MEASURING BANK BRANCH EFFICIENCY USING DATA ENVELOPMENT ANALYSIS: MANAGERIAL AND IMPLEMENTATION ISSUES

ABSTRACT

Data Envelopment Analysis (DEA) is used to assess the efficiency of branches of a major Canadian bank ("Canbank"). First, a DEA model of American branch bank efficiency is utilized to build a model with Canbank data, then model outcomes are compared to the outcomes of the US study and the differences explained. Subsequently, the model is revised to represent the particular circumstances of Canbank's western, urban branches. Differences in outcomes between the revised model and the initial model are identified, then analysis with the revised model is conducted. Observations on implementing DEA in a work environment are also provided.

Keywords: Data envelopment analysis, efficiency measurement, linear programming.

(ProQuest Information and Learning: ... denotes formulae omitted.)

1. INTRODUCTION

Since the seminal contribution by Charnes, Cooper and Rhodes (1978), data envelopment analysis (DEA) has been widely applied for efficiency measurement. In particular, it has been applied in many countries for measuring efficiency in branch banking. Cooper, Seiford and Tone (1999) explain DEA modeling in detail, while Berger and Humphrey (1997) survey DEA studies applied to financial institutions.

Canadian branch banks are presently under intense competition to improve efficiency and transform banking service delivery into networks encompassing traditional branches, automated tellers, telephone banking and the Internet. Because no template exists to guide this transformation, they have experimented by reducing service hours, closing underperforming branches and introducing new and cheaper ways of banking. DEA may be able to help banks improve efficiency in delivering banking services.

In this paper we discuss our application of DEA to a major bank in Canada, which for anonymity we call Canbank. First, we provide background discussion on DEA and brief comments on several Canadian studies. second, we build a Canbank model based upon the US model of Golany and Storbeck (1999) in order to learn how their model might apply to Canbank. Subsequently, we assemble a DEA model from first principles to prepare a customized model for Canbank. Next, for illustrative purposes we exercise our reformulated model to respond to hypothetical claims by branch managers that their efficiency measures are biased or wrong. Finally, we summarize what has been learned from building our DEA models and exercising the reformulated model.

2. BACKGROUND DISCUSSION

2.1 Traditional and DEA Measures of Branch Bank Efficiency

The major measures of branch efficiency presently used by Canbank are all accounting-based: comparison of actual managed operating profit to its budget, ratios such as the productivity measure, and scorecards. These measures do not allow branch performance to be represented in terms of controllable variables and uncontrollable variables, and they measure relative branch performance only along one dimension unless arbitrary weights are assigned to define an aggregate performance measure. To overcome these disadvantages, DEA is used in this study. Ferrier and Lovell (1990) and Berger and Humphrey (1997) discuss the relative advantages and disadvantages of DEA versus econometrics for measuring efficiency. Charnes, Cooper, Lewin and Seiford (1994, p. 8) focus on specific advantages of DEA. As notable advantages for our work, DEA is relatively easy to explain to branch managers and the only assumptions bank staff must accept are that the data are accurate, that their branch performance will be compared to an appropriate reference or peer group of branches, and that they will be assessed on efficiency criteria that are under their control.

2.2 Canadian Studies of DEA Applied to Branch Banking

Published DEA studies of branch banking are numerous but Canadian studies are relatively few. A recent paper by Mclntosh (2002) wrongly asserts that there are none. To place our work into perspective, we review several Canadian studies. The first, Parkan (1987), applies DEA to 35 branches of a major Canadian bank, utilizing six inputs and six outputs. Only eleven branches were inefficient, implying that 69% were efficient. Although the study was pioneering, its results permitted no distinction among efficient branches, allowed few comparisons among inefficient branches and their reference sets of efficient branches, and were not conclusive regarding the sources of the inefficiencies.

Schaffnit, Rosen and Paradi (1997) provide brief discussion of the Canadian banking industry, review the fundamentals of branch banking efficiency measurement, provide a critical review of the literature, and implement DEA models with output multiplier constraints. They also develop a model with input multiplier constraints to measure allocative efficiency and they pay particular attention to the impact of model choice upon the results. Finally, they utilize some statistical tests for their post-hoc analysis.

Alirezaee, Howland and van de Panne (1998) utilize data from 1282 branches of a large Canadian bank to conduct numerical experiments relating DEA results to sample size. They found that the average branch efficiency score varied inversely with the number of DMUs and directly with the total number of inputs and outputs. They also cautioned that using relatively small sample sizes in a model with as few as three inputs and three outputs could lead to a substantial upward bias in efficiency scores.