A bootstrap approach to analyse productivity growth in European banking

Abstract

This paper analyses productivity growth for European banks over the 1995–2001 period. In contrast to previous literature, our study covers the majority of current European Union (EU) countries—all except Greece and those joining the EU in 2004. We also use resampling methods so as to gain statistical precision, which turns out to be especially important due to the limitations of the database. In order to be consistent, we use additional nonparametric methods to disentangle why productivity differentials might exist. Results show that productivity growth has occurred in most countries, mainly due to improvements in production possibilities. The bootstrap analysis yields further evidence given that for many firms and countries productivity growth or decline is not statistically significant. The two-stage analysis provides some extra insights, suggesting that the relevance of environmental variables found in other studies focusing on efficiency could be lessened when focusing on productivity.

Keywords:

• bank
• bootstrap
• Malmquist productivity index

Acknowledgements

All three authors acknowledge the financial support of the Ministerio de Ciencia e Innovacion (SEJ2005-08269/ECON, ECO2008-03813/ECON and ECO2008-05908-C02-01/ECON). Emili Tortosa-Ausina also acknowledges the financial support of the Fundació Caixa Castelló-Bancaixa (P1-1B2008-46) and Generalitat Valenciana (PROME-TEO/2009/066). A previous version of this article circulated as a FUNCAS working paper (207-2005). We also thank an anonymous referee for helpful comments.

Notes

1 This assertion parallels one of the claims by Färe et al (1994b) in their study on the productivity of 17 arbitrarily selected OECD countries.

2 Note that we do not differentiate between the EU and Europe. Furthermore, the notion EU refers to the EU-15, not the enlarged EU. In our particular setting, we will speak about EU-14 as Greece was not considered.

3 See, for instance, the article on the history of the MPI by Grosskopf (2003).

4 The existing literature is already substantial, including studies for many countries such as Australia (Avkiran, 2000), Greece (Noulas, 1997), Japan (Fukuyama and Weber, 2002), South Korea (Gilbert and Wilson, 1998), Italy (Casu and Girardone, 2004), Portugal (Canhoto and Dermine, 2003), Norway (Berg et al, 1992), Spain (Tortosa-Ausina et al, 2008), Turkey (Isik and Hassan, 2003), or the US (Wheelock and Wilson, 2009).

5 However, these assumptions are far less restrictive than the requirements demanded by parametric methods such as SFA.

6 The choice of B is limited by the speed of the computer but we need at least 2,000 replications for yielding a reasonable approximation to confidence intervals.

7 Each value of efficiency is estimated under constant returns to scale because the index only correctly measures the productivity change if the true technology exhibits constant returns to scale everywhere (Grifell-Tatjé and Lovell, 1995).

8 By reversing t₁ and t₂ in Equation (13) we obtain

9 A more exhaustive decomposition of MPI, as in Simar and Wilson (1999b), might imply that although changes in productivity might be significant, the sources of productivity could themselves be nonsignificant.

10 See Simar and Wilson (1999a) for a comprehensive guideline of the algorithm.

11 We only illustrate the case of the Malmquist productivity index, but the procedure is identical for each component making up the index.

12 A11 computations have been performed using codes created by the authors. Specifically, we have applied Matlab codes for the computation of the different components of the Malmquist indices and the efficiency scores, as well as Gauss codes for the bootstrap procedure. Our codes are available upon request.

13 The DGP is defined by considering i.i.d. (independent and identically distributed) observations within each subgroup, but not necessarily across them.

14 In order to test significance for the geometric mean of each country we require the empirical distribution of each index, as in Equation (17), but for each country. This is obtained by averaging for each index, the corresponding B bootstrap values of all national banks and afterwards we test for the presence of unity in their percentile confidence interval.

15 If we had followed the output oriented version of the Malmquist TFP change index, interpretation of results would reverse. This is possible due to the constant returns to scale (CRS) assumption.

16 Similar possibilities exist for the case of productivity decline.

17 Following Simar and Wilson (1999b), when averaging bank estimates over time, we also average the corresponding bootstrap values over time to obtain estimates of significance for the complete period.

18 A previous version of the paper contained tables providing summaries for each particular country. They have been skipped in this version for space reasons but are available from the authors upon request.

19 Such as the recent takeover of Abbey National by Banco Santander, Spain's largest bank.

20 We consider six economic neighbourhoods, namely: (i) The Netherlands, Belgium, and Luxembourg; (ii) Sweden, Finland, and Denmark; (iii) UK and Ireland; (iv) Austria and Germany; (v) France and Italy; and (vi) Portugal and Spain. Whereas in most cases they are clearly a reality, in some others they are not so apparent.

21 We consider three groups: (i) The Netherlands, Belgium, Luxembourg, Italy, France, Germany, UK, Ireland, and Denmark; (ii) Portugal and Spain; and (iii) Sweden, Finland, and Austria. Although the first group contains countries which joined the EU at different points in time, we consider that sufficient time has passed to assume their financial systems are not going to become much more integrated—at least in the short run.

22 We have confined the analysis to the analysis of Malmquist productivity indices, omitting their decomposition into efficiency and technical change, so as to save space. Results are available upon request.