Bank fragility in Africa: GMM dynamic panel data evidence

Abstract

This study investigates the impact of bank-level and macroeconomic variables on bank fragility using a dynamic two-step GMM panel estimator on 433 banks in 46 African countries over the period 1997–2012. The study finds that both bank characteristics and macroeconomic variables are key drivers of bank fragility. The past experience of higher levels of non-performing loans (NPLs) significantly and positively determines current levels of NPLs. The growth of gross loan is negative and significant but economic growth leads to higher NPLs. The equity to assets ratio and the log of assets of banks are negatively associated with NPLs suggesting their potential to provide buffers to banks. Equally, total assets reduce bank fragility. These findings have important policy implications. The study shows that credit risk management initiatives, bank operation oversight and regulations should not be restricted in the times of financial crises, even during positive economic growth episodes in the business cycle.

Keywords:

• Bank characteristics
• macroeconomic variables
• regulation
• NPLs
• GMM

JEL Classification:

• G01
• G21
• G28

Acknowledgements

This work was supported by DFID and ESRC, Grant Ref. ES/N013344/2, and the AXA Chair at SOAS University of London, to Murinde. Useful comments were provided by seminar participants at the African Development Bank, University of Leicester and University of Birmingham. We retain responsibility for all errors.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 See Koutsomanoli-Filppaki and Mamatzakis (2009); Moshirian (2008) and Tropeano (2010).

2 Another study by Salas and Saurina (2002) includes both microeconomic and macroeconomic variables as determinants of NPLs in the Spanish banking system. They also suggested that any future changes in NPLs can be highly identified by bank-specific variables more in saving than in commercial banks.

3 Baltagi (2001) documented that in a dynamic relationship, fixed effect or random effect estimation techniques provide biased and inconsistent estimates, particularly when N is quite larger than T; normally a fixed and random effect estimation technique applies in a static relationship.