## _wp05173

## Source details

**Canonical URL:** [_wp05173](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05173.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05173.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05173.pdf.json)

---

### I. Introduction — purpose and approach
- Motivation: financial crises of the late 1990s revealed large macroeconomic costs of weak financial systems (negative effects on growth, policy effectiveness, and fiscal consequences).
- Objective: develop robust measures of the quality of different aspects of financial policies and extend analysis of their relationship with financial stress to a system-wide perspective.
- Methodological features:
  - Construct an index of financial stress (measures severity of stress, not predicting crises).
  - Use longitudinal data-estimation techniques that control for time-invariant unobserved heterogeneity.
  - Sample: 68 countries or economies with completed FSAP at time of writing (Appendix Table 2).

### II. Literature overview — theoretical and empirical context
- Four generations of models summarized:
  - First-generation: macro fundamentals and speculation.
  - Second-generation: self-fulfilling expectations.
  - Third-generation: incentives/opportunities for risky lending; twin crises.
  - Fourth-generation: institutional factors (politics, rule of law, trust, property rights, legal origin, governance, quality of financial policies).
- Conceptualization of financial stress:
  - Adopts Illing and Liu (2003): stress is continuous; extreme realizations are crises.
  - Symptoms: falling asset prices; exchange rate depreciation/loss of reserves; insolvency; defaults (including sovereign); rising interest rates; increased volatility.
- Empirical evidence: more recent FSAP-based indices find a positive link between regulatory governance and financial system soundness.

### III. Measures constructed — indices and sectoral methodology
- Overall design:
  - Sector-wise indices of financial stress combined into overall indices weighted by sector size.
  - Two overall indices:
    - FSI5: banking, foreign exchange, equity markets, domestic and external government debt markets (valid observations for 22 economies).
    - FSI3: banking, foreign exchange, equity markets (used in regressions; broader coverage).
- Sector-wise measurement principles:
  - Use variables capturing first and, when possible, second moments; include volatility measures (GARCH(1,1) conditional variances for one-day nominal returns in stock and foreign exchange markets).
  - Transform series into percentiles across countries and time; sector score = average percentile across indicators.
- Banking sector stress indicators:
  - Preferred measures: real overnight interbank rate and ratio of central bank credit to total bank assets (avoid aggregate balance-sheet NPL ratios).
- Foreign exchange market stress:
  - Use volatility measure; use real effective exchange rate; use net international reserves to imports ratio; use levels rather than arbitrary period percentage changes.
- Stock market stress:
  - Separate real stock market index level and volatility of nominal stock returns.
- Government debt markets:
  - External: spread between country foreign debt return (or hedged domestic debt return) and comparable-maturity U.S. government securities.
  - Domestic: quarterly real treasury bill rate.
- Index of quality of financial policies (QFPI):
  - Based on FSAP principle-by-principle BCP and OPSR assessments.
  - Sector-wise indices for banking and securities markets; combined index weighted by sector sizes (deposit money banks total assets to GDP; stock market capitalization to GDP).
  - Within each sector, constituent principles grouped into four areas: governance; regulatory practices; prudential framework; financial integrity and safety net arrangements.
  - Equal weights across four areas; combined index = simple average of sector scores.

### IV. Regression analysis — model, data, and specification
- Data and frequency:
  - Panel data at annual frequency over 1998–2003.
  - Dependent variable: overall index of financial stress (FSI3) converted from quarterly to annual by taking maximum quarterly realization per year.
  - Number of countries: 55; number of observations: 271.
- Conceptual model:
  - FSIit = αi + Σn ξn Fnit + Σk δk Ckit + Σj βj Ij + Σ interactions of institutions with fundamentals/contagion + εit
  - Emphasis: interaction terms where institutions (QFPI) modify impact of macro fundamentals and contagion on financial stress.
- Contagion channels:
  - Direct and indirect contagion via trade and financial flows (direct contagion; indirect/common-third-country channels: competitive devaluation, common lender channel, portfolio rebalancing).
- Estimation method:
  - Arellano and Bover (1995) / Blundell and Bond (1998) GMM-SYS estimator (Roodman 2003 STATA module).
  - Manage instrument count (one instrument per variable and lag distance).
  - Include lagged dependent variable to form a partial adjustment model.

### V. Key regression results and magnitudes (preferred GMM-SYS specification for FSI3)
- Model fit and diagnostic tests:
  - Overall R2: 0.62.
  - Arellano-Bond AR(1) test in first differences: -2.92 **.
  - Arellano-Bond AR(2) test in first differences: -1.6.
  - Number of instruments: 30.
  - Sargan test of overidentifying restrictions: 28.3 (statistically insignificant at 99%).
  - Number of countries: 55; number of observations: 271.
- Estimated short-run coefficients (robust standard errors in parentheses; significance):
  - Lagged value of FSI3: 0.52 (0.11) **.
  - Constant: 19.73 (7.19) **.
  - General government balance as percent of GDP (GovBal): -2.11 (0.95) *.
  - Interaction QFPI_x_GovBal: 0.023 (0.01) *.
  - Index of terms of trade (TOT): 0.15 (0.04) **.
  - Interaction QFPI_x_TOT: -0.002 (0.0005) **.
  - Measure of contagion among constituents of the MSCI Emerging Markets Equity Index (ContMSC): -27.25 (9.91) **.
  - Interaction QFPI_x_ContMSC: 0.37 (0.13) **.
  - Time trend (TIME): -4.39 (1.10) **.
  - Interaction QFPI_x_TIME: 0.059 (0.01) **.
  - ** denotes statistically significant at 99% level; * denotes statistically significant at 95% level.
- Static long-run solution (reported expression):
  - FSI3 = 40.89 – 4.36 GovBal + 0.05 QFPI_x_GovBal + 0.32 TOT – 0.003 QFPI_x_TOT – 56.47 ContMSC + 0.77 QFPI_x_ContMSC – 9.11 TIME + 0.12 QFPI_x_TIME
- Interpretation of interactions:
  - Higher QFPI reduces impact of government deficits (GovBal) and terms-of-trade improvements (TOT) on FSI3.
  - ContMSC coefficient (-27.25 **) implies that, on average in 1998–2003, higher realized stress in one MSCI Emerging Markets constituent tended to decrease financial stress in peer countries (consistent with capital reallocation away from the troubled country into other emerging markets).
  - Interaction QFPI_x_ContMSC (0.37 **) implies the contagion effect becomes less negative and can reverse for higher QFPI values.
- Quantified scenario effects (changes reported in basis points of FSI3; Table 2):
  - GovBal falls below its sample mean by one percentage point (TOT, ContMSC, TIME at sample means):
    - QFPI at sample mean (67.66): Short run 0.52; Long run 1.08.
    - QFPI at 25-th percentile (56.49): Short run 0.78; Long run 1.62.
    - QFPI at 75-th percentile (80.89): Short run 0.21; Long run 0.43.
  - TOT rises above its sample mean by ten points (GovBal, ContMSC, TIME at sample means):
    - QFPI mean: Short run 0.42; Long run 0.88.
    - QFPI 25th: Short run 0.61; Long run 1.26.
    - QFPI 75th: Short run 0.21; Long run 0.42.
  - ContMSC doubles from its sample mean (GovBal, TOT, TIME at sample means):
    - QFPI mean: Short run -0.56; Long run -1.16.
    - QFPI 25th: Short run -1.62; Long run -3.36.
    - QFPI 75th: Short run 0.69; Long run 1.43.
  - TIME increments by one period from its sample mean (GovBal, TOT, ContMSC at sample means):
    - QFPI mean: Short run -0.40; Long run -0.82.
    - QFPI 25th: Short run -1.06; Long run -2.19.
    - QFPI 75th: Short run 0.38; Long run 0.80.
- Descriptive statistics (selected from Table 3):
  - FSI3 (basis points): Valid observations 271; Mean 52.12; Standard deviation 16.99; Minimum 5.73; Maximum 92.65.
  - GovBal (%): Valid observations 271; Mean -2.30; Standard deviation 4.51; Minimum -24.61; Maximum 13.73.
  - TOT (basis points): Valid observations 271; Mean 112.43; Standard deviation 58.01; Minimum 53.11; Maximum 532.73.
  - ContMSC (basis points): Valid observations 271; Mean 0.26; Standard deviation 0.91; Minimum 0; Maximum 6.14.
  - TIME: Valid observations 271; Mean 4.01; Standard deviation 1.42; Minimum 2; Maximum 6.
  - QFPI (basis points): Valid observations 271; Mean 67.66; Standard deviation 18.44; Minimum 24.73; Maximum 100.

### VI. Key substantive findings and implications
- Institutions-moderated effects:
  - Higher QFPI mitigates effects of government deficits and terms-of-trade shocks on overall financial system stress.
  - Short-run and long-run effects of macro changes on FSI3 are larger in absolute terms at lower QFPI percentiles (25th) than at higher percentiles (75th); higher QFPI attenuates adverse effects.
- Contagion and capital flows among emerging markets:
  - No support for portfolio rebalancing contagion channel among emerging markets in 1998–2003; higher stress in one MSCI constituent tended to coincide with lower stress in peers (consistent with capital reallocation across emerging markets).
  - Negative effect of contagion-driven influx of foreign funds on recipient emerging markets’ stress is more pronounced for low QFPI; for QFPI above the 55-th percentile, high stress in one country leads to increases in peer-country stress.
- Time trend:
  - Autonomous time effects: slight downward trend of average financial stress among countries below the 55-th percentile of QFPI; slight upward trend among countries above the 55-th percentile.
  - Short-run effects of time on FSI3 are small (less than two basis points per year).
- Fit and robustness:
  - Preferred specification yields R2 = 0.62 and passes tests for second-order serial correlation and overidentifying restrictions.

### VII. Conclusions and data implications
- Constructed multi-country indices:
  - FSI3: banking sector, foreign exchange, equity markets using first- and second-moment measures and percentile transformation.
  - QFPI: based on FSAP principle-by-principle assessments for banking and securities regulation, aggregated across four policy areas.
- Main conclusion:
  - Financial system stress depends on interplay between macro fundamentals, cross-border contagion, and institutional quality; higher-quality financial policies mitigate adverse impact of deficits and terms-of-trade shocks on financial stress.
- Data and future research:
  - Wider availability of data over time could permit stress indices that incorporate additional financial sectors and enable more robust regression analysis using repeated measurements of quality of financial policies.

### Appendix highlights — sector indices, contagion, and macro fundamentals
- Sample frequency and span for sector-wise indices:
  - Quarterly frequency; period 1993–2003; typically 44 quarterly observations per country.
- Missing-value treatment:
  - Do not extrapolate for missing at beginning/end or pervasive gaps.
  - Otherwise: replace by arithmetic average for non-trending series; extrapolate by constant quarterly growth (geometric mean) for trending series.
- Weighting schemes tested:
  - Variance-equal weights method (Illing and Liu, 2003) — limitations noted.
  - Percentiles transformation method (chosen): transform values into 0–99 percentiles across countries and quarters; final sector index = average of percentiles.
- Coverage counts:
  - Banking sector: 68; Stock market: 52; Foreign exchange market: 70; External government debt market: 33; Domestic government debt market: 41.
  - Overall indices based on five sectors: 22; Overall indices based on banking, stock, foreign exchange (FSI3): 65.
- Contagion measures: constructed as (sector-wise stress in country Y) × (importance of trade/financial links of Y for X) × (relative importance of the financial system segment in X); measures for FX (trade partners), banking (BIS consolidated claims), stock (MSCI Emerging Markets membership), external debt (J.P. Morgan EMBI membership).
- Macroeconomic fundamentals (sources and definitions):
  - Log real PPP GDP per capita (1995 international U.S. dollars); inflation (annualized quarterly CPI growth in percent); index of terms of trade (1995=100); general government balance as percent of GDP; gross fixed capital formation as percent of GDP; broad money to international reserves net of gold; trade openness; number of capital account restrictions as percent of applicable restrictions; exchange rate regime dummies; dummy for explicit deposit insurance in 1999.

*Source: _wp05173 — Excerpt from IMF working paper (3. Groupings of Principles of Main Regulatory IFS Standards, pages 43–24 of provided PDF content).*

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

### _wp05173 - References..............................................................................................................

### Tables
- 1. Determinants of Financial System Stress ............................................................................19
- 2. Effects of Changes in Explanatory Variables on Financial System Stress ..........................21
- 3. Descriptive Statistics of Regressors.....................................................................................21

### Figures
- 1. Intertemporal Dynamics of the Overall Indices of Financial Stress ....................................14
- 2. Histograms of Variables in Regression Model of Financial System Stress.........................22

### Appendices
- I.    Sector-Wise Indices of Financial Stress ............................................................................32
- II.   Sector-Wise Indices of Quality of Financial Policies ........................................................40
- III. Measures of Contagion ......................................................................................................44
- IV. Macroeconomic Fundamentals ..........................................................................................47

### Appendix Tables
- 1. Country Coverage of Financial Stress Indices.....................................................................38
- 2. Regulatory IFS Standards Assessed in Countries with Completed FSAPs .........................41

*Source: _wp05173 - References..............................................................................................................*

### 3. Groupings of Principles of Main Regulatory IFS Standards ...............................................43

### 3. Groupings of Principles of Main Regulatory IFS Standards

### I. Introduction — purpose and approach
- Financial crises of the late 1990s highlighted large macroeconomic costs of weak financial systems, including negative effects on growth, policy effectiveness, and fiscal consequences.
- Paper objective: develop more robust measures of the quality of different aspects of financial policies and extend analysis of their relationship with financial stress beyond banking and foreign exchange to a system-wide perspective.
- Key methodological features:
  - Construct an index of financial stress (aimed at measuring severity of stress, not predicting crises).
  - Use longitudinal data-estimation techniques that control for time-invariant unobserved heterogeneity.
  - Sample: 68 countries or economies with completed FSAP at the time of writing (Appendix Table 2).

### II. Literature overview — theoretical and empirical context
- Theoretical evolution summarized in four generations of models:
  - First-generation: macro fundamentals and speculation.
  - Second-generation: self-fulfilling expectations.
  - Third-generation: incentives/opportunities for risky lending; twin crises.
  - Fourth-generation: institutional factors (politics, rule of law, trust, property rights, legal origin, governance, quality of financial policies).
- Financial stress conceptualization:
  - Adopts Illing and Liu (2003): stress is continuous; extreme realizations are crises.
  - Symptoms of disruption include: falling asset prices; exchange rate depreciation/loss of reserves; insolvency; defaults (including sovereign); rising interest rates; increased volatility.
- Empirical evidence on institutions:
  - Early cross-country studies found limited direct effects of supervision quality on aggregate bank soundness.
  - More recent studies (Das, Quintyn, and Chenard, 2004; Podpiera, 2004) using larger samples and FSAP-based indices find a positive link between regulatory governance and financial system soundness.

### III. Measures constructed — indices and sectoral methodology
- Overall design:
  - Construct sector-wise indices of financial stress and combine into overall indices weighted by sector size relative to combined sectors considered.
  - Two overall indices:
    - FSI5: encompassing banking sector, foreign exchange, equity markets, domestic and external government debt markets (valid observations for 22 economies).
    - FSI3: encompassing banking sector, foreign exchange, and equity markets (used for regression analysis due to broader country coverage).
- Sector-wise measurement principles:
  - Use variables capturing first and, when possible, second moments of price data; include volatility measures (GARCH(1,1) conditional variances for one-day nominal returns in stock and foreign exchange markets).
  - Transform series into percentiles across countries and time; sector score equals average percentile across indicators.
- Banking sector stress indicators:
  - Do not use aggregate balance-sheet indicators (NPL ratios) because of data availability/timeliness and the long build-up of vulnerabilities.
  - Preferred measures: real overnight interbank rate and ratio of central bank credit to total bank assets.
- Foreign exchange market stress:
  - Deviations from standard pressure indices: add volatility measure; use real effective exchange rate; use net international reserves to imports ratio; use levels rather than arbitrary period percentage changes.
- Stock market stress:
  - Decouple CMAX-style volatility-loss measure into two separate variables: level of real stock market index and volatility of nominal stock returns.
- Government debt markets:
  - External market stress: spread between return on country foreign debt (or hedged return on domestic debt) and comparable-maturity U.S. government securities.
  - Domestic market stress: quarterly real treasury bill rate (due to lack of daily domestic bond yield data).
- Index of quality of financial policies (QFPI):
  - Based on principle-by-principle FSAP assessments of Basel Core Principles (BCP) and IOSCO Objectives and Principles of Securities Regulation (OPSR).
  - Construct sector-wise indices for banking and securities markets; combined index weighted by sector sizes (total assets of deposit money banks to GDP; stock market capitalization to GDP).
  - Within each sector, group constituent principles into four areas and average:
    - governance;
    - regulatory practices;
    - prudential framework;
    - financial integrity and safety net arrangements.
  - Equal weights across the four areas to produce sector scores; combined index is simple average of sector scores.

### IV. Regression analysis — model, data, and specification
- Data and frequency:
  - Panel data at annual frequency over 1998–2003.
  - Dependent variable: overall index of financial stress (FSI3) converted from quarterly to annual by taking maximum quarterly realization per year.
  - Number of countries used in regression: 55; number of observations: 271.
- Conceptual model:
  - FSIit = αi + Σn ξn Fnit + Σk δk Ckit + Σj βj Ij + Σ interactions of institutions with fundamentals/contagion + εit
  - Emphasis on interaction terms: institutions (QFPI) modify the impact of macro fundamentals and contagion on financial stress.
- Contagion channels constructed:
  - Direct and indirect contagion via trade and financial flows (direct contagion; indirect/common-third-country channels such as competitive devaluation, common lender channel, portfolio rebalancing).
- Estimation method:
  - Arellano and Bover (1995) / Blundell and Bond (1998) GMM-SYS estimator (Roodman 2003 STATA module) to handle endogeneity, dynamic panel, and country fixed effects.
  - Use option to keep instrument count manageable (one instrument per variable and lag distance).
  - Static specification showed serial correlation in disturbances; a lagged dependent variable was added producing a partial adjustment model.

### V. Key regression results and magnitudes (preferred GMM-SYS specification for FSI3)
- Model fit and tests:
  - Overall R2: 0.62 (squared correlation between FSI3 and model predictions).
  - Arellano-Bond AR(1) test in first differences: -2.92 ** (statistically significant).
  - Arellano-Bond AR(2) test in first differences: -1.6 (not significant at 99%).
  - Number of instruments: 30.
  - Sargan test of overidentifying restrictions: 28.3 (statistically insignificant at 99%).
  - Number of countries: 55; number of observations: 271.
- Estimated short-run coefficients (robust standard errors in parentheses; significance):
  - Lagged value of FSI3: 0.52 (0.11) **.
  - Constant: 19.73 (7.19) **.
  - General government balance as percent of GDP (GovBal): -2.11 (0.95) *.
  - Interaction QFPI_x_GovBal: 0.023 (0.01) *.
  - Index of terms of trade (TOT): 0.15 (0.04) **.
  - Interaction QFPI_x_TOT: -0.002 (0.0005) **.
  - Measure of contagion among constituents of the MSCI Emerging Markets Equity Index (ContMSC): -27.25 (9.91) **.
  - Interaction QFPI_x_ContMSC: 0.37 (0.13) **.
  - Time trend (TIME): -4.39 (1.10) **.
  - Interaction QFPI_x_TIME: 0.059 (0.01) **.
  - ** denotes statistically significant at 99% level; * denotes statistically significant at 95% level.
- Static long-run solution (reported expression):
  - FSI3 = 40.89 – 4.36 GovBal + 0.05 QFPI_x_GovBal + 0.32 TOT – 0.003 QFPI_x_TOT – 56.47 ContMSC + 0.77 QFPI_x_ContMSC – 9.11 TIME + 0.12 QFPI_x_TIME
- Interpretation of interactions:
  - Higher QFPI reduces the impact of government deficits (GovBal) and inflationary pressures from terms-of-trade improvements (TOT) on FSI3.
  - Contagion via ContMSC shows a direct negative coefficient (ContMSC: -27.25 **), implying that, on average in 1998–2003, higher realized stress in one MSCI Emerging Markets constituent tended to decrease financial stress in peer countries (consistent with capital being channeled away from the troubled country into other emerging markets). Interaction with QFPI is positive (0.37 **), implying the contagion effect becomes less negative and can reverse for higher QFPI values.
- Quantified scenario effects (Table 2; changes reported in basis points of FSI3):
  - GovBal falls below its sample mean by one percentage point (with TOT, ContMSC, TIME at sample means):
    - Change in predicted FSI3 with QFPI at sample mean (67.66): Short run 0.52; Long run 1.08.
    - QFPI at 25-th percentile (56.49): Short run 0.78; Long run 1.62.
    - QFPI at 75-th percentile (80.89): Short run 0.21; Long run 0.43.
  - TOT rises above its sample mean by ten points (with GovBal, ContMSC, TIME at sample means):
    - QFPI mean: Short run 0.42; Long run 0.88.
    - QFPI 25th: Short run 0.61; Long run 1.26.
    - QFPI 75th: Short run 0.21; Long run 0.42.
  - ContMSC doubles from its sample mean (with GovBal, TOT, TIME at sample means):
    - QFPI mean: Short run -0.56; Long run -1.16.
    - QFPI 25th: Short run -1.62; Long run -3.36.
    - QFPI 75th: Short run 0.69; Long run 1.43.
  - TIME increments by one period from its sample mean (with GovBal, TOT, ContMSC at sample means):
    - QFPI mean: Short run -0.40; Long run -0.82.
    - QFPI 25th: Short run -1.06; Long run -2.19.
    - QFPI 75th: Short run 0.38; Long run 0.80.
- Descriptive statistics (selected from Table 3):
  - FSI3 (basis points): Valid observations 271; Mean 52.12; Standard deviation 16.99; Minimum 5.73; Maximum 92.65.
  - GovBal (%) : Valid observations 271; Mean -2.30; Standard deviation 4.51; Minimum -24.61; Maximum 13.73.
  - TOT (basis points): Valid observations 271; Mean 112.43; Standard deviation 58.01; Minimum 53.11; Maximum 532.73.
  - ContMSC (basis points): Valid observations 271; Mean 0.26; Standard deviation 0.91; Minimum 0; Maximum 6.14.
  - TIME: Valid observations 271; Mean 4.01; Standard deviation 1.42; Minimum 2; Maximum 6.
  - QFPI (basis points): Valid observations 271; Mean 67.66; Standard deviation 18.44; Minimum 24.73; Maximum 100.

### VI. Key substantive findings and implications
- Institutions-moderated effects:
  - Countries with higher quality of financial policies (higher QFPI) are better able to contain the effects of government deficits and inflationary pressures (from terms-of-trade improvements) on the overall level of financial system stress.
  - The short run and long run effects of changes in macro variables on FSI3 are larger in absolute terms when evaluated at lower percentiles of QFPI (25th) than at higher percentiles (75th); higher QFPI attenuates adverse effects.
- Contagion and capital flows among emerging markets:
  - Evidence does not support a portfolio rebalancing contagion channel among emerging markets in 1998–2003; instead, higher stress in one MSCI Emerging Markets constituent tended to coincide with lower stress in peers (consistent with reallocation of capital among emerging markets).
  - The negative effect of contagion-driven influx of foreign funds on recipient emerging markets’ stress is more pronounced for low QFPI values; for higher QFPI values (above the 55-th percentile), high stress in one country leads to increases in peer-country stress.
- Time trend:
  - Autonomous time effects: slight downward trend of average financial stress among countries below 55-th percentile of QFPI; slight upward trend among countries above 55-th percentile. Short run effects of time on FSI3 are small (less than two basis points per year).
- Fit and robustness:
  - Preferred specification yields a good fit (R2 = 0.62) and passes tests for second-order serial correlation and overidentifying restrictions.

### VII. Conclusions and data implications
- Constructed multi-country indices:
  - Overall financial stress index (FSI3) encompassing banking sector, foreign exchange, and equity markets using first- and second-moment measures and percentile transformation.
  - Index of quality of financial policies (QFPI) based on FSAP principle-by-principle assessments for banking and securities regulation, aggregated across four policy areas.
- Main conclusion:
  - Financial system stress depends on the interplay between macroeconomic fundamentals, cross-border contagion, and institutional quality; higher-quality financial policies mitigate the adverse impact of deficits and terms-of-trade shocks on financial stress.
- Data and future research:
  - Wider availability of data over time could permit stress indices that incorporate additional financial sectors and more robust regression analysis using repeated measurements of quality of financial policies.

*Source: Excerpt from IMF working paper (3. Groupings of Principles of Main Regulatory IFS Standards, pages 43–24 of provided PDF content).*

### REFERENCES

### _wp05173 - REFERENCES

### References (bibliographic listings)
- Comprehensive bibliography of works cited covering financial crises, systemic risk, econometric methods, financial stability indicators, and related empirical studies. Notable recurring themes/authors include:
  - Financial crisis theory and history: Kindleberger (1978); Eichengreen and Portes (1987); Sachs, Tornell, and Velasco (1995).
  - Contagion and crisis propagation: Dornbusch, Park, and Claessens (2000); Van Rijckeghem and Weder (2000); Forbes and Rigobon (2002).
  - Financial stability measurement and indicators: IMF publications (Financial Soundness Indicators; Global Financial Stability Report); Illing and Liu (2003); Borio and Lowe (2002).
  - Econometric and panel-data methods: Arellano (2003); Blundell and Bond (1998); Bond (2002); Roodman (2003).
  - GARCH/volatility modeling and ARCH literature: Bollerslev, Chou, and Kroner (1992); Bera and Higgins (1993).
- The references include working papers, IMF and World Bank reports, NBER papers, BIS publications, central bank working papers, and published journal articles.

### Appendix I — Sector-wise indices of financial stress: overview and methodology
- Sample frequency and span:
  - Analysis carried out at quarterly frequency.
  - Eleven year period 1993–2003.
  - Each country typically has 44 quarterly observations (minus any missing values).
- Rationale for choices:
  - Quarterly aggregation of raw daily data mitigates seasonality and reduces revisions relative to monthly data.
  - Long sample start date chosen to include major 1990s crises and to improve distributional assumptions for weighting schemes.

### Transformations and treatment of missing values
- Directional transformations:
  - Variables that signal higher financial stress when low (e.g., real asset prices, level of international reserves) are multiplied by -1 so that increases in transformed values indicate greater stress.
- Missing-value treatment (two cases):
  - Do not extrapolate when missing values are at the beginning, end, or permeate the whole series.
  - Otherwise:
    - Series that do not exhibit time trends: replace missing values by the arithmetic average of pre- and post-missing-period realizations.
    - Trending series: extrapolate assuming constant quarterly growth equal to the geometric mean implied by pre- and post-missing-period realizations.

### Variables by sector (definitions, frequencies, and sources)
- Banking sector:
  - Real Overnight Interbank Rate: (R_t / (1 + π_t) + 1)? [expression shown in source]. Quarterly frequency.
  - R_t — nominal money market rate (quarterly). Source: IMF’s International Financial Statistics; CEIC; Bloomberg; country websites.
  - π_t — inflation (quarterly). Annualized growth rate of quarterly CPI: CPI_Infl = 4⋅(ln(CPI_t) − ln(CPI_{t-1})), where CPI base = 100 in 2000. Source: IMF’s International Financial Statistics.
  - Central bank credit to deposit money banks (DMB) as ratio of DMB’s total assets (quarterly).
  - Central bank credit to DMB (quarterly). Source: IMF’s International Financial Statistics; unpublished country data.
  - DMB’s total assets (extrapolated quarterly data 27)—sum of all asset types in DMBs’ aggregate balance sheet. If sum of assets exceeds sum of liabilities by more than one percent, value set to not available. Source: IMF’s International Financial Statistics.
  - Note on extrapolation: Quarterly data extrapolated from annual data under assumption described in footnote 27.

- Foreign exchange market:
  - International reserves net of gold in U.S. dollars as ratio of imports in U.S. dollars multiplied by -1 (quarterly).
  - Real effective exchange rate (REER—nominal exchange rate measured in foreign currency units per unit of domestic currency, expressed in real terms) multiplied by -1 (quarterly).
  - Quarterly averages of the conditional variance (h_t) of one-day nominal returns (percent) on the foreign exchange market (domestic currency units per US$1), modeled as GARCH(1,1) for all countries with exchange rate regimes other than fixed:
    - r_t = ζ_0 + u_t
    - u_t = ε_t h_t^{1/2}, ε_t | F_{t-1} ~ N[0,1]
    - h_t = α_0 + α_1 u_{t-1}^2 + β_1 h_{t-1}, t = 2, ..., T
    - α_0 ≥ 0, α_1 + β_1 ≥ 0, and α_1 + β_1 < 1
    - r_t = 100⋅ln(E_t − E_{t-1}); E_t – nominal exchange rate (domestic currency units per US$1) (daily frequency). Source: Bloomberg and Datastream.
  - For fixed-exchange-rate countries: use quarterly averages of realized squared one-day nominal foreign exchange returns instead of GARCH-based conditional variance. Exchange rate regime classifications from IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions. Fixed regimes include exchange arrangements with no separate legal tender, currency board arrangements, and conventional pegs.

- Stock market:
  - Real stock market index = (SMI_t ⋅ CPI_t^{-1}), scaled by the average of its quarterly values in 2002 (average 2002 = 100), then multiplied by -1 (quarterly).
  - SMI_t — average quarterly nominal stock market index derived from daily data. Sources: Bloomberg, Datastream, CEIC, country exchange websites.
  - CPI_t — Consumer Price Index (base = 100 in 2000). Source: IMF’s International Financial Statistics.
  - Quarterly averages of conditional variance (h_t) of one-day nominal stock market returns (percent), modeled as GARCH(1,1) with the same specification constraints as FX returns:
    - r_t = 100⋅ln(SMI_t − SMI_{t-1}); SMI_t daily frequency. Sources as above.

- External government debt market:
  - Emerging markets: quarterly averages of stripped spreads between return on countries’ U.S. dollar-denominated foreign debt and comparable-maturity U.S. government securities (daily). Source: J.P. Morgan Emerging Markets Bond Index.
  - Developed countries: quarterly averages of spread between the 12-month hedged return in U.S. dollars of benchmark government bond index and the 12-month return of the U.S. benchmark government bond index (daily). Source: J.P. Morgan Government Bond Index.

- Domestic government debt market:
  - Real 3-Month Treasury Bill rate: (R_t / (1 + π_t) + 1)? [expression shown in source]. Quarterly frequency.
  - R_t — nominal 3-Month Treasury Bill rate (quarterly). Source: IMF’s International Financial Statistics.
  - π_t — inflation (quarterly). Annualized growth rate of quarterly CPI using the same formula as above. Source: IMF’s International Financial Statistics.

### Weighting schemes for sector-wise indices
- Two weighting schemes tested:
  1. Variance-equal weights method:
     - Weights each variable by the inverse of its sample standard deviation (Illing and Liu, 2003).
     - Under normality, weighted variables have equal variances to prevent dominance by any single variable.
     - Limitations noted:
       - Penalizes variables with higher within-country variance (which are often better indicators of stress because volatility tends to rise during crises).
       - Ill-suited for heterogeneous country sets; may assign higher index scores to low-variance countries (developed) than to crisis-prone emerging markets.
       - Best suited for constructing intermediate indices used to define binary crisis dummies for homogeneous country sets.
  2. Percentiles transformation method (chosen):
     - Uses the sample cumulative distribution function of each variable across all countries and quarters to transform values into percentiles.
     - Definition: p-th percentile is the value with p percent of measurements below it and (100-p) percent above it. Minimum transformed value = 0; maximum = 99.
     - Advantages:
       - Does not rely on normality.
       - More transparent and intuitively interpretable country scores.
       - More liberal treatment of missing values; allows assignment of scores to more countries.
       - Because percentiles transformation assigns equal weights dependent only on number of variables, weights can be redefined in presence of missing values as the reciprocal of the number of variables with valid observations in a quarter (approach used in this paper).

- Final sector-wise index:
  - Obtained by averaging the percentiles-transformed series for each country in each quarter.

### Coverage of financial stress indices (Table summary)
- Column counts (number of countries with valid observations by sector/index):
  - Banking sector: 68
  - Stock market: 52
  - Foreign exchange market: 70
  - External government debt market: 33
  - Domestic government debt market: 41
  - Overall indices based on all five sectors/markets: 22
  - Overall indices based on the banking sector, and stock and foreign exchange markets: 65
- Table lists country-by-country existence of valid observations (marked “X”) for each sector-wise index and for two types of overall financial stress indices. Notes state: “X” indicates the existence of valid observations on the particular index of financial stress for a given country or economy.

*Source: _wp05173 - REFERENCES (IMF PDF content provided).*

### 2. An empty cell indicates that a particular index could not be estimated due to lack of data.

### _wp05173 - 2. An empty cell indicates that a particular index could not be estimated due to lack of data.

### Note on missing-data notation
- An empty cell indicates that a particular index could not be estimated due to lack of data.

*Source: _wp05173 - 2. An empty cell indicates that a particular index could not be estimated due to lack of data.*

### 3. The overall indices of financial stress could have valid observations for a given country or economy, even when

### _wp05173 - 3. The overall indices of financial stress could have valid observations for a given country or economy, even when some of the indices used in their construction do not, if the weights assigned to these indices are zero.

### Sector-wise indices of quality of financial policies: construction and grouping
- Primary data: categorical variables from principle-by-principle BCP and OPSR assessments showing grades assigned to countries’ observance of constituent principles.
- Four broad areas of financial policies (mapping broadly based on Joint Forum, 2001; see IMF (2004)):
  - Regulatory governance: capacity of supervisory agencies; objectives of regulation; independence and adequate resources; enforcement powers and capabilities; clarity and transparency of regulatory process; external participation.
  - Regulatory practices: practical application of laws, rules, and procedures; group-wide supervision; monitoring and on-site inspection; reporting to supervisors; enforcement, cooperation and information sharing; confidentiality; licensing, ownership transfer, and corporate control; qualifications.
  - Prudential framework: rules, directives, and regulatory requirements guiding firms’ risk management and internal controls; risk concentration; capital requirements; corporate governance.
  - Financial integrity and safety net arrangements: policies and instruments to promote fairness and integrity and safeguards for depositors, investors and policyholders.
- Index construction (banking sector and securities market):
  - Equal weights assigned to countries’ performance in the four areas.
  - For each regulatory standard and area, calculate the proportion of principles (under that grouping) that have been assigned a particular assessment grade, out of the total principles in that grouping found applicable and assessed.
  - Multiply proportions by numeric weights for assessment grades:
    - Compliant/Implemented—100
    - largely compliant/Broadly implemented—66.67
    - materially noncompliant/Partially implemented—33.33
    - Noncompliant/Nonimplemented—0
  - Sum resulting terms to obtain area score; countries’ final sector-wise scores are simple averages of the four area scores.
- Note: No index created for insurance sector due to insufficient information to construct an index of financial stress.

### Coverage of BCP and IOSCO standards (summary statistics from Table 2)
- Column counts (memorandum item): 68, 66, 46.
- Footnotes:
  - Excluding assessments conducted prior to the development of the assessment methodology by the Basel Committee on Banking Supervision.
  - Excluding qualitative assessments that do not assign assessment grades to individual principles.
- Source noted: IMF internal databases.

### Groupings of principles by area (excerpted mapping, Appendix Table 3)
- Banking sector and securities markets: constituent principles are grouped under the four areas (I. Regulatory governance; II. Prudential framework; III. Regulatory practices; IV. Financial integrity and safety nets).
- Specific BCP and OPSR principle identifiers are listed under each grouping in the source (e.g., BCP Principle 1(1) ... BCP Principle 25; OPSR Principle 1 ... OPSR Principle 30).

### Measures of contagion: methodology and specific constructs
- General approach: All contagion measures constructed in the spirit of Eichengreen, Rose, and Wyplosz (1996). For a country X in a given year, each contagion measure equals:
  - (sector-wise index of financial stress in another country Y) × (importance for country X of trade or financial links with Y) × (relative importance of the financial system segment through which stress spills over in country X).
- Foreign exchange market (contagion via trade):
  - Top three trading partners for each country identified using IMF country weights applied in real multilateral effective exchange rates.
  - Germany identified as a top-three trading partner for 54 countries in the sample.
  - Two trade-contagion measures:
    - Maximum quarterly realization of index of financial stress in the foreign exchange market in the top trading partner, multiplied by that partner’s weight in total external trade of the sample country; result multiplied by relative share of the foreign exchange market in the financial systems of sample countries.
    - For countries with Germany as one of their three top trading partners: maximum quarterly realization of FX stress in all countries that have Germany as a top-three partner, multiplied by the weight of Germany in total external trade of each of these countries (value set to “0” in the country where the maximum realization of stress occurred); result multiplied by relative share of the foreign exchange market. For all other sample countries: contagion measure set to “0.”
- Banking sector (contagion via international bank lending):
  - BIS (2004) data used to calculate average shares over 1993–2003 of each of the 24 BIS reporting countries in consolidated international claims on each sample country; three BIS reporting countries with highest shares identified.
  - France identified as one of the top three international bank lenders for 40 countries in the sample.
  - Two banking-contagion measures:
    - Maximum quarterly realization of index of financial stress in the banking sector in the top international bank lender, multiplied by that lender’s weight in consolidated international claims on the sample country; result multiplied by relative share of the banking sector in the financial systems of sample countries.
    - For countries with France as one of their three top international bank lenders: maximum quarterly realization of banking-sector stress in all countries that have France as a top-three lender, multiplied by the weight of France in consolidated international claims on each of these countries (value set to “0” in the country where the maximum realization occurred); result multiplied by relative share of the banking sector. For all other sample countries: contagion measure set to “0.”
- Stock market contagion:
  - Focus on sample countries’ inclusion in the MSCI Emerging Markets Index.
  - For sample countries that are members of the MSCI Emerging Markets Index: maximum quarterly realization of index of financial stress in the stock markets of all index members, multiplied by the weight of each sample country in that index (value set to “0” in the country where the maximum realization occurred); result multiplied by relative share of the equity market in the financial systems of sample countries. For other countries: contagion measure set to “0.”
  - MSCI Emerging Markets Index sample constituents listed in source.
- External government debt market contagion:
  - Focus on sample countries’ inclusion in the J.P. Morgan Emerging Markets Bond Index.
  - For members of the J.P. Morgan EMBI: maximum quarterly realization of index of financial stress in external government debt markets of all index members, multiplied by the weight of each sample country in that index (value set to “0” in the country where the maximum realization occurred); result multiplied by relative share of the foreign exchange market in the financial systems of sample countries (the weight is not the relative share of the external government debt market because the financial stress index does not cover that segment). For other countries: contagion measure set to “0.”
  - J.P. Morgan EMBI sample constituents listed in source.

### Macroeconomic fundamentals: definitions and data sources (Appendix IV)
- Variables and measurement:
  - Log of real PPP GDP per capita, measured in constant 1995 international U.S. dollars (annual frequency). Source: World Bank’s World Development Indicators.
  - Average value of inflation, measured at quarterly frequency in percent. Inflation at quarterly frequency estimated as the annualized growth rate of quarterly CPI in percent. Source: IMF’s International Financial Statistics.
  - Index of terms of trade in goods and services (1995=100, annual frequency). Unit value of exports as a ratio of unit value of imports. Source: Unpublished IMF data, derived from IMF’s International Financial Statistics.
  - General government balance as percent of GDP (annual frequency). Source: Unpublished IMF data, derived from IMF’s International Financial Statistics.
  - Gross fixed capital formation as percent of GDP (annual frequency). Source: Unpublished IMF data, derived from IMF’s International Financial Statistics.
  - Ratio of broad money (converted in U.S. dollars) to international reserves net of gold in USD (annual averages of quarterly data). Source: IMF’s International Financial Statistics.
  - Imports of goods and services plus exports of goods and services as percent of GDP (annual frequency). Source: Unpublished IMF data, derived from IMF’s International Financial Statistics.
  - Number of capital account restrictions in place as percent of the total number of capital account restrictions judged to be applicable for a given country (annual frequency). Source: IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions.
  - Dummy variables for countries’ exchange rate regimes (annual frequency). Source: IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions:
    - Fixed exchange rate regime—exchange arrangements with no separate legal tender; currency board arrangements; and conventional pegs.
    - Crawling peg exchange rate regime—pegs within horizontal bands; crawling pegs; crawling bands.
    - Floating exchange rate regime—managed floating and independently floating.
  - Dummy for explicit deposit insurance scheme in 1999. Source: Demirguc-Kunt and Sobaci, 2000.
- Footnote on capital account restrictions: list of 11 categories (1) restrictions on transactions with capital market securities; (2) restrictions on transactions with money market instruments; (3) restrictions on transactions with collective investment securities; (4) controls on derivatives and other instruments; (5) restrictions on commercial credits; (6) restrictions on financial credits; (7) restrictions on guarantees, sureties, and financial backup facilities; (8) controls on direct investment; (9) controls on liquidation of direct investment; (10) controls on real estate transactions; and (11) controls on personal capital transactions.

*Source: IMF internal databases.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05173.pdf_