000			01688cam a2200181 4500500
005			20251012014549.0
041			_afre
042			_adc
100	1	0	_aDurand, Pierre _eauthor
700	1	0	_a Le Quang, Gaёtan _eauthor
700	1	0	_a Vialfont, Arnold _eauthor
245	0	0	_aNonlinearities and interactions between variables: Insights from interpretable machine learning methods for banking regulation
260			_c2025.
500			_a94
520			_aThe aim of this article is to illustrate the usefulness of interpretable machine learning methods in the specific case of banking economics. In particular, we rely on a gradient boosting model to determine the optimal regulatory capital ratio within the framework of prudential banking regulation. To this end, we develop, on the one hand, a classification model whose purpose is to determine the impact of capital ratios on the probability of bank default, and, on the other hand, a regression model aimed at assessing the costs to banks’ performance associated with increased capital requirements. Using various interpretability tools (permutation importance, Shapley values, partial dependence plots, accumulated local effects), we found the following optimal values: 15% for the capital adequacy ratio and 10% for the leverage ratio. The determination of these values relies on highlighting the non-linear effects and interaction effects that characterize the relationships between the various variables studied.
786	0		_nRevue d'économie politique | 134 | 6 | 2025-03-17 | p. 893-922 | 0373-2630
856	4	1	_uhttps://shs.cairn.info/journal-revue-deconomie-politique-2024-6-page-893?lang=en&redirect-ssocas=7080
999			_c1531363 _d1531363