IRB Overview

The Standardized VS Internal Model Based (IRB) Approach

According to the BIS, it is the main incentive of IRB approach is to provide a more risk-sensitive approach to measuring credit risk. It rewards stronger and more accurate risk measurement by using banks’ own internally generated measures. The idea applies to credit Risk, operational Risk and market Risk. However, the name and methodology both are called differently. Advanced Measurement Approach (AMA) and Internal Model Approach (IMA) is for operational risk and market risk respectively.

Under IRB approach, the banks develop risk model to predict the expected loss (EL). It therefore leads to estimate the losses above expected level as unexpected loss (UL). The function of capital is to protect against any unexpected loss. Higher level of capital reserve provides more buffer to protect a bank’s debt holders against uncertainty over years.
The IRB approach is composed of:
• Probability of default (PD) denotes the average percentage of obligor default per rating grade in one year.
• Loss given default (LGD) represents percentage of exposure the bank might loses given obligor default.
• Exposure at default (EAD) denotes the estimate of the amount outstanding or exposure if the obligor defaults happen

There are two sub-types of IRB: Foundation IRB (F-IRB) and Advanced IRB (A-IRB). The F-IRB banks estimate probability of default (PD), loss given default (LGD) and exposure at default (EAD) for all the retail exposures. For non-retail segment, EAD and LGD are decided and set by the regulators. The A-IRB banks develop their own internal models to estimate PD, LGD and EAD.

Comparison between Standardized and IRB

The following diagram illustrates the overall process of how to calculate capital under different approaches:

Implementation Snapshot

Choice of model

As the IRB model is developed by the banks, the choice of model can depend on data availability, credit assessment and policy, infrastructure of rating system as well as model governance. There are 3 main types of model:
• Heuristic model applies a set of judgmental rules in order form a knowledge structure. It further constitutes a set of credit rating methodology. One of common examples is via rating questionnaires prepared by credit expert. Those questionnaires will be answered by relevant staffs like credit assessor or loan approver. All of the inputs summed up into points representing credit worthiness per characteristics.
• Statistical modelling is a mathematically way to estimate reality under a variety of assumptions. The result (ie estimation) can apply to make predication. Credit scoring can be based on logistic regression to predict customer credit rating for bank decisioning.
• Causal models derive direct analytical links to creditworthiness based on financial theory. Cash flow model is primarily on the future cash flows arising from the assets financed. The economic value of the company can be calculated to determine the possibility of fault.
All the implementation must follow the regulator requirements.

Practicality

Model practicality should be carefully considered during the concept phase of IRB implementation. For some banks, the loan process may be simple and highly judgmental. The data preparation can be difficult as the system does not record at various point. The information is usually stored at multiple systems as required to link across different storage.
The PD model is to predict one year horizon so that minimum one year data is collected with reasonable sample size. LGD arguably requires data over a long period of time including down turn economy. EAD is measured at point in time. The risk profile requires homogenous data prior to modelling.

Attestation

IRB approach is required by support of senior management and the board of directors. The understanding is a key success to implement. A significant number of new challenges includes model validation techniques, cost benefit between approaches, adaptability of usage as well as best practices within organization. It includes extensive requirements for model performance and monitoring to govern the role and responsibility. Some key areas can be highlighted below:
• Model scope, governance and implementation plan;
• Data availability
• Model design, assumption and limitations
• Performance analysis if the model exists
• Model process including software, resource and timeline
• Model application
• Model Validation from internal and external
• Post implementation monitoring

According to national authorities, almost half of capital requirement are calculated using IRB approach. Only 24% of institutions deploy the IRB models. As IRB adoption is not mandatory in Europe, It is likely because internal models are usually developed by middle to large sized banks.

Source: National Authorities, end-2013