Stress test models – A framework for model rationalization

Stress test models – A framework for model rationalization

By Soner Tunay, Principal Director, Quantitative Analytics Lead, Accenture Consulting.

Ahead of Risk Americas 2018, Soner Tunay provided us with an insightful opinion piece.

Soner, can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?

Currently, I am leading the Quantitative Analytics in the Finance & Risk practice in Accenture. I am coming from a banking background. Until very recently, I was the Head of Risk Analytics at Citizensbank. Prior to Citizensbank, I lead model development and validation functions in various US banks, including an FBO and a G-SIB.

These days Banks are seeking a number of newer tools either to gain an edge in a competitive marketplace or derive efficiencies in a compressed spread environment. We at Accenture aim to be in the new by innovating and developing assets and capabilities to provide new applications. Among these, I can count modeling methodologies for CCAR and CECL, model execution platforms for CECL and Loss Forecasting, and applications of AI and ML in finance, including data lineage and triage tools, financial crime analytics and a variety of Robotic Process Automation techniques to deliver efficiencies across data, model development, model implementation and ongoing monitoring.

CCAR and Stress Test models in general are still the biggest and most influential changes in the modeling world in the last 20 years. We are constantly seeking ways to improve models and at the same time increase their applicability beyond regulatory compliance. Working closely with the business partners, covering everything in between the loan origination systems and portfolio management capabilities, we find many uses of the scenario-driven models, like the CCAR models.

At the Stress Testing USA: CCAR & DFAST Congress you shared your insight on ‘BAU applications of Stress Test Models’ – Why is this a key talking point right now?

In the last few years, Banks made sizable investments in developing CCAR models. At this point in the evolution of CCAR programs, we have a timely opportunity to turn these models and processes to drive better business decisions and integrate these risk models in day to day risk management activities.

Initial contribution of the CCAR models was primarily for regulatory compliance purposes in assessing the minimum capital requirements. After 6 years of intense work, most of the banks in the industry are in a better position with respect to their loss forecasting and capital estimation capabilities. But in our opinion, this was the first leg of the race. The second, and arguably the more challenging phase is to integrate these models for the business uses. There are certain adjustments needed to remove the conservative bias built into the stress testing process and to make the stress test models work just as well in the benign periods.

How can financial institutions best manage utilizing CCAR/DFAST models for multiple end goals?

As mentioned above, scenario-driven models present a fantastic opportunity to relate to business activities, including loan origination strategies, BAU loss forecasting, limit setting, capital management and oncoming change in the loan loss reserving methodologies with the CECL standards.

The first generation of CCAR models industry developed sit on top of the previous risk rating models, particularly Advanced Internal Ratings Based (AIRB) models. We often see multiple risk models which may or may not point to the same direction on the same obligors and the portfolio. Each model is built for a very specific purpose, but does not necessarily align with each other at the enterprise level. Instead, what we would like to see is that a unified model construct that can serve multiple end goals.

Our proposal is a framework to rationalize the risk models. This process would help reduce the number of models and streamline the existing ones into a seamless view of the risk.

Without giving too much away, can you outline the key challenges that arise when we have too many models?

Having many assessments of credit risk for a single obligor may create more risk for the business unit and the level and measure of that risk might change at every stage of the process, from underwriting to loss forecasting to capital assessment.

Developing alternative views of the risk particularly in the identification phase is a healthy risk management process. But the issue of having many credit models, we mention here can be explained when an input model and the user model does not have the same construct in terms of risk driver or risk segments. To further illustrate the model basis risk consider a situation, where loan origination a stand-alone model, but pricing is based on the Economic Capital model that utilizes PDs from the AIRB approach and finally, loss forecasting and strategic planning is carried out using CCAR models. This certainly looks like a case where risk management function is handicapped because of many risk models for the same portfolio. Model risk as described here could be eliminated with a structured approach to model rationalization.

Can you give a brief overview of the best practices in model rationalization?

Model rationalization is relatively a new concept. Some refer to it as model harmonization which is also an acceptable term.

We recommend a phased approach, where long term integrated model requirements and blueprints are drawn, but also short-term needs are serviced as well. Once the target state of streamlined models is defined with all possible uses and user groups, the end-state needs to be compared to the current inventory of models. The evaluation process could be based on materiality, age, cost of development, performance and coverage of the existing models. During the journey to the target state, a series of modify, new build and drop decisions need to be made. The goal of the model rationalization process is to minimize the number of models, build multi-use models and thereby reduce the model risk. We recommend a structured approach to achieve this goal that starts with the business buy-in, and leverages the efficiencies in integrated data and technology platforms.

What, in your opinion does the future hold for stress testing professionals, and how can they keep up with the increasing changes in the industry?

Stress Testing is still evolving since 2009 when it was first used to assess the capital adequacy with the Supervisory Capital Assessment Program (SCAP).

It has developed both in terms of the scenarios, models and capital planning purposes. While this progress might be satisfactory on the regulatory compliance front, there is so many efficiencies to be realized when the credit and revenue models are embedded in the business applications. This is a great opportunity for the stress testing professionals to continue to evolve their models around the business needs. Oncoming changes in the loan loss reserve methodology (CECL) will certainly be an additional catalyst. But we believe that even without the change in accounting standards there is plenty of opportunities to embed the Stress Test models in the day to day business making decisions. For example, scenario-driven loss models could be perfect choices for loss forecasting and strategic planning. These models would certainly be helpful in CECL process. Pre Provision Net Revenue (PPNR) models can aid the origination strategies and product offerings. And finally, scenario-driven loss simulation process could be a credible challenger to the current Economic Capital methodologies.