## _wp0618 — Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)

## Source details

**Canonical URL:** [_wp0618 — Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2006/_wp0618.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2006/_wp0618.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2006/_wp0618.pdf.json)

---

### Research question, scope, and data
- Focus: How entry of foreign banks affects financial sector development in poor countries (World Bank classification: low income and lower middle income).
- Time coverage in main analyses: 1995-2002; key averages for outcomes use 1999-2002 (private credit and cost efficiency) and 2003-04 for access indicators.
- Sample: 89 countries in the low- and lower-middle-income group; regressions use smaller samples depending on data availability. Four outliers excluded from regressions for private credit: China P.R., Jordan, Eritrea, and Thailand.
- Main dependent variables:
  - Commercial bank credit to private sector to GDP (IFS line: deposit money banks only).
  - Cost-efficiency: average ratio of overhead costs (OH) to total assets (Bankscope), averages over 1999-2002.
  - Access indicators: branch and ATM networks, number of loans and deposit accounts (2003-04 coverage).
- Measurement of foreign bank presence: share of foreign bank assets to total bank assets (Micco, Panizza, and Yañez (2004); bank = foreign if ≥ 50 percent nonresident capital); earliest available year used in cross-section (1995–1998 depending on country).
- Controls: GDP per capita, transition-country dummy, inflation, lack of corruption, creditor information, enforcement speed, state-owned bank share, concentration, banking crises, regional dummies.

### Theoretical framework and welfare implications
- Agent types: {B, S, H} with proportions {μB, μS, μH}, 1 = μB + μS + μH.
- Monitoring technologies:
  - Hard-information monitoring cost cH (identifies H types); assumed cH < cS.
  - Soft-information monitoring cost cS.
- Equilibrium types (Proposition 1):
  - (A) Pooling equilibrium: Total credit = 1; Foreign bank share = 0.
  - (B) Separating equilibrium: Total credit = μH + μS; Foreign bank share = 1 (foreign banks lend to H types).
  - (C) Semi-pooling equilibrium: Total credit = 1; Foreign bank share = μH.
  - (D) Credit-constrained equilibrium: Total credit = μH; Foreign bank share = 1.
- Foreign/domestic bank comparative costs:
  - Foreign banks: lower cost of hard-information monitoring (cH − ∆) and higher cost of soft-information monitoring (cS + ∆′).
- Monitoring-costs per unit lent (as reported in source):
  - Pooling: monitoring costs/total credit = 0; foreign bank share = 0.
  - Partially pooling: monitoring costs/total credit = (∆−H c) μH.
  - Separating: monitoring costs/total credit = S H S H c c μμ μμ +∆−)( (notation preserved).
  - Credit-constrained: monitoring costs/total credit = ∆−H c ; foreign bank share = 1.
- Welfare implications (summary):
  - Foreign entry is not unambiguously welfare-improving.
  - If initial equilibrium is non-pooling, entry benefits H types (lower rH) and welfare rises (gains concentrated on less opaque borrowers).
  - If initial equilibrium is pooling, foreign entry can move the economy to equilibria (B), (C), or (D), often reducing aggregate lending and producing ambiguous welfare effects; soft-information borrowers (S) are never better off with foreign entry.

### Key empirical findings — private credit and foreign bank presence
- Cross-sectional OLS (LICs):
  - Robust, negative association between foreign bank share and private credit across specifications.
  - Magnitude example: an increase in foreign share by one standard deviation → decline in private credit of about 6 percentage points of GDP (about one-third of average private credit/GDP in sample).
  - Results robust to alternative foreign-share measures, use of historical GDP, banking crisis controls, and settlers’ mortality/legal origin controls.
- Credit-growth regressions (changes 1994-96 to 1999-2001):
  - Larger foreign presence associated with lower subsequent growth in credit to the private sector.
  - Example magnitude: increase in foreign share by one standard deviation → decline in growth rate of credit of about 20 percent (based on Table 4 column cited).
  - Inflation and share of state banks negatively and significantly affect credit growth; lack of corruption positive.
- Panel regressions:
  - Fixed-effects regressions with lagged dependent variable: foreign bank share significantly and negatively associated with private credit in most specifications.
  - System GMM estimates: foreign bank presence remains significantly and negatively associated after instrumentation; lagged dependent variable coefficient close to one.
  - Panel statistics (selected):
    - Panel A (Fixed effects): lagged dependent variable coefficients 0.47; 0.53; 0.5; 0.48 with t-stats [9.90]***, [10.26]***, [9.22]***, [8.72]***; foreign ownership coefficients include -0.68; -0.3; -0.38; -0.42 with t-stats [3.55]***, [1.78]*, [2.15]**, [2.35]**.
    - Panel B (System GMM): lagged dependent variable coefficients 0.87; 0.88; 0.87 with t-stats [32.09]***, [29.68]***, [23.69]***; foreign ownership -0.44; -0.24; -0.39 with t-stats [2.04]**, [1.64]*, [1.67]*.
    - Sargan p-values reported: 0.341; 0.836; 0.991; serial correlation order-2 p-values include 0.616; 0.665; 0.75.
- Instrumental variables (2SLS) — Table 6 (selected):
  - Instruments: log population, share Muslim, share Buddhist, share Christian Orthodox (various specifications).
  - Second-stage foreign ownership coefficients: -21.37; -20.89; -24.79; -21.57; -28.54 (robust z statistics reported).
  - First-stage log population coefficients: -0.048; -0.047; -0.047; -0.048; -0.024 with [1.80]*, [1.73]*, [1.87]*, [1.80]*, [1.71]* respectively.
  - Partial R-squared: 0.25; 0.26; 0.27; 0.27; 0.24. F-statistics: 9.26; 13.56; 11.85; 13.31; 7.92. p-values of F test: 0.0000; 0.0000; 0.0000; 0.0000; 0.0005.
  - Hansen J statistic p-values reported (e.g., 0.23; 0.51; 0.64; 0.5; 0.52).
  - 90 percent confidence intervals robust to weak instruments (Moreira, 2003) for Foreign ownership: [-25.3 , -21.1]; [-23.7 , -19.7]; [-27.6 , -23.7]; [-24.5 , -20.6]; [-33.3, -28.0].
  - Observations: 57; 59; 59; 59; 39; R-squared: 0.58; 0.55; 0.51; 0.55; 0.58.
- Interpretation: IV results support a negative effect of foreign bank presence on private credit in LICs; IV coefficients larger in magnitude than OLS (consistent with attenuation bias in OLS).

### Access to financial services, overhead costs, and heterogeneity
- Access indicators and foreign ownership (Table 8 correlations):
  - Sample stated: 62 low-income and lower-middle-income countries (Table 1).
  - Bivariate correlations (examples):
    - Foreign Ownership vs Demographic branch penetration: -0.3969 (P-value 0.0124).
    - Foreign Ownership vs Geographic branch penetration: -0.413 (P-value 0.009).
    - Foreign Ownership vs Demographic ATM penetration: -0.2868 (P-value 0.1001).
    - Foreign Ownership vs Geographic ATM penetration: -0.3777 (P-value 0.0277).
    - Foreign Ownership vs Loans per capita: -0.3607 (P-value 0.1414).
    - Foreign Ownership vs Deposits per capita: -0.4642 (P-value 0.0223).
  - Cross-sectional OLS (Table 9) — selected coefficients:
    - Foreign ownership: demographic branch penetration -5.34 [3.10]***; -4.6 [2.92]***; -13.34 [2.38]** in different specs.
    - Foreign ownership: deposit accounts per capita -926.77 [3.12]***; -601 [2.52]**; -892.7 [3.23]*** in different specs.
    - Observations and fit for deposit accounts: Observations 24; R-squared 0.58; 0.68; 0.58 (across columns).
  - Interpretation: All access indicators are negatively correlated with the foreign share, and most correlations are large and statistically significant; negative correlation remains after controlling for GDP per capita, population density, inflation, and corruption.
- Overhead costs and cost efficiency (Table 7 and Table 10):
  - Regressors explain at most 34 percent of variation in overhead costs (maximum R-squared reported up to 0.34 in Table 7).
  - Inflation robustly and positively correlated with overhead costs; better availability of information about debtors tends to reduce overhead costs.
  - Foreign ownership coefficients in overhead regressions: 0.02; 0.027*; 0.032**; 0.029*; 0.024*; 0.01; 0.015; 0.013 across specs with respective t-stats preserved.
  - In cross-country Table 10, foreign ownership effect on overhead costs: 0.007 [0.77]; 0.017 [1.46]; -0.003 [0.29] across panels.
  - Using changes in cost efficiency yields poor fit and no robust relationship with foreign bank presence.
- Heterogeneity by income:
  - Extending sample to all 102 countries: coefficient on foreign banks much smaller and no longer significant in private credit regressions; effect disappears when higher-income countries included.
  - Interpretation: foreign bank presence appears to have different effects in poor countries versus advanced countries; local knowledge and relationship lending matter more in poor countries.

### Quantitative summary of selected regression coefficients and diagnostics (preserved)
- Cross-section OLS (Table 2, selected):
  - GDP per capita coefficients (columns (1)-(8)): 7.04, 8.16, 9.05, 6.99, 5.65, 5.17, 6.38, 6.56.
  - Foreign ownership coefficients (columns (1)-(8)): -6.59, -15.3, -22.56, -19.6, -18.88, -21.5, -23.4, -23.78.
  - Observations: 62, 62, 61, 61, 59, 59, 45, 45.
  - R-squared: 0.21, 0.34, 0.56, 0.61, 0.63, 0.64, 0.66, 0.67.
- Panel estimation (Table 5, selected):
  - Initial depth: -0.01; -0.02; -0.02; -0.02; -0.02; -0.02 with t-stats [3.38]***, [4.57]***, [4.83]***, [5.12]***, [3.84]***, [4.49]***.
  - Foreign ownership (panel): -0.55; -0.56; -0.68; -0.52; -0.58; -0.54 with t-stats [2.29]**, [2.42]**, [2.77]***, [2.05]**, [2.12]**, [2.33]**.
  - Observations: 59; 59; 59; 59; 50; 59. R-squared range reported: 0.23 to 0.38 across columns.
- Instrumental variables (Table 6, selected diagnostics):
  - Partial R-squared: 0.25; 0.26; 0.27; 0.27; 0.24.
  - F statistic: 9.26; 13.56; 11.85; 13.31; 7.92.
  - Hansen J p-values: e.g., 0.23; 0.51; 0.64; 0.5; 0.52.
  - 90 percent weak-instrument-robust CIs for Foreign ownership: [-25.3 , -21.1]; [-23.7 , -19.7]; [-27.6 , -23.7]; [-24.5 , -20.6]; [-33.3, -28.0].

### Interpretation and policy-relevant implications
- Empirical interpretation: Evidence consistent with theoretical “cream-skimming” — foreign banks concentrate on less informationally opaque (high-quality) borrowers, reducing credit to opaque borrowers and decreasing private credit depth in LICs.
- Trade-offs of foreign bank entry:
  - Potential benefits: improved cost-efficiency for hard-information lending, better risk management, competition, banking system stability gains in some cases.
  - Potential costs: decreased credit availability to opaque borrowers (SMEs, small firms), reduced branch/deposit outreach, slower credit growth, possible welfare losses for some parameter configurations.
- Policy considerations:
  - Reforms to improve property rights, contract enforcement, borrower information (credit registries), and collateral enforcement may be necessary in LICs to ensure foreign entry benefits broader financial inclusion.
  - Monitor changes in access indicators (branches, deposits per capita, SME lending) after liberalization of foreign entry.
  - Avoid relying solely on bank cost-efficiency (OH) as a welfare metric; use multiple indicators: private credit to GDP, credit growth, access measures, and distributional outcomes across borrower types.

*Italic source: Excerpted content from “5. Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)” and related tables and appendices in IMF Working Paper _wp0618.*

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

### _wp0618 - References..............................................................................................................

### Tables
- 1. Sample Summary Statistics ....................................................................................... 27
- 2. Financial Depth and Foreign Bank Presence in Poor Countries: OLS Regressions.. 28
- 3. Financial Depth and Foreign Banks in Poor Countries: Robustness Tests................ 29
- 4. Foreign Bank Presence and Changes in Financial Depth in Poor Countries ............. 30
- 5. Financial Depth and Foreign Bank Presence in Poor Countries: Panel Estimation .. 31
- 6. Financial Depth and Foreign Bank Presence in Poor Countries: Instrumental 
    Variables Estimation ............................................................................................... 32
- - 3 -
- 7. Bank Cost Efficiency and Foreign Bank Presence in Poor Countries ....................... 33
- 8. Access to Financial Services and Foreign Bank Presence in Poor Countries: 
    Bivariate Correlations ............................................................................................. 34
- 9. Access to Financial Services and Foreign Bank Presence in Poor Countries: 
    Cross-Sectional OLS Regressions .......................................................................... 35
- 10. Financial Depth, Cost Efficiency, and Foreign Bank Presence: OLS Regressions 
    Including All Countries........................................................................................... 36

### Figures
- 1. Impact of Foreign Bank Entry on Welfare and Aggregate Lending.......................... 37
- 2. Bank Credit to the Private Sector Overhead Costs in Lower-Income Countries 
     by Region (Percent of GDP) .................................................................................. 38
- 3. Foreign Bank Presence by Region in Selected Lower-Income Countries ................. 39
- 4. Private Credit to GDP and Foreign Bank Presence ................................................... 40

*Source: _wp0618 - References..............................................................................................................*

### 5. Change in Private Credit to GDP and Change in Foreign Bank Presence

### 5. Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)

### Introduction and research question
- Focus: How entry of foreign banks affects financial sector development in poor countries (World Bank classification: low income and lower middle income).
- Time coverage in main analyses: 1995-2002; key averages for outcomes use 1999-2002 (private credit and cost efficiency) and 2003-04 for access indicators.
- Sample: 89 countries in the low- and lower-middle-income group; regressions use smaller samples depending on data availability. Four countries (China P.R., Jordan, Eritrea, and Thailand) are excluded from regressions as outliers for private credit.
- Central hypothesis developed and tested: foreign bank entry can produce “cream-skimming” — foreign banks concentrate on less informationally opaque (high-quality) borrowers — which can reduce aggregate private credit and access for opaque borrowers.

### Theoretical framework: adverse selection, monitoring, and equilibria
- Agents: three entrepreneur types {B, S, H} with proportions {μB, μS, μH}, 1 = μB + μS + μH.
- Two monitoring technologies:
  - Hard-information monitoring: cost cH (identifies H types); cheaper than soft monitoring (assumed cH < cS).
  - Soft-information monitoring: cost cS (identifies S types).
- Banks offer contract menus (interest factor r and monitoring strategy). Four equilibrium types identified (Proposition 1):
  - (A) Pooling equilibrium: Total credit = 1; Foreign bank share = 0. Condition: R > rP and rP < rH (i.e., pooling break-even holds).
  - (B) Separating equilibrium: Total credit = μH + μS; Foreign bank share = 1 (foreign banks lend to H types). Banks monitor both hard and soft information; B types excluded.
  - (C) Semi-pooling equilibrium: Total credit = 1; Foreign bank share = μH (hard-information firms borrow from foreign banks; S and B pooled).
  - (D) Credit-constrained equilibrium: Total credit = μH; Foreign bank share = 1; only H types receive credit.
- With foreign and domestic banks:
  - Foreign banks assumed to have lower cost of hard-information monitoring (cH − ∆) and higher cost of soft-information monitoring (cS + ∆′), with 0 > ∆ and 0′ > ∆′ as parameter signs in the paper.
  - Foreign entry makes pooling less likely; hard-information firms gravitate to foreign banks; equilibria with greater foreign penetration tend to have less total lending (Proposition 2).
- Monitoring-costs per unit lent (as reported):
  - Pooling: monitoring costs/total credit = 0; foreign bank share = 0.
  - Partially pooling: monitoring costs/total credit = (∆−H c) μH; foreign bank share = μH (indeterminate in some formulations).
  - Separating: monitoring costs/total credit = S H S H c c μμ μμ +∆−)( (as presented in text); foreign bank share = μH+μS / μH? (notation preserved from source).
  - Credit-constrained: monitoring costs/total credit = ∆−H c ; foreign bank share = 1.
- Welfare implications:
  - Foreign entry is not unambiguously welfare-improving. If initial equilibrium is non-pooling, entry benefits H types (lower rH) and welfare rises, but gains accrue to less opaque borrowers.
  - If initial equilibrium is pooling, foreign entry can:
    - Move economy to separating (B): aggregate lending falls (B types no longer get credit); welfare ambiguous.
    - Move to semi-pooling (C): aggregate credit unchanged, welfare unambiguously declines due to monitoring costs with B projects still financed.
    - Move to credit-constrained (D): total lending falls; welfare ambiguous.
  - Soft-information borrowers (S) are never better off with foreign entry and can be worse off.

### Empirical methodology and data
- Main dependent variable: commercial bank credit to private sector to GDP (IFS line used: deposit money banks only).
- Cost-efficiency dependent variable: average ratio of overhead costs (OH) to total assets, computed from Bankscope bank-level data aggregated at country level; averages over 1999-2002.
- Access-to-financial-services indicators: branch and ATM networks, number of loans and deposit accounts (Beck, Demirgüç-Kunt, and Martinez Peria (2005)); information refers to 2003-04; coverage uneven.
- Measurement of foreign bank presence: share of foreign bank assets to total bank assets from Micco, Panizza, and Yañez (2004) using Bankscope plus outside sources for 1995-2002 (bank classified as foreign if ≥ 50 percent nonresident capital). Coverage changes over time; earliest available year used in cross-section (1995–1998 depending on country).
- Controls: GDP per capita (level of development), transition-country dummy, inflation, freedom from corruption, Doing Business measures (availability of information to creditors; time to enforce contracts), and other robustness checks (state-owned bank share, concentration, banking crises, regional dummies).
- Estimation strategies:
  - Cross-sectional OLS (benchmark: yi = α + β fi + γXi + ui), with extensive robustness checks.
  - Changes regressions: Δ ln(y) on initial y and foreign share to reduce endogeneity.
  - Panel regressions with country and time fixed effects and system GMM (Arellano and Bover, 1995) to address dynamic panel bias and endogeneity.
  - Instrumental variables: population size and religious composition (share practicing Islam, Buddhism, Greek Orthodox) as instruments for foreign bank presence. First-stage: instruments highly significant; partial R-squared > 20 percent; F-test on joint instrument significance above or close to 10 (Staiger and Stock (1997) rule).

### Key empirical findings
- Cross-sectional OLS (private credit to GDP):
  - Strong, robust negative association between foreign bank share and private credit across specifications for LICs.
  - Magnitude: an increase in foreign share by one standard deviation → decline in private credit of about 6 percentage points of GDP (about one-third of average private credit/GDP in sample).
  - Results robust to alternative foreign-share measures (supervisor survey), to using GDP in 1980 or 1970, to adding banking crisis controls, and to inclusion of settlers’ mortality/legal origin controls.
  - No clear threshold effect with institutional quality within the LIC sample.
- Credit-growth regressions (changes between 1994-96 and 1999-2001):
  - Larger foreign presence associated with lower subsequent growth in credit to the private sector.
  - Magnitude example: based on column 3 in Table 4, increase in foreign share by one standard deviation → decline in growth rate of credit of about 20 percent.
  - Controls with negative and significant effects on credit growth: inflation and share of state banks; lack of corruption positive.
- Panel regressions:
  - Fixed-effects regressions with lagged dependent variable: foreign bank share significantly and negatively associated with private credit in most specifications.
  - System GMM estimates: foreign bank presence remains significantly and negatively associated with private credit after instrumentation; lagged dependent variable coefficient close to one.
  - Robustness checks (Sargan, serial correlation tests) support validity of instruments in GMM.
- Instrumental-variable regressions:
  - Instruments (population, religion) are strongly correlated with foreign-share in first stage (partial R-squared > 20 percent; F-statistic around or above 10).
  - IV results support negative effect of foreign bank presence on private credit in LIC sample.
- Access to services and other dimensions:
  - Foreign bank presence negatively associated with indicators of access to financial services (branch penetration, number of deposits per capita).
  - Cost-efficiency relationship ambiguous in cross-country sample: monitoring costs and overhead (OH) may rise when foreign entry induces more monitoring — OH can increase even if welfare improves; thus OH regressions alone are insufficient to judge welfare impact.

### Interpretation and heterogeneity
- Overall interpretation: evidence consistent with theoretical “cream-skimming” mechanism — foreign banks focus on less opaque borrowers, reducing credit to informationally opaque firms and decreasing private credit depth in LICs.
- Heterogeneity by income:
  - Effect disappears (becomes statistically insignificant) when higher-income countries are included, consistent with model prediction that monitoring-ability differences and informational opacity are less marked in richer countries.
- Country heterogeneity:
  - Countries differ in monitoring costs and project returns; these differences lead to different equilibria (A–D) and explain variation in private credit and foreign penetration across countries.

### Policy-relevant implications and considerations
- Foreign bank entry carries trade-offs:
  - Potential benefits: improved cost-efficiency for hard-information lending, better risk management, competition, banking system stability gains (e.g., Mexico: improved capitalization, reduced NPLs and OPEX).
  - Potential costs: decreased credit availability to opaque borrowers (SMEs, small firms), reduced branch/deposit outreach, slower credit growth, possible welfare losses for some parameter configurations.
- Policy design should account for country-specific features:
  - In LICs where informational opacity and distance constraints matter, regulatory and institutional reforms to improve property rights, contract enforcement, borrower information (credit registries), and collateral enforcement may be necessary to ensure foreign entry benefits broader financial inclusion.
  - Monitoring of changes in access indicators (branches, deposits per capita, SME lending) is needed following liberalization of foreign entry.
- Evaluation metrics:
  - Relying only on bank cost-efficiency (OH) is insufficient to assess welfare—monitoring-induced increases in OH can coincide with either welfare gains or losses.
  - Use multiple indicators: private credit to GDP, credit growth, access measures, and distributional outcomes across borrower types.

*Italic source: Excerpted content from “5. Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)” (IMF working paper content provided).*

### 16.9 different countries on average, as opposed to 8.4 countries on average for the Asian

### _wp0618 - 16.9 different countries on average, as opposed to 8.4 countries on average for the Asian

### Foreign bank presence and private credit in poor countries
- Empirical finding: In the sample of poor countries, a larger foreign bank penetration is associated with a shallower banking sector and slower credit growth.
- Robustness:
  - Negative association between foreign bank presence and private credit is economically and statistically significant and robust across specifications.
  - Relationship persists in panel estimation with country fixed effects and system GMM to address joint endogeneity of the foreign penetration variable.
  - Instrumental variable estimation in the cross-section yields similar results; IV estimates make the negative coefficient larger in size (suggesting possible attenuation bias in OLS due to measurement error).
- Interpretation: Results are consistent with a cream-skimming mechanism by foreign banks—foreign banks lend mainly to large, transparent firms and governments rather than to smaller, opaque businesses that require local knowledge.
- Relevant numeric values from sample and correlations:
  - Sample consists of 62 low-income and lower-middle-income countries (Table 1).
  - Bivariate correlations table entries: Private Credit mean 18.09; Overhead costs mean 4.87; Private credit standard deviation 15.71; Overhead costs standard deviation 2.80; Private credit minimum 0.86; maximum 93.26; Overhead costs minimum 1.00; maximum 16.23.
  - Correlation notes: Overheads correlated with Private credit at -0.29 with [0.0083] p-value; other correlation entries preserved in Table 1.

### Overhead costs and cost efficiency
- Theoretical expectation: Foreign bank entry may increase overhead costs if entry leads to a switch away from pooling and to monitoring in equilibrium.
- Empirical results:
  - Regressors explain at most 34 percent of the variation in overhead costs (R-squared up to 0.34).
  - Inflation is robustly and positively correlated with overhead costs, implying lower inflation allows banks to improve efficiency.
  - Better availability of information about debtors tends to reduce overhead costs.
  - GDP per capita and the transition dummy do not seem to affect cost efficiency, although they were robustly correlated with depth.
  - Lack of corruption has the expected negative sign but is significant only in some specifications.
  - Banking concentration and country size do not have explanatory power for overhead costs.
  - Surprisingly, LICs with a larger presence of state-owned banks have more cost-efficient banking systems (contrasts with bank-level studies that control for bank size).
- Relationship with foreign banks:
  - The share of foreign banks has a positive sign for overhead costs and is significant in a number of specifications, but loses significance when controlling for the presence of state-owned banks.
  - Using changes in cost efficiency as the dependent variable yields poor fit and no robust relationship with foreign bank presence.

### Financial access indicators and foreign bank presence
- Measures examined: size of the branch network, number of loans, number of deposits (financial access indicators).
- Key observations from Table 8 correlations and regressions (summary in text):
  - Access indicators are not strongly correlated with private credit in the sample, suggesting they capture a separate dimension of financial development.
  - Some access indicators are strongly correlated with one another (e.g., demographic branch and ATM penetration).
  - All access indicators are negatively correlated with the foreign share, and most correlations are large and statistically significant.
  - The negative correlation between access and foreign bank presence remains after controlling for GDP per capita, population density, inflation, and corruption (Table 9).
- Interpretation: Foreign bank presence in poor countries is associated with less access to financial services; smaller, more opaque customers, possibly in remote regions, suffer when foreign banks enter.

### Heterogeneity by income level / Extending the sample
- When the sample is extended to include all 102 countries for which data are available:
  - In regressions for private credit using the full sample, the coefficient on foreign banks becomes much smaller and is no longer significant, although it remains negative.
  - The change reflects that a larger foreign bank presence is associated with more private credit in high-income countries (coefficient not significant).
  - For cost efficiency, there is no significant relationship with foreign bank presence as the sample is extended to include more advanced countries.
- Interpretation: Foreign bank presence appears to have different effects in poor countries versus advanced countries; local knowledge and relationship lending are more important in poor countries where soft information matters.

### Theoretical model and overall conclusions
- Theoretical model summary:
  - Foreign banks have a cost advantage in lending to larger, more transparent firms but not to smaller, more opaque customers.
  - Foreign bank entry can improve total lending, cost efficiency, and welfare in some parameter configurations, but not universally.
  - For some parameter configurations, foreign bank entry results in cream-skimming that hurts opaque businesses, increases operating costs, and lowers aggregate welfare.
- Empirical conclusion:
  - In poor countries, empirical evidence supports the theory of cream-skimming: more foreign bank penetration correlates with shallower banking sectors and slower credit growth.
  - No robust relationships are identified for cost efficiency in the cross-country sample of poor countries.
- Policy implication (implicit in conclusions):
  - Hopes that foreign banks will automatically replace inefficient domestic/state banks and boost financial development in poor countries are not supported by these findings.
  - Concerns about loss of “relationship capital” and availability of credit to SMEs are heightened when large international banks operate in poor countries where cultural and geographic distances are large.

### Key regression coefficient and test values preserved from the source
- Table 2 (selected coefficients, OLS cross-section for poor countries, dependent variable: ratio of bank credit to private sector to GDP):
  - GDP per capita coefficients across columns (1)-(8): 7.04, 8.16, 9.05, 6.99, 5.65, 5.17, 6.38, 6.56 with respective t-statistics in brackets preserved as in source.
  - Foreign ownership coefficients across columns (1)-(8): -6.59, -15.3, -22.56, -19.6, -18.88, -21.5, -23.4, -23.78 with t-statistics preserved in brackets.
  - Transition dummy coefficients in some columns include -14.33, -8.88, -6.99, -5.2, -4.82, -10.02, -8.92 with t-statistics preserved.
  - Inflation coefficients in several columns: -7.02, -5.92, -5.76, -5.53, -4.3, -4.41 with t-statistics preserved.
  - Lack of corruption coefficients include 10.54, 8.45, 9.42, 9.84, 8.92 with t-statistics preserved.
  - Creditor information coefficients shown include 1.48, 1.46, 0.8, 0.84 with t-statistics preserved.
  - Enforcement speed coefficients include 423.6, 411.41, 367.52, 392.44 with t-statistics preserved.
  - State banks coefficients include -5.42, -11.64, -12.76 in specified columns with t-statistics preserved.
  - Observations reported: 62, 62, 61, 61, 59, 59, 45, 45; R-squared values reported: 0.21, 0.34, 0.56, 0.61, 0.63, 0.64, 0.66, 0.67 (all preserved exactly).

- Table 1 sample and summary statistics preserved exactly as presented (see above for selected entries).

*Source: Excerpt from IMF Working Paper _wp0618 (chapter/section content as provided).*

### 2002. Initial depth is the ratio of bank credit to the private sector measured in 1995-98. Foreign ownership is the

### _wp0618 - 2002. Initial depth is the ratio of bank credit to the private sector measured in 1995-98. Foreign ownership is the

### Panel estimation: Financial Depth and Foreign Bank Presence in Poor Countries (Table 5)
- Dependent variable: ratio of bank credit to the private sector to GDP.
- Key regressors and coefficient estimates (selected columns shown):
  - Initial depth: -0.01; -0.02; -0.02; -0.02; -0.02; -0.02
    - Robust t statistics: [3.38]***, [4.57]***, [4.83]***, [5.12]***, [3.84]***, [4.49]***
  - Foreign ownership: -0.55; -0.56; -0.68; -0.52; -0.58; -0.54
    - Robust t statistics: [2.29]**, [2.42]**, [2.77]***, [2.05]**, [2.12]**, [2.33]**
  - Inflation: -0.09; -0.08; -0.1; -0.11; -0.08; -0.12
    - Robust t statistics: [2.13]**, [1.98]*, [2.68]***, [2.87]***, [1.84]*, [2.42]**
  - State banks: -0.35; -0.4; -0.38; -0.37; -0.37; -0.43
    - Reported t-stat bracketed lines include [1.73]*, [2.05]**, [1.85]*, [1.80]*, [1.62], [2.26]**
  - Lack of corruption (included in later columns): 0.3; 0.33; 0.34; 0.32; 0.29
    - t-statistics: [2.44]**, [2.69]***, [2.87]***, [2.01]*, [2.37]**
  - Banking crisis 1990-93 and 1994-2002 appear in some specifications:
    - Banking crisis 1994-2002: 0.16; 0.13 with t-stat [2.01]**, [1.52]
  - Africa dummy in one specification: -0.2 with t-stat [1.76]*
  - Change in mortality: -0.01 with t-stat [0.06]
  - Hyperinflation: 0.2 with t-stat [0.88]
- Sample sizes and fit:
  - Observations: 59; 59; 59; 59; 50; 59 (across columns)
  - R-squared: 0.23; 0.29; 0.34; 0.38; 0.29; 0.3
- Note: Robust t statistics in brackets. Significance: * 10 percent; ** 5 percent; *** 1 percent.

### Panel A (Fixed effects) and Panel B (System GMM) summary (Table 5 continuation)
- Panel A (Fixed effects regressions):
  - Lagged dependent variable coefficients (selected): 0.47; 0.53; 0.5; 0.48
    - t-stats: [9.90]***, [10.26]***, [9.22]***, [8.72]***
  - Foreign Ownership coefficients in Panel A: -0.68; -0.3; -0.38; -0.42; -0.3
    - t-stats: [3.55]***, [1.78]*, [2.15]**, [2.35]**, [1.43]
  - Log(inflation): -0.1; -0.1; -0.09 (t-stats: [5.44]***, [4.84]***, [4.79]***)
  - GDP per capita in some fixed-effects specs: 0.44; 0.48 (t-stats reported [1.69]*, [1.82]*)
  - Observations: 422; 419; 392; 392; 392
  - Country fixed effects: yes; Year fixed effects: yes in reported columns
  - R-squared: 0.92; 0.94; 0.94; 0.94; 0.94
- Panel B (System GMM):
  - Lagged dependent variable coefficients: 0.87; 0.88; 0.87
    - t-stats: [32.09]***, [29.68]***, [23.69]***
  - Foreign Ownership: -0.44; -0.24; -0.39
    - t-stats: [2.04]**, [1.64]*, [1.67]*
  - Log(inflation): -0.13; -0.12; -0.14 (t-stats: [4.10]***, [3.71]***, [4.46]***)
  - State banks coefficient reported: -0.24 (t-stat [1.99]*)
  - Observations: 392; 392; 392
  - Sargan test (p value) reported: 0.341; 0.836; 0.991 (across columns)
  - Serial correlation p-values (order 1 and order 2) reported; order 1 values include 0.000, 0.02, 0.44, 0.46, 0.38 in some fixed-effects columns; Panel B order 1 values: 0.039, 0.047, 0.042; order 2 values include 0.616, 0.665, 0.75.
- Note: Two-step system GMM uses Windmeijer (2005) finite sample correction with up to three lags for instruments. t statistics in brackets; significance levels as above.

### Instrumental Variables estimation (Table 6)
- Approach: 2SLS estimation; dependent variable: ratio of bank credit to the private sector to GDP. Instruments: log population, share Muslim, share Buddhist, share Christian Orthodox (and others in different specifications).
- Second-stage foreign ownership coefficients (first row listed under "Second stage"):
  - Foreign ownership: -21.37; -20.89; -24.79; -21.57; -28.54
    - First-stage related coefficient on log population: -0.048; -0.047; -0.047; -0.048; -0.024 with [1.80]*, [1.73]*, [1.87]*, [1.80]*, [1.71]* respectively.
- Other second-stage coefficients (selected):
  - Transition: -5.56; -6.85; -8.18; -6.66 (first-stage / instrument lines show associated statistics)
  - GDP per capita: 5.17; 6.64; 8.09; 6.59; 5.48 (t-stats: [3.24]***, [4.62]***, [6.00]***, [4.69]***, [2.28]**)
  - Inflation: -5.79; -5.94; -6.76; -5.93; -6.7 (t-stats across columns: [4.30]***, [4.27]***, [4.50]***, [4.23]***, [2.91]***)
  - Lack of corruption: 5.36; 7.88; 7.83 (reported t-stats include [1.64], [2.38]**, [2.34]**)
  - Enforcement speed reported in one specification: 504.43 with [3.68]***; another: 521.34 with [3.32]***.
  - Creditor information: 1.45 with [2.25]**, and 1.34 with [1.96]* in another specification.
- Diagnostics and fit:
  - Partial R-squared: 0.25; 0.26; 0.27; 0.27; 0.24 (across columns)
  - F statistic: 9.26; 13.56; 11.85; 13.31; 7.92
  - p-value of F test: 0.0000; 0.0000; 0.0000; 0.0000; 0.0005
  - Hansen J statistic and p-values: reported (e.g., Hansen J stat 4.29 p-val 0.23; 2.31 p-val 0.51; 1.71 p-val 0.64; 2.35 p-val 0.5; 1.29 p-val 0.52)
  - 90 percent confidence intervals robust to weak instruments (Moreira, 2003) for Foreign ownership: [-25.3 , -21.1]; [-23.7 , -19.7]; [-27.6 , -23.7]; [-24.5 , -20.6]; [-33.3, -28.0]
- Observations: 57; 59; 59; 59; 39 (across columns)
- R-squared: 0.58; 0.55; 0.51; 0.55; 0.58
- Robust z statistics in brackets. Significance levels: * 10 percent; ** 5 percent; *** 1 percent.

### Bank cost efficiency and foreign bank presence (Table 7)
- Dependent variable: ratio of bank operating costs to assets (aggregated bank-level).
- Selected coefficients (across specifications 1–8):
  - GDP per capita: -0.002; -0.009; -0.012; 0.007; 0.024; 0.01; 0.007; 0.006
    - t-stats: [0.07], [0.40], [0.57], [0.29], [0.93], [0.39], [0.21], [0.18]
  - Foreign ownership: 0.02; 0.027*; 0.032**; 0.029*; 0.024*; 0.01; 0.015; 0.013
    - t-stats: [1.34], [1.71], [2.21], [1.89], [1.76], [0.72], [0.89], [0.70]
  - Transition: reported in some columns with 0.013* and other small negative values; t-stats include [1.75], [0.23], [0.45], [1.18], [1.39], [1.11], [1.19]
  - Inflation: 0.007***; 0.006**; 0.007**; 0.008***; 0.008***; 0.008*** (t-stats: [2.69], [2.26], [2.59], [3.03], [3.00], [2.96])
  - Lack of corruption: -0.016**; -0.016*; -0.011; -0.009; -0.009 (t-stats: [2.01], [1.79], [1.55], [0.98], [0.97])
  - Creditor information: -0.003*; -0.004*; -0.004**; -0.004** (t-stats: [1.76], [1.99], [2.07], [2.14])
  - State banks: -0.028**; -0.025**; -0.022 (t-stats: [2.58], [2.18], [1.61])
  - Concentration and country size reported in later columns with non-significant coefficients (e.g., concentration -0.007, -0.012 with t-stats [0.35], [0.55]; country size (pop.) -0.002 [0.39])
- Observations and fit:
  - Observations: 63; 63; 62; 62; 60; 60; 58; 58
  - R-squared: 0.03; 0.07; 0.17; 0.21; 0.26; 0.34; 0.34; 0.34
- Robust t statistics in brackets. Significance levels: * 10 percent; ** 5 percent; *** 1 percent.

### Bivariate correlations: Access to financial services and foreign bank presence (Table 8)
- Variables: Foreign Ownership; Private Credit; Demographic branch penetration; Geographic branch penetration; Demographic ATM penetration; Geographic ATM penetration; Loan accounts per capita; Deposit accounts per capita.
- Selected correlation coefficients and p-values:
  - Correlation between Foreign Ownership and Demographic branch penetration: -0.3969 (P-value 0.0124)
  - Foreign Ownership and Geographic branch penetration: -0.413 (P-value 0.009)
  - Foreign Ownership and Demographic ATM penetration: -0.2868 (P-value 0.1001)
  - Foreign Ownership and Geographic ATM penetration: -0.3777 (P-value 0.0277)
  - Foreign Ownership and Loans per capita: -0.3607 (P-value 0.1414)
  - Foreign Ownership and Deposits per capita: -0.4642 (P-value 0.0223)
  - Correlations among access indicators (examples):
    - Demographic branch penetration and Geographic branch penetration: 0.3592 (P-value 0.0195)
    - Demographic branch penetration and Demographic ATM penetration: 0.6962 (P-value 0)
    - Geographic branch penetration and Geographic ATM penetration: 0.3643 (P-value 0.0314)
    - Loan accounts per capita correlations with some access measures: ranges reported (e.g., Loans per capita and Geographic ATM penetration 0.4222 with P-value 0.0637)
- Observations: not explicitly listed in table body; p-values presented inline.

### Access to financial services: Cross-sectional OLS regressions (Table 9)
- Dependent variables: demographic branch penetration; geographic branch penetration; loan accounts per capita; deposit accounts per capita (alternative specifications).
- Selected coefficients:
  - GDP per capita:
    - Demographic branch penetration columns: 2.33 [4.72]***; 2.13 [4.89]***; -1.28 [0.70]
    - Loan accounts per capita columns: 30.42 [1.91]*; 18.33 [1.43]; 28.08 [2.00]*
    - Deposit accounts per capita columns: 337.85 [4.06]***; 252.12 [2.96]***; 359.53 [3.01]***
  - Population density:
    - Demographic branch penetration: 0 [1.45]; 0 [1.92]*; 0.02 [1.22] in some specs
    - Loan/deposit accounts: values small, some significance marginal
  - Foreign ownership:
    - Demographic branch penetration: -5.34 [3.10]***; -4.6 [2.92]***; -13.34 [2.38]**
    - Geographic branch penetration: -12.99 [2.50]**; -15.14 [2.52]**; -13.34 [2.38]**
    - Loan accounts per capita: -91.42 [1.42]; -143.35 [1.68]; -77.83 [1.38]
    - Deposit accounts per capita: -926.77 [3.12]***; -601 [2.52]**; -892.7 [3.23]***
  - Inflation and lack of corruption appear intermittently (e.g., inflation 7.56 [0.52] in some loan/account specs; inflation 193.54 [2.35]** in deposit accounts specification)
- Observations and fit:
  - Demographic/geographic branch regressions: Observations 39; R-squared 0.47; 0.49; 0.44; 0.42; 0.46; 0.44 (across columns)
  - Loan accounts regressions: Observations 18; R-squared 0.32; 0.41; 0.33
  - Deposit accounts regressions: Observations 24; R-squared 0.58; 0.68; 0.58
- Robust t statistics in brackets. Significance levels: * 10 percent; ** 5 percent; *** 1 percent.

### Cross-country OLS including all countries: Financial Depth, Cost Efficiency, and Foreign Bank Presence (Table 10)
- Two dependent variables panels: Private Credit and Overhead Costs; subgroups: Low-Income and Lower-Middle-Income; All Countries; High-Income and Upper-Middle-Income.
- Selected coefficients (Private Credit panel):
  - GDP per capita:
    - Low-Income and Lower-Middle-Income: 5.65 [3.40]***
    - All Countries: 5.84 [2.32]**
    - High-Income and Upper-Middle-Income: 22.33 [1.49]
  - Foreign ownership:
    - Low-Income and Lower-Middle-Income: -18.88 [3.09]***
    - All Countries: -3.98 [0.32]
    - High-Income and Upper-Middle-Income: 24.46 [0.76]
  - Inflation:
    - Private Credit: -5.76 [4.13]***; -5.56 [2.30]**; -3.94 [0.57]
  - Lack of corruption:
    - Private Credit: 8.45 [2.00]*; 12.23 [2.50]**; 5.45 [0.60]
  - Creditor information: 1.48 [2.34]**; 1.96 [1.58]; 3.79 [1.14]
  - Enforcement speed: 423.6 [2.78]***; 724.48 [1.65]; 351.28 [0.31]
  - State banks: -0.015*; -0.023**; 0.016 (reported t-stats [1.79], [2.33], [0.84])
- Overhead Costs panel (selected):
  - GDP per capita: 0.003 [1.17]; 0.003 [0.79]; -0.003 [0.58]
  - Foreign ownership: 0.007 [0.77]; 0.017 [1.46]; -0.003 [0.29]
  - Inflation: 0.012***; 0.015***; 0.006 (t-stats [5.20], [5.00], [1.37])
  - Creditor information: -0.003***; -0.005***; 0.002 (t-stats [2.78], [3.29], [1.25])
  - Enforcement speed: -0.381*; -0.372; -0.206 (t-stats [1.79], [1.59], [0.49])
  - State banks: reported significance in some columns [1.79], [2.33], [0.84]
- Observations and R-squared:
  - Private Credit: Observations 59 (Low-Income and Lower-Middle-Income), 102 (All Countries), 43 (High-Income and Upper-Middle-Income); R-squared 0.63; 0.71; 0.55
  - Overhead Costs: Observations 101; 57; 44; R-squared 0.5; 0.52; 0.44

### Figures: conceptual and empirical visuals (Figures 1–5)
- Figure 1: Impact of Foreign Bank Entry on Welfare and Aggregate Lending
  - Describes welfare as a function of monitoring cost H_c; pooling equilibrium welfare does not depend on H_c; other equilibria welfare increases linearly with H_c. Labeled equilibria A, B, C, D with descriptions: in many cases "Lending falls" when switching away from pooling; in some cases "Lending does not change."
- Figure 2: Bank Credit to the Private Sector in Lower-Income Countries by Region (Percent of GDP)
  - Depicts cross-country percent of GDP values for regions: Sub-Saharan Africa, Latin America, Middle East and North Africa, Europe and Central Asia, South and East Asia; country names listed and axis 0–90 percent scale.
- Figure 3: Foreign Bank Presence by Region in Selected Lower-Income Countries
  - Shows foreign bank penetration (scale 0–0.8) across regions and selected countries; source: Micco, Panizza, and Yanez (2004).
- Figure 4: Private Credit to GDP and Foreign Bank Presence
  - Scatter of countries (list provided) plotting Foreign Bank Presence (x-axis) versus Private Credit to GDP (y-axis); fitted values line included.
- Figure 5: Change in Private Credit to GDP and Change in Foreign Bank Presence (1995-2002)
  - Scatter of country changes (1995-2002) in Foreign Bank Presence (x-axis) and Change in Private Credit to GDP (y-axis); fitted values line included.

*Source: IMF Working Paper content provided in the supplied PDF excerpts.*

### Appendix I. Welfare Comparison under Alternative Equilibria

### Appendix I. Welfare Comparison under Alternative Equilibria

### Welfare expressions by equilibrium
- Pooling equilibrium (A):
  - S = pR + μA(1 − μB) + μB? (source expression: () 1 )( −⋅++=RpS BSHA μμμ)  
    (Exact source expression: () 1 )( −⋅++=RpS BSHA μμμ)
- Separating equilibrium (B):
  - S = cS + cH + R S + ... (source expression: ()()() SSHHSHB ccRS+⋅−+⋅−⋅+=11 )( μμμμμ)
    (Exact source expression: ()()() SSHHSHB ccRS+⋅−+⋅−⋅+=11 )( μμμμμ)
- Semi-separating equilibrium (C):
  - S = cH + R p S + μH μB ... (source expression: () () HHBSHC cRpSμμμμ−−⋅++=1)
    (Exact source expression: () () HHBSHC cRpSμμμμ−−⋅++=1)
- Credit-constrained equilibrium (D):
  - S = cH + R S + ... (source expression: ()() HHD cRS+−⋅=1 )( μ)
    (Exact source expression: ()() HHD cRS+−⋅=1 )( μ)

(Note: expressions above are reproduced exactly as in the source text where symbols and ordering appear.)

### Comparison: Pooling (A) vs Separating (B)
- Difference between expected net output in the separating and pooling equilibrium:
  - () 4434421 43421 0 0 )()( 1 < > −−−⋅=− SSHHBAB ccpRSSμμμ .  
    - Interpretation in source: first term = welfare gain from not financing type B anymore; second term = welfare cost of monitoring soft and hard information. Overall effect ambiguous.
- Special case when PH rc = +1:
  - () 4434421 44443444421 0 0 )()( 1 )( < < =+ −−⋅− ++ =∆=− HSSB BSH B BA rc B ccRp p p SSS PH μμ μμμ μ
    - Source note: first term is negative because BSH P p r R μμμ++ => 1.

### Comparison: Pooling (A) vs Semi-separating (C)
- Difference:
  - 0 )()()( ≤−=∆=− HHCAC cSSSμ
    - Source conclusion: welfare falls because all projects are financed, but monitoring costs are paid in equilibrium.

### Comparison: Pooling (A) vs Credit-constrained (D)
- For PH rc = +1:
  - ()() )()()( 1 )( APBSDA rc D SrRpSSS PH <−⋅+−=∆=− =+ μμ
- For 0 = H c :
  - ()() )()(
    0 )( ˆ APBSA c D SrRpSSS H >−⋅+−= = μμ
    - Note: P r ˆ is the interest rate charged by banks pooling type S and type B together; in this case the interest rate pooling types )(B and )(A is too high to sustain such a pooling equilibrium (R r P > ˆ).

### Comparison: Separating (B) vs Semi-separating (C)
- Difference:
  - ()() BSSSBCB cRpSSμμμμ+++⋅−−=−1 )()( 
- Condition and conclusion:
  - () S BS SCB pR cSS μ μμ −+ < + ⇔ > − 1 10 )()(  
    - Source conclusion: welfare in the separating equilibrium is always higher than welfare in the semi-pooling equilibrium whenever the separating equilibrium is feasible.

*Source: _wp0618 - Appendix I. Welfare Comparison under Alternative Equilibria*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2006/_wp0618.pdf_