Since the very inception of the concept of an expected loss standard back in 2012, the CECL accounting standard has changed significantly over time, from the “three-bucket approach,” which attempted to converge with the International Accounting Standards Board (IASB), to a concept that focused on a discounted cash flow approach, to a final draft standard that is highly principles-based with very few bright lines throughout.
CECL has been touted as the “biggest change to bank accounting,” and financial institutions across the country have been keeping a close eye on changes and delays to the standard to ensure that their institution is prepared for its effective date. Since the Financial Accounting Standards Board (FASB) issued the final CECL standard in 2016, we have refined and tested the various methodologies and models with institutions all over the country and have learned many lessons, some through careful analysis, some through trial and error, and some from others in the industry. While no truths in CECL are universal, there are several key lessons learned over the years that your institution should know.
1. For most institutions and portfolios, vintage analysis is probably not the best modeling approach.
Vintage analysis has probably been presented as an example in more CECL presentations than any other method due to the simple and effective presentation of many of the key concepts in the guidance (lives of loans, forecasting conditions, reversion to the mean, etc.). If your institution manages large, homogenous portfolios where age and vintage of a loan may be the best single indicator of future loss experience, the vintage analysis approach may prove effective. However, you’ll find that for the vast majority of other institutions and portfolios, there are a variety of issues that arise when attempting this approach. For one, many smaller financial institutions have infrequent losses or default events in the majority of their portfolios. Trying to further segment these losses by vintage year simply exacerbates this problem by further dividing an already thin history. Furthermore, for many commercial credits, the vintage and year of the loan likely has no discernible or provable relationship with the expected remaining losses on that credit and using these variables may actually lead to an expected loss model that is actually less predictive than one based on other criteria.
2. Overly granular loan segments may do more harm than good for an institution’s CECL estimation process.
To kickstart your CECL journey, you’ll likely begin with applying the guidance related to grouping together loans with similar risk characteristics. This was certainly encouraged in the guidance itself, with an entire laundry list of potential risk characteristics listed as ways that instruments might be segmented. Certainly, there is legitimacy to the fact that loss expectations on a given loan may be affected by many different factors, including collateral types, risk ratings, geography, industry, structure, and term. However, there are two main issues that arise when these segmentations are applied to excess. One, you may find that your institution lacks sufficient loss or default data to inform even very broad segments, much less very granular ones. Using broader categories with supportable historical information would likely be preferable to trying to sub-segment a limited loss history and causing it to become statistically meaningless. Secondly, if your institution does not have sufficient loss history and will have to rely on some form of peer data, it may be impossible to find peer or industry information that is as granular as the segmentation that has been created. Both of these situations likely lead to either collapsing the segments anyway, pay for and justify the use of expensive industry data, or justify extensive qualitative adjustments to derive their expected loss allowance.
3. The sum of a very long list of very small numbers is probably still a small number.
One of the biggest CECL challenges your institution may encounter is the lack of meaningful loss or default history. This challenge can present itself in two ways. In some cases, you may have data challenges regarding the actual loan and loss history available. Sometimes, this is due to past core conversions or a lack of data archiving in previous years. In these instances, even there is a history of losses, it may not be in an accessible format for modeling purposes. In other cases, your institution may maintain good data history for their loans and losses, thus containing so few losses that it makes any type of analysis on it largely meaningless. For example, an institution that has 15 years of loss history, but has a 0% annual loss rate in 8 of 15 years, and never more than 0.03% in the others, is going to produce a maximum cumulative expected loss rate of 0.21% (assuming these are very long-lived loans with little prepayment activity) – and very likely much less than that. This can be a surprising revelation for institutions that are holding between 1% and 1.25% in current incurred loss reserves. Today, the vast majority of that allowance number is made up of qualitative factors for many of those institutions.
For either of these situations, you have two real options. First, you could utilize an expected loss estimation method that would allow for an input of external loss information directly into the model. Typically, this would narrow the choices to a discounted cash flow (DCF) approach or the weighted average remaining life (WARM) method. Either of these methods can utilize external or peer information as inputs to form an expected loss calculation. The second option is to perform an expected loss calculation using the institution’s own data and then, on top of that low estimate, justify significant qualitative factors in order to achieve a reasonable expected loss estimate. Neither solution is going to be perfect, and both will require significant documentation around the decisioning to get to the ultimate result.
These are just a few of the lessons learned over the last few years of implementing CECL. Hopefully this advice can help guide your institution – no matter where it may be in the transition process – to make some CECL decisions more quickly. Unfortunately, we have seen a number of institutions travel very far down pathways that were ultimately fruitless and ended up wasting valuable time in their implementation process. By understanding and avoiding common obstacles and making logical decisions based on available data early in the process, you may be able to help accelerate your institution’s implementation timeline and avoid costly missteps along the way.
Chris Emery is Director of Strategy and Engagement at Abrigo.