For risk management purposes within banks, it is important to assess the credit risk of customers taking loans and monitoring if the credit risk remains acceptable to a bank. Specifically, it is important for banks to assess if their customers are likely to default on their obligations, meaning they are unable to pay back and possibly leading to a loss for the bank. However, the bank would like to be able to assess the credit risk of its clients before a default occurs, to take mitigating actions at an early stage possible. Since credit risk needs to be assessed prior to default, modelling is involved to determine the credit risk of individual clients.
In general, there are three risk measures that are typically used for the assessment of credit risk: Probability of Default (PD), Loss Given Default (LGD) and Exposure at Default (EAD). The PD quantifies the probability that a non-defaulted customer defaults over a particular horizon (typically 1 year). The LGD quantifies the loss to the bank, given that a customer defaults within a particular horizon (typically aligned with the horizon for PD). The EAD quantifies the Exposure of the client to the bank at the default moment, which might be unknown in case of undrawn credit lines.
This page focusses on PD to assess credit risk. In general, predicting PDs for loan takers in banks requires statistical modelling or expert-based models. One uses historical information on customers prior to default (for instance information on delinquency) to predict whether a default event will occur within the risk horizon. This results in a relative ranking in customers, where customers with a higher PD should have a higher likelihood to default ensuring risk differentiation (meaning the model has discriminatory power). Furthermore, the PD model should be calibrated to ensure accurate predictions of PDs (calibration accuracy).
Aside from the business case of PD models, regulation also prescribes the use of PD models. Under the Internal Ratings-Based (IRB) approach for credit risk capital requirements prescribed in the Basel guidelines, PD models need to be developed. Furthermore, under IFRS 9, PD models are widely used for Loan Loss Provisioning (LLP) purposes. Under the IRB approach, the risk horizon is 1 year, and estimated PDs should be Through-the-Cycle (TtC) meaning that they are calibrated on long-run average 1-year default rates. Under IFRS9, the risk horizon can be multiple years, and estimated PDs should be Point-in-Time (PiT) meaning that estimated PDs follow macro-economic developments.
RiskQuest has developed a multitude of PD models for various Dutch banks. We have acquired expertise on PD models used for internal risk management, PD models for credit risk capital calculations under the IRB approach and PD models under IFRS 9 used for LLP purposes. RiskQuest combines the technical expertise to build a statistically optimal model, the business experience to ensure that the model leads to intuitive outcomes and the knowledge of relevant regulations to satisfy compliance. Due to its experience in PD model development, RiskQuest is capable of facilitating the development and approval processes related to PD models.
RiskQuest also has experience validating PD models for various banks. Independent validation ensures that the PD models are fit for purpose and producing correct results. Additionally, the validation process can also be used to identify improvements in the PD model. Furthermore, PD models are compared to the market standard to identify any shortcomings.
To summarize, RiskQuest provides all tools to develop a PD model that satisfies all business and regulatory requirements. Furthermore, RiskQuest’s experience and knowledge of the market practices makes RiskQuest the ideal partner to perform independent validations of (re-)developed PD models. Here you can find more about our knowledge and experience on all things related to Credit Risk.