Managing Credit Risk for Retail Low-Default Portfolios
By Gabriele Sabato
“My goal is to analyze a possible realistic methodology to develop scoring or rating systems for the retail portfolios where the number of defaults is low or equal to zero.“
Low-default portfolios (LDPs) can be defined as portfolios where the bank has no or a very low level of defaults and is therefore unable to estimate and validate probability of default (PD), loss given default (LGD) and exposure at default (EAD) on a basis of a proven statistical significance.
Several types of portfolios may have low numbers of defaults due to different reasons: for example, portfolios generally considered to be low risk (such as exposures to sovereigns, banks, insurance companies or highly rated corporates) or commonly small in size, either globally or at an individual bank level (such as project finance or shipping). However, most of the literature does not consider retail portfolios in this list or considers only retail mortgages.
Unfortunately, in the real world, banks often have to face the problem of relatively sparse default data for many retail portfolios. Contingency issues (e.g. banks are recent market entrant for a given retail portfolio or have started to collect data only from a small time period) or the nature of the product itself (e.g. mortgages) are the most common reasons. Considering that retail banking is generally based on transactional data (“hard” information, as Berger (2006) observes) and needs advanced credit risk management tools to be managed in an efficient way, the lack of internal data to develop meaningful credit risk models must be regarded as more dangerous for retail portfolios than for non-retail ones.
Although retail mortgages, one of the best examples of low-default portfolio, are commonly considered a low-risk product, they usually represent a very high percentage (often more than 80%) of bank retail assets. Indeed, banking organizations should pay a special attention to the way they manage this business, either in the application or in the behavioural process, since any kind of inefficiency (such as the lack of a scoring system or the use of a low quality one) can have significant effects on the overall bank profitability.
Financial institutions should be aware that credit risk management for retail LDPs is becoming a strategic issue in general and for Basel II purposes in particular. The new Basel Capital Accord and concerns raised by the industry that the lack of sufficient statistical data and the resulting difficulty in backtesting risk parameters will result in LDPs being excluded from the IRB treatment have caused a special attention to this topic. Many recent studies, mainly focusing on the bank’s wholesale portfolio, conclude that it seems inconsistent with the spirit of the new Basel Accord, to exclude LDPs from the IRB treatment only because they have suffered so few defaults. However, I believe that managing LDPs in a more efficient way, using scoring systems based as much as possible on internal data, will likely be the next challenge for banks in order to improve their internal efficiency and profitability more than only to be allowed to apply the Basel II IRB approach.
The prime objective of this paper is to present a new technique that allows banks to use internal data to develop default prediction models even if the number of defaults in the selected retail portfolio is low or equal to zero. In LDPs, the dependent variable (if a regression analysis is used) or the variable to define the different groups (if the multivariate discriminant analysis is used) cannot be defined due to the lack of information about one group, defaulted clients. Usually, in the past, banking organizations have solved the problem by using generic scoring models (expert scorecards based on subjective weights or developed on pooled data) or having their processes based on simple policy rules, all manually checked by the employees. Both these solutions are extremely inefficient in a retail context. The performance of generic scorecards is frequently very low and often the selected variables are not able to satisfactorily explain the credit risk of a specific retail product. The use of policy rules to manage retail credit risk is highly cost inefficient considering the large volume and the low profit margin of retail products.
In this study, I focus on retail portfolios containing products for private individual clients, but I believe that a similar methodology can be applied, with some realistic assumption, also to SME clients. Actually, SMEs have been included by banks in the retail segment only recently, often forced by national regulators and to follow the new Basel Capital Accord rules. This means that many SME portfolios have a small history within the bank and contain few or zero defaults, hence we can reasonably consider most of them as LDPs. Moreover, many recent studies (see for example Schwaiger (2002), Saurina and Trucharte (2004), Udell (2004), Berger (2006), Jacobson et al. (2004), Kolari and Shin (2004), Altman and Sabato (2005) and Altman and Sabato (2006)) demonstrate that the use of scoring systems for SMEs clients is a significant strategic and competitive issue for banking organizations in order to achieve internal efficiency and maximize profits linked to the SME business. For this reason, the credit risk management for SME LDPs can be reasonably considered another important challenge for banks.