The unstable boundary: conditional classification of non-performing loan determinants and its implications for banking regulation

Abstract Research on the determinants of non-performing loans (NPLs) is organised around fixed classifications that assign each driver to a single category—macroeconomic, bank-specific or loan-specific. This paper argues that the assignment is not fixed, and that the instability has direct consequences for banking regulation. Whether a determinant operates through an endogenous or an exogenous channel—terms used here in a mechanism sense, relative to a specified unit of analysis, time horizon and institutional, capital or policy regime, rather than in the econometric sense—is conditional on those three dimensions. The argument is drawn from thirteen purposively selected primary studies whose reported findings, verified against the source text, are read together. Bank capital is reported to reduce NPLs in one setting, to raise them in another, and to do both either side of an estimated threshold within a single study. Economic growth is the dominant determinant in one study, insignificant in a second, positively signed in a third, and reverses its causal direction with the horizon in a fourth. Regulation is reported to bear no predictable relationship with NPLs. The paper distinguishes heterogeneity of effect from a change of channel, proposes a two-part test for the latter, and finds that the strongest cases rest on within-study variation in which sample, measure and estimator are held fixed. It argues that treating regulation, capital and provisioning as exogenous—across capital frameworks and across incurred-loss and expected-credit-loss provisioning regimes—risks misstating the relationships on which supervisory evaluation, provisioning policy and stress testing depend.