CECL and DFAST: Do You Know the Difference?

by Garver Moore

How The Current Expected Credit Loss (CECL) will differ from Dodd Frank Act Stress Testing (DFAST).


In June of 2016, the Financial Accounting Standards Board (FASB) introduced Topic 326 in update 2016-13, detailing the use of the current expected credit loss (CECL) notion for presentation of an Allowance for Credit Losses (ACL, or ALLL, used interchangeably). For many financial institutions, the implementation of this standard requires a level of quantitative rigor usually beyond the scope of predominant ALLL practices. Those that have implemented – or are beginning to implement – a stress testing capability to comport with Dodd Frank Act Stress Testing (DFAST) requirements rightly want to explore whether the modeling efforts already used for their DFAST program can be used in the context of ACL measurement under the CECL standard. Entities that allocated significant resources to developing the capability to prepare DFAST submissions are looking to use those resources – analysts, data stores, models, etc. in other practices.

When financial institutions are executing or preparing to execute DFAST submissions, it is important to ensure that the modeling approaches are broadly comparable and, whenever possible, to attempt to reconcile any differences in results. Further, DFAST capabilities that are already in place can be used to serve as “benchmarks”, especially under the baseline scenario, to calibrate models for the CECL capability. This article discusses the specific similarities and differences between the two exercises.

DFAST – CECL Similarities

The primary conceptual change from current ALLL practices under the Incurred Loss (ILM) notion to the future standard is implementation of a forward-looking notion under CECL. An Allowance for Credit Losses prepared under the CECL standard involves some forecasting effort. The activity of preparing a stress test is also fundamentally forward-looking. To that end, models that correlate a given econometric outlook to financial performance are employed in both exercises. 

Development of such models can be a data-intensive exercise; in order to determine what might happen to a loan in an economic downturn it is, to put it lightly, strongly advisable to have data from an economic downturn. Further, risk identification for non-retail portfolios can require high statistical counts in order to produce a stable, meaningful measurement. To that end, data storage and validation requirements for both exercises can be similar. Institutions with data warehouse capabilities in DFAST scope have found those capabilities were sufficient for CECL scope. 

Development and deployment of such models as might be used in these exercises requires similar skillsets and capabilities. Whether an entity works with a vendor platform, develops tools in-house, outsources some or all of the effort, or takes a combination of these approaches, the resulting models will require an understanding of the nuance, art, and science involved. Such models rarely produce static “answers”; rather, the language of ranges and confidence intervals is employed. It is not just management’s right, but its responsibility to interpret these results in both cases. 

Finally – and this point is easy to lose sight of – the results of the DFAST and CECL modeling will not determine how an entity performs. Forecasts and scenarios under both exercises are not prophecy. Being adequately reserved or under-reserved does not change whether an entity’s borrowers can repay any more than Allowance practices control asset prices for underlying collateral. 

At this point, however, the material similarities end and the important divergences begin.

Audience and authority

DFAST is a supervisory activity, monitored by a regulatory authority and prescribed by an Act of U.S. Congress. The intended purpose of the stress test is to demonstrate institutional safety and soundness in the United States banking system; to that end, satisfactory submissions will be methodologically defensible and, ideally, demonstrate capital adequacy in specific circumstances. However, criticism of DFAST modeling efforts, assumptions, and preparation does not impact the realities of bank performance or users of financial statements.  

In contrast, CECL is an accounting activity prescribed by the FASB and part of U.S. GAAP. Its application will be evaluated by audit firms who answer to their own reputational risk, the PCAOB, etc. CECL estimation will impact the institution’s balance sheet ACL and income statement PCL lines, which are presented to investors and may have tax implications. Issues with the CECL implementation can cause material weakness findings in an audit.

While this contrast is generally well-understood, the implications are subtle. Putting the modeling efforts for CECL and DFAST in lock-step creates an environment where criticism of one function imperils the other. Considering the audiences for the exercises may have interests that are not in line – safety and soundness versus defensibility – the interprogram risk should not be taken lightly. 


The CECL standard for preparation of an Allowance for Credit Losses applies to assets reported at amortized cost. While for bank financial institutions this scope represents an entire side of the balance sheet, or nearly so, DFAST modeling must consider the entirety of both sides. In many cases, the modeling effort for DFAST may include autoregressive factors (outputs are predicted by prior period outputs), or correlations to other outputs, even on the other side of the balance sheet. Further, and perhaps most critically, loan production and retention must be correctly modeled for DFAST to produce a meaningful result, while these assumptions are explicitly disallowed under the accounting standard – mercifully, we do not have to allocate reserves for loans that have not been made.

Forecasts versus Scenarios

While both exercises have a forward-looking element, the Federal Reserve provides the model input “answers” in the form of the baseline, adverse, and severely adverse scenarios. Institutions must correlate these economic indicators to their portfolio performance, but are not responsible for defining the indicators themselves. CECL requires both – that important indicators are identified and predicted. The scenarios under DFAST are hypothetical, projections under CECL are intended to be likely. For that reason, forward-looking models employed under CECL should only consider, as inputs, factors for which the institution can create or source a reasonable and supportable forecast, regardless of the strength of correlation of the factor to the bank’s credit loss experience. As an illustrative but perhaps facile example, death and divorce rates may correlate to a bank’s losses – but it is unlikely the bank will want to be in the position of defending such forecasts to their auditor… as much as it would make financial institution MD&A more entertaining reading.


The difference here is straightforward – as of the date of publication, DFAST requirements apply to financial institutions with more than $10B of assets. CECL applies to all institutions with assets in scope, with different phase-in requirements and required adoption dates depending on an institution’s status as an SEC filer or Public Business Entity. Congress may act to adjust which institutions must perform the stress testing activity, but has been historically reluctant to address accounting standards. 

Contract versus Crisis

The CECL standard requires institutions to predict lifetime credit loss expectations for assets presently on the balance sheet, with the contract lifetime being adjusted either implicitly or explicitly for pre-payment behavior. In contrast, regulatory stress testing prescribes institutions compute at least a two-year outlook to determine the institution’s ability to survive a crisis of that duration. The synthesis of these differences is important, as it describes the most fundamental hurdle to program comparability – DFAST models and results assume some level of renewal and some level of asset production, while CECL implementation requires the opposite assumption. Depending on the implementation of the stress testing program and the ACL program, it may not be possible to “unwind” these effects. 


Garver Moore, Advisory Services at Sageworks.