Featured Product

    New Impairment Model: Governance Considerations

    Although full CECL implementation is several years away, banks must begin preparing now to meet the impending requirements. In this article, we review some of the most important model governance considerations, including how to approach new modeling needs, key differences between models for CECL and models for AIRB and DFAST, and the differing expectations for less complex banks.

    As US banks prepare for new financial instrument impairment standard implementation of the current expected credit loss (CECL) model, governance in general and model governance in particular will take center stage. Because CECL will have a direct impact on current period financial statements, banks especially will need to ensure the impairment processes and models used in allowance calculations are appropriate for that purpose. Banks should give serious consideration to model governance standards in the Federal Reserve’s SR 11-7 supervisory letter and ensure that any models built for Basel advanced internal ratings-based (AIRB) or stress testing frameworks are not merely recycled for CECL estimates. Without appropriate challenge, validation, documentation, and auditing specific to the purposes of CECL, banks may not be able to demonstrate the appropriateness of the models for the new purpose.

    CECL Governance Now?

    It is still very early in the CECL implementation process, with a principle-based standard from the Financial Accounting Standards Board (FASB) issued and compliance deadlines three to four years out. Is it too early to raise the topic of model governance?

    Various organizations have already raised the topic of governance related to IFRS 9. The Global Public Policy Committee issued a paper on IFRS 9 in June 2016, opening the document with a significant section on governance and controls. We agree that banks should consider the models, governance policies, and processes that will be needed so they can plan both the model development and approval frameworks for CECL.

    But Do There Need to Be Models?

    For some firms, especially for small non-complex banks, the current allowance for loan and lease losses (ALLL) process may be simple and highly judgmental. Since the FASB and regulatory agencies have already indicated that there will be flexibility on the required sophistication of the models, depending on the size and complexity of the bank and portfolio, it could be argued that those banks could use non-modeled approaches to set the ALLL reserve.

    We believe that all approaches to estimating expected credit losses – even simple spreadsheet approaches – will likely be considered models. The existing model governance guidance from the Federal Reserve and Office of the Comptroller of the Currency (OCC), as laid out in SR 11-7 and OCC Bulletin 2011-12, provides a good perspective on how banks should approach the governance that will be required for any CECL estimates.

    Consider the following aspects of CECL in the context of SR 11-7 guidance:

    • Forward-looking CECL estimates: Technically, a forward-looking aspect is not part of the definition of a model. However, at the heart of the definition is a quantitative estimate of an uncertain quantity, and a forward-looking risk measure such as expected credit losses is by its very nature a quantitative estimate of an uncertain quantity.
    • Assumptions, data, and statistical and mathematical methods: Even simple historical averages of losses rely on assumptions and data. The assumption that future losses will be reasonably similar to the historical loss rate is key to this approach. Moving to the life-of-loan approach required for CECL will require stretching the assumptions around historical loss rates, or applying more rigorous statistical and mathematical techniques. All of these elements are covered under SR 11-7, and banks at the very least will need to demonstrate their assumptions and methods are appropriate for CECL estimation.
    • Materiality: CECL will affect current period financial statements, especially the income statement, but also the balance sheet through ALLL and the knock-on effect on retained earnings. Technically, materiality is also not part of the SR 11-7 definition of a model. But as a practical matter, models with important and significant uses are best treated as separate models on the inventory, so that appropriate model risk management isn’t impeded by an effort to accommodate a condensed inventory.
    • Appropriateness of parameter quantification: Any parameter quantifications used in CECL estimation have to be appropriate for the portfolio or individual loan in question. Even simpler methodologies using historical charge-off rates or expected loss (EL) rates should be appropriate to the life of loan loss estimation. More sophisticated methods employing probability of default (PD), loss given default (LGD), or exposure at default (EAD) will likewise need to be appropriate for CECL estimation.

    Another practical reason to have a separate model on the inventory is that all models under SR 11-7 require ongoing monitoring. Monitoring PD, LGD, and EAD separately, devoid of the connection as factors for use in CECL allowances, wouldn’t be appropriate.

    CECL Governance for Banks Subject to AIRB and DFAST

    Even Basel AIRB and stress testing banks will likely need to generate “new” models from a governance perspective. Although the AIRB and stress testing requirements overlap significantly with CECL requirements – all of them require an estimate of expected credit losses of some sort – the differences between the three directives mean that governance of the models will need to be tailored to the goals of the effort.

    Banks will need to ensure that CECL models are appropriate for their intended purpose, so it is likely that banks cannot reuse AIRB or stress testing models as they are. Basel or stress testing models may provide foundational elements for CECL, but there are material differences between CECL, the Basel EL calculation, the stress testing credit losses, and provision projections. The key differences include:

    • Credit loss horizons: Basel AIRB considers a one-year horizon, and stress testing considers a 13-quarter projection horizon. CECL will require banks to estimate an expected impairment value over the life of the loan. Banks will need to consider the assumptions that go into all the component CECL models and ensure that they are appropriate to the life-of-loan calculation.
    • Parameters (PD/LGD/EAD): Basel AIRB requires through-the-cycle (TTC) PDs and downturn LGDs and credit conversion factors (CCF). Stress testing expects point-in-time (PIT) or scenario-specific parameter values. CECL is closer to stress testing in that all parameter values should match the scenarios used, but banks should be wary of assuming stress testing parameters are appropriate for CECL. In particular, CECL requires a “life of loan” estimate of losses, and the parameter treatment will be particular to the context of each portfolio, or perhaps even each loan. In some cases, CECL may require PD and LGD curves to match loan cash flows over the life of the loan.
    • Scenarios: IFRS 9 requires banks to generate ECL estimates with consideration of current and potential future conditions. While there is not an explicit requirement for a scenario-based approach, it is likely that many banks will utilize their existing stress testing scenario framework in the CECL context. But we caution: The Comprehensive Capital Analysis and Review (CCAR) and Dodd-Frank Act Stress Test (DFAST) baseline scenarios (regulatory or internal) may not be appropriate or adequate for life-of-loan forecasting.
    • Conservatism: In both the Basel and stress testing frameworks, model deficiencies or limitations can be addressed by “topping up” the credit loss estimates, such as through management adjustments. With CECL, because the goal is to get to an expected impairment estimate, oneway adjustments (upward) may not be appropriate. With ALLL generally, the goal is to have the right amount of reserves (and quarterly provisions) rather than a generous buffer. In the current reserving framework, banks must justify their provision and reserves and ensure that they are not manipulating earnings. As guidance develops around CECL, it will become clearer what regulators will expect in terms of adjustments for modeling deficiencies. But at this point, banks should not assume that the conservative management adjustments for AIRB and stress testing can be applied in CECL.

    Generally, AIRB and stress testing banks will need to look at all aspects of CECL modeling separate from their existing frameworks. While many components of those frameworks may be appropriate in CECL, banks should check all assumptions for appropriateness in the new CECL context.

    CECL Governance for Less Complex Firms

    Less complex banks – those using the Basel standardized approach and below the DFAST threshold – will likely be able to use less complex approaches for CECL estimation. But SR 11-7 already recognizes that scale and complexity impact appropriate modeling approaches of risk estimation. For banks that are starting without AIRB or stress testing frameworks, the governance will require that they ask the same basic question: Are the assumptions, data, models (even spreadsheets), and overall process appropriate to the estimation of expected lifetime impairment of loans and leases?

    While these banks may not have experience in setting up the required governance elements, they should be able to draw on existing industry experience from the earlier Basel and stress testing efforts and modify what other (larger, more complex) banks have done to meet their own needs for CECL.

    Basic Considerations for CECL Governance

    Given the materiality of CECL numbers and the impact on a bank’s financial reporting, we expect that banks will need to develop governance programs that address the following aspects:

    • Appropriateness of data, methods, and models for CECL purposes
    • Reconciliation of portfolio positions
    • Benchmarking of CECL estimates against Basel and/or stress testing results
    • Sensitivity of models to assumptions and limitations, with adjustments appropriate for an expected measure of credit losses (not just adding a cushion for conservatism)
    • Understanding of models by senior management and the board of directors
    • Internal processes for challenge, validation, review, approval, backtesting, and outcomes analysis
    • Tracking of results from quarter to quarter, to understand movements in outcomes and whether they conform to expectations (i.e., sensitivity of models)


    For the CECL process and all of the modeling elements associated with it, AIRB and stress testing banks will need to take a fresh look at the methods, avoiding the assumption that models that were good enough for other purposes will meet the needs of CECL. And less complex banks that are building completely new frameworks will need to address these elements as well, even for the simpler approaches.

    Banks should recognize in advance the importance of two items:

    • Documentation: Key documentation needs to include model development, validation, model use and maintenance, and ongoing monitoring.
    • Policies: Existing policies for Basel or stress testing should be modified to meet CECL’s specific needs. Given the primary role of the finance function (and the chief financial officer) in banks’ ALLL calculations, policies will need to address the enterprise-wide nature of the CECL effort and clearly define authority, approval, and decision-making powers. Banks should consider what will be needed for both internal and external auditors to provide their opinion statements with regard to CECL estimates.

    Board of Governors of the Federal Reserve System. “SR 11-7: Guidance on Model Risk Management.” April 4, 2011.

    Global Public Policy Committee. “The implementation of IFRS 9 impairment requirements by banks.” June 17, 2016.

    Office of the Comptroller of the Currency. “OCC Bulletin 2011-12: Sound Practices for Model Risk Management.” April 4, 2011.

    Featured Experts
    As Published In:
    Related Articles

    Machine Learning Interpretability Techniques in Credit Risk Modeling

    We explore constructing and applying Machine Learning Techniques that are both transparent and interpretable. Also discussed are leveraging a model-agnostic perspective and demystifying and socializing chosen ML risk models.

    November 2019 Pdf Dr. Craig Peters

    Comparing DFAST 2014 Estimates for CCAR Banks Under the FRB's Severely Adverse Scenario

    This quantitative analysis of CCAR 2014 Severely Adverse scenarios, Moody's Analytics finds that the Federal Reserve Bank's (FRB's) and banks' own modeled estimates of capital ratios, revenue, net income, and loan credit losses are generally well aligned, although variations in all measures and across all banks are evident. In addition, the FRB's estimates are generally more conservative than those of the individual banks, reflecting differences in the FRB's industry-based models vs. the banks' portfolio specific models, treatment of missing or invalid data in the FRB's modeling approach, and assumptions about projected balance sheet volumes. The wide variation among bank modeled estimates and their overall alignment with FRB modeled estimates argues against banks targeting general industry benchmarks (such as average loss rates) and in favor of building models around their own business models and portfolio characteristics.

    July 2014 Pdf Danielle Ferry, Daniel Brown, Anna Krayn
    RESULTS 1 - 2 OF 2