Modeling Severity Under CECL


As many financial institutions enter into CECL parallel runs in 2019, they need to have a solid handle on their loss forecasting model inputs. To ensure compliance with the standard, regulators and auditors will be scrutinizing these inputs and assumptions, including severity, or loss given default (LGD).

Regardless, whether a loss forecasting model takes the form of a discounted cash flow, vintage, or pooled approach, its foundation is historical loss data. In this paper, we define the multiple components making up a credit loss and discuss how this definition could differ between FAS 114 and CECL. We primarily focus on collateralized loans, though much of our discussion may still apply to unsecured loans.

A loss can be calculated simply as [Defaulted Balance + Total Costs – Total Proceeds]. Even before a loan defaults, it begins to incur costs. Uncollected interest may accrue, collateral may be appraised or maintained, and taxes or insurance might be advanced on behalf of the borrower. With respect to proceeds, a lender would typically recover defaulted balance and incurred costs by liquidating collateral, claiming on any insurance, and collecting from the borrower. Severity, or LGD, is defined as loss as a percentage of defaulted balance.

FAS 114 broadly defines how costs may be included in a loss estimate.(1) However, CECL clarifies that credit loss should only consider direct costs-to-sell, such as taxes and brokers’ commissions.(2)(3) This excludes several costs that may be currently included in a FAS 114 approach.

We analyze a sample of Fannie Mae single family loans for vintages 2000 through 2012 and reporting periods 2000 through 2015. This portfolio experienced $8B of defaults, or 2.7% of the total $308B. Below, we compare total loss and severity between a FAS 114 approach and a CECL approach. Costs are presented in grey and proceeds in blue.

We see that, over the analyzed period, this difference remains directionally consistent and relatively constant.


For this portfolio, historical LGD would be lower by approximately 9 percentage points in a CECL model versus a FAS 114 model. Because of the narrower definition of costs allowable, this institution would lower its loss reserve estimate under CECL, all else being equal. The impact of CECL adoption on LGD would differ for other vintages, products, asset classes, and portfolios, however. We recommend:

  1. Assess your historical loss data. Work with your data providers to understand the reliability, completeness, and granularity of your loss data. Are key data elements stored in different warehouses? Do key data elements reconcile with each other and the financial statements?
  2. Align and document LGD definitions with stakeholders. Having a documented consensus among Model Development, Model Validation, and Accounting Policy can make the difference during an audit or regulatory review. Has the CECL Standard been interpreted by Accounting Policy? Are these interpretations reflected in your model?


(1) ASC 310-10-35 broadly defines costs-to-sell, and is superseded by ASU 2016-13 (ASC 326)

(2) ASC 326 no longer specifies what should be considered costs-to-sell and cites ASC 360

(3) ASC 360 defines selling costs as incremental direct costs to transact a sale