The effects of ownership on bank efficiency in Latin America

Abstract

In recent years many countries have privatised their state-owned banks and encouraged foreign investment. This article investigates the roles of state and private ownership and foreign and domestic ownership on the performance of banks across Latin America. Using a range of financial and economic ratios, data envelopment analysis (DEA) and regression modelling, the study reveals that by 2001 there was surprisingly little difference in performance between state-owned and privately-owned banks and between foreign and domestically-owned banks. The study also reports significantly different levels of bank performance in different Latin American countries, suggesting that country differences outweighed ownership differences in explaining performance.

Notes

¹Cuba is the exception. It remains a centrally planned economy.

²When considering the impact of privatization on bank efficiency, Nakane and Weintraub (2005) have also examined the different forms of privatization. In their article about bank privatization and productivity in Brazil, they claim that straight privatization appears to be a superior strategy to restructuring, followed by privatization.

³Haber and Kantor (2003) went further and claimed that the lack of foreign entry at the outset of privatization of banks in Mexico ‘proved fatal for the banking system’ (p. 28), because banks tended to be undercapitalized and Mexican bankers had, generally, little bank management expertise.

⁴Some studies, e.g. Bonin et al. (2005) adopt a panel data analysis but using ownership data for only one year during the period covered. This approach we believe is flawed because ownership is not consistent over time in Latin America.

⁵Argentina was excluded from the study due to the fact that it was in a severe financial crisis in 2001 and this could bias the results. Small Caribbean islands, such as Bermuda and the Cayman Islands, were also excluded because of the atypical environment in which their banks operate, dealing largely with international funds.

⁶The method to calculate IRS was adopted from Unite and Sullivan (2003, p. 2329) and involves calculating the difference between the ratio of interest income on loans to total loans and the ratio of interest expense on total deposits to total deposits.

⁷A translog cost function analysis was also estimated using restricted ‘Seemingly Unrelated Regressions’, to allow for the influence of the cost share equations, derived from the Shepard's lemma, as , where i=1, 2 and SH_i denotes the two cost shares functions. The results obtained are in line with those obtained from the DEA and SCF analyses reported below. For reasons of space we do not report these results, but they can be obtained from the authors.

⁸The translog cost function is based on a flexible form production function that places few restrictions on the underlying production technology. It is, therefore, the most appropriate cost function to use when studying firms that may have differing production technologies.

⁹The detailed results can be obtained from the authors.

¹⁰The interest rate data were extracted from the IMF's database, International Financial Statistics.

¹¹The CRS results can be obtained from the authors.

¹²SCF results are also available on this. However, as the results obtained provide generally similar conclusions to those obtained by DEA, they are not reported here. They are available from the authors at the reader's request.

¹³Brazilian banks are used as a benchmark in the analysis due to the fact that Brazil is the biggest economy in Latin America.

¹⁴In four out of the eight countries presented, the sample does not include banks which are at least 50% state owned.

¹⁵These were computed from the Bankscope data and therefore correctly only reflect concentration within the sample of banks included.

¹⁶Country variables for Colombia and Panama and institutional variables for specialized governmental credit institutions and the fiscal burden of the government were included in an alternative regression but were statistically insignificant. These results are not reported.

¹⁷In both sets of estimated results, some of the variables have been excluded. They proved to be statistically insignificant in the initial estimation of the models and were therefore dropped. A restriction test on each set of estimation was carried out and confirmed the validity of the restrictions imposed.