- Design
- Longitudinal multilevel analysis with years nested within banks and banks situated within countries. Time-varying predictors use a within-between decomposition; year and observed bank/country controls account for measured differences.
- Outcome and measure
- Natural logarithm of financing reported for financial institutions, insurance, real estate and consumer-durable activities.
- Coverage
- The current harmonised public Atlas aggregate contains 4,794 observed bank-year outcomes across 437 unique banks during 2000–2022. The earlier source abstract reports 491 banks for one analysis. The discrepancy remains unresolved; neither count has been silently substituted. Model-specific complete-case exclusions apply.
- Public interpretation
- Islamic-bank status is associated with lower relative exposure in several preferred specifications, while interactions and debt-like channels indicate continued sensitivity to the surrounding financial regime.
Limitations
- The outcome is a broad allocation proxy and is sensitive to bank scale.
- Conditional association is not a design-based causal effect.
- Contract quality, off-balance-sheet exposure and realised loss incidence are not fully observed.
- The coefficient for a bank-type indicator should not be compared directly with differently scaled predictors.