Symposium on Risk Management in the Retail Financial Services Sector

This one-day open symposium was held at Imperial College London on the 17th April 2007. There were over 80 delegates from banks, financial institutions and academia. It was led by researchers from Imperial College London, the University of Southampton and the University of Edinburgh, and research in Credit Scoring, modelling Loss Given Default (LGD), plastic card fraud, predicting consumer behaviour, choosing interventions and incorporating economic effects into models was described.

Programme

Presentations

1. New Methods of Estimating LGD for Consumer Loans
   Jonathan Crook, Tony Bellotti, Galina Andreeva and Jake Ansell
2. Loss Given Default process modelling [Slides (PDF)]
   Anna Matuszyk, Lyn Thomas and Christopher Mues
3. Predicting consumer behaviour with ATMs [Slides (PDF)]
   Adam Brentnall, Martin Crowder and David Hand
4. Plastic Card Fraud Detection using Peer Group analysis [Slides (PDF)]
   David Weston
5. Incorporating macroeconomic variables into consumer credit analysis [Slides (PDF)]
   Tony Bellotti and Jonathan Crook
6. Modelling the Credit Risk of Consumer Loans through Survival Analysis [Slides (PDF)]
   Madhur Malik and Lyn Thomas
7. Product selection in the presence of selectivity bias [Slides (PDF)]
   I-Ding Wu and David Hand
8. Choosing credit limits to maximise Profit: Markov Decision Process approach [Slides (PDF)]
   Meko So and Lyn Thomas
9. Modelling of SME default over different definitions of Financial distress [Slides (PDF)]
   Jake Ansell, Galina Andreeva and Shu Min Lin

Abstracts

1. New Methods of Estimating LGD for Consumer Loans
   Jonathan Crook, Tony Bellotti, Galina Andreeva and Jake Ansell

   This paper will review approaches to modelling corporate credit risk to see what lessons can be learned for modelling LGD for consumer accounts. In particular, we discuss structural models, both first generation Merton models and second generation models and also reduced form intensity models. We then review the empirical studies which have tried to explain and predict LGD at the level of the corporate borrower. Finally we propose some methods for estimating LGD at the level of the individual consumer account.

2. Loss Given Default process modelling
   Anna Matuszyk, Lyn Thomas and Christopher Mues

   Loss Given Default is one of the components used to calculate credit risk capital. LGD can be defined as a mix of random events and decisions made by the lender to decide what kind of collection strategy should be used. These decisions affect the results and need to be separated out using modelling collection process. As well as this it is necessary to estimate the uncertainty on how much can be recovered at each step of the strategy. We describe a two stage model for personal loans to predict LGD recovered in house. Results received and the structure of the model will be presented.

3. Predicting consumer behaviour with ATMs
   Adam Brentnall, Martin Crowder and David Hand

   Models of consumer behaviour based purely upon empirical relationships in data can perform well in the short term, but often degrade rapidly with changing circumstances. Superior longer-term performance can sometimes be attained by developing models for the deeper processes underlying the consumer behaviour, and using them within a customer management strategy. This talk will describe the development of random-effects models for ATM (Automated Teller Machine) withdrawals. Preditive performance will be assessed using a real data set.

4. Plastic Card Fraud Detection using Peer Group analysis
   David Weston

   Peer group analysis is an unsupervised method for monitoring behaviour over time. For each credit card account a 'Peer Group' of accounts is determined; These are accounts that exhibit similar behaviour. As time evolves, it is assumed the behaviour of an account is tracked by those accounts in its Peer Group. An account whose subsequent behaviour deviates strongly from its Peer Group is considered to have behaved anomalously and is flagged as potentially fraudulent. Since it is unlikely that an account will be tracked indefinitely by its Peer Group, we investigate how tightly accounts are tracked by peer groups. The real-time practical issues of both summarizing the behaviour of an account and comparing accounts are also described.

5. Incorporating macroeconomic variables into consumer credit analysis
   Tony Bellotti and Jonathan Crook

   Survival analysis can be applied to build models for time of default on debt. In this paper we report an application of survival analysis to model default on a large data set of credit card accounts. We show that survival analysis is competitive for prediction of default in comparison with logistic regression. We explore the hypothesis that probability of default is affected by general conditions in the economy over time. These macroeconomic variables cannot ordinarily be included in logistic regression models. However, survival analysis provides a framework for their inclusion as time-varying covariates. Various macroeconomic variables, such as interest rate and unemployment index, are included in the survival model as time-varying covariates. We show that inclusion of these indicators does affect probability of default and provide a statistically significant improvement in predictions of default on an independent test set.

6. Modelling the Credit Risk of Consumer Loans through Survival Analysis
   Madhur Malik and Lyn Thomas

   One of the issues that the Basel Accord highlighted was that though techniques for estimating the probability of default and hence the credit risk of loans to individual consumers are well established, however there were no models for the credit risk of portfolios of such loans. Motivated by the reduced form models for credit risk in corporate lending, we will seek to exploit the obvious parallels between behavioural scores and the ratings ascribed to corporate bonds to build consumer lending equivalents. We incorporate both consumer specific ratings and macroeconomic factors in the framework of Cox Proportional Hazard models. Our results show that default intensities of consumers are significantly influenced by macro factors. Such models then can be used as the basis for simulation approaches to estimate the credit risk of portfolios of consumer loans.

7. Product selection in the presence of selectivity bias
   I-Ding Wu and David Hand

   In many circumstances it is desirable to select t he banking product for a new customer to yiel d the greatest profit. If we regard the assignment of each available product to the customer as an action then the task is to compare the utility between available actions and select an optimal one. The utility for each action is estimated using information from customers who were previously assigned the corresponding product. Typically, previous product assignments were not made randomly. That is, the product selection for all previous customers was deterministically related to standard customer descriptors. In such situations selectivity bias might occur and result in biased estimation. Biased estimation has the potential to lead to sub-optimal product selection for new customers. We first demonstrate situations where such bias occurs and then describe standard adjustment methods. The performance of these methods is limited by an arbitrary assumption. We introduce two adjustment approaches that overcome this problem and further improve the performance.

8. Choosing credit limits to maximise Profit: Markov Decision Process approach
   Meko So and Lyn Thomas

   Spending on credit cards remains a vital part of the financial banking industry. Lenders have recognized that maximizing profit is as important as minimizing default rates. They have also recognized that their operating decisions are crucial in determining how much profit is achieved from a card. Foremost among these operating decisions is the credit limit chosen on the card. In this study, we applied a Markov Decision Process (MDP) model to develop optimal credit limit policies to maximize the profit gained from customers. Results for a case study are discussed in the presentation.

9. Modelling of SME default over different definitions of Financial distress
   Jake Ansell, Galina Andreeva and Shu Min Lin

   Risk associated with lending to small businesses, which forms the subject of this presentation, shares the features of both retail and corporate sectors, and this has been recognised by Basel II provisions. Driven by Basel II, the presentation introduces a number of risk-rating models for the U.K. small businesses using an accounting-based approach, which employs financial ratios to distinguish between defaulting and non-defaulting firms and to predict corporate bankruptcy. A common problem with default prediction consists in a small number of bankruptcies or real defaults available for analysis. The presentation contemplates adopting less strict definitions of default by considering different levels of financial distress and investigates the impact of different definitions on the choice of predictor variables and predictive accuracy of the model. The analysis demonstrates that each definition considered leads to a different model. The presentation describes the variable composition of these models, compares their predictive accuracy and comments on the suitability of each approach used.

Further Information..

Is available from the Symposium website qfrmc.imaa.ic.ac.uk/qfrmc/symposium .