Decision making in an uncertain environment: An application of ROC analysis for credit scoring in the mobile telephone m

  • PDF / 429,743 Bytes
  • 14 Pages / 597.972 x 767.348 pts Page_size
  • 78 Downloads / 200 Views

DOWNLOAD

REPORT


Marc Hoogenberg works as a business consultant at Paul Postma Marketing Consultancy in the Netherlands. He provides solutions and training in data analysis and data modelling problems for customer relationship management (CRM).

Sander auf dem Brinke works as a business consultant at Paul Postma Marketing Consultancy in the Netherlands. He advises in analytical CRM projects and has several years of experience in a variety of sectors, including telecommunications.

Abstract This paper describes the use of relative operating characteristics (ROC) analysis for decision making under uncertain circumstances. It is presented in the framework of credit scoring for a mobile phone operator.

Sander auf dem Brinke Paul Postma Marketing Consultancy B.V., Edisonbaan 14A, 3439 MN, Nieuwegein, The Netherlands. Tel: ⫹31 (0)30 2598460; Fax: ⫹31 (0)30 2598484; e-mail: sander.aufdembrinke@ ppmc.nl

INTRODUCTION Now that the cards have been dealt, and the market share of mobile telephone operators has converged to a steady state, their marketing focus is shifting from the brute force acquisition of new customers to a more refined strategy. It is becoming more and more important to focus on customers (potentially) generating high value. A problem that perpetuates, however, is that of ‘noncredible’ customers. Every year, customers who can, or will, no longer pay their monthly subscription fee and/or telephone bill are a large source of costs for mobile telephone operators. Within the data mining community, the problem of predicting and detecting (deliberate) refusal of payments has become a major topic in applied research. It is often called the credit-scoring problem. This paper discusses the use of relative operating characteristics (ROC) analysis

䉷 Henry Stewart Publications 1741–2447 (2004) Vol. 11, 3, 241–254

in the setting of credit scoring for a mobile phone operator. It shows that ROC analysis is a particularly suitable instrument for decision making in uncertain situations like the mobile telephone market. PROBLEMS IN CREDIT SCORING Credit scoring can be applied at two stages.1 The behaviour of existing customers can be analysed to detect credit risk. On the other hand, at the application stage, new potential customers (prospects) can be assessed in order to predict credit risk. The authors will concentrate on the latter in this paper. With application scoring, there is a basic choice whether to accept a prospect as a customer, or to decline him or her, based on some credit risk score. When the risk is too high, the applicant is rejected. As in most practical

Database Marketing & Customer Strategy Management

241

Hoogenberg and auf dem Brinke

applications, many decision-making parameters are uncertain or even unknown, and most of the parameters vary over time. Such parameters include the cost of making a wrong decision (and the profit for making a right one), and assumptions about the underlying patterns of the data. Usually a solution to the problem is accompanied by a sensitivity analysis to changes in the underlying a