Understanding Expected Credit Loss for Enterprises Finance

Objective

The goal of the Expected Credit Loss (ECL) method is to assess credit risk and determine provisions for potential losses within co-lending portfolios. By adhering to IFRS 9 guidelines, the method ensures timely and accurate provisioning based on the risk profile of loans, enhancing the ability to manage and mitigate credit risk effectively.

Abstract

The Expected Credit Loss (ECL) model is a forward-looking framework that assesses credit risk by considering a borrower’s likelihood of default, the exposure at the time of default, and the expected recovery following default. Utilizing historical data, borrower characteristics, and predictive modelling techniques, the ECL method provides precise estimates of potential credit losses. This empowers us to proactively allocate provisions, ensuring compliance with regulatory standards, and enabling more informed, data-driven decision-making.

Method

Definition: ECL is a method of accounting for credit risk that is based on the loss that is likely to occur on a loan or portfolio of loans.

Expected Credit Loss = PD * EAD * LGD

• Probability of Default (PD): Calculated by estimating the forward-looking probability of default for each loan.
• Loss Given Default (LGD): The percentage loss that is expected to occur if the borrower defaults.
• Exposure at Default (EAD): Expected loss for each loan.

Data Preparation

A decision tree model is created where the target variable is Probability of Default (PD)— specifically, whether a borrower defaulted within 12 months or not. The model will segment borrowers based on the likelihood of default. The process works as follows:

1. Root Node Selection: The decision tree selects the most critical factor (e.g., borrower’s credit score) to split the borrowers into two groups—those likely to default and those who are not.
2. Branching: For each group, further splits are made based on other factors like POS to disbursement ratio, age or payment history.
3. Leaf Nodes: At the end of the tree (leaf nodes), predictions are made regarding the likelihood of default for each borrower.

Probability Estimation

Once the decision tree is trained, each borrower is assigned a Probability of Default (PD) based on the leaf node they land in. The PD for each borrower corresponds to the likelihood that the borrower will default within the next 12 months, based on their features. The leaf node where a borrower lands indicates their risk level: If a borrower ends up in a leaf where many others defaulted, a higher PD is predicted. If they end up in a leaf where defaults were rare, a lower PD is predicted.

Interpretation of the Decision Tree

The decision tree provides insights into which factors are most predictive of default. By interpreting the tree, financial institutions can make informed lending decisions. For example, if the model shows that borrowers with lower credit scores and higher loan amounts are more likely to default, this can be used to adjust lending policies.

EAD and LGD Calculation Using the Decision Tree

After building the decision tree with PD as the target variable, the next step is to calculate EAD (Exposure at Default) and LGD (Loss Given Default). For each loan in the training set, input the actual EAD (loan principal) and LGD (expected loss given default) values into the decision tree.

• Calculate Expected EAD: The Expected EAD is calculated by aggregating the exposure at default, relative to the principal amount of the defaulting loans in each risk category (leaf node) of the tree. Loans with a higher PD contribute more to the Expected EAD.
• Calculate Expected LGD: The Expected LGD is calculated by aggregating the loss given default for each risk category (leaf node) of the decision tree, based on the sum of latest outstanding and exposure at the time of first default. Borrowers with a higher PD typically have a higher LGD, indicating a greater loss in the event of default.