Copyright is owned by the Author of the thesis. Permission is given for a copy to be downloaded by an individual for the purpose of research and private study only. The thesis may not be reproduced elsewhere without the permission of the Author. AN EXAMINATION OF NEW ZEALAND BANK EFFICIENCY A dissertation presented in partial fulfilment of the requirements for the Degree of Doctor of Philosophy m Banking Studies At Massey University Palmerston North New Zealand David William Lethbridge Tripe 2005 Abstract: This research explores the effic iency of the New Zealand banking system over the period 1 996 to 2003 using Data Envelopment Analysi s (DEA). DEA is used as a lack of data on prices and the relatively smal l cross-sect ions (because of the l imited number of banks in the Tew Zealand market) pose difficulties for the use of parametric methods. This i s the first major research to make use of the data-set provided under the New Zealand banking disclosure regime, and the first major attempt at contrasting the relative efficiency of banks in Austral ia and New Zealand. The research discusses the problems of analysis of efficiency in smal l banking markets and proposes a solution through use of panel data. Analysis on this basis highl ights problems that arise from changing enviromnental condit ions ( specifical ly from changes in the general l evel of interest rates), but a lso produces a reasonably consistent ranking of the efficiency of New Zealand banks. The research finds that equity IS an important input to the study of bank efficiency, and that it is a cause of differences in relative efficiency between New Zealand and Austral ian banks. Acknowledgements I need to thank a number of people for their support and assistance in getting this research completed. Firstly I want to thank members of my extended family for their support, particularly my wife Marion, and my sister-in-law Robyn. I owe thanks also to my col leagues, pm1icularly Claire Matthews and Chris Moore, other people in the Department who have assisted with support and suggestions. part icularly Fong Mee Chin, and people from elsewhere within the University, past and present, who have provided support, such as Astrid Baker, Jack Dowds, Abdul lah Mamun, Hector Perera, and Andrew Trl in. Thanks are a lso due to my supervisors, Larry Rose and Srikanta Chatterjee. I would also l ike to thank Kim Wil l iams and Sharon Henderson for practical suppm1. Thanks are also due to some current and former students, with whom I have been able to col laborate on research, and from whose effm1s I have been able to learn about the Jssues I have been researching. I pm1icularly thank James Fahey. Benjamin Liu. David MmTay. Rohan Pathirage, David Sara, Robert Smith, Huong Minh To and Prasad V edula. There is also a wider network of col leagues interested in banking and efficiency studies, who have provided encouragement and support : important people have been Tecmi A vkiran, Al ien Berger, Rayna Brown, Lesl ie Hul l , A lexander Karmann, Justin Ken, Knox Lovel l , Geof Mortlock, John Owens, Chris Plantier, Paul Rouse, Mi l ind Sathye, Jolm Simpson, Michael Skully, Greg Walker, John Whitelaw, and Andrew Wo11hington. 11 Table of Contents Page Abstract : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i Acknowledgen1ents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i i Table o f Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i i i L ist of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v i L ist of F igures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix 1 . 1 . 1 1 .2 1 .3 1 .4 1 . 5 2 . 2 . 1 2 .2 ? " _,.) 2.4 2 . 5 3 . 3 . 1 3 .2 ") ") .) . .) 3 . 4 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ] Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 The proble1n . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . 2 Aim and objectives of this research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The importance of thi s research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 An outl ine of the dissertation . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . .. . . . . . . . 3 New Zealand and its banking system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 4 The situation prior to deregulation . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 The transfom1ation of the banking sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 0 1 996 and since . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 The banking sector and the New Zealand economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 What does this mean for research? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Previous research and theoretical issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 What is efficiency? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . .. . . . . . . . . . . .. . . . . . . . . . . . . . . . 30 Efficiency in F inancial l nstitutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Approaches to efficiency measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Practical i ssues in F inancial Institution efficiency measurement . . . . . . . . . . . . . . . . . . 56 3 .4 . 1 Returns to scale in DEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 3 .4.2 Model orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 9 3 .4 .3 Specification of inputs and outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3 .4 .4 The impact of environmental factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 Ill 3 .4 .5 Choosing the data set in which efficiency i s to be measured . . . . . . . . . . . 65 3 . 5 Surnmary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4. Data and Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4 . 1 The data set studied . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4 .2 Data: \Nhat i s reported . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4 .2 . 1 The Income Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4 .2 .2 The Balance Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4 .2 .3 Sun11nary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4 .3 General methodological i ssues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1 4 .4 The individual studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4 .4. 1 Use of the cost to income ratio relative to multivariate approaches for measuring bank efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4 .4.2 Effic iency trends through time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.4.3 A lternative input variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4 .4.4 An introductory cross-country study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 4 .5 Sun1n1ary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5 . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 5 . 1 Use of the cost to income ratio to measure bank efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5 . 1 . 1 The Malmquist Index approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5 . 1 .2 The panel data approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 02 5 . 1 .3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 04 5 . 2 Efficiency trends through time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 06 5 .2 . 1 DEA with gross interest expense as an input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 06 5 .2 .2 Testing the impact of interest rates on efficiency scores . . . . . . . . . . . . . . . 1 1 3 5 . 3 Alternative input variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 6 5 .3 . 1 Including equity as an input along with gross interest expense . . . . . 1 1 7 5 .3 .2 Using an adjusted interest cost figure as an input . . . . . . . . . . . . . . . . . . . . . . . . . . 1 20 5 .3 .3 Inc lusion of off-balance sheet items as an input . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 23 5 . 3 .4 Analysis of al l the banks together . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 26 5 .3 .5 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 3 8 5 .4 The cross-country study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 42 5 .4 . 1 Does a common frontier apply? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 42 iv 5 .4 . 2 A model with equity a s an input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 47 5 .4 . 3 Model without equity as an input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 50 5 .4 .4 What do the results from these two models mean? . . . . . . . . . . . . . . . . . . . . . . . . 1 52 6 . Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 54 6 . 1 A review of the research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 54 6 .2 What this research has not done . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 5 8 6 .3 Limitations of this research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 60 6.4 Future research challenges and opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 62 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 65 V List of Tables Page Table 1 : Costs and Revenues for New Zealand banks in 2003 70 Table 2: Reporting of non-interest income and expense under New Zealand's disclosure regime 76 Table 3 : umber of sectors used for reporting funding and lending by sector under New Zealand ' s disclosure regime 79 Table 4: Descriptive statistics for inputs and outputs, study comparing Malmquist and DEA of panel data 85 Table 5 : Correlations between input and output variables, study comparing Malmquist and DEA of panel data 86 Table 6: Descriptive stati stics - individual bank studies 89 Table 7 : Correlations between inputs and outputs - individual bank studies 9 1 Table 8 : Correlations between i nputs and outputs, all banks together 93 Table 9 : I nputs and outputs - cross-country study 94 Table 1 0 : Statistics on Input data ( in AUD) - cross-country study 95 Table 1 1 : Correlations between inputs and outputs - cross-country study 96 Table 1 2 : Cost to Income ratios for New Zealand banks 99 Table 1 3 : Results from appl ication of Malmquist Index - annual data for ew Zealand banks 1 00 Table 1 4 : Results from panel data analysis - annual data for New Zealand banks 1 02 Table 1 5 : Relative changes in banks' efficiency scores, based on panel data results 1 04 Table 16: Efficiency scores for studies of individual banks- gross interest expense as an input (Constant returns to scale) 1 07 Table 1 7 : Effic iency scores for studies of individual banks- gross interest expense as an input (Variable returns to scale) 1 08 Table 1 8 : Estimated scale efficiency scores for studies of individual banks? gross interest expense as an input vi 1 09 Table 1 9 : Returns to scale status- model studying individual banks, gross interest expense as an input 1 1 2 Table 20 : Regression results - logits of efficiency scores from constant returns to scale model s for individual banks, gross interest expense as an input 1 1 5 Table 2 1 : Efficiency scores for studies of i ndividual banks - gross interest expense and equity as inputs 1 1 8 Table 22 : Efficiency scores for studies of individual banks with adjusted interest cost and equity as inputs 1 2 1 Table 23 : Comparison of means of models for individual banks with gross interest expense and adjusted interest expense as inputs 1 23 Table 24: Efficiency scores for individual bank models with gross interest expense as input, and including off-balance sheet items as output 1 24 Table 2 5 : Efficiency scores for individual bank models with adjusted interest expense as input, and including off-balance sheet items as output 1 25 Table 26 : Average efficiency scores for banks studied individually before and after addition of off-balance sheet items as an output: gross interest expense as an input 1 25 Table 27 : Average efficiency scores for banks studied individually before and after addition of off-balance sheet items as an output : adjusted interest expense as an input 1 26 Table 28 : Efficiency scores from model with all banks studied together, \IVith equity and gross interest expense as inputs (CCR) 1 27 Table 29 : Efficiency scores from model with al l banks studied together, with equity and gross interest expense as inputs (BCC ) 1 29 Table 30 : Average scale efficiency estimates and returns to scale status from model with al l banks together with equity and gross interest expense as inputs 1 30 Table 3 1 : Difference in efficiency between banks - equity and gross interest expense as inputs 1 32 Table 3 2 : Efficiency scores from model with al l banks together with equity and adjusted interest expense as inputs (CCR) 1 34 Table 3 3 : Difference in efficiency between banks - equity and adjusted interest expense as inputs 1 3 5 Table 34 : Efficiency scores from model with a l l banks together with equity and gross interest expense as inputs, off-balance sheet items among outputs (CCR) 1 35 v i i Table 3 5 : D ifference in efficiency between banks - equity and gross interest expense as inputs, off-balance sheet items as an output 1 36 Table 36 : Efficiency scores from model with al l banks together with equity and adjusted interest expense as inputs, off-balance sheet items among outputs (CCR) I 3 7 Table 37 : Difference in efficiency between banks- equity and adj usted interest expense as inputs, off-balance sheet items as an output 1 3 8 Table 38 : Impact of incl usion of Australian regional banks in efficiency comparisons against Austral ian major banks 1 44 Table 39 : Impact of incl usion of New Zealand banks in efficiency comparisons against Austral ian banks 1 45 Table 40: Frequency with which types of banks appear in reference sets 1 46 Table 4 I : Cross-country study, model 1 results ( i . e . with capital as an input) - constant returns to scale 1 4 7 Table 42 : Cross-country study, model 1 results ( i .e . with capital as an input)- variable returns to scale 1 48 Table 43 : Cross-country study, model I results ( i .e . with capital as an input) - measures of scale efficiency I 49 Table 44: Cross-country study, scale effects for model I (with equity as an input) 1 49 Table 45 : Cross-country study, model 1 results- summary 1 50 Table 46 : Cross-country study, model 2 results ( i .e . without capital as an input) 1 5 1 Table 4 7 : Cross-country study, model 2 results - summary I 5 1 Table 48 : Spread ratios to show the impact of use of equity capital as an input 1 52 Table 49: Median efficiency scores for Australian majors and their New Zealand operations 1 53 VIII List of Figures Page Figure 1 - 90 day bank bi l l rates between New Zealand and Australia 23 Figure 2- Total asset trends - major banks 24 Figure 3 -Cost to income ratios in New Zealand and Australia- major banks 2 8 F igure 4 - Relative returns on equity - Australia and New Zealand 29 F igure 5 - X-efficiency and its decomposition into technical and al locative effic iency 3 3 Figure 6 - Efficiency scores through time - New Zealand banks 1 03 Figure 7 - Individual bank efficiency scores with inputs gross interest expense and non-interest expense (CCR model) 1 1 0 F igure 8 - Individual bank efficiency scores with inputs gross interest expense and non-interest expense (BCC model) 1 1 1 F igure 9 - Individual bank efficiency scores with equity included as an input 1 1 7 F igure 1 0 - Spread ratios for individual banks showing the effect of adding equity as an input F igure 1 1 - Individual bank efficiency scores - model with adjusted interest 1 1 9 expense 1 20 F igure 1 2- Efficiency score trends for all banks together - gross interest expense as an input (CCR model) 1 28 Figure 1 3 - Volumes of transactions through the New Zealand payments system ix 1 59 1. Introduction 1.1 Background At an immediate impression, the structure of the New Zealand banking market at the beginning of 2005 differs somewhat from the perfectly competitive model . There is a small group of four banks (sharing five brands), which dominate the retai l banking market. Because the banks' pricing of their products is inclined to be relatively close to each other, thi s commonly leads to allegations of monopol i stic practice, that there is no real competition between the banks, and that the banks are able to price their products in ways that allow them to extract excessive profits from a public that lacks alternative channels to access financial services. This issue i s aggravated in the New Zealand context by the dominance of foreign, particularly Austral ian, ownership. As at 30 June 2004, 98 .5% ofNew Zealand banks' assets were foreign owned, with 86.6% of banking system assets under Australian ownership. This is incl ined to raise questions in the publ ic mind as to whether the major Austral ian banks are using their dominance of the New Zealand market to charge excessive prices, and to deprive New Zealanders of easy access to low-cost and efficient financial services. This sort of perception is incl ined to be encouraged by the relatively high profits being achieved by the Australian banks in their New Zealand businesses, whether these profits are measured by return on assets or return . I on eqUJty. At the same time, banks in New Zealand, and some of the Australian-owned banks in particular, have put a lot of emphasis on their costs in recent years, and have sought to reduce these. Significant reductions have been achieved, both in terms of reduced cost to income ratios and reduced ratios of costs to assets. Thi s has been accompanied by extensive branch closures, with the number of retail bank branches in New Zealand having reduced from 1 ,447 in 1 995 to 832 by 200 1 . Thi s has been another i ssue of concern to the wider publ ic , who have been incl ined to associate the reduct ion in the 1 This is discussed funher in Section 2 . 5 below. scale of banks' branch networks with a general deterioration m customer service standards . 1.2 The problem This l i sting of negative perceptions for the performance of the New Zealand banking system has not, in general, been tested in any empirical way, which means that a number of questions are left about the performance of the ew Zealand banking system. If it is assumed that an efficient banking system is a good thing, there should be a publ i c benefit in the banking system becoming more efficient . If the costs incurred in the overall operations of New Zealand banks have been reduced as they have sought to enhance their profitabi l i ty , does this mean that the banks have become more efficient? Does the strong profit performance of the New Zealand banks mean that the banks are efficient? Are there any special issues that arise as a result of the high degree of foreign ownership of the New Zealand banking system? 1.3 Aim and objectives of th is research The aim of this research i s already indicated in the l isting of problems in the previous section. and in the dissertation? s title : how efficient are New Zealand banks, and have they become more efficient over time? As efficiency is inclined to be a relative measure, thi s wi l l also entai l attempting some assessment of the efficiency of the major banks in Australia. Thi s research therefore looks at v,;hat efficiency is , at the different approaches that may be appl ied to its measurement, and at the problems that arise with these types of analysis. l t reviews the ew Zealand banking system in greater depth, to gain an understanding of how the cunent posit ion has been reached, and to appreciate i ssues that arise with the data used to try and measure the banks ' efficiency . It affirms the desirabi l ity of using a multivariate approach to measurement of bank efficiency, rather than merely relying on ratios. The analysis uses a number of different approaches, and the dissertation seeks to compare and contrast the results obtained. 2 1.4 The importance of this research This research i s important as the first major study of New Zealand bank efficiency, and also as the first major piece of research that has been undertaken making use of the data provided by New Zealand banks' disclosure regime. It is also the first major research to try to compare and contrast the relative performance of the Austral ian and New Zealand banking systems. There are other ways in which thi s research should also make a contribution. The study of bank efficiency has led to relatively few points of agreement in relation to method and interpretation of results, and this dissertation seeks to outline some approaches that may be fol lowed in small economies with relatively concentrated banking systems (and with relatively smal l number of banks in particular). It wi l l also make suggestions for i ssues which have posed problems for other researchers in areas of bank efficiency, particularly in the use of Data Envelopment analysis, which i s the method used for this research. 1.5 An outl ine of the d issertation Because much of the dissertation reports on the analysis of quantitative data, it is written in the third person. Where the author' s views are being repo11ed. these are identified accordingly. The rest of the dissertation i s structured as fol lows. Chapter Two revie,vs the New Zealand banking system, and at how it has developed to its cunent situation. Chapter Three contains a d iscussion of the meaning of efficiency, and then reviews relevant previous approaches to researching it in financial institutions (and banks in particular). Chapter Four outlines the data that is used, primari ly in respect of the New Zealand banking system, but also in respect of Australia, and sets out the individual studies that are undertaken in an attempt to answer the questions posed. Chapter Five rep011s the results obtained, and di scusses these. Chapter Six concludes and provides some suggestions for areas where fol low-up research might be unde11aken in the future. 3 2. New Zealand and its banking system One of the remarkable features of the New Zealand economy has been the extent of deregulation that has occurred, particularly since a change in government fol lowing the 1 984 general election. Deregulation has occuned in many sections of the economy, with the changes in the financ ial sector being among the most noteworthy.2 S ignificant changes were effected in monetary policy and the way in which it was implemented, but there were also changes in the activities financial institutions were permitted to unde11ake, and a consequent series of changes in the ways in which they have operated. This disse11ation focuses on one pm1icular class of financial i nstitutions, the banks, a group of firms within the financial sector which has grown substantial ly since 1 984, and which now have a much more imp011ant role in the financial system.3 A new system for bank registration came into effect in 1 987, whi l e further changes have included the adoption of what i s regarded international ly as a new and innovative approach to the prudential supervi sion of banks. Thi s is based on public disclosure, and the information it provides has contributed a major source of data for this research. More or less contemporaneously with what was happening in New Zealand, a process of deregulation was a lso occurring in Australia, although the process started a l ittle earl ier in Australia than it did in ew Zealand, and occurred at a rather l ess frenetic pace. Deregulation in Austral ia also impacted quite significantly on the banking sector, and there are some interesting parallels. This dissertation thus also has regard to the Australian market, the importance of which is highl ighted by the fact that, as at 3 1 March 2003 , 66 .3% of New Zealand bank assets were under Austral ian ownership.4 By 3 1 December 2003 , fol lov.:ing , the acquisit ion of the (formerly British- 2 See Evans et al ( 1 996) for a review of the deregu lation/economic reform process in New Zealand. 3 Th is would seem to be particularly the case with loans- see Thorp & Ung (2000), Table 2 (p 20). No particular d i fference is evident in terms of the proportion of household fi n ancial assets held with deposit-taking institutions, at least up to 2000. Note, however, that today ' s five largest banking groups are identified as holding a significant ly increased proportion of household financ ial assets. 4 See Tripe & Matthews ( 2 00 3 ) for a review of the international expansion into New Zealand and other markets undertaken by the major A ustral ian banks. 4 owned) National Bank of New Zealand by the ANZ Banking Group Limited, the proportion of Australian ownership had increased to 86.4%. The fol lowing sections of this Chapter l ook at aspects of the New Zealand financial system prior to deregulation, and at the changes that followed, through to the structure of the New Zealand banking system as at early 2003 . This outl ine provides a background against which the research reported in this d issertation can be set, whi le also identifying and describing the main subjects for the research. 2.1 The s ituation prior to deregu lation The New Zealand banking system as it existed prior to 1 984 was firmly segmented. Fol lowing the merger of two Austral ian banks with New Zealand operations (the Bank of New South Wales and the Commercial Bank of Australia) to form Westpac in 1 982 , there were four trading banks . 5 There were al so savings banks, of three types, sh011-term money market dealers, merchant banks and building societies, with each group having its own defined role. In addition to these, there was a range of other institutions operating in the financial sector, some of which appeared to owe their role to restrictions placed on the operations of the c lasses of financial institutions reported above. Over and above the strict segmentation of the ew Zealand financial system there was a range of other regulatory control s that appl ied. There were direct controls on a range of their activities, and mandatory ratios applying to different c lasses of assets . There were strict l imitations in the entry of new fim1s to the more controlled sectors (banking in particular). and l imits on foreign ownership, although the trading banks were already mostly foreign owned, as they had been more or less ever since the Union Bank (a British overseas bank) had establ ished a branch in Wel l ington in 1 840. The four trading banks that operated in 1 984 were the ANZ Banking Group (New Zealand) Ltd (ANZ), Bank of New Zealand (BNZ), the National Bank of New 5 These were the banks that provided services to both personal and business c ustomers, and which operated the infrastructure for cheque c learance in the New Zealand economy. 5 Zealand Limited (NBNZ), and the Westpac Banking Corporation (Westpac) . The ANZ was incorporated in New Zealand and had a New Zealand sharemarket l i sting, with approximately 25% of its shares owned by New Zealand investors, but the control l ing shareholding was with its Australian parent bank.6 The BNZ had been owned by the government since 1 945, and was the government' s transactional bank. The NBNZ had been original ly constituted as a British overseas bank, but the remaining independent shareholding was purchased in the late 1 960s to make the bank a wholly owned subsidiary of Lloyds Bank Ltd (Holmes. 1 999) . Westpac operated as a branch of its Australian parent bank, a publ icly l isted company on the Australian stock exchange. A market controlled by four banks may commonly be assumed to be uncompetitive. although there i s no necessity for that to be the case. In practice, however, competition between the banks m New Zealand was severely constrained. They operated together a jointly-owned computer system, Databank Systems Limited, which not only provided clearing for interbank transactions but which also provided account keeping and almost all other computing functions for the banks' New Zealand operations. Banks' interest rates on deposits and loans were subject to controls (most particularly during the years immediately prior to the 1 984 election when these controls were part of a broader wage and price freeze), which l imited banks ' abil ity to compete against each other on price. Fixed ratio requirements and other controls. which tended to limit banks' access to wholesale funding, l imited banks' abil ity to compete by mounting aggressive lending campaigns, and thus the maj or area of competition was incl ined to be service. Thi s was reflected most particularly in the extent of the banks' branch networks (Harper, 1 986, pp 26-27; Harper & Karacaoglu, ] 987, p 209) . Competit ion was also constrained by barriers to entry. At a regulatory level, entry of a new bank would have required an Act of Parliament, while at a practical, economic level, a new bank would have required a very substantial investment in the development of a branch network and technological infrastructure if it was to be able 6 The bank h ad been pressed to float a portion of its New Zealand business in the late 1 970s in response to a regulatory request: i t had previously operated as a branch of its parent bank. The change of status is recorded by Deane et al ( 1 9 8 3), p 3 J. The parent bank, A ustra l ia and New Zealand B anking Group Limited, was publicly l i sted on the Austra l ian stock exchange. 6 to compete against existing participants. 7 Even if thi s investment were made, the new banks would be l ikely to be severely hampered in acquisition of customer business from existing market participants (as was found by new entrants to the Australian market fol lowing the i ssuance of new banking l icences in 1 985) . 8 The trading banks did have some advantages relative to other classes of financial institutions, however. They were the only institutions which could offer cheque accounts for businesses, and they also had a monopoly of foreign exchange deal ing (up until August 1 983 ) . i chol l & King ( 1 985, p 23 7 ) note how profitable this was for the banks. There were three classes of savmgs banks: the government-owned Post Office Savings Bank (POSB). which was part of the Post Office, the trustee savings banks and the private savings banks (of which there were four, one owned by each of the trading banks). Each of these classes of savings banks was covered by its own legislation, and there were different restri ctions applying to each of them. The POSB had an extensive nation-wide network through the Post Office, whereas the 1 2 trustee banks each had their regional focus and did not compete \vith each other (they also shared computer systems). The private savings banks ,.vere separately incorporated companies. although they generally operated as diYisions of their parent banks and uti l i sed parent bank infrastructure. The POSB and trustee savings banks offered an extensive range of savings accounts and cheque accounts for the personal sector (they also dealt with non-profit organisations). Their lending was predominantly in residential mortgage loans, particularly after they were the beneficiaries of some l iberali sation in the mid 1 970s. The private savings banks tended to be restricted to passbook savings, but as Nicholl and King ( 1 985 , p 1 85 ) point out, other services were avai lable to private savings banks' customers through the parent trading bank. All the savings banks tended to 7 This was before the days when a viable E-banking option coul d be offered. There is also an argument that ex isting banks' branch networks m ight have constituted a barrier to entry for new participants. See Evanoff ( 1 98 8) . Th is is a lso consistent with the finding of To & Tripe (2002) that foreign banks with a long-standing presence in the N ew Zealand market performed more successfully. 8 See Ferguson ( 1 990), H ogan ( 1 99 1 ) . 7 have quite high proportions of their assets invested in govermnent securities, while their deposits were guaranteed (Grimes, 1 998, p 295) . The official shm1-term money market dealers were a category of institution that existed largely because of regulation, and which did not survive the removal of the regulatory boundaries which had protected them. They were d istinguished from the trading banks by being al lowed to pay interest on deposits for periods of less than 30 days, whi le also enjoying the benefit of access to the Reserve Bank as lender of last resort. The quid pro quo for this was being quite severely restricted in the assets in which they could invest (Nicholl & King, 1 985. p 1 88) . The merchant banking sector comprised the unofficial short-term money market, although the merchant banks undertook a range of other activities as wel l . A number of the merchant banks obtained foreign exchange deal ing l icences when these were l iberal ised in August 1 983 . The merchant banks often represented the corporate business activities of finance companies, which played an important role in New Zealand credit markets through their hire purchase and other instalment financing activities.9 New Zealand had originally had two types of building societies, permanent and terminating, but legislation in 1 98 1 prohibited the sale of further tem1inating shares . Fol lowing this. which stimulated some rational isation in the sector, the number of building societies had shrunk to 33 by 1 983 . although 66% of the sector' s assets were in the hands of just two mutually-owned societies, the United and Countrywide Building Societies (Nicholl & King, 1 985. pp 206-207) . In 1 982, 69.9% of the sector ' s assets were in mo11gage loans, mainly secured over residential property (Nichol l & King, 1 985 , p 209). In addition to the classes of financial institutions described above, there were also some special i st government-owned institutions. The Housing Corporation of New Zealand specialised in housing lending, with a particular concentration on first-home buyers and others who might otherwise have difficulty in purchasing accommodation. 9 As Grimes ( 1 99 8. p 296) notes, the development of finance companies had been ass isted by regu latory controls on the act i v it ies of other classes of financial institut ions. 8 The Rural Banking and Finance Corporation (Rural Bank) played a similar role in supporting farming, primary industry and related service industries. The Development F inance Corporation (DFC) targeted i ts activities at new and expanding industries, particularly those with an export or regional development focus (Ni choll & King, 1 985 , pp 22 1 -222) . There were some significant consequences of the restrictions imposed on the operations of the institutions l isted above. In particular, interest rate restrictions on commercial lending were circumvented by use of the commercial bi l l market. Would? be borrowers would issue a commercial bil l , which they would then get discounted, with the discount rate not being classed as a controlled interest rate . 1 0 I nvestors could buy commercial bi l ls and earn a non-taxable capital gain by holding them to maturity. Restrictions on housing lending were surmounted in a different way, using sol icitors' nominee companies. Borrowers would typical ly be able to obtain advances secured by a first mortgage, with the funding provided by the nominee company , which would have a pool of investors \Vho wanted to earn higher returns than were readi ly avai lable from the banks (and until 1 982, the interest rates were uncontrolled) . 1 1 The controls applying to the financial system were identified as having a number of negative consequences. Money and capital markets were perceived as underdeveloped, which made it difficult for both businesses and households to satisfy thei r credi t and investment needs (Harper. 1 986, p 28) . Financial inst itutions ? costs and margins w?ere perceived as higher than they would have been i f markets were more free, while the uneven application of monetary pol icy through quantitative controls meant that. for example, finance companies gained market share, despite being relatively high-cost institutions ( Harper, 1 986, pp 29-30) . The question then arose as to how deregulation might assi st in overcoming these negative effects. 10 This occurred part icu larly during the period of interest rate restrict ions associated with the wage and price freeze prior to the 1 984 general election. The commercia l bi l l market had original ly been developed by the merchant banks in the 1970s, as it provided a veh icle for them to fund their lend ing. 11 Thorp & Ung (2000. p 22) suggest that the sol ic i tors ' market had funded more than a th ird of total housing loans at its he ight in the early 1 970s, but by 1 984 the share of sol ic i tors' loans in household financial l iabil it ies was 1 7% (p 20). 9 2.2 The transformation of the banking sector When the new Labour government was elected in July 1 984, it found itself fac ing a financial crisis hinging on the value of the New Zealand dollar. Dealing with this i ssue required attention to a range of other i ssues as wel l . Although the initial focus of these was on monetary and exchange rate policy, a number of the changes had operational impacts on banks and other financial institutions, in terms of increasing their opportunities to compete against each other, and in providing them with new challenges in terms of servicing their customer base. Thus, over the fol lowing months, interest rate restrictions and l imitations on offshore borrowing were removed, as were compulsory ratios applying to financial institutions' balance sheets. The ew Zealand dol lar was floated, and the restrictions on the foreign ownership of financial institutions were abolished. In late 1 985 the Reserve Bank mmounced pol ic ies which would provide for the entry of new banks to the New Zealand market. 1 2 The banking market was thus l iberalised both in terms of the activities that banks could undertake and in terms of the number of banks. 1 3 The announcement that new banks would be al lowed into the New Zealand market became the Reserve Bank of New Zealand Amendment Act 1 986, which came into effect on 1 April 1 987 . The key to this was a system of bank registrat ion. with the four trading banks reclassified as registered banks as at I April 1 987 . Applications were opened for other institutions to seek regi stration. and, as of 22 July 1 987, a further seven institutions were granted registration. These were Barclays New Zealand Limited (Barc lays), Broadbank Corporation (Broadbank), C IBC New Zealand Limited (CIBC), Citibank NA (Citibank), the Hong Kong and Shanghai Banking Corporation (HSBC), Jndosuez New Zealand Limited (lndosuez), Macquarie Bank Limited (Macquarie), and NZI Financial Corporation (which became known in due course as NZI Bank) . 1 4 12 For a fu l l chronology, see Harper ( 1986), pp 40-4 3 . 13 One o f the principles underpinning this approach was the theory o f contestable markets, which implied that m arket pat1icipants would be obl iged to act in a competit ively optimal and efficient fashion (Doughty, I 986, p I I 3) . 14 Th is detai l and a significant portion of the detai l that fol lows derive from the Reserve Bank of New Zealand's ?'L i st of registered banks in New Zealand - past and present .. , available at hnp://www.rbnz.govt.nzlbanking/nzbanks/0029 I 34 .htm l#TopOfPage. 1 0 Except for CIBC (which was a subsidiary of the Canadian Imperial Bank of Commerce, one of the big five Canadian banks), all of these had some history of participation in New Zealand financial markets. 1 5 Bm?clays had had a merchant banking operat ion through New Zealand United Corporation. Citibank had entered New Zealand with a merchant banking authority during the early 1 980s, as had HSBC (operating as Wardleys). Indosuez had also had a merchant banking operation, with outside shareholders as had been required under the rules l imiting institutions to a maximum of 70% foreign ownership. NZI Financial Corporation was a long-established merchant bank, also with significant finance company activity, which was pa11 of the New Zealand Insurance group, then one of New Zealand' s major companies and l isted on the ew Zealand stock exchange. It was the only one of the new banks that was not part of a major international banking group. Broadbank was also a long-establi shed finance company and merchant banking group, which had been owned by the New Zealand conglomerate Fletcher Challenge unti l 1 985 . when it was sold to the Government Life Insurance Office. They, in turn, on sold 74% of it to the National Australia Bank group in early 1 997 . Broadbank Corporation Limited thus changed it name to National Australia Bank (NZ) Ltd (NAB(NZ)) in December 1 987, and this became the New Zealand vehicle for the National Australia Bank group (NAB. which was a major Austral ian bank, l isted on the Australian stock exchange). Other institutions followed in successfully obtaining New Zealand banking regi stration. The Countrywide Bui lding Society enl i sted some outside investors and then obtained registration as Countrywide Banking Corporation in December 1 987 . 1 6 Security Paci fi c New Zealand Limited (Security Pacific) , part of the Los Angeles based Security Pacific Bank, and which had been a shareholder in a New Zealm1d merchant bank, was also registered in December 1 987 . Bankers Trust obtained registration as BT New Zealand (Holdings) Ltd (BT) in June 1 988 . 1 5 N icholl & Sm ith ( 1 9 8 5 ) report some deta i l s o n t h e previously exist ing merchant bank s and foreign exchange dealers on pages 1 9 1 (Table 3 . 1 2) and 236 (Table 3 .3 9 ) respectively. 16 Th i s was associated with deregulation of t h e building soc iet ies - s e e Spencer & Carey ( 1 9 8 8 , p p 1 1 - 1 2) . 1 1 The forms in which these institutions were registered reflect a difference between the New Zealand and Austral ian approaches to the registration of new banks. In Austral ia, between World War I I and 1 985, foreign banks had only been allowed to operate as money market corporations: in 1 985 an initial l imit was imposed of 1 6 new foreign commercial banks, and one of the criteria for approval related to their potential contribution to Australia. New banks in Australia were also required to be separately incorporated, and many of them formed joint ventures with local participants, such as those between the Royal Bank of Canada and National Mutual, and between Chase Manhattan Bank and AMP. No such restrictions applied in New Zealand, although this was not perhaps clear at the time appl ications were being made. Thus it was perceived that C IBC, and perhaps a lso Security Pacific, had sought registration in New Zealand because of their fai lure to obtain registration in Australia. Many of the new banks were also establi shed as subsidiaries of their parent banks : as they came up against the d isadvantages of subsidiary status in terms of name recognition and access to funding, 1 7 many of the new entrants converted to branches of their parent banks. Fol lowing the opening up of registration. changes were also occurring in the savings bank sector. As restri ctions were removed, the separate exi stence of the private savings banks could no longer be justified, and these began to be reabsorbed into the operations of their parent banks. 1 8 The trustee savings banks began a process of consol idation and unification, although the Taranaki Savings Bank (now TSB Bank Limited, or TSB) preferred to remain separate, and achieved registration in its own right in June 1 989. On 1 July 1 986, the 1 1 remaining trustee banks announced the formation of Trusteebank Holdings Limited, which was to take over the functions of the Trustee Banks Association ofNew Zealand, and provide the basis for a new. more unified structure. This entity became Trust Bank New Zealand Limited on 30 September 1 988 (Burns, 1 989, p 1 69) . Government restrictions were relaxed to al low the banks to engage in a broader range of activities (Carew, 1 987 , p 44) . Trust Bank New Zealand Limited (Trust Bank) and its member banks were registered in December 1 989. 1 7 Some o f t h e n e w entrants sought to overcome thi s d isadvantage b y obtain ing formal letters of guarantee from their parent banks, which m ight be used to support a Trust Deed. 1 8 This process was finally completed in 1993/94 (Thorp & Ung, 2000. p 32). 1 2 While that process was occurring, however, the largest of the banks, the Auckland Savings Bank (now ASB Bank Limited, or ASB), decided to withdraw from the path being fol lowed by the other trustee banks, and it was fol lowed in this by the Westland Savings Bank (Westland Bank). In early 1 989, the community trust that (by now) owned ASB sold 75% of the bank to the Commonwealth Bank of Austral ia (CBA). 1 9 The ASB was thus able to accelerate its path towards registration, which it obtained in May 1 989, with Westland Bank (which had been the smallest of the trustee savings banks, and which depended on ASB for a number of services) fol lowing in March 1 990. The operations of the POSB had initially been total ly integrated with those of the Post Office, but when the Post Office was corporatised, it was spl i t into three state-owned enterprises, New Zealand Post, Telecom, and PostBank (representing the banking operations) . The government identified PostBank as able to be privat ised, and it was purchased by ANZ in late 1 988 . PostBank obtained registered bank status in August 1 989, although the ANZ continued to run it as a separate bank for a number of years after that. A number of other institutions also obtained banking regi stration, general ly with the objective of supporting other financial services businesses. The Austral ian conglomerate Elders was the ultimate parent company of Elderbank Limited (Elderbank) , which obtained initial registration in March 1 989. Austral ian l ife insurer National Mutual was the parent company for National Mutual Bank New Zealand Limited (National Mutual Bank). ,;.,,hich was original ly registered in June 1 989. The special i st rural l ender, Primary Industry Bank of Australia Limited ( PJBA) obtained registration in May 1 989. The Rural Bank, which had by then been sold by the government to Fletcher Challenge, obtained registration in August 1 990. The United Bui lding Society also converted to bank status (becoming known as United Bank), obtaining registration in June 1 990, fol lowing its acquisition by the State Bank of South Australia ( Sykes, 1 996, p 505) . 1 9 T h e C B A w a s o n e o f t h e four major ful l-service banks in A ustral ia (a longs ide ANZ. A B and Westpac), although i t was at that time st i l l wholly owned by the Austral ian government. I ts entry to the New Zealand market meant that al l four of the A u stral ian major banks then had substantial operations in New Zealand. 1 3 With BNZ Finance (a finance company majority owned by the BNZ) obtaining registration in January 1 99 1 , most of the finance company sector was then part of registered banking groups . With the broader powers to engage in a wider range of activities now available to the banks, there was less of a role for separate stand-alone finance companies, although these institutions started to again find a di stinct special ist role towards the end of the 1 990s. 20 Not only was registration easier than had been expected : getting out of the New Zealand market was also relatively easy. The first to leave was Security Paci fic, whose business in ew Zealand had been relatively small in any case, but whose departure was also influenced by the parent bank' s problems in attempting to bui ld a global wholesale banking business. It sold its business to the State Bank of South Australia, which had gai ned registration in December 1 988 . C IBC did not remain particularly long in New Zealand either, with registration being rel inqui shed in July 1 989 . The number of registered banks peaked in 1 990 and 1 99 1 , and from that time on there were further withdrawals . Elderbank withdrew from the ew Zealand market in August 1 990 as i ts parent conglomerate came under pressure at home. National Mutual Bank rel inqui shed its l icence in December 1 990, again in response to financial pressures on its parent. Such finance company business as National Mutual had in New Zealand was sold to the ANTs finance company, UDC.2 1 Macquarie Bank found that it gained no pa11icular advantage from being a registered bank. and rel inquished its registration in January 1 99 1 . although it has continued to operate in New Zealand since that t ime. NZI Bank had been badly affected by the 1 987 stock market crash and its aftermath: it ran its business down and relinquished regi stration in February 1 992. The State Bank of South Australia also ran into trouble in its home market, and it finally rel inquished its New Zealand banking l icence in July 1 994. Mergers and acquisitions also had their impact in reducing the numbers of registered banks. Thus PostBank was absorbed by the ANZ, Westland Bank by the ASB, the 20 See Thorp (2003 ) for a more exten sive discu ssion. 2 1 l t had been proposed that National Mutual 's financ ial weakness woul d have been remed ied b y a m erger with the ANZ, but penn ission for this was denied by the A u stral ian Federal Treasurer, thus giv ing rise to the ?'six pi l lars'? pol icy. 1 4 Rural Bank by the National Bank, NAB(NZ) by the BNZ (reflecting the acquisition of the BNZ by the NAB), and United Bank by Countrywide. This l ast acquisition reflected the problems ofUnited Bank' s parent, the State Bank of South Austral ia . The absorption of the BNZ into the NAB group arguably reflected the i l l preparedness of such New Zealand-owned financial institutions for the deregulated environment they now faced. The BNZ incurred substantial lending losses in the aftermath of the 1 987 sharemarket crash, first in New Zealand, and then in Austral ia . This forced the government to effectively bai l the bank out twice, despite its having been only partly government owned since a partial float in 1 987 . The government sold its shareholding in the bank to the NAB in 1 992, and the NAB then succeeded in buying up the remaining minority shareholdings. Another New Zealand owned ent ity which had wanted to become a bank, DFC, had fai led in October 1 989, a victim of problems in its commercial and corporate lending, many of which were associated with the 1 987 stock market crash. No new banks were registered between January 1 99 1 and the end of 1 995 : the New Zealand banking sector was consolidating and settling down after the first rush of enthusiasm when the market was first opened up to new appl icants. As of the end of 1 995 , there were 1 5 banks registered, compared with 22 at the end of 1 990. Even though seven of those 1 5 banks had ceased to be registered by the end of 2003, the New Zealand banking system at the begin11ing of 1 996 looked a lot more stable. with most of the excesses of the 1 980s removed from bank balance sheets (although it was not unti l September 1 996 that the BNZ' s levels of impaired assets fell to levels in l ine with those of other banks). At the beginning of Apri l 1 987 , two of the four banks were Australian ovmed, one was Brit ish owned, and one was New Zealand owned. Some of the new banks to enter the market were or had been New Zealand owned entities, but by the end of 1 995, only two banks remained in New Zealand ow?nership : TSB and Trust Bank New Zealand. New Zealand owned NZI Bank had entered the market and fai led. ASB, Country,vide Bank and United Bank had acquired foreign shareholders prior to conversion to bank status. The financing of Countrywide ? s acquisition of United Bank 1 5 obliged it to become whol ly foreign-owned. PostBank had been owned by the ANZ before it even gained registration as a bank. The Rural Bank was owned by New Zealand conglomerate Fletcher Chal lenge when it was registered, but it was later acquired by the United Kingdom owned NBNZ. A number of the Australian entrants to the New Zealand market had also been and gone, with PIBA, which had been Austral ian owned, sold to the (Netherlands-owned) Rabobank group in late 1 994. Elderbank, National Mutual Bank and the State Bank of South Australia had all otherwise been forced to contract as a result of pressures on their parent companies. At the begirming of 1 996, therefore, only 2 out of 1 5 banks were New Zealand owned, and these 2 banks accounted for 1 0 . 1 2% of the assets of the New Zealand banking system as at 3 1 March 1 996. The beginning of 1 996 also saw the introduction of a new system of banking supervision, based on quarterly public disclosure by the banks. Thi s has generated significant quantities of information on the New Zealand banking system, and these data provide a basis for the research reported in this dissertation. 2.3 1 996 and s ince Another way of looking at the New Zealand banking market at the beginning of 1 996 is to look at the activities the banks were undertaking, and at their relative significance in the market. Of the 1 5 banks that were registered, eight undertook significant retail banking business, with a collective market share of 92 . 1 2% of total assets. This group was sti l l dominated by the four original trading banks, which had a combined market share of 67 .89% of total assets, and which also undertook significant amounts of corporate and commercial banking business. A further two banks were special ists (PIBA and BNZ Finance). Of the other five banks, two (C itibank and HSBC) had a small amount of retail business, although they were primarily concentrated on the corporate market . Barclays and Indosuez were also focused on the corporate market, although Barclays also unde11ook some sharebroking business. BT had two main strands to its business - funds management and trading in the foreign exchange and money markets . 1 6 In 1 996 the New Zealand banking system was in a process of transition in other ways. One of these was with the payments system. New Zealanders had traditionally been high users of cheques, but among the new activities embarked on during the mid 1 980s were some experiments with EFTPOS.22 Once a single interoperable system was established in the late 1 980s, EFTPOS volumes began to grow, and by the end of 1 993 all banks had come to be participants in the system. Over the next few years, EFTPOS volumes grew rapidly, with at least part of this growth being a replacement for cheques, which are relatively expensive to process. Then, after 1 998, credit card usage also started to grow, at least in part in response to the banks ' development and promotion of card loyalty programmes.23 Transformation had also occurred in the banks' computer processing. During the early 1 980s, the four trading banks that owned Databank had decided to confront the changing economic environment and the opportuni ties offered by technological advances in computing through a banking redevelopment project. Following its sign? off in 1 983 , this came to be known as IBIS (Integrated Banking Information System). The project came to be subj ect to a number of problems, not least of which were its scale and complexity, but the factor that finally led to its abandonment in 1 989 was that such a co-operative computing project no longer made sense in a market where there was a range of new competitors with significantly lower cost computer systems. Moreover, the new competitors were not stuck in the straitjacket of a joint computer processing environment. and could thus respond much more readily to changing market conditions. In the new operating environment. the four banks that owned Databank were no longer interested in co-operating on computing. but wanted to be in a position to compete against each other, for which they needed to be able to control their own computing future.24 The transition for each of the four banks to running their own computer systems took t ime, but by 1 996 the process had been largely completed, and part of the Databank system had become the foundation for an all -bank clearing system. This was under the control of a new company, Interchange and Settlement Limited ( ISL) , which was 22 A lthough the introduction of E FT POS coincided with deregu lation, it was not dependent on or a consequence of it. 23 Credit cards had first been issued by New Zealand banks i n 1 97 9 . 24 This history i s discussed in greater depth in Manhews & Tripe ( 2004 ). 1 7 owned collectively by al l the settlement banks (defined in terms of transaction volumes and use of their own accounts at the Reserve Bank for interbank settlement) . This company contracted the actual processing to EDS, who had bought the Databank business from the banks that had owned it. A number of the banks had undergone transformation in other ways. By the begi1ming of 1 996, the BNZ had completed its absorption of NAB (NZ), Countrywide had completed the absorption of United Bank, the National Bank had completed the absorption of the Rural Bank, and the A Z had largely completed the absorption of PostBank. Since the ASB had separated itself from the rest of the Trust Bank group it had developed a national network outside its home base of Auckland and Northland (although its branches were only in the larger towns and cities, and it did not try to achieve the breadth of network enjoyed by the four former trading banks). Trust Bank had managed to largely transform itself into a single national bank, with a portion of the bank having been floated on the New Zealand stock exchange in 1 994. When Trust Bank was floated, 87% of the shareholding had been retained by the nine community trusts which had previously been the bank ' s sole owners. These community trusts had agreed that they would retain their shareholdings for at least two years, but as that period reached i ts end in early 1 996, speculation began to mount that Trust Bank could be acquired by one of its competitors. Westpac was the successful bidder. with NBNZ being unsuccessful. and later in 1 996 Westpac adopted the name WestpacTrust to reflect the merger of the two entities.2 5 Following the completion of this transaction, more than 99% of the New Zealand banking sector was foreign-owned, and the number of banks with significant retail business was reduced to seven. It is not obviously a consequence of the new disclosure regime, but after the beginning of 1 996 a number of new banks obtained registrati on. Rabobank obtained registration for a branch of the parent bank in Apri l 1 996. Bank of Tokyo-Mitsubishi (Australia) Ltd obtained registration as a branch in September 1 996, followed i n 25 This was an anempt t o preserve the association with the Trust Bank name, wh ich was seen a s having a strong association with the New Zealand community. A decision to drop ?'Trust'? from the name and revert to operat ing as Westpac was announced in late 2002 . 1 8 November by a branch of Deutsche Banl<. Banque Nationale de Paris was registered as a branch in March 1 997, with Kookmin Bank (from Korea) gaining registration as a branch in July 1 997. ABN Amro obtained registration for a branch in March 1 998, which it used to take over the business of Barclays, which was under pressure internationally, and whi ch rel inquished its New Zealand registration later in March 1 998 . In August 1 998, lndosuez, which was by now trading as Credit Agricole l ndosuez, also rel inquished its New Zealand registration. on the basis that it could cont inue to service its New Zealand cl ient base from Austral ia. From the retail banking perspective, the big event of 1 998 was the acquisition of Countrywide by the NBNZ, which gave the NBNZ the bigger role in the retail market that it had hoped to achieve with the purchase of Trust Bank. Following its acquisition of the Rural Bank in 1 992, the NBNZ had had a relatively large proportion of its portfol io in rural sector lending (22 .8% as at 30 June 1 998), and a relatively small proportion of its portfol io in housing ( 32 .9% as at 30 June 1 998) . This balance was now restored somewhat, with housing increased to 47.2% and rural exposures decreased to 1 6.2% as at 3 1 December 1 998 . The Countrywide banking l icence was rel inquished in November 1 998 . 1 998 also saw Citibank sel l ing i t s retail loan portfolio to AMP Banking, \Vho obtained regi stration for the purpose in October 1 998. AMP' s other lending business. which had mainly been developed through a business called Ergo, was switched across to the bank towards the end of 1 999. In June 1 999, BT rel inqui shed its regi stration, fol lowing the acquisition of BT's business worldv.?ide by Deutsche Bank. There was another rearrangement of banking l icences in July 1 999, ,;vith the regi stration of Rabo Wrightson Finance Limited, which later changed its name to Rabobank New Zealand Limited. This fol lowed the rel inquishing of PIBA 's l icence at the end of June 1 999, although it had in effect ceased doing business some time previously. Rabobank New Zealand Limited was to be the entity that undertook most of the Rabobank group' s rural lending, which it joined with business it had acquired with the purchase of Wrightson Farmers Finance. Data for Rabobank New Zealand Limited are consolidated into the financial reports for the Rabobank branch : the addit ional business undertaken by the branch is essentially corporate banking related. 1 9 The CBA (no longer government-owned fol lowing its privatisation during the 1 990s) had been a long-term investor in ASB, but as at 1 October 2000 it moved to buy out the minority shareholding held by the ASB Community Trust, and convert the bank to a whol ly owned subsidiary. In June 2000 it had registered a branch to cover its other business in New Zealand (which had not been consol idated into ASB), and since that t ime the CBA group has had two banks registered. The ASB figures are consolidated into the figures for the branch, however, although it is generally noted that the CBA does not have a lot of business outside of ASB (Note that the CBA's branch figures also include the group' s insurance interests in New Zealand) . I n March 200 1 , Banque Nationale de Paris, which was by then known as BNP Paribas, relinquished its registration, on the basis that it could service its New Zealand c lient base just as wel l from Austral ia . B Z Finance, the minority shareholders in which had been bought out in the mid 1 990s, rel inqui shed i ts reg istration in June 200 1 , with the business being absorbed into a division of the parent bank. Two new banks have been regi stered more recently, with a primary focus on retai l banking. Kiwibank Limited (Kiwibank. originally registered as New Zealand Post F inancial Services Limited, and owned by the New Zealand government through New Zealand Post ) was registered in November 200 1 . St George Bank New Zealand Limited was regi stered in February 2003 , and operates as a joint venture with supermarket chain Foodstuffs . St George Bank is a major retai l bank in Austral ia, the fifth largest bank in Australia overalL with a strong concentration of its business in the states ofNew South Wales and South Australia. In late 2002, AMP Banking announced that it would be sell ing its New Zealand banking business (as wel l as its banking business in the UK). This decision was a reflection both of the problems being experienced by the parent company and the Jack of profitabi l ity being achieved by its operations in New Zealand. I ts residential and retai l portfolio was sold to HSBC, its c redit card business to American Express, most of its commercial loan portfolio to GE (who had earl ier purchased the Austral ian and New Zealand business of AGC from Westpac), with the balance to Strategic Finance Ltd, and its rural loan portfol io to Rabobank. 2 0 There have also been changes in the monetary policy implementation and interbank settlement arrangements over the period 1 996 to 2003, which have impacted on the performance of New Zealand financial institutions. Key events in this respect have been the introduction of a real-time gross settlement (RTGS) system in 1 998, and a change in the monetary policy regime by the adoption of an Official Cash Rate (OCR) system in 1 999. The introduction of the OCR has been accompanied by a reduction in volati lity in financial markets (Brookes & Hampton, 2000) . As at 3 1 March 2003 , there were 1 8 banks registered in New Zealand . Only two of these were New Zealand owned, with a combined market share of total bank assets of 1 .2 1 %. A further 66 .33% was Austral ian owned, 22 .37% British owned26, and 6 .53% German owned. Remaining portions of the banking system assets were owned by banks from the Netherlands, the USA, Korea and Japan. There were five major banks which undertook a full range of activities with both personal and business customers - ANZ, ASB. BNZ, National Bank and Westpac . TSB also had a branch network, although this was l imited to Taranaki , and relationships with business customers tended to be l imited in scope. There were two new banks focussed on personal retai l customers, which were looking to operate branch networks using other rganizations' faci l ities - Kiwibank and St George Bank. Other special i st retai l banks included Rabobank (NZ) Limited. focussed on farming business, and Kookmin Bank, whose business was largely with the Korean community. AMP Banking could also be classed as retail , although it was in the process of exiting the New Zealand market. The other seven banks - ABN Amro, Bank of Tokyo/Mitsubi shi , CBA, C it ibank, Deutsche Bank, Hong Kong Bank, and Rabobank Nederland - were essential ly all specialist wholesale corporate banks, although Hong Kong Bank had a moderate amount of retail business, which expanded somewhat as a result of acquiring the retai l banking portion of the AMP Banking business. Towards the middle of 2003 , Lloyds TSB announced that it wished to review its ownership of the NB lZ. A number of the major Austral ian banks expressed an 26 Th i s i s o n the bas i s that l-IS BC can b e c lassed as a British-owned bank, a lthough t h e N e w Zealand branch is actual ly a branch of the ban k ' s Hong Kong business. 2 1 interest in buying it, but in the end the only purchaser to carry through and obtain a c learance from the Commerce Commission was the ANZ Banking Group Ltd, with the acquisition mmounced on 24 October, and settlement occurring on 1 December. The actual combining of the operations has been subject to number of restrictions by the Reserve Bank of New Zealand, and the formal combination of the two legal entities (and lapsing of the NBNZ's registration) d id not occur until late June 2004. At that stage it was envisaged that the new bank, ANZ- ational, would retain two separate brands, at least for a significant period into the future . 2.4 The bank ing sector and the New Zealand economy The changes and developments i n the structure and operation of the banking sector were not occurring independently of changes in the environment within which the banks were operating, and were accompanied by changes in the volumes of banking business. One of the i ssues that confronted the newly-elected Labour government in 1 984 was a high base level of inflation in the New Zealand economy, even though this had to some extent been suppressed by the wage and price freeze imposed by the previous government. Part of the process of l ibera l i sation was therefore directed at making monetary pol icy a more effective tool for the control of inflation. Market interest rates were therefore allovied to increase so that real interest rates became positive. and these interest rates became the key measure of monetary conditions. The combating of inflation was fom1al i sed in the Reserve Bank of New Zealand Act 1 989. which provided for a target inflation level to be specified in a Policy Targets Agreement, with the initial inflation target set at 0 to 2%. Because inflation had persisted since the early 1 970s. there was a very substantial level of inflation expectations in the New Zealand psyche, and relatively high levels of interest rates were therefore necessary to achieve reductions in the rate of inflation. The key market interest rate, the 90-day bank bi l l rate, thus peaked at 35 . 5% on 8 March 1 985 (exacerbated by the effects of the floating of the New Zealm1d dollar), 22 a lthough there have been many reductions since that time. The 90-day bil l rate was last above 1 5% on 1 8 October 1 990, and last above 1 0% on 1 0 October 1 996.27 For the period covered by this research, from 1 996 to 2003, interest rates were still incl ined to be relat ively high, particularly when compared with other developed countries, and they were also inclined to be relat ively volati le . The comparison with Australia, with data derived from the respective central bank web-sites (www. rbnz govt .nz and www.rba.gov au), is shown in Figure 1 . Figure 1 - 90 day bank b i l l rates between New Zealand and Austra l ia 12 10 E 8 :::J c c "' Oi 6 c. E -NEW ZEALAND -AUSTRALIA "' ? "' Q. 4 2 0 Another striking feature of New Zealand bank ing since the mid 1 990s has been the very substant ial growth in bank assets, with a particular growth in housing lending. The growth in housing lending was partly in react ion to the restrict ions applying during the regulated period prior to 1 985, when housing fmance had been d i fficult to obtain through formal c hannels, as discussed in section 2 . 1 . Banks relished their new? found freedom to lend in this area, and were encouraged to do so by the preferent ial risk-weighting for lending secured by residential mortgage under the ( 1 988) Base! capita l accord (although mortgage-secured lending as a proportion of total assets did not change significantly over the period 1 996 to 2003) . 2 7 These data are obtained from the Reserve Bank o f New Zealand ' s web-site, www. rbnz.govt.nz, Table 82. 23 There has nonetheless been a very substantial growth in banks' total assets over the period of this research, as is evident in Figure 2, which shows the end-of-quarter figures for the five major banks (with retail branch networks), derived form their conso lidated statements of risk-weighted assets as part of the banks' quarterly d isclosure statements. Other things being equal, one would expect this sort of bank asset growth to have had some impact on the way in which banks have gone about their business, and this is one of the issues explored as part of this research. F igure 2 - Total asset trends - major banks This asset growth has led to other changes in the way New Zealand banks operate, with a key change being evident in bank funding. I n 1 988, the assets of the New Zealand banking system were largely funded by New Zealand deposits, but since that time there have been changes in the patterns of bank funding. Particularly in the late 1 990s, there has been very substantial growth in foreign funding of the banking system. 2.5 What does this mean for research? Deregulation of New Zealand fmancial markets was undertaken with a number o f objectives, and in anticipation of a number o f outcomes being achieved. Many of these outcomes were reported by Harper ( 1 986), and i t i s appropriate t o look at what 24 he suggested as rationales for and potential consequences of deregulation, and at the extent to which some of the outcomes he suggested have been real i sed. The first major argument offered by Harper was that the inputs used by financial firms are not general ly highly specific to those financial firms. Major inputs are identified as capital ( in the form of fixed assets), labour, materials, intangibles (such as information and brand-name capital) and financial inputs. Because of the flexibi l ity in the way in which most of these inputs can be used (and, in cost terms, financial inputs are the most significant), it is relatively easy for financial firms to change their use of inputs and the outputs they generate from them in response to market conditions, such as would have arisen from the financi al reform process. Harper highl ighted the differences between financial firms and other types of firms with much more substantial investments in specialised plant in this respect (p 60). From this argument, Harper therefore argued that there ought to be economies of scope in the production of financial services, with economies of scope defined as the abi l ity to produce multiple outputs in a s ingle firm at lower cost than producing the same outputs in individual specialist firms. An example of thi s would be in a single financial firm being able to process transactions to both current and savings accounts more cheaply than institutions which are only able to operate one type of account. Harper noted that economies of scope were perceived as being important as the previous functional regulation ofNew Zealand financial institutions had l imited firms in the range of activities they could undertake. Information and kno\v-how \Vere ident ified as key inputs into financial institutions' production processes, giving rise to economies of scope. Economies of scope are contrasted with economies of scale, which ari se when the cost of producing an output increases less than proportionately with an increase in the quantity of output. Harper assumed that there ought to be economies of scale in the production of financial services, although the amount of research that had been undertaken international ly to that stage was relatively l imited?8 He promoted the desirabil ity of research into economies of scale and scope for New Zealand financial 28 Subsequent i nternat ional research on economies of scale and scope in financi a l i nst itutions i s d iscussed in Chapter 3 . 2 5 markets (p 67) , but then noted that the small size of the New Zealand market constrained firms from enjoying economies of scale, which meant that they ought therefore to be pursuing economies of scope (p 67) . A key factor in why information is so important to economies of scope was identified in being the difficulty in transferring it from one firm to another. This was seen as l eading to the formation of conglomerates, and it i s interesting to note that one of the examples Harper used was the National Bank group (p 88) . At the time he was writing, this consisted of four separate entities - the ational Bank, Southpac ( a merchant bank), General F inance ( a finance company), and Equus (an investment company). Within five years of Harper' s description of this situation, the only surviving standalone entity was the National Bank, into which significant portions of the other businesses had been absorbed .29 A final point Harper suggested was that conglomerates might not always result in efficiency. L imits might be imposed on diversification from diseconomies of scale and scope. These might ari se from excessive demands on management, and from the increased complexity of the firms. Since Harper wrote his report in 1 986, there has been very l ittle exploration of the i ssues he rai sed in respect of the efficiency of financial institutions, that might have been expected to have arisen as a consequence of deregulation. That alone would provide justification for the proposed research. but there are additional reasons as to why one ought to be interested in financial institution efficiency. One of the key steps in the process of financial sector deregulation was the enactment of the Reserve Bank of New Zealand Act 1 989. There are two separate places \V here the Reserve Bank of New Zealand Act refers to a "sound and efficient" banking system - Section 1 0, in the context of monetary policy, and Section 68, in terms of the supervisory oversight of banks. Dawe ( 1 990) suggests that Section 1 0 of the Act means that monetary policy should not be managed in such away as to cause 29 The major exception to this was the remaining p011ion of the General F inance business, which was sold to Nat ional M utual, and which thus became pan of N at ional M utual Bank. The h istory of the National Bank of New Zealand over this period is reponed in Holmes (2003). 26 instabil ity or inefficiency in financial markets (p 33) , while Mayes ( 1 998), in highl ighting the expression "sound and efficient" (p 1 3 ), seems to be more interested in sound rather than efficient. The Reserve Bank has not defined what it means by efficient: one of the i ssues to be explored in thi s thesis is what efficiency might mean, which wil l provide a foundation for exploring the extent to which the ew Zealand banking system might be regarded as efficient. As part of what was in effect a study of the consequences of deregulation, Diewert & Lawrence ( 1 999) looked at productivity in the New Zealand financial sector. This was part of a broader study looking at productivity in the New Zealand economy as a whol e for the period 1 972- 1 998, and which was generally concemed at the allegedly slow rate of productivity growth in the ew Zealand economy. Diewe11 and Lawrence broke the data for the New Zealand economy down into 20 separate sectors. Almost all sectors of the economy showed improvement, but the outstanding disappointment was the financial services sector, which showed a 2 . 1 1 % per annum decline in productiv ity over the period 1 978- 1 998 (p 76) . Diewert and Lawrence noted that the reforms in the financi al sector in the 1 980s. together with the rapid change ( improvement) in the range and quality of services offered make this result implausible (p 74). lt was suggested that the problems might l ie in the \vay both investment and output were measured, which made it difficult to identify changes in qual ity (although this problem is not pecul iar to the New Zealand market) . They also suggested that the estimated capital stock for the sector had increased at an implausibly fast rate. Their finding was also inconsistent with the measure of bank effic iency that is most popular amongst bank managements, the cost to income ratio . The trend in bank cost to income ratios for Austral ia and New Zealand, based on the author' s calculations from the OECD bank profitabi l ity reporting (OECD, 2002) is shown in Figure 3 . Other things being equal, this graph would suggest that efficiency had improved i n both countries, and one of the issues explored in this research i s thus t o look at the val idity of the cost to income ratio as a measure of efficiency. 27 Another problem identified by Diewert and Lawrence in respect of the productivity of the financial sector was in the defmition of financial sector output. This is an area which has been subject to significant theoretical review and empirical investigation, and it will be discussed further later in this research. Figure 3: Cost to income ratios in New Zealand and Austra l ia - a l l banks 1 990 1 99 1 1 992 1993 1 994 1 995 1 99 6 1 99 7 1 998 1 999 2000 2 0 0 1 There is a further po int made by Diewert and Lawrence in their conc lusion, which provides support and justificat ion for this research. As part of the process of enhancing understanding ofNew Zealand ' s productivity performance, they suggest : "Benchrnarking and DEA (data envelopment analysis) projects that would examine particular New Zealand industries or firms and compare their performance with international best practice. This would prov ide information on productivity levels as wel l as growth rates, while ensuring like is being compared with like" (p 1 60) . A range of issues has thus been identified for further investigation, in relation to the efficiency of the New Zealand fmancial system and the consequences of deregulat ion. A major thrust of this study is to look at the efficiency of New Zealand fmancial institutions (banks in particu lar), as a step to trying to understand something about the productivity of the New Zealand fmancial sector. There is a desire to find out how the 2 8 productivity of the New Zealand fmancial sector has developed through time, and a lso to see how the efficiency or productivity ofNew Zealand banks compares with those in Australia ( and particularly those banks in Australia that own banks in New Zealand) . There is also a need to have regard to competit ive condit ions, because of the impact they can have on efficiency, although initial research (Smith & Tripe, 200 1 ) suggests that New Zealand fmancial markets have been relatively contestable. The comparison with Australia is also important because a comparison of returns on equity, as seen in F igure 4,30 shows much higher returns being earned in New Zealand than are earned by either the major or regional banks in Australia. This issue is explored further in section 5 .4 . Figure 4: Relative Returns on Equity - Australia and New Zealand 30% ---New Zealand 1 5% +----------"'......._11"""'--------?-?"""-----='!:lil(-- --Austra lian maJors ""*-Australian regiona ls 5%+-------------------------- 0% --?--?--?--?--?--?---?-? 1 996 1 997 1 998 1 999 2000 2001 2002 2003 This chapter has looked at the background of the New Zealand banking system relat ive to an evaluation ofNew Zealand banks' relative efficiency. The next chapter of this dissertation looks at prior research on measuring efficiency, and wil l provide some principles that can be applied in measuring the efficiency of New Zealand banks. 30 These are based on the author's calculations from the respective banks' annual reports. The New Zealand figures cover ANZ, ASB, BNZ, National Bank and TSB; the Australian majors are ANZ, CBA, NAB and Westpac; the set of Australian regional banks comprises Adelaide Bank, Bank of Queensland, Bank West, Bendigo Bank, St George Bank and Suncorp-Metway. Further discussion of the Austral ian banking system is provided in section 4. 1 below. 29 3 . Previous research and theoretical issues This chapter deals with a number of questions. It begins by looking at what might be meant by efficiency, and then goes on to look at efficiency in the specific context of financial i nstitutions. This provides a basis for an examination of specific methods that have been used for studying efficiency, and efficiency in financial institutions in particular. The latter part of the chapter then looks at some of the specific practical i ssues which otherwise get in the way of research, and at how these have been dealt with 111 previous research. This section has more of a focus on one of those techniques m particular, Data Envelopment Analysis, as this i s the teclmique used in this research, but it also identifies i ssues that need to be addressed in other approaches. Following thi s outline of previous research, the fol lowing chapter explains the actual methods to be used in this research, results from which are reported in Chapter 5 . 3.1 What is effic iency? The concept of efficiency may be regarded as one of the fundamental precepts of economics, and one which also has welfare connotations. Effic iency may be defined as the ratio of the \veighted sum of outputs to the weighted sum of inputs (Boussofiane et a!, 1 99 1 ) . In general terms, a fin11 may be said to be operating efficiently i f it caru1ot produce more output without a corresponding relative increase in inputs, or if it caru1ot reduce its inputs without a corresponding relative decrease in output .3 1 More generally, a decision-making unit (DMU) wil l be 1 00% efficient if there i s no scope for improvement in the ratio in which it conve11s inputs to outputs. 32 3 1 This assumes that there i s no change or d i fference in the qual ity of inputs or outputs. 32 This i s consi stent with what Cooper et al (2000) refer to as Pareto-Koopmans effici ency: a unit " is fu l ly efficient i f and only if i t i s not possible to i mprove any input or output without worsening some other input or output" (p 4 5 ) . For the background to the term i nology on e ffi c iency, refer to their d iscuss ion on pp 68-69. See also Charnes et al ( 1 98 5 ) . 3 0 Efficiency can be discussed i n a variety of different forms, not all of them necessarily total ly consistent with the previous definition. Traditional microeconomic theory has l ong talked of economies of scale, where increased volumes of output are supposed to be able to be produced with less than proportionate i ncreases in quantities of inputs ( increasing returns to scale) . In due course, however, economies of scale wi l l be exhausted, and increased output wil l require a more than proportionate increase in inputs, a situation described as diseconomies of scale ( decreasing returns to scale) . 3 3 This description of economies of scale i s consistent with what is referred to as a U? shaped average cost curve, one of the implications of which i s that there is a part icular l evel of output consistent with a minimum level of average cost. Under such a view, there is l ikely to be a flat portion in the middle of the U, characterised by constant returns to scale, where there is a fixed (and minimum cost) relationship between output and uti l isation of inputs. I t i s more common in pract ice to focus on the left side of the U-shaped average cost c urve, where increases in outputs are associated with l ess than proportionate increases in inputs. A possible source of such positive scale economies in banking might arise from using a computer system to process customer accounts: more accounts can be processed without a corresponding increase in computing costs (Mester, 1 987) . Another type of efficiency i s economies of scope . The essence of these i s that firms should be able to produce multiple outputs from the same group of inputs at l o\ver c ost. in tern1s of inputs. than if they specialised in producing only one type of output. Clark ( 1 988) identified economies of scope as existing where the total costs from j oint production of al l products in the mix were less than the sum of the costs of producing each product independently (p 1 8) . In the context of a financial institution, one might be looking at a situation where a firm produced both l oans and deposit services, using the same staff and branch networks, rather than special i sing in just one of these functions by itself. 33 This discussion i s d irected at sca l e economies in a stat ic context. I ssues relating to changes in e ffic iency arising from changing production functions through t i m e are discussed funher below. 3 1 Mester ( 1 987) notes that, at least in financial services, economies of scale and scope may arise at the same time. The use of a computer to reali se economies of scale in account processing may be accompanied by use of the same computer system to process several different types of account simultaneously. Clark ( 1 988) identified a relationship between economies of scale and scope and the structure of firms in an industry. If the avai lable teclmology allows for both economies of scale and scope, the industry wil l tend to be made up of large diversified firms, producing at lower unit costs, and using this advantage to gain market share.34 I f the technology does not allow for economies of scale or scope, smal l specialised firms wil l dominate the industry. If there is an absence of significant economies of scale or scope, there is l ikely to be a mixture of larger diversified firms and smal ler specialised firms (p 1 7). These discussions of economies of scale and scope in the previous paragraphs may be construed as assuming that firms are operating on some so11 of production possibil ity frontier, and that it is only a matter of achieving an effic ient level of production or mix of outputs. This will often not be the case, thus providing a basis for the concept of X-efficiency, as proposed by Leibenstein ( 1 966). If a firm is X-inefficient, it is l ikely to be capable of producing more output for any given l evel of inputs, perhaps by a better uti l isation of resources, reorganisation of the production process so as to make better use of avai lable technology. better purchasing of inputs, enhancing staff motivation. or by any one of a range of other improvements. 3 5 X-inefficiency is commonly broken down into 2 elements, al locative inefficiency and technical i nefficiency, in terms of the approach outl ined by Farrel l ( 1 957) . 36 Technical efficiency might be conceived in s imple terms as a measure of whether the firm is max imising product ion from the inputs it is using, while allocative efficiency looks at whether the best combination of inputs is being used, having regard to their relative cost. Frei et al (2000) dist inguish X-efficiency, suggesting that : 34 Th i s is the so-called Efficient Structure hypothesis, wh ich may be contrasted with the Structure Conduct Performance hypothesis . See, for example, B erger ( 1 995) for a review of these. 35 Stigler ( 1 976) suggested that d i fferences in X-efficiency should be attributed to d i fferences in technology. 36 A lthough Fan?el J referred to al locati ve effic iency as price effi c iency. 32 " . . . X-efficiency describes all technical and allocative inefficiencies of individual firms that are not scale/scope dependent. Thus X-efficiency is a measure of how wel l management is aligning technology, human resources and other assets to produce a given level of outputs ." (p 260) . Attempts to measure X-efficiency generally occur relative to an efficiency frontier, with firms' teclmical efficiency being defined in tem1s of their relative distance from the frontier (which then becomes the benchmark for optimum performance) . Al locative efficiency wil l then be identified according to whether firms are producing at that point on the efficient frontier that minimises input costs. This can be explained using Figure 5 . Figure 5 : X-effic iency a nd its decomposition into techn ical a nd a l locative efficiency. X2 A S' ' Xl Figure 5 i s commonly referred to a s the Farrell diagram, after Farrell ( I 95 7, Diagram I , p 254) . X I and X2 represent inputs, with the line SS' representing a fixed quantity of production with minimum uti l isation of inputs (an iso-quant l ine, which makes up an efficient frontier) . At point P, the producer is inefficient, as they could be using less of both inputs, and the producer may be said to be inefficient by the ratio of the 3 3 distance from the ongm (which we wil l refer to as 0) to point Q relative to the di stance from the origin to point P. At point P the producer may be said to be producing with efficiency OQ/OP, which wil l be less than one, reflecti ng an over? uti l isation of inputs. This is a measure of teclmical efficiency. The line AA' reflects the relative prices of the two inputs, and where this l ine is tangential to the iso-quant l ine (SS ' ) at Q', the cost of production wil l be minimised. Although point Q is technically effic ient, it i s apparent that the cost of production could be further reduced if the point of production could be moved to Q' (which wil l have the same cost of production a s point R, even though R is not teclmical l y feasible). The move from Q to Q' represents a fm1her source of inefficiency, referred to as allocative i nefficiency, which arises from use of a less than-cost-minimising combination of inputs. The amount of allocative inefficiency i s reflected in the distance QR, with the measure of allocative inefficiency being OR/OQ. Total inefficiency thus comprises OR/OP, but with this able to be decomposed into teclmical and allocative inefficiency. Siems & Barr ( 1 998) distinguish teclmical and allocative efficiency as fol lows, with X-efficiency referred to as economic efficiency: "Allocative efficiency i s about doing the right things, productive efficiency i s about doing things right, and economic efficiency is about doing the right things right"' (p 1 3 ) . Efficiency defined as above, relative to a frontier, i s generally only a measure of relative, rather than absolute efficiency (Methods for specifying the efficient frontier are reviewed in section 3 .3 below). Measurement of absolute efficiency would require knowledge of the optimum productive process ( in an engineering sense), which would then become the benchmark against which other units' efficiency could be compared : there i s no such agreed model for banking firms. When measurement i s only of relative efficiencies, it is not in general valid to compare efficiency estimates derived from different samples. 34 It should also be noted that efficiency, as a measure of productivity, is not the same as profitabi l ity, although the concepts are related to each other. Grifell-Tatje & Lovel l ( 1 999) noted that a change in a firm ' s profits may derive from a number of sources, including a change in input or output prices. The other sources of profit change are l ikely to be measured using the teclmiques of efficiency analysis, including technical change leading to an increase in output without any increase in resource uti l isation, an improvement or decline in operating efficiency (X-effic iency), or a change in output proportionately greater or less than input uti l isation, reflecting economies or diseconomies of scale. Further sources of profitabi lity improvement include changes in product or resource mix, associated with economies of scope and allocative efficiency. This review of efficiency so far has only been on a static basis, whereas it is arguable that one ought to look at changes in efficiency through time. There is an assumption that technological development should result in improved efficiency, and that one should therefore see a steady improvement in efficiency through time, but prior research suggests this is not always found, at least in financial services.37 If researchers are concerned with the welfare benefits of efficiency, the dynamic context is important, as thi s is what will provide for improvements in welfare through time. In broad terms, consistent with the concepts espoused by Schumpeter ( 1 943) , innovation and technological progress wi l l thrive in a competitive market. and this i s one reason why researchers should be concerned about competitive conditions.38 In terms of identifying the impact of technological progress and innovation, however, there are other issues to be appreciated. What technological progress might mean is that the efficient frontier, against which a firm ' s teclmical or X-efficiency i s measured, might be expected to move through time. There i s then a question as to 37 See, for example, the findings of Men des & Rebelo ( 1 999) in respect of the Portuguese market. I n some cases a t least, this may b e a consequence o f deregulation and i t s effect o n increasing interest costs ( H umphrey, 1 99 1 ; Lozano- Vi vas, 1 99 7 ; Lozano- Vi vas, 1 998), and thus, once the process of deregulat i on h as run its course, effi ciency may start to improve ( K umbhakar et al, 200 I ). l t is a lso possible that, in financial services. technolog i cal change may be reflected in improvements in service and product qual ity, such as through te lephone and interne! banking, which may be much harder to measure. 38 See, for example the discussion of Schumpeter' s dynamics in Molyneux & Shamroukh ( 1 999), pp 8 8-9 1 . 3 5 how an individual firm's efficiency, in terms of outputs relative to inputs, might change through time as the efficient frontier i s itself moving. These changes might also be impacted by changes in the scale of production (Balk, 200 1 ) . One technique for exploring this process further is via the Malmquist Index, which is d iscussed in greater depth as part of section 3 .3 . 3.2 Effic iency i n F inancia l I nstitut ions I t fol lows from the previous section that there are three types of effic iency that one might want to explore in looking at financial i nstitutions. One might be interested in economies of scale, economies of scope and X-efficiency, while one might also want to see how measures of efficiency have changed through time. Efficiency in financial institutions i s and should be a matter of public concern, as not only can more efficient financial institutions be expected to be more profitable, but one should also expect financial institution efficiency to lead to greater amounts of funds being intermediated, and better service at lo,ver prices for consumers. Other things being equal , more efficient financial institutions should exhibit greater safety and soundness (Berger et al . , 1 993b, pp 22 1 -222 ), whi le also showing better credit quality in the loan portfol io (Berger & De Young, 1 997). These i ssues also provide a basis for looking at the efficiency implications of both bank mergers and acquisitions and government pol icy initiatives. Farrell ( 1 95 7) made the point that. to properly measure efficiency one needs to look at the range of inputs that a firm uses, and make allowance for all of these. I t was of no use just to look at the productivity of labour, for example, without regard to the range of other inputs used (p 253 ) . Some attempts had been made to get around this by use of index numbers, but his intention \;l,1as to get around the problems of these . This point appl ies just as much to efforts to measure efficiency in financial institutions. Common approaches to measurement of efficiency in banks are incl ined to focus on ratios of non-interest costs to average total assets (cost to assets), or non? interest costs to gross income (the cost to income ratio). As has been discussed el sewhere (De Young, 1 997b: Tripe, 1 998) , these ratios have a number of deficiencies, most particularly in that they don 't take account of differences in the business that 3 6 banks undertake, which wi l l in turn be reflected in different combinations of inputs and outputs. De Young ( 1 998) suggests that blind pursuit of accounting-based benchmarks might reduce a bank' s cost efficiency by cutting back on those expenditures necessary to run the bank properly (pp 5-6) . There i s also the i ssue that ratios may be contradictory: if a bank performs very well in terms of one ratio but poorly in terms of another, how does one determine an overall performance ranking (Golany & Storbeck, 1 999)?39 To take account of these i ssues, one therefore needs to look at banks on a multiple input and multi -product basis . Banks use a mixture of inputs to produce a mixture of outputs, and their reported average cost figures wi l l depend j ust as much on the mix of inputs and outputs as on the rate at which those inputs are used to produce outputs (Mester, 1 987) . In looking at bank costs one needs to consider more than j ust operating costs, which account for only part of overall bank costs. Berger & Humphrey ( 1 992b) highl i ght the effect of tradeoffs between price and service, and note that a bank with a less extensive branch network may pay higher interest costs to attract deposits, although it wi l l have lower operating costs (p 559). As a general rule, l arger banks will have lower (non-interest) operating costs but higher interest costs, reflecting their dependence on bono wed funds. The extent of banks' branch networks is impm1ant for simi lar reasons (Humphrey, 1 990) : v;here a bank does not have branches, it can be argued as reducing its operating costs by transfening these to its customers . S imi lar issues can arise on the output side. Noulas et al ( 1 990) show that attempts to aggregate outputs into a s ingle index are inval id, and that one i s therefore required to take account of the multi -product nature of large banks in seeking to measure their efficiency. Resti (2000) notes that techniques aimed at summarising multiple products had been proven to be too good to be true, since they required separabi l ity conditions not usual ly suppm1ed by empirical data. 39 See also the cri t ic ism of use of rat ios for the analysis of the perfonnance of bank branches in Schaffn it et al 1 997), p 273 . 37 More generally, Berger et al . ( 1 993 b) note that financial ratios may be misleading because they do not control for product mix or input prices.40 Moreover, use of a simple ratio cannot distinguish between X-efficiencies and scale and scope efficiencies (p 233 ). Thanassoul is et al ( 1 996) suggest that use of a multivariate approach gives a more balanced approach to performance measurement than ratio? based performance indicators . Siems & Barr ( 1 998) found positive relationships between bank efficiency, profitability and CAMEL ratings. The first p1ece of empirical analysis undertaken, described in section 4 .4 . 1 and rep011ed in section 5 . 1 , looks to provide a comparison of multivariate approaches to efficiency measurement and the cost to income ratios for a group of New Zealand banks. Initial studies of bank efficiency were inclined to be focused on looking for economies of scale, although some attention was also given to economies of scope (Mester, 1 987; Clark, 1 988) . Despite what Humphrey ( 1 985 ) refers to as the conventional wisdom, earl ier studies were general ly unable to find evidence for economies of scale beyond a relatively small size for a financial institution (assets greater than $ 1 00 mill ion). Evidence for economies of scope was also weak (Berger et al , 1 987) . A number of methodological difficulties were identified with thi s earl ier research, which cast doubt on the rel iabil ity of some of the results obtained . Humphrey ( 1 985 ) noted the problem that could ari se from looking at unit banks and banks with branches together: scale economies should be expected to be observed quite differently between the different classes of banks. Clark ( 1 988) noted the difficulties in defining bank costs and outputs, and problems with data and statistical methodology. Thus, as Berger Hunter & Timme ( 1 993 ) noted, studies whi ch looked at larger banks found the minimum average cost point to be associated with rather larger banks, with assets between $2 and $ 1 0 bi l l ion. In more recent work, however, using data from the 1 990s 40 Thus Chu & L i m ( 1 998) argue that it is more appropriate to consider interest expense as an input in effic iency studies, rather than deposits, because not a l l deposits car y the same interest expense (p 1 5 8). 3 8 as opposed to the 1 980s, Berger & Mester ( 1 997) have suggested that the most efficient scale size might be rather larger. 4 1 McAll ister & McManus ( 1 993 ) explained what they perceived to be some of the methodological reasons for the earl ier results, with a particular focus on the use of the translog cost function.42 Using a different approach, they suggested that banks could operate at minimum constant average cost at asset levels between $500 mi l l ion and $ 1 0 bil l ion. Also using a different approach, S iems & Clark ( 1 997) found banks' scale efficiencies to be essentially invariant above a relatively small balance sheet size. Another problem identified by Berger et al . ( 1 993b) was the importance of using a method which only measured scale efficiency for firms that were on the efficient frontier: this was not necessari ly common, particularly prior to the work of Berger et al ( 1 987) . If one was not focusing on firms on the frontier for assessing economies of scale, one was in danger of confusing scale efficiencies and X-inefficiency. Berger & Humphrey ( 1 99 1 ) found that X-inefficiency was a much more significant component of overal l inefficiency than was scale inefficiency. Scale inefficiencies accounted for only 5% of costs, whereas X-inefficiencies were around 20%. Humphrey ( 1 990) highlighted the importance of the definition of costs to be used when trying to measure economies of scale. Larger banks are general ly less able to ful ly fund themselves through retail deposits, and wil l thus have higher i nterest costs but lower non-interest costs than smal ler institutions. lf a researcher tries to measure scale economies using only operating costs, minimum costs wi l l be found at a larger asset size than if al l costs are considered . Berger et al . ( 1 993b) also identified methodological problems with prior work looking at scope efficiencies. The first problem they identified was with the translog cost function, which caused problems for zero output levels. A second problem related to 4 1 This wou ld be consi stent with the suggestion of Noulas et al ( 1 990) to the effect that opt imum scale might change for year to year, and that this m ight have been a particular i ssue in the 1 98 0 s when deregulation was st i l l working its way through the system in the U nited States. 42 I ssues with the translog cost function w i l l be d i sc ussed in greater depth as part of section 3 . 3 , below. H umphrey ( I 985, 1 990) pointed out that the translog was a cons iderable i mprovement on approaches used in earl ier research, as it a l lowed for a U-shaped cost curve, which the Cobb- Douglas approx imat ion of a production function d id not . 3 9 difficulties in finding specialised firms: if one is to test for economies of scope, one needs firms producing single outputs to allow comparison with firms producing them jointly. The third problem also applied to the search for scale economies; to test for scope economies one needed to be comparing firms that were on an efficient frontier. The beginning of a trend to investigation of X-inefficienc ies was noted by Evanoff & l srailevich ( 1 99 1 ) , who also noted that a major cause of X-inefficiency might be regulation. Berger et al . ( 1 993b) also called for further research into X-inefficiency, reflecting the dearth of such research to that stage, as wel l as more comparisons of X? efficiencies across borders, reflecting the increased level of competition between countries and the (sometimes) relatively large X-inefficiencies within countries. Measurement of X-ineffic iency requires banks' performance to be assessed relative to an efficient frontier, and by 1 997, Berger & Humphrey were able to report on 1 30 financial institution studies, covering institutions in 2 1 countries. Since that time, the number of studies undertaken has grown very substantial ly. The overwhelming emphasis in the studies reviewed by Berger & Humphrey ( 1 997) was in looking at banks i n the United States. although there were also studies of savings and loans and credit unions. There were also studies which focused on the relative efficiency of branches of pa11icular financial institutions.43 Most of the non? United States studies \Vere for countries in Europe. although studies were also reported from Canada. lndia, Japan, Mexico, Saudi Arabia, Tunisia and Turkey. There were also six studies which looked at multiple countries simultaneously, five of which were focussed on banks. Berger & Humphrey did not report on any Australasian studies, whereas a number have now been publ ished. New Zealand publications include Liu & Tripe (2002) and Tripe (2003) . Study of financial institution efficiency in Austral ia has been more extensive, with studies of banks by Walker ( 1 998) , Avkiran ( 1 999a) , Avkiran (2000), Sathye (200 1 ) , and Sathye (2002). There has been rather more study of non-bank financial institutions in Australia, including, in respect of building societies, Esho & Sharpe ( 1 995 and 1 996) and Worthington ( 1 998b ) , and in respect of credit unions, 43 Athanassopoulos ( 1 99 8 ) notes that these can have a role in exploring the sources of ineffic iency in banks as a whole. 40 Brown & O 'Connor ( 1 995) , Esho (200 1 ), Garden & Ralston ( 1 999), and Worthington ( 1 998a, 1 999a, 1 999b, 2000 and 200 1 ) . The more recent international research generally finds lower levels of X-inefficiency than was reported in earl ier studies. This could be a reflection of the globalisation of financial markets and the accompanying competitive pressures making it more difficult for a financial institution to survive if it was significantly inefficient relative to its competitors (Mester, 1 994 ; Bauer et al, 1 998) .44 In some countries. this may be a reflection of deregulation.4 5 On the other hand, it may be a reflection of the finding by Berger et al . ( 1 993a) that larger banks are generally less X-inefficient than smaller banks.46 Increased bank size and the greater number of non-US studies, which do not inc lude the same number of smal ler banks, may thus explain the lower levels of X? inefficiency observed.47 Berger & Humphrey ( 1 997 ) noted that there was an important role for cross-country comparisons of bank efficiency because of harmonisation in European markets for financial services. and because of the more general consequences of globalisation (p 1 87 ) . The difficulty with cross-country comparisons i s that the regulatory and economic environments are l ikely to differ betv,'een countries, as wi l l expectations regarding product offerings and customer service. These mean that it may not be val id to assume a common frontier for measurement of efficiency. Berger et al (2000) elaborate on the i ssue in greater depth, and fu11her note the potential impact of differences in the intensity of competition v,;ithin countries. Along with the other factors rep011ed above, thi s means that a finding of greater X-efficiency for banks in one country cannot be construed as meaning that banks from that country would be equal ly efficient were they to operate in some other country (p 49).48 44 See Tortosa-Ausina (2002a. 2 002c) for further exploration of this issue, although some of the issues raised are outside the scope of t h is research. 45 See, for example, Chen (200 I ) for a discussion of Taiwan . 46 Thus M il ler & Noulas ( I 996) fi nd that large U S banks (defined as those w ith more than $ 1 b i l l ion of assets) are incl ined to show a h i gher level of X-effi c iency, but some signs of decreasing returns to scale. 4i See A l am ( 2 00 1 ) for a d iscussion of the relat i onship between the number of firms used to est imate effic ienc ies and the scores generated. 48 See also the findings of Claessens et a! (200 I ) on the different interest margins earned by fore ign and domestical ly-owned banks in d i fferent countries . 4 1 Operation in a different cow1try would in any case be l ikely to be under different conditions from those which might be experienced in a bank' s home country. Various attempts have been made to overcome some of these problems . Berg et al ( 1 993) in their study of Norway, Sweden and F inland used both separate frontiers for each country and then a common frontier, against which they conducted further tests for the robustness of the results they obtained . A number of similar approaches have been attempted, with a basic assumption that such efficiency differences as arise reflect differences in the technology that is used in different countries (Chaffai et al, 200 1 ). This is a particular i ssue for studies that have used a common frontier, such as Fecher & Pestieau ( 1 993), Alien & Rai ( 1 996), Altunbas & Molyneux ( 1 996), Altunbas et a! (200 1 ), Cavallo & Rossi (200 1 ), Caval lo & Rossi (2002) and Maudos et al (2002) . Pastor et a1 ( 1 997) estimated separate functions for each country, but then attributed the very substantial differences in efficiency observed to differences in the technology employed . Carbo et a! (2002) used a single frontier for their study of European savings banks, but this allowed them to look at banks in some countries for which purely national samples would have been too smal l to allow meaningful measures of relative efficiency. 49 Athanassopoulos et al (2000) compared bank branch networks across three separate countries - the UK Greece and Cyprus. They first constructed separate frontiers for each country. and then compared the performance of the efficient branches in each country against each other. They found the UK branches to be most efficient. but also. despite the much higher degree of regulation in Cyprus, that branches in Cyprus were not less efficient than those of the Greek bank studied. Chaffai et al (200 1 ) and Lozano-Vivas et a! ( 2002) have argued that many of these earl ier bank-level studies have not properly taken account of country-specific conditions or norms. Dietsch & Lozano-Vi vas (2000) looked at the French and Spanish banking sectors relative to a common frontier, which incorporated country? specific conditions. Previous approaches could mis-state the relative efficiency of firms from different countries, because they did not account for c ross-country 49 I ssues relating to sample size are d iscussed fut1her in Section 3 .3 , below. 4 2 differences in demographic, regulatory and economic conditions beyond the control of firm managers within their inputs and outputs. These previous approaches might generate artificial ly low efficiency scores for banks in countries where operating conditions were unfavourable, and high scores where conditions were favourable '0 (Chaffai et al, 200 1 , pp 1 4 7 - 1 48) . ) I n their study, Chaffai et al (200 1 ) used separate frontiers for each country, but then looked at the differences between the frontiers applying to the four countries studied. Environmental differences were found to be greatly more important than technological differences, whi le Gem1an banks were the most efficient, fol lowed by French and then Ital ian banks, with Spanish banks least efficient. Lozano-Vivas et al (2002 ) first analysed banks from 1 0 European countries usmg individual country frontiers. They then used a straightforward cross-country frontier, which generated much lower average efficiency scores for banks from each country than from the individual countries looked at separately. They then developed a cross? country model incorporating environmental factors (with a procedure specified for identifying and incorporating these). Some significant improvements in average country scores then became evident, which highlighted the disadvantages of operating conditions in particular countries. The environmental factors found to be most relevant were density of demand (value of deposits per square ki lometre), income per branch, equity over total assets and salary per capita. Casu & Molyneux (2003 ) used both separate country functions and a combined one for the five countries whose banks' efficiencies were compared - France. Germany, Italy, Spain and the UK. Regression was then used in an effort to find relevant country specific factors, which they referred to as teclmological. They described these as reflecting a legacy of different banking regulations and different managerial responses to the changing environment (p 1 873) . 5 0 Th is appears t o a lso relate t o t h e d isc ussion b y M ester ( 1 997) - a point spec ifica l ly noted by Lozano? Vivas et al (2002) - of the d i fferences in average X -efficiency between the third d i strict and the whole of the United States, where average X- inefficienc ies in the more nan?owly-defined third d istrict were rather lower. She rejected use of a s ingle cost function model for the whole country. See the furt her d i scussion in Section 3 . 4 . 5 below. 43 One aspect which has received relatively l itt le attention in this area i s in making some fom1al attempt to ascertain whether a common frontier might in fact apply in cross? country analyses . Edvardsen & F0rsund (2003) have looked at this i ssue in the context of the electricity distribution business in different countries, but there is l imited other material . 5 1 This research wil l explore the appropriateness of assuming a common frontier between Australian and New Zealand banks, according to a method outl ined in section 4 .4.4, with results reported i n section 5 .4 . 3.3 Approaches to effic iency measu rement Because there is no agreed set of engineering relationships defining a standardised set of production processes in banking, there is no simple, readi ly agreed approach for specifying the production function and related efficient frontier. Attempts to determine the position of the efficiency frontier are thus dependent on use of accounting information and any other measures of input or output volume that may be avai lable. Berger & Humphrey ( 1 997) identify five different approaches to determining the efficiency frontier. The three main parametric approaches to specification of the efficiency front ier are the stochastic frontier approach (SF A, also !mown as the econometric frontier approach), the distribution-free approach (OF A) and the thick frontier approach (TF A). while the t\vo non-parametric approaches are data envelopment analysis (DEA) and the free d isposal hull (FDH) method .52 53 A maJor chal lenge for both sets of approaches i s in distinguishing random error. ari sing from accounting practice or some other source, from inefficiency. Each of the parametric approaches has d ifferent ways of dealing with random error, whereas the 5 1 E lyasian i & Mehdian ( 1 992) suggest that the hypothesis of a common frontier can be tested by comparing the probabi l ity density funct ions of the two group, a lthough it wou ld seem that we m ight not know whether a d i fference was caused by a d ifferent e fficient frontier, or by a difference in the relative effic iency of the banks being studied. 52 L ists of approaches to frontier analysis often omit the FDH approach. which may be regarded as a special case of D E A . Thral l ( 1 999) has crit i c i sed FDH as being devoid of economic mean ing, but note a lso the defence of FDH by Cherchye et al (2000). 53 Some work has been attempted more recently using a neural network approach. See Sant in et al (2004 ) for a review ofthis . 4 4 non-parametric approaches have general ly fai led to deal with it at a l l (although recent work is exploring ways of dealing with it within DEA). 54 SF A reqmres specification of a functional form for the cost, profit or production relationship between inputs, outputs and environmental factors, but problems may then arise from the way the functional form has been speci fied, in that it presupposes the shape of the efficiency frontier, and for the commonly used translog approximation, this may impose minimum costs over a relatively narrow size range, thus generating misleading interpretations in relation to economies of scale in particular (Berger & Humphrey, 1 997; McAll i ster & McManus, 1 993) . One solution to this has been use of the Fourier-tlexible form, but one is sti l l then faced with the problem of dist inguishing random error from inefficiency. 55 A common approach to this has been to assume that the errors fol low a half-normal di stributi on, although Berger & Humphrey ( 1 997) describe this as relatively inflexible. Other d istributions include the truncated normal and gamma d istribut ions, although use of these may lead -6 to other problems (Berger & Humphrey, 1 997, p 1 78) . ) The DF A takes a different approach, assuming that random error v;i l l average out to zero over time, and thus requires use of panel data covering a sufficiently long t ime period for this to be possible. 57 The differences between individual bank performance and the frontier v;i l l thus provide a measure of inefficiency. An advantage of DF A is that i t imposes a less restrictive form on the frontier production function than does the SF A. but it cannot capture X-efficiency changes within a firm over t ime (Berger, 1 993 , p 265) . Moreover, its X-efficiency estimates for a bank apply only in respect of 54 Note, however. the point made by Grosskopf ( 1 996) that both the parametric and non-parametric approaches may be impacted by outl iers . She notes u se of chance-constrained programming as a solution to this in DEA. 55 Berger & Humphrey ( 1 99 1 ) refer to the need for ad hoc assumptions to disentangle ineffic iency differences from temporary or random fl uctuations in costs or output. Rime & Stiroh (2003 ) note that the Fourier flexible fom1 requires the estimation of additional parameters, and it can thus be difficult to use for smal l samples. 56 Berger & De Young ( 1 99 7 ) note some significant d ifferences in the resul ts obtained from using a Fourier-flexible form rather than a translog cost function, and the truncated normal rather than the half? normal error distribution. Wheelock & Wi lson (200 I ) outl ine some of the problems ident i fi ed with use of the Fourier flex ible form (pp 66 1 -663 ). 57 On the other hand, the t i m e period must st i l l be short enough so that the level of X -effi c iency for individual banks does not change (De Young, 1 997a) . This is a more general i ssue with analysis of panel data, as is discussed further below. 4 5 a period of time, rather than in respect of individual (typically annual) observations (Berger & De Young, 1 997) . The TF A specifies a functional form, and assumes that differences within the highest and lowest performance quartiles of observations (strati fied in size c lass) represent random error, while differences between the highest and lowest performance quartiles represent inefficiency. TF A does not provide estimates of inefficiency for particular banks, but i s intended to provide an estimate of the general level of overall ineffic iency (Berger & Humphrey, 1 997, pp 1 78- 1 79). Berger & Humplu?ey ( 1 99 1 ) identi fied a key advantage of TF A as being the Jack of a requirement that the inefficiencies be orthogonal to the outputs and other regressors specified in the cost function (p 1 22) . By contrast, Mester ( 1 996) notes that the division into quartiles i s to some extent arbitrary,58 while there is also a potential econometric problem because of the pre-scoring having been based on average cost, which is a dependent variable. Bauer et al ( 1 998) suggest that using only 25% of the data for any given year to estimate the cost function may add significant noise to the model, relative to other approaches. DEA is a l inear programmmg technique where the frontier i s assembled on a piece\:vise basis from the best practice observations (which will then be c lassified as 1 00% efficient) . 59 I t does not specify any functional form for the data, allowing this (reflected in the weights for the inputs and outputs) to be determined by the data.60 A major problem is that measurement error and luck are assumed av;ay, with no allowance for random error. so that all variations are treated as inefficiency (Berger & Humphrey, 1 99 1 , p 1 20) .6 1 This means that the position of the frontier may end up somewhat artific ial , leading to higher estimates of inefficiency than might be derived using a parametric approach. 58 Lozano- Vi vas ( 1 997) experimented with th irds and quint i l es, and found that the l arger the subsets used to measure best and worst practice, the smal ler was the average inefficiency. 59 F D H is a special case of D EA where the frontier is constructed as a series of steps between the points of opti mal performance (whereas DEA assumes that the l inks, also known as facets, are straight l ines). 60 E lyasiani & Mehdian ( 1 99 2 ) also suggest that DEA avoids the mult icol l inearity problems that plague the paramen?ic approaches. 61 Thus, as Favero & Papi ( 1 995, p 3 8 8) note, it is important to check the robustness of the results obtained using DEA, using techniques such as looki n g at d i fferent input and output variables, and to be wary of out l i ers, to which DEA is much more sensit ive. On the other hand, Chu & L i m ( I 998) suggest that, if audited financial statements are used, such errors w i l l not be a problem. 46 The basic (multipl ier form of the) DEA problem,62 in the constant returns to scale version, can be expressed as a requirement to maximise efficiency, for output weights u and input weights v, for ; inputs x and } outputs y (with bold to indicate vectors) . I f the weighted sum of inputs is set a s 1 , in mathematical notation this gives a requirement to subject to maxuv (uyj) VXi =} ll)0 - VXi < 0 U, V > 0 ( 1 ) Evanoff & I srai levich ( 1 99 1 ) noted that use of DEA l imited scope to unde11ake statist ical inference. A major reason for this is that the distribution of efficiency scores is neither known nor specified (Ferrier & Hirschberg, 1 997), whi le they wi l l a lso be dependent on each other. Efficiency scores wi l l a lso be l imited to the range 0 to 1 , with a tendency for scores to be c loser to 1 , which means that. if one wishes to regress efficiency scores against environmental factors, one should use a logit or preferably tobit regression (Coel l i et a l , 1 998, pp 1 70- 1 7 1 ). OLS regression i s not appropriate (Grosskopf, 1 996). 63 Some progress i s now being reported in overcoming these l imitations. Contributions in this area, based on the bootstrapping of efficiency scores, and aimed at developing a di stribution for each DMU' s score, include Ferrier & H irschberg ( 1 997) and S imar & Wil son ( 1 998, 2000). Casu & Molyneux (2003) summarise some of this debate and uti l i se the resultant method in a study of banks from five European countries . An alternative line of research looks at the distribution of scores of al l DMUs (Tortosa? Ausina, 2002b). 62 For a more extens ive discuss ion of DEA mathematics, refer to Avkiran ( 1 999b), Coe l l i et al ( 1 998), and Cooper et al (2000), including the further references they provide. 63 Note that OLS regression i s sometimes used, but the parameter estimates will not be able to be rel ied on. Note also that any regression w i l l have no regard to infonnat ion contained in s lacks and surpl uses, potential ly further b iasing parameter estimates ( Fried et al , 1 999, p 25 1 ) . Bhanacharyya et al ( 1 997) adopt a quite different solution: in their i n itial analysi s they used DEA, but then used output from SF A on the same data as the dependent variable for the regressions i n which they explored the sources of e fficiency d i fference. 1t is not immed iately obvious that th i s w i l l overcome the l im itation caused by the d istribution of effic iency scores, which m ust in any case be truncated at I . 47 Note, however, that bootstrapping approaches will not deal with the problem of random error or outliers in the data from which DMUs' scores have been estimated.64 There is no obvious approach to the identification of outliers which are inefficient ( in that one cmmot necessari ly distinguish between inefficiency and random error in the case of low efficiency scores), but for firms that show as fully efficient, random error effects may be located by use of the super-efficiency model (Anderson & Peterson, 1 993) .65 The super-efficiency model generates an efficiency score for each DMU based on a frontier that comprises al l the other DMUs in the set, and DMUs that were c lassed as CCR-efficient may thus achieve a super-efficiency score greater than 1 . If a super? efficiency score exceeds 2, it is suggested that the DMU in question can be identified as an outl ier, and that it should therefore be omitted from the analysis (Hartman et al, 200 1 ) . Although the super-efficiency scores are no longer censored at 1 , caution i s sti l l required in undertaking statistical analysis, a s the efficiency scores wil l sti l l be dependent on each other. Note that testing for the significance of differences in average DEA scores for groups of observations may be undertaken using the non-parametric Mann-Whitney test ( Cooper et al, 2000; Casu & Molyneux, 2003 ) . This may be preceded by use of the Kruskal-Wall is test to look for differences in the efficiency di stributions for the d ifferent groups (Athanassopoulos et al, 2000). Other tests used include the median test, Wi lcoxon test, Savage scores test, and Van der Waerden test (Grabowski et al, 1 993 ; Fukuyama, 1 993) .66 Drake & Weyman-Jones ( 1 996) suggest that DEA provides more straightforward observation-speci fic measures of inefficiency ( i .e . at the level of the individual firm) .67 The abil ity to identify a peer group whose operations may be emulated or 64 Thi s point was highl ighted by W i l l iam Greene i n d iscussion of a paper he presented at A PPC 2004 in Br isbane, Austra l i a on 16 J u ly 2 004. 65 The super-effic iency model exists in more than one form - see Tone (2002 ). Note that the software D EA-Solver described by Cooper et al (2000) uses the Slacks-Based Super-effic iency model described by Tone (2002). 66 G rosskopf ( 1 996) notes that this l atter group are generated as standard output from SAS procedures. I nterestingly, she makes no mention ofthe Mann-Wh itney test. 67 We thus see Rezvan ian & Mehdian ( 2 002) run ning OEA in addit ion to a parametric approach, so that they could spec i fically investigate the production perfonnance of a sample of Singapore banks. We 4 8 targeted provides a c learer agenda for the analyst as to what might be done to improve efficiency at the individual firm level (A vkiran, 2002, p 1 60). 68 This is at least in part because DEA i s a methodology directed towards frontiers, in contrast with the focus on central tendencies in the regression approaches that underpin the parametric methods (Seiford & Thral l , 1 990). Thus, as Charnes & Cooper ( 1 985 ) explain it, "DEA optimizes on each observation, whereas the usual stat istical regression optimizes across al l observations" (p 6 1 ) . 69 Another aspect of this i s that DEA provides scope for specific analysis of slacks and surpluses (Fried et a! , 1 999). DEA also allows study of jointly-produced multiple outputs, whereas the parametric approaches are normal ly l imited to focussing on a single dependent variable, such as cost, revenue or profit70 (Avkiran, 2002 , p 50) . The different outputs used in DEA may reflect different firm objectives, which might not so obviously relate to economic optimisation, such as service quality7 1 (Soteriou & Stavrinides, 1 997) or market value, particularly if the analysis i s to support a benchmarking exercise (Bergendahl, 1 998) . Chu & Lim ( 1 998) look at growth in assets as a performance objective. Resti (2000) notes the potential value of looking at intangible outputs such as service qual ity, customer and employee satisfaction, although he notes that the difficulty in using these i s that thei r measurement i s incl ined to be difficult and uncertain . In thi s regard also, DEA does not require the assumptions of cost minimisation or profit maximisation (Alam, 200 1 ) . Another i ssue i s use of information on the prices of inputs to bank production. Price information i s general ly regarded as being necessary for the parametric techniques, whereas DEA can sti l l be used to provide assessment of relative efficiency without this information, although as Evanoff & I srai levich ( 1 99 1 ) and Berger & Mester also see S i ems & B an ( 1 998) identi fying DEA as a useful tool for benchmarking exerc i ses, while Fukuyama ( 1 993) j ustifies use of DEA to focus on results at the individual bank leve l . 68 In fact, as Avkiran ( 1 999b) notes, relative efficiency measures in DEA are only relat i ve to the peer group of efficient firms aga inst which an ineffic ient firm is being compared, rather than relative to the whole set of efficient fi rms. DEA also al lows identi fication of a global leader, spec i fied as the effic ient DMU that most often appears in the reference set of ineffic ient D M U s . 69 E lyasiani & Mehdian ( 1 990) t h u s note that effi c iency measures a n d t h e rate o f technolog ical change w i l l be measured relative to the average performance of the sample, rather than rel at i ve to sample best pract ice. 70 Resti (2000) identifies th is as a particular strength of DEA, a lthough Berger et a ! ( 1 993a) note the advantage of the profit function in identifying i neffic iencies on both the i nput and output s ides. 7 1 Note that Athanassopoulos (2000) has developed a set of models, the first stage of which generates a service qual ity measure as an output, which then becomes an input into the second stage DEA model. 49 ( 1 997) note, one can therefore only study technical efficiency with no investigation of allocative efficiency possible. 72 More particularly, as Cooper et al (2000) describe it, for DEA the measurement units of the different inputs and outputs do not need to be congruent (p 22), allowing stock and flow variables to be dealt with in the same model . DEA can address both quantitative and qual itative data and d iscretionary and non-di scretionary variables (A vkiran, 1 999b, p 2 1 3 ; Golany & Storbeck, 1 999, p 1 5) . 73 Another issue in the choice of approach, although not d iscussed by Berger & Humphrey ( 1 997), i s sample size . The parametric techniques require significant numbers of observations for their regressions, which wil l be of l imited value if the number of observations in the data set is not significantly greater than the number of parameters estimated. 74 By contrast, as Evanoff & I srailevich ( 1 99 1 ) note, use of DEA al lows one to work with less data, fewer assumptions and a smaller sample (p 22) . A rule of thumb commonly used with DEA suggests that the number of observations in the data set should be at least three times the sum of the number of input and output variables (Cooper et al , 2000, p 252) . 75 That i s not to say, however, that DEA vli l l not generate better results with larger data sets, and Berger et al ( 1 997) identify a major problem with prior studies of bank branch efficiency as the small number of observations relat ive to the input, output and environmental variables (p 1 45 ) . Where a sample i s smal l , i t i s possible that a high proportion of firms will be classed as efficient, some of which would not otherwise show as efficient if a larger sample were used. As Nunamaker ( 1 985) has identified, inclusion of additional ( input 72 In th is case, the efficient fTont ier may be described as a production frontier ( Pastor et ai, 1 997) . Berger & M ester ( 1 997) thus suggest that non-parametric tec hniques thus focus on technological rather than economic optim isation (p 905 ) . Some researchers using DEA who have tested for al locative effic iency have found it to be signi ficant (e .g. Rezvanian & Mehdian, 2 002), wh i l e others have found it to be insignificant (e.g. Aly et a l , 1 990; Ferrier & Lovel l , 1 990). On the other hand. E lyasiani & Mehd ian ( I 990) h ighl ight the unre l i abi l ity of input price data, even if it is avai lable. 73 DEA may not face some of the stat istical and econometric constraints that might impact on parametric approaches. Gong & Sic kles ( I 992 ) thus note that DEA may be preferred over SF A where inputs are conelated with technical effic iency. 74 Th is proved to be a constraint for Walker ( 1 998), who looked at I 2 A u stral ian banks for I 3 years. 75 There is an alternat ive expression by Dyson et al (200 I . p 248), which says that the number of observations should be at least twice the product of the number of i nputs and outputs. Avkiran (2002) suggests a further rule of thumb - that a sample is large enough if the n umber of ful ly efficient DM U s does not exceed one third o f t h e sample. 50 or output) variables in a DEA model cannot cause reported efficiency scores to be reduced, and may results in DMUs appearing to be more efficient. 76 This pattern of choice as to the approach to be fol lowed is incl ined to resemble the wider international experience reported by Berger & Humphrey ( 1 997), who found 69 applications of non-parametric techniques (almost all DEA) and 60 of parametric techniques amongst the studies they reviewed. They were otherwise unable to specify which approach might be regarded as best, although it is noted that their own research has almost always used parametric methods. The previous studies of financial institution effic iency in New Zealand (Liu & Tripe, 2002 ; Tripe, 2003) used DEA, as have most Australian banking studies. The exception was Walker ( 1 998), who used SF A on panel data, although his research raised questions as to the homogeneity of the cross-section of banks and the possibi l ity of technological change over the period studied. (These i ssues are d iscussed further below) . One approach to companng techniques for efficiency measurement was the development of a set of consistency conditions by Bauer et al ( 1 998) . They suggested that the efficiency scores generated by different approaches should have comparable means, standard deviations and other distributional properties; that the different approaches should rank the institutions in roughly the same order. with mostly the same institutions ranked as most and least efficient. Al l approaches should shov?l reasonably stable efficiency scores for the same institutions tlu?ough time, efficiency scores should be consistent with market conditions, and reasonably consistent with other measures of bank performance. When they tested these, they found them to general ly apply, apart from the much lower scores reported under DEA, for which they had no certain explanation. A similar analysis, but only companng three approaches, was applied to five European countries (against separate frontiers) by Wei l l (2004). He found that the results of the parametric studies d id not correlate with the DEA results although the DEA efficiency scores appeared to be more c losely related to other measures of bank 76 This is consistent with the broader issue of d imensional ity, d iscussed by H ughes and Yaiswarng ( 2 004). 5 1 performance. His conclusion was to affirm that of Bauer et al ( 1 998), that it was wise to try different techniques to affirm the results of any efficiency analysis . Some other studies have used DEA because of smal l cross-sectional samples, as with Chu & Lim ( 1 998) in their study of the six major banks in Singapore, where the small sample appears to have led to unduly high efficiency scores. Canhoto & Dermine (2003) made specific mention of sample size in explaining their choice of DEA to study Portuguese banks, as did Isik & Hassan (2003) in their study of Turkish banks. One of the ways in which sample size IS expanded in non-parametric studies is through use of panel data, which can also al low study of how particular institutions' efficiency i s changing through time (Tulkens & Vanden Eeckaut, 1 995) . 77 One approach to this is window analysis, which entai ls using data for firms in d ifferent time periods as if they were separate decision-making units. This al lows a financial institution to be compared both with other institutions in the current period, and with itself and other institutions in other time periods (Charnes & Cooper, 1 985 ; Boussofiane et a l , 1 99 1 ; Yue, 1 992; Cooper et a l , 2000, pp 272-276). Lovel l ( 1 993, p 4 7) notes the role of window analysis in rel ieving degree of freedom pressures, 78 and also that it can assist in detection of outl iers: an average score from a \Vindow analysis covering a number of time periods v-ri l l be less impacted by the random enor that may othenvise be regarded as a problem for the non-parametric approaches. Thi s i s because \Vindow analysi s al lows multiple estimates of efficiency for each DMU for each period. This might be particularly appl icable in the case of quarterly data - where a smal l number of sequential quarters are being compared, changes in technology are unl ikely to cause major changes in efficiency. 79 77 Panel data has also been used to expand sample s ize to enable parametric stud ies using S F A, although as noted above in the discussion of OF A, there i s then the question of whether the technology is suffic iently invariant over the t ime period in quest ion (wh ich is a l so an issue when using DEA - see below). 78 This is the d imensional ity e ffect refetTed to by H ughes & Yaiswarng (2004). 79 Window analysis general ly assumes that, since al l un its w ithin a window are measured aga inst each other. there i s no technical change occurring within that window (Asmi ld et al. 2004, p 70). On the other hand, panel data can be analysed without this assumpt ion, and i f a change in effic iency scores is observed, the question can then be asked as to whether t h i s i s attributable to technical change. See also Tulkens & V anden Eeckaut ( 1 995) . 52 Users of window analysis in banking include Yue ( 1 992) who looked at banks in the U . S. state of Missouri , and As mi ld et a! (2004) who looked at Canadian banks. Use of DEA for panel data has been reported by Drake (200 1 ) in looking at the UK market, where the cross-section of banks was not large enough for single year studies. Canhoto & Dem1ine (2003) used both a panel and individual year cross-sections in their study of Portuguese banks. Bhattacharyya et a! ( 1 997) used a panel (which they describe as a grand frontier) for their study of Indian banks, noting the advantage such an approach brought in terms of rel ieving degrees of freedom pressures and increasing the variation in calculated efficiencies. The more common l inear-programming based technique for examining changes in efficiency through t ime, and which potentially caters for technical progress, i s the Malmquist Index, the first application of which in banking was undertaken by Berg et al ( 1 992) . The Malmquist Index uses panel data to derive a measure of total factor productivity change, which c an in tum be broken down into change in technical efficiency and technical change (which would cause the efficient frontier to shift) (Coel l i et a!, 1 998 , p 222) . 80 The Malmquist Index thus caters satisfactorily for firms that are not on the efficient frontier. 8 1 The Malmquist Index works b y comparing the quantity o f output produced i n period t+ 1 with that which would have been produced in period t+ 1 using the period t teclmology. The index i s thus abl e to have a score greater than 1 if there has been technical progress, if the firm has improved its technical efficiency, or if both of those effects have occurred. In mathematical tem1s the Malmquist Index is specified in tem1s of distance functions:82 where x and y refer to inputs and outputs respectively, an input-oriented Malmquist productivity change index may be formulated as follows (where the D refers to an input-oriented distance function) : M 1 1+ 1 1- 1 , _, 1 = [D'r 1 -'- 1 )+ I)ID'r , -' l * D'+ I1 1-'- 1 ,_._ J)I D'-'- 11 , ' l) J/2 IY ' X ' y ' X / IY . X IY ' X / IY . X IY ' X/ (2) 80 Technical change and techn ical effic iency are alternatively described by A lam (200 I ) as " innovati on and im itation" (p 1 22). Drake (200 I ) refers to them as the frontier shi ft and catching up effects (p 560). The Malmquist I ndex i s a geometric mean of these two effects. For a fUither discussion of the decompos it ion of the Malmqui st index, refer to Love ll (2003). 81 This provides an advantage over some other index number approaches, wh i ch require the assumption that firms are cost m i n i m i sers and revenue maxim isers (Coe l l i et a l , 1 998, p 22 1 ) . 82 For a more extensive d iscussion, refer to Fare & Grosskopf ( 1 997) and Coe l l i et al ( 1 998) . 53 The Malmquist Index has become increasingly popular, and significant work has now been undertaken in banking uti l ising i t . Looking at banks in Norway, Berg et al ( 1 992) found the greatest productivity improvement amongst the least efficient banks, with relatively l ittle improvement in efficiency for the most efficient banks. This finding would appear to be consistent with the suggestion in Section 3 .2 above that differences in bank efficiency might be reducing in response to competitive conditions. Looking at the United States banking market, Wheelock & Wilson ( 1 999) found that banks became less technically efficient over the period 1 984- 1 993 . They attributed this result to teclmical progress having occurred faster than bank-level efficiency improvement. Using a different approach, Bauer et a! ( 1 993) also found negative total factor productivity growth for U .S . banks over the period 1 977 to 1 988 : possible reasons suggested for this included flow-on effects from high interest rates in the late 1 970s, the phasing out of interest rate controls . the effects of non-bank competition and qual ity improvements. I sik & Hassan (2003 ) looked at the effic iency of the Turkish banking system over the period 1 98 1 to 1 990. which covered the period of deregulation. They found that there was initia l ly a reduction in efficiency in some cases, but that, after around 1 986/87, efficiency began to improve again . Over the whole period, pure technical efficiency and (to a lesser extent) scale effic iency both improved, although there was technical regress. Leightner & Lovell ( I 998) and Gilbert & Wilson ( 1 998) looked at the Thai and Korean banking systems respectively, over periods when deregulation was occurring. Considerable productivity improvement was found in both cases, pm1icularly in Korea where events were reviewed over a 1 5-year period. Leightner & Lovel l noted that productivity gains were more important when there was a focus on bank profi tabi l ity rather than on Bank of Thailand objectives of increasing financial i ntennediation. Gi lbert & Wilson found the most impor1ant changes to be in terms of teclmology used, which they associated with changes in mix of inputs and outputs. 54 For the Austral ian market, Avkiran (2000) looked at the period 1 987 to 1 995, and found that teclu1ical change had more of an impact than pure technical effic iency in respect of the observed improvement in total factor productivity. Sathye (2002) looked at the period 1 995 to 1 999 and also found some improvement in total factor productivity, which could in this case be attributed to pure technical efficiency. As with any other analyses, if the Malmquist Index i s to provide meaningful results, individual year cross-sections must be large enough to be able to distinguish inefficient banks. This was a constraint for Drake (200 1 ) in his study of UK banks, where he attempted to overcome the problem by using multiple year (window) base periods. One also needs to l ook at a long enough time period for significant teclmical change to occur. This appears to have been a constraint for Noulas ( 1 997) in his study of Greek banks, where he looked for changes over a period of only two years . Market conditions can also be a factor. Fukuyama ( 1 995) looked at Japanese banks over the period 1 989 to 1 99 1 , and found evidence of technical progress in the first year, to 1 990. but technical regress between 1 990 and 1 99 1 . lt was suggested that this was a consequence of the bursting of the economic bubble in 1 990, with the start of an economic downturn. Bhattacharyya et al ( 1 997) decided against usmg the Malmquist Index for their analysis of Indian banks, preferring instead to run a DEA study on a grand frontier of Indian banks for the period 1 986- 1 99 1 . They noted the advantages of this approach in providing a single benchmark against which to evaluate performance and its change through time. Such an approach also gets around the problem of an unbalanced panel, caused by the entry and exit of banks from the market . Asmil d et al (2004) have tried usmg the Malmquist Index and window analysis together in a study of the five major Canadian banks, a group which would be too small by itself to be able to be studied using individual year cross-sections. The problem they found was that because the window analysis-derived cross-sections covered a period of time, it was not possible to uniquely define the base period 55 technology to which the Malmquist Index should be applied. They studied a 20-year period, within which they used five-year windows, which they regarded as small enough to minimise the problems of unfair comparisons over time, but sti l l l arge enough to provide a reasonable sample size. 3.4 Pract ical issues i n F inanc ia l I nstitut ion effic iency measurement The mere identification of the different techniques with which effic iency analysis might be undertaken is not the end of the process of preparation for effic iency analysis . There is a range of other decis ions required of researchers and relevant i ssues are reviewed in this section . There i s incl ined to be some emphasi s on Data Envelopment Analysis (DEA) in this section reflecting the use of this technique in the research reported l ater in this thesis, but many of the required choices relate to a broader range of approaches to financial institution efficiency measurement. This section begins by looking at how scale economies can be investigated within DEA, reflecting a choice in the type of model to be used. The section then looks at another DEA i ssue, model orientation, although there is a related i ssue that needs to be considered in the parametric approaches, in terms of whether the analyst uses a cost, profit or revenue function. 83 Another i ssue that applies to almost a l l anempts to analyse financial inst itution efficiency i s the ident ification of the inputs and outputs of the production process. Further i ssues to be considered in thi s section include the speci fication of the group of DMUs whose efficiency is to be compared, and the problems of time effects and interest rates identified in Tripe (2003) . 3.4. 1 Retu rns to scale in DEA An important i ssue that has to be decided with DEA i s the type and orientation of the model to be used. DEA was original ly developed on a constant returns to scale (CRS) basis by Charnes et al ( 1 978), and then extended to variable returns to scale (VRS) 8 3 For a d iscussion o f cost a n d profit functions and the d ist inctions between them. s e e Berger & M ester ( 1 997), pp 897-904 . Note that some researchers (e .g. B erger & M ester, 1 997; Lozano- Vi vas, 1 997) have used alternative profit functions, wh ich take account of market conditions that are not perfectly compet it ive, by treating output volume as fixed. 56 fonn by Banker et al ( 1 984), with the VRS version allowing for the separate identification of technical and scale inefficiency, as explained below.84 The VRS approach also al lows i dent ification of whether a DMU is operating at increasing, constant, or decreasing returns to scale. By implication, a DMU's optimal scale of production can thus be identified. Scale efficiencies are most commonly measured by running the same data through both the constant and variable returns to scale models: scale efficiency is found by d ividing the CRS score by the efficiency score from the VRS model .85 The difference between the constant and variable returns to scale models is that the VRS model envelopes the data points more tightly than the CRS model, and the efficiency scores from the VRS efficiency must therefore be greater than or equal to those from the constant returns to scale model . Scale efficiency measures wil l thus be in the range zero to one. 86 Coell i et al ( 1 998, p 1 50) note that the VRS model has been most commonly used since the beginning of the 1 990s, and that would also be the case with DEA studies of financial institutions. Caution must be exercised in use of VRS models, however, particularly where cross sections are small , and where there is diversity in the size among the institutions being studied. As Dyson et al (200 1 ) note. if a VRS model is used, smal l and large units will tend to be over-rated in the efficiency assessment. Thi s means that scale inefficiencies identified for such institutions may be spurious, with the actual cause of i nefficiency being X-inefficiency. 87 If a CCR model is being used and it is found that a DMU is inefficient, it may be difficult to ascertain whether 84 The CRS and V R S models are common ly refened to as the C C R and BCC models respectively. after the authors of the original art i c les. 85 Th is generates whcrt Schaffn i t et a l ( 1 997) refer to as a spread ratio. which they define as the rat io of two effic iency scores for the same D M U . Although th i s i s commonly used to provide measures of scale e ffic iency, it may a l so be used to assess models w ith d ifferent numbers of variables, data points, etc : Schaffnit et al suggest that the Malmquist I ndex is also a type of spread ratio when it is being used to compare different technologies. These i ssues are d iscussed again i n the analysis which i s described in section 4 .4 .4, and reported on i n section 5 .4 . Note that such a decomposit ion i s often also appl ied to the M almquist Index, to identi fy the impact of scale e ffects on product ivity changes through time, although this may raise issues with respect to the effects of changing output m ix ( Balk, 200 I ) . See Love l l (200 3 ) for further discussion of the val idity of t h e scale decomposition. 86 Note that th is contrasts with approaches to identifying econom ies of scale under the parametri c approaches, which look a t t h e aggregate data set, rather than focusing o n t h e leve l of t h e individual fi rm (Athanassopoulos, 1 998, p 1 8 7) . 87 For a further discussion of t h i s in a pract ical banking context, see Tripe (2003). This issue is explored further in section 5 .2 below. 57 the source of that inefficiency i s scale or X- inefficiency. 88 Lovel l (2003) has thus noted that scale diseconomies may be perceived as a departure of the best practice technology from the benclunark teclmol ogy. Avkiran ( 1 9998) has suggested that researchers should run both CRS and VRS models. Then, "if the majority of DMUs are assessed as having the same efficiency under both methods, one can work with CRS without being concerned about scale efficiency confounding the measure of technical efficiency'? (p 2 1 1 ). A further feature of VRS model s (in most DEA software) is that they report whether a DMU i s operating at increasing, constant or decreasing returns to scale. Constant returns to scale will apply when the CRS and VRS efficiency frontiers are tangential with each other: in other words, when the ( local) slope of the efficiency frontier is equal to the ratio of input(s) to output(s) ( Cooper et a!, 2000, pp 1 1 6- 1 1 7) . Increasing returns to scale must apply below that level, as the slope of the efficient frontier (which reflects the marginal rate of transformation of inputs to outputs) wil l be greater than the average rate of conversion (which is equivalent to the average cost ) . Likewise, decreasing returns to scale must apply must apply above the zone in which constant returns to scale apply. DMUs not on the efficient frontier must first be projected onto the efficient frontier before their returns to scale status can be assessed. 89 The way in which returns to scale status is commonly determined is by running a non? increasing returns to scale (NIRS) mode l . A N IRS model can only show constant or decreasing returns to scale. and thus, if the efficiency score under N IRS is the same as under a VRS model , the DMU must be operating at decreasing returns to scale. If the scores are different, with the NIRS score lower than the VRS score, the DMU i s operating at increasing returns to scale (Coel l i e t a ! , 1 998, pp 1 5 1 -2 ; A vkiran, 1 9998, pp 2 1 1 -2 1 2 ; Seiford & Zhu, 1 999; Avkiran, 2002, p 58) .90 88 This problem is compounded by the tendency for X-effic iency to increase with a financial institution ? s s ize ( Berger H ancock & Humphrey, 1 993), as d iscussed above in section 3 .2 . 89 Golany & Yu ( 1 997) highl ight some of the complexit ies and potentia l inconsistenc ies i n this process. This relates to the problem discussed in section 3 .2 above of defining the returns to scale status of firm s that were not on the frontier (Berger et a l , I 987; Berger et a l , I 993b ) . 90 Some D E A software reports return to scale status analysis from Y R S model output, rel ieving the researcher of the need to undertake thi s separate analysis . 5 8 Despite the methodological concerns, there is some consistency in suggestions that l arge banks may be scale inefficient, as was found by Drake (200 1 ) in respect of the big four UK clearing banks, which were all found to be operating with decreasing returns to scale. Drake & Hall (2003) found a simi lar effect for the ( large) Japanese c ity banks. Christopoulos et al (2002) found that large banks in Greece were generally less efficient than small and medium-sized banks. Thi s is also consistent with the d iscussion by Berger & Mester ( 1 997) where they note that previous studies (many of which used parametric methods) had found that cost scale economies were exhausted well below $ 1 0 bil l ion of assets (p 927). 3 .4 .2 Model orientat ion DEA models wil l commonly have either an input or an output orientation. An input orientation aims at reducing the input amounts as much as possible while keeping at l east the present output levels, while an output orientation aims at maximising output levels without increasing use of inputs (Cooper et al , 2000, p 1 03). The focus on costs in banking means that input-oriented models are most commonly used, although in the CRS case, the same efficiency scores are generated by both approaches .9 1 This choice i s also a reflection of what management is able to change. For a financial institution as a whole, it is easier to reduce inputs than it is to increase outputs, growth in which would be likely to be constrained by aggregate demand in the market as a whole (particularly if the market is not characterised by perfect competition), and which are not under management control .92 Where inputs are not control lable by management, for example, for a comparison of a financial institution? s branches, an output-oriented approach may be more appropriate (A vkiran, 2002, p 57) . If a DMU i s identified as inefficient under an input-oriented approach, i t wil l shov.' as over-uti l i sing inputs, but it may also show as under-producing one or more outputs . 9 1 Even in V R S models, t h e sam e D M Us w i l l b e ident ified as e ffic ient o r ineffic ient, but the effi ci ency measures w i l l d i ffer (Coe l l i et a l , 1 998) . Thi s is because the efficiency score i s being measured relative to different axes, according to the orientation of the model . 92 See also the discussion of input and output orientat ions relat ive to the M almquist Index in Pastor et al ( 1 997). 59 This indicates an output slack. Similar circumstances can identify input s lacks when an output-oriented approach is being used (A vkiran, 1 999b ). There is some argument that restricting the choice to either an input or output orientation may neglect major sources of technical inefficiency in the other direction, which provided a justification for Berger et al. ( 1 993a) to look at profit efficiency. There i s therefore some interest in uti l isation of the slacks based approach (Cooper et al, 2000), which is one of a set of what De Borger et al ( 1 998) refer to as global efficiency measures, because they treat input and output dimensions simultaneously. In his fonnal outl ine of the slacks-based model, Tone (200 1 ) shows that a slacks? based measure i s a product of input and output measures of CCR inefficiency, which means that any slacks-based measure must be less than or equal to a CCR efficiency measure with either input- or output-orientation. Tone further notes that the slacks? based model i s thus providing measures of profit efficiency, rather than the ratio efficiency measures provided by the CCR model . The slacks-based model allows for a further decomposition of inefficiency. in that a slacks-based efficiency score i s a product of a CCR score and a mix-efficiency measure. This is argued as being appropriate and necessary because a CCR efficiency measure is based on a Fanell rather than a Koopmans approach to inefficiency, taking account only of radial slacks, and not of non-radial slacks (Ruggiero, 2000 ) . 3 .4.3 Specification of inputs and outputs Another choice to be made in model l ing bank efficiency i s in specifying the inputs and outputs of the production process : differences in the input and output variables chosen are commonly found to impact on the efficiency scores generated, while Wheelock & Wilson ( 1 995) state that umeliable estimates of efficiency can be generated by models that omit key features of bank production . Tortosa-Ausina (2002b) suggests that conclusions relative to the efficiency and potentially the competitive viabi l ity of firms in the industry could depend on the model chosen. 60 A distinction i s made between the production and intermediation models, with the i ntermediation model existing in a number of different forms.93 Under the production approach, banks are regarded as using labour and capital to produce deposits and loans, with both inputs and outputs typically measured as physical magnitudes, rather than in dollars. The intermediation approach sees deposits and other funds being transformed into loans (Sealey & Lindley, 1 977), with its different versions including the asset approach, which uses funds as inputs and loans as outputs, the user cost approach, which looks at the various contributions to banks' net revenue, and the value added approach, where inputs and outputs are identified according to their share of value added (Berger & Humphrey, 1 992a) . Even though the inputs may include actual money costs, because the production approach focuses on physical measures of inputs and outputs, the relevant data can be hard to obtain (except for the sorts of studies where the performance of a financial institution' s branches are being compared). But the production approach also takes no account of the cost of deposits, which can be of particular importance because of the potential trade-off betv,,een interest and non-interest expense (Humphrey, 1 99 1 ; Berger & Humphrey, 1 992b).9? On the other hand, the production approach may be more appropriate for examination of the comparative efficiency of bank branches, particularly where there are differences in the patterns of transactional act ivities. The majority of such studies of bank branches have used the production approach.95 I f financial institutions were primarily or solely engaged in receiving funds at interest and using these to make loans, the asset approach might provide a fair enough description of their activities . It fai ls to take account of the other activities that banks undertake, however, for example in providing transaction services, \;>,1hich cause non? interest expense, and which in most cases contribute to non-interest revenue. If one looks at New Zealand banks, for example, it is found that approximately 35% of gross 93 For some discussion of the origin of the different approaches, see Shennan & Gold ( 1 98 5 ) . 94 Thus Berger e t a l ( 1 987) note that the intenned iation approach is t o be preferred for compet itive viabi l ity analysis because it i s inclusive of both operating and interest costs (and a competit ive firm would be seeking to m inim ise the sum of these for any given level of output). 95 Athanassopoulos ( 1 997) suggests that bank branch studies have e ither used DEA and the production approach, or parametric methods and one of the intermediation approaches (pp 302-3 03) . 6 1 d ?' b . . 96 d f h . Id mcome I S accounte 10r y non-mterest mcome, an omiSSion o t IS wou misrepresent the activities that banks undertake (Favero & Papi , 1 995) .97 Berger & Humphrey ( 1 992a, p 24 7) note a further problem with the asset approach, giving the example of two banks that merge. Prior to the merger, one bank lends substantial ly to the other through the interbank market, but once the merger occurs, thi s output is no longer recorded, making it appear as if the merger has caused a reduction in aggregate outputs. Berger & Humphrey ( 1 992a, p 248) describe the user-cost approach as determining whether a financial product i s an input or an output on the basis of its net contribution to bank revenue.98 A major difficulty with this approach is with services that are not charged for, or where interest rates also include a component for other financial services. An example of this is with the non-payment of interest on current accounts to support the non-charging of transaction or account keeping fees. In the USA there has often been a requirement for compensating balances as a condition for loans. Another complication with the user cost approach is that interest rates on both deposits and loans include some allowance for risk.99 It may thus be difficult to distinguish whether higher interest rates on loans are a reflection of efficient pricing or of greater credit risk. Higher interest rates on deposits may be a reflection of deposits being taken for a longer term, with reduced exposure to l iquidity risk, rather than being a reflection of inefficiency in raising funds. 1 00 The value added approach considers all asset and l iabi lity categories to have some output characteristics, with those categories with greatest added value being recorded 96 B ased on data for the 6 banks that dominate retai l business. for the period 1 996-2002 . 97 Rogers ( 1 998) goes further than this in highl ighting the broad range of off-balance sheet activities that banks undertake, and the i mpact of these on m easures of efficiency. 98 This c oncept originated in the work of H ancock ( 1 986, 1 99 1 ) . 99 This was a basis for McAII ister & M c M anus ( 1 99 3 ) to include capital as an i nput in their model, although M ester ( 1 996) proposes a number of further arguments for its inc l usion. See also Hughes & M ester ( 1 99 3 ) and the discussion in Berger & M ester ( 1 997), pp 909-9 1 0. As an alternative approach to gening around this problem (and the potential effect of ol igopoly power in the relative markets), Resti ( 1 997) recommends using figures for deposit and loans, rather than for the interest Jlows relating to them . 100 This i ssue may be mitigated by est imating profit efficiency rather than cost effic iency (Berger & De Young, 2 0 0 I ) . 62 as outputs. Deposits are thus commonly recorded as outputs, although Hughes & Mester ( 1 993) show that they ought more appropriately be classified as inputs. 1 0 1 In practice, as Favero & Papi ( 1 995) argue, a l l of these approaches have their strengths, but none of them is necessarily perfect. To properly account for a financial institution 's performance, it i s reasonable to also take account of off-balance sheet items (although these might be reflected in non-interest income), and capital . Thus, in their version of the asset approach, Favero & Papi include non-interest income as an output, to proxy for the services provided by banks . By contrast, Altunbas et al (200 1 ) include total off-balance sheet items (measured i n nominal terms). 1 02 The importance of off-balance sheet business as a source of non-interest income has been shown to have an impact on efficiency by Siems & C lark ( 1 997) and C lark & S iems (2002). Siems & Clark ( 1 997) suggested that the way in which off-balance sheet business was measured could have a significant impact on the efficiency scores generated . In practice, the imp011ance of some of the taxonomic distinctions may be overstated. One key factor that will determine what input and output variables are used will be what can be measured, and it most cases it i s not possible to obtain data at a bank level for numbers of (deposit or loan) accounts or transactions processed. 1 03 One may also want to take note of input and output variables used in previous research, and the impacts of the variables chosen on the results obtained. 1 04 A more important i ssue i s one that i s highlighted by Dyson et a! (200 1 ) . particularly where using DEA that the input/output set should cover the full range of resources used and outputs created, particularly if one real ly wants to assess a financial institution' s efficiency at converting inputs to outputs. which should result in adding value. At the same time, the researcher '"'i l l also want to be mindful of degrees of freedom constraints, and wil l want to avoid using these up by using input or output 1 0 1 Sealey & L indley ( I 9 7 7 ), who are regarded as the originators of the intermediation approach, were firm in their view that deposits should be c l assed as inputs. 102 Simi lar arguments mi ght apply in respect of non-perfonn ing loans. See Berger & De Young ( I 997). 1 03 One therefore often notes. with cross-country studies, that variables may be l i m ited to what is reported on the B ankscope database. This is acknowled ged, for example, by Maudos et a l (2002). 1 04 N unamaker ( I 9 8 5 ) warns of the danger of choosing i nput and output variables so a s to m axim ise individual firms' effic iency scores. This can be more of an issue under DEA than under other approaches, as one is using DEA to maxim ise effic iency scores under the DMU ' s best d i m ension. 63 variables which don' t contribute to the identification of bank effic iency. Common sense and expert judgement can play a role in this. It is important to include key resources as inputs and to include in outputs those objectives which are regarded as key to the DMU' s success (Avkiran, 1 999b). I n practical terms, the way of investigating these i ssues is to look at the stat istical relationships between the variables. Although some researchers report use of regression (Golany & Storbeck, 1 999), it is general ly regarded as satisfactory to look at correlation coefficients between the inputs, the outputs, and the inputs and outputs together. If inputs are highly correlated with each other, they are not going to effectively identify potential efficiencies (trade-offs) in input usage, and the d i scriminatory power of the model wi l l be reduced accordingly. 1 05 S imi lar c onsiderations apply with outputs. 1 06 With respect to the conelations between inputs and outputs together, a typical c riterion for inclusion i s that correlations should exceed 0.7 (Avkiran, 2002) or even 0 . 8 . The basis for requiring a high conelation coefficient is that this demonstrates that ut i l isation of a particular input is l ikely to have an effect on the quantity of output. This may not be the end of the process of selecting variables, however. Hughes & Yaiswarng (2004) note that there may be several variables that reflect a DMU' s activities, although no combination of these may ful ly capture all aspects o f the group of DMUs' activities. In such a context it may be appropriate to try a range of variables and look for consistency in the ranking of results. Where the results are consistent, one can have greater confidence in the reliabi l ity of DEA models. 3 .4.4 The impact of environmenta l factors Checking correlations between variables i s not a total ly rel iable basis for i dentification of appropriate input and output variables, however. Tripe (2003) i dentified a problem that arose \Vith use of interest cost as an input variable, with 1 05 Th is process should not be taken too far. however - see N unamaker ( 1 985) . 106 A vk iran (2002, p 40) notes that there wi l l be no distortion in efficiency scores, but rather that the d i scri minatory power of the model wi l l be reduced if h ighly con?elated inputs (or outputs) are retained. This is the degrees of freedom problem, which tends to be much more of an issue in DEA where sample sizes are often quite smal l . 64 interest cost being highly correlated with the general level of i nterest rates. With net interest income as an output, the efficiency scores generated by the models used in longitudinal studies ended up being negatively correlated with the general level of interest rates. The suggested finding that New Zealand banks had improved their efficiency through time could not therefore be rel ied on. If this particular combination of inputs and outputs had been tested usmg the correlation procedure outl ined in section 3 .4 .3 , they might well have been rejected from the model . This issue wil l be explored in greater depth in section 5 .2 below. This issue has been identified as particularly applying where there have been fluctuations in the general level of interest rates, but it may also apply in other c ircumstances as wel l . Care is required on the part of the researcher when there i s found to be a relationship between efficiency scores and enviro1m1ental variables. Note that some of these issues ought to be able to be overcome in a DEA context by use of weight restrictions in a constrained multipl ier model . Unconstrained DEA models al low each firm to be evaluated in the best possible light. but this can cause fim1s to appear to be efficient in ways that might be hard to justify in a logical approach. Constrained multiplier models incorporate prior knowledge and j udgement into the evaluation of each firm. In their study, Siems & Barr ( 1 998) obtained weights to apply to bank inputs and outputs from a survey of bank examiners . 3 .4.5 Choosing the data set in which efficiency is to be measu red An important requirement for meaningful comparison of firms' efficiency is that the firms be sufficiently similar to make comparisons meaningful . This is particularly the case with DEA, where Dyson et al (200 1 ) have developed \Vhat they describe as a series of homogeneity assumptions. The first of these is that the units the perfonnance of which is being compared should be undertaking similar activities and producing comparable products or services so that a common set of outputs can be defined. This might be extended to a requirement to use common technologies, but it is suggested that this should not be a binding 65 constraint (p 247) . This also relates to the i ssues of whether or not a common frontier applies to all the firms whose performance is being compared (noting that this is a particular issue when comparing banks in different countries). There can be a role for the exerc ise of judgement by the researcher in this area: if there is a DMU in respect of which there is some doubt as to whether it ought to be included in a study, it may be appropriate for its effect on efficiency scores to be tested. 1 07 If it shows as overwhelmingly efficient, it may be an indication that its operations are different and that it ought to be omitted from the study. The second homogeneity assumption (and a closely related implicit third assumption) is that a s imilar range of resources is avail able to all the units, and they operate in a s imilar environment. If the environments are different, these might need to be specifical ly accounted for in the analysis (p 24 7) . Thus, for example, in their study of branches of a large bank, Golany & Storbeck ( 1 999) started off with 200 branches, but reduced this to 1 82 because some of the branches were " . . outliers, performing unique activities that other branches did not perform?' (p 1 6). In terms of banking studies, the consequences of use of heterogeneous samples have been demonstrated by Mester ( 1 997) . She found that studies that looked at only a single Federal Reserve d istrict showed significantly higher average efficiencies than studies which looked at the United States as a whole . These issues would be l ikely to be of even greater significance in comparing the efficiency of banks across international borders. These points raise a number of practical i ssues for research, particularly in DEA, where exceptional results are able to distort overall efficiency measures. This would particularly apply when comparing the performance of banks in a country which has its system dominated by a smal l number of large banks, such as New Zealand. 1 08 Banks could be c lassed as inefficient j ust because they perfonn d i stinct (more expensive) services (T011osa-Ausina, 2 004). Apart from the approach used by 1 07 Thi s testing m ay be undertaken using the super-efficiency model (disc ussed above), although a lternative approaches include ident ification of D M U s which are in the reference sets for an overwhelming proportion of the inefficient DM U s . 1 08 l s ik & H assan (200 3 ) tested for whether domestic and foreign commerc ial banks in Turkey had identical techno logies, and found that it was appropriate to construct a common frontier by pooling data. 66 Elyasiani & Mehdian ( 1 992), however, there has been very l ittle testing of the appropriateness of assumptions relative to whether common frontiers apply : thi s is one of the issues that the research repo11ed in this d issertation attempts to address. 3.5 Summary This chapter has sought to review a range of previous research, with a very specific focus on issues that would be expected to come to the surface in a study of the ew Zealand market. It has thus not sought to provide an exhaustive literature review, but has summarised key points, and focused in more detail and depth on issues that wi ll be discussed in subsequent chapters. Key issues on which attention was addressed thus included the method to be used to specify the efficient frontier, studies of efficiency in cases where the number of firms whose efficiency is to be compared is smal l , and c ross-country and inter-temporal studies. This background review should provide some justification for the use of Data Envelopment Analysis, whi le it should also have made clear that the relatively small number of banks in the New Zealand market poses a particular challenge for research i nto their relative efficiency. It was because of these constraints that it was necessary to review approaches to undertaking cross-country studies. and studies covering multiple time periods. But there was also an interest in studies covering multiple time periods because of questions over whether banks have become more efficient through time, reflecting the changes that have occurred in the New Zealand banking sectoL some of which we reported on in Chapter 2 . With the relative theoretical background having been recorded i n this chapter, the next chapter goes to look in more detail at the method to be used for the research, and at the data that are available to support it . 67 4. Data and Method A number of the issues relating to the method used and data employed for this research have already been discussed in the previous chapter of this disse11ation. This chapter seeks to justify the approaches fol lowed in this research relative to that previous discussion. Thus, because of the relatively small number of banks with retail branch networks in the New Zealand and Australian markets, DEA is used, with an input-oriented model, to accord with the focus of many Australasian banks in trying to reduce costs, and to acknowledge that, in most cases, at the whole bank level, banks are much more readily able to influence costs than to affect revenues. Partly in response to the wide d ivergence in asset size among the banks included in the research, but also reflecting some of the difficulties that can arise with variable returns-to-scale models, a constant returns-to-scale model has generally been used in the first instance, with the idea that a variable returns-to-scale model could be appl ied later to investigate the existence of scale effects. In a number of cases variable returns-to-scale models are run to check for scale effects, and these situations wil l be commented on in the description of results in the next chapter. This research reports the results of a number of separate studies \:vhich build upon each other. These individual studies are discussed in greater depth later in this chapter. The chapter begins by defining the data set and then reviewing the data that is available on the New Zealand and Austral ian banking systems at an indiYidual bank leve l . The chapter goes on to explore some general methodological issues before introducing the individual studies that make up this research. 4.1 The data set stud ied In section 3 .4 .5 , the problem of sample homogeneity was discussed, and i t was noted that the firms involved in a study needed to be sufficiently similar to each other for their performance to be able to be meaningfully compared. This relates to the question of whether or not a common efficient frontier applies. 6 8 The primary focus of this research is on the six banks operating in New Zealand with extensive branch networks and a significant focus on retai l banking: ANZ, ASB, BNZ, NBNZ, TSB Bank Limited (TSB) and Westpac NZ. 1 09 Although the New Zealand Government has established Kiwibank tlu?ough New Zealand Post, it only commenced business in early 2002, and its relatively short period of operation and its failure to earn consistent profits during the period of the study would make it unfair to include it in the study, despite its extensive branch network. 1 1 0 Superbank, which commenced business only in February 2003 , and which uses New World supermarkets and other Foodstuffs outlets as its public face, has been omitted from the study for similar reasons. Some previous research (e.g. Liu & Tripe, 2002) has included a wider selection of New Zealand banks in its data set, to get around problems that would arise from use of a narrower data set, in terms of the models not having sufficient d iscriminatory power. In retrospect, it is not clear that that was necessari ly a correct choice, in that the range of firms included in that previous research were too diverse in tem1s of the types of business that they were undertaking. 1 1 1 For this study it is regarded as preferable to use a narrower data set, which can be expanded by use of multiple periods of data in a single DEA model (as with window analysis or panel data approaches), so as to achieve an adequate sample size. These concerns over homogeneity in the group of banks to be studied can be demonstrated by looking at a cross-sectional comparison of New Zealand banks' performance for financial years ending in 2003 , and reported in Table 1 below. 1 1 2 There are significant and major differences in the percentages of net interest income, non-interest income and operating costs relative to average assets for the different banks, according to whether they have branch netv-wrks, whi le it is al so evident that there are significant differences in bank size. 109 Although TSB is a great deal smaller than the other bank s in the sample, the resu lts generated do not suggest that it suffers in terms of the efficiency scores generated for it by the analys is : if anything, it shows as being more e ffic ient. There is therefore no obvious basis for excluding it from this ana lysis. 1 10 This approach can be j ustified in terms of previous research - see De Young & H asan ( 1 998) . 1 1 1 It was thus found that, i n L i u & Tripe (2002), the set of effi c ient firms contained relat ively I in le representation from the banks focused on in this research. 1 1 2 No data are inc luded for Superbank, as that bank had been operat ing for less than a fu l l year, and was in any case very small as at its 30 September 2 0 0 3 balance date ( with total assets o f only $76 m i l l ion). 69 Table 1 : Costs and Revenues for New Zealand ban ks in 2003 Net I nterest Non-interest Non-interest Average I ncome/ I ncome/ Expense/ Assets Average Average Average ($000) Assets (%) Assets (%) Assets (%) Fu l l-service banks with branch networks ANZ Banking 2 .68 1 . 95 2 . 32 28 ,354 ,000 Group (NZ) Ltd ASB Bank Ltd 2 .40 0 .89 1 . 57 25 ,880, 1 50 Bank of New Zealand 2 .40 1 .4 1 1 .71 36 ,754, 000 National Bank of New 2 .60 0 .95 1 .78 40,593 ,000 Zealand TSB Bank Limited 3 . 1 0 0 .48 1 .63 1 , 727 ,749 Westpac Banking 2 .72 1 .42 1 .72 37, 969,800 Corporation Specialised and new banks ABN Amro 1 . 08 4 . 90 5.48 782,461 AMP Banking 1 . 1 5 1 . 1 5 2 .31 1 ,037, 534 Bank of Tokyo- 1 . 93 0 .36 1 .28 2 1 9 ,43 1 Mitsubishi C itibank 0 .5 1 0 .85 0.62 2 ,424 ,997 Commonwealth Bank 0 .25 0 .72 0.22 9 1 9 ,268 of Austral ia Deutsche Bank 3 .05 0 .63 0.27 1 3 , 1 1 2 , 500 HSBC 1 .42 0 .34 1 . 1 6 5 ,820 ,792 Kiwibank 2 . 1 7 8 . 33 1 2 .90 474,288 Kookmin Bank 0 .9 1 2 . 1 2 1 .21 2 1 8, 774 Rabobank Nederland 1 . 89 0 .25 1 . 1 5 3 ,4 1 1 , 568 Rabobank (NZ) Ltd 1 . 99 0 . 1 3 1 . 34 2 ,802 ,9 1 9 Source: KPMG (2004) Data for the 6 banks that are the primary focus of this study, figures for which are shown in the upper pm1 of Table 1 , have been obtained from their quarterly d isc losure statements, although some of the studies rely on annual financial results only (particularly where New Zealand and Australian performance is being compared, but a lso when an effort i s made to eliminate some of the statistical noise that can otherwise arise in some of the qum1erly figures) . These banks together typically account for around 85% of the New Zealand banking market. In those cases where the performance of New Zealand banks was being compared with that of Australian banks, figures in New Zealand dollars were converted to Australian dollars at the h ? h . d . . 1 1 3 average exc ange rate 10r t e peno 111 questiOn. 1 1 3 Exchange rate information has been obtained from the Reserve Bank of New Zeal and ' s web-site at www.rbnz. govt.nz, Table 8 1 . 70 The Australian banks inc luded from time to time in thi s research are the four major banks - ANZ, CBA, NAB and Westpac - and the s ix so-called regional banks, each of which has a strong network presence in one or more states - Adelaide Bank (Adelaide), Bank of Queensland (BoQ), BankWest, Bendigo Bank (Bendigo), St George Bank (St George), and Suncorp-Metway Bank (Suncorp) . Financial statements are publ ic ly available for each of these banks as they are li sted companies. 1 1 4 These 1 0 banks together have typically accounted for around 80% of the assets of the Austral ian banking system. There was some concern as to whether it was val id to include Suncorp, because of that bank' s very significant general insurance business, which can make it look rather d ifferent from its peers. As is demonstrated in Section 5 .4 below, however, its inclusion can be justified. A lthough qum1erly data has been available for New Zealand banks smce the beginning of 1 996, data for the Australian banks has been annual, and the research has therefore been restricted to using annual results for the Australian banks. It is not believed that the diversity of balance dates has led to any significant distortion of results, and where, in the case of Bank West, accounting periods were for other than 1 2 months, figures have been adjusted. 1 1 5 Data for this group of Australian banks has been for the whole bank ( i .e. including their New Zealand operations) . reflecting what the banks publ ish. 1 1 6 Although the study looks at the performance of banks over a period of 8 years from 1 996 to 2003, data for the banks i s used at current, rather than constant prices. Rates of inflation in both New Zealand and Austral i a have been low throughout the period of the study, and any differences in inflation rates between the two countries ought to be reflected in differences i n the exchange rates between the two countries? 1 14 Bank West ceased being l i sted d uring 2003 fol lowing its acq uisit ion by its parent, H BOS, which previously held only a partial stake. 1 1 5 One effect of this is that the results for Bank West described as being for 2002 and 2003 are actually for periods up to December 200 1 and 2002 respectively. 1 1 6 This is a potential d ist011ion, forced by the data, although, as noted by Tripe & Matthews (2003 ), the New Zealand business comprises only a relatively smal l part of the banks ' overall business. S ince its acquisit ion of the N BNZ, the ANZ has had a much larger proportion of its business in New Zealand, but that is outside the period covered by this research. 7 1 currencies. Moreover, because all inputs and outputs are measured as monetary amounts, and because DEA looks at the ratios between the weighted inputs and outputs, there i s no obvious distortion l ikely to arise from not converting these amounts to constant prices. It is fm1her noted that any attempt to adjust would be l ikely to open up arguments as to what price index ought to apply. Note that an advantage of not using data from earlier than 1 996 i s that the efficiency scores generated wil l not be impacted by the effects of deregulation. A number of other studies (some of which were discussed in section 3 .3 ) have found that, because of the l iberal isation of interest rates, the effects of deregulation can be hard to predict. Now that the set of banks whose efficiency i s to be reviewed in this research has been specified and justified, the next part of this chapter goes on to look at the data that are actuall y avai lable for those banks, and at i ssues that may arise with the data. 4.2 Data : what is reported The introduction of the disclosure regime in New Zealand at the begi1ming of 1 996 has been a boon for researchers, as it has forced banks to publish financial statements and a range of other disc losures on a quarterly basis. 1 1 7 In particular. it has meant that d isclosures have had to be made by all banks, not just those which were raising deposits from the public (the previous requirement), while the quarterly disclosures have meant that it is now possible to compi le a snapshot of the \\'11ich differs from that reported in Model 3 2 by the addition o f one output variable, off-balance sheet items. Efficiency scores are 1 74 Westpac showed the h i ghest average spread ratio (as per Schaffnit et al, 1 997) of 1 .036. which indicates the greatest change in effi c iency scores from the introduction of the additional output variable. 1 3 6 again expected to be higher, reflecting the reduction of degrees of freedom from the inclusion of the additional output variable. By comparing the average efficiency scores, it can see that this in fact the case, although it is noted that there is no change in the average efficiency scores for either ASB or BNZ. A review of the s lacks once again shows that, for these two banks , except for the two quarters where the BNZ was c lassed as 1 00% efficient, a shortfa l l was recorded in the production of off-balance sheet items. Off-balance sheet items were not a constraint in the determination of efficiency. Table 36 : Effic iency scores from model with a l l banks together with equity and adjusted interest expense as inputs, off-balance sheet items among outputs (CCR) ANZ ASB BNZ NBNZ TSB Westpac J un-96 0 .888802 0. 77303 0 .83358 1 0 .584 1 04 0.972281 Sep-96 0.9907 1 5 0. 752685 0 .83 1 265 0. 553722 1 1 Dec-96 1 0 . 807508 0. 95747 1 0 .606798 1 1 Mar-97 0 .9892 1 9 0 . 862529 0 .871 284 0.695979 0.75 1 99 0 .946605 Jun-97 1 0 . 847258 0 .840773 0 . 703495 0 .828043 0 .944 164 Sep-97 1 0 .735696 0 . 82078 0 .645434 0. 879845 0 .9722 14 Dec-97 0 . 764066 0 . 772272 0 .845926 0.6637 1 9 0 .91 7746 1 Mar-98 0 . 82 1472 0. 780928 0. 794004 0 .7 1 9327 1 1 Jun-98 0 . 768864 0 .7450 1 9 0 .875084 0 . 724402 1 1 Sep-98 0 .92342 0 . 73308 0.7 1 2 1 67 0 . 787552 0 .869923 1 Dec-98 0 . 854935 0 .766367 0. 796 1 03 0. 73745 0 .81 1 585 0 .968635 Mar-99 0 . 9 1 9509 0 .763207 0 . 830505 0 .851 282 0.744978 1 Jun-99 0 . 872229 0 .766268 0 .835701 0. 759505 0.82799 0 . 955504 Sep-99 1 0 . 7951 1 4 0 .725959 0.833745 0.896026 1 Dec-99 1 0 .7473 1 3 0 . 792695 0.788958 0 .932933 0 .924063 Mar-00 0. 92437 0 .802763 0 .8 1 9 1 39 0 .877907 0 .947086 0. 889982 Jun-00 0 . 962988 0 .776623 0 . 838671 0. 937983 1 0 .9493 1 3 Sep-00 0 . 763338 0 . 83 1 096 0 . 842462 0 .86 1 307 1 0 .946226 Dec-00 0 . 879578 0 .803623 0 .847208 0 .84359 1 1 Mar-0 1 0 .933292 0 .8 1 5848 0 .835285 0 .942622 0 .839 1 65 0.9865 1 3 Jun-01 0 . 997339 0 . 924 1 94 0 . 8 1 4 1 3 8 0.934409 0 .968548 1 Sep-0 1 1 0 . 832663 0 . 785388 0. 898356 0 .905954 1 Dec-01 0 . 945855 0 .86 1 323 0 .844972 0 . 7552 1 6 0. 982671 0.89 1 7 1 8 Mar-02 0 .95 1 248 0 .849467 0. 87493 0 .945 176 0 .979599 0. 946409 Jun-02 0 .942748 0 .88 1 823 0 . 908656 0 .944 1 59 1 0 .9465 1 6 Sep-02 0 .94291 1 0 .89457 0 .820492 0 .92034 1 1 1 Dec-02 0 .939946 0 .907886 0 .89 1 1 44 0 . 89279 0.9637 1 2 1 Mar-03 0 .94 1 775 0 .91 4971 1 1 0. 927423 1 Jun-03 0 . 922745 0 .932 1 74 1 1 0 .984627 1 Sep-03 0 . 995609 0 .928909 0 .83 1 96 0 .9 1 9637 1 0.9068 1 2 Dec-03 0 .937 1 68 0 .8765 1 6 0. 964449 0.975982 Average 0.927899 0 .82398 0. 844976 0 .81 0965 0. 932 148 0. 971 689 1 3 7 For the other banks, the differences in efficiency scores were not significant at the 5% level, although the d ifference was significant at the 1 0% level for Westpac. 1 75 Note that there appears to be a fair degree of similarity between the results reported in Tables 34 and 36 . As with the model reported in Table 32 , TSB shows the lowest level of scale efficiency, at 0.970, but this was not found to be significant . 1 76 Tests were undertaken for differences in efficiency between the banks. Median efficiency scores, and the significance of any difference, in terms of the p-values applying to the differences between the medians, are rep011ed in Table 3 7. Westpac i s again the most effic ient, reflecting i t s relatively greater output of off-balance sheet items. lt is fol lowed by A Z and TSB, 1 77 with ASB, B Z and NBNZ less efficient. Table 37: Difference in efficiency between banks - equ ity and adj usted interest expense as inputs , off-balance sheet items as an o utput ANZ ASB BNZ Median .94226 . 80751 .83570 ANZ . 0000** .0001 ** ASB . 1 857 BNZ NBNZ TSB * * md 1cates s1gnlficance a t t h e I % level * indicates s ign ifi cance at the 5% level 5.3.5 Discussion NBNZ TSB Westpac .83867 . 96445 .99326 .000 1 ** . 5851 .003 1 ** . 9253 .0000** .0000** . 5885 .000 1 ** .0000** .000 1 ** .0000** .0566 A number of i ssues have surfaced in the previous sub-sections' results, which are now discussed. I ssues to be considered include the existence of scale effects, the impact of changes in the general level of interest rates and the use of an adjusted interest cost variable, the causes and meaning of apparent d ifferences in efficiency between the banks, and approaches to formulation of a best model for the analysis of bank efficiency. Throughout sections 5 .2 and 5.3 there have been indications of scale effects, particularly through time, but also for TSB, which were not evident in section 5 . 1 . 1 75 Westpac showed the h ighest average spread ratio (as per Schaffni t et al , 1 997) , of 1 .03 1 , which indicates the greatest change in effic iency scores from the i ntroduct i on of the addit ional output variable. 1 76 This was tested by looking at the d i fference between the efficiency scores from the CCR and BCC models, using the M an n-Whitney test. 1 77 This rel ies on Westpac showi ng as more effi c ient than TSB at the I 0% level . 1 3 8 One conclusion i s that findings of scale effects are incl ined to be dependent on the input and output sets selected. The finding of scale effects through time is only observed where gross i nterest expense i s an input (in section 5 .2) . The inabi l ity to satisfactori ly confirm thi s result using regression analysis, and its disappearance once equity was included as an additional input variable, combine to suggest that the apparent improvement in scale efficiency through time, as banks became l arger, is a spurious effect . Where an improvement in efficiency through time was found in sub? section 5 . 3 .4, this had to be attributed to improvements in X-effic iency. It is thus considered more l ikely that the apparent improvements in scale efficiency are a consequence of the way VRS models envelope the data more tightly. The other suggested example of scale efficiency was in the case of TSB, although evidence for this was s ignificant only i n models that use gross interest expense as an input and not in mode ls that used adjusted interest expense. One is inclined to the view this apparent evidence for TSB being scale inefficient as a consequence of VRS model characteristics, as discussed by Dyson et al . (200 1 ) . The potential impact of changes in the general l evel of interest rates on efficiency scores has also been establi shed. It has been found, more particularly in the models that look at data for al l the banks together, that use of an adjusted interest cost figure as an altemative input can al leviate some of the adverse effects. 1 78 Issues that would be l ikely to impact such a variable \:VOuld include whether or not there were significant d ifferences in interest rates, the extent to which interest costs might constrain the estimat ion of efficiency scores, and the availabil ity of a basis for adjustment. The significance of any effect might be determined by looking at the relationship between raw efficiency scores and the general level of interest rates. The all-bank models reviewed in sub-section 5 . 3 .4 indicated a remarkabl e degree of consistency in banks' efficiency rankings. This consi stency in the results a l lows greater confidence in their rel iabi lity, consistent with the arguments proposed by Bauer et a] ( 1 998). For the CCR models, Westpac was most efficient, fol lowed by 1 78 This i s even though the d i fference between the two sets of efficiency scores for the models repm1ed in Tables 28 and 32 is not statist ical ly significant : the p-va lue generated by the non-parametric Mann? Whitney test is 0 . 1 7 1 7 . 1 39 TSB, ANZ, BNZ and NBNZ, with ASB least effic ient. For BCC models, TSB general ly appeared to be more efficient than Westpac. The question thus arises as to what other factors might characterise these banks' performance, and impact on their est imated efficiency scores. A significant contributor to TSB's superior performance is l i kely to be its superior generation of net interest income: its average level of net interest income for the period from the June quarter 1 996 to the December quarter 2003 was 3 .25% of average total assets. The next best performer was Westpac : over the period from the September quarter 1 996 to the December quarter 2003 its net interest income relative to average total assets averaged 2 .63%, whereas the average for the banks which dominate retail business, as a whole, was 2.46%. This contrasts with the relatively poor perfonnance of BNZ and NBNZ in particular, at 2 . 35% and 2 .38% respectively of average total assets for the period from the September quarter 1 996 to the December quarter 2003 . This is consistent with suggestions that the approach in this research is more directed at revenue efficiency, rather than cost efficiency (Chu & Lim, 1 998) . Westpac ' s efficiency score was also enhanced by its relatively low level of equity, particularly during the earl ier part of the period, and this accounts for the difference between the trend evident in the results reported in section 5 . 1 and those reported in sub-section 5 .3 .4 . l t could be argued that the apparent improvement in efficiency observed in Tables 1 3 and 1 4 was at least in part a reflection of a fai lure to take account of equity as an input. The impact of including equity as an input will be further explored in section 5 .4 below. Other effects are evident if banks' gross incomes are compared over the period of the study. For the period from the September quarter 1 996 to the December quarter 2003 , average gross income (relative to average total assets) was 4 .28% for ANZ and 3 .97% for Westpac, compared to 3 .4 1 % for NBNZ and 3 .52% for ASB. Review of these revenue figures helps with understanding their significance in determining bank 1 40 efficiency in a multivariate sense, subject, of course, to the inputs and outputs used in efficiency models . 1 79 It i s also possible to compare the ranking of efficiency scores with a ranking by cost to income ratio. For the period from June 1 996 to December 2003 , the highest average cost to income ratios were for ASB at 56 .5% and ANZ at 56.0%, whereas the lowest were for TSB at 52 .0% and BNZ at 53 .6%. Using the ratio of cost to assets, the effects are just as striking : for the period from September 1 996 to December 2003 , the lowest ratio was for NBNZ at 1 . 84%, whereas the highest ratios were for ANZ at 2 .40% and Westpac at 2 . 1 5%. Thi s evidence tends to support the proposition that simple cost ratios provide a less than complete picture of bank efficiency. The final point to be discussed at this stage is the potential to i dentify a best model for analysis of bank efficiency. The range of results obtained so far, which have at times been inconsistent with each other, suggest that it may not always be easy to identify a single best set of inputs and outputs, although it can be said that inclusion of shareholders ' equity seems to make a positive contribution to particular models ' abi l ity to discriminate. This i s a reasonable finding. As other researchers have found (e.g. Berger & Mester, 1 997, 2003) , equity is an important resource used by banks, both as a source of funding, and as a foundation for ri sk-taking, which wil l al lov., them to take up opportunities to earn revenue. It is l ikely that bank management do not look at equity in this way in reviewing the performance, including their perceptions of efficiency, for the banks that they manage. They are incl ined to regard equity as expensive, and as a cost imposed on them by regulation. Because of this, they are frequently trying to reduce costs by I ssuance of debt-type instruments with only a bare sufficiency of equity-type characteristics to achieve regulatory capital compliance. What thi s research suggests is that efficient use of equity can make a rather more positive contribution to bank performance. 1 79 The importance of revenue is h ighl ighted in the way in some of the models, such as that reported in Table 22, the ANZ showed re latively higher efficiency prior to the change i n account ing pol icy that came into effect after September 1 997, and which reduced non-interest revenues. 1 4 1 It has also been found that, for New Zealand banks, simply adding off-balance sheet i tems to the output set does not necessari l y have an impact, although a more meaningful distinction may be obtained using a different measure for off-balance sheet exposures. It is not c lear whether off-balance sheet items are a less important part of New Zealand banks' outputs than they are for banks in other countries (e .g . Siems & Clark, 1 997; Rogers, 1 998) or whether alternative measurement proxies might show them as having greater impact. 1 80 Other things being equal, and in terms of the argument put forward by Berger & Humphrey ( 1 992b ), use of a proxy for uti l i sation of funds that considers the cost of those funds should be superior to one which does not, such as total interest-bearing l iabi l ities. If there are reasons for interest costs to be impacted by other factors, such as operation in different countries at different t imes, however, such an approach may not be possible. Such a situation is examined in the next section ofthis dissertation. 5.4 The cross-country study This analysis i s in four parts. The first part looks at whether or not it i s val id to apply a common frontier to the analysis of ew Zealand and Australian banks. The second part considers a model which has equity as an input. and then in the third part a model is used which does not have that input. ln the fourth part. the results from the two separate models are compared. 5.4.1 Does a common frontier app ly? The high degree of simi larity between the New Zealand banks and the Australian major banks (who are in most cases their owners) would seem to be supportive of arguments that there was a measure of integration betv,reen the Austral ian and New Zealand banking systems. It could be argued similarly that as the Austral ian regional banks operate in the same market, and also have branch networks, that a l l three groups of banks should be sufficiently similar in their operations to be studied in a 1 80 Some prior research has used non-interest income as a proxy for off-balance sheet business, whereas other research has used total off-balance sheet commitments or the risk-we ighted equivalent (Leong et a l . , 2003 ). The New Zealand case is perhaps unusual in having a wider range of publ i shed data to choose between. 1 42 single efficiency model relative to a common frontier. Before the analysis of banks' efficiency rel ative to each other proceeds, however, one should explore the proposition of whether a common frontier applies. Very l ittle in the way of precedent can be found for approaches to try and resolve this question. 1 8 1 One approach i dent ified is a teclmique of mapping individual banks, one at a time, onto a frontier compris ing a set of banks for which i t can reasonably be assumed that a common frontier might apply. 1 82 Such relevant groups of banks might include the set of six New Zealand banks studied in sections 5 . 1 to 5 . 3 above, or a set of the major Austral ian banks. The effect of introducing a new bank can be examined in tem1s of changing the efficiency scores of banks already in the sample. If the new bank in the sample caused a change in efficiency scores for a particular bank or banks, it could be inferred that the newly introduced bank was on the same portion of the frontier as the bank(s) whose efficiency i s changing. The analysis starts with the four Austral ian maJor banks, and then exammes the impact of adding Austra l ian regional banks, one at time. Amongst other things, this al lows an assessment of whether or not Suncorp-Metway properly belongs in the sample of Austral ian banks, an i ssue identified in section 4 . 1 . The New Zealand banks are then added, one at a time, to the set of all Austral ian banks, to see whether a common frontier can be applied to New Zealand and Austral ian banks together. This i nitial analysis i s undertaken using Model 1 (which has equity as an input). Table 38 reports the sum of changes in efficiency for each of the major Austral ian banks, across the whole 8-year period of the study, as a result of the introduction of each of the Austral ian regional banks, one at a time and once only, to the data set of the four major Australian banks. The total impact of changes from each bank' s i ntroduction i s in the second t o bottom row o f the table, whi le the right-most column of the table totals the changes for each of the major banks, and suggests which of these banks are most c losely related to the regional banks in terms of position on the efficient frontier. The bottom row of the table shows the mean efficiency score for the 1 8 1 The approach by Cook et al ( 1 998) of looking at h ierarch ies and groups i s not considered app l icable to this problem. 1 82 This approach fol lows a suggest ion made by Alexander Karmann. A s imi lar technique has been used by Ste inmann et al (2004). 1 43 regional banks, in the model that contai ns only the majors and the specific regional banlc Table 38: Impact of inc lusion of Austra l ian reg iona l banks in efficiency comparisons against Austral ian major banks Adelaide Bank of Bendigo Bank St Suncorp- Total Bank Queensland Bank West George Metway ANZ 0 0. 0024 0 0 0 .0039 0 . 0532 0 .0595 CBA 0 0 0 0 0 0 . 1 508 0 . 1 508 NAB 0 0 0 0 0 0 .2637 0 .2637 Westpac 0 .01 1 0 0 .01 1 5 0 .0053 0.02 1 6 0 .01 72 0 .0733 0 . 1 399 Total 0 .01 1 0 0 .01 39 0 .0053 0 .02 1 6 0 .02 1 1 0. 5409 Mean 0.9436 0.9601 0 .8738 0 .9396 0 .8649 0 .99 1 9 efficiency This analysis demonstrates that the regional banks have relatively l ittle impact on the efficiency scores of the majors, with the exception of Suncorp-Metway. There are two possible interpretations for this. One i s that the regional banks are general ly in a d ifferent part of the efficient frontier to the maj ors. The alternative interpretation i s that the regional banks might be significantly less efficient than the majors, being consi stently off the frontier, and thus not having any impact on the position of the efficient frontier. To explore the second of those possible interpretations, the mean efficiencies for each of the regional banks were recorded . The mean efficiency for the four major banks together is 0 .9538 . If the average effic iency scores for the regional banks were c lose to or higher than that average, it could confidently be asse11ed that they were not significantly less efficient than the majors, and the reason for a lack of impact on the efficiency scores for the majors was that the regionals operated in a different pm1 of the efficient frontier. It can be concluded that Adelaide Bank, Bank of Queensland and BankWest are on a different part of the efficient frontier to where the majors are located. For Bendigo Bank and St George the situation is less c lear, and one thus cannot be sure of the extent to which they are comparable with the other Austral ian banks . To investigate the position of Bendigo and St George fm1her, a DEA study was undertaken of j ust the Austral ian regional banks . Although these two banks are less 1 44 efficient on average, they make a reasonable contribution to the reference sets of the other regional banks and it is considered reasonable to study them together. By contrast, the inclusion of Suncorp-Metway impacts rather more significantly on the efficiency scores of the major banks, and more pm1icularly CBA and NAB (which would be likely to be the majors with the most significant insurance business) . This suggests that there is a significant degree of comparabil ity between Suncorp-Metway and the major banks. 1 83 It is also noted that whi le Suncorp-Metway impacts less strongly on the efficiency of ANZ and Westpac, the efficiency scores of these two banks are more ?trongly impacted by the other regional banks, suggesting that ANZ and Westpac are s l ightly closer, relative to the frontier, to these other regional banks. Table 39 shows the sum of changes in efficiency for each Australian bank, across the 8-year period of the study, as a result of the introduction of each of the New Zealand banks, one at a time and once only, to the data set of all I 0 Australian banks. Table 39: Impact of inc lus ion of New Zealand banks in efficiency comparisons against Austra l ian banks ANZ ASB Bank of NBNZ TSB Westpac Total ( NZ) Bank NZ Bank (NZ) Adelaide Bank 0 .2487 0 .4607 0 . 1 546 0. 1 028 0. 1 642 0 . 3286 1 .4596 Bank of Queensland 0 . 3877 0 .0402 0.3697 0. 0003 0 .01 62 0. 73 1 7 1 . 5458 BankWest 0 . 5 149 0 .3448 0 .3842 0 . 0303 0 .01 1 0 0 .6826 1 . 9678 Bendigo Bank 0 .2936 0 .01 1 9 0 .3 1 92 0 . 0009 0 .0001 0 . 5036 1 . 1 293 St George 0 .0405 0.0020 0 . 0 147 0 . 0501 0 . 1 954 0 . 1 084 0 .4 1 1 1 Suncorp 0 . 0772 0 .03 1 4 0 .0471 0 0 .0404 0 . 1 781 0. 3742 ANZ (G lobal ) 0 . 1 438 0 .0001 0 . 0534 0 .0024 0. 0048 0 . 1 525 0 .3570 CBA 0 . 1 3 1 4 0 0 . 0762 0 0 0 . 1 242 0 . 33 1 8 NAB 0 . 0270 0 0 .0046 0 0 0 .0 1 32 0 .0448 Westpac (G lobal ) 0 . 2245 0 .0 1 1 8 0 .0941 0 .00 1 5 0 . 0034 0 . 1 63 1 0.4984 Total 2 . 0893 0 . 9029 1 . 5 1 78 0 . 1 883 0.4355 2 . 9860 The total impact of changes from each bank's introduction is shown in the row at the bonom of the table, whi le the right-most column of the table sums the changes for 1 83 On the other hand, noting the h igh average efficiency score for Suncorp-Metway, there may be a quest ion as to whether it properly belongs in the same data set. l t seem reasonable to argue that it does, however: it shows as less that fu l ly efficient in 3 out of 8 cases, and although its 2000 figures provide the largest single contribution to the references sets of the ineffic ient banks, cases for other banks also contribute signi ficantly. 1 4 5 each of the Australian banks, and suggests which of these banks are most c losely related to the New Zealand banks in terms of position on the efficient frontier. The results show the strongest effect for the Austral ian regional banks, Adelaide Bank, Bank of Queensland, BankWest and Bendigo in particular. It appears reasonable to assume that these banks are the ones most comparable to the New Zealand banks. It is al so interesting to note that the NAB is virtually unaffected by the inclusion of the ew Zealand banks, and that the other major banks are affected less, suggesting that the Australian major banks may be on a d ifferent part of the frontier to the New Zealand banks. To further explore the i ssue of comparabi l i ty and appropriateness of a common frontier further, the reference sets for ineffic ient banks in both groups were examined. This was expected to show the extent to which New Zealand banks have Austral ian majors in their reference sets, and therefore the extent to which their perfom1ance i s compared to the Austral ian banks, and vice versa. Table 40 reports the numbers of times efficient banks from each country appear in the references sets of the inefficient New Zealand and Austral i an major banks. 1 84 Table 40: Frequency w ith which types of banks appear in reference sets Austral ian major banks New Zealand banks Austra l ian banks 73 30 New Zea land banks 3 1 95 Although the other country ? s banks appear a considerable number of times in individual banks' reference sets, one cannot accept a hypothesis that there are no country-speci fic concentrations in the reference sets cl statistic i s 48 .9, which has p? value 0.000). Though the proportion of New Zealand banks in the reference sets of NBNZ and TSB, neither of which were owned by a major Australian banking group i s greater than for the New Zealand banks a s a whole, when these banks are removed from the analysis, the x2 stat istic remains highly significant. 184 Note that each ineffic ient bank can have up to 4 (the total number of inputs and outputs less one) banks in its reference set, which means that the number of reference set references may be substantially larger than the number of banks i n the study. The ineffic ient New Zealand banks have more other banks in their reference sets in total as there are 6 banks, with 33 bank year observat ions showing as inefficient compared with just 4 Austral ian majors, with 29 bank year observations showing as inefficient). 1 46 The overall conclusion i s that the three groups of banks are on different relative points on the efficient frontier, with the Australian majors and Suncorp-Metway general ly in one group, and the New Zealand banks and the Australian regionals in another part of the frontier. This does not mean that it i s inappropriate to compare the banks' efficiencies relative to each other, but rather that care may need to be exerci sed in interpreting any results. 5.4.2 A model with equ ity as an i nput This section looks at the actual efficiency scores generated by the first model, and seeks to interpret them. Efficiency scores for each bank for each year are reported in Table 4 1 , along with the averages for each bank for each year. Table 41 : Cross-country study, model 1 resu lts ( i .e . with cap ital as an input) - constant retu rns to scale 1 996 1 997 1 998 1 999 2000 2001 2002 2003 Average Ade la ide 0.846 0 .835 0 . 836 0.750 0 .692 0 .777 0 .777 0. 884 0 .800 BoQ 0.985 0.873 0 . 8 14 0.820 0 .81 1 0 .758 0 .723 0. 726 0 . 8 14 Ban kW est 0 .925 0. 902 0. 848 0 .791 0 .780 0 . 794 0 .764 0 .721 0 . 8 1 6 Bendigo 1 . 000 0. 8 1 3 0 .761 0 .740 0 .766 0 .759 0 . 754 0 .759 0. 794 St George 0 .950 0 .860 0 .787 0. 782 0 .773 0 .792 0.790 0 .875 0 .826 Suncorp 0.872 0 .868 0 . 935 0 .929 1 . 000 0 .965 1 .000 1 . 000 0 .946 ANZ 0 .969 0 .942 0 . 888 0 .922 0 .949 0 .944 1 . 000 0 .970 0 .948 CBA 1 . 000 0 .935 0 . 9 1 5 0.925 0 .874 0 .872 0 .921 0 . 824 0.908 NAB 1 .000 0 .965 0 .955 0 .940 0. 950 0 .840 0 .858 0 .868 0. 922 Westpac 0.909 0 .898 0 .907 0 .876 0. 906 0 .926 0.81 5 0 .877 0 .889 ANZ (NZ) 1 . 000 0 .932 0 .84 1 0 .868 0.866 0 .963 1 . 000 0 .998 0 .933 ASB 0. 821 0 .830 0 .829 0.795 0.806 0 .895 0 .928 1 . 000 0 .863 BNZ 0 .931 0 .969 0 .870 0. 8 1 5 0. 834 0 .857 0.956 0.960 0 .899 NBNZ 0 .691 0 .78 1 0 .775 0 .868 0 .949 0 .94 1 1 .000 0 .936 0 .868 TSB 1 . 000 1 . 000 0 . 976 0.937 0 .939 1 . 000 1 . 000 1 . 000 0 .982 Westpac (NZ) 1 000 1 . 000 1 . 000 1 . 000 0 .94 1 0 .883 0 .905 1 .000 0 .966 Average 0 .931 0 .900 0 .871 0 .860 0 .865 0 .873 0 .887 0 .900 A number of points can be observed in these results . In the first place, when one examines the average figures, there is no obvious trend through time. There is a suggestion of a decrease in efficiency from 1 996 to 1 998 , although these results are perhaps a consequence of trends an1ongst the Austral ian banks. 1 85 One also notes the 1 85 l t a lso raises the quest ion a s t o whether there was something unusual about the results for some of the Australian banks in 1 996. I t i s interesting to note that Sathye (2002), in looking at Austra l ian banks for the period 1 99 5 to 1 999, generally found more posit ive results for productivity change. He used 1 4 7 significant improvement in efficiency through time for the NBNZ (and to a lesser extent ASB), consistent with results reported in section 5 . 1 and sub-section 5 .3 .4 . 1 86 I t is also noted that TSB and Westpac (NZ) show as the most efficient among the New Zealand banks studied, consistent with the results obtained in section 5 . 3 .4 . As TSB was also the smal lest of the banks in the study, this again raises questions as to whether it i s likely that there were any benefits in operating at increased scale . The data used to generate the results repm1ed in Table 40 were therefore run through a BCC model, with the results reported in Table 42 . Table 42 : Cross-country study, model 1 resu lts ( i .e. with capital as an input) - variable returns to scale 1 996 1 997 1 998 1 999 2000 2001 2002 2003 Average Adelaide 0 .848 0 .84 1 0 .850 0 .750 0 .694 0 .790 0 .778 0 . 906 0 .807 BoO 0 .991 0 .873 0 . 8 1 4 0 .823 0 .837 0 .781 0 .735 0 .727 0 .823 BankWest 0.933 0. 9 1 7 0 . 862 0 . 796 0 .785 0 .801 0 .768 0 .72 1 0 .823 Bendigo 1 .000 0 .829 0 . 764 0 . 743 0 .767 0 .76 1 0 .755 0 .763 0 .798 St George 1 .000 0 .923 0 . 829 0 .800 0.779 0 .794 0.790 0 .876 0 .849 Suncorp 0 .901 0. 908 0 .936 0 .931 1 . 000 0.965 1 .000 1 .000 0 .955 ANZ 1 . 000 1 . 000 0 .946 0 . 955 0 .977 0.965 1 . 000 1 . 000 0 .980 CBA 1 . 000 0 .959 0 .972 1 . 000 0 .875 0 .922 0 . 967 0 .847 0 .943 NAB 1 .000 0 .995 1 . 000 0 . 997 1 . 000 0 .999 1 . 000 1 000 0 .999 Westpac 0 . 9 1 2 0. 905 0 .9 1 9 0 .901 0 .941 0 .977 0 .921 0 .977 0 .931 ANZ (NZ) 1 . 000 0 . 933 0 .842 0 .869 0 .867 0 .967 1 . 000 1 . 000 0 . 935 ASB 0.822 0 . 830 0 .833 0 .800 0 .8 1 0 0 .899 0 .928 1 . 000 0 .865 BNZ 0 .931 0 . 972 0 .872 0 . 8 1 5 0 .836 0 . 857 0 . 956 0 .960 0 .900 NBNZ 0.694 0 .784 0 .784 0 .878 0 .953 0 .942 1 . 000 0 .998 0 .879 TSB 1 .000 1 . 000 0 .976 0 . 983 0.992 1 . 000 1 . 000 1 .000 0 . 994 Westeac (NZ) 1 . 000 1 . 000 1 . 000 1 . 000 0 .942 0 .885 0 . 906 1 .000 0 . 967 Average 0.939 0 .9 1 7 0 .887 0 .878 0 .878 0 . 894 0 . 906 0 .923 This has in turn allowed a test for scale efficiency, with the results reported in Table 43 . Contrary to the expectations of bank management, scale efficiency does not appear to be important overal l , with average scale effic iency across all banks of 0.98 1 9 . Returns to scale status for each bank are reported in Table 44. 1 8 7 interest expense as an input, however, and some of the effect he found may be a consequence of a reduction in the general level of Austra l ian interest rates over that period. 1 86 The improvement in efficiency for the NBNZ in this case i s less than that reported in Table 27, which reflects the use of total deposit l iabi l ities as an input rather than interest cost. Th i s provides further support for the argument that the N BNZ's performance during 1 996 and 1 997 in pa11icular, was adversely affected by interest costs. 187 The figures reported are for the total of efficient and projected returns to scale est imates, noting that returns to scale status, when us ing DEA, is strictly defined only for firms on the effic ient frontier. 1 4 8 There is no evidence for increasing returns to scale, 1 88 and there is a question as to whether the major Australian banks, pm1icularly NAB and Westpac, with their relatively lower levels of scale efficiency, particularly towards the end of the period, may be suffering from decreasing returns to scale. 1 89 Table 43 : Cross-country study, model 1 resu lts ( i .e . with capital as an input) - measures of scale efficiency Adelaide BoO BankWest Bendigo St George Suncorp ANZ CBA NAB Westpac ANZ (NZ) ASB BNZ NBNZ TSB Westpac (NZ) Average 1 996 0 .997 0 .995 0 .992 1 . 000 0 .950 0 .968 0 .969 1 . 000 1 . 000 0 .997 1 . 000 0 .999 1 . 000 0 .995 1 . 000 1 . 000 0 .991 1 997 0.993 1 . 000 0. 983 0.98 1 0 .931 0.956 0.942 0.975 0.970 0.993 0. 999 1 . 000 0.998 0 996 1 . 000 1 000 0. 982 1 998 0 . 984 1 . 000 0 . 984 0 .996 0 .949 0 .998 0.939 0 .94 1 0 .955 0. 987 0. 998 0 .996 0. 998 0 .988 1 . 000 1 . 000 0 . 982 1 999 2000 200 1 2002 2003 1 . 000 0. 997 0.983 0. 999 0 . 976 0 .997 0. 969 0 .970 0. 984 0 .998 0 .994 0. 994 0.99 1 0.995 1 . 000 0 .997 0 .999 0 .998 0. 999 0 .995 0 .978 0 .993 0 .997 1 .000 1 . 000 0 .997 1 . 000 1 .000 1 .000 1 . 000 0 . 965 0 . 972 0.978 1 .000 0 .970 0 .925 0 .998 0.945 0.953 0 .973 0 .943 0 .950 0 .84 1 0. 858 0 .868 0 .972 0 .963 0.948 0. 885 0 .897 0 .999 0.999 0 .997 1 . 000 0 .998 0 .993 0. 996 0 .995 1 . 000 1 .000 1 . 000 0. 998 1 . 000 1 . 000 1 . 000 0 .988 0.995 1 . 000 1 000 0 .938 0 .953 0 . 947 1 . 000 1 .000 1 . 000 1 . 000 0 .999 0 .998 0.999 1 . 000 0 .98 1 0. 986 0 .978 0.979 0 . 976 There are indications of some differences in efficiency between the banks. Average 0 .991 0 .989 0. 992 0. 996 0 .975 0 .990 0 .967 0 .964 0 .923 0 .955 0 .999 0 .997 0 .999 0 .987 0 . 988 1 . 000 Table 44: Cross-country study, scale effects for model 1 (with equity as an input), as reported by the BCC model I ncreasing returns to Constant returns scale sca l e Adelaide 6 BoO 8 BankWest 2 Bendigo 8 St George 3 Suncorp 6 ANZ 1 CBA 1 NAB 1 Westpac 1 ANZ (NZ) 7 ASB 4 BNZ 6 NBNZ 5 TSB 8 Westpac (NZ) 7 Tota l 0 74 1 88 This is consistent with the resu lts found in sub-section 4 . 1 .2 . to Decreasing returns scale 2 6 5 2 7 7 7 7 1 4 2 3 54 1 89 With the very smal l sample s ize. the M ann-Whitney test fai l s to identify a significant d ifference between the effic iency scores from the CCR and BCC models, at the 5% level . 1 49 to In this case it is considered meaningful to test for differences in efficiency between the groups of banks. Median efficiency scores for the groups, across the whole time period, are rep011ed in Table 45. Because the d istribution of efficiency scores i s censored at one, and because the distribution i s not normal , testing for the significance of any apparent differences between the groups has to be undertaken using the non? parametric Mann-Whitney test (Cooper et al, 2000; Casu & Molyneux, 2003 ) . In the first place, a s ignificant difference i s found between the efficiency of the New Zealand banks and the Austral ian banks as a whole, with a p-value of 0 .0006. If the New Zealand banks are compared with the Austral ian major banks only, the efficiency differences are not significant. There i s, however, a significant difference between the efficiency scores for the Austral ian majors relative to the Austral ian regional banks, with a p -value of 0.0000. Table 45: Cross-country study, model 1 resu lts - summary All Austral ian banks Austral ian reg ional banks Austral ian major banks New Zea land banks Med ian efficiency scores 0 .875 0 8 1 2 0 .922 0 .938 5.4.3 Model without equ ity as an i nput The actual efficiency scores generated by the second model can now be examined, and an effort made to interpret these. Efficiency scores for each bank for each year are reported in Table 46, along with the averages for each bank for each year. The efficiency scores are lower overal l than those rep011ed in Table 4 1 , reflecting the reduced number of inputs and the consequent increase in degrees of freedom. Although there is no obvious trend through time in the average efficiency scores, the New Zealand banks as a whole do show an improvement, consistent with the results rep011ed in Section 5 . 1 (and Figure 5 in particular) . 1 90 190 Note, however, that individual effic iency scores w i l l d iffer, even though the input and outputs sets are the same, because of the en larged data set aga inst wh ich the re lative efficiencies are being measured. 1 50 Table 46: Cross-country study, model 2 results ( i .e . without capital as an input) ?-- ---- - ------ --- ------ - -- - -- --- ---- ?----- -? 1 996 1 997 1 998 1 999 2000 2001 2002 2003 Average Adelaide 0 .81 3 0 .805 0 . 802 0.7 1 3 0 .657 0.707 0 .7 1 1 0. 752 0 .745 BoO 0 .963 0.828 0 . 774 0 . 784 0 .769 0. 7 1 5 0 .697 0. 7 1 2 0. 780 BankWest 0 . 883 0 .858 0 . 809 0. 754 0 .747 0. 746 0 .726 0.697 0.777 Bendigo 0 . 892 0.745 0 . 723 0. 709 0 .742 0 .735 0 . 734 0 .740 0.753 St George 0 . 930 0 .860 0 .787 0. 782 0 .773 0 .789 0 . 788 0 .875 0.823 Suncorp 0 .822 0 .831 0 . 899 0.886 1 . 000 0 .965 0 .994 1 . 000 0.925 ANZ 0 .870 0. 867 0 .87 1 0.92 1 0.948 0 .943 1 .000 0 .970 0 .924 CBA 1 .000 0 . 927 0 . 894 0.906 0.874 0. 872 0 .92 1 0 .824 0. 902 NAB 1 .000 0. 964 0 . 954 0 .940 0 .950 0 .840 0 . 858 0. 868 0.922 Westpac 0 .909 0. 894 0 . 900 0 .874 0 .905 0 .923 0 .800 0.874 0.885 ANZ ( NZ) 0 .872 0. 809 0 .8 1 0 0 .846 0 . 852 0 .952 0 . 993 0 . 974 0. 888 ASB 0.784 0 .788 0 . 787 0 .749 0.753 0 . 8 1 1 0 . 852 0 . 9 1 7 0.805 BNZ 0.881 0 922 0 . 828 0 .784 0 .798 0 .823 0 . 934 0 .947 0.865 NBNZ 0.672 0 .755 0 . 746 0.835 0 .930 0 .932 1 .000 0 . 9 1 7 0 .848 TSB 1 . 000 0 .999 0 . 970 0 .931 0 .934 1 .000 1 .000 1 . 000 0. 979 West12ac {NZ) 0 .722 0 . 804 0 .896 0. 904 0.899 0 .877 0 . 905 1 .000 0. 876 Average 0 .876 0.853 0 .84 1 0. 832 0.846 0 .852 0 . 870 0. 879 The pattern of individual bank efficiency scores i s now a l ittle different. Median scores for the groups of banks, for Model 2, across the whole time period, are reported in Table 47. 1 9 1 Table 47 : Cross-country study, model 2 resu lts - summary Al l Austra l ian banks Austra l ian reg ional banks Austral ian major banks New Zealand banks Median efficiency scores 0 . 859 0 783 0 . 905 0.889 Note that there now appears to be a greater degree of d ifference in relative efficiency between the New Zealand and Australian major banks. When the Mann-Whitney test i s used to look for differences in efficiency between the groups, it is found that the New Zealand banks as a whole are more efficient than the Australian ones (with p? value 0.044 1 ) . The difference in efficiency between the Austral ian majors and the Austral ian regionals remains highly significant (with p-value 0.0000). For the New Zealand banks and the Austral ian majors, the difference is not significant. 1 9 1 The lower level s of average effi c iency scores are l ikely to be a result of the reduction in the number of inputs and outputs, from 5 to 4, compared with model I . 1 5 1 5.4.4 What do the resu lts from these two models mean? The only difference between the models that have generated the two previous sets of results ( in sub-sections 5 .4 .2 and 5 .4 . 3 ) has been one input - equity. It is therefore reasonabl e to compare the results from the two models using the technique described by Schaffnit et al ( 1 997) as the spread rat io . This is calculated by dividing the results obtained in Table 4 1 by those obtained in Table 46, which did not use equity as an input. Values for the spread ratio will be greater than or equal to 1 : the larger the ratio, the more impact the inclusion of equity capital has on the efficiency score. Other things being equal banks that are more strongly capitali sed wi l l show a lower value for their spread ratio. 1 92 The effect of l ower capital levels is most evident for Westpac (New Zealand) during the earlier part of the period, when it was making the most of its branch status and holding minimal 1eve1s of equity in New Zealand. Results for the spread ratio, for each bank, for each year, are reported in Table 48 . Table 48 : Spread ratios to show the impact of use of equ ity capita l as an input 1 996 1 997 1 998 1 999 2000 2001 2002 2003 Average Adela ide 1 .040 1 .038 1 .042 1 . 0 5 1 1 . 053 1 099 1 .093 1 . 1 77 1 .074 BoO 1 .024 1 .055 1 .051 1 .046 1 . 055 1 . 06 1 1 .037 1 . 0 1 9 1 . 043 BankWest 1 .048 1 .05 1 1 .049 1 .050 1 .045 1 . 064 1 . 054 1 . 035 1 . 049 Bendigo 1 . 1 2 1 1 .092 1 . 053 1 .044 1 .032 1 . 033 1 . 027 1 . 026 1 . 053 St George 1 . 02 1 1 .000 1 . 000 1 .000 1 000 1 . 003 1 . 002 1 . 001 1 . 003 Suncorp 1 . 061 1 . 045 1 . 039 1 . 049 1 .000 1 . 000 1 . 006 1 . 000 1 . 025 ANZ 1 . 1 1 4 1 . 086 1 . 020 1 . 002 1 .001 1 . 001 1 . 000 1 . 000 1 .028 CBA 1 . 000 1 .009 1 . 023 1 . 021 1 . 000 1 . 000 1 . 000 1 . 000 1 . 007 NAB 1 . 000 1 . 001 1 . 001 1 . 000 1 .000 1 . 000 1 . 000 1 . 000 1 .000 Westpac 1 . 000 1 .004 1 . 008 1 . 002 1 .002 1 . 003 1 0 1 8 1 . 003 1 .005 ANZ (NZ) 1 . 1 47 1 . 1 52 1 . 039 1 . 027 1 .0 1 6 1 . 0 1 2 1 . 007 1 . 025 1 . 053 ASB 1 . 048 1 .052 1 . 053 1 . 06 1 1 .070 1 . 1 04 1 .090 1 . 090 1 . 071 BNZ 1 . 057 1 .05 1 1 . 051 1 . 040 1 045 1 . 042 1 .024 1 . 0 1 3 1 . 040 NBNZ 1 . 028 1 .035 1 . 038 1 . 039 1 .02 1 1 . 0 1 0 1 .000 1 . 02 1 1 . 024 TSB 1 . 000 1 .001 1 . 006 1 . 007 1 .006 1 . 000 1 .000 1 . 000 1 . 002 Westeac (NZ) 1 . 384 1 .243 1 . 1 1 6 1 . 1 07 1 . 047 1 . 007 1 .000 1 .000 1 . 1 1 3 Average 1 . 068 1 . 057 1 . 037 1 . 034 1 .025 1027 1 022 1 . 026 These results show that the Austral ian major banks are more highly capitalised than New Zealand banks, and certainly more highly capital i sed than those New Zealand banks owned by the Austral ian majors. The question then arises as to whether the New Zealand banks are holding less capital than they might otherwise need were they 1 92 Note that this is a straight capital i sation rat io, not adjusted for risk, or any other factors, such as s ignificant holdings of goodwi l l (which has forced St George, for example, to hold a relatively h igher level of equity). 1 52 not owned by their Australian parents, and thus able to enjoy the benefits of their parent bank reputation and credit rating. This could also provide an explanation for the differences in returns on equity highl ighted in F igure 4 ( in Chapter 2) . This explanation i s best examined using model s l ike models 1 and 2 , but with the data sets comprising only the Australian major banks and their New Zealand operations. Once the data sets are modified, the M ann-Whitney tests can be run, comparing only the four New Zealand banks owned by the Austral ian majors with their parent banks. Median scores are reported in Table 49. Table 49 : Median efficiency scores for Austra l ian majors and their New Zealand operations Austra l ian major banks New Zealand banks owned by Austra l ian majors Med ian score from Model 1 . 9253 . 9296 Median score from Model 2 . 9080 .8856 It can be seen that there is no significant difference between the efficiency scores for banks in the two countries in Model 1 , where al lowance i s made for use of equity capital . 1 93 By contrast, in Model 2, where the effect of capital is ignored, effic iency scores for the New Zealand banks are lower, with the difference significant at the 5% level (p-value i s 0.0 1 77) . It can be concluded that the New Zealand operations of Austral ian banks are gaining the benefit of the capital level s held by their Australian parent banks. This and other issues are explored in the next and final chapter of thi s d issertation. 193 The p-value applying to the M ann-Whitney test i s 0 .930 I . 1 5 3 6 . Summary and conclusion 6 .1 A review of the research The focus o f the study was on financial institution efficiency, looking at New Zealand banks. Substantial amounts of research have been published on financial institution efficiency in recent years, and a major part of this d isser1at ion entailed reviewing that research and identifying a range of issues that would have to be dealt with as part of thi s . Pm1icular features of the previous research which required more serious attention included the problems of working with smal l cross-sectional samples and an attempt to solve thi s by analysis of panel data as single data sets, the d ifficulties in ident ifying and selecting suitable input and output sets, and difficulties in undertaking cross? border studies. Some of the approaches fol lowed in this research, such as the prevalent use of panel data, differ from those commonly fol lowed in previous research, but it i s bel ieved that they are appropriately j ustified in terms of previous research. Thi s dissertation began with a review of the New Zealand banking system prior to the deregulation of the 1 980s. and went on to out l ine some of the events that have occurred subsequently in turning the New Zealand financial system from what it was and how it performed in the early 1 980s to \Vhat it has become in the 2 1 51 century. Over that time there has been significant change to the participants in the financial system, to the activities that they undertake, and to the regulatory structure within which they operate. One of the major recent changes to the regulatory structure has entai led adoption of a regime based on public d isc losure of information, and i t is the information disc losed under this regime that has provided the data to underpin thi s research. This i s the first dissertation to have been able to make use o f that data set, but as that data set increases in size, it is providing scope for further research oppor1unities. In looking at the New Zealand banking system, however, this dissertation has not provided a detailed study of every bank that has operated in New Zealand over the primary period at which the study has been directed ( 1 996-2003) . The research has 1 54 foc used on a core group of six banks: ANZ, ASB, BNZ, NBNZ, TSB and Westpac. These banks were in business continuously throughout the period of the study, and they were all names which were generally fami liar to the majority of New Zealand' s population as banks with which they could do business. They a l l conducted a reasonably broad range of business, through branch networks, even though TSB' s physical network wa?s l imited t o Taranaki (a provincial area with a population under 1 00,000), and the range of business it entered into was less extensive than that of the other five banks. They thus constituted a group whose performance was expected to be able to be validly compared. The period of the study was also general ly appropriate for this group of banks: it began with Westpac ' s acquisition of Trust Bank New Zealand, and ended with the ANZ's acquisition of the NBNZ, which cemented the Australian dominance of the New Zealand banking system. The research itself comprised a range of research questions. I t first looked at the appropriateness of the multivariate approach to the measurement of bank efficiency, comparing and contrasting the Malmquist Index and DEA analysis of panel data, and explaining the difference between the results from those approaches and use of the more common cost to income ratio for measurement of bank efficiency. It then went o n to look at a quite specific i ssue which appl ies for the analysis of bank efficiency in d ifferent time periods or in different locations: what happens if interest cost is used as an input, and there are differences in the general level of interest rates applying to d i fferent decision-making units. In such an environment, there is a danger that d i fferences in efficiency may only reflect differences in the general level of interest rates (or other environmental factors) applying to different fim1s. From identifying a problem with one particular input variable, the research went on to l ook at some of the impacts of the selection of a range of different input and output variables, and identified the importance and value of including equity capital as an i nput. The l ast part of the research then set out to explore one of the i ssues that arises from the foreign ownership of the New Zealand banking system: are New Zealand banks more or less efficient than Austral ian banks, particularly as New Zealand banks appear to record rather higher levels of return on equity. 1 94 The conclusion was that 1 94 This also relates to the tax i ssue d i scussed in sub-section 4 .2 . 1 . above. ! 5 5 the differences were l ikely to be a reflection of different l evels of equity being appl ied. In the course of this analysis i t has become c lear that equity is an important component to the model l ing of bank efficiency. The research has also fai led to find evidence to support some commonly accepted myths about the way the banks work and prosper, even if these myths have not been specifical ly debunked. Very l ittle evidence was found for banks having got more efficient through time, apart from in those cases where banks had been particularly inefficient at the start of the period of the study. Banks may have got bigger, and their operating costs may not have increased at the same rate as their size, but that does not mean that they have got more efficient. Linked to this was the l ack of any convincing evidence for the existence of economies of scale in the commonly-expected shape of increasing returns to scale. Common sense encourages one to think that these ought to exist, but in this study at least, no rel iable evidence has been found to support this supposition. This undermines arguments that might otherwise be adduced in support of bank mergers, and which have been used in support of mergers in Australia and New Zealand. A key reason why the results from this research differ from those suggested by a focus on the cost to income ratio is that the mult ivariate approaches al low a greater emphasis on the contribution of revenue to bank efficiency. The banks that show as more efficient general ly show higher levels of revenue, which makes an interesting contrast with the prevalent attitude of banks in emphasi sing cost control . Fm1her advantages from use of multivariate DEA arise from its ability to handle trade-offs between both inputs and outputs. After al l , there should be no economic advantage to a bank in reducing non-interest costs if it does not also focus on interest costs, which general l y comprise a greater part ofNew Zealand banks' costs overal l . S imi lar i ssues can arise with non-interest income, increases in which are sometimes i dentified as an objective by bank management. It is revenue as a whole that is important to efficient bank performance, even if there is not the same trade-off evident between the two major sources of revenue (as outputs) . If the concerns raised by De Young & Rice (2004) could be substantiated for the New Zealand market, there may be distinct risks in emphasising non-interest income. 1 56 Another outcome of this research is an i l lustration of some of the potential advantages that may fol low from use of DEA as a technique for measuring bank efficiency. The research has suggested that a review of patterns of slacks can infonn as to the sources of inefficiency, while it can also assist in identifying factors that may act as constraints on efficiency . The research has also shown the ways in which addition of inputs, outputs and DMUs can impact on efficiency scores and their composition, and the potentially useful information that can be obtained in consequence. The composition of reference sets and identification of peers can also be important and provide useful information. These outcomes of the research exercise can be related to the problems, aims and objectives outlined in Chapter 1 of the d issertation. Although New Zealand banks' costs have reduced over the period studied (as shown in Figure 3, in section 2 .5) , this has not been reflected in corresponding improvements in effic iency. All that can be said is that New Zealand banks have maintained their levels of efficiency, and that this level of efficiency has been at least comparable to that achieved by Austral ian banks. When one compares the New Zealand banks to the Australian regional banks. the New Zealand banks shov?l as more efficient. The research has also identified an issues arising from the foreign ownership of the New Zealand banking system, pm1icularly for those banks that are Austral ian-owned. The Austral ian-owned New Zealand banks appear to make less use of capital as an input than do their parent banks . This is l ikely to reflect an abi lity to rely on their parent banks' names to conduct business in ew Zealand, and the effect can be seen in the New Zealand subsidiaries having the same credit ratings as the global business of their parent banks. As discussed in the next section, however, there are questions that this research has not been able to answer satisfactori ly : the research has looked at the efficiency of the individual banks that together account for most of the assets of the ew Zealand banking system, but it has not looked a the efficiency of the system as a whole . 1 5 7 6 .2 What th is research has not done This research i s not the l ast word on the efficiency of New Zealand banks or of the New Zealand financial system. It has not managed to look at the efficiency of New Zealand' s financial system as a whole, but only at the efficiency of a smal l number of firms that make up a part of it, even if the part they comprise is relatively large in percentage terms. Moreover, the measure of efficiency used i s only a relative measure against observed best practice : it is not known whether there is some technology avai lable, adoption of which would al low even the best performing firms to improve. No answer has therefore been provided to the conundrum identified by Diewert & Lawrence ( 1 999) about the lack of apparent productivity improvement in the financial services sector, although the findings obtained suggest that there may not in fact have been the productivity improvements that one might otherwise have expected to occur. One of the ways in which such productivity improvements might have been expected to be evidenced would be through the transformation of the payments system that has occuned s ince the early 1 990s. New Zealand used to have a payments system that was heavi ly based around paper, panicularly cheques, but the growth in the use of EFTPOS and credit cards in the 1 990s has seen dramat ic changes in the patterns of use of the payments system. As is demonstrated in F igure 1 3 , 1 95 the number of payment transactions undenaken in New Zealand has increased significantly, and because such a high proportion of those transactions are now undenaken using e lectronic methods, one would expect that the cost of those ought to have decreased, generating a welfare (economic) benefit. This has not been able to be measured, however. Measurement of these so11s of effects would require use of a production approach to modell ing those financial institutions that provide such payment services, and the data at individual bank level that would make this possible are not available. I t would be desirable also to look at the efficiency of the New Zealand payments system compared to those that operate in other countries. Is New Zealand ' s relatively i ntegrated electronic system such an advantage, and what i s the rel ationship between competitive conditions and effic iency in the payments system? What are the 195 Data for this have been obtained rrom the New Zealand Bankers ' Associat ion at hnp://www.nzba.org.nz/ 1 5 8 complications of trying to measure these sorts of effects across international borders? This has been a topic of debate in Australia since the publication of reports by the Reserve Bank of Australia, but it is not an issue that has been addressed in New Zealand (where payment system standards have been essentially determined by arrangement between private businesses) . 2,000,000,000 1 ,800 ,000,000 1 .600,000,000 1 ,400 .000,000 1 ,200 ,000,000 1 ,000,000,000 800,000,000 600.000,000 400,000,000 200,000,000 0 Figure 13 : Annual volumes of transactions through the New Zealand payments system .._ 1993 ? -------- ----? ____/ _,- ? - --- - -- - - 1 994 1995 1996 1997 1 998 1999 2000 2001 2002 2003 1--MICR -Electronic credits D?rect deb1ts Cred1t card -ATM --EFTPOS -TOTAL 1 -- .. -- >V 2004 There are other aspects of the efficiency of New Zealand ' s financial system which it would also have been desirable to try and measure. Does the disclosure regime provide an efficient approach to the regulation of banks, and does the OCR regime provide an efficient basis for the conduct of monetary policy. I nitial research ( Petro et al, 2003 ) suggests that the monetary policy regime is rather more effective that the regime that operated previously, but further work is warranted to c larify this conclusion. Another aspect of the efficiency ofNew Zealand 's fmancial system is in the extent to which the fmancial intermediation services it provides give appropriate support to the country's economic development. This might most often be conceived as being reflected in the support that is provided to smal l business, which is assumed to provide a key foundation for economic growth. This is an area in which a lmost no research has been undertaken, and yet its importance is acknowledged international ly in that we see some of the same people ( such as Alien Berger) involved in researching 1 5 9 both financial institution efficiency and the obstacles to the effective distribution of credit to small business. 1 96 I f it i s exceptional ly difficult for smal l businesses to access funding through the banking system, it is hard to argue that the financial system is therefore operating efficiently . 6.3 L im itations of this research Some of the l imitations of this research have been identi fied in the previous section where there was a review of what the dissertation did not address. However, even in the areas where some attempt has been made to cover i ssues, the treatment has not been as successfu l of effective as might have been hoped. This section seeks to identify some of these l imitations, which provide a l ink into the future research challenges and opportunities discussed in the next section. An obvious l imitation arises from the small cross-sectional sample of New Zealand banks, which has forced the use of panel data. Although this has some precedent in previous research, and it has been affirmed as a val id approach by Tulkens and Van den Eeckaut ( 1 995), it raises a conundrum in interpreting effic iency in different time periods, in terms of distinguishing the effects of efficiency change and technical change. Attempts to c larify thi s through use of the Malmquist Index must be open to question because of the smal l cross-section. For all that. it is bel ieved that the approach fol lowed in using panel data i s the best avai lable under the circumstances. Another l imitation has been in use of logit regressiOn to explore the impact of changing levels of interest rates on effic iency measures. Where logit regression has been used for second stage regression of efficiency scores in previous research. this has general ly been appl ied to efficiency scores generated under the SF A approach (e .g. M ester, 1 996), which wil l not be distributed in the same way as scores generated by DEA. A particular problem arises with DEA in that the logit of an efficiency score of 1 cannot be defined . I t i s l ikely that tobit regression, as recommended by Coel l i et al ( 1 998), would be a more satisfactory approach. 196 See, for example, Berger & Udell (2002) and the references contained therein. 1 60 It a l so appears that the proxy chosen for off-balance sheet business, in terms of an untransformed total of the face value of off-balance sheet items, but not including interest rate and foreign exchange contracts, may not be optimal . This may be why the research has fai led to establish the significance for off-balance sheet business found in other research (Siems & C lark, 1 997; Clark & Siems, 2002). Some options for overcoming this are discussed in the next section. Another area of l imitation has been in attempting to compare the efficiency of banks in New Zealand and Austral ia. There is no publ ished data which reports the financial statements of just the Austral ian business of the major Australian banks. and the data used for the Australian maj ors has therefore included their New Zealand operations. Because the New Zealand operations have comprised only a relatively small part of the Australian banks ' business the distortions this might have caused would only be expected to be minor, although thi s can be expected to change in the future, for the ANZ at least, where, fol lowing its acquisition of the NBNZ, operations in New Zealand now compri se a much larger prop011ion of its overall business . This provides us with a t imely reminder that the quality of any research must ultimately be constrained by the qual ity of the data used to undertake it . Although the data provided by the New Zealand disclosure regime is a great resource, it i s subj ect to l imitations, and there have sometimes been suggestions that some of the reporting is not as helpful as it might be. Moreover, the data avai lable is only financial statement data, and does not provide for any breakdown of financial magnitudes into quantities and prices, such as would be necessary to make assessments of allocative efficiency, or to follow orthodox procedures for investigation of profit effic iency ( see Berger & Mester, 1 997) . This is also one of the reasons why this research had to use DEA. Another aspect of l imitation arising from the data i s the failure of DEA, as used in this study, to account for random error. It is believed that judicious selection of cases for inclusion and checking using the super-efficiency model have allowed us to avoid inclusion of anything that would particularly di stort the results obtained, but the use of 1 6 1 non-parametric techniques provides less confidence m this regard that might be adduced if parametric methods had been used. There are some lesser l imitations as wel l . No satisfactory explanation i s immediately evident for the correlations between the inputs and outputs to the models used to study the ANZ and Westpac in sections 5 .2 .and 5 .3 (as reported in sub-section 4 .4 .3) . This also reflects the question of the appropriateness of correlation analysis as a basis for confirming input and output selections. This i s a suitable subject for further research, along with the issues out l ined in the next section. 6.4 Futu re research cha l lenges and opportun ities Future research opportunities do not al l have to be as broad-ranging as those identi fied in section 6 .2 . One outcome wi l l be the opportunity to look at the efficiency of New Zealand financial institutions over a longer period of t ime, which may also al low us to observe the performance of financial institutions in economic circumstances which are less favourable than have been enjoyed over the period of thi s research. It would also give us the opportunity to assess the effect on the efficiency of individual institutions of the ANTs acqui sition of the NBNZ at the end of 2003, and the potential effect of the gro\'-'lh and development of the business undertaken by Kiwibank and Superbank (St George Bank New Zealand Ltd ) , both of which were only in start-up phase at the end of the period covered by thi s research. It would also be desirable to investigate the impact of different input and outputs sets. and of d ifferent approaches to efficiency analysis . Some initial exploratory work i s already in train to investigate the impact of customer service quality and asset gro"-1h measures on efficiency scores, taking advantage of the flexibi lity of DEA to deal with non-financial inputs and outputs. There is a risk otherwise that some of the results obtained in this research appear to be relatively trivial . Future research might also look at alternative ways of measuring off-balance sheet business, such as through total risk-weighted assets : some exploratory work is under way to l ook at the usefulness of 1 62 such a measure, which might provide a more generally val id risk-adjusted measure of a bank' s output. 1 97 There i s also scope to try a range of different types of DEA models, such as the slacks-based and super-effic iency models (see Cooper et al, 2000). There are indications, for example, that the mix inefficiencies as measured using the slacks? based approach might highl ight some of the issues that arose when off-balance sheet items were included in the al l -bank models in sub-section 5 . 3 .4 . The super-efficiency model might provide a more satisfactory basis for second stage regression analysis (Lovel l et al , 1 994). Also, despite the l imitations imposed by the relatively smal l cross-section and the lack of data on pri ces, it may be possible to attempt use of parametric approaches (the distribution free approach - DF A - in particular). There i s a worthwhile corpus of l iterature that has used panel data for exploration of bank efficiency. There are some other challenges as wel l . There i s a body of research which has looked at the relationship between bank efficiency and competitive conditions, in terms of identifying the relative merits of the structure-conduct performance hypothesis and effic ient structure hypothesis (e .g . Berger, 1 995 ; Goldberg & Rai , 1 996) : some proper analysis of bank efficiency in New Zealand should provide a foundation for further work in thi s area. It might also be that there is a feedback effect from competitive conditions back to efficiency, such as is identi fied under the "quiet l i fe" hypothesis (Berger & Haru1an. 1 998) . There is also the problem of economies of scale : it seems so obvious that they ought to exi st, but \\'hy i s it so hard to find any actual evidence to supp011 them. 1 98 A lthough thi s research has been undertaken with and future research may fol low with a New Zealand focus, one should not assume that the research should lack wider interest or application. The distinctive feature of the New Zealand market i s the extent of foreign ownership, and as global isation causes banking systems in other countries to become increasingly foreign-owned, the questions being addressed in New Zealand 1 97 Relevant data i s avai lable on a quarterly basis as part of New Zealand's d i sclosure regime. 1 98 This finding is consistent with earl ier research : see, for example, Humphrey ( 1 985) . 1 63 wil l come to be of increasing imp011ance elsewhere. The rat ionale and justification for thi s research i s assured ! 1 64 References Alam, I . M. S . (200 1 ) . A nonparametric approach for assessing productivity dynamics of large U.S . banks. Journal of Money, Credit, and Banking 33 ( 1 ) . 1 2 1 - 1 39 . Al ien, L. & Rai , A. ( 1 996). Operational efficiency in banking: an international comparison. Journal of Banking and Finance. 20. 65 5-672 . A ltunbas, Y.; Gardener, E . P . M . ; Molyneux, P. & Moore, B . (200 1 ) . Efficiency in European banking. European Economic Review. 45. 1 93 1 - 1 955 . A ltunbas, Y . & Molyneux, P . ( 1 996). Cost economies in EU banking systems. Journal of Economics and Business. 48. 2 1 7-230. Aly, H. Y. ; Grabowski . R. ; Pasurka, C. & Rangan, N . ( 1 990). Teclmical, scale and allocative efficiencies in U .S . banking: an empirical investigation. The Review of Economics and Statistics. 72. 2 1 1 -2 1 8 . Andersen, P. & Petersen, N . C . ( 1 993). A procedure for ranking efficient units in data envelopment analysis . Management Science. 39 . 1 26 1 - 1 264. Asmild, M. ; Paradi, J. C . ; Aggarwall , V. & Schaffnit, C . (2004) . Combining DEA window analysis with the Malmquist Index approach in a study of the Canadian banking industry. Journal of Productirity Analysis. 2 1 . 67-89. Athanassopoulos, A. D. ( 1 997) . Service qual ity and operating efficiency synergies for management control in the provision of financi al services: evidence from Greek bank branches. European Journal of Operational Research. 98. 300- 3 1 3 . Athanassopoulos, A . D. ( 1 998 ) . Nonparametric frontier models for assessing the market and cost effic iency of large-scale bank branch networks. Journal of Money. Credit, and Banking 30 (2) . I 72- 1 92 . Athanassopoulos, A . D. (2000). An optimisation framework of the triad : service capabi lities, customer satisfaction and performance . Chapter 9 in Harker, P. T. & Zenios, S. A., Pe1jormance of Financial Institutions -Efficiency. Innovation and Regulation. (pp 3 1 2-3 35) . Cambridge University Press. Athanassopoulos, A. D. ; Soteriou, A. C. & Zenios, S. A. (2000). Disentangl ing within- and between-country efficiency differences of bank branches. Chapter 1 0 in Harker, P. T, & Zenios, S. A. , Pe1jormance of Financial Institutions ? Efficiency, Innovation and Regulation. (pp 336-363) . Cambridge University Press. Avkiran, N. K. ( 1 999a). The evidence on effic iency gains: the role of mergers and the benefits to the publ i c . Journal of Banking and Finance. 23 . 99 1 - 1 0 1 3 . 1 65 A vkiran, N K. ( 1 999b ) . An appl ication reference for data envelopment analysis in branch banking: helping the novice researcher. International Journal of Bank Marketing. 1 7 (5 ) . 206-220. Avkiran, N. K. (2000). Rising productivity of Australian trading banks under deregulation 1 986- 1 995 . Journal of Economics and Finance. 24 (2) . I 22 - I 40. Avkiran, N . K . (2002). Productivity Analysis in the Service Sector with Data Envelopment Analysis. (Second Edition). Camira, Queensland : K Avkiran. Balk, B. M. (200 I ) . Scale efficiency and productivity change. Journal of Productivity Analysis. 1 5 . I 59- I 83 . Banker, R . D . ; Charnes, A . W . & Cooper, W . W. ( I 984). Some models for estimating teclmical and scale inefficiencies in Data Envelopment Analysi s. Management Science. 30 (9) I 078- I 092. Bauer, P . W. ; Berger, A. . ; Ferrier, G . D. & Humphrey, D . B . ( 1 998) . Consistency conditions for regulatory analysis of financial institutions: a comparison of frontier efficiency methods. Journal of Economics and Business. (50) . 8 5 - I I 4 . Bauer, P . W. ; Berger, A. N. & Humphrey, D . B . ( 1 993 ) . Efficiency and productivity growth in U.S . banking. Chapter I 6 in (eds) Fried, H . 0. ; Lovel l , C . A . K. & Schmidt, S . S . , The !11easurement of Productive Efficiency. (pp 3 86-4 1 3 ) . New York : Oxford University Press. Berg, S. A . ; Forsund, F . ; Hjalmarsson, L. & Suominen, M. ( 1 993) . Banking efficiency in the Nordic countries . Journal of Banking and Finance. 1 7 . 3 7 1 -388 . Berg, S . A . ; Forsund, F . R. & Jansen. E . S . ( 1 992 ) . Malmquist indices of productivity growth during the deregulation ofNorwegian banking, 1 980-89 . Scandinavian Journal a./Economics. 94 (Supplement). 2 I I -228 . Bergendahl, G . ( 1 998) . DEA and benchmarks - an application to Nordic banks. Annals of Operations Research. 28 . 233 -249. Berger, A. N. ( 1 993). "Distribution-free'? estimates of efficiency in the U .S . banking industry and tests of the standard distributional assumptions. Journal of Productivity Analysis. 4 . 2 6 1 -292. Berger, A. N. ( 1 995, May). The profit-structure relationship in banking - tests of market-power and efficient-structure hypotheses. Journal o.l Money, Credit, and Banking. 27 (2) . 404-43 I . Berger, A . N . & DeYoung, R . ( I 997). Problem loans and cost efficiency in commercial banks. Journal o.l Banking and Finance. 2 I . 849-870. Berger, A. N. & DeYoung, R. (200 1 ) . The effects of geographic expansion on bank efficiency. Journal of Financial Services Research. 1 9 (2/3) . I 63 - 1 84 . 1 66 Berger, A. N . ; DeYoung, R. ; Genay, H . & Udel l , G. F. (2000). Global izat ion of financial institutions: evidence from cross-border banking performance. Brookings-Wharton Papers on Financial Services. 3. 2 3 - 1 58. Berger, A. N.; Hancock, D. & Humphrey, D. B. ( 1 993a). Bank efficiency derived from the profit function. Journal of Banking and Finance. 1 7 . 3 1 7-34 7 . Berger, A. N. ; Hanweck, G. A. & Humphrey, D . B . ( 1 987) . Competitive viability in banking - scale, scope, and product mix economies. Journal of Monetary Economics. 20. 50 1 -520. Berger, A. N. & Hannan, T. H. ( 1 998). The efficiency cost of market power in the banking industry : a test of the "quiet l ife" and related hypotheses . Review of Economics and Statistics. 80. 454-465 . Berger, A. N . & Humphrey, D. B . ( 1 99 1 ) . The dominance of inefficiencies over scale and product mix economies in banking. Journal of Monetary Economics. 28 . 1 1 7- 1 48. Berger, A. N. & Humphrey, D. B . ( 1 992a). Measurement and efficiency issues in commercial banking. Chapter 7 in Gril iches, Z. ; Berndt, E . R . ; Bresnahan, T. F . & Manser, M . , Output Measurement in the Service Sectors. (pp 245-279) . Chicago : University of Chicago Press. Berger, A. N. & Humphrey, D. B. ( 1 992b). Megamergers in banking and the use of cost efficiency as an antitrust defence. The Antitrust Bulletin. 37 . 54 1 -600. Berger, A . N. & Humphrey, D. B. ( 1 997). Effic iency of financial institutions: international survey and directions for future research. European Journal of Operational Research. 98 . 1 75-2 1 2 . Berger, A . N . ; Hunter, W. C . & Timme. S . G . ( 1 993b). The efficiency of financial institutions: a review and preview of research past, present and future. Journal of Banking and Finance. 1 7 . 22 1 -249. Berger, A. N. ; Leusner, J. & Mingo, J. ( 1 997) . The efficiency of bank branches. Journal of Monetary Economics. 40 ( 1 ). 1 4 1 - 1 62 . Berger, A. N. & Mester, L . J . ( 1 997). Inside the black box : what explains differences in the effic iencies of financial institutions? Journal of Banking and Finance. 2 1 . 895-947. Berger, A . N. & Mester, L . J . (2003) . Explaining the dramatic changes in performance of US banks: technological change, deregulation, and dynamic changes in competition. Journal of Financial Intermediation. 1 2 . 57-95. Berger, A . N. & Udel l . G . F . (2002) . Smal l business credit avai labil ity and relationship lending: the impm1ance of bank organisational structure. The Economic Journal. 1 1 2 . F32-F53 . 1 67 Bhattacharyya, A . ; Lovel l , C . A. K. & Sahay, P . ( 1 997). The impact of l iberalization on the productive efficiency of Indian commercial banks. European Journal of Operational Research. 98 . 332-345 . Boussofiane, A . ; Dyson, R. G . & Thanassoulis, E . ( 1 99 1 ) . Applied data envelopment analysis . European Journal of Operational Research. 52 . 1 - 1 5 . Brookes, A. & Hampton, T. (2000). The Official Cash Rate one year on. Reserve Bank of New Zealand Bulletin. 63 (2 ) . 53 -6 1 . Brown, R. & O'Connor, I . ( 1 995) . Measurement of economies of scale in Victorian credit unions. Australian Journal of Management. 20 ( 1 ) . 1 -24. Burns, F . ( 1 989) . Built on Trust. Waikanae : The Heritage Press. Canhoto, A. & Dermine, J. (2003 ), A note on banking efficiency in Portugal , new vs old banks. Journal of Banking and Finance. 27 . 2087-2098. Carbo, S. ; Gardener, E . P . M. & Will iams. J . (2002) . Efficiency in banking: empirical evidence from the savings bank sector. The Manchester School. 70 (2) . 204- 228 . Carew, E. ( 1 987) . New Zealand 's Money Revolution. Wel l ington: Alien & Unwin!Po11 Nicholson Press. Casu, B . & Molyneux, P. (2003 ) . A comparative study of effic iency in European banking. Applied Economics. 35 . 1 865 - 1 876 . Cavallo, L & Rossi, S . P . S . (200 I ) . Scale and scope economies in the European banking systems. Journal of Multinational Financial Management. 1 1 . 5 1 5 - 53 1 . Caval lo. L & Rossi. S . P . S . (2002 ) . Do enviro1m1ental variables affect the performance and teclu1ical efficiency of the European banking systems? A parametric analysis using the stochastic frontier approach. The European Journal a/Finance. 8 . 1 23 - 1 46 . Chaffai , M. E. ; Dietsch, M . & Lozano-Vivas, A. (200 1 ) . Teclmological and environmental differences in the European banking industries. Journal of Financial Sen?ices Research. 1 9 (2/3 ) . 1 4 7- 1 62. Charnes, A. & Cooper, W. W. ( 1 985) . Preface to topics in Data Envelopment Analysis . Annals of Operations Research. 2 . 59-94. Charnes, A.; Cooper, W. W. ; Golany, B. & Seiford, L . ( 1 985) . Foundations of data envelopment analysis for Pareto-Koopmans effic ient empirical production functions. Journal of Econometrics. 30. 9 1 - 1 07 . 1 68 Charnes, A . : Cooper, W. W. & Rhodes, E . ( 1 978) . Measuring the efficiency of decision making units. European Journal of Operational Research. 2 . 429- 444. Chen, T-Y. (200 1 ). An est imation of X-inefficiency in Taiwan' s banks. Applied Financial Economics. 1 1 . 23 7-242 . Cherchye, L . ; Kuosmanen, T. & Post, T. (2000). What i s the economic meaning of FDH? A reply to Thral l . Journal of Productivity Analysis. 1 3 . 263-267. Chri stopoulos, D, K . ; Lolos, S . E . G. & Tsionas, E . G. (2002) . Efficiency of the Greek banking system in view of the EMU: a heteroscedastic stochastic frontier approach. Journal o.f Policy Modelling 24. 8 1 3-829 . Chu, S . F. & Lim, G. H. ( 1 998) . Share performance and profit efficiency of banks in an ol igopolistic market : evidence from Singapore. Journal of Multinational Financial Management. 8 . 1 55- 1 68 . Claessens, S . , Demirgily-Kunt, A. , Huizinga, H. (200 1 ) . How does foreign entry affect domestic banking markets? Journal of Banking and Finance. 25 . 89 1 -9 1 1 . Clark, J . A . ( 1 988, September/October). Economies of scale and scope at depository financial institutions: a review of the l iterature. Economic Review (Federal Reserve Bank of Kansas Cif)1. 1 6-33 . Clark, J . A . & Siems, T. F . (2002, November). X-efficiency in banking: looking beyond the balance sheet. Journal of Money, Credit and Banking 34 (4). 987- 1 0 1 3 . Coel l i , T . ; Prasada Rao, D . S . & Battese, G. E . ( 1 998) . An Introduction to Ef iciency and Productivity Analysis. Boston : Kluwer Academic Publishers. Cook, W.D. ; Chai, D . ; Doyle, J. & Green, R. ( 1 998) . Hierarchies and groups in DEA. Journal of Productivity Analysis. 1 0 . 1 77- 1 98. Cooper, W. W. ; Seiford, L . M. & Tone, K. (2000) . Data Envelopment Analysis. Boston : Kluvver Academic Publ i shers. Dawe, S. ( 1 990). Reserve Bank of New Zealand Act 1 989 . Reserve Bank o.fNew Zealand Bulletin. 53 ( 1 ) . 29-36. De Borger, B. ; Ferrier, G. D . & Kerstens, K. ( 1 998) . The choice of a teclmical efficiency measure on the free disposal hul l reference technology: a comparison using US banking data. European Journal of Operational Research. 1 05 . 427-446. Deane, R. S . ; Nichol l , P. W. E . & Smith, R. G. ( 1 983 ) . Monetary Policy and the New Zealand Financial System. (Second Edition). Wel l ington : Reserve Bank of New Zealand . 1 69 Demirgi.i9-Kunt, A. & Huizinga, H . (200 1 ) . The taxation of domest ic and foreign banking. Journal of Public Economics. 79. 429-453 . De Young, R. ( 1 997a). A diagnostic test for the distribution-free efficiency estimator: an example using U .S commercial bank data. European Journal of Operational Research. 98 . 243-249. De Young, R. ( 1 997b ) . Measuring bank cost effic iency: don' t count on accounting ratios. Financial Practice and Education. Spring/Summer. 20-3 1 . De Young, R . ( 1 998). Management quality and X -efficiency in national banks. Journal of Financial Services Research. 1 3 ( 1 ) . 5-22. De Young, R. & Hasan, I ( 1 998) . The performance of de novo commercial banks : a profit efficiency approach. Journal of Banking and Finance. 22 . 565-587 . De Young, R . & Rice, T. (2004, Quarter 4) . How do banks make money? The fal lacies of fee income. Economic Perspectives (Federal Reserve Bank of Chicago). 34- 5 1 . Dietsch, M. & Lozano-Vivas, A . (2000) . How the environment determines banking efficiency: a comparison between the French and Spanish industries. Journal of Banking and Finance. 24. 985 - 1 004. Diewert, E. & Lawrence, D. ( 1 999). Measuring New Zealand 's Productivity. New Zealand Treasury Working Paper 99/5 . Doughty, A . J . ( 1 986). New Banks and Financial Structure Reform. Chapter 7 in Financial Policy Reform. ( pp 1 1 1 - 1 23) . Wel l ington : Reserve Bank of New Zealand. Drake, L. (200 1 ) . Efficiency and productivity change in UK banking. Applied Financial Economics. 1 1 . 55 7-57 1 . Drake, L . & Hal l , M. J . B . (2003) . Efficiency in Japanese banking: an empirical analysis . Journal of Banking and Finance. 27 . 8 9 1 -9 1 7 . Drake, L . & Weyman-Jones. T. G . ( 1 996). Productive and a l locative inefficiencies in U.K. building societ ies : a comparison of non-parametric and stochastic frontier techniques . The Manchester School. 64 ( 1 ). 22-3 7 . Dyson, R . G . ; Allen, R. ; Camanho, A. S . ; Podinovski, V . V . ; Sarrico, C. S . & Shale, E. A. (200 1 ) . Pitfa l l s and protocols in DEA. European Journal of Operational Research. 1 32 . 245-259. Edvardsen, D . F . & F0rsund, F . R . (2003) . l ntemational benchmarking of electricity distribution util i ti es . Resource and Energy Economics. 25 . 3 53-37 1 . 1 70 Elyasiani, E . & Mehdian, S. ( 1 990). A nonparametric approach to measurement of efficiency and teclmological change: the case of large U.S . commercial banks. Journal of Financial Services Research. 4. 1 57- 1 68. Elyasiani, E . & Mehdian, S . ( 1 992 ) . Productive efficiency performance of m inority and nonminority-owned banks: a nonparametric approach. Journal of Banking and Finance. 1 6 . 933-948. Esho, N . (200 1 ) . The detem1inants of cost efficiency in cooperative financial institutions: Australian evidence. Journal of Banking and Finance. 25 . 94 1 - 964. Esho, N. & Sharpe, I . G. ( 1 995) . Long-run estimates of technological change and scale economies in a dynamic framework: Australian permanent bui lding societies, 1 974- 1 990. Journal of Banking and Finance. 1 9 . 1 1 35 - 1 1 57 . Esho, N . & Sharpe, I . G . ( 1 996) . X-efficiency of Australian pem1anent bui lding societies, 1 974- 1 990. The Economic Record. 72. 246-259. Evanoff, D. D. ( 1 988) . Branch banking and service accessibility. Journal of Money Credit and Banking 20 (2 ) . 1 9 1 -202 . Evanoff, D. D. & Israilevich, P . R. ( 1 99 1 ) . Productive efficiency in banking. Economic Perspectives (Federal Reserve Bank of Chicago). 1 1 -3 2 . Evans, L . ; Grimes, A . ; Wilkinson, B . & Teece, D . ( 1 996. December). Economic reform in Nev?/ Zealand 1 984-95 : the pursuit of efficiency. Journal of Economic Literature. 34 . 1 856- 1 902. Hire, R . & Grosskopf. S . ( 1 996). Jntertemporal Production Frontiers: With Dynamic DEA. Boston : K1uwer. Farre l l , M. J . ( 1 957) . The measurement of productive efficiency. Journal of the Royal Statistical Society. 1 20 . 253 -28 1 . Favero, C . A. & Papi , L . ( 1 995) . Teclmical effic iency and scale efficiency in the Ital ian banking sector: a non-parametric approach. Applied Economics. 27 . 3 85 -395 . Fecher, F . & Pestieau, P . ( 1 993). Efficiency and competition in O.E.C .D financial services. In Fried, H. 0.; Lovell , C. A. K. & Schmidt S . S . (eds), The Measurement of Productive Efficiency. (pp 374-3 85 ) . New York : Oxford University Press. Ferguson, R . A. ( 1 990, September). Foreign banks in Australia - a strategic reassessment. Economic Papers (Economic Society of A ustralia). 1 -8 . Ferrier, G . D. & H irschberg, J . G. ( 1 997) . Bootstrapping confidence intervals for l inear programming effic iency scores: with an i l lustration using Italian banking data. Journal of Productivity Analysis. 8 . 1 9-33 . 1 7 1 Ferrier, G. D. & Lovel l , C. A. K. ( 1 990). Measuring cost efficiency in banking: econometric and l inear programming evidence. Journal of Econometrics. 46. 229-245 . Frei, F . X . ; Harker, P . T . & Hunter, L . W . (2000). Inside the b lack box : what makes a bank efficient? Chapter 8 in Harker, P . T, & Zenios, S . A. , Pe1jormance of Financial Institutions -Efficiency, Innovation and Regulation. (pp 258-3 1 1 ) . Cambridge University Press. Fried, H. 0. ; Schmidt, S . S. & Yaiswarng, S. ( 1 999). Incorporating the operating environment into a nonparametric measure of technical efficiency. Journal of Productivity Analysis. 1 2 . 249-267. Fukuyama, H . ( 1 993) . Teclmical and scale efficiency of Japanese commercial banks : a non-parametric approach. Applied Economics. 25 . 1 1 0 1 - 1 1 1 2 . Fukuyama, H . ( 1 995 ) . Measuring efficiency and productivity growth in Japanese banking: a nonparametric frontier approach. Applied Financial Economics. 5 . 95 - 1 07 . Garden, K. & Ralston, D. ( 1 999). The X-efficiency and allocative efficiency effects of credit union mergers . Journal of International Financial Markets, Institutions and Money. 9. 285 -3 0 1 . Gi lbert, R. A. & Wi 1son , P. W. ( 1 998). Effects of deregulation on the productivity of Korean banks. Journal of Economics and Business. 50. 1 33 - 1 5 5 . Golany, B . & Storbeck, J . E . ( 1 999, May-June). A Data Envelopment Analysis of the Operational Efficiency of Bank Branches. Inteifaces. 29. 1 4-26. Golany, B. & Yu, G . ( 1 997) . Estimating returns to scale in DEA. European Journal of Operational Research. 1 03 . 28-37 . Goldberg, L. G. & Rai . A. ( 1 996). The structure-performance relationship for European banking. Journal of Banking and Finance. 20 . 745-77 1 . Gong, B . & Sickles, R. C . ( 1 992). Finite sample evidence on the performance of stochastic frontiers and data envelopment analysis using panel data. Journal of Econometrics. 5 1 . 2 59-284. Grabowski, R . ; Rangan, N. & Rezvanian, R. ( 1 993) . Organizational forms in banking: an empirical investigation of cost efficiency . Journal of Banking and Finance. 1 7 . 53 1 -538 . Grifell-Tatje, E . & Love l l . C . A. K. ( 1 999). Profits and productivity. Management Science. 45 (9) . 1 1 77 - 1 1 93 . Grimes, A . ( 1 998). L iberal i sation of financial markets in New Zealand. Reserve Bank of New Zealand Bulletin. 6 1 (4). 29 1 -306. 1 72 Grosskopf, S . ( 1 996). Statistical inference and nonparametric efficiency: a selective survey. Journal of Productivity Analysis. 7 . 1 6 1 - 1 76 . Hancock, D. ( 1 986). A model of the financial firm with imperfect asset and deposit elasticities. Journal of Banking and Finance. I 0 . 3 7-54. Hancock, D. ( 1 99 1 ) . A Theory ofProductionfor the Financial Firm. Boston: Kluwer Academic Publishers. Harper, D. A. ( 1 986). The Financial Services Industry: Effects of Regulatory Reform. Research Paper 35 . Wellington: New Zealand Institute of Economic Research. Harper, D. A. & Karacaoglu, G. ( 1 987) . Financial Policy Reform in New Zealand. Chapter 1 0 in Bollard, A & Buckle, R (eds), Economic Liberalisation in New Zealand. (pp 206-235) . Well ington: A lien & Unwin. Harrison, I. ( 1 996, June) . Disclosure of registered banks' market risks. Reserve Bank Bulletin. 59 (2) . 1 46- 1 54 . Hartman, T. E . ; Storbeck, J . E . & Byrnes, P . (200 1 ) . Allocative efficiency in branch banking. European Journal of Operational Research. 1 34. 232-242. Hi l l -Cone, D. (2004, 3 September). The great Aussie bank swindle. National Business Review. Pp 1 8- 1 9 . Holmes, F. W. ( 1 999). The Thoroughbred amongst Banks: 1 945- 198-1: Volume 2. Well ington : The National Bank ofNew Zealand L imited. Holmes, F . W. (2003) . The Thoroughbred among Banks in New Zealand: Banking in New Territory. 1 98-1-2000. (Volume 3). Wel l ington: The National Bank of New Zealand L imited. Hogan, W. ( 1 99 1 , March) . New banks: impact and response. Economic Papers (Economic Society of Australia). 1 1 -3 3 . Hughes. A. & Yaiswarng, S . (2004). Sensitivity and dimensional ity tests of DEA efficiency scores . European Journal of Operational Research. 1 54 . 4 1 0-422 . Hughes, J . P . & Mester, L . J . ( 1 993) . A qual ity and risk-adjusted cost function for banks: evidence on the ?'Too-big-to-fail" doctrine. The Journal of Productivity Analysis. 4 . 293 -3 1 5 . Humphrey, D . B . ( 1 985) . Costs and scale economies in bank intermediation. Chapter 24 in Aspinwal l , R. C & E isenbeis, R. A. (eds), Handbookfor Banking Strategy. (pp 745-783). Humphrey, D. B . ( 1 990, September/October). Why do estimates of bank scale economies differ? Economic Review (Federal Reserve Bank of Richmond) . 3 8- 50. 1 73 Humphrey, D . B . ( 1 99 1 , March! Apri l ) . Productivity in banking and effects from deregulation. Economic Review (Federal Reserve Bank of Richmond) . 1 6-28 . Isik, l . & Hassan, M. K. (2003) . Financial deregulation and total factor productivity change : an empirical study of Turkish commercial banks . Journal of Banking and Finance. 27 . 1 455- 1 485 . KPMG. (2004). Financial Institutions Pe1jormance Survey. Wellington: KPMG. Kumbhakar, S . C. ; Lozano-Vivas, A.; Lovell , C. A. K. & Hasan, I . (200 1 , February). The effects of deregulation on the performance of financial institutions: the case of Spanish savings banks. Journal ofMoney, Credit and Banking 33 ( 1 ). 1 0 1 - 1 20. Leibenstein, H . ( 1 966). Allocative efficiency vs . "X-efficiency. American Economic Review. 56. 392-4 1 5 . Leightner, J . E . & Lovell, C . A . K . ( 1 998). The impact of financial liberalization on the performance of Thai banks. Journal of Economics and Business. 50. 1 1 5- 1 3 1 . Leong, W. H. ; Dollery, B . & Coel l i , T . (2003). Measuring the technical efficiency of banks in Singapore for the period 1 993-99. ASEAN Economic Bulletin. 20 (3 ) . 1 95 -2 1 0 . Liu, B . & Tripe, D. W. L . (2002) . New Zealand bank mergers and efficiency gains. Journal of Asia Pac?fic Business. 4 (4). 6 1 -8 1 . Llewellyn, I . (2004. 1 4 August) . Banks probed on tax records . New Zealand Herald. P C3 . Lovell , C . A . K. ( 1 993) . Production frontiers and productive efficiency. Chapter 1 in (eds) Fried, H . 0.; Love! ! , C . A . K. & Sclm1idt. S . S . . The Jvfeasurement of Productive Efficiency. (pp 3-67) . New York : Oxford University Press. Lovell , C . A. K. (2003 ) . The decomposition of Malmquist productivity indexes . Journal of Productivity Analysis. 20. 43 7-458 . Lovel l , C. A . K . & Rouse, A . P . B . (2003 ) . Equivalent standard model s to provide super-efficiency scores. Journal ofthe Operational Research Society. 54. 1 0 1 - 1 08 . Lovel l , C . A . K . ; Waiters, L . C . & Wood, L . L . ( 1 994) . Strat ified models o f education production using modified DEA and regression analysis . Chapter 1 7 in Chames, A. ; Cooper, W. W. ; Lewin, A. Y. & Seiford, L . M. (eds), Data Envelopment Analysis: Theory. Methodology and Application. (pp 329-35 1 ). Boston : Kluwer Academic Publ i shers. 1 74 Lozano-Vivas, A . ( 1 997) . Profit efficiency for Spanish savings banks. European Journal of Operational Research. 98 . 38 1 -394 . Lozano-V i vas, A . ( 1 998) . Efficiency and technical change for Spanish banks. Applied Financial Economics. 9. 289-300. Lozano-Vivas, A.; Pastor, J . T. & Pastor, J. M. (2002). An efficiency comparison of European banking systems operating under different environmental conditions. Journal of Productivity A nalysis. 1 8 . 59-77 . Matthews, C .D. & Tripe, D. W. L . (2004) . Bank computing in a changing economic ------..e?llvironment : the IBIS project in New Zealand. Accounting Business and Financial History. 1 4 (3 ) . 3 0 1 -3 1 5 . Maudos, J . ; Pastor, J . M . ; Perez, F . & Quesada, J . (2002). Cost and profit efficiency in European banks. Journal of International Financial Jvfarkets. Institutions and Money. 1 2 . 3 3 -58 . Mayes, D. G. ( 1 998) . Improving Banking Supervision. Bank of Finland Discussion Paper 23/98 . McAll ister, P. H. & McManus, D. ( 1 993 ) . Resolving the scale efficiency puzzle in banking. Journal of Banking and Finance. 1 7 . 389-405 . Mendes, V . & Rebelo. J . ( 1 999). Productive efficiency. technological change and productivity in Portuguese banking. Applied Financial Economics. 9. 5 1 3 -52 1 . Mester, L . J . ( 1 987, January/February) . Efficient production of financial services: scale and scope economies. Business Reriew (Federal Reserve Bank of Philadelphia). 1 5-25. M ester, L . J . ( 1 993 ). Efficiency in the savings and loan industry. Journal of Banking and Finance. 1 7 . 267-286. Mester, L . J . ( 1 994 , January/February). How efficient are third district banks? Business Reriew (Federal Reserve Bank of Philadelphia). 3 - 1 8 . Mester, L . J . ( 1 996). A study of bank efficiency taking into account risk-preferences. Journal of Banking and Finance. 20. 1 025 - 1 045. Mester, L . J . ( 1 997). Measuring efficiency at U.S. banks : accounting for heterogeneity is important. European Journal of Operational Research. 98. 230-242. Mi l ler, S. M. ( 1 996). The technical efficiency of large bank production. Journal of Banking and Finance. 20 . 495-509. Molyneux, P . & ShamroukJ1, N . ( 1 999). Financial Innovation. Chichester: John Wiley & Sons. 1 75 Mortlock, G . ( 1 996, March) . New disclosure regime for registered banks. Reserve Bank Bulletin. 59 ( 1 ) . 2 1 -29. Nicholl, P, W. E. & King, M. F . ( 1 985 ). Financial institutions and markets in New Zealand. Chapter 3 in Skully, M. T (ed), Financial Institutions and Markets in the Southwest Pacific. (pp 1 60-244) . New York : St Martins Press. oulas, A. G. ( 1 997). Productivity growth in the Hel lenic banking industry : state versus private banks. Applied Financial Economics. 7. 223-228. Noulas, A. G.; Ray, S . C . & Mi ller, S . M . ( 1 990, February) . Returns to scale and input substitution for large U.S . banks. Journal of Money, Credit and Banking 22 ( 1 ) . 94- 1 08 . Nunamaker, T. R. ( 1 985) . Using Data Envelopment Analysis to measure the efficiency of non-profit organizations: a critical evaluation. Managerial and Decision Economics. 6 ( 1 ), 50-58 . OECD (Organisation for Economic Cooperation and Development). (2002) . Bank Profitability: Financial Statements of Banks. Paris : OECD. Pastor, J . M . ; Perez, F . & Quesada, J . ( 1 997). Efficiency analysis in banking firms : an international comparison. European Journal of Operational Research. 98 . 395 -407. Petro. B ; McDermott, .J , & Tripe, D. W. L . (2003 ) . The link between rhe Official Cash Rate (OCR) and market interest rates - a New Zealand perspectire. Paper presented at t11 New Zealand Finance Col loquium. Avai lable at : http ://centre? banking -studi es.massey .ac .nzlresearch-pro g . asp. Resti, A . ( 1 997). Evaluating the cost-efficiency of the Ital ian banking system: what can be learned from the joint appl ication of parametric and non-parametric techniques. Journal of Banking and Finance. 2 1 . 22 1 -250. Resti , A . (2000). Efficiency measurement for multi-product industries: a comparison on c lassic and recent techniques based on simulated data. European Journal of Operational Research. 1 2 1 . 5 59-578 . Rezvanian, R . & Mehdian. S . (2002) . An examination of cost structure and production perfom1ance of commercial banks in S ingapore. Journal of Banking and Finance. 26. 79-98 . Rime, B . & Stiroh, K . (2003 ) . The performance of universal banks : evidence from Switzerland. Journal of Banking and Finance. 27 . 2 1 2 1 -2 1 50. Rogers, K. E. ( 1 998) . Nontraditional activities and the efficiency of US commercial banks . Journal of Banking and Finance. 22. 467-482. Ruggiero, .J . (2000). Measuring technical efficiency. European Journal of Operational Research. 1 2 1 . 1 3 8 - 1 50 . 1 76 Santin, D . ; Delgado, F. J . & Val ino, A. (2004). The measurement of technical efficiency: a neural network approach. Applied Economics. 36. 627-63 5 . Sathye, M . (200 1 ) . X-efficiency in Australian banking: an empirical investigation. Journal of Banking and Finance. 25 . 6 1 3-630 . Sathye, M . (2002). Measuring productivity changes in Australian banking: an appl ication ofMalmquist indices. Managerial Finance. 28 (9). 48-59. Schaffnit, C . ; Rosen, D. & Paradi , J . C. ( 1 997). Best practice analysis of bank branches: an application ofDEA in a large Canadian bank. European Journal of Operational Research. 98. 269-289. Schumpeter, J . ( 1 943 ). Capitalism, Socialism and Democracy. Alien & Unwin. Sealey, C . W. & Lindley, J . T. ( 1 977) . Inputs, outputs, and a theory of production and cost at depository financial institutions. Journal of Finance. 32 ( 4) . 1 25 1 - 1 266. Seiford, L . M. & Thrall , R. M. (1 990). Recent developments in DEA: the mathematical programming approach to frontier analysis . Journal of Econometrics. 46. 7-3 8 . Seiford, L .M. & Zhu, .J. ( 1 999). An investigation of retums to scale in data envelopment analysis . Omega. 27 . 1 - 1 1 . Sherman, H . D. & Gold, F . ( 1 985) . Bank branch operating efficiency: evaluation with data envelopment analys is . Journal of Banking and Finance. 9. 297-3 1 5 . Siems, T. F. & Barr, R. S . ( 1 998, December) . Benclm1arking the productive effic iency ofU.S . banks. Financial Industry Studies (Federal Reserve Bank of Dallas) . 1 1 -24. Siems, T. F. & Clark. J . A. ( 1 997, December) . Rethink ing bank efficiency and regulation : how off-balance-sheet activities make a difference. Financial Industry Studies (Federal Reserve Bank of Dallas) . 1 - 1 2 . Simar, L . & Wilson, P . W. ( 1 998) . Sensitivity analysis of effic iency scores: how to bootstrap in nonparametric frontier models . . Management Science. 44 ( 1 ). 49- 6 1 . Simar, L . & Wilson, P . W. (2000). Statistical inference i n nonparametric frontier models : the state of the art . Journal of Productivity Analysis. 1 3 . 49-78 . Smith, R. & Tripe, D . (200 1 ). Competition and contestability in New Zealand 's banking system. Paper presented at the 1 4111 Australasian Finance and Banking Conference, Sydney. Available at http ://centre-banking- studies.massey .ac.nz/research _ detai ls .asp?output_id=3 1 4 1 77 Soteriou, A. C . & Stavrinides, Y . ( 1 997). An internal customer serv ice qual ity data envelopment analysis model for bank branches. International Journal of Operations and Production Management. 1 7 (8 ) . 780-789 . Spencer, G . & Carey, D . ( 1 988) . Financial Policy Reform - the New Zealand Experience, 1 984-1 98 7. Reserve Bank ofNew Zealand Discussion Paper G88/ 1 . Staff. ( 1 997) . Bank defends disclosure regime. The Dominion. 22 January. P 1 8 . Steinmann, L . ; Dittrich, G . ; Karmann, A . & Zweifel, P . (2004). Measuring and comparing the (in)e.fjiciency of German and Swiss hospitals. Mimeo. Stigler, G. J. ( 1 976, March) . The Xistence ofX-efficiency. American Economic Review. 66 ( 1 ). 2 1 3-2 1 6 . Sykes, T. ( 1 996). The Bold Riders. (Second Edition). St Leonards, NSW: Al ien & Unwin Thanassoul is , E . ; Boussofiane, A . & Dyson, R. G. ( 1 996). A comparison of data envelopment analysis and ratio analysis as too ls for performance measurement. Omega. 24 (3 ) . 229-244. Thorp, C . (2003, June). Financial intermediation beyond the banks: recent developments . Reserve Bank of New Zealand Bulletin. 66 (2) . 1 8-28 . Thorp, C. & Ung, B. (2000. June). Trends in household assets and l iabil ities since 1 978 . Reserve Bank of New Zealand Bulletin. 63 (2) . 1 7-37 . Thral l . R. M. ( 1 999). What i s the economic meaning of FDH? Journal of Productivity Analysis. 1 1 . 243-250 . To, H.M. & Tripe, D . (2002) . Factors influencing the performance of foreign-owned banks in New Zealand. Journal of International Financial Markets. Institutions and Money, 12, 34 1 -3 5 7 . Tone, K. (200 1 ) . A slacks-based measure of effic iency in data envelopment analysis . European Journal of Operational Research. 1 30. 498-509. Tone, K. (2002). A slacks-based measure of super-efficiency in data envelopment analysis. European Journal of Operational Research. 1 4 3 . 32-4 1 . Tortosa-Ausina, E . (2002a). Financial costs, operating costs, and specialization of Spanish banking firms as d istribution dynamics . Applied Economics. 34. 2 1 65- 2 1 76. Tortosa-Ausina, E. (2002b). Exploring efficiency differences over time in the Spanish banking industry. European Journal of Operational Research. 1 39 . 643-664. 1 78 Tortosa-Ausina, E . (2002c). Bank cost efficiency and output specification. Journal of Productivity Analysis. 1 8 . 1 99-222. T011osa-Ausina, E. (2004). An alternative conditioning scheme to explain efficiency differentials in banking. Economics Letters. 82. 1 47- 1 55 . Tripe, D . W. L . ( 1 997, January). New Zealand banks in the September quarter, 1 996. Palmerston North: Centre for Banking Studies. Tripe, D . W. L . ( 1 998). Cost to Income Ratios in Australasian Banking. Available at http:/ I centre-banking -stud ies.massev .a c . nz/researc h prog.asp Tripe, D . W. L (2002). Factors affecting New Zealand bank interest margins. Paper presented at the 1 5111 Australasian Banking and Finance conference. Avai lable at http://centre-banking-studies.massey.ac.nz/mainh.htm Tripe, D. W. L . (2003) . Trends in New Zealand bank efficiency over time. Applied Econometrics and International Development. 3 ( 1 ) . 55-80. Tripe, D . W. L. (2005) . Book review (of Holmes, F . W., "The thoroughbred among banks in New Zealand: banking in new territory, 1 984-2000). Forthcoming in Australian Economic History Review. Tripe, D. W. L. (Various). The pe1:[ormance of New Zealand banks in the . . . . . . quarter (Series). Avai lable at http ://centre-banking-studies .massey.ac.nz/mainh.htm. Tripe, D, & Matthews, C. (2003) . The international expansion of Australian banks. In Lonnborg, M : Olsson, M . ; Rafferty, M . & Nalson, I (eds), Money and Finance in Transition. Huddinge, Sweden: Sodertorns hogskola. (pp 1 5 5 - 1 80). Tulkens, H. & Vanden Eeckaut, P . ( 1 995) . Non-parametri c efficiency, progress and regress measures for panel data: methodological aspects. European Journal of Operational Research. 80 . 4 74-499. Walker, G. ( 1 998, March) . Economies of scale in Australian banks 1 978 - 1 990. Australian Economic Papers. 37 . 7 1 -87 . Wei l l , L . (2004) . Measuring cost efficiency in European banking: a comparison of frontier techniques. Journal of Productivity Analysis. 2 1 . 1 33 - 1 52 . Wheelock, D . C . & Wilson, P . W . ( 1 995, July/ August). Evaluating the efficiency of commercial banks: does our view of what banks do matter? Review (Federal Reserve Bank ofSt Louis). 39-52. Wheelock, D . C & Wilson, P . W. ( 1 999, May). Technical progress. inefficiency, and productivity change in U . S . banking, 1 984- 1 993 . Journal of Money, Credit, and Banking. 3 1 (2) . 2 1 2-234. 1 79 Wheelock, D . C . & Wilson, P. W. (200 1 ). New evidence on returns to scale and product mix among U. S. commercial banks. Journal of Monetary Economics. 47 . 653-674. White, L. J . ( 1 997). The lessons of the 1 980s for bank regulation : an overview of the overview. In History ofthe Eighties: Lessons for the Future. (Volume 2 ) . Washington DC: FDIC. Pp 7 1 -78 . Worthington. A. C . ( 1 998a). The determinants of non-bank financial institution efficiency : a stochastic cost frontier approach. Applied Financial Economics. 8 . 279-287. Worthington. A. C . ( 1 998b). Effic iency in Austral ian bui lding societies: an econometric cost function approach using panel data. Applied Financial Economics. 8 . 459-467. Worthington, A. C. ( 1 999a). Measuring technical efficiency in Australian credit unions. The Manchester School. 67 (2) . 23 1 -248 . Worthington, A. C . ( 1 999b ) . Malmquist indices of productivity change in Austral ian financial services. Journal of International Financial Jvfarkets, Institutions and Money. 9. 303-320. Wm1hington, A. C. (2000). Cost efficiency in Australian non-bank financial institutions: a non-parametric approach. A ccounting and Finance. 40. 75-97 . Wonhington, A. C . (200 1 ) . Efficiency in pre-merger and post-merger non-bank financial institutions. Managerial and Decision Economics. 22. 439-452 . Yue, P . ( 1 992, January/February). Data envelopment analysis and commercial bank performance : a primer \;vith applications to Missouri banks . Reriew (Federal Reserve Bank ofSt Louis). 3 1 -45 . 1 80