Comparison procedure of predicting the time to default in behavioural scoring

Abstract

The paper deals with the problem of predicting the time to default in credit behavioural scoring. This area opens a possibility of including a dynamic component in behavioural scoring modelling which enables making decisions related to limit, collection and recovery strategies, retention and attrition, as well as providing an insight into the profitability, pricing or term structure of the loan. In this paper, we compare survival analysis and neural networks in terms of modelling and results. The neural network architecture is designed such that its output is comparable to the survival analysis output. Six neural network models were created, one for each period of default. A radial basis neural network algorithm was used to test all six models. The survival model used a Cox modelling procedure. Further, different performance measures of all models were discussed since even in highly accurate scoring models, misclassification patterns appear. A systematic comparison ‘3 + 2 + 2’ procedure is suggested to find the most effective model for a bank. Additionally, the survival analysis model is compared to neural network models according to the relative importance of different variables in predicting the time to default. Although different models can have very similar performance measures they may consist of different variables. The dataset used for the research was collected from a Croatian bank and credit customers were observed during a 12-month period. The paper emphasizes the importance of conducting a detailed comparison procedure while selecting the best model that satisfies the users’ interest.

Introduction

Credit scoring is a process of determining how likely applicants are to default with their repayments (Hand & Henley, 1997). It includes various decision models and tools that assist credit managers in making credit decisions. The aim is to assess the risk of default associated with a credit product/decision. Credit scoring models can be divided into two types: (i) credit scoring in targeting and application, which deals with new applicants and (ii) credit scoring in managing existing accounts called behaviour scoring. Behaviour scoring models are used for making on-going decisions on open accounts. It is used to describe the behaviour of existing customers by using behavioural scoring variables and also to predict future purchasing behaviour or credit status of existing customers (Hsieh, 2004). They are based on dynamic account performance information and are used to manage limits during the lifetime of an account, to increase account usage, to set risk-based collections and recovery strategies, to retain future profitable customers, in retention and incentive programs, to predict accounts that are likely to close, become inactive or settle early, to identify short-life accounts, to predict future response and help to improve targeting of offers, to pre-screen mailing lists of new customers, to gain market share and long term profitable accounts, to optimize telemarketing operations, to predict fraudulent activity, and to manage recovery debt (Hand and Henley, 1997, McNab and Wynn, 2000). Behaviour scoring models take into account the current and recent behavioural data of a client (consumer, corporate, and SME) and predict future risk. They can also be built to predict other dimensions of future behaviour, such as revenue, propensity, usage, attrition, contractibility, and fraudulent activity. In such a way they can influence decisions across the whole credit cycle (McNab & Wynn, 2000).

With the inclusion of the dynamic component, credit scoring modelling is getting more and more important in default prediction due to several reasons (Baesens, Gastel, Stepanova, Poel, & Vanthienen, 2005): (i) the ability to compute the client’s profitability over a lifetime; (ii) the ability to estimate the default levels over time which enables debt provisioning; (iii) the ability to decide upon the term of the loan; and (iv) the ability to incorporate changes in economic conditions. If a bank can predict default times for different periods (30, 60, 90 days, etc.) for each client, it enables it to take certain actions in order to prevent undesirable behaviour and therefore to protect itself from potential borrowers with high default risks in timely manner (Lim & Sohn, 2007). The model for predicting ‘when’ may provide financial companies with an estimate of the default levels over time (Noh, Roh, & Han, 2005).

In 2005 the Bank for International Settlement produced a revised framework of the International convergence of capital measurement and capital standards – Basel 2. According to this new accord, banks are allowed to determine their own calculation of risk parameters, such as probability of default, loss given default, exposures at default, and maturity in order to calculate risk weighted assets. Probability of default is derived from application credit scoring for new clients, and from behavioural scoring for existing clients. This makes credit scoring modelling even more important than ever before. Fifth quantitative impact study (BIS, 2006) shows that banks in the EU are expected to reduce capital requirements up to 26.6% due to the new accord if they use internal ratings-based (IRB) approach. This is a great motivation for banks to develop their own credit risk models. The better models they have the greater advantages they will accomplish. The new accord influences consumer credit modelling in three ways (Thomas, Oliver, & Hand, 2005): (i) minimum requirements that a bank has to fulfil in order to use the IRB approach; (ii) validation of risk parameters; and (iii) models for credit risk portfolio.

The paper deals with personal open-end accounts where clients can make different kinds of payments to the accounts and withdrawals from the accounts, as well as use revolving credits. The main purpose of the paper was to model time to default using survival analysis and neural networks, as well as to establish a procedure for selecting one model over another according to model accuracy and its efficiency for the bank. In order to accomplish these purposes, the models predicting defaults in 30, 60, 90, 120, 150, and 180 days have been created using a dataset collected from a Croatian bank. Separate neural network models were designed for each period of default using the radial basis function algorithm. Survival analysis that deals with censored observations is used to estimate hazard functions for each period of default. On the basis of the results of all developed models, a selection procedure is suggested and the best model is discussed.

The structure of the paper is as follows. A description of variables and data is given in the next section. Modelling and results section provides a description of methodology used including survival analysis and neural networks, followed by the results of survival analysis and neural networks as well as their comparison. Finally, some important features of credit scoring modelling are discussed.