EDEA: An expert knowledge-based tool for performance measurement

This thesis presents an improved measurement tool for evaluating performance of branches within a major Canadian bank. While there have been numerous previous studies of performance at a branch level, within the banking industry, this study is different in a very significant way: specifically two kinds of data are used to develop the model.

The first type of data is standard transaction data available from any bank. Such data have formed the basis of previous studies. The second type of data, obtained from the site studied, is what can be called classification information, based on branch consultant/expert judgment as to good or poor performance of branches.

The purpose here is to develop an expert knowledge-based version of an existing benchmarking model. Data Envelopment Analysis (DEA), and to show how this tool is applied in the banking industry. To reflect this extension of the basic DEA model, we adopt the acronym EDEA.

Chapter 1 presents the context of the research and briefly describes knowledge acquisition techniques.

Chapter 2 introduces the DEA theory, with its major models, and describes three different discriminant techniques, namely: (1) Logistic regression, which is based on the Maximum Likelihood concept; (2) Discriminant analysis, based on centroids and groups; (3) Goal programming, a powerful extension of linear programming.

Chapter 3 builds classification concepts into the additive DEA model. It demonstrates how DEA measures can be enhanced, by incorporating expert judgement into the structure. This enhancement facilitates variable selection, as part of the modeling exercise. This new methodology is tested using a set of data provided by a major Canadian bank.

Chapter 4 extends the ideas of Chapter 3 to a nonlinear (input-oriented ) DEA model structure. As well, this chapter extends the expert system structure, by adding further knowledge information in the form of a specification of the status (output or input), of a subset of the variables.

Chapter 5 investigates a number of extensions of the models of the two previous chapters. Specifically, an investigation is performed regarding the imposition of certain constraints into the earlier models.

Conclusions are presented in Chapter 6.