Reducing the Stress in Stress Testing

Completing the Comprehensive Capital Analysis and Review (CCAR) and Dodd-Frank Act Stress Test (DFAST) exercises in a timely, efficient, and accurate manner has presented US banks with a number of challenges since the inception of the capital plan review process in 2009. One of the more difficult aspects of the stress testing exercise is coordinating the inputs and outputs from various business units in a controlled environment that is auditable, repeatable, and allows for increased automation. The Federal Reserve Bank (the Fed), in their August 2013 paper, Capital Planning at Large Bank Holding Companies: Supervisory Expectations and Range of Best Practices, outlined some of the weaker practices at many CCAR banks, including a lack of coherent integration among credit loss models, pre-provision net revenue (PPNR) calculations, and the final FR Y-14A reporting. This common theme was again highlighted in their response to the 2014 submissions, noting that the stress tests and capital plans still lack sufficient detail and explanation.

In summarizing the most recent round of stress test reviews for 2014, the Fed emphasized the need for banks to develop transparent, repeatable, and well-supported processes to generate the stress test results. Recent guidance suggests the Fed wants banks to transform the bi-annual stress testing process into a tool for continuous risk management. Furthermore, the Fed has recently issued Matters Requiring Attention (MRAs) and Matters Requiring Immediate Attention (MRIAs) to banks that currently fall short of the Fed’s expectations from a process and governance standpoint. In order to integrate the evolution of credit defaults, losses, and recoveries with the associated impact on net interest income and other non-interest revenues and expenses, banks have developed several approaches.

Moody’s Analytics works with many CCAR and DFAST banks to solve process challenges related to their stress testing and capital planning programs. Financial institutions of all sizes told us that their most pressing concern is to further improve their conditional loss modeling, which they would like to do in a stress testing environment with an auditable and flexible workflow. We have also found that most clients seek assistance with idiosyncratic scenario design, regulatory reporting, or infrastructure related projects (e.g., scenario, model, and business process management and integration). Our experience thus far suggests that, while a number of best practices for managing the CCAR/DFAST process are starting to emerge, the industry overall is still in the nascent stage of developing a full-scale approach to stress testing automation.

Figure 1. How do banks use their stress testing results?

Source: Moody's Analytics

This inaugural survey serves as both an early benchmark and a stimulus for ideas on how institutions can improve their information technology to build a more sustainable, repeatable, and efficient stress testing process, reduce manual processes, maximize their investment in staff and resources, and utilize the results for strategic growth.

One of the more difficult challenges facing banks during the stress testing exercises is the need to standardize, integrate, and reconcile data from both finance and treasury. Underdeveloped business processes, technological challenges, and poor coordination between functional groups conspire to complicate the exercise. A common theme among the banks we surveyed was that a stress testing platform must have losses and cash flows computed at the instrument level (bottom-up). Some banks use a top-down approach because it is quicker to implement, but the bottom-up approach is the preferred methodological target for modeling the impact of stresses on a bank’s portfolio. The top-down approach is inevitably a less sophisticated modeling technique, and cannot identify the idiosyncratic risk factors driving stressed losses or capture non-linear loss reaction to macroeconomic stresses across the portfolio. However, banks reported that they are looking to model their portfolios at the loan-level and thus need more granularity in the data driving their forecasts.

The majority of banks use their ALM system to forecast balances. However, many financial planning systems don’t possess the granularity or calculation rigor of the ALM systems, or simply absorb as input the output of the ALM platform. While this is not the universal standard, the majority of banks we talked to viewed both the ALM and financial planning and analysis (FP&A) systems as critical components of the CCAR process, and expect the systems will have to reconcile at some point in the process.

Advanced hierarchy management across numerous systems, however, along with the data required for bottom-up loss estimation and PPNR modeling, remain elusive. Finally, reconciling data between the ALM, FP&A, and credit modeling systems is also an aspect with which banks continue to struggle.

Forecasted cash flows are a key component of the data required to perform regulatory stress testing. However, the ALM systems utilized by the banks we surveyed are typically not able to project forecasted cash flows at the same granularity as their current loan portfolios. Most banks use their ALM system for the forecasted balance sheet and income statement in order to support the FR Y-14A output. In each case to date, all credit work occurs outside of the ALM models. Programs are often utilized as part of the credit conditioning of forecasted cash flows. When forecasting new business volume, the most common case is that assumptions are qualitatively estimated using data sourced from an FP&A (top-down aggregation) system. New business assumptions are then incorporated into an ALM system to support net interest income calculations. Credit assumptions are provided to the ALM system via spreadsheets. There is recognition of a need to move toward credit-adjusted new business volumes. New business volume assumptions are not often “credit” adjusted, and banks generally assume new business adopts the current position risk profile.

Current ALM systems are not adept at handling the credit modeling portion of the exercise. Several banks we surveyed would like to incorporate credit-adjusted cash flows into their stress testing engine, though many have not chosen a solution that can aggregate both sets of models, along with the requisite data inputs and results.

A majority of the banks surveyed indicated that they are generally reluctant to replace or add duplicative systems to their current processes and infrastructure. This is understandable given the heavy amount of investment they have made in recent years in ALM, FP&A, and credit modeling systems. However, they need to consolidate data, develop workflows to manage and audit the process, and extend the bank’s internal and external reporting capabilities, all while utilizing elements of their current infrastructure. In some cases, ALM systems are being used as a final aggregation point for reporting. However, this can lead to stress testing and workflow process constraints. Other banks report that they are attempting to use their FP&A system as a final aggregation point, though again with constraints on data granularity and auditability. Overall, banks remain interested in the integration of risk with balance sheet management. They are, though, still looking for potential solutions to emerge, which also include data management, model and scenario management, auditability, and regulatory reporting features.

There is an emerging need to more tightly integrate the data handoffs between multiple bank systems throughout the stress testing process. One example of the need for transparent, auditable workflows is within the forecasting of the balance sheet. Most banks opined that in today’s environment, a common practice is to hold the balance sheet composition (i.e., the “mix”) constant in the various scenarios, although the overall size of the balance sheet may be reduced in the stress scenarios. Various “views” of the balance sheet will be passed to credit risk or credit administration, who will then estimate non-performing asset levels in the various scenarios, create the conversion of economic losses into charge-offs, and pass these assumptions back to either finance or treasury for inclusion in reporting. ALM is often the source system for prepayments as required by various schedules, and for forward-period valuations. In some cases, these assumptions are passed through various committee structures, such as the loss forecasting committee, credit portfolio management committee, or similar governance structures.

New business volumes don’t appear to generally contain a credit dimension, although most banks agreed that this would be a welcome addition to their modeling framework. In some cases, pro-forma RWA calculations are also being performed within the ALM system. It is more common, however, for banks to pass a detailed output file to the capital team for estimation of RWAs. Given that balance sheets are generally held constant, RWAs change in accordance with balance sheet shrinkage, not in the nature of a shifting risk distribution. With the Fed now estimating balance sheets and RWAs for each of the CCAR banks over the nine-quarter forecast horizon, this is a practice that many banks plan to refine.

Managing the workflows between the risk, finance, and treasury groups presents a tremendous challenge. A majority of banks would like to move away from “spreadsheet risk” and toward an integrated system that can more efficiently automate and document the assumptions utilized throughout the process.

Most banks have not implemented – let alone developed – a plan for integrating their operations with an end-to-end stress testing solution. Some are building roadmaps for integration, but are not yet certain what the end-state will look like. This is largely because most banks have not adopted the stress testing exercises into their capital planning and business management processes on a consistent basis. The industry is still early in the process of determining what the end-state for a stress testing solution will look like. At this point, a few emerging trends are starting to take shape, such as granular data management, integration of cash flows and credit models, the leveraging of current infrastructure, and automated workflows to manage the process.

Many banks are beginning to develop a “roadmap” or “glide path” toward building a more automated, controlled, and transparent stress testing platform to provide more consistency in the process. This involves taking manageable steps in the near term to automate certain FR Y-14A schedules, while looking for ways to leverage IT to use stress testing for more strategic business planning purposes in the medium term. This approach, while highly practical from a resource-planning standpoint, has the added effect of demonstrating real incremental progress to the regulators.

Figure 2. Does your bank have a stress testing roadmap?

Source: Moody's Analytics

The Fed again noted the need for banks to develop transparent, repeatable, and well-supported processes to generate the stress test results, and outlined a lack of coherent integration among credit loss models, pre-provision net revenue calculations, and the final FR Y-14A reporting as one of the weaker practices industry wide.

This is consistent with what financial institutions of all sizes told us. While their most pressing concern is improving their conditional loss modeling, they would like to improve their modeling capabilities in a stress testing environment.

Regulators have indicated they will look at a number of areas as they continue to refine their focus for the stress testing and capital planning exercise. These include expecting more detail in the banks’ capital policies, better documentation for model assumptions, and adjustments and explanations across the process, from their models through their idiosyncratic scenarios and capital plans.

To continue complying with the evolving regulatory requirements and making their stress testing processes more efficient through automation, banks may spend significantly on infrastructure over the next few years. To get more out of the stress testing exercise and use the results for more than just regulatory compliance, banks need speed and repeatability – which they will not get without revamping key elements of their existing infrastructure. As stress testing becomes more ingrained within the bank’s current risk and finance infrastructure, the process should become less “stressful” and more of a consistent application for capital planning.

• A stress testing platform must have losses and cash flows computed at the instrument level
• Forecasted cash flows are a key component of the data required to perform regulatory stress testing
• Banks would like to incorporate credit-adjusted cash flows into their stress testing engine
• Banks are generally reluctant to replace or add duplicative systems to their current process and infrastructure, but remain interested in the integration of risk with balance sheet management
• A majority of banks would like to move away from “spreadsheet risk” and toward an integrated system that can more efficiently automate and document the assumptions used throughout the process