Procyclicality: CECL versus Incurred Loss Model
Notes from the Vault
Larry D. Wall
The United States is on the verge of implementing a major change in the way lenders measure the expected losses on their loans for financial accounting purposes. The change would move the country from a loan loss measurement system called incurred loss, which is backward looking, to the new measure, current expected credit loss (CECL), which is forward looking. The move has provoked strong objections from various quarters (for example, see here, here, here, here, and here).
One of the major concerns about CECL is that it will lead to more procyclical lending, that is, magnifying the fluctuations in an economic cycle. An increase in a bank's loan losses will, all else being equal, result in the bank reporting lower regulatory capital ratios. Such a reduction in the capital ratio reduces a bank's capacity to lend while remaining in compliance with regulatory capital rules. Critics claim that CECL will result in higher reported loan losses during a recession than the incurred loss model, resulting in a larger reduction in lending during a downturn.
This post begins by providing a little background on the change. Next, it explains how the new procedure differs from current practice and reviews some of the evidence supporting the need for a change in accounting standards. The post then discusses why some observers are concerned about procyclicality, the evidence on whether the standards are more or less procyclical, and the policy options assuming the standards are procyclical.
A brief history on how CECL was adopted
One year after the failure of Lehman Brothers, the G20 Leaders Statement: The Pittsburgh Summit agreed on the need for "forward-looking provisioning" for loan losses. The two major accounting standards setting bodies for the developed world responded with forward-looking methods that took a more statistical, model-based approach to measuring losses. The International Accounting Standards Board (IASB) did so with its 2014 approval of IFRS 9, and the U.S. Financial Accounting Standards Board (FASB) took a more radical approach with its 2016 approval of Accounting Standards Update 2016-13, which adopted CECL.
FASB provided banks and other lenders several years to prepare for the implementation of CECL. As the date for CECL implementation has approached, concerns about the new loan loss model have grown. In response to those concerns about the difficulty of implementing CECL, FASB deferred the effective date of the new standard from calendar year 2020 to calendar year 2023 for smaller public and private banks. However, a comment letter from the American Bankers Association to FASB calls for a delay in the adoption of CECL for all companies pending a quantitative impact study of CECL's impact on bank lending. The letter raises several concerns about CECL, with an emphasis that it will make bank lending more procyclical.
What is the loan loss reserve?
The FASB establishes what is often called "generally accepted accounting principles" for financial reporting by U.S. corporations. Among the many standards issued by FASB are those related to loan loss accounting. According to FASB's "Statement of Financial Accounting Concepts 1," accounting standards are intended primarily to provide "information that is useful in making business and economic decisions…especially by external users who lack the authority to prescribe the information they want."
FASB's project page on credit losses describes the allowance for credit losses (loan loss reserve) as "a valuation account that is deducted from the amortized cost of the financial asset to present the net amount expected to be collected." This loan loss reserve is management's estimate of the expected credit losses on outstanding loans.
The call for forward-looking provisioning gets at the process by which management estimates expected losses. The method in effect back in 2009 and in effect for all U.S. banks through the end of 2019 is called the incurred loss model. Under this model, management estimates the losses on loans in which a loss has already been incurred. That is, no loss is recorded on a loan until it is "probable" that the bank has already incurred losses based on information available at that time. However, once a loss is probable, then the expected value of the estimated loss given default is calculated and incorporated into the allowance.
CECL changes the standard for measuring expected losses so that instead of waiting until it is probable that a loss has been incurred, the bank estimates the expected loss over the remaining life of all outstanding loans. Moreover, whereas the incurred loss model relies on information up to that date, CECL requires that the bank take account of expected future developments, including future economic conditions. In general, CECL is expected to result in banks reporting a larger allowance for loan and lease losses.
Why was a more forward-looking loss reserve adopted?
An often expressed concern about the incurred loss model is that it resulted in loan loss provisions during the 2007–10 crisis and recession that were "too little, too late." The finding that losses had been understated has been supported by a wide variety of studies. At a very basic level, a study by Federal Reserve Board economists Bert Loudis and Ben Ranish shows that loan losses provisions started increasing in 2007, but the peak of the loan loss provisions was deferred until the end of the recession in 2009 and the recovery's start in 2010. The peak in the loan loss allowance was also in 2010.
Moreover, a simple comparison of book-based (accounting) capital ratios and market value of equity capital ratios shows that the incurred loss model lacked credibility in financial markets. A speech by Bank of England executive director Andrew G. Haldane compared the Tier 1 capital ratios of an international set of globally important banks with the ratio of their market capitalization to book value of total assets. The market capitalization ratio started plunging in early 2007. However, the Tier 1 ratio, which relies on accounting measures of capital, remained remarkably flat over the May 2002 to November 2008 period, even for Haldane's set of so-called crisis banks (banks that failed, required government assistance, or were taken over in distress circumstances).1 Moreover, it was not just equity investors who lacked confidence in the accuracy of banks' financial statements. A paper by Federal Reserve Board economist Jonathan D. Rose documents runs on a number of large banks during the crisis, including runs on banks such as Washington Mutual and Wachovia that were categorized as well capitalized immediately before their failure.
The lack of credibility of the largest banks' loan loss estimates was resolved in 2009 when the Federal Reserve forced these banks to supply forward-looking loss estimates as a part of the Supervisory Capital Assessment Program (also known as the stress tests). As I discussed in an earlier paper, these tests helped restore bank credibility by showing that the losses not yet recorded in large banks' books were manageable.
Moreover, several recent studies find similar evidence that it was not just large banks that had understated their losses. The Office of the Inspector General of the various bank regulatory agencies conducted material loss reviews of many of the banks that failed during the crisis. A paper by Gillian G.H. Garcia (2010) reviewed 50 of the reports from crisis era failures. Her sample of large and small banks found that the inspector generals listed an "inadequate allowance for loan and lease losses" as one of the causes of failure in 42 cases.
Two other papers provide evidence consistent with the incurred loss model failing to provide accurate loan loss information for samples that included a substantial number of small banks. A paper by Clemson University professor Lucy Chernykh and DePaul University professor Rebel A. Cole (2011) tested failure prediction models using a sample that included 767 failed banks between 2007 and 2010. They found that a measure of nonperforming loan coverage outperformed a variety of regulatory capital ratios based on financial accounting statements, including the ratio of equity to total assets. Similarly, a paper by California State University–East Bay professor Robert Loveland (2016) finds that "failed banks consistently underreported loan impairments during the crisis."
The message from these postcrisis studies accords with the concerns expressed shortly after the crisis. The incurred loss model had failed the test posed to it by the global financial crisis. These results strongly supported the G20 leaders' call for a more forward-looking provisioning for loan losses.
The only direct impact resulting from a change from the incurred loss model to CECL is on a bank's financial statements. A larger loan loss provision (increase in this period's loan loss allowance) results in a lower net income. All else being equal, a reduction in income results in the bank reporting lower retained earnings, which also means it reports lower equity capital. None of these changes affects a bank's legal rights to collect on its loans, nor do they result in a requirement that a bank set aside cash to cover the loss.2 Moreover, given that the cash flows from the loan are unchanged, the reduction in income and capital only change the timing of when the bank reports the losses on any given loan.
Nevertheless, when a bank reports the losses may interact with other regulatory requirements and affect the bank's willingness to extend new loans to good borrowers. Bank regulators set minimum capital adequacy standards that require banks to equal or exceed certain ratios of capital relative to the risks on their balance sheet. Increases in the loan loss allowance has the effect, all else being equal, of reducing a bank's capital ratio. The potential problem is that loan losses are higher during economic downturns than during robust growth. As a result, banks have less capital and lending capacity during a recession than during a boom (holding all else equal). The result could be that bank loan loss accounting has a procyclical effect on lending, allowing banks to lend more during boom times but discouraging them from lending to good borrowers during recessions. The result could be that loan loss accounting would amplify both the peaks and the troughs of economic cycles.
As was illustrated during the 2007–10 period, the incurred loss model resulted in higher provisions during a recession, albeit a large fraction of the recession losses were not reported until late in the downturn or after the start of the recovery. In theory, CECL could mitigate or exacerbate the problem of procyclicality.
CECL could mitigate procyclicality concerns by requiring banks to build some loan loss allowances even before a loss becomes probable (that is, meets the requirements for the incurred loss model). Given that every loan has some possibility of a credit loss, CECL requires a bank to increase its loss provisions starting in the quarter the loan is made (sometimes referred to as day one provisions). Moreover, CECL requires additional provisioning as expected credit loss on the loan increase.
However, CECL could also exacerbate the procyclicality problem by forcing banks to recognize losses as they occur rather than requiring them to defer reporting the losses until they become probable. The result may be that CECL results in a larger increase in the loss allowance at the start of a downturn despite banks already having built up bigger reserves during the expansion. In effect, CECL smooths losses before a recession starts, and the incurred loss model smooths losses after the recession starts.
Given that theory cannot say whether CECL would be more procyclical than the incurred loss model, several empirical papers have looked at this question and found mixed results. Several papers have found that CECL would result in higher loan loss provisions in a recession, including:
- • A paper by doctoral student Jorge Abad and professor Javier Suarez from CEMFI
- • A paper by professors Steffen Krüger and Daniel Rösch from the University of Regensburg and professor Harald Scheule from the University of Technology–Sydney
- • A paper by Bank Policy Institute economists Francisco Covas and William Nelson
- • An admittedly less thorough paper by the American Bankers Association that finds stressed losses under CECL are likely to be substantially higher than during benign times.
However, CECL was found to result in lower recession-related provisions in three papers:
- • A paper by Benjamin H. Cohen, the head of financial markets at the Bank for International Settlements and Gerald A. Edwards Jr., CEO of JaeBre Dynamics
- • A paper by Moody's senior director Cristian DeRitis and chief economist Mark Zandi
- • A paper by Federal Reserve Board economists Sarah Chae, Robert F. Sarama, Cindy M. Vojtech, and James Wang.
Additionally, Loudis and Ranish focus specifically on lending and find that CECL would have led to higher provisions before the crisis and possibly during the crisis. However, looking at the net effect on lending, their paper finds that CECL would have less of a procyclical effect.
The mixed results from these studies are not necessarily surprising, as they analyze somewhat different data sets and adopt varying assumptions about the implementation of CECL. The findings that are most consistent across the studies is that the extent to which CECL would be procyclical depends on the extent to which bank managers accurately forecast the economy and, in some cases, on modeling assumptions.3 If bank managers have good forecasts, CECL would be less procyclical, as managers would see the recession coming earlier and not try to project economic conditions too far into the future.4 Presumably, the dependency of the results on assumptions about managerial foresight and modeling assumptions would also apply to the American Bankers Association proposal for a quantitative impact study before implementation of CECL.
Although procyclicality is clearly a consideration in the implementation of CECL, the implications of loan loss accounting for procyclicality of bank lending can be overstated. A paper by Tulane University professor P. Barrett Wheeler notes that bank supervisors may exert pressure to reduce lending on banks that meet minimum regulatory capital standards but which the supervisors determine are holding inadequate amounts in their loan loss allowances. In analyzing a sample of publicly traded banks from 1990 to 2014, he finds evidence that supervisors were exerting pressure on banks that resulted in lower loan growth, even though minimum capital requirements were not binding for the vast majority of his sample. Wheeler also finds some evidence that timely loan loss disclosure is associated with lower growth rates of loans, but his results suggest this was due to market pressure rather than failure to meet supervisory capital requirements.
Meanwhile, the 16 of the largest and most important banks are subject to the Federal Reserve's Comprehensive Capital Analysis and Review's (CCAR's) annual stress test, and another 17 banks are subject to biannual stress tests. As part of these tests, the banks are required to demonstrate that they have sufficient loan loss reserves and capital to absorb nine quarters of estimated accounting losses associated with a severe, but plausible economic downturn and still not be forced to limit their lending due to inadequate capital. This way of structuring the tests results in the loan loss allowance and capital becoming substitutes for absorbing loan losses in meeting CCAR's requirements.
Additionally, the Federal Reserve Board's "Policy Statement on the Scenario Design Framework for Stress Testing" is designed to be countercyclical, with larger increases in the unemployment rate when that rate is at its peak (robust expansions) and smaller increases when the rate is at its trough (recessions, especially late recessions). Thus, at a minimum, to the extent that CECL will require more procyclical increases in the loss allowance, the effect should be mitigated by the stress test requirements for those large banks subject to CCAR (those with a majority of assets in the banking system).5
Indeed, the standard for passing CCAR highlights what should be the real issue with regard to procyclicality in the CECL debate, insuring that banks have sufficient loss-absorbing capacity so that they can absorb the losses from a recession and still have sufficient economic capital to accommodate loan demand by good borrowers. Looked at from this perspective, the underlying issue is not how to measure loan losses but how to make sure banks have sufficient loss-absorbing capacity. Yet if the problem is that of maintaining sufficient loss-absorbing capacity, it is a situation that individual banks can solve without help from FASB or the bank regulators. Any bank concerned that loan loss accounting is causing procyclicality in its lending has the option of voluntarily maintaining additional capital so reported loan loss provisions are unlikely to cause regulatory capital requirements to bind.6
However, the regulators have set minimum capital requirements for a reason; absent the requirements, many banks would not hold sufficient capital to absorb the losses from a major recession. Thus, the problem can be reframed as what the regulators could do to offset the procyclicality of loan loss provisions. One option would be for the regulators to adopt countercyclical capital requirements that are higher in good times but decline during recessions.7
If banks do not voluntarily hold sufficient capital and supervisors do not require banks to hold it, there is one more way to avoid procyclical lending. The supervisors can forbear, allowing banks with insufficient loss-absorbing ability to continue lending. This solution implicitly relies on the Federal Deposit Insurance Corporation (which is ultimately backed by the taxpayers) to absorb additional losses if a bank should fail. However, in this situation the taxpayers and their elected representatives should be provided with timely information about a bank's losses and not a measure of losses that has proven to be "too little, too late."
All of the above focuses on the procyclicality of the loan loss allowance on existing loans. However, there is one other way in which the loss reserve may have a procyclical effect on loss reserves: the requirement that an allowance for expected losses on a new loan be added to existing reserves on the "first day" (or more accurately, the end of the first quarter after the loan is made). This requirement would apply to all new loans, both those made in economic expansions and those made in a recession. The degree to which CECL would have a procyclical effect on capital requirements depends upon the extent to which the new loans made during a recession are estimated to be more risky than those made during economic expansions. Thus, the extent to which CECL is likely to generate procyclical "first day" losses will likely depend to a significant degree on the modeling choices made by banks.8
Large public U.S. banks are scheduled to adopt a new way of measuring loan losses called CECL in 2020, with smaller public and private banks to follow in 2023. This change has proven to be controversial for a variety of reasons. This post examines one of the objections to CECL, that it will cause bank lending to become more procyclical than under the existing incurred loss model. The post finds that the incurred loss model failed the test of the 2007–09 financial crisis and that the G20 leaders were correct to call for a forward-looking method of measuring loan losses. It further finds that the evidence on whether CECL is more procyclical is mixed and depends to a significant degree on how it is implemented. Finally, the post observes that the underlying issue concerning the procyclicality of CECL is how to make sure banks build up enough loss-absorbing capacity in good times to keep lending to good borrowers in a recession despite higher credit losses. To the extent that CECL aggravates the procyclicality problem, arguably the best policy response is the adoption of countercyclical capital requirements.
Larry D. Wall is executive director of the Center for Financial Innovation and Stability at the Atlanta Fed. The author thanks Mark Jensen for helpful comments. The view expressed here are the author's and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. If you wish to comment on this post, please email firstname.lastname@example.org.
Chernykh, Lucy, and Rebel A. Cole. "How Should We Measure Bank Capital Adequacy for Triggering Prompt Corrective Action? A (Simple) Proposal." Journal of Financial Stability 20 (2015): 131–143.
Garcia, Gillian G.H. "Failing Prompt Corrective Action." Journal of Banking Regulation 11, no. 3 (2010): 171–190.
Krüger, Steffen, Daniel Rösch, and Harald Scheule. "The Impact of Loan Loss Provisioning on Bank Capital Requirements." Journal of Financial Stability 36 (2018): 114–129.
Loveland, Robert. "How Prompt Was Regulatory Corrective Action during the Financial Crisis?" Journal of Financial Stability 25 (2016): 16–36.
Wheeler, P. Barrett. "Loan Loss Accounting and Procyclical Bank Lending: The Role of Direct Regulatory Actions." Journal of Accounting and Economics 67, no. 2–3 (2019): 463–495.
1 To be fair, part of the reason for the drop in market values reflected weakness in banks' accounting for losses in their securities portfolio.
2 Sometimes the assertion is made that the loan loss reserve is the amount of cash a bank is required to set aside for bad loans. The absurdity of this view can be seen by thinking about a hypothetical situation in which you made a loan to a friend. If it becomes clear that the friend cannot repay the loan, you have suffered an economic loss. However, you would not compound that loss by taking cash equal to the loan value and putting it in a safety deposit box as a loss reserve.
3 A paper by Joseph L. Breeden of Prescient Models LLC disagrees with the conclusion that foresight is critical. However, even Breeden's results show that how CECL is implemented is critical, with the important difference in his results coming from other modeling decisions.
4 Although CECL requires management to estimate expected losses over the life of the loan, managers are only required to predict economic conditions over the period in which the forecasts are "reasonable and supportable." Over longer horizons, CECL permits management to assume that expected losses will trend toward their long-run average. See the paper by Loudis and Ranish for a discussion of this issue.
5 However, to the extent that CECL is procyclical, its adoption would force CCAR banks to hold more capital during nonrecession periods.
6 The discussion of how much loss-absorbing capacity is sufficient necessarily depends upon some assumptions about the severity of the recession. The only way that a bank can make sure no capital requirement ever binds is to have a loan loss allowance plus equity capital equal to 100 percent of its maximum losses.
8 The same potential for "first day" loss recognition in the stress tests may also exist for CCAR banks depending on the specific modeling assumptions. The principal difference is that CCAR also provides for the inclusion of the revenue from the loans during the stress test horizon. This recognition of the revenue via pre-provision net revenue, however, likely only provides a partial offset as it is implicitly based on average loan rates over the models' estimation period.