Whether financial services organizations are prepared for the impact of the Current Expected Credit Loss (CECL) standard is a hot topic of conversation. The standard, issued in June 2016 by the Financial Accounting Standards Board (FASB) and set to take effect in 2020 for SEC registrants and 2021 for all other banks, represents a significant shift in how organizations account for expected credit losses. CECL requires that strong, validated models are in place to maintain the integrity of valuable data.
During a recent webinar we conducted, What Matters Now in CECL Model Validation, we asked the more than 200 attendees to gauge their readiness and comfort levels with the process changes needed within their organizations to meet the new CECL requirements. This was an audience made up of nearly equal percentages of credit risk and finance/treasury professionals, with a handful representing model risk management roles. We were not entirely surprised to hear a mix of responses: A few have gotten their CECL models in parallel runs, some expect to reach that milestone this year, and many expect to cross that bridge in 2019. Twenty percent of respondents were not able to say where they stand with their CECL readiness.
CECL Models and Methodologies
In a nutshell, the new CECL model requires financial services organizations to use historical information, current conditions and reasonable forecasts to estimate the expected loss over the life of the loan. The transition to CECL will have significant data requirements and require major changes to methodologies to estimate expected losses accurately. In addition, the new accounting standard has more rigorous allowance disclosure requirements.
Perhaps the most important questions to ask now are the following: Can the models our organization relies on today handle the forward-looking information required by CECL? Can our models (ALLL, Basel II, stress testing models) handle more than two years of data?
In addition to modeling decisions, methodology is also an important determination. There are several potential new methodologies for CECL compliance, which were covered in more detail during the webinar, accessible online. They include:
- Estimating expected credit losses using a loss-rate approach
- Estimating expected credit losses using a vintage-year basis
- Approaches based on Probability of Default (PD) and Loss Given Default (LGD) parameters
- A discounted cash flow approach
Institutions have the flexibility to use any of these approaches to estimate their credit reserves. Given a financial institution’s data availability, portfolio type, business requirements and model purpose, there are a number of questions to be answered in order to select the correct credit loss forecast methodology. For example, a credit card company’s most likely choice would be a loan-level model with logistic regression for PD and fractional logit for LGD purposes. In addition, payment rate models may be needed to estimate the speed of balance payoff. When selecting a methodology, it is important to keep in mind that data has to be highly repurposed for CECL.
We asked the webinar audience what methodology they expect to use for CECL estimation. The answers were as follows:
- Expected loss rate (18.7 percent)
- Vintage analysis (2.3 percent)
- PD and LGD modeling (13.5 percent)
- Cash flow approach (5.8 percent)
- Other or multiple methodologies (15.8 percent)
- Unsure (43.9 percent)
The high “unsure” response rate is not unexpected at this point in the CECL preparation timeline and reflects what we hear from clients. Each methodology has pros and cons, and many lenders have simply not yet tackled the groundwork needed to make these important methodology decisions. In fact, when we asked our webinar audience to rank their biggest challenge in validating CECL models, whether it’s data quality, CECL framework, methodology selection, CECL model or overall ALLL estimation accuracy, the majority (60.1 percent) selected “all of the above.”
Given the CECL deadline for public filers is less than 20 months away, we suggest starting with a gap analysis to identify where solid processes are in place and where change is needed. Once that analysis is complete, the real work of implementation can begin.