Menu

Risk Weighted

How to Address CECL and IFRS 9 Weaknesses Exposed by COVID-19

The pandemic has illuminated deficiencies in existing accounting standards for expected credit losses. What are these shortcomings, and is there a better, investor-friendly, countercyclical approach?

Friday, August 28, 2020

By Tony Hughes

Advertisement

During the 2008/09 global financial crisis, loan-loss accounting methods were unable to provide timely, accurate information to investors about the quality of loans held by banks. CECL and IFRS 9 were introduced to address these concerns but, for different reasons, have failed to transmit useful signals in the COVID-19 economy.

The way things are going, these protocols may struggle to survive their very first recession. Given this possibility, it's logical to explore the CECL and IFRS 9 difficulties that the coronavirus recession has exposed, and then reconsider an alternative, data-driven approach for projecting credit losses.

tony-hughes
Tony Hughes

I want to avoid two traps that are common in public commentary on these matters.

The first is the tendency to bash the accounting standard setters without suggesting a valid alternative. The second is to give the false impression that because CECL (in particular) is flawed, the incurred-loss framework - which calls for banks simply to wait until a default is assured before estimating losses and recording the expense - must necessarily be superior.

The upheaval in accounting standards that followed the GFC was all about making reported loan-loss allowances more forward looking - rising during booms, falling as subsequent recessions bite. Both IFRS 9 and CECL attempt this with a heavy reliance on macroeconomic forecasts and statistical credit risk models.

CECL is the more radical departure from the old standard (which required very little engineering), because lifetime modelled losses need to be recorded for every loan on day one. The staging mechanism at the heart of IFRS 9, on the other hand, dilutes the importance of the models, because one-year expectations replace lifetime losses for all unblemished loans.

If CECL is found to be poisonous, however, IFRS 9 (a less-concentrated version of CECL) may also be deemed toxic. Countries that are using IFRS 9 should therefore take a keen interest in the fate of CECL as a harbinger of possible deficiencies with their own standard.

Modeling Obstacles: Past and Present

Many studies have considered how CECL and IFRS 9 would have performed during the global financial crisis (GFC) of 2008-09, with most finding that they would have been less procyclical than the incurred-loss model - but still procyclical. To achieve true countercyclicality in the context of an approach built around loss modelling, you needed to assume a superhuman ability to predict the future path of the economy.

Well-specified credit models, on which CECL and IFRS 9 crucially depend, have a chance of working when banks originate the loans that ultimately cause the recession, which is what happened during the subprime crisis. The COVID-19 shock, in contrast, was external to the banking sector, completely unpredictable and cataclysmic in its effect.

In reporting their virus-affected Q2 financials, some U.S. bank executives have ridiculed CECL, citing model predictions far in advance of observed rates of non-performing loans. Many analysts, moreover, are finding that their models are struggling with the dichotomous effect of the shock - with some sectors of the economy thriving while others have been decimated. They are also having trouble capturing the impact of government assistance programs on observed borrower behavior.

In short, the models currently have a large residual, and are probably experiencing a seismic structural break. They are missing key variables that weren't needed six months ago and that might not be needed again for a century or more.

Investors, who the new accounting rules were meant to aid, have been left scratching their heads. They need to determine, for a given peer group, whether anomalies in the relative performance of companies reflect differences in portfolio make-up or vagaries in the model-building process. It's impossible to exonerate the models at the moment, so investors cannot easily make a judgment on the relative merits of the underlying portfolios.

This is a big problem. The unavoidable fickleness of recession-era credit modeling makes CECL (and probably IFRS 9, as well) unhelpful - but going back to incurred loss is also a very poor option.

TTC: An Alternative, Investor-Friendly Approach

The solution (which I initially proposed in 2019) downplays the use of models and forecasts, especially those with the potential to fail at certain points in the business cycle. Instead, it emphasizes using through-the-cycle (TTC) estimates of lifetime loss for each loan, as opposed to the point-in-time specifications required by CECL/IFRS 9.

While still model based, TTC numbers are preferred in this context, because they have well-known properties that don't change very much from expansion to recession and back again. TTC models perform the necessary task - if a bank originates a huge quantity of poor-quality loans in a particular quarter, its average baseline allowances will rise sharply relative to those of more conservative peers.

The second element that can be applied is a business cycle adjustment factor, common to all banks, to the loan-level allowances calculated in the first step. If industry loan growth is high in a particular period, calculated allowances are scaled up for a given loan; if the industry is in recession, allowances decline relative to baseline.

“Industry” here is defined in a granular manner with different regions, rating grades/score bands and product segments potentially considered. This mechanism (which does not require the use of models) acts like a tax on excessive lending - if it's growing like a weed, it's probably a weed. The “taxes” collected are retained by each bank and returned to shareholders if losses turn out to be lower than anticipated.

I wouldn't want a mechanism like this to go live at the start of a COVID-19-like recession. Unlike CECL, IFRS 9 and incurred loss, this alternative approach actually is, by construction, countercyclical. If this were in place right now, for example, average reserves would be declining for many products with falling balances.

This is because the proposed protocol relies on adequate reserves being in place before the first day of the first recession. Aside from this issue, though, the system would currently be working well.

The TTC numbers would still be providing investors with intelligible data on portfolio makeup (despite the craziness of the economic reality) - and this is the most important task the loan-loss accounting system should currently be performing.

Parting Thoughts

The practice of investing revolves around making successful predictions about future trends in business and society. If you like fro-yo and think demand will come roaring back, put your money where your mouth is. If you think COVID-19 has run its course and that banks will suffer low loan losses in 2021, lay your bets accordingly. If you're right on either score, you'll make far more money than the person making the opposite bet.

To make these predictions accurately, investors need good, clean data. If they want help making predictions, a wide variety of very high-quality forecasting services are available.

The strange thing about CECL and IFRS 9 is that they force all investors to consume a particular set of forecasts. As we have learned through COVID-19 and other crises, successful investors are no longer those making the best predictions, but, rather, those who can most accurately interpret what the other predictors are predicting.

Ideally, we want a countercyclical loss-projection system that provides clean data, leaves prediction up to investors and reacts quickly to changed circumstances.

A system based on credit growth rates and through-the-cycle credit models gets much closer to this ideal than anything else that's currently on the table.

 

Tony Hughes is a risk modeler and economist. He is a credit risk analytics expert and thought leader with many years of experience building high-caliber risk modeling and data science teams.




Advertisement

We are a not-for-profit organization and the leading globally recognized membership association for risk managers.

weChat QR code.
red QR code.

BylawsCode of ConductPrivacy NoticeTerms of Use © 2024 Global Association of Risk Professionals