CECL and Stress Tests: A Dangerous Mix

CECL and Stress Tests: A Dangerous Mix

This post was revised on September 18, 2019.[1]

Last year, BPI published a paper that analyzed the impact of the “current expected credit loss” (CECL) methodology for accounting for loan losses on bank regulatory capital ratios had it been in place during the 2007–09 financial crisis.[2] The paper reported that CECL would have been highly procyclical, as it would have resulted in banks adding to their loan allowances (and constricting lending) when their support for the economy was the most needed. In short, CECL is procyclical when incorporated in non-stressed capital requirements because of the rapid shift in expectations about the deteriorating state of the economy at the onset of an economic downturn. A similar mechanism is at play under the stress tests because of the sudden change from banks’ own baseline scenarios at the beginning of the planning horizon to the Federal Reserve’s severely adverse scenario. As will be shown in this blogpost, stressed capital requirements for banks would have increased by a significant amount had the Federal Reserve incorporated CECL in the 2018 comprehensive capital analysis and review (CCAR).[3]

At the end of 2018, the Federal Reserve announced that it would not account for the current expected credit loss methodology in CCAR until 2022 at the earliest. Before that announcement, CECL was to be included in the 2020 stress test cycle. Nonetheless, the Federal Reserve is expecting banks to incorporate CECL into their own company-run stress tests. This blogpost shows that introducing CECL in CCAR will generally lead to increases in capital requirements. CECL pulls forward loan loss provisions and leads to a greater decline in peak-to-trough regulatory capital ratios under stress. Also, the front-loading of loan loss provisions would vary across portfolios and assumptions on scenario design and would have a greater impact on banks with a higher concentration of retail portfolios and portfolios with longer tenors. For instance, the unemployment rate peak and the house price trough, which are key drivers of expected losses on retail portfolios, are typically reached later in the scenario. Therefore, the earlier recognition of losses under CECL has a larger impact on retail portfolios.[4] In summary, incorporating CECL in CCAR will increase the stringency of stress tests, especially for banks that hold loans with longer maturities.

The objective of this note is to estimate the magnitude of the impact of incorporating CECL in the stress tests. Note that CECL has no impact on the total amount of loan losses over the nine-quarter stress horizon; thus, the impact on banks’ stressed regulatory capital ratios should in theory be minimal. However, CECL affects the mapping between projected loan losses and loan loss provisions. In particular, because of the front-loading of provisions under CECL, especially for loans with longer tenors, I estimate that the peak-to-trough decline in the aggregate common equity tier 1 (CET1) ratio of the 35 banks that participated in the 2018 stress tests would have increased by 0.9 percentage points relative to the current methodology used by the Federal Reserve to calculate allowances on loans.

This exercise is done here in three steps. First, it projects loan losses under the Federal Reserve’s 2018 severely adverse scenario. Second, it calculates CECL-based allowances and provisions using those projections. Third, it estimates the impact on the aggregate CET1 ratio using projections for pre-provision net revenue and other assumptions on corporate taxes and dividend payouts. Consistent with the Federal Reserve’s recent stress capital buffer proposal, the analysis assumes the bank’s balance sheet size and risk-weighted assets remain constant under stress.[5]

Loan Losses Under the 2018 Severely Adverse Scenario

As noted earlier, the framework used to calculate allowances on loans does not have an impact on the projections of loan losses. This section estimates loan losses under the Federal Reserve’s 2018 severely adverse scenario, which are identical under the two frameworks used to calculate loan allowances over the first 13 quarters of the planning horizon. Under the current framework, banks don’t need to estimate loan losses beyond 13 quarters, but they would under CECL. Covas and Nelson (2018) show how to estimate CECL-based allowances for each of the 15 loan types available on bank Call Reports under a given macroeconomic scenario. This blog post uses the same approach detailed in that paper to estimate aggregate loan losses during the first 13 quarters and beyond.

Naturally, BPI’s loan-loss projections under stress yield different forecasts relative to those generated using the Federal Reserve’s own models, but the main objective is to infer the impact of incorporating CECL through the lenses of supervisory models. Thus, this analysis goes one step further and makes an adjustment to loan loss projections in the blog post to more closely match the Federal Reserve’s projections for loan losses in Dodd-Frank Act stress tests (DFAST) 2018. Specifically, projected loan losses from DFAST 2013 through DFAST 2017 and those estimated using BPI’s top-down models are used to make portfolio-level adjustments to projected losses under the 2018 severely adverse scenario. These two bar charts compare loan losses (both in billions and as a percent of loans) obtained using BPI’s own models and the Federal Reserve’s projections under DFAST. Overall, the differences in the projections for total loans are very small—less than 1 percent (left panel). At the portfolio level, BPI’s top-down models understate loan losses for first-lien mortgage loans and overestimate them for consumer loans, but the differences in projections approximately cancel each other out (right panel).

CECL-Based Allowances and Provisions

To show the increase in capital requirements under CECL in the stress tests, these two line charts depict the projected allowance for loan and lease losses (ALLL, left panel) and the projected reserve build (right panel) under the current incurred framework (orange lines) and under CECL (purple lines). The projected ALLL under CECL equals the sum of lifetime losses across all loan types. Under “FRB incurred”, projected ALLL is equal to the next four quarters of net charge-offs. The line chart on the left shows ALLL in quarter zero until quarter 9, where quarter zero represents the end of period zero allowances or, equivalently, the beginning of period 1 allowances. CECL-based allowances start at 4.7 percent of loans in quarter zero and decline steadily until reaching 2.4 percent in quarter 9. In contrast, allowances under FRB incurred show a “hump-shaped” pattern, peaking at 3.3 percent in quarter 4 and ending at 2.2 percent in quarter 9.

The line chart to the right depicts the reserve build, which corresponds to the change in ALLL during a given quarter of the stress test planning horizon. As the macroeconomic scenario used to project CECL-based allowances transitions from a baseline scenario to a severely adverse one, the projections for lifetime losses under CECL increase approximately $230 billion in the first quarter of the stress horizon (purple line). This occurs because perfect foresight forces banks to assume that the depth and duration of the recession is known with certainty at the onset of the stress tests.[6] After the first quarter of stress, CECL-based allowances start to decline as banks release their reserves; thus, the purple line in the right panel lies below zero from quarter 2 onward.[7] In contrast, under the FRB incurred framework for calculating allowances, the increase in reserves is much more gradual over the nine-quarter horizon, and it is also almost always above the zero line.

Although the assumption of perfect foresight is quite extreme, it was used in the European Banking Authority and the UK stress tests this past year. In the UK stress tests, the Bank of England neutralized the impact of IFRS 9 (the European analogue to CECL) on the stress tests by reducing each bank’s hurdle rates (the level of capital at the end of the trough of the projection required to pass the test) to offset the increase in provisions under IFRS 9 for each bank. In addition, the ameliorating factor in Europe is that the accounting framework does not apply lifetime estimates for most loans. In particular, it limits expected losses to 1 year for non-defaulted loans (or stage 1 loans). Lifetime losses only apply for stage 2 and 3 loans (those close to default and defaulted loans, respectively). The Federal Reserve’s incurred loss methodology assumes incurred-based allowances should cover four-quarters-ahead of projected loan losses, which is a lot like IFRS 9.[8]



Estimated CET1 Impact

To estimate the impact on common equity capital ratios, projections for pre-provision net revenue (PPNR) are generated using the methodology in Hirtle et al. (2015). For brevity, I do not describe the PPNR projections, but I also match overall revenues over the nine-quarter horizon (projected to be $492.3 billion in CCAR 2018).[9] Furthermore, the analysis assumes a corporate tax rate of 21 percent and eliminates net operating loss carrybacks to account for the changes made in the Tax Cuts and Jobs Act. Under these assumptions, the path of the CET1 ratio under DFAST is depicted in this line chart. Under the Federal Reserve’s incurred methodology, I project a peak-to-trough decline in the CET1 ratio or the stress capital buffer (SCB) of 3.7 percentage points. The actual decline in DFAST 2018 was 4.4 percentage points, but my lower estimate reflects the constant balance sheet assumption and the elimination of prefunding of dividends.[10] The results using CECL-based reserves under perfect foresight are represented by the purple line, which shows a SCB of 4.6 percentage points, 0.9 higher than under the current incurred methodology. This is entirely driven by the introduction of CECL, since total losses are the same under the two frameworks used to calculate allowances on loans.

Relaxing the Assumption of Perfect Foresight

The assumption of perfect foresight is not realistic in practice, because the odds a bank attaches to the occurrence of the Fed’s severely adverse scenario are extremely low based on the macroeconomic conditions that prevail before the start of the stress tests. After a bank observes the first quarter of the stress scenario, projections for future periods would then be updated to reflect the new information. It is not unreasonable to expect that real-time forecasts would take a few quarters to converge to the Fed’s severely adverse scenario. Although the adjustment speed of real-time forecasts is a topic that deserves its own separate analysis, for simplicity I assume that it would take four quarters for bank CECL-based allowances projections done in real time to catch up to those provided in the Fed’s severely adverse scenario and that the convergence occurs in equal steps over the four quarters.



These two line charts illustrate the impact on CECL-based allowances and the CET1 capital ratio under a four-quarter delayed build. The top line chart shows that CECL-based allowances calculated under real time would take 4 quarters to converge to those projected under perfect foresight. Moreover, because total provisions don’t change and the number of quarters it takes for bank CECL-based allowances projections to catch up with those calculated under perfect-foresight projections is less than or equal to the quarter in which the bank reaches its trough in regulatory capital, the SCB remains unchanged. Thus, the difference in the SCB between CECL and the incurred methodology is still 0.9 percentage points under a four-quarter delayed build, as shown in the bottom line chart.

Relaxing the perfect-foresight assumption could still result in a reduction in capital requirements. For that to happen, the number of quarters it takes for a bank’s CECL-based allowance projections to catch up to those under obtained perfect foresight needs to exceed the quarter in which that bank reaches its trough in regulatory capital. For instance, assuming a four-quarter delayed build would reduce capital requirements for firms that reach the trough in regulatory capital during any of the first three quarters of the planning horizon.

There are other changes to the stress test assumptions that would reduce the impact of CECL on banks’ capital requirements under stress that are realistic and avoid the use of arbitrary assumptions.[11] First, the Federal Reserve could offset the impact of CECL in CCAR by reducing the severity of the severely adverse scenario in the stress tests. This approach would mitigate the shift of the types of loans than tend to perform badly under a severe stress, such as small business loans, to the unregulated nonbank sector. In addition, the Fed could also assume that loans originated under stressful conditions—those that replace the loans that mature during the 9-quarter planning horizon—have lower loss rates. Under CECL, banks must reserve for newly originated loans. If those loans are assumed to be underwritten under tighter loan standards, the amount of CECL-based allowances would be lower than under the Fed’s incurred methodology.


In summary, the Federal Reserve’s decision to delay the implementation of CECL in CCAR reflects concerns that it could lead to sizable and unwarranted increases in capital requirements, especially for firms that make loans with longer tenors or banks with a greater share of loans in retail portfolios. The mechanism by which CECL has an impact under stress is driven by the high sensitivity of loan losses to the economic outlook and the procyclical nature of CECL. The results outlined in this blogpost also underscore the importance of offsetting the impact of CECL for non-stressed (or spot) capital requirements.

[1] This post was revised to include results that relax the perfect-foresight assumption. In addition, changes in assumptions resulted in a reduction of the average impact of CECL on a bank’s stress capital buffer from 1.6 to 0.9 percentage points relative to the current methodology used by the Federal Reserve.

[2] The CECL methodology for accounting for loan losses requires a bank to reserve for expected loan losses over the life of the loan at the time the loan is made. After the loan is originated, banks raise the loan loss reserve if the probability of default of the loan increases and vice versa.

[3] The stress tests are conducted over the first nine quarters of the planning horizon. However, in practice banks need to know the path of the macroeconomic variables over 13 quarters to be able to estimate loan loss provisions.

[4] See Financial Stability Report, Bank of England, November 2018, Issue No. 44, pages 11–14.

[5] Moreover, according to Vice Chair Randal K. Quarles’s speech on September 5, 2019, banks will no longer be required to prefund four quarters of planned common stock dividends in their stress capital buffer.

[6] Only in CCAR is perfect foresight even possible, because the Federal Reserve provides banks with 13 quarters of information about the future that would otherwise not be known with any degree of certainty.

[7] The shape of the reserve build under CECL depends on the reasonable and supportable horizon used. The analysis in the note uses 9 quarters, which is the same as the stress-test planning horizon.

[8] They are approximately equal. There is also a problem associated with perfect foresight under the FRB incurred methodology, as noted on page 69 of the DFAST 2018: Supervisory Stress Test Methodology and Results (June 2018). However, the true-up in ALLL is considerably smaller, and the Federal Reserve smooths the difference into the provision projection over the nine quarters of the planning horizon. In this analysis, we smoothed the true-up under the current framework equally over the nine quarters.

[9] For simplicity, I don’t model trading and counterparty losses and other losses—those are directly deducted from PPNR. Also, I assume that trading and counterparty losses are incurred in the first two quarters of the stress horizon.

[10] Under DFAST 2018, risk-weighted assets were projected to grow under stress and included nine quarters of dividends.

[11] The Fed currently smooths the difference between the firm’s actual level of ALLL at the beginning of the planning horizon and the ALLL at the start of the stress horizon under the severely adverse scenario into the provisions projection over the nine quarters of the planning horizon.