Bank Branch Location: a Count Analysis

Abstract

This study undertakes an analysis of the association between socio-economic variables and the spatial distribution of bank branches in South Africa. To analyse retail bank branch location, parametric Poisson, negative binomial, Poisson-hurdle, and finite-mixture count models that accommodate unobserved heterogeneity have been estimated with a data set of bank branches in municipalities in South Africa. The key finding is that aggregate income in a municipal area is a statistically significant determinant of the number of bank branches. The results also suggest that the four core banks tend to cluster their branches in the same areas—suggesting the possibility of oligopolistic collusion. A more recent entrant locates its urban branches slightly away from the cluster and in some areas where the core banks have no presence. In addition, several socio-economic variables, some of which are unique to South Africa, do not appear to affect branch location.

Situation des filiales de banques: une analyse à comptage

RÉSUMÉ La présente étude entreprend une analyse de l'association entre des variables socio-économiques et la répartition de filiales de banques en République Sud Africaine. Pour analyser l'emplacement des filiales du secteur de la banque de détail, on a évalué des modèles de comptage de Poisson, de probabilités binomiales négatives, au taux étalon de Poisson, et aux mélanges finis, qui tiennent compte de l'hétérogénéité non observée au moyen d'un ensemble de données de filiales de banques dans des municipalités en Afrique du Sud. La principale conclusion est que le revenu total dans une zone municipale est un déterminant significatif statistiquement du nombre de filiales de banques. Les résultats semblent également indiquer que les quatre principales banques ont tendance à grouper leurs filiales dans les mêmes régions, ce qui semblerait indiquer la possibilité d'une collusion oligopolistique. Une banque arrivée plus récemment a réparti ses filiales légèrement hors de ce groupement, o[ugrave] les principales banques ne sont pas représentées. En outre, plusieurs variables socio-économiques, certaines desquelles sont particulières à l'Afrique du Sud, ne semblent pas affecter l'emplacement des filiales.

Ubicación de sucursales bancarias: un análisis de recuento

RÉSUMÉN Este estudio realiza un análisis de la asociación entre las variables socioeconómicas y la distribución espacial de sucursales bancarias en Sudáfrica. Para analizar la ubicación de sucursales de bancos minoristas, se han estimado modelos de recuento como el paramétrico de Poisson, binomial negativo, Poisson-hurdle (modelo en dos partes) y mezcla finita, que conllevan una heterogeneidad inadvertida, con un conjunto de datos de sucursales bancarias de municipalidades de Sudáfrica. El descubrimiento clave consiste en que los ingresos agregados en un área municipal representan un determinante estadísticamente significativo del número de sucursales bancarias. Los resultados también sugieren que los cuatro bancos principales tienden a agrupar sus sucursales en las mismas áreas, indicando la posibilidad de una colusión oligopolística. Un participante más reciente sitúa sus sucursales urbanas algo más alejadas del grupo, y en algunas áreas donde los bancos principales no están presentes. Asimismo, ciertas variables socioeconómicas, algunas de las cuales son exclusivas de Sudáfrica, no parecen afectar a la ubicación de la sucursal.

Keywords:

• Bank branches
• count analysis
• South Africa

JEL CLASSIFICATION:

• C31
• N27
• R32

Acknowledgements

This study has been funded by a grant (project 6185) from the Trustees of the FinMark Trust entitled: ‘An Economic Analysis of Financial Intermediation Instrument Location in South Africa’. I wish to thank Mark Napier, Chief Executive of the FinMark Trust, and the Trustees for commissioning this study. I am particularly grateful to Mark Dowdeswell of the University of the Witwatersrand who worked very ably as a researcher on this project. I am grateful to the South African Banking Risk Intelligence Centre (SABRIC) for providing access to its data on branch location. I wish also to thank and acknowledge Fabian Obgonna, Michael Adeyemo, and Bongiwe Kanzi for the development of the ECONBRANCH database. Finally, I wish to thank the anonymous referees of this journal—their comments have improved this paper. None of the above are responsible for any errors or omissions which may remain.

Notes

1. The core banks in South Africa (whose bank branch network are the subject of this study) have been existence in some form or other for about a century. See Okeahalam (2004) for a fuller description and some financial metrics of their size. They have strong commercial ties and cross-holdings with a number of other firms in Europe, the USA and Asia. As a result, they provide the full range of corporate finance, capital markets and securities trading products in the wholesale market. The retail market provides a similar product suite as would be found in most high street banks in Europe and the USA. The national payments system complies fully with the prudential requirements of the Committee of Payment and Settlement Systems of the Bank for International Settlements (see Okeahalam, 2003). In addition, the Johannesburg Stock Exchange is (currently) the 14th largest stock exchange by market capitalization. Yet, despite this, a large proportion of the citizens of South Africa cannot make ready access to a bank branch.

2. Okeahalam & Jefferies (1999) use co-integration techniques to illustrate that the South African financial market is integrated with the major markets of the world. However, South Africa also has an exchange control regime which limits the level of fixed and portfolio capital flows. Accordingly, South African banks were (with hindsight, fortunate) not to be able to participate in a number of activities such as the sub-prime market—which have contributed to the current financial crisis. While there was an increase in credit provision over the 2002–2006 period, the introduction of the National Credit Act of 2007(which coincided with the beginning of the financial crisis) led to a reduction in access to credit. Therefore, South African banks have not been as adversely affected by financial contagion as banks in some other parts of the world. Yet the global financial crisis is beginning to affect the real economy. Demand for natural resources and minerals has slowed and this is impacting upon bank balance sheets. In the retail banking market, there has been an increase in the level of credit defaults by clients of the four core banks. Over the 1-year period February 2008—February 2009 the level of non-performing loans increased from R11 billion (at current exchange rates approximately US$1.1 billion) to R24 billion (US$2.4 billion).

3. The Financial Sector Charter, which took effect in October 2003, stipulates certain access targets and the quality of physical access. Specifically, the distance to a branch for everyone should be less than 15 km and the distance to an ATM should be less than 10 km.

4. Crime is an issue that influences the supply and location of bank branches. Okeahalam (2006) notes that security measures taken by banks to prevent armed robberies and other forms of crime are a drag on the level of production efficiency of bank branches in South Africa. The police authorities in South Africa do not readily provide statistics on crime at the spatial level required to conduct an analysis that could be used to inform the relationship between crime and branch location. So although it would be an interesting factor to analyse, I have not been able to obtain statistics on crime at the municipal spatial level at which the bank branch location analysis has been conducted. As I show later in the results, income is the primary determinant of bank branch location. Yet intuitively it is likely that both crime and branch location are strongly influenced by the level of income in a spatial area. This is a view confirmed by Witt et al. (1999) and Chen (2009). Thus it could be argued that banks would still locate in places with high income, despite high crime levels. This is in fact what happens in South Africa—high-income areas have a large number of branches and, given the high levels of crime, banks take a range of measures to ensure that they can continue to operate in those areas. This is feasible, because, as Okeahalam (2007) illustrates, South African retail banks have market power. Therefore, it is likely that they are able to pass on to customers the increase in costs of operating in spatial areas with high crime levels.

5. A parametric approach was used as it was the only method that was revealed in the literature that would fully handle the count data. Hurdle and mixture models are necessarily parametric. A meaningful alternative to a fully parametric approach that could be used to evaluate stochastic models that are represented by a mixing distribution might be to use the methodology of Brännäs & Rosenqvist (1994)—who use a semi-parametric estimator in an overdispersed Poisson regression model.

6. Note that the overdispersion parameter has been parameterized slightly differently here. This has been done to preserve positivity and to allow for unconstrained optimization of the maximum likelihood function.

7. Mullahy (1986) provides an introduction to hurdle count regression models with an application to daily beverage consumption. Those hurdle finite mixture models differ from the finite mixture models introduced above—which allow mixing of both zeroes and positive counts from two or more densities.

8. The Poisson, NB1, and NB2 models for the aggregate counts (not reported here) show significant lack of fit, possibly due to heterogeneity introduced by ‘mixing’ the individual institutional distributions.

9. Capitec is the fifth largest retail bank and together the five banks account for almost 100% of the retail bank branches in South Africa. One large insurance company has opened a handful (literally) of bancassurance (insurance and banking) branches. The data constitute a complete record of all the bank branches that were open at the time when the database was compiled (end of 2006).

10. A list of metropolitan areas and district councils is provided below. Provinces are in capital letters; * = metropolitan. South Africa has nine Provinces, six metropolitan municipalities, and 46 district municipalities. WESTERN CAPE 1. Cape Town * 2. West Coast 3. Cape Winelands 4. Overberg 5. Eden 6. Central Karoo. EASTERN CAPE: 7. Nelson Mandela * 8. Cacadu 9. Amatole District Municipality 10. Chris Hani 11. Ukhahlamba 12. OR Tambo 13. Alfred Nzo. FREE STATE: 14. Xhariep 15. Motheo 16. Lejweleputswa 17. Thabo Mofutsanyane 18. Northern Free State. NORTHERN CAPE: 19. Namakwa 20. Karoo 21. Siyanda, 22. Frances Baard 23. Kgalagadi. NORTH WEST: 24. Bojanala Platinum 25. Central 26. Ruth Mompati 27. Southern. GAUTENG: 28. West Rand 29. Johannesburg * 30. Sedibeng 31. Ekhuruleni * 32. Metsweding 33. Tshwane * LIMPOPO: 34. Mopani 35. Vhembe 36. Capricorn 37. Waterberg 38. Sekhukhune. MPUMALANGA: 39. Gert Sibande 40. Nkangala 41 Ehlanzeni. KWA ZULU NATAL: 42. Amajuba 43. Zululand 44. Umkhanyakude 45. uThungulu 46. Umzinyathi 47. Uthukela 48. Umgungundlovu 49. iLembe 50. eThekwini * 51. Ugu 52. Sisonke.

11. In terms of GDP, provinces in South Africa are considerably smaller than are the states in the USA and regions in Europe. Data from the International Monetary Fund (IMF) World Economic Indicators Fact Book indicates that in 2005 the GDP of South Africa was US$264.3 billion. The GDP of Gauteng province, which is the largest GDP of the nine provinces, is approximately US$68.7 billion. In comparison, the GDP of the US state with the smallest economy, Louisiana, was US$79 billion. Three of the six metropolitan municipalities in the country are located in Gauteng.

12. The SABRIC data set includes 1,835 bank branches, of which 507 are ABSA branches, 537 are FNB branches, 357 are Nedbank branches and 434 are Standard Bank branches. The SABRIC database does not contain data on Capitec branches. The estimation results from the SABRIC GIS data and all the estimates of the models which were not selected by the AIC/BIC procedure are available from the author or from www.baragh.com

13. Estimates from Statistics South Africa indicate that in early 2007 the population of South Africa was 47.9 million, an increase from the census of 2001 count of 44.8 million. Black Africans are the majority at just over 38 million, making up 79.6% of the total population. The white population is estimated at 4.3 million (9.1%), the coloured population at 4.2 million (8.9%) and the Indian/Asian population at approximately 1.2 million (2.5%). Nine of the country's 11 official languages are African. The population distribution by language is Afrikaans 13.3%, English 8.2%, isiNdebele 1.6%, isiXhosa 17.6%, isiZulu 23.8%, Sepedi 9.4%, Sesotho 7.9%, Setswana 8.2%, siSwati 2.7%, Tshivenda 2.3%, and Xitsonga 4.4%.

14. The term ‘black’ is used as an alternative to the emotive apartheid classification (‘non-white’). Therefore, ‘black’ covers people of African (black), mixed origin (coloured) and Asian (Indian) origin. It is also in keeping with the race classification used by Stats SA.

15. A recent announcement by ABSA that it intends to invest R220 million to launch 115 new outlets, of which 80% will be in previously underserved areas, is a welcome initiative and may encourage the other three core banks to follow suit.