With the advent technology and data availability more and more financial institutions are opting Advanced Internal Rating Based (A-IRB) approach. This enables them to determine their own estimates of Probability of Default (PD), Loss Given Default (LGD) and Exposure at Default (EAD).
For this approach risk management team should have a clear understanding of the business requirements. They should have complete understanding of input data and application of output.
In the case of Loss Given Default (LGD) this issue becomes critical. It is because LGD is dependent on recovery and data on recovery is quite fragmented and hence unreliable.
Empirically it has been observed that recovery rate (and hence LGD) is dependent on
- The bank’s behavior in terms of debt renegotiation with debtors, compromise and settlements which are country specific.
- The quality of collateral attached to loans
- Firm specific capital structure: Seniority standing of debt in the firm’s overall capital structure, leverage etc.
- Industry tangibility: The value of liquidated assets dependent on the industry of the borrower.
- Macro-economic factors: industrial production, GDP growth, unemployment rate, interest rate and other macro-economic factors have strong influence on LGD.
But the recovery is also dependent on following operational factors:
- Financial Institution’s understanding about the business of the obligor
- Recovery agent and relationship history of the obligor with the financial institution
- Negotiation of restructuring of loans
- Optimal combination of “time of resolution” and price received by selling collateral
- Current microeconomic factors which are impacting both the parties
- Indirect costs which include cost of maintaining the recovery staff etc.
Imagine two default cases of two similar shopkeepers (here their shop is collateral) on a same lane. Recovery of one shop may be better (in terms of time to resolution and recovered amount) just because more parking space is available for customers around it in comparison other one.
Data associated with such specific information is thin hence development of a robust model is a challenging task. It is never guaranteed that two deals with same collateral (also obligor, industry, macroeconomic situation) will result in same recovery. Also there is no exhaustive list variable available on which LGD depends. This leaves scope of subjective judgments in the right measurement and prediction of LGD.
Thus a robust model monitoring framework is mandatory. Monitoring has a more qualitative nature and focuses on portfolio dynamics, quality of the data and use of the model. One can monitor changes at the data level (population stability), loss of model performance, or calibration issues (i.e., by comparing estimated against observed LGD at the pool level).
Because of the subjectivity in LGD measurements, monitoring framework for LGD models need to be taken in a different paradigm.
We provide a list of questions which analyst should be exploring so that he can logically document the assumptions he makes and articulate logical explanations why he is choosing certain real world characteristics in his modeling exercise and ignoring others. Finally this information will help him in setting up a robust model monitoring framework for the developed LGD model.
Business objective related questions
- What are the businesses which the model caters?
- Size of the business?
- Who are the target customers and their ratings? What is their reputation?
- Does the business/sales team have expertise in understanding the business of their customers? How is this understanding reflected in the recovery process?
- Is the business directly lending the end customer or is a third party vendor? Who guarantees the loans? How reliable is the guarantor?
- How is the default defined?
- What are the collaterals?
- What are the underwriting procedures/criteria?
- What are the legal agreements around the loan agreement?
Data Quality related questions
- How can the available modeling data can be triangulated so that its correctness is verified
- How often the data base is updated?
- What are the noise/exclusion/outlier criteria?
Model related questions
Model chosen should have right combination of accuracy and simplicity. Models which are accurate in the given data set may be over fitted and complex. Their performance in the future will be doubtful. Too simple model may over generalize the business problem and may not capture the specific business risks.
Outcome analysis level questions
- Typically LGDs are used for Allowance of Loan and Lease Losses (ALLL), Capital allocation, Stress testing and underwriting. Is the data used for modeling compliant with these requirements?
- How the LGDs are benchmarked? What are the benchmarking criteria?
- Benchmarking is a challenging task, because it should ensure right seniority, rating of the debt.
- What will be the impact on the output LGD on the output of other models which are going to use these LGDs as an input?
A detailed questionnaire can be found at A Questionnaire to Address Subjectivity in LGD Modeling
LGD is one concept in risk management which is dependent not only on objective variables but on subjectivities because recovery varies from business to business, firm to firm, time to time etc. There is no end to this list.
Due to this variation in recoveries we find most varieties in modeling prepositions in LGD modeling. On a given data set one analyst may propose a simple linear regression, another may propose Decision trees, and others may propose Neural Networks, genetic algorithm or any other exotic fuzzy logic based models. Interestingly these analysts may not agree on the choice of variables as well.
In fact LGD modeling is all about business and data understanding. This understanding is important for the model developer to develop a right framework of model and model monitoring.
To develop that understanding he should be prepared to ask questions to gather information so that he can develop right monitoring framework to evaluate an existing model. We provided some list of questions which can help that analyst to clarity around the business requirements and hence develop right risk solution.