A Scorecard for Pay/No Pay Decision-Making in the Retail Banking Industry

The financial sphere covers a wide set of pivotal areas in the actual society, where credit decision-making assumes great relevance. In the retail banking industry, analysts’ judgment prevailed in credit decision-making, without alternative, for long time. In the last decades the emergence of classification methods took place in this area. In the sixties, the expansion of credit cards has lead to the development of appropriated models. Nowadays, the banking sector accelerates the implementation of new models, fitted to the type of credit and segments of customers and operative efficiency, converging to the Basel II Accord requirements. The ubiquity of digital communications has led to the generalization of online payments in individuals’ Demand Deposit Accounts (DDAs). Retail banks have to assure a prompt answer for those payment requests, which can be millions a day. When the DDA has not sufficient balance the bank has to decide whether to pay the debit transaction (a pay/no pay decision-making). This pay/no pay decision must be performed by the end of the day, to fit the Financial Net Settlement System service level’s requirements. Optimizing this decision-making entails the decision to be consistent, objective and fast, with the minimum of mistakes and losses. Currently at a retail Portuguese bank, most of the pay/no pay decisions are automatically managed with behavioural models and models that attempt to reproduce human judgement, while critical decisions are left for manual assessment. However, the automatic behavioural scoring models in use were developed for predicting default in a six-month period or more; furthermore, to keep the implementation straightforward, they do not entirely emulate human reasoning. Therefore, some distinctive features of the problem are not materialized on them. Both customers’ income and payments cycles take one month to be completed. Hence, if a ‘pay’ decision is made, it is expected that the DDA cures within 30 days. This led us to consider the development of a specific model to classify short-term credit risk for mass-market customers of this retail bank. In this work several classification models are built on this assumption. We start by assessing binary scorecards, assigning credit applicants to good or bad risk classes according to their record of defaulting. The detection of a critical region between typical good and bad risk classes, together with the opportunity of manually classifying some of the credit applicants, led us to develop a tripartite scorecard, with a third output class, the review class, in-between the good and bad classes. With this model, 87% decisions can be made automatically, which compares favourably to the actual scorecards, with an automation of 79%.