A global analysis of bank profitability factors

The study addresses the determinants of bank profitability in a highly globalized banking environment, noting that most prior research is country- or region-specific and often excludes a broad, global comparison. Given banks’ central role in financial intermediation and monetary policy transmission, understanding profitability drivers is critical for economic growth and financial stability. The paper fills a gap by providing a comprehensive global and regional analysis using 2,091 commercial banks across 110 countries during 2018–2021, grouped into eight world regions. It aims to identify internal (bank-specific) and external (macroeconomic) factors that shape profitability and to compare global versus regional models to assess whether region-specific determinants improve explanatory power and practical applicability.

The literature distinguishes internal (bank-specific) and external (macroeconomic/market) determinants of profitability. Internal factors include capital adequacy, deposits, loans, liquidity, efficiency, size, and margins. Evidence on capitalization is mixed: higher capital can enhance flexibility and profitability, yet lower risk-taking can reduce returns. Deposits often lower funding costs and may raise profitability, but effects can be negative or insignificant under weak loan demand or poor liquidity management. Larger loan volumes can lift income but may harm profits under adverse macro conditions; non-performing loans (NPLs) are commonly linked to lower profitability. Efficiency tends to correlate positively with profitability; bank size often shows positive associations via scale economies, while liquidity is usually linked to lower returns. External factors include GDP per capita and growth, inflation, unemployment, interest rates, market development, sector size, and concentration. Prior studies report mixed effects: competition during expansions can compress margins; downturns raise credit losses and reduce profits. Interest rate effects vary by development level; stock market development can either complement (information, fee income) or substitute for bank intermediation. Recent research highlights COVID-19’s impacts, with more diversified income helping mitigate shocks. Many studies are confined to specific regions (EU, US, developing countries) or fewer countries; only a few adopt a global scope, motivating broader coverage and updated data.

Data: Panel of 2,091 conventional commercial banks from 110 countries, 2018–2021, sourced primarily from Orbis Bank Focus by Moody’s (bank-level financials). Macroeconomic indicators are from the World Bank’s World Development Indicators and the Bank for International Settlements. Banks with total assets below €5 billion were excluded. After cleaning and deduplication, the final sample yields ~8,366 bank-year observations. Countries are grouped into eight regions: Africa, Eastern Europe, Far East and Central Asia, Middle East, North America, Oceania, South and Central America, and Western Europe.

Variables: Profitability proxies are ROAE (net income/average equity, %) and ROAA (net income/average assets, %). Internal factors: LIS (listed indicator), TALN (ln total assets), NPL (impaired loans/gross loans), RSKC (cost of risk), EFR (efficiency ratio), NIM (net interest margin over loans), GRM (gross margin over loans), ETAR (equity-to-assets), CLR (cash-to-liabilities), CDP (customer deposits, deflated), CLO (customer loans, deflated), CLOC (aggregate loans of three largest banks, deflated), BCR (country bank rank by assets). External factors: INF (inflation), UNEM (unemployment), GPC (GDP per capita), GDPG (GDP growth), CBIR (central bank policy rate), DCPS (domestic credit to private sector, % GDP). Definitions appear in Table 2 of the paper.

Model specification: Following Demirgüç-Kunt and Huizinga (1999), panel regressions: P_ijt = α_i + α_ij B_ijt + β_j X_jt + E_ijt, where P is ROAE or ROAA; B bank-specific variables; X country-level variables.

Data treatment and diagnostics: Outliers were trimmed by deleting observations outside (Q1 – 3×IQR, Q3 + 3×IQR). Panel unit root testing used Levin-Lin-Chu; all variables rejected the null of a common unit root (p<0.05). Model selection employed F-tests (fixed effects vs pooled) and Hausman tests (fixed vs random effects). F-tests favored fixed over pooled; Hausman favored random over fixed, so random-effects models were adopted. Residual diagnostics supported normality, linearity, homoscedasticity (plots) and no autocorrelation (Durbin–Watson). Multicollinearity was not a concern (VIF<5). Two model variants were estimated: Model 1 with internal factors only; Model 2 with both internal and external factors, at global and regional levels.

Global results (Table 7):

• For ROAE (Model 2): Significant internal determinants include LIS (+0.631***), TALN (+0.316***), NPL (−66.426***), RSKC (−131.736***), EFR (−8.951***), NIM (−4.260***), GRM (+2.647***), ETAR (+9.509***), CLOC (−0.001***), BCR (+0.014***). CLR and CDP/CLO were not significant. Significant external determinants: INF (+0.435***), UNEM (−0.161***), GPC (−0.001***), GDPG (+0.172***), CBIR (+0.194***), DCPS (−0.011***). Model fit: F=92.123***, R2=27.63% (Adj. 27.33%), RMSE=5.94, MAPE=93.30%.
• For ROAA (Model 2): Significant internal determinants include LIS (+0.056**), NPL (−2.115***), EFR (−0.718***), GRM (+0.415***), ETAR (+7.861***), CLR (+0.231**), CLOC (−0.001***), BCR (+0.001***). TALN, RSKC, NIM, CDP, CLO not significant. Significant external determinants: INF (+0.044***), UNEM (−0.022***), GPC (−0.001*), GDPG (+0.018***), CBIR (+0.019***); DCPS not significant. Model fit: F=127.881***, R2=34.14% (Adj. 33.87%), RMSE=0.64, MAPE=97.68%.
• Interpretation: Listed banks, higher gross margin, and stronger capitalization are associated with higher profitability, while greater NPLs, higher cost/inefficiency, and larger aggregate loans of the top three banks (CLOC) reduce profitability. Macro environments with higher inflation and interest rates and stronger GDP growth support profitability; higher unemployment and higher GDP per capita correlate with lower profitability. DCPS is negatively associated with ROAE but not ROAA. TALN and RSKC matter for ROAE but not ROAA; CLR matters for ROAA only.

Regional patterns (Tables 8–9, synthesis):

• Variables broadly significant across regions: RSKC and EFR are significant in virtually all regions; NPL is significant in all regions for ROAE except North America; ETAR is significant in most regions (for ROAE not in the Middle East); CLOC is not significant in Oceania for ROAE. GRM is widely significant for ROAA except Middle East and North America. NIM’s significance is region-specific (notably Far East & Central Asia, North America, Western Europe for ROAA). INF is significant in some regions (e.g., North America, Western Europe for ROAA) but not universally.
• Region-specific effects: BCR is significant primarily in Western Europe; UNEM is significant in South & Central America. TALN shows negative effects in North America and Western Europe. CLO is significant in Africa and South & Central America. GDPG significantly affects profitability in South & Central America and Western Europe. Overall, region-specific models show higher explanatory power than global models, indicating heterogeneous profitability drivers by region.

The findings confirm that both internal bank characteristics and macroeconomic context shape profitability. Globally, profitability improves with listing status, higher gross margins, and stronger capitalization, and deteriorates with impaired assets and inefficiency, aligning with prior evidence on efficiency and capitalization effects. The robust negative association of NPLs with profitability underscores credit risk management as central to performance. External conditions—higher inflation and policy rates alongside stronger GDP growth—are associated with higher profitability, while higher unemployment depresses it; the negative link with GDP per capita suggests intensified competition in more developed markets that compresses margins. Some results corroborate earlier global studies (e.g., efficiency, capitalization, GDP growth), while others differ across regions (e.g., NPL insignificance in North America), possibly reflecting scale economies, market structures, and regulatory environments. Distinct regional variable sets significantly enhance model fit compared with global models, highlighting the importance of tailoring profitability strategies and regulatory assessments to regional conditions. Deposits and individual banks’ loan volumes were not globally significant, but showed mixed regional effects, indicating heterogeneity in funding structures, credit demand, and intermediation across regions.

The paper provides comprehensive global and regional evidence on bank profitability determinants for 2,091 banks in 110 countries (2018–2021). Globally, key internal drivers include listing status, impaired loans, efficiency, gross margin, and capitalization; external drivers include countries’ rank by bank assets, inflation, unemployment, GDP growth, and policy rates. Regionally, RSKC and EFR are consistently important, with NPL, ETAR, and CLOC also key but varying by region, and several variables (BCR, UNEM, TALN, CLO, GDPG) exerting effects in specific regions only. Regional models generally outperform global ones, indicating heterogeneous determinants and the value of region-specific modeling. Practical implications include focusing managerial attention on credit risk and efficiency, optimizing capitalization, and incorporating macro conditions into profitability planning. Policymakers can use these insights for regulatory calibration and to support market development without disintermediating banks. Future research should broaden determinants (e.g., legal enforcement, IT/governance), apply alternative empirical methods (e.g., GMM, machine learning, qualitative approaches), and examine fintech/digitalization impacts on bank profitability and strategy.

• Variable scope: Only selected internal financial and macroeconomic indicators were included; legal/regulatory enforcement, IT adoption, governance, and other structural factors were not modeled.
• Methodological scope: The study employs random-effects panel regressions; alternative methods (e.g., GMM, neural networks, qualitative analyses) could address endogeneity, nonlinearity, and unobserved heterogeneity differently.
• Measurement and data constraints: Some variables may be imperfect proxies; outlier trimming and data availability constraints (licensing and minimum asset size threshold) may affect generalizability.
• Temporal scope: 2018–2021 covers the COVID-19 period; longer horizons could capture different cycles and structural shifts.
• Regional aggregation: Grouping countries into broad regions may mask within-region heterogeneity.