## wpiea2021149-print-pdf

## Source details

**Canonical URL:** [wpiea2021149-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021149-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021149-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021149-print-pdf.pdf.json)

---

### I. INTRODUCTION — competition and banking sector relevance
- Competition linked to secular trends in corporate market power and labor share (Diez, Leigh, and Tambunlertchai, 2018; IMF, 2019; Syverson, 2019; Autor et al., 2017; Aghion et al., 2019; De Loecker, Eeckhout, and Unger, 2020).
- Financial-sector competition matters because banks:
  - mobilize savings, allocate credit, and provide payment services;
  - enable firms and households to invest and cope with uncertainties;
  - transmit monetary policy, so changes in banking competition can affect policy effectiveness.
- Trade-offs of banking competition:
  - Benefits: lower cost of services, improved efficiency, better credit allocation.
  - Risks: erosion of franchise value, increased risk taking, threats to financial stability, weaker incentives for relationship lending, reduced access for SMEs.
- Post-GFC structural forces ambiguous for banking market power:
  - Consolidation and higher compliance fixed costs could increase market power.
  - Greater non-bank intermediation and fintech could reduce banks’ market power.
- Empirical focus:
  - Document evolution of competition measures (primary focus: Lerner index) between 2000 and 2017 in 67 countries (27 AEs and 40 EMDEs).
  - Key preview: Lerner index rises significantly for AEs, especially after the GFC; other indicators (market concentration, profitability) are more muted.

### II. THE LERNER INDEX: definition and relation to monetary policy
- Lerner index (textbook): Lerner = (p − mc) / p = 1 − mc / p, where p = revenue / Q_b,t and mc = ∂C_b,t / ∂Q_b,t = ε_b,t * C_b,t / Q_b,t, with ε_b,t the elasticity of costs to quantity.
- Cost function and elasticity estimation (country-specific OLS trans-log):
  - log C_b,t = α log Q_b,t + δ/2 (log Q_b,t)^2 + Σ_j β_j w_j,b,t log Q_b,t + Σ_j σ_j w_j + Σ_jk σ_jk w_j,b,t w_k,b,t + γ X_b,t + μ_b + π_t + ε_b,t
  - w_j are bank-specific input costs (logs): total interest expenses over deposits, personnel expenses over assets, other operating expenses over assets.
  - X_b,t includes capitalization (equity over assets), focus on lending (loans to assets), loan quality (NPLs over loans).
  - ε_b,t = α + δ log Q_b,t + Σ_j β_j w_j,b,t
- Lerner reformulation: Lerner = 1 − ε_b,t * θ_b,t where θ = Costs / Income; with ε_b,t fixed (or slow-moving), Lerner increases when cost-to-income ratio decreases.
- NIM relation:
  - NIM = Net Interest Income / Average Earning Assets = R_l − R_d, where R_l = interest earned on assets and R_d = interest paid on liabilities.
  - Lerner depends on ratio R_d / R_l (under simplifying assumptions); NIM depends on difference R_l − R_d. They can move differently when interest rates change.
- Low-rate environment caveat:
  - When interest rates are close to or below zero, Lerner becomes uninformative because R_d / R_l mechanically moves toward zero and Lerner approaches one.
- Numerical example reported (median, AEs):
  - Interest paid declined from approximately 2 percentage points (pp) to 0.5 pp.
  - Interest earned declined from 5.8 pp to 3.8 pp.
  - NIM of a hypothetical bank would decrease from 3.8 pp to 3.3 pp.
  - Ratio R_d / R_l would go from 0.34 to 0.13.
  - Therefore, NIM decreases while Lerner would increase substantially.

### III. EMPIRICAL ANALYSIS — data construction and key findings
- Data construction:
  - Bank-level information primarily from Fitch Connect; end-year unconsolidated statements used where available, consolidated otherwise (consolidated statements account for about 40% of observations).
  - Exclude banks with missing/zero operating income or missing key variables (assets, costs, operating expense, income over assets, interest expenses over deposits, wages over assets).
  - Winsorize variables at the 2.5 and 97.5 percentiles (by year).
  - Compute bank-level Lerner index per Section II; aggregate to country level using asset-weighted averages.
  - Sample restrictions for country-level indices: balanced bank sample and at least 10 banks per country-year.
  - Complement bank data with macro indicators (real GDP growth and monetary policy rates) from IMF WEO and IFS.
- Main empirical findings:
  - Traditional Lerner:
    - Marked increase in Lerner since early 2000s, with temporary decline around late 2000s.
    - Steep rise particularly observable in AEs after the GFC; EMDEs oscillate around similar level since mid-2000s.
  - Other indicators contrast with Lerner:
    - Market concentration and profitability indicators do not show the same upward trend.
    - Market concentration broadly downward; profits declined after GFC and income remained relatively flat, especially in AEs.
  - Role of monetary policy (regression of Lerner on policy rates and GDP growth, controlling for bank fixed effects):
    - In AEs, a one standard deviation (SD) decrease in the policy rate leads to a 0.05 higher Lerner index over a two-year period.
    - In EMDEs, a one SD decrease in the policy rate leads to a 0.02 higher Lerner index.
    - After the GFC (AEs), one SD lower policy rate → 0.06 higher Lerner index.
    - A one SD increase in GDP growth → 0.04 increase in Lerner (post-GFC, AEs).
    - Economic significance: policy-rate effect in AEs represents about a fifth of the average Lerner index and more than a third of its country-level standard deviation.
  - Adjusted Lerner:
    - Using residuals from the regressions (purging policy-rate and cyclical factors) yields a “statistically-adjusted” Lerner.
    - Adjusted Lerner shows market power has been fairly constant in AEs.
    - In EMDEs, traditional and adjusted Lerner more closely co-move, consistent with smaller estimated policy-rate coefficients.

### IV. HETEROGENEITY AND ROBUSTNESS NOTES
- Heterogeneity:
  - Increase in unadjusted Lerner among AEs is common across bank types: commercial vs investment, large vs small banks.
  - Similar unadjusted Lerner increases in countries with many M&As and those without; market concentration increased only in high M&A countries—suggesting common macro drivers (policy rate, cyclical factors) could be driving Lerner increases rather than structural competition changes.
- Robustness:
  - Imposing additional structure on the cost function (e.g., homogeneity of degree one in input prices) does not significantly affect results.
  - Operating expenses and non-interest income are important but less affected by monetary policy; these are included in estimations though ignored in illustrative algebra.
- Denominator effect and alternative adjustment:
  - Expansionary monetary policy lowers both interest earned and interest paid; because Lerner depends on their ratio, very low rates produce a “denominator effect” that can mechanically raise Lerner.
  - Alternative Lerner (subtract interest expenses from both numerator and denominator) — price p_b,t = (revenues − interest expenses) / assets; costs C_b,t = (total expenses − interest expenses) — does not grow over time, confirming the unadjusted Lerner rise is caused by the “denominator effect”.
  - Alternative Lerner appears impacted by business cycle fluctuations, whereas the statistical adjustment explicitly controlling for economic growth does not.

### V. POLICY IMPLICATIONS AND CONCLUDING REMARKS
- Measurement and monitoring:
  - Traditional Lerner can be misleading in low-interest-rate environments and when comparing across policy regimes.
  - Supervisors and policymakers should rely on a range of indicators and adopt a holistic approach in monitoring banking-sector competition and market power.
- Interpretation:
  - Part of the post-GFC rise in the Lerner index for AEs appears driven by low policy rates rather than a pure increase in market power; adjusted measures that control for policy rates show little secular increase in market power in AEs.
- Uncertainty and forward-looking considerations:
  - If rates stay at the ZLB for long, pressure on profits may intensify and possibly trigger consolidation in the banking sector.
  - Structural challenges such as the rise of fintech could affect banks’ ability to exert market power.
  - At least so far, there is little indication that banks’ profitability has been adversely affected in a low interest rate environment.
- Monitoring recommendation:
  - Policymakers should monitor behavior of the Lerner index at the zero lower bound (ZLB).

### KEY STATISTICS AND REGRESSION RESULTS
- Table 1A (bank-level summary statistics):
  - Lerner Index: # obs 312,566; Mean 19.9130; Std. Dev. 67.96
  - NIM: # obs 312,566; Mean 3.69; Std. Dev. 2.1; Min -0.24; Max 19.64
  - ROE: # obs 312,566; Mean 6.51; Std. Dev. 12.4; Min -160.37; Max 41.48
  - ROA: # obs 312,566; Mean 0.71; Std. Dev. 1.3; Min -8.37; Max 7.8
  - Cost to income: # obs 312,566; Mean 82.9; Std. Dev. 17.3; Min 0; Max 169.8
- Table 1B (country-level summary statistics):
  - Lerner, traditional: # obs 1,108; Mean 0.24; Std. Dev. 0.13; Min 0.00; Max 0.66
  - Lerner, adjusted: # obs 1,108; Mean 0.00; Std. Dev. 0.06; Min -0.23; Max 0.26
  - HHI: # obs 1,107; Mean 1427; Std. Dev. 906; Min 141; Max 7,564
  - Real GDP growth: # obs 1,108; Mean 0.03; Std. Dev. 0.03; Min -0.06; Max 0.10
  - Policy rate: # obs 1,108; Mean 0.05; Std. Dev. 0.04; Min -0.01; Max 0.16
- Table 2 (Cyclical Drivers of the Lerner Index — selected coefficients and statistics):
  - Column (1) Whole period: Policy rate = -0.87*** (p-value (0.00)); GDP growth = 0.92*** (p-value (0.00)); Observations 1,108; R-squared 0.77; # countries 67; Country FEs Yes; Within R-squared 0.17
  - AE vs EMDE effects:
    - Policy rate - AEs: -2.75*** (p-value (0.00)); Growth rate - AEs: 1.50*** (p-value (0.00))
    - Policy rate - EMDEs: -0.44** (p-value (0.00)); Growth rate - EMDEs: 0.97*** (p-value (0.00))
  - Pre-GFC and post-GFC regressions (2010 first post-GFC year): Observations 582 (pre-GFC) and 525 (post-GFC); R-squared 0.86 and 0.88 respectively.

### SAMPLE COMPOSITION AND HETEROGENEITY
- Sample covers 67 countries (27 advanced economies and 40 emerging markets and developing economies).
- Bank counts:
  - Advanced economies total banks: 21,043 (examples: United States 13,281; Germany 2,501; Italy 934; Japan 823)
  - EMDEs total banks: 3,734 (examples: Russian Federation 1,136; India 315; Brazil 200)
- Country-level observation counts for panels:
  - Advanced economies sample total obs: 249 (typically 18 per country; exceptions shown like Singapore 11)
  - EMDEs sample total obs: 693 (typically 18 per country; exceptions included)
- Appendix figures document heterogeneity across bank groups, M&A intensity, and regional patterns (Appendix Figure A1 and Appendix Figure A2: Panel C shows adjusted Lerner flattens more for EU than US).

*Source: wpiea2021149-print-pdf - References .............................................................................................................*

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

### wpiea2021149-print-pdf - References .............................................................................................................

### I. INTRODUCTION — competition and banking sector relevance
- Competition has been linked to secular trends: rise in corporations’ market power measured by markups and profitability (Diez, Leigh, and Tambunlertchai, 2018; IMF, 2019; Syverson, 2019) and related to the decline in the labor share (Autor et al., 2017; Aghion et al., 2019) with implications for wage stagnation and migration (De Loecker, Eeckhout, and Unger, 2020).
- Financial-sector competition matters for the real economy because banks:
  - mobilize savings, allocate credit, and provide payment services;
  - enable firms and households to invest and cope with uncertainties;
  - transmit monetary policy, so changes in banking competition can affect policy effectiveness.
- Competition in banking has trade-offs:
  - Benefits: lower cost of services, improved efficiency, better credit allocation.
  - Risks: erosion of franchise value, increased risk taking, threats to financial stability, weaker incentives for relationship lending, reduced access for SMEs.
- Post-GFC structural forces had ambiguous effects on banking market power:
  - Consolidation and higher compliance fixed costs could increase market power.
  - Greater non-bank intermediation and fintech could reduce banks’ market power.
- Empirical focus of the paper:
  - Document evolution of competition measures (primary focus: Lerner index) between 2000 and 2017 in 67 countries (27 AEs and 40 EMDEs).
  - Key finding preview: Lerner index rises significantly for AEs, especially after the GFC; other indicators (market concentration, profitability) are more muted.

### II. THE LERNER INDEX: definition and relation to monetary policy
- Lerner index definition (textbook): Lerner = (p − mc) / p = 1 − mc / p, where p is ratio of revenue to quantity Q_b,t (assets) and mc = ∂C_b,t / ∂Q_b,t = ε_b,t * C_b,t / Q_b,t, with ε_b,t the elasticity of costs to quantity.
- Cost elasticity estimated from a trans-log cost function (country-specific OLS):
  - log C_b,t = α log Q_b,t + δ/2 (log Q_b,t)^2 + Σ_j β_j w_j,b,t log Q_b,t + Σ_j σ_j w_j + Σ_jk σ_jk w_j,b,t w_k,b,t + γ X_b,t + μ_b + π_t + ε_b,t
  - where w_j are bank-specific input costs (logs): total interest expenses over deposits, personnel expenses over assets, other operating expenses over assets;
  - X_b,t includes capitalization (equity over assets), focus on lending (loans to assets), loan quality (NPLs over loans);
  - μ_b and π_t are bank and year fixed effects.
- Elasticity formula: ε_b,t = α + δ log Q_b,t + Σ_j β_j w_j,b,t
- Lerner index reformulation: Lerner = 1 − ε_b,t * θ_b,t where θ = Costs / Income.
  - Therefore, with ε_b,t fixed (or slow-moving), Lerner increases when cost-to-income ratio decreases.
- Net Interest Margin (NIM) relation:
  - NIM = Net Interest Income / Average Earning Assets = R_l − R_d, where R_l is interest earned on assets (gross interest income over assets) and R_d is interest paid on liabilities (interest expenses over assets).
  - Lerner depends on ratio R_d / R_l (under simplifying assumptions); NIM depends on difference R_l − R_d.
  - Hence Lerner and NIM can move in different directions when interest rates change.
- Low-rate environment caveat:
  - When interest rates are close to or below zero, Lerner becomes uninformative because R_d / R_l mechanically moves toward zero and Lerner approaches one.
  - When R_d approaches 0 and banks charge positive rates to borrowers, R_d / R_l → 0, making the Lerner index uninformative.
- Numerical example reported:
  - Median interest paid (AEs) declined from approximately 2 percentage points (pp) to 0.5 pp.
  - Interest earned declined from 5.8 pp to 3.8 pp.
  - NIM of a hypothetical bank would decrease from 3.8 pp to 3.3 pp.
  - Ratio R_d / R_l would go from 0.34 to 0.13.
  - Therefore, NIM decreases while Lerner would increase substantially.

### III. EMPIRICAL ANALYSIS — data construction and key findings
- Data construction:
  - Bank-level information primarily from Fitch Connect; end-year unconsolidated statements used where available, consolidated otherwise (consolidated statements account for about 40% of observations).
  - Exclude banks with missing/zero operating income or missing key variables (assets, costs, operating expense, income over assets, interest expenses over deposits, wages over assets).
  - Winsorize variables at the 2.5 and 97.5 percentiles (by year).
  - Compute bank-level Lerner index per Section II; aggregate to country level using asset-weighted averages.
  - Sample restrictions for country-level indices: balanced bank sample and at least 10 banks per country-year.
  - Complement bank data with macro indicators (real GDP growth and monetary policy rates) from IMF WEO and IFS.
- Main empirical findings:
  - Traditional Lerner:
    - Marked increase in Lerner since early 2000s, with temporary decline around late 2000s.
    - Steep rise particularly observable in AEs after the GFC; EMDEs show oscillation around similar level since mid-2000s.
  - Other competition indicators contrast with Lerner:
    - Market concentration and profitability indicators do not show the same upward trend.
    - Market concentration broadly downward; profits declined after GFC and income remained relatively flat, especially in AEs.
  - Role of monetary policy:
    - Decline in policy rates after the GFC reduced bank interest expenses; interest income on loans declined almost in tandem in AEs, explaining fairly flat NIMs.
    - Regression analysis (Lerner on policy rates and GDP growth controlling for bank fixed effects):
      - In AEs, a one standard deviation (SD) decrease in the policy rate leads to a 0.05 higher Lerner index over a two-year period.
      - In EMDEs, a one SD decrease in the policy rate leads to a 0.02 higher Lerner index.
      - After the GFC (AEs), one SD lower policy rate → 0.06 higher Lerner index.
      - A one SD increase in GDP growth → 0.04 increase in Lerner (post-GFC, AEs).
    - Economic significance: policy-rate effect in AEs represents about a fifth of the average Lerner index and more than a third of its country-level standard deviation.
  - Adjusted Lerner:
    - Using residuals from the regressions (purging policy-rate and cyclical factors) yields a “statistically-adjusted” Lerner.
    - Adjusted Lerner shows market power has been fairly constant in AEs.
    - In EMDEs, traditional and adjusted Lerner more closely co-move, consistent with smaller estimated policy-rate coefficients.

### IV. Heterogeneity and robustness notes
- Heterogeneity:
  - Increase in unadjusted Lerner among AEs is common across bank types: commercial vs investment, large vs small banks.
  - Similar unadjusted Lerner increases in countries with many M&As and those without; however, market concentration increased only in high M&A countries—suggesting common macro drivers (policy rate, cyclical factors) could be driving Lerner increases rather than structural competition changes.
- Robustness:
  - Imposing additional structure on the cost function (e.g., homogeneity of degree one in input prices) does not significantly affect results.
  - Operating expenses and non-interest income are important but less affected by monetary policy; these are included in estimations though ignored in illustrative algebra.

### V. Policy implications
- Measurement and monitoring:
  - The traditional Lerner index can be misleading in low-interest-rate environments and when comparing across policy regimes.
  - Supervisors and policymakers should rely on a range of indicators and adopt a holistic approach in monitoring banking-sector competition and market power.
- Interpretation:
  - Part of the post-GFC rise in the Lerner index for AEs appears driven by low policy rates rather than a pure increase in market power; adjusted measures that control for policy rates show little secular increase in market power in AEs.

*Source: wpiea2021149-print-pdf - References .............................................................................................................*

### Appendix Figure A2). However, the adjusted Lerner flattens more for European countries

### Appendix Figure A2). However, the adjusted Lerner flattens more for European countries

### Denominator effect and alternative adjustment procedure
- Expansionary monetary policy exerts downward pressure on both the interest rate earned on assets and the interest paid on liabilities.
- Different margin measures react differently: NIM = interest earned − interest paid; Lerner index depends on the ratio between the two.
- When interest rates are very low, the two measures can move in different directions because the ratio becomes uninformative as the funding rate approaches zero.
- Alternative adjustment: subtract interest expenses from both numerator and denominator rather than including interest expenses in the denominator.
  - Price p_b,t computed as (revenues − interest expenses) / assets (instead of revenues / assets).
  - Costs C_b,t computed as (total expenses − interest expenses) (instead of total expenses).
- Empirical result: the alternative Lerner does not grow over time, confirming the sharp rise of the unadjusted Lerner index is caused by the “denominator effect”.
- The alternative Lerner appears impacted by business cycle fluctuations, whereas the statistical adjustment does not because it explicitly controls for economic growth.

### Statistical adjustment, regression setup, and main findings
- Statistically adjusted Lerner index is computed as the residual from a regression of the traditional Lerner index on policy rates and GDP growth (see Table 2).
- Regressions are estimated with ordinary least squares with robust standard errors and include country fixed effects.
- For each variable, reported coefficient is the sum of contemporaneous and lagged terms; p-values are from a joint F-test for the sum of coefficients.
- Main interpretation: the significant drop in policy rates after the global financial crisis (GFC) can explain the apparent increase in the unadjusted Lerner; after adjustment, market power is fairly constant in advanced economies.
- The adjusted Lerner flattens more for European countries than for the US (Panel C, Appendix Figure A2).

### Policy implications and concluding remarks
- The Lerner index shows a marked increase in market power in advanced economies, especially after the global financial crisis, but other indicators are more muted.
- Adjusting the Lerner index for policy rates and cyclical factors reveals market power has been fairly constant in advanced economies, consistent with other indicators and with patterns in emerging markets.
- Policymakers should monitor behavior of the Lerner index at the zero lower bound (ZLB).
- Since the GFC, traditional measures of mark-ups in banking appear to have increased largely due to the significant drop in policy rates; once accounted for, the Lerner for advanced economies shows no significant upward trend.
- Uncertainty going forward:
  - If rates stay at the ZLB for long, pressure on profits may intensify and possibly trigger consolidation in the banking sector.
  - Structural challenges such as the rise of fintech could affect banks’ ability to exert market power.
- At least so far, there is little indication that banks’ profitability has been adversely affected in a low interest rate environment.

### Key statistics and regression results (preserve reported numbers exactly)
- Table 1A (bank-level summary statistics):
  - Lerner Index: # obs 312,566; Mean 19.9130; Std. Dev. 67.96
  - NIM: # obs 312,566; Mean 3.69; Std. Dev. 2.1; Min -0.24; Max 19.64
  - ROE: # obs 312,566; Mean 6.51; Std. Dev. 12.4; Min -160.37; Max 41.48
  - ROA: # obs 312,566; Mean 0.71; Std. Dev. 1.3; Min -8.37; Max 7.8
  - Cost to income: # obs 312,566; Mean 82.9; Std. Dev. 17.3; Min 0; Max 169.8
- Table 1B (country-level summary statistics):
  - Lerner, traditional: # obs 1,108; Mean 0.24; Std. Dev. 0.13; Min 0.00; Max 0.66
  - Lerner, adjusted: # obs 1,108; Mean 0.00; Std. Dev. 0.06; Min -0.23; Max 0.26
  - HHI: # obs 1,107; Mean 1427; Std. Dev. 906; Min 141; Max 7,564
  - Real GDP growth: # obs 1,108; Mean 0.03; Std. Dev. 0.03; Min -0.06; Max 0.10
  - Policy rate: # obs 1,108; Mean 0.05; Std. Dev. 0.04; Min -0.01; Max 0.16
- Table 2 (Cyclical Drivers of the Lerner Index — selected reported coefficients and statistics):
  - Column (1) Whole period: Policy rate = -0.87*** (p-value (0.00)); GDP growth = 0.92*** (p-value (0.00)); Observations 1,108; R-squared 0.77; # countries 67; Country FEs Yes; Within R-squared 0.17
  - Columns allowing AE vs EMDE effects report:
    - Policy rate - AEs: -2.75*** (p-value (0.00)); Growth rate - AEs: 1.50*** (p-value (0.00))
    - Policy rate - EMDEs: -0.44** (p-value (0.00)); Growth rate - EMDEs: 0.97*** (p-value (0.00))
  - Columns (3) and (4) report pre-GFC and post-GFC separate regressions (2010 as first post-GFC year); Observations: 582 (pre-GFC) and 525 (post-GFC); R-squared 0.86 and 0.88 respectively.

### Sample composition and heterogeneity
- Sample covers 67 countries (27 advanced economies and 40 emerging markets and developing economies) with more than 10 observations in many tables and figures.
- Bank counts:
  - Advanced economies total banks: 21,043 (examples: United States 13,281; Germany 2,501; Italy 934; Japan 823)
  - EMDEs total banks: 3,734 (examples: Russian Federation 1,136; India 315; Brazil 200)
- Country-level observation counts for panels:
  - Advanced economies sample total obs: 249 (typically 18 per country; exceptions shown like Singapore 11)
  - EMDEs sample total obs: 693 (typically 18 per country; exceptions included)
- Heterogeneity exhibited across bank groups and countries:
  - Appendix Figure A1 shows heterogeneity across bank groups (by business model and by size) for 27 advanced economies.
  - Appendix Figure A2 panels:
    - Panel A: Lerner index trajectories for low vs high post-GFC M&A countries.
    - Panel B: Median HHI for low vs high post-GFC M&A countries.
    - Panel C: Median country-specific Lerner index and statistically adjusted Lerner index for United States vs EU countries; adjusted Lerner flattens more for EU than US.

*Sources: Fitch Connect; IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021149-print-pdf.pdf_