.
  - Significance indicators used throughout: *** p<0.01, ** p<0.05, * p<0.1.
  - Source note repeated for regression tables: IMF Staff estimates.

_Italic: Source: DataStream, Moody'sKMV, IFS, Global Insight, and IMF staff calculations._

### REFERENCES

### REFERENCES

### Monetary policy, credit channel, and transmission
- Altunbas, Yener, Otabek Fazylov, and Philip Molyneux, 2002, “Evidence on the bank lending channel in Europe,” Journal of Banking & Finance, Vol. 26, Chapter 11, pp. 2093–2110.
- Altunbas, Yener, Leonardo Gambacorta, and David Marqués, 2007, “Securitization and the Bank Lending Channel,” European Central Bank Working Paper Series, No. 838 (Frankfurt: European Central Bank).
- Angeloni, Ignazio, and Michael Ehrmann, 2003, “Monetary policy transmission in the euro area: any changes after EMU?,” European Central Bank Working Paper Series, No. 240, (Frankfurt: European Central Bank).
- Bernanke, Ben, and Alan Blinder, 1988, “Money, Credit and Aggregate Demand,” American Economic Review, Vol. 78, pp. 901–21.
- Bernanke, Ben, and Mark Gertler, 1995, “Inside the Black Box: The Credit Channel of Monetary Policy Transmission,” The Journal of Economic Perspectives, Vol. 9, pp. 27–48.
- Bayoumi, Tam, and Andrew Swinston, 2007, “Foreign Entanglements: Estimating the Source and Size of Spillovers Across Industrial Countries,” IMF Working Paper 07/183 (Washington: International Monetary Fund).

### Financial crises, stability, and restructuring
- Bernanke, Ben, 1983, “Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression,” American Economic Review, Vol. 73, pp. 257–76 (June).
- Hunter, William C., G.G. Kaufman, and T.H. Krueger (eds.), 1999, “The Asian Financial Crisis: Origins, Implications and Solutions,” Boston: Kluwer.
- Lindgren, Carl, and others, 1999, “Financial Sector Crisis and Restructuring. Lessons from Asia,” IMF Occasional Paper No. 188 (Washington: International Monetary Fund).
- International Monetary Fund, 2008, Global Financial Stability Report (Washington).
- Cihak, Martin, and Petya Koeva Brooks, 2008, From Subprime Loans to Subprime Growth? Evidence from the Euro Area, IMF Staff Country Report No. 08/263, pp. 5–25 (August) (Washington: International Monetary Fund).
- De Nicoló, Gianni, Robert Corker, Alexander F. Tieman, and Jan-Willem van der Vossen, 2005, European Financial Integration, Stability and Supervision, IMF Staff Country Report No. 05/266, pp. 113–146 (August) (Washington: International Monetary Fund).

### Asset pricing, credit risk, and financial modeling
- Black, F., and M. Scholes, 1973, “The Pricing of Options and Corporate Liabilities,” Journal of Political Economy, Vol. 81, No. 3, pp. 637–54.
- Merton, R.C., 1974, “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates,” Journal of Finance, Vol. 29, No. 2, pp. 449–70.
- Vassalou, Maria, and Y. Xing, 2004, “Default Risk in Equity Returns,” Journal of Finance, Vol. LIX, Chapter 2, pp. 831–68.

### International transmission, forecasting, and empirical analysis
- Esponiza, Raphael, Fabio Fornari, and Marco Lombardi, 2008, “The Role of Financial Variables in Predicting Economic Activity and in the International Transmission of Shocks,” mimeo.
- Noland, Markus, L. Liu, S. Robinson, and Z. Wang, 1998, “Global Economic Effects of the Asian Currency Devaluations,” Policy Analyses in International Economics 56, (Washington: Institute for International Economics).
- Bagehot, W., 1873, “Lombard Street,” London: Henry S. King and Company.

*Source: REFERENCES*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09198.pdf_