Muneera Carr | EVP, Controller and Chief Accounting Officer | Comerica Bank
Dan Harach | Vice President and Portfolio Analytic Manager | Western Alliance Bancorporation
Of the National ALLL Conference’s vast array of perspectives on CECL, Muneera Carr’s and Dan Harach’s were uniquely that of preparers. As financial professionals, they look at CECL from a practical standpoint, addressing what they should be doing to prepare for CECL. They shared their preparations, discussed relevant considerations and examined some of what are certain to be unanticipated consequences of CECL.
Muneera Carr – Comerica Bank
- Once the standard is released, we will plan how our groups are going to come together and we’ll start our models.
- From our modeling exercises, in some circumstances our reserves would not go up under CECL and that was a surprise.
- Over – segmentation and too many models are problematic.
Dan Harach – Western Alliance Bancorporation
- Don’t waste any time before starting your CECL preparations. Begin now with a high level plan, identify key milestone dates and organize a working group within your institution.
- Our focus is on data management, aligning portfolio data with stress testing model data, performing some high level model testing and taking a fresh look at our portfolio segmentation.
- Start amassing as much data as you can – having the data gives you the flexibility to determine which models might be most suitable for your portfolio.
On initial reactions: We were hoping that FASB wouldn’t make us recognize losses from day one and over the life of the loan, but have accepted the fact that it is going to
be that way.
On progress in preparing: We don’t want to front-run it too much; just paying attention and preparing, picking up some steam with internal discussions in recent months. We rub shoulders with the largest banks. Everyone seems to be waiting for someone else to do something – looking for some magical solution.
How formal is your planning? Our auditors gave us a questionnaire to complete. They want to know how we’re going to get there. A formal plan yet? No, but once the standard is released, we will plan how our groups are going to come together and we’ll start testing models. Our modelers say it will take about a year to change from our current models. In the past month, my policy team went back 20 years to create a blend of reality and assumptions. Our loans average 2.5 to 4 years in length, so we looked at the last 20 years to see how they behave over every four years and use that to build our models. It was different from what we expected. We found that in certain circumstances our reserves will not go up under CECL and that was a surprise. Our LEP is nine quarters, so we’re using nine for all loans relative to life of loan; do the simple math and see where it takes you. For future expectations, after two years, go to your mean, that is, normal charge-offs.
What about segmentation? My SBA loans behave one way, so pooling helps. Over segmentation and too many models are problematic. Will we reconsider our segments? Yes, and use cluster analysis to see commonalities – but we don’t want to go too far to too many different segments.
Will we use early adoption? We will analyze our numbers so we know what they are and can make sense of them, that is, the difference between what the mean was and our expectation of the future. We don’t plan to but could be forced to.
What about economic forecasting? Our chief economist believes one thing, the Fed something else, others something else, so we often choose to use different forecasts to achieve different purposes. We will have to decide where our forecasts will come from. Our chief economist is not involved today.
Overall, we are sure to come across new challenges that we’re not aware of today, so we’ll try things out, and when we get questions, we’ll get them in front of the TRG (Transition Resource Group).
What are you doing to prepare for CECL? We’re going through our multiple data sources, designing processes and creating primary keys to bring information together and making sure the information is quality data that can be leveraged for modeling. Initially our focus is on data management, aligning portfolio data with stress testing model data, then doing some high level model testing and eventually building out some form of a CECL model. It’s all about data at this point. Within our organization, when senior management hears the prolonged implementation timeline, they lose interest. At this point we have put together a working CECL group and are addressing expectations year to year.
What about modeling? We currently use a net charge-off model as our primary ALLL model. Our portfolio through the last economic cycle is so different now than it was – different product lines, different geographies and organic growth that was augmented with a few bank and portfolio acquisitions. We did do some preliminary model testing using some of the recommended models outlined in the CECL guidance from the regulators. Under a vintage scenario, our reserves would have been less than current levels. You really have to spend the time and evaluate which model best suits your bank or credit union – and leverage what you can. We thought we’d use bottom up models for stress testing and got pushback from our regulators. We’re doing a lot of testing of various models – top down and bottom up. We’ll probably bootstrap to our
DFAST PD/LGD models.
What about early adoption? If we did it would be to enhance the sophistication of our modeling efforts – upsizing from a net charge-off model to a PD/LGD model. This may become prevalent given the significant changes in our portfolio over the years. We will increase staff for modeling but will probably not be looking to add an economist. To supplement econometric information, we’ll look to vended solutions.
What do you advise others to do? Start amassing as much data as you can. You probably have it or can find it: loan level charge-offs, recoveries, and loan level default triggers. Once you start bringing it all together and aligning portfolio data with credit risk data, you can provide value to your company in the model selection
and implementation process. Once the data is aligned, it will be a lot easier to start testing various portfolio segmentations under different model types. You should
also spend the time to get acquitted with the final rule and craft a plan that encompasses data gathering, model selection and model implementation.