Cost efficiency of Kazakhstan and Russian banks: results from competing panel data models¹

Abstract

In this paper, we estimate cost efficiency of the Kazakhstan and Russian banks. A stochastic frontier approach based on a panel data for 2002–6 is used. The Kazakhstan banking system is traditionally assumed to be more advanced compared to the Russian system. Empirically we do not find any significant differences in the cost efficiency of banks between these two countries during the period of our study. This result is found to be quite robust across several alternative and competing models. We also find that many of the banks in both countries operate below their optimal size.

Keywords:

• stochastic frontiers
• cost efficiency
• persistent inefficiency
• scale economy

JEL classification:

• D21
• G21
• F30

Acknowledgements

The authors are grateful to Subrata Sarkar, Sergei Golovan, Alexander Karminsky, Oleg Eismont, Laurent Weill, Iikka Korhonen, Zuzana Fungacova, participants of the BOFIT seminar, the participants of the 7^th INFINITI Conference on International Finance in Dublin (June 2009), and the participants of the XI European Workshop on Efficiency and Productivity Analysis in Pisa (June 2009), for helpful comments. The usual caveats apply.

Notes

1. This paper replaces the earlier version of the paper under the title ‘Bank cost efficiency in Kazakhstan and Russia' (Peresetsky, 2010).

2. Bureau van Dijk Electronic Publishing, http://www.bvdinfo.com/Products/Company-Information/International/Bankscope.aspx

3. www.raexpert.ru/researches/banks/retail/part1/.

4. As calculated by AFN data, this does not correspond to Tables 1 and 2. The reason is that Table 2 presents data on consolidated balance sheets of Kazakhstan banks, but Table 1 presents data from the AFN site, based on unconsolidated balance sheets. The main difference is due to two of the ‘Big Three’ (Kazkommertsbank, Bank TuranAlem). Expert RA data differ from AFN data for the same reason.

5. This specification was suggested by Kumbhakar, Ghosh and McGuckin (1991) and Battese and Coelli (1995).

6. Models similar to Kumbhakar and Heshmati (1995) were also used in Kumbhakar and Hjalmarsson (1993, 1995).

7. Note that the Kumbhakar, Lien and Hardaker (2011) model (Model 6 in their paper) is for a production function whereas our present model is for a cost function.

8. To conserve space here we show the plots for truncated normal distributions in Models 1 and 2 (scale3 and scale6). Plots for other cases, viz., scale1, scale2, scale4 and scale5 are similar.