Efficiency in the Greek Banking Industry: A Comparison of Foreign and Domestic Banks

Abstract

This study uses a sample of foreign and domestic banks operating in Greece during 1999–2004 to examine the impact of ownership on efficiency. We estimate an input oriented data envelopment analysis (DEA) model under variable returns to scale with inputs and outputs selected on the basis of a profit‐oriented approach. The results indicate an average pure technical efficiency equal to 0.7325 showing that the banks in sample could improve their efficiency by 26.75%. Over the same period, scale efficiency was equal to 0.6830. The comparison of the efficiency scores by group of ownership shows that domestic banks have higher pure technical efficiency and lower scale efficiency; however, the differences are not statistically significant. A DEA window‐analysis confirms the results of the cross‐section estimations. We also estimate a Tobit regression model but consistent with the univariate results we find no evidence to support the argument that ownership has a statistically significant impact on efficiency.

Key Words:

• Banks
• DEA
• Efficiency
• Foreign
• Greece

JEL classifications:

• D61
• G21

Notes

1. See for example Noulas (2001), Tsionas et al. (2003), Pasiouras (2008a), Pasiouras et al. (2008) and Delis et al. (2009), among others.

2. For example, using data from 1993 to 2000 on 49 nations, Berger et al. (2004) find that the cost and profit efficiency ranks of community banks are associated with faster economic growth. In a more recent study, Hasan et al. (2007) also provide evidence of a positive association between bank‐level profit efficiency and regional economic growth, using a large sample of banks operating in 254 different NUTS 2 regions in the EU‐25. See Levine (2005) for a more general discussion on the relationship between efficiency of financial intermediaries and economic growth.

3. For other recent applications of DEA in banking see among several others Webb (2003), Havrylchyk (2003), Drake et al. (2006), Pasiouras (2008a, 2008b).

4. Our analysis focuses on the 1999–2004 period due to data (un)availability in the website of the Hellenic Bank Association. Data before 1999 and after 2004 were available only for a very limited number of cases, not allowing us to proceed to a meaningful analysis. Information from the 2004 annual report of the Bank of Greece indicates that at end‐2004 there were 23 foreign and 21 domestic banks in Greece, showing that our sample is quite representative.

5. As an anonymous referee suggested, additional inputs/outputs that could be considered are loan loss provisions (LLP) and off‐balance‐sheet items (OBS). Unfortunately, such information was not available in our case for foreign banks. Furthermore, including OBS could be inconsistent with past studies that use the profit‐oriented approach. To some extent, non‐traditional activities such as OBS are captured by net commission income as in Rogers (1998) and Stiroh (2000) among others. Furthermore, keeping the number of inputs and outputs low, allows us to meet various rules of thumb. Soteriou and Zenios (1998) and Boussofiane et al. (1991) state that the number of units should be larger than the product of the number of inputs and outputs. Dyson et al. (2001) argue that the number of units should be at least twice the product of the number of inputs and outputs. Nunamaker (1985) mentions that the sample size should be at least three times larger than the sum of the number of inputs and outputs.

6. We would like to thank an anonymous referee for recommending this analysis.

7. Experimenting with other values of α in the range of 0.4 to 0.8 does not alter the presented curve.

8. We would like to thank an anonymous referee for suggesting this analysis.

9. Since our sample covers fives years in total, it was not possible to consider a five‐year window as in Asmild et al. (2004) and Webb (2003). Charnes et al. (1985) actually use a window of 3 months, since their analysis is performed on a monthly basis between October 1981 and May 1982.

10. The p‐values in the case of the K‐W test were as follows: Window 1 (PTE: 0.218, SE: 0.275), Window 2 (PTE: 0.697, SE: 0.030), Window 3 (PTE: 0.471, SE: 0.263), Window 4 (PTE: 0.431, SE: 0.121). To conserve space we do not present details on the efficiency scores, which are available from the authors upon request.

11. Bank’s size is measured by the logarithm of total assets. As for time, we use dummy variables with 1999 as the base year. Several other bank specific characteristics could be potentially used in the regression as control variables. However, we include only size and ownership for two reasons. First, to minimize (to the extent that it is possible) potential endogeneity problems. Second, to preserve the quality of the regression due to the small number of observations in the sample (see Weill, 2003). During the Tobit regression we excluded 3 observations due to missing values for total assets.

12. The same results were obtained in a univariate Tobit regression with ownership being the only independent variable.