CECL-Ready: Loan portfolio segmentation
By Tim McPeak, Executive Risk Management Consultant, Abrigo
One of the steps financial institutions will need to take as they transition to the current expected credit loss model, or CECL, will be to determine their segmentation, or the way they break their loan portfolio into pools for the purpose of estimating their allowance for loan and lease losses (ALLL). In the end, whether they make wholesale changes or instead determine that their current segmentation is optimal for CECL, institutions will need to document how they arrived at their decisions.
In practice, it is common to find institutions that have tinkered with their loan pools over the years as the result of input and guidance from regulators and auditors (particularly during the financial crisis era as institutions attempted to deal with high-default portions of their portfolios). As such, the transition to CECL provides an opportunity to revisit the portfolio with a fresh perspective. The following are some points to keep in mind in considering your segmentation for CECL.
Granularity vs. significance in CECL segmentation
The CECL standard states: “Segmentations or pools should have similar risk characteristics. These pools should be as granular as possible while maintaining statistical significance.” (emphasis added). It is this balance between granularity and significance (in terms of pool size) that is a key friction point in segmentation elections. In most cases, finding the balance will require a blend of art and science.
In considering risk characteristics that may warrant break-outs in the portfolio, it is easy to find yourself digging deeper and deeper into the details. However, with each additional filter added, the smaller each loan pool becomes, and as a result, the higher the risk of the pools lacking significance. Keep in mind there is no “right” way to determine this balance; each institution will need to evaluate these decisions for itself.
Keep data controls in mind when segmenting the portfolio
As noted above, there is likely no shortage of data fields that institutions could utilize in defining risk characteristics for segmentation purposes. However, it is important to keep in mind the controls around each of these data points, particularly as the end result of the allowance estimate is a financial statement entry.
For example, many institutions have noted that metrics such as loan to value (LTV) and borrower debt service coverage ratio (DSCR) can be important indicators of credit risk and so they consider using them as segmentation metrics. While this intuitively makes sense, such metrics rely on data points like collateral values and borrower financial information, but institutions may not have adequate controls to ensure accuracy and timeliness of those data points. It would likely be preferable to rely on credit risk grades/ratings as the segmentation metric since these ratings systems should already have metrics like LTV and DSCR baked into them.
Mind the “counts” in portfolio segments
When considering the size of each portfolio segment, it is important to consider not only the size in dollars but also in terms of units (or “counts” of loans). The segmentation in the end serves as the format for the application of an institution’s loss rate methodologies, qualitative adjustments and forecasts. Since default and loss events are the “raw material” that most loss rate methodologies are based on, having enough observations is critical.
For example, if we are attempting to determine a “lifetime” probability of default (PD) for a particular loan type, we may follow groups of loans (or “cohorts”) for a time period equal to our average life expectation for the segment. We could then observe how many loans experienced a default event (which could be defined by payment delinquency, a move to non-accrual status, etc.) for various cohorts and use that to determine our expectation for a lifetime PD. However, if we have over-segmented our loan portfolio down to pools with low-loan counts, there is an increased chance of getting “weird” results that don’t make intuitive sense or even pools with little to no default or loss events (yielding a loss rate of zero).
Where to start CECL segmentation?
A reasonable place to begin the search for the “right” segmentation of the portfolio is the Federal Call Code. This is not to say that the granularity of the existing Call Report codes will be sufficient for all institutions, but utilizing the call code is a common starting place that virtually all institutions are already tracking (with the notable exception of credit unions). Additionally, by basing the segmentation on call code, institutions may find it easier down the road to utilize aggregated peer data in segments with little or no loss experience due to the standardized nature of call reporting.
Remember, these is no one “right” way to solve this challenge of segmenting the loan portfolio for CEL, and a blend of both art and science may be required. Carefully consider your institution’s loan portfolio and the concentrations within, and keep the above points in mind as you move down the path toward the transition to CECL.
Learn more about revisiting loan pools for ALLL calculations under CECL during the webinar, “Navigating loan pool segmentation under CECL.”
Tim McPeak is Executive Risk Management Consultant at Abrigo, where he advises on risk and portfolio management with financial institutions nationwide. Previous to his current position, Tim led Abrigo’ strategic partnership program, through which the company partners with consulting, loan review, accounting, and other professional services firms.