As a crucial part of capital planning, Pre-Provision Net Revenue (PPNR) modeling has been under the spotlight for careful review and scrutiny from the Federal Reserve Board (FRB) since 2012 for the annual Comprehensive Capital Assessment and Review (CCAR) submission. Although bank holding companies (BHCs) have become better equipped at utilizing PPNR models for capital planning purposes in recent years, large and small BHCs continue to face challenges in meeting the FRB’s increasingly stringent requirements, as well as the banks’ own business needs.
In a recent white paper, we took a detailed look at two of the challenges BHCs commonly face when looking to develop and validate conceptually sound and robust PPNR models — challenges we frequently have encountered in our own work with clients over the past few years.
The first challenge is whether to choose a quantitative or a qualitative approach to PPNR modeling. At the early stages of the modeling process, it is important to determine whether the modeled PPNR components for any specific portfolio are sensitive to intuitive macroeconomic factors. Building quantitative models without first examining the nature and historical trends of PPNR components such as asset size, interest income, non-interest income, etc. could lead to costly model development cycles that fail the model validation process or receive scrutiny from regulators.
From our experience, one of the reasons for poor modeling results is that the components have a low actual correlation with macroeconomic variables. In theory, many PPNR subcomponents, such as asset balance, deposit balance and interest income, should be sensitive to changes in the macroeconomic environment and therefore, can be projected using quantitative models based on macroeconomic variables. In practice, however, whether a BHC’s PPNR component is correlated with the macroeconomic environment or not, the results depend on other factors, such as the bank’s business strategy, customer profile, footprint, product nature and segmentation granularity of the PPNR component. In most cases, a BNC will use a qualitative approach for some components (to project checking account balance, for example) and a quantitative approach for others (projecting the balance of a CD account, which depends on macroeconomic factors).
After determining the modeling approach for each segment, the next step of model development is the selection of variables. BNCs should take into account both qualitative and quantitative perspectives to select the final set of variables for the model. The explanatory variable selection process entails finding the balance between model specifications built solely based on statistical property by automated algorithms and incorporating key business and macroeconomic drivers that are intuitive and robust in both the short and long-term.
To achieve this balance, a BHC can start with a general variable selection process by applying automated variable selection algorithms to identify several combinations of statistically significant variables, then sharing those combinations of variables with business stakeholders and finalizing the selection based on business intuition and test model performance.
We suggest that modelers include business stakeholders in every stage of the development process — including the early stages. Through early-stage discussions, the modelers can learn the nature of the portfolio and the potential macroeconomic drivers of the portfolio based on the opinions from business stakeholders. Then, modelers can take into account business knowledge, along with the results of statistical analyses, to identify several groups of business-intuitive and statistically significant explanatory variables to further discuss with business stakeholders.
To learn more about Protiviti’s comprehensive Advanced Analytics practice, which includes our team of Ph.D.-level Model Risk Management professionals, visit our website or email us.
