Guest Post – CECL Pointers: What to Do Now, What You May Have Missed For Later
AuditOne, LLC Co-CEO Jeremy Taylor has prepared a summary of the proactive measures financial institutions need to consider now to better prepare for the new Current Expected Credit Loss (CECL) standard. We’re delighted to host this guest blog post.
A lot is being written these days about the new Current Expected Credit Loss (CECL) standard for the ALLL and what it’s going to do to bankers’ lives. There are plenty of summaries available out there. We’re going to stick here to two angles.
- What you need to do now to prepare. For many institutions (the non-public), there’s still about three years until you need to be reporting your loan loss reserving in accordance with CECL. Which means temptation to postpone. But there are a couple of things all institutions should be doing right now to lay the groundwork, even if time can still be taken for other things (like considering alternative calculation methodologies, available vendor models). That’s because #2 below will require a lot of planning, to ensure you have those needs fully anticipated and ready to go. Which will in turn become the top agenda item for #1.
- Form a CECL Committee. At a smaller institution, the obvious participants are the CCO, CFO and COO/CIO (or their designees), all of them having direct interests in the process. At this earlier stage, the Committee will have an education role for the bank, and will need to be gathering information for future decisions on models, methodologies, et al. But its key near-term responsibility will be to:
- Identify and arrange for collection of all required data. This applies both in terms of time series (i.e., as far back as can reasonably be gathered) and
cross-sectionally (i.e., a broader range of data series than currently required). It applies both to internal data (i.e., loss and other performance characteristics for the institution’s loan portfolio, down to the borrower and loan level) and external (e.g., macroeconomic conditions in relevant markets, peer bank loan performance metrics). It should be noted that identification of data needs will require at least some sense of how reserve requirements will be calculated (modeled).
- What may not have registered. The 2016 guidance on CECL was deliberately vague as to how to go about setting up a CECL-compliant approach. This was appropriate simply because of the vast differences across the US financial system in size, sophistication, data availability, MIS capabilities, in-house expertise/understanding, etc., etc. But there are some key features or characteristics of CECL whose significance and implications may not have fully registered, that we thought might be helpful to highlight.
- The general vs. specific reserving distinction (i.e., FAS 5 vs. 114) is going away. That’s because the current approach to impairment analysis is in line with the general CECL approach (whatever the loan quality) – i.e., estimating potential loss over remaining life of the loan. So the carve-out of impaired loans, with their own manual of requirements, will no longer be needed.
- But there will still be pooling. CECL envisages estimation of potential loss on the basis of pooling assets with similar (risk of loss) characteristics, similar to today’s approach. That could apply to impaired assets, such as mortgages or consumer loans with common borrower and structural features and common drivers of credit impairment. But it is likely that larger commercial loans that are adversely graded will continue to be handled and reported individually.
- CECL will apply not just to loans but also to securities. But not to a trading portfolio. For HTM securities, you’ll need to estimate a lifetime credit loss, just like for loans. For AFS, rather than the current requirement of (irreversible) OTTI assessment, there will be a valuation adjustment to reflect the difference between fair value and amortized cost. Estimation of lifetime expected loss can be done on a pooled basis for securities with similar risk characteristics.
- When you book a new loan or security, you book the expected credit loss as an expense right away. It’s no longer the incurred loss approach of booking when a loss is deemed probable. Rather, it’s an up-front estimation as to how much might be lost actuarially, given the mortality (i.e., default and recovery) characteristics of that type of borrower and loan. On average you’re going to lose a little making a given type of loan; recognizing this with a day one loss provision is entirely appropriate. Doing so will help remind us that our credit spread is intended to cover that expected loss amount (with capital there to protect against outlier (“unexpected”) losses).
- CECL’s impact on reserve levels may be material – but shouldn’t be excessive. Intuitively, moving from losses already incurred (which in practice is typically calculated based on a one-year loss horizon) to a life of loan should boost the required reserves; it means a longer period over which losses might occur. True, but there are offsetting effects. Most importantly, smaller financial institutions today are typically carrying booked reserves in excess of required (i.e., calculated) levels – and that’s after using Q-factors to push up the required levels. The move to CECL will push up required loss reserves, but for many institutions that may still lie below the current actual reserve level.
- Regulators recognize that CECL implementation will vary widely. For large institutions, splitting probability of default (PD) from loss given default (LGD) will be expected, along with more powerful migration or vintage analysis approaches. Smaller institutions, on the other hand, should be able to build on their current ALLL methodology in order to satisfy regulators – e.g., still starting with historic loss rates, but looking back over a longer time horizon; still adding on Q-factor adjustments, but looking out over a longer (remaining life) horizon. However:
- More institutions will find vendor software worth considering – as much for managing the more onerous data expectations as for increases in complexity of calculations required.