## _wp1351 - Appendix I

## Source details

**Canonical URL:** [_wp1351 - Appendix I](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1351.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1351.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1351.pdf.json)

---

### I. Introduction
- Net interest margin (NIM) defined: NIM_{i,c,t} = (total interest revenue_{i,c,t} − total interest expense_{i,c,t}) / total assets_{i,c,t}.
- Regions analyzed: EAC, WAEMU, CEMAC, SACU; ECCU included as comparator.
- Four hypotheses tested (Beck and Hesse, 2009):
  - Risk-based hypothesis: informational imperfections, adverse selection, moral hazard → higher margins.
  - Small financial system hypothesis: fixed costs prevent scale economies → higher margins.
  - Market structure hypothesis: imperfect competition → economic rents raise NIMs.
  - Macroeconomic hypothesis: exchange rate, interest rate, real-economy fluctuations increase NIMs.
- Policy implication overview:
  - If fixed costs dominate, promoting competition may be ineffective.
  - If information asymmetries dominate, solutions include credit bureaus and improved asset recovery.
- Study objective: assess relative contribution of the four hypotheses to NIMs across regional groupings using measures of lending risk, monopolistic rents, economies of scale and business-cycle movements.

### II. Literature review (selected empirical findings)
- Macroeconomic environment:
  - Ho and Saunders (1981) model implies NIMs rise with variance of interest rates.
  - Inflation and high treasury bill rates positively correlated with bank spreads in multiple studies.
- Bank competition:
  - Theory: more competition → lower spreads; empirical evidence mixed.
  - Some studies find market concentration significant; proxies can be fragile.
- Bank-specific factors:
  - Higher operating costs positively associated with higher NIMs.
  - Reliance on fee income may negatively affect NIMs.
  - Higher non-performing loans can raise margins via risk premia.
  - Capitalization affects funding costs and thus NIMs.
- Institutional setting:
  - Strong legal environment and creditor rights lower NIMs.
- SSA-specific findings:
  - Volatile macro environment and small financial systems challenge region.
  - Beck and Hesse (2009): strong evidence for risk-based view; some for macroeconomic view; little for market-structure view.
  - Ahokpossi (2012): lending risk, interest rate risk, bank equity and non-interest activity matter; inflation significant with non-linear effects.

### III. Evolution of NIMs (1997–2011) — stylized facts and accounting decomposition
- Sample period: "1997–2011."
- Sample construction: original BankScope sample "400 banks in 30 countries over the period 1997-2011." Final sample after exclusions: "213 banks" forming an unbalanced panel.
- Cross-sectional/time-series patterns:
  - EAC had the highest margins throughout 1997–2011.
    - EAC margins were on average over 2 percentage points greater than WAEMU.
    - EAC margins: 7.7 percentage points in 2001; declined to 5.7 in 2008; rose to 6.2 percentage points from 2009 to 2011.
    - Loan-loss provisions in EAC: 2.7 percentage points of the margin in 2002 and 0.4 percentage points in 2011.
    - Operating costs fell from 6.9 percentage points in EAC to 5.7 percentage points (period implied between early period and 2011).
  - WAEMU margins declined from 5.2 percentage points in 1998 to 3.5 percentage points in 2011.
    - WAEMU decline associated with falling loan-loss provisions and rising non-interest income, not operating cost decline.
  - CEMAC and SACU experienced considerable volatility in NIMs between 1997 and (text ends).
  - ECCU margins oscillated between 3.7 and 2.5 percentage points over 1997–2011 and were fairly constant.
  - CEMAC: "1.3 percentage points in 1998" and "rose to 3.6 percent in 2011."
- Decomposition evidence: "operating costs and loan-loss provisions seem to have been associated most closely with movements in NIMs across the regions."
- Operating-cost contribution over whole period: operating costs were the largest contributor to margins in all regions except CEMAC, ranging "between 41 (SACU) and 46 percent (ECCU)" on average.

### IV. Empirical strategy, variables, and estimation
- Panel specification (dynamic): y_ict - y_{ict-1} = α y_{ict-1} + β x_{ict} + γ z_{ct} + μ_{ic} + η_t + ε_{ict}.
- Bank-level explanatory variables: operating costs, loan-loss provisions, non-interest income, liquid assets, equity, size of loan book.
- Market-structure measures: Lerner Index and HHI (assets and deposits).
- Macroeconomic controls: inflation, exchange rate, GDP per capita, treasury bill rate / central bank rate.
- Institutional variables: six WGI measures plus WGI composite (first principal component).
- Estimation techniques: pooled OLS (POLS), fixed effects (FE), system GMM (Arellano–Bover/Blundell–Bond) with Windmeijer robust two-step standard errors and collapsed instruments.

### V. Main empirical findings (statistical and economic significance)
- Robust determinants across specifications:
  - Operating costs/assets (log)
    - POLS coefficient examples: "3.054***" (column (i)); other POLS values "2.331***", "2.250***".
    - System GMM coefficients: e.g., "1.582***", "1.655***", "2.323***".
    - Table 7 effect of 1 S.D. (Mean S.D. "1.560.47"): POLS 1.05, FE 0.77, GMM 1.21 percentage points increase in NIMs.
  - Equity/assets (log)
    - POLS: "1.354***", other POLS values "1.501***", "1.310***".
    - GMM: "0.985***" (and other values around "0.626", "1.431***" in some specs).
    - Table 7 effect of 1 S.D. (Mean S.D. "2.420.61"): POLS 0.80, FE 0.74, GMM 0.98 percentage points increase in NIMs.
- Other findings:
  - Loans (log) positive in POLS (e.g., "0.164**", "0.225***") and FE but often insignificant in system GMM (suggesting simultaneity).
  - Loan impairment charge/assets (log): POLS "0.252***"; FE "0.265***"; often smaller/insignificant in GMM.
  - Non-interest activity/assets: negative and significant in POLS/FE (e.g., "-0.424***", "-0.281***", "-0.274***"); mostly insignificant in GMM.
  - Lerner Index: POLS "0.562**" (in one column) and later "1.230***"; GMM shows significance in some specs (e.g., "0.785**"); economic impact smaller than operating costs/equity.
  - Inflation and T-Bill rate: positive and significant in POLS; T-Bill rate remains significant in GMM (e.g., "0.0929***"); inflation often loses significance in FE/GMM.
  - USD exchange rate (% change): negative and significant in POLS (e.g., "-0.0164**") but not robust in FE/GMM.
  - GDP per capita: generally not significant in POLS; can be weakly significant under FE in some specs.
- Dynamic persistence:
  - Lagged net margin significant in GMM: lag coefficient ("α") ranges "0.39 to 0.44" (e.g., "0.390***", "0.442***", "0.391***"), indicating persistence.
- Institutions and operating costs (indirect channel):
  - Pooled OLS regression with operating costs (log) as dependent variable shows:
    - Size of operations (log): coefficient around "-0.0584***" to "-0.0609***".
    - WGI Composite: "-0.0917***" (Table 3 column (vii)) — better institutions reduce operating costs.
    - Robustness (Table 4): WGI Composite remains significant with values like "-0.100***", "-0.0794***", "-0.0842***", and "-0.0692*".
  - Interpretation: Poor institutional quality indirectly influences NIMs through higher operating costs (costly credit appraisal, monitoring, asset recovery).

### VI. Quantitative and economic significance (Tables 7–9, selected entries preserved)
- Table 7: effects of a one standard deviation change (baseline specification; GMM long-run effect beta/(1-alpha) with alpha = 0.391).
  - Operating costs/assets (log): Mean S.D. "1.560.47"; effect of 1 s.d. on margins: POLS 1.05, FE 0.77, GMM 1.21.
  - Equity/assets (log): Mean S.D. "2.420.61"; effect of 1 s.d. on margins: POLS 0.80, FE 0.74, GMM 0.98.
  - Loans (log): Mean S.D. "4.531.59"; effect of 1 s.d. on margins: POLS 1.59, FE 0.36, GMM 0.71 (table shows "0.38" as one entry).
  - Lerner Index: Mean S.D. "-0.060.48"; effect of 1 s.d. on margins: POLS 0.59, FE 0.57, GMM 0.62 (system GMM interpretation: 0.62 percentage points).
  - T-Bill rate: Mean S.D. "8.214.06"; effect of 1 s.d. on margins: POLS 0.51, FE 0.28, GMM 0.62 (system GMM long-run effect).
  - WGI Composite / institutions: indirect impact reported in Table 8 (see below).
- Table 8: indirect impact on NIMs via operating-cost channel (POLS result in Table 4 column (vii) for operating costs; indirect impact computed using GMM Table 6 column (vi)):
  - Loans (log): operating-cost channel POLS coefficient "-0.0609***"; Mean S.D. "4.531.59"; indirect impact on margins: -0.25 percentage points.
  - Non-performing Loans (log): operating-cost channel POLS coefficient "0.0142"; Mean S.D. "1.671.38"; indirect impact on margins: 0.05 percentage points.
  - Lerner Index: operating-cost channel POLS coefficient "0.0199"; Mean S.D. "-0.060.48"; indirect impact on margins: -0.02 percentage points.
  - WGI Composite: operating-cost channel POLS coefficient "-0.0917***"; Mean S.D. "-0.931.38"; indirect impact on margins: -0.33 percentage points.
    - Interpretation: A one standard deviation improvement in the WGI Composite predicts NIMs to fall by 0.33 percentage points via lower operating costs.
- Table 9: regional decomposition (2011 averages × system GMM coefficients; permanent effects beta/(1-alpha)):
  - EAC (East African Community)
    - Operating costs/assets (log): 2.37
    - Equity/assets (log): 1.90
    - T-Bill rate: 1.48
    - Lerner Index: 0.15
    - Implied net interest margin: 5.75
    - True net interest margin (2011): 6.19
    - Error: 0.44
  - CEMAC
    - Operating costs/assets (log): 1.45
    - Equity/assets (log): 1.33
    - T-Bill rate: 0.61
    - Lerner Index: 0.19
    - Implied net interest margin: 3.39
    - True net interest margin (2011): 4.16
    - Error: 0.64
  - ECCU
    - Operating costs/assets (log): 1.31
    - Equity/assets (log): 1.64
    - T-Bill rate: 0.90
    - Lerner Index: -0.27
    - Implied net interest margin: 3.84
    - True net interest margin (2011): 2.84
    - Error: -1.17
  - UWAEMU
    - Operating costs/assets (log): 2.08
    - Equity/assets (log): 1.47
    - T-Bill rate: 0.65
    - Lerner Index: 0.18
    - Implied net interest margin: 4.20
    - True net interest margin (2011): 3.67
    - Error: -0.70
  - SADC
    - Operating costs/assets (log): 1.84
    - Equity/assets (log): 2.17
    - T-Bill rate: 0.97
    - Lerner Index: -0.14
    - Implied net interest margin: 4.98
    - True net interest margin (2011): 4.93
    - Error: -0.24
  - Interpretation: Operating costs and equity/assets together explain the vast proportion of predicted NIMs in every region (example for EAC: operating costs 2.4 percentage points, equity/assets 1.9 percentage points, T-Bill rates 1.5 percentage points → implied 5.8 vs actual 6.2 in 2011).

### VII. Conclusions and policy implications (as presented)
- Main empirical conclusions:
  - Operating costs and equity-to-asset ratios are both statistically and economically significant determinants of NIMs in SSA and the Eastern Caribbean.
  - Operating costs are strongly and inversely related to the size of bank loan portfolios, indicating significant fixed costs to lending and scope for economies of scale.
  - Treasury bill (short-term) rates have a statistically significant but quantitatively smaller role.
  - Market-structure effects (Lerner index) show some significance but are economically smaller than operating costs and equity.
- Supported hypotheses:
  - Evidence supports the risk-based and small-financial system views (importance of operating costs, institutions, economies of scale).
  - Some evidence for the macroeconomic hypothesis (short-term interest rate) and limited evidence for the market-structure hypothesis.
- Institutional channel and policy implications:
  - Poor institutional quality raises operating costs (via costly credit appraisal, monitoring, asset recovery) and indirectly increases NIMs.
  - Policy measures implied by findings:
    - Reduce operating costs and fixed costs of lending (promote scale economies, reduce procedural costs).
    - Improve institutional quality (regulatory quality, rule of law) to lower operating costs.
    - Enhance credit information (e.g., credit bureaus) and asset-recovery mechanisms to address information asymmetries.
    - Recognize that promoting competition alone may be less effective if fixed costs and institutional weaknesses dominate.

*Source: _wp1351 - Appendix I (selected content).*

### Appendix I .............................................................................................................

### _wp1351 - Appendix I

### I. INTRODUCTION
- Net interest margin (NIM) is defined as the value of a bank‘s net interest revenue as a share of its total assets.
- Within Sub-Saharan Africa (SSA), NIMs are on average considerably higher than in most developed economies.
- Regional groups analyzed: EAC, WAEMU, CEMAC, SACU; ECCU included as comparator.
- Four hypotheses for high NIMs (Beck and Hesse, 2009):
  - Risk-based hypothesis: informational imperfections, adverse selection, moral hazard lead banks to charge higher margins.
  - Small financial system hypothesis: fixed costs of providing financial services prevent scale economies, requiring high margins.
  - Market structure hypothesis: imperfect competition allows banks to earn economic rents, raising NIMs.
  - Macroeconomic hypothesis: exchange rate, interest rate, and real economy fluctuations (e.g., high domestic interest rates reflecting macro risks) increase NIMs.
- Policy implication overview:
  - If fixed costs dominate, promoting competition may be ineffective.
  - If information asymmetries dominate, solutions include credit bureaus and improved asset recovery.
- Study objective: assess relative contribution of the four hypotheses to NIMs across regional groupings, using measures of lending risk, monopolistic rents, economies of scale and business-cycle movements.
- Key empirical focus: explain cross-sectional and time-series variation in NIMs among EAC, WAEMU, CEMAC, SACU, and ECCU.

### II. LITERATURE REVIEW (summary of empirical findings)
- Macroeconomic environment:
  - Ho and Saunders (1981) dealership model implies NIMs rise with variance of interest rates.
  - Maudos and de Guevara (2004) find interest-rate volatility significant.
  - Inflation and high treasury bill rates positively correlated with bank spreads (Hoanahan, 2003; Ahokpossi, 2012; Poghosyan, 2010).
- Degree of bank competition:
  - Standard theory: more competition leads to lower spreads.
  - Empirical evidence mixed; proxies for market structure can be fragile.
  - Poghysan (2010) and Martinez-Peria and Moody (2004) find significant effect of market concentration.
- Bank-specific factors:
  - Higher operating costs positively associated with higher NIMs.
  - Reliance on fee income may negatively affect NIMs.
  - Higher non-performing loans could lead to higher margins via risk premia.
  - Capitalization can affect funding costs and thus NIMs (Demirguc-Kunt, Laeven and Levine, 2004).
- Institutional setting:
  - Strong legal environment and creditor rights lower NIMs (Demirguc-Kunt, Laeven and Levine, 2004; Poghosyan, 2012).
- Findings specific to SSA:
  - Honohan and Beck (2007): volatile macroeconomic environment and small financial systems challenge region.
  - Beck and Hesse (2009): find role for macroeconomic variables; faster GDP growth, rising inflation and increasing treasury bill rates associated with higher margins; lower spreads linked to better institutions and larger balance sheets. Strong evidence for risk-based approach; some evidence for macroeconomic view; little evidence for market structure view.
  - Ahokpossi (2012): lending risk, interest rate risk, bank equity and non-interest activity matter; inflation statistically significant with non-linear effects.

### III. EVOLUTION OF NET INTEREST MARGINS (1997–2011)
- Definition used in the study:
  - NIM_{i,c,t} = (total interest revenue_{i,c,t} − total interest expense_{i,c,t}) / total assets_{i,c,t}
  - Total interest revenue, total interest expense and total assets taken from bank income statement for year t.
- Accounting decomposition:
  - Net interest revenue minus personnel and other non-interest operating costs, provisions for loan losses, and profits are related by accounting identity; dividing by total assets expresses NIMs as components: profit before tax, loan-loss provisions, non-interest income, operating costs, and an error term (normalized by total assets).
- Cross-sectional and time-series patterns (1997–2011):
  - Throughout 1997–2011, EAC had the highest margin levels of all regions.
    - On average, EAC margins were over 2 percentage points greater than the next-highest region, WAEMU.
    - By 2011, CEMAC, ECCU and WAEMU had similar levels of margins.
    - By 2011 SACU had the lowest margins of all five regions at 2.72 percentage points.
    - By 2011 EAC still had the highest NIMs at above 6 percentage points.
  - ECCU margins:
    - Fairly constant over 1997–2011, oscillating between 3.7 and 2.5 percentage points.
- Regional trend highlights:
  - EAC:
    - Sustained decline in margins between 2001 and 2008: from 7.7 percentage points in 2001 to 5.7 in 2008.
    - Margins rose somewhat to 6.2 percentage points from 2009 to 2011.
    - Declining operating costs and loan-loss provisions largely associated with this trend.
      - Loan-loss provisions accounted for 2.7 percentage points of the margin in EAC in 2002 but only 0.4 percentage points in 2011.
      - Operating costs fell from 6.9 percentage points in EAC to 5.7 percentage points (period not explicitly restated; implied between early period and 2011).
  - WAEMU:
    - Sustained decline in margins from 5.2 percentage points in 1998 to 3.5 percentage points in 2011.
    - Unlike EAC, decline not associated with operating costs; associated with falling loan-loss provisions and rising non-interest income.
  - CEMAC and SACU:
    - Experienced considerable volatility in NIMs between 1997 and (text ends).

### IV. MAIN EMPIRICAL FINDINGS (from paper overview)
- Strong evidence that:
  - Bank operating costs are both statistically and economically significant determinants of NIMs in all regions.
  - Equity-to-asset ratios are both statistically and economically significant determinants of NIMs in all regions.
- Indirect institutional effect:
  - Poor institutional quality indirectly influences NIMs through higher operating costs.
  - Consistent with view that poor regulatory quality and weak rule of law cause banks to engage in costly credit appraisals and monitoring.
- Scale economies:
  - Operating costs are strongly and inversely related to the size of the loan portfolio.
  - Fixed costs to lending are non-trivial, creating scope for banks with larger loan portfolios to exploit economies of scale.
- Overall interpretation:
  - Results are consistent with both the risk-based and small financial system hypotheses:
    - Small nature of African financial systems and risk factors, amplified by inefficiency of judicial systems, are key determinants of high margins.
  - Less supportive evidence for market-structure hypothesis in the overview presented.

### V. STRUCTURE OF THE FULL PAPER (as presented)
- Section II: brief literature review, motivation for explanatory variables, contribution.
- Section III: stylized facts and accounting decomposition of NIMs (1997–2011).
- Section IV: econometric strategy to test the four hypotheses.
- Section V: estimation results.
- Section VI: conclusions with policy recommendations and areas for future work.

*Source: _wp1351 - Appendix I (selected content).*

### 2011. CEMAC initially had low NIMs (1.3 percentage points in 1998), coupled with very

### _wp1351 - 2011. CEMAC initially had low NIMs (1.3 percentage points in 1998), coupled with very

### Key empirical facts and stylized findings
- CEMAC net interest margins (NIMs) were "1.3 percentage points in 1998" and "rose to 3.6 percent in 2011."
- The original BankScope sample: "400 banks in 30 countries over the period 1997-2011." Final sample after exclusions: "213 banks" forming an unbalanced panel.
- Sample period: "1997–2011."
- Decomposition evidence: "operating costs and loan-loss provisions seem to have been associated most closely with movements in NIMs across the regions."
- Over the whole period, "operating costs were by far the largest contributor to margins in all regions except CEMAC," with operating-cost contribution to the margin ranging "between 41 (SACU) and 46 percent (ECCU)" on average.
- Equity-to-asset ratio correlation: regression of NIM on log(equity-to-asset) yields "R-squared of 0.125" in repeated cross-sections.

### Empirical strategy and model specification
- Panel specification (dynamic) used: y_ict - y_{ict-1} = α y_{ict-1} + β x_{ict} + γ z_{ct} + μ_{ic} + η_t + ε_{ict} (presented in equation (3) form in the text).
- Bank-level explanatory variables (following Ho and Saunders (1981)): operating costs, loan-loss provisions, non-interest income, liquid assets, equity, size of loan book.
- Market-structure measures used: Lerner Index and HHI (assets and deposits).
- Macroeconomic controls: inflation, exchange rate, GDP per capita, treasury bill rate / central bank rate.
- Institutional variables: six WGI measures plus a WGI composite (first principal component).
- Estimation techniques: pooled OLS, fixed effects, system GMM (Arellano–Bover/Blundell–Bond), with Windmeijer robust two-step standard errors and collapsed instruments to limit instrument proliferation.

### Data processing and sample construction
- Eliminations: microfinance institutions, holding companies, banks with fewer than four consecutive yearly observations, and outliers (top and bottom percentiles of bank-specific variables except loan-book size).
- Variable transformations: operating costs, equity-to-asset ratios, loans and loan loss provisions entered in logs.
- Institutions summary: "a summary index of institutions using principal components analysis... weights derived from the first principal component."

### Main pooled OLS results (highlights from Table 2)
- Operating costs/assets (log): coefficient "3.054***" (column (i)), with alternative specifications showing values "2.331***", "2.250***", and a range around "2.23-2.24" in later columns.
- Equity/assets (log): coefficient "1.354***" (column (i)); other columns report "1.501***", "1.310***", and values around "1.29-1.31".
- Loans (log): "0.164**" (column (i)) and up to "0.225***" in other specs.
- Loan impairment charge/assets (log): "0.252***" (column (i)) and similar across columns (e.g., "0.263***", "0.241***").
- Non-interest activity/assets: negative and significant, "-0.424***" (column (i)) and around "-0.28 to -0.32" in other columns.
- Liquid/total assets: not robustly significant (e.g., "0.0123**" in column (i) but often insignificant elsewhere).
- Lerner Index: "0.562**" in column (i); becomes "1.230***" and "1.131***" when market-structure and macro controls added; authors note coefficient rises by "over 50 percent" when macro variables added.
- Inflation: positive and significant in pooled OLS (e.g., "0.0350**").
- T-Bill rate: positive and significant (e.g., "0.125***").
- USD exchange rate (% change): negative and significant (e.g., "-0.0164**"), interpreted as exchange rate depreciation associated with lower NIMs in pooled OLS (not robust under fixed effects/GMM).
- GDP per capita: not significant in pooled OLS.

- Sample and fit notes from Table 2:
  - Observations vary by specification, e.g., "1524", "1400", "1399", "1320", "1320", "1320", "1320", "1320", "1320", "1320".
  - R-squared values: e.g., "0.417", "0.419", "0.474", "0.467" (various specifications).
  - Number of banks ranges around "213", "210", "209" across columns.

### Institutions and operating costs (pooled OLS auxiliary results)
- Direct pooled OLS: institutions not statistically significant in direct NIM regressions when included with full controls (columns (iv)–(x) of Table 2).
- Indirect effect via operating costs: pooled OLS regression with operating costs (log) as dependent variable shows:
  - Size of operations (log): coefficient around "-0.0584***" to "-0.0609***" (Table 3 column (i) and later).
  - WGI measures: each WGI measure (Voice and Accountability; Political Stability; Government Effectiveness; Regulatory Quality; Rule of Law; Control of Corruption) is negative and statistically significant in separate specifications (e.g., "WGI Voice and Accountability -0.151***", "WGI Government Effectiveness -0.219***").
  - WGI Composite: " -0.0917***" (Table 3 column (vii)) indicating better institutions reduce operating costs.
- Robustness with additional controls (Table 4):
  - WGI Composite remains significant: "-0.0917***" (column (i)) and still significant after adding controls, with values like "-0.100***", "-0.0794***", "-0.0842***", and "-0.0692*" (columns (ii)–(v)).
  - Tertiary Enrollment shows a negative coefficient and weak significance: "Tertiary Enrollment -0.00985*" (Table 4).
  - Size of operations remains negative for operating costs in most specifications (e.g., "-0.0609***").

- Interpretation: Better institutional quality reduces operating costs (consistent with risk-based view that poor institutions induce costly monitoring/recovery), supporting an indirect institutional channel to NIMs through operating costs.

### Fixed effects and system GMM results (highlights from Tables 5–6)
- Presence of bank fixed effects is statistically supported (Breusch-Pagan LM and Hausman tests discussed).
- Fixed effects results: signs and significance of bank-specific variables largely preserved relative to pooled OLS; inflation and exchange-rate significance weakened; GDP per capita can become significant at the 10 percent level under fixed effects.
- System GMM (dynamic panel) results:
  - Lagged net margin significant in GMM: lag coefficient ("α") ranges from "0.39 to 0.44" (e.g., "0.390***", "0.442***", "0.391***"), indicating persistence.
  - Under system GMM the robustly significant bank-specific determinants are primarily:
    - Operating costs/assets (log): coefficients e.g., "1.582***", "1.655***", "2.323***" across GMM columns.
    - Equity/assets (log): remains significant in many GMM specifications (e.g., "0.985***", with other GMM values like "0.626", "1.431***").
  - Variables that lose statistical significance in system GMM (relative to pooled OLS/fixed effects): loans (log), loan impairment charge, non-interest income (mostly insignificant, with one weak significance at the 10 percent level).
  - Short-term rate (T-Bill) remains significant in GMM (e.g., "0.0929***").
  - GDP per capita generally loses significance in GMM (exceptions in some alternative market-structure specs where GDP per capita shows " -6.38e-05*").

- Market structure robustness (Table 6):
  - Lerner Index remains significant in some GMM specifications (e.g., "Lerner Index 0.785**").
  - HHI (assets) reports coefficient "-2.265" (insignificant); HHI (deposits) reports "0.99" (insignificant). Authors note lack of robustness across measures and caution in interpreting market-structure effects.

- GMM diagnostic tests:
  - Hansen J test p-values reported (e.g., "0.579", "0.267", "0.294" in Table 5); AR(2) test p-values e.g., "0.390", "0.441", "0.655" indicate no rejection of valid instruments in many specs.
  - Authors apply Windmeijer correction and Roodman 'collapse' to limit instrument proliferation.

### Quantitative/ economic significance (summary points drawn from tables)
- Operating costs exert the largest quantified effect on NIMs across specifications, with pooled OLS coefficients as high as "3.054***" (Table 2 column (i)) and system GMM coefficients in the "1.3–2.3" range depending on specification (Tables 5–6).
- Equity-to-asset ratios are positively associated with higher NIMs across pooled OLS, fixed effects and GMM, with pooled OLS values around "1.3–1.5" and GMM values around "0.985***" in preferred specifications.
- Size (loans) has positive association in pooled OLS and fixed effects (e.g., "0.164**", "0.483***") but becomes insignificant under system GMM, suggesting simultaneity/endogeneity concerns.
- Institutional quality reduces operating costs: WGI composite coefficient around "-0.0917***", with robustness to additional controls (Table 4).

### Interpretations emphasized by the authors
- Operating costs are key to both short-run movements and long-run differences in NIMs across regions.
- Declines in loan-loss provisions are associated with declines in margins in some regions (EAC and WAEMU) and represent a broader trend across regions.
- Equity-to-asset ratio correlation with NIMs (R-squared "0.125") provides prima-facie evidence equity may be important for explaining NIMs.
- System GMM suggests persistence in margins and that some bank-specific variables significant under pooled OLS may reflect simultaneity rather than causal effects.
- Institutional quality likely affects NIMs indirectly via operating costs (better institutions lower operating costs, reducing NIMs).

*Source: Excerpt from the provided IMF working paper content (1997–2011 bank-level analysis).*

### 8. Table 7 shows the effect on NIMs of a one standard deviation change in one of the

### _wp1351 - 8. Table 7 shows the effect on NIMs of a one standard deviation change in one of the

### Effects on Net Interest Margins (Table 7; baseline specification, regressions (iii) in Table 3 and (iii)/(vi) in Table 6)
- Estimation methods considered: Pooled OLS (POLS), fixed effects (FE), system GMM (GMM).
- For system GMM, the long-run (permanent) effect is calculated as beta/(1-alpha), where alpha = 0.391 (estimated coefficient on the lagged dependent variable).
- Operating costs and equity dominate the magnitude of effects on NIMs.

Key coefficient estimates and computed effects of a 1 standard deviation (S.D.) change on NIMs (table entries preserve source values):
- Operating costs/assets (log)
  - Coefficients: POLS 2.250***, FE 1.655***, GMM 1.582***
  - Mean S.D.: 1.560.47 (as presented)
  - Effect of 1 s.d. on margins: POLS 1.05, FE 0.77, GMM 1.21
  - Interpretation: A one standard deviation increase in operating costs (rise in the log of operating costs-to-assets of 0.47) is associated with an increase in NIMs between 0.77 and 1.21 percentage points.
- Equity/assets (log)
  - Coefficients: POLS 1.310***, FE 1.217***, GMM 0.985***
  - Mean S.D.: 2.420.61
  - Effect of 1 s.d. on margins: POLS 0.80, FE 0.74, GMM 0.98
  - Interpretation: A one standard deviation increase in equity (increase in the log of the equity/asset ratio of 0.61) is associated with a rise in NIMs between 0.74 and 0.98 percentage points.
- Loans (log)
  - Coefficients: POLS 0.225***, FE 0.449***, GMM 0.145
  - Mean S.D.: 4.531.59
  - Effect of 1 s.d. on margins: POLS 1.59, FE 0.36, GMM 0.71, (table shows "0.38" as one entry)
- Liquid/total assets
  - Coefficients: POLS -0.006130, FE 0.0000927, GMM -0.0121
  - Mean S.D.: 27.9016.02
  - Effect of 1 s.d. on margins: POLS -0.10, FE 0.00, GMM -0.32
- Loan impairment charge/assets (log)
  - Coefficients: POLS 0.241***, FE 0.265***, GMM 0.056
  - Mean S.D.: -0.461.18
  - Effect of 1 s.d. on margins: POLS 0.28, FE 0.31, GMM 0.11
- Non-interest activity/assets
  - Coefficients: POLS -0.281***, FE -0.274***, GMM -0.221
  - Mean S.D.: 3.172.01
  - Effect of 1 s.d. on margins: POLS -0.56, FE -0.55, GMM -0.73
- Lerner Index
  - Coefficients: POLS 1.230***, FE 1.174***, GMM 0.785**
  - Mean S.D.: -0.060.48
  - Effect of 1 s.d. on margins: POLS 0.59, FE 0.57, GMM 0.62
  - Interpretation: A one standard-deviation increase in the Lerner Index raises NIMs by 0.62 percentage points (system GMM).
- Inflation
  - Coefficients: POLS 0.0350**, FE -0.00695, GMM -0.00198
  - Mean S.D.: 5.844.42
  - Effect of 1 s.d. on margins: POLS 0.15, FE -0.03, GMM -0.01
- T-Bill rate
  - Coefficients: POLS 0.125***, FE 0.0701***, GMM 0.0929***
  - Mean S.D.: 8.214.06
  - Effect of 1 s.d. on margins: POLS 0.51, FE 0.28, GMM 0.62
  - Interpretation: A one standard deviation change in T-Bill rates of 4.06 percentage points raises NIMs by 0.62 percentage points in the long run (system GMM).
- USD exchange rate (% change)
  - Coefficients: POLS -0.0164**, FE -0.00684, GMM -0.00566
  - Mean S.D.: -2.049.57
  - Effect of 1 s.d. on margins: POLS -0.16, FE -0.07, GMM -0.09
- log GDP per capita (USD)
  - Coefficients: POLS 0.0000281, FE -0.000241*, GMM -0.000018818
  - Mean S.D.: 811.032611.66
  - Effect of 1 s.d. on margins: POLS 0.07, FE -0.63, GMM -0.08

Notes (as presented in source):
- The economic impact of net interest margins was found by multiplying the standard deviation of each explanatory variable by the respective coefficient estimate.
- For system GMM, the table shows the permanent effect, calculated as beta/(1-alpha).

### Economic significance of operating cost determinants (Table 8; indirect impact on NIMs)
- Method: Effect of each variable on operating costs calculated using POLS result in Table 4, column (vii); indirect impact on NIMs then calculated using GMM results in Table 6, column (vi). Table entries show permanent effect, beta/(1-alpha).
- Selected indirect impacts (operating-cost channel), preserving table values:
  - Loans (log)
    - Operating costs (log) POLS coefficient: -0.0609***
    - Mean S.D.: 4.531.59
    - Indirect impact on margins: -0.25
    - Interpretation: A one standard deviation in the log of the size of the loan book reduces NIMs, through lowering operating costs, by 0.25 percentage points.
  - Non-performing Loans (log)
    - Operating costs (log) POLS coefficient: 0.0142
    - Mean S.D.: 1.671.38
    - Indirect impact on margins: 0.05
  - Lerner Index
    - Operating costs (log) POLS coefficient: 0.0199
    - Mean S.D.: -0.060.48
    - Indirect impact on margins: -0.02
  - WGI Composite (synthetic indicator of institutions)
    - Operating costs (log) POLS coefficient: -0.0917***
    - Mean S.D.: -0.931.38
    - Indirect impact on margins: -0.33
    - Interpretation: A one standard deviation improvement in the synthetic indicator of institutions (roughly equivalent to the difference between Burundi and South Africa in 2011, for instance) predicts NIMs to fall by 0.33 percentage points via lower operating costs.

### Regional comparison (Table 9; 2011 values and system GMM estimates)
- Approach: For each region, obtain arithmetic average of bank-specific explanatory variables for 2011; multiply by system GMM coefficient estimates; table shows permanent effects (beta/(1-alpha)).
- System GMM implied contributions to region-average NIMs (selected entries preserved):
  - EAC (East African Community)
    - Operating costs/assets (log): 2.37
    - Equity/assets (log): 1.90
    - T-Bill rate: 1.48
    - Lerner Index: 0.15
    - Implied net interest margin: 5.75
    - True net interest margin (2011): 6.19
    - Error: 0.44
  - CEMAC
    - Operating costs/assets (log): 1.45
    - Equity/assets (log): 1.33
    - T-Bill rate: 0.61
    - Lerner Index: 0.19
    - Implied net interest margin: 3.39
    - True net interest margin (2011): 4.16
    - Error: 0.64
  - ECCU
    - Operating costs/assets (log): 1.31
    - Equity/assets (log): 1.64
    - T-Bill rate: 0.90
    - Lerner Index: -0.27
    - Implied net interest margin: 3.84
    - True net interest margin (2011): 2.84
    - Error: -1.17
  - UWAEMU
    - Operating costs/assets (log): 2.08
    - Equity/assets (log): 1.47
    - T-Bill rate: 0.65
    - Lerner Index: 0.18
    - Implied net interest margin: 4.20
    - True net interest margin (2011): 3.67
    - Error: -0.70
  - SADC
    - Operating costs/assets (log): 1.84
    - Equity/assets (log): 2.17
    - T-Bill rate: 0.97
    - Lerner Index: -0.14
    - Implied net interest margin: 4.98
    - True net interest margin (2011): 4.93
    - Error: -0.24
- Interpretation: Operating costs and equity/assets together explain the vast proportion of predicted NIMs in every region. Example for EAC: operating costs account for 2.4 percentage points, equity/assets 1.9 percentage points, and short-term interest rates 1.5 percentage points of NIMs, yielding an implied net interest margin at 5.8 compared to the actual margin in 2011 of 6.2 (source text statements preserved).

### Concluding remarks (Section VI)
- Main empirical findings:
  - Operating costs and equity-to-asset ratios are both statistically and economically significant determinants of NIMs in SSA and the Eastern Caribbean.
  - Operating costs are strongly and inversely related to the size of bank loan portfolios, indicating significant fixed costs to lending and scope for economies of scale for larger banks.
  - Treasury bill (short-term) rates have a statistically significant but quantitatively smaller role in explaining NIMs.
  - Market-structure effects (Lerner index) show some significance, but with lower economic importance relative to operating costs and equity.
- Interpretations and hypotheses supported:
  - Evidence supports the risk-based and small-financial system views (importance of operating costs, institutions, and economies of scale).
  - Some evidence for the macroeconomic hypothesis (short-term interest rate significance).
  - Some evidence for the market-structure view (Lerner index significance).
  - Overall, macroeconomic and market-structure hypotheses are less important than bank-specific operating costs and equity in the regions studied.
- Institutional channel:
  - Institutional factors may indirectly contribute to higher operating costs (through regulatory quality and rule of law gaps), increasing costs of credit appraisal, monitoring, and asset recovery, which in turn raise margins.

*Source: _wp1351 - 8. Table 7 shows the effect on NIMs of a one standard deviation change in one of the (IMF Working Paper PDF content provided)*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1351.pdf_