Banking efficiency and competition in low income countries: the case of Uganda

Abstract

There is a concern that the state dominated, inefficient and fragile banking systems in many low-income countries, especially Sub-Saharan Africa, are a major hindrance to economic growth. In this context, this article systematically analyses the impact of the far-reaching banking sector reforms undertaken in Uganda on banking sector competition and efficiency. Using models of banking competition and efficiency that have been predominantly estimated in industrial countries, we find that the level of competition has significantly increased and has been associated with a rise in efficiency of the sector. Moreover, on average, larger banks and foreign-owned banks are more efficient than others while smaller banks have fallen back in efficiency with the increase in competitive pressures.

Acknowledgements

Thanks are due to Leslie Teo, Yuji Yokobori, Martin Cihak, Richard Podpiera, Peter Allum and Jan Mikkelsen for useful comments. The views expressed in this article do not necessarily represent those of the IMF.

Notes

¹ This paper draws partly on Hauner (2005) and Peiris (2005).

² Starting in October 2001, the BOU aligned its loan classification criteria with international standards and issued circulars in April and July 2002 to adequately capture NPLs in the system.

³ The NPL ratio rose through end-2003 because of the default of a large trading company. Changes in NPLs can be relatively large because of the concentration of exposures.

⁴ Funds placed abroad largely comprise deposits in the correspondent accounts of the large foreign-owned banks (FB).

⁵ Ugandan banks have higher overhead costs than comparable banks in Kenya, partly because they do more outreach and have recently invested in physical infrastructure, such as branches and ATMs. Cross-country comparisons show that smaller banks have higher overhead costs because they find it difficult to exploit economies of scale and scope. This is confirmed by a significant positive correlation between the share of deposits and loans below U Sh 3 million in total deposits and loans and overhead costs, as well as the relatively low ratio of loan and deposit volume per branch in Uganda.

⁶ See Gelos and Roldos (2002) and Buchs and Mathisen (2005) for weaknesses of the equilibrium test.

⁷ See Gelos and Roldos (2002) (Central Europe and Latin America), Belaisch (2003) (Brazil), Levi Yeyati and Micco (2003) (Latin America) and Claessens and Leaven (2003) (cross-country).

⁸ The BOU categorizes banks as small or large based on their asset size as of a particular date, which in this case was 2002. A FB is one with majority foreign ownership.

⁹ Note that the H-statistic in the post-privatization period is statistically significantly different from the pre-privatization level at the 5% significant level in the first two specifications and at the 10% significance level under the interest-revenue specification (Table 3).

¹⁰ See Hauner (2005) for a survey of studies in industrial countries.

¹¹ Empirical applications necessarily refer to relative efficiency. When outputs are held fixed and inputs are to be minimized, relative efficiency is attained by a production unit if and only if according to the available evidence none of the inputs can be reduced without increasing at least another input.

¹² The most common parametric approaches are the stochastic frontier approach (SFA), the thick frontier approach (TFA) and the distribution-free approach (DFA). The main trade-off between parametric and nonparametric approaches concerns their assumptions on random errors and the functional form of the cost frontier. While DEA fails to distinguish between inefficiency and random errors, it does not presume a particular functional form of the frontier. Parametric approaches, in turn, distinguish between random errors and inefficiency, but do so along the lines of somewhat arbitrary assumptions about their respective distributions and, in addition, impose a particular functional form, which, if misspecified, risks overstating inefficiency. In practice, bank efficiency studies have used nonparametric and parametric methods similarly frequently (Berger and Humphrey, 1997).

¹³ For further details on DEA, the reader is referred to Seiford and Thrall (1990). The DEAP software used here is described in Coelli (1996).

¹⁴ A ‘radial’ reduction cuts all inputs by the highest proportion λ possible for all of them at given output levels. After radial reduction, there could be remaining ‘slack’ in some inputs (Farrell efficiency is necessary but not sufficient for Pareto efficiency). Here, this problem is circumvented by amending the linear program so that it meets Koopman's efficiency, which also fulfils the Pareto criterion. See Lovell (1993) for a more extensive treatment and Coelli (1996) on how the software used here solves the problem.

¹⁵ Alternatives for dealing with M&A would have been to delete all banks involved, resulting in a selection bias, or to add up the figures of the merging banks for the years before the merger, counterintuitively implying a decline in the combined assets after consolidation.

¹⁶ When some banks push the frontier inward, and others are left behind, then average efficiency decreases.

¹⁷ Bootstrapping would be one way to check whether this change is statistically robust. Bootstrapping is a way to analyse the sensitivity of efficiency scores relative to the sampling variations of the estimated frontier.

¹⁸ It should, however, be noted that stress testing conducted during the recent FSAP Update mission suggest that even a 50% reduction in government securities yield would not severely damage small banks.

¹⁹ See Molyneux et al . (1996) and Claessens and Laeven (2003) among others.

²⁰ In fact, the equilibrium test for periods further away from the privatization period are more robust in accepting the null hypothesis of banking system equilibrium.