This study examines whether the accumulated stock of private credit provides early-warning information for subsequent deterioration in banking-sector asset quality. It combines annual Passport banking indicators with World Development Indicators for 58 countries over 2010–2024; the preferred sample contains 746 country–year observations. A second-order dynamic fixed-effects model links log(1 + NPL), where NPL denotes the non-performing loan ratio, to lagged private credit to gross domestic product (GDP), real credit growth, lending rates, bank capital, GDP growth, inflation, and unemployment. Its preferred credit-depth coefficient is 0.00377, implying that a 10-percentage-point increase is associated with approximately 0.15 percentage points more NPLs one year later at the sample median. To operationalize early-warning calibration without claiming a universal cutoff, the paper reports the sample credit-depth quartiles and estimates a country fixed-effects linear probability model using the European Banking Authority’s 5% gross-NPL supervisory trigger. In that alternative outcome, a 10-percentage-point increase in credit depth is associated with a 2.78-percentage-point higher conditional probability of NPLs reaching 5% or more (p = 0.002). On a strictly common 609-observation sample, the credit-depth coefficients at one-, two-, and three-year horizons are 0.00501, 0.00960, and 0.01266. Lending rates and unemployment are positive, whereas annual credit growth and capital ratios are not robust predictors. Pooled interactions do not reject equal slopes across broad country partitions. System generalized method of moments (GMM) passes conventional tests but violates a persistence-bound credibility check. The evidence supports an early-warning interpretation, not a causal claim.

This study examines whether digital commerce deepening and digital financial use are associated with banking fragility through the household leverage channel. Using country-year data from Euromonitor International Passport for 16 economies over 2015-2025, the analysis links bank nonperforming loans to household debt, app-based mobile commerce, internet banking, smartphone possession, and government effectiveness. The empirical strategy applies dynamic two-way fixed effects models with country and year effects, clustered standard errors, Driscoll-Kraay sensitivity checks, restricted housing-stress controls, crisis-year exclusions, alternative winsorization, mechanism regressions, and placebo leads. The findings show strong persistence in banking fragility and a positive household-debt signal, although the effect is strongest in robust covariance and alternative winsorization specifications. App-based mobile commerce is negatively associated with nonperforming loans in the dynamic models, suggesting that digital commerce may capture formalization, payment efficiency, or digital maturity rather than mechanical overborrowing. Internet banking and the household-debt-by-government-effectiveness interaction are not robust predictors. Overall, digitalization does not mechanically amplify banking fragility; the more consistent channel is household leverage, moderated only weakly by institutional execution in the available panel.

This study examines whether the accumulated stock of private credit provides early-warning information for subsequent deterioration in banking-sector asset quality. It combines annual Passport banking indicators with World Development Indicators for 58 countries over 2010–2024; the preferred sample contains 746 country-year observations. A second-order dynamic fixed-effects model links log(1 + NPL) to lagged private credit to GDP, real credit growth, lending rates, bank capital, GDP growth, inflation, and unemployment. Its preferred coefficient is 0.00377, implying that a 10-percentage-point increase in private credit to GDP is associated with approximately 0.15 percentage points more NPLs one year later at the sample median. On a strictly common 609-observation sample, the credit-depth coefficients at one-, two-, and three-year horizons are 0.00501, 0.00960, and 0.01266; a stacked country-clustered Wald test rejects equality, although these are horizon-specific predictive projections rather than cumulative causal effects. Lending rates and unemployment are positive, whereas annual credit growth and capital ratios are not robust predictors. Pooled interactions do not reject equal slopes across broad country partitions. System GMM passes conventional tests but violates a persistence-bound credibility check. The evidence therefore supports an early-warning interpretation, not a causal claim.