By Nav Vaidhyanathan, Group VP, Head of Model Validation and Governance, M&T Bank
CECL 2019 is taking place in New York City on 27-28 March, 2019 – find out more here www.cefpro.com/cecl
Can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is? What, for you, are the benefits of attending a conference like CECL?
I have been in the role of developing or validating models for the last 13 years across various banks. Over last few years, my focus has been model validation. I have been managing model risk departments for the past 8 years. CECL has been the area of focus during the last year. Machine learning / AI are the areas that will likely consume a lot of our time this year and for the foreseeable future. Additionally, making model validation more agile to reduce bottlenecks and redundancy is another challenge that we are looking into.
CECL today is where CCAR was 4-5 years ago. Today, CCAR is business-as-usual (BAU) after going through years of iterative improvements. A conference on CECL is a step towards accelerating CECL to a business-as-usual state by leveraging industry best practices and understanding common challenges.
What key considerations need to be made when testing models for CECL use?
In most cases, CCAR models have been adapted for CECL use. The principle of conservativeness was dominant with CCAR models. This is not the case with CECL. Accuracy becomes more important than conservativeness. Additionally, with CECL there are several sources of uncertainty. These include – macroeconomic forecasts, reasonable and supportable forecasts, reversion, historical losses, and qualitative adjustments. Further, these sources of uncertainty do not exist in isolation. They are highly interconnected. While testing individual components are important, it is necessary to test the overall forecasts from all the components coming together.
How can financial institutions develop effective governance processes to manage model risk in CECL?
CECL is a quarterly exercise and for internal purposes a monthly one. Given this frequency, an effective governance process should not be too onerous on several participants across the organization. As this exercise is carried out from one period to the next, the governance processes can focus mainly on what material changes occurred from one period to the next. A comprehensive attribution analysis would help. A couple of areas where the governance process should focus on include – macroeconomic forecasts and qualitative adjustments. For other areas, including reasonable and supportable forecasts, reversion, and historical loss, an appropriate ongoing monitoring program with appropriate controls, triggers, and actions would help In limiting the governance required.
Can you give an overview to model limitations under CECL and how qualitative adjustments can be used?
All models have deficiencies and limitations. A framework for qualitative adjustments should be based on the principle of CECL neutrality. This means, if certain qualitative adjustments are made for model deficiencies or limitations, then after those limitations and deficiencies are remediated, the CECL estimate post remediation should be the same as CECL estimate pre-remediation with qualitative adjustments. Of course, the practical implementation of this is not that straightforward. An overly simplistic assumption can be that all model deficiencies and limitations are manifested in uncertainties captured in backtesting and ongoing monitoring. The results from backtesting and ongoing monitoring can become the quantitative basis for qualitative adjustments.
How can institutions validate assumptions and what impact can this have on model risk?
The thumb rule is – are assumptions appropriate for use and if they are, how are they be supported. Empirical evidence, testing, and governance supporting assumptions should be evaluated. Further, sensitivity and scenario analysis should be performed for key assumptions. This ensures that assumptions are valid for a range of scenarios. Additionally, benchmarking should be performed for material assumptions. The validity of assumptions should be monitored periodically. CECL is laden with assumptions outside of the traditional model but these assumptions impact the uncertainty from the model. For example, application of universal assumptions (for all portfolios) like – length of reasonable and supportable forecast, methodology for reversion, and historical loss methodology, life of loan, FIFO/LIFO, etc. could produce different impact for difference portfolios and models. Understanding these impacts will lead to remediating issues and/or accounting for them in qualitative adjustment.
How do you see the impact of CECL evolving over the next 6-12 months?
Models used for CECL will likely become the standard for other uses. There will likely be a convergence of models used for CECL, CCAR, budgeting and planning, risk based capital allocation, pricing, etc. Work will begin on 2nd generation of CECL models. There will likely be some convergence of approaches in the industry for reasonable and supportable forecasts, reversion, historical losses, and qualitative adjustments.