Portal de Periódicos da CAPES

The Effect of Board Size and Composition on the Efficiency of UK Banks

2011; Taylor & Francis; Volume: 18; Issue: 3 Linguagem: Inglês

10.1080/13571516.2011.618617

1466-1829

Sailesh Tanna, Fotios Pasiouras, Matthias Nnadi,

Efficiency Analysis Using DEA

Abstract We examine a sample of 17 banking institutions operating in the UK between 2001 and 2006 to provide empirical evidence on the association between the efficiency of UK banks and board structure, namely board size and composition. Our approach is to use data envelopment analysis to estimate several measures of the efficiency of banks, and then to use panel data regressions for investigating the impact of board structure on efficiency. After controlling for bank size and capital strength, we find some evidence of a positive association between board size and efficiency, although this is not robust across all our specifications. Board composition, by contrast, has a robustly significant and positive impact on all measures of efficiency. Key Words: Board SizeBoard CompositionBanksCorporate GovernanceEfficiencyNon-ExecutivesJEL classifications: G21G34 Notes 1. Adams and Mehran (Citation2008) provide evidence and explanations for a positive effect of board size on performance (proxied by Tobin's Q) for the US banking industry, although, as discussed in Section 2, the evidence for European banks is not positive. Similarly, the evidence on the impact of board composition is mixed. 2. Berger and Humphrey (Citation1997) in their survey of the efficiency of financial firms identified 130 studies dealing with frontier techniques, of which 69 employed the non-parametric Data Envelopment Analysis (DEA) that we use in this study, while Fethi and Pasiouras (Citation2010) identify over 150 DEA applications between 1998 and early 2009. 3. Berger and Mester (Citation1997) use a sample of US commercial banks and examine the relation between a bank's highest holder registration for public trading with SEC and the proportion of stock owned by insiders and outsiders with cost and profit efficiency. Isik and Hassan (Citation2003) investigate whether the affiliation of the CEO and public trading of banks have an impact on efficiency in the Turkish commercial banking sector. Amess and Drake (Citation2003) investigate UK building societies but focus on the relationship between total factor productivity change and executive remuneration rather than on board size and composition and efficiency. There are other studies, such as Hardwick et al. (Citation2003), Zelenyuk and Zheka (Citation2006), and Destefanis and Sena (Citation2007), that relate corporate governance issues with efficiency but provide evidence from non-banking sectors in the UK, Ukraine, and Italy respectively. There are also several bank-level studies that define corporate governance more broadly and examine the link between ownership and bank efficiency (e.g. Berger et al., Citation2005). These studies actually compare the performance of different types of banks (such as cooperative with savings and commercial banks, government-owned with private banks, listed with non-listed banks, foreign with domestic banks) and consequently do not examine the board structure aspects of corporate governance mechanisms. 4. Lipton and Lorsch (Citation1992) recommend a number of board members between seven and eight, which is supported also by Jensen (Citation1993). However, board size recommendations tend to be industry-specific, since Adams and Mehran (Citation2003) indicate that bank holding companies have board size significantly larger than those of manufacturing firms. 5. The investigation of the impact of corporate governance mechanisms on bank risk taking (see, e.g., Akhigbe and Martin, Citation2008; Pathan, Citation2009) is outside the scope of this paper. However, considering the interest of regulators on this topic, we discuss in the concluding section the relationship between efficiency and risk, and propose an avenue for future research. 6. The sample includes the following banking institutions:Alliance & Leicester Commercial Bank plc, Arbuthnot Latham & Co. Ltd, Barclays plc, Bradford & Bingley plc, Consolidated Credits Bank Ltd, Co-operative Bank plc, HSBC Bank plc, Julian Hodge Bank, Reliance Bank Ltd, Ruffler Bank plc, Schroder & Co Ltd, Standard Chartered Bank, Standard Life Bank Ltd, Unity Trust Bank plc, HBOS plc, Lloyds TSB Group plc, and Royal Bank of Scotland Group plc. Thus we include most of the large UK banks, while the excluded institutions (due to data unavailability) are smaller and most specialized ones such as Tesco Personal Finance Ltd, Vanquis Bank Ltd, Southsea Mortgage & Investment Co Ltd, Marks and Spencer Financial Services plc, Smith & Williamson Investment Management Ltd, and so on. Thus their omission from the analysis is also justified on the basis of their specialization and it should not bias the obtained results. Apparently, some of the banks in our sample conduct business only or mainly in the UK (e.g. Arbuthnot), while others have an international presence (e.g. HSBC). However, as mentioned in the main text, they are all classified as UK ones in the Bank of England's (Citation2006) "Institutions included within the United Kingdom banking sector – nationality analysis". A point raised by an anonymous referee is that banks with an international presence may use different production technologies, an issue that it is important in the context of efficiency assessment. While acknowledging this issue, it should be mentioned that it was not possible to split the sample and estimate separate frontiers for at least two reasons. The first is the already small sample we have had to use. The second is that, after estimating separate frontiers, it is by definition then not appropriate to compare the efficiency of the banks with international presence with those of the non-international banks. Furthermore, we believe that the issue of international or no international presence can have only a marginal impact on the results of our study. The reason is that the banks with international presence will tend to be larger than the ones with a domestic focus. The estimation of efficiency under a VRS assumption ensures (with OTE being the only exception) that the ith bank is not "benchmarked" against units that are substantially larger than it (i.e. possibly banks with an international presence and different technology), although it may be compared with smaller units. 7. The alternative is to estimate an output-oriented measure of technical efficiency, which addresses the question: "By how much can output quantities be proportionally expanded without altering the input quantities used?" (Coelli et al., Citation2005, p. 137). The vast majority of banking studies obtain efficiency estimates under the input-oriented approach (Fethi and Pasiouras, Citation2010). 8. According to Maudos et al. (Citation2002), "Of all the techniques for measuring efficiency, the one that requires the smallest number of observations is the non-parametric and deterministic DEA, as parametric techniques specify a large number of parameters, making it necessary to have available a large number of observations" (p. 511). 9. It should be noted that, under constant returns to scale, the input- and output-oriented models will provide the same value. The results differ only when variable returns to scale is assumed. However, as pointed out by Coelli et al. (Citation2005), since linear programming does not suffer from statistical problems such as simultaneous equation bias, the choice of orientation is not as crucial as it is in the case of econometric models, and in many instances, it has only a minor influence upon the scores obtained (Coelli and Perelman, Citation1996). 10. Some studies propose the use of an additional output, namely non-interest income (e.g. Tortosa-Ausina, Citation2003) to account for off-balance sheet and other non-traditional activities of banks. Non-interest income, however, is generated from both on-balance sheet and off-balance sheet activities. With limited data availability, it was not possible for us to determine the sources of non-interest income. However, if we assume that an important proportion of non-interest income is generated by on-balance sheet business, then its effect would already be captured in the "other earning assets" output. In that case, including both other earning assets and non-interest income in the model would lead to a large amount of double counting. To avoid this difficulty, we estimate a traditional model that includes loans and other earning assets, which is the most common approach followed in the literature. 11. Given that DEA efficiency is a relative measure, it might be appropriate to use a balanced sample to avoid potential bias from the entry and exit of banks over the period of examination. However, including only banks with complete data across the whole period would reduce our sample size further. We therefore rely, as in the vast majority of DEA studies in the banking literature, on the use of annual frontiers. Isik and Hassan (Citation2002) argue that this approach has two advantages. First, it is more flexible and thus more appropriate than estimating a single multiyear frontier for the banks in the sample. Second, it alleviates, at least to an extent, the problems related to the lack of random error in DEA by allowing an efficient bank in one year to be inefficient in another, under the assumption that the errors owing to luck or data problems are not consistent over time. Nevertheless, to partly address any concerns, we estimate our DEA models and present the results after including in all the annual frontiers, banks for which we had at least one year of corporate governance data. Obviously, this reduces the variability of the sample composition among the years. 12. For example, Apergis and Rezitis (Citation2004) and Rezitis (Citation2006) examine six banks, Pasiouras et al. (Citation2008) examine 10 banks, Chu and Lim (Citation1998) examine as few as six banks, Neal (Citation2004) examines 12 banks, while in a UK study, Drake (Citation2001) examines only nine banks. 13. For the banks with efficiency score equal to one, we subtract a small figure (i.e. 0.005) from BEFi,t to allow this transformation. 14. Some studies use simultaneous equations estimation methods like two- and three-stage least squares to examine interdependence of relationship between corporate governance variables and firm valuation. However, as Banhart and Rosenstein (1998) point out, theory provides little guidance as regards the specification of the models, and the misspecification of any of the equations in a system may result in serious bias in all of the equations, whereas OLS tends to be less sensitive to misspecification error (Rhodes and Westbrook, Citation1981). 15. Equity could potentially be included as an input in DEA to control for different risk characteristics of banks. However, adopting this approach would be a deviation rather than the norm in the banking literature that uses DEA for the estimation of efficiency. We are actually aware of four studies that have used equity as an input (Chu and Lim, Citation1998; Luo, Citation2003; Pasiouras, Citation2008b; Sturm and Williams, Citation2004), but these studies examine technical rather than cost efficiency. One problem with the calculation of cost efficiency is to obtain a reliable and accurate measure of the input price (or cost) of equity. In view of this difficulty, we have used equity to assets in the second stage of our analysis, consistent with Casu and Girardone (Citation2004), Casu and Molyneux (Citation2003), Isik and Hassan (Citation2003), Pasiouras (Citation2008a), among others. 16. The time trend T takes the value 1 for 2000, 2 for 2001, and so on. We also estimated our specifications with year dummies instead of the time trend. The results remain the same. To conserve space, we do not present them here, but they are available from the authors upon request. 17. The yearly averages of board size are as follows: 12.17 (2001), 12.23 (2002), 12.58 (2003), 11.93 (2004), 11.73 (2005), and 12.15 (2006). 18. The yearly averages of board composition are as follows: 61.51% (2001), 55.67% (2002), 56.80% (2003), 55.73% (2004), 51.74% (2005), and 57.06% (2006). Averages in other studies are 64.4% (Staikouras et al., Citation2007), 68.7% (Adams and Mehran, Citation2003), 69% (Adams and Mehran, 2008), 71% and 81% (Busta, Citation2007). In Zulkafli and Samad (Citation2007), the proportions for individual countries range from 9.09% (Taiwan) to 60.46% (Korea), with an overall average of 32.29%. 19. The reduction in the sample size is 33 observations due to missing values for BCOMP. We also re-estimated the model of column 1 with 46 observations as in models 2 and 3, and found an insignificant effect of LNBSIZE on all measures of efficiency, suggesting that the impact of board size is possibly affected by the smaller sample size. It is possible that with a larger sample, both board size and composition may have a positive effect on efficiency, since the low correlations in Table 1 indicate that the results are not susceptible to multicollinearity problems. 20. We would like to thank an anonymous referee for making this point and for motivating the analysis discussed in this subsection. 21. In the case of Model 1, the coefficients (t-test) for LNBSIZE are equal to 1.453 (3.148) for the input-oriented and 1.451 (3.184) for the output-oriented specification. In the case of Model 2, the corresponding results for BCOMP are 0.019 (1.921) and 0.019 (1.860) for the input- and output-oriented specifications respectively. 22. The coefficient estimates of LNBSIZE and BCOMP included simultaneously in the regressions for profit-orientated efficiency (i.e. Model 3) are 2.080 (t-test = 3.552) and 0.017 (t-test = 3.162) in the case of the input-oriented model, and 1.789 (t-test = 2.727) and 0.013 (t-test = 2.381) in the case of the output-oriented model.