Credit Risk Scoring Models
By Gabriele Sabato
Credit scoring models play a fundamental role in the risk management practice at most banks. They are used to quantify credit risk at counterparty or transaction level in the different phases of the credit cycle (e.g. application, behavioural, collection models). The credit score empowers users to make quick decisions or even to automate decisions and this is extremely desirable when banks are dealing with large volumes of clients and relatively small margin of profits at individual transaction level (i.e. consumer lending, but increasingly also small business lending). In this article, we analyze the history and new developments related to credit scoring models.
We find that with the new Basel Capital Accord, credit scoring models have been remotivated and given unprecedented significance. Banks, in particular, and most financial institutions worldwide, have either recently developed or modified existing internal credit risk models to conform with the new rules and best practices recently updated in the market.
Moreover, we analyze the key steps of the credit scoring model’s lifecycle (i.e. assessment, implementation, validation) highlighting the main requirement imposed by Basel II. We conclude that banks that are going to implement the most advanced approach to calculate their capital requirements under Basel II will need to increase their attention and consideration of credit scoring models in the next future.
Credit scoring models play a fundamental role in the risk management practice at most banks. Commercial banks’ primary business activity is related to extending credit to borrowers, generating loans and credit assets. A significant component of a bank’s risk, therefore, lies in the quality of its assets that needs to be in line with the bank’s risk appetite. In order to manage risk efficiently, quantifying it with the most appropriate and advanced tools is an extremely important factor in determining the bank’s success.
Credit risk models are used to quantify credit risk at counterparty or transaction level and they differ significantly by the nature of the counterparty (e.g. corporate, small business, private individual). Rating models have a long term view (Through The Cycle) and have been always associated with corporate clients, financial institutions and public sector. Scoring models, instead, focus more on the short term (Point In Time) and have been mainly applied to private individuals and, more recently, extended to small and medium sized enterprises (SMEs) .
In this article, we will focus on credit scoring models giving an overview of their assessment, implementation and usage. Since the 1960s, larger organizations have been utilizing credit scoring to quickly and accurately assess the risk level of their prospects, applicants and existing customers mainly in the consumer lending business. Increasingly, midsize and smaller organizations are appreciating the benefits of credit scoring as well. The credit score is reflected in a number or letter(s) that summarizes the overall risk utilizing available information on the customer. Credit scoring models predict the probability that an applicant or existing borrower will default or become delinquent over a fixed time horizon. The credit score empowers users to make quick decisions or even to automate decisions and this is extremely desirable when banks are dealing with large volumes of clients and relatively small margin of profits at individual transaction level.
Credit scoring models can be classified into three main categories: i.e. application, behavioural and collection models, depending on the stage of the consumer credit cycle in which they are used. The main difference between them lies in the set of variables that are available to estimate the client’s creditworthiness, i.e. the earlier the stage in the credit cycle the lower the number of specific client information available to the bank. This generally means that application models have a lower prediction power than behavioural and collection models.
Over the last 50 years, several statistical methodologies have been used to build credit scoring models. The very simplistic univariate analysis applied at the beginning (late 1950s) was replaced as soon as academic research started to focus on credit scoring modeling techniques (late 1960s). The seminal works in this field of Beaver  and Altman  introduced the multivariate discriminant analysis (MDA) that became the most popular statistical methodology used to estimate credit scoring models until Ohlson , for the first time, applied the conditional logit model to the default prediction’s study. Since Ohlson’s research (early 1980s), several other statistical techniques have been utilized to improve the prediction power of credit scoring models (e.g. linear regression, probit analysis, Bayesian methods, neural network, etc.), but the logistic regression remains still the most popular method. Lately, credit scoring gains new importance with the New Basel Capital Accord.
The so called Basel II replaces the current 1988 Capital Accord and focuses on techniques that allow banks and supervisors to evaluate properly the various risks that banks face. Since credit scoring contributes broadly to the internal risk assessment process of an institution, regulators have enforced more strict rules about model development, implementation and validation to be followed by banks that wish to use their internal models in order to estimate capital requirements. The remainder of the article is structured as follows. In Section 2, we review some of the most relevant research related to credit scoring modeling methodologies. In Section 3, following the model lifecycle structure, we analyze the main steps related to the model assessment, implementation and validation process.