Measuring bank profit efficiency

Abstract

To date, work concerned with the potential determinants of credit institutions’ profit inefficiency levels has addressed this issue in either a single-step or multi-step process. In the former, inefficiency scores are conditioned by region and bank-specific indicators, while in the latter, generated inefficiency scores are subsequently regressed on a set of potential correlates. The approach proposed here allows these issues to be explored jointly in a statistically consistent manner. The model is applied to a sample of banks from Ireland, the UK, Canada and Australia.

Acknowledgement

The authors are grateful to Karl Whelan and CBFSAI seminar participants for comments on an earlier draft.

Notes

¹ A similar application in an agricultural context was proposed by Rahman (2003).

² See Vander-Vennet (2002) for a discussion of the merits of the alternative profit function vs. the standard specification in the context of credit institutions.

³ Produced by Bureau Van Dijk (BVD).

⁴ International financial systems are frequently distingushed between the Anglo-Saxon model (i.e. the UK, North America, Canada, Australia and New Zealand, etc.) and the continental European model. The difference between these models lies in the manner in which ownership, control and regulation are organized. For more on this, see Franks and Mayer (1994).

⁵ The complete list of institutions used is available, upon request, from the authors.

⁶ We use total assets instead of total employees as the relevant denominator owing to the absence of employee data for many credit institutions in the sample. As noted by Maudos et al. (2002) this definition can be interpreted as labour cost per worker adjusted for differences in labour productivity as PE/TA = PE/NE × NE/TA where PE is personnel expenses, NE is number of employees and TA is total assets.

⁷ We treat financial equity capital as a fixed input as it is slow to adjust and price data, for the variables are often difficult to measure.

⁸ OECD Economic Outlook Number 75–Statistical Annex tables.

⁹ The definition of specific and general provisions varies across countries as do bank discretions in provisions for loans. Provisions tend to be based on historical averages of the institutions, consequently, they do not exhibit much within-group temporal variation. In addition, loss provisions were not available for all institutions which would have reduced the size of the sample further.

¹⁰ The correlation coefficient between the GDP and unemployment rates was −0.02% suggesting that multi-collinearity between these variables was not a potential problem. We also explored the use of other variables within the 1 vector such as output gap estimates, money market rates and the SD of money market rates for each country. However, our results were relatively unchanged by their addition/omission and given the relatively large number of estimated parameters within the inefficiency model already (59), we elected to omit these variables.

¹¹ The p-values from the Likelihood Ratio (LR) tests for the stochastic and deterministic specifications were 0.622 and 0.444, respectively. Full regression results are available from the authors upon request.

¹² They are available from the authors upon request.

¹³ This results in 18 credit institutions in both the big and small category and 19 in the medium group.

¹⁴ The price of labour in UK institutions is, on average, 58%, 20% and 2% greater than the relative figures of Irish, Australian and Canadian banks for the sample.

¹⁵ The significance of the country-specific dummies, particularly in the profit function, may suggest evidence of different production technologies across the different countries.