Bank Loan Loss Provisions Research: A Review

We review the recent academic and policy literature on bank loan loss provisioning (LLP) to identify several advances in the literature, to highlight some challenges in LLP research and suggest possible directions for future research with some concluding remarks. Among other things, we observe some major advancement in country-specific and cross-country analyses and substantial interaction between LLPs and existing prudential, accounting, institutional firm characteristic, cultural, religious, tax and fiscal framework. We observe that managerial discretion in provisioning does not necessarily generate LLP estimates that reflect the true and underlying economic reality of banks’ credit risk exposure but rather managerial discretion in provisioning is strongly linked to income smoothing, capital management, signalling and other objectives. We also address several issues including the ethical dimensions of income smoothing, motivations and constrains to income smoothing, methodological issues in the bank loan loss provisions literature and the dynamic loan loss provisioning experiment. Moreover, we suggest several avenues for further research such as: finding a balance between sufficient LLPs which regulators want versus transparent LLPs which standard setters want; the sensitivity of abnormal (specific and general) LLPs to changes in equity; the persistence of abnormal LLPs following CEO exit; country-specific interventions that induce LLP procyclicality in emerging countries; investigating LLP behaviour in the post-financial crisis sample period; the impact of Basel III on banks’ provisioning discretion; LLP behaviour among systemic and non-systemic financial institutions; etc. We conclude that, because provisioning models are only as good as the assumptions underlying such models as well as the accuracy of the inputs included in such models, regulators need to pay attention to how much discretion banks and lending institutions should have in determining reported provision estimates, and this has been a long standing issue