CECL: Simply Complex in the Real World

By Michael Gullette

Here it comes!

After five years and five different proposals, the Financial Accounting Standards Board is set to issue a final Current Expected Credit Loss standard for impairment of loans and debt securities. With an effective date for fiscal year 2019 for SEC registrants and 2020 for all others, CECL is bound to make a big splash – but not necessarily in ways first anticipated. While estimates made by regulators in 2011 indicated that CECL would result in a 30-50 percent increase in the allowance for loan and lease losses, a recent KBW article believes the increase will be much lower, around 6-8 percent. So, it may not be the huge hit to capital that many feared, but it will be the jump in complexity that presents the biggest challenge to bankers.

While acknowledging that systems will need to be reconfigured in order to perform CECL’s life-of-loan loss estimates, both regulators and FASB members want a standard that is scalable to banks of all sizes, not requiring complex modeling. Anticipating CECL’s release, the Federal Reserve staff recently held an “Ask the Fed” webinar to show bankers how a vintage analysis (one that analyzes charge-offs by year of origination) may be used to estimate expected credit losses over the life of a loan portfolio. Presumably, this is a good example of a non-complex model.


Never mind that forecasts over a year in the future are notoriously unreliable, even those made by economists. The LOL concept introduced in CECL is an enormous change from current accounting and has potential complexities waiting at every turn. It is not the examiners, though, that that bankers should fear. In the real world, auditors, complying with their regulator (the Public Company Accounting Oversight Board), will require detailed documentation to support the specific financial impact of banker long-term forecasts of future losses. No shortcuts. No, Monte Carlo simulations will not be needed, and thirty-year forecasts will not be required. If this were 1995, it might not be a big deal; in this post-financial crisis age of too-little-too-late reserves, earnings management and Sarbanes-Oxley, however, quantifying such estimates will be a steep hill to climb for many bankers.

Forecasts of the future: Too hard for too long

During their webinar, the Fed staff presented historical life-of-loan charge-off rates per vintage. While different, this vintage approach seems straightforward – more work, yet straightforward nonetheless. The fine print is that these averages are merely the starting point – the real challenge is what bankers must do in order to arrive at an actual loss expectation during a real life audit. Current PCAOB inspections have auditors scrambling to obtain specific quantitative support for the so-called “Q factor” qualitative adjustments applied to historical loss rates. Justifying that the rise in local unemployment will result in a 10 basis point (bp) increase in losses (and not 20 bps) is an arduous task now, and it will get significantly harder under CECL. Q factors are currently used to adjust historical loss rates to reflect current conditions. Under CECL, they must reflect current conditions, as well as conditions a year from now, two years from now, and so on over several years. How will banks do that?

Think of the complications of estimating future losses. For example, an assumption that general unemployment will increase over the next two years will impact one vintage differently than another vintage. That will need to be modeled. Remember also that losses on collateral-based loans also do not normally occur until loan to value ratios hit 100 percent. That must be considered, too. Needless to say, future loss expectations based on forecasts of future economic indices will often get pretty hairy. The regulators may allow certain short cuts. Auditors will, nevertheless, require bankers to include such factors into their estimates.

Piling on top of these complexities is the extended time period that a LOL forecast requires – the financial impact of your assumptions will be multiplied over years. Bankers now struggle in explaining to auditors why a Q factor resulted in a 25 basis point bp adjustment and not one of 50 bps. Imagine the struggle in justifying a 100 bp lifetime loss adjustment, when other legitimate assumptions could result in 250 bp adjustment!

Historical data: To use or not to use

Reminiscent of mutual fund disclosures that warn “past performance is not indicative of future results”, the Fed staff acknowledges that some historical experience may not be relevant to current vintages. For example, a bank’s current loss expectations should not generally be based on losses experienced during the financial crisis. They also point out that expectations may differ, based on increases or decreases in underwriting standards and other circumstances that may cause certain vintages to be unique. That sounds reasonable, but which history WILL be relevant and how will banks support that assumption? Will a separate analysis be required for each vintage of history to be used?

Currently, auditors merely ask that the last X number of years of historical loss data be consistent from year to year. This is because recent history, by its nature, is related to current losses. Under CECL, however, such a relationship does not exist because CECL does not address actual losses, but expected losses over the lifetime of the loan. Under CECL, virtually every year of historical data used will be scrutinized.

Metrics: When higher delinquencies and rising unemployment are good things

Bankers typically look at different credit metrics, such as delinquencies or nonaccrual loans, to get an idea of where loan loss provisions should go. An increase in delinquencies normally equates to an increase in loss provisions. This allows bankers to also communicate with their boards and their investors. Unfortunately, such a relationship will no longer exist under CECL. Since loan loss provisions are initially recorded at origination, provisions will decrease, even as delinquencies increase, if such delinquencies are less than what was expected at origination. Increases in unemployment can cause a decrease in provisions if the increase was less than expected and provided for.

With this in mind, banks will need to create metrics of their own to determine whether their original loss expectations require change. They will also need to figure out how to then communicate these metrics to board members and investors in a way that goes beyond “trust us.” This specific issue has yet to be addressed by regulators.

Both FASB and banking regulators are intent on implementing CECL in a “scalable” fashion that does not require complex models. Unfortunately, it is the auditing firms which may make CECL a multi-headed monster only a “quant” could love. So far, the regulator of the auditing firms, the PCAOB, has not commented publicly on CECL on what kind of documentation will be considered reasonable and supportable. However, until they do, FASB will not know how many heads this monster will have.

About Michael Gullette

Michael Gullette
Michael Gullette is vice president for accounting and financial management policy at ABA.
More from Michael Gullette >