By Raj Kunwar, Director, Model risk management, Bank of America and Guoning Yang, Director, Quant Analytics, Fifth Third Bank.
Guoning, can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?
I’m responsible for credit risk modeling and analytics for Stress Testing and CECL of Fifth Third Consumer book (Auto loan, Credit Card, Home Equity, Mortgage and Personal Lending). I oversee models and analytics for long-term loss forecasting and stress testing, as well as production, submission and regulatory review support of stress testing. Starting this year, I was assigned to lead Consumer CECL modeling in partnership with Accounting, Finance and Capital Planning. Before Fifth Third, I worked at Capital One leading loss forecasting modeling and Basel modeling of their credit card portfolio.
Raj, can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?
My current focus is in validating high materiality retail credit risk models. My team is responsible for model validation of the auto loan, credit card, mortgage and credit lending models . In the past I have worked at HSBC in developing PPNR revenue and loan balance forecasting models for both the banking and trading book. Prior, to that I have worked in the Enterprise modeling center of excellence at GE capital primarily developing PPNR and credit risk models and validating treasury market risk models.
At the CECL Congress 2018, you will be speaking on your insight regarding, ‘Considerations for CECL model approaches: New approaches vs. updating internal infrastructure’ Why is this a key talking point in the industry right now?
Guoning: CECL accounting standard is the most impactful accounting change in over a decade for the banking industry and will continue to transform the way banks operate business in the future. From this perspective, banks may want to consider new approaches that reflect the fundamental changes associated with CECL. On the other hand, many banks have developed advanced approaches that can be leveraged or even directly adopted for CECL purpose over the years fulfilling Basel, Economic Capital, CCAR and DFAST. For efficiency and consistency, these banks may want to update the existing approaches for CECL. Therefore, banks need to be strategic with their decision on approaches, balancing factors like what’s the anticipated regulatory expectation on the bank, how’s their current approach compliant with CECL and is the remaining gap manageable through quantitative or qualitative adjustments, is their current approach robust enough for CECL tweak, is there any existing regulatory or managerial concerns on the current approaches that may aggravate if used for CECL, any road blocks prevent the bank from new approach such as data limitations, and etc. Of course, underneath all these factors are the banks’ portfolios. Some portfolios, such as short-term loans, may be indifferent to the change from the old to new approach while others, such as credit card, may result in crucial differences. Therefore, this is a very complicated consideration. I hope the discussion will help the peers to assess their position as well as meet regulatory expectation.
Raj: The FASB’s new accounting changes to credit losses will require additional capital and therefore it requires careful consideration towards loss forecasting approaches and infrastructure. The new FASB ASC 326 requires a move to allowances based on “current expected credit losses” (CECL) affecting banks along with both public and private insurers and financial institutions. The implementation date is at the end of 2019 for most large institutions and therefore it requires more discussion among the industry players. Many major banks are planning to leverage existing CCAR, DFAST, Basel models towards full filling CECL requirements. For these banks the focus is on identifying key areas, gaps and allocating resources to full fill these requirements. Organizational silos, limited data capacity/constraints, and limited resources are some of the key challenges that needs to be efficiently managed towards a successful CECL implementation.
In your opinion, what do you think the limitations would be going forward?
Guoning: Like for all modeling, the new standard may impose new challenges to data requirements, especially since CECL considers credit risk over the life of loan, which implies longer period of data needed and model/assumption of prepayment behavior. Therefore, I do expect data availability to define the adopted modeling approaches for some banks. In addition, CECL standard is more prescriptive than CCAR/DFAST. This increased expectation may limit the flexibility to tailor the approach for business use. Some workaround solutions in stress testing may not be CECL compliant. Another limitation is IT infrastructure, which may lag behind CECL standard for some banks. This includes data sourcing, extract, and construction, ECL computation and CECL reporting in a timely manner.
Raj: One of the major limitation will be data quality and data infrastructure. Presently, many major banks have separate risk and finance data systems that do not communicate well with each other. Since CECL is in the cross hair of risk and finance it will require better interaction between the two systems. Most of the banks will find this a big limitation in developing CECL model forecasts. Updating internal infrastructure will be a necessity towards CECL implementation. Back testing models for life time loan loss forecasting use is also going to be challenging and may require banks to leverage industry data in addition to internal data.
What are the key considerations that need to be made when examining the preliminary results under each approach?
Guoning: I think the key considerations should reply on CECL standard and CECL compliance should be the overarching criterion assessing the alternative approaches. In addition and secondary to CECL compliance comes the business reasonableness and model performance, which add incremental perspectives. Last but not least, since the intent of CECL is to address the too little too late weakness of the current ALLL framework, I would think twice if any approach generates reserve lower than the current ALLL.
Raj: CECL requires life time loan loss estimation upfront and front load it in the process. This requires a better integration between the various data systems that provide the input variables to the calculation engine and communicates back to the various data reporting systems for analysts to calculate the final numbers. The approaches have to be well integrated to the existing bank infrastructure for it to work in a sustainable manner. There will be upfront capital investments for developing the model, data, process, project planning and in addition there will be an ongoing cost associated with running the BAU process. Different banks depending on their individual needs will find one approach more suitable than the other. At the end all the key CECL requirements needs to be met in a sustainable way.
Can you provide a brief overview of reconciliation and explanation of differences?
Guoning: We leveraged the existing CCAR PD/LGD approach but rebuilt some models to focus on accuracy under various economic conditions instead of sensitivity to stressed economic scenarios. In addition, prepayment models were built to enhance the existing framework to capture the behavioral life of loan. Alternative approaches were tested for the expected credit losses beyond reasonable and supportable period. In addition, discounted cash flow approach was also considered and the benefit from discounting was assessed. There are noticeable differences across the different approaches and we do observe a wide range of ECL estimates but they are consistent with the methodological distinctions.
Raj: We leveraged our existing PD/LGD approaches for life time loan forecasts, we plan on using some of the vended products for near term forecasting and extending them for longer term benchmarking purpose. The back testing becomes challenging for longer horizon forecasting with limited historical data. Different approaches seem to work in near term versus longer horizon forecasting. We might need more than one approach to cover the full life time loan forecasts.
Finally, what challenges do you foresee with CECL implementation over the coming years and how can institutions best plan to meet deadlines?
Guoning: The data challenges will remain for a while and assumptions may be needed to work around the challenges. While institutions can continue to refine approaches to mitigate the reliance on these assumptions, new data should be collected starting now and a gap analysis on a regular basis is necessary to assess data readiness so that when the time comes, the institution can upgrade their approaches to rely on the data instead of the assumptions. Another challenge is regulatory expectations, which remains unclear for some CECL areas.
Raj: Several challenges come to mind first is the data requirements, model development, limited quant resources, project planning, governance and control process. Banks will have to plan out the complete CECL development and ongoing execution in order to carry out a successful CECL implementation. There is a whole industry learning curve that individual institutions will have to undertake in order to satisfy CECL regulatory requirements.