This paper examines the relationship between banks’ business models, bank performance and banks’ business risks by employing both panel regression and long short-term memory (LSTM) neural networks. The bank business model definition addresses the fundamental endogeneity problem by strictly separating causal constructs from outcome variables. Building on previous efforts to develop continuous classifications, the authors address the limitations of categorical classifications and introduce a continuous ‘bank business model index’ (BBMI) that captures banks’ strategic balance sheet structures from retail- to market-oriented banks. An empirical analysis of 111 Eurozone banks from 2014 to 2023 reveals that retail-oriented banks outperform market-oriented banks in terms of bank performance. A mediation analysis demonstrates that bank business risks serve as a significant partial mediator in the relationship between banks’ business models and their performance. The study examines risk at the business model level rather than at the institution level. The empirical results show that the LSTM network achieves a higher prediction accuracy than panel regression. Using marginal effects, the authors introduce an approach to address the ‘black box’ limitation of LSTM networks and examine possible nonlinear effects of risk. The results provide valuable insights for regulators, bank managers and researchers to conduct cause-and-effect analyses with a measurable bank’s business model construct and marginal effects to explain deep learning outputs. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.
Herdt et al. (Sun,) studied this question.