promontory.com INFOCUS OCTOBER 12, 2018 BY WILLIAN LANG WITH RYAN CHAREST A Fundamental Shift in Models Used for Estimating Loan-Loss Reserves The new U.S. accounting standard for current expected credit losses represents a fundamental change in financial institutions approach to estimating required loan-loss reserves and provisions. The Financial Accounting Standards Board created the CECL standard as a forward-looking measure for recognizing potential credit losses for loans and debt securities. CECL will generally take effect in 2020 for Securities and Exchange Commission registrants and in 2021 for financial institutions that are not registered with the SEC, with early adoption permitted as soon as 2019. William Lang is a managing director at Promontory and is an authority on financial risk management, quantitative analysis, and bank supervision and advises Promontory clients on stress testing, model validation, risk measurement, and capital planning. A central goal of CECL is to mitigate the procyclicality inherent in the current approach to loan-loss provisioning. Over the past 40 years, the allowance for loan and lease losses has been measured using the incurred-loss concept, where the ALLL reflects existing impairments to outstanding loans. Under CECL, the ALLL will be measured as the difference between the financial assets amortized cost basis and the net amount expected to be collected on the financial assets (i.e., expected lifetime credit losses). Thus, CECL reflects existing impairments and expected future losses for the portfolio. In addition to managing this accounting change, institutions will need to grapple with highly technical model-development and model-validation challenges. Specifically, the new standards require thoughtful identification and review of suitable sources of data; selection of appropriate modeling methodologies; as well as appropriate governance, documentation, challenge, and review of models and tools used to generate loss estimates. While there are some similarities and potential synergies between stress testing and CECL, there are significant differences and institutions must be alert to the unique challenges that CECL may pose. The Importance of Getting CECL Right The ALLL and loan-loss provisions are critical components of an institution s financial statements that impact net income and capital. Provisions are an expense that directly affect net income. Because the ALLL is a component of tier 2 capital, increases in the estimated ALLL reduce tier 1 capital and common equity. 1 Regulators are concerned about the appropriateness of the ALLL. The SEC is focused on accuracy of financial statements, and insufficient or excessive provisioning raises red flags. Financial institutions prudential supervisors also expect an appropriately estimated ALLL, but are typically more concerned about potential under-provisioning that can threaten safety and soundness. If a regulator 1 There is a cap on the amount of the ALLL eligible for recognition in tier 2 capital. For banks subject to the cap, an increase in the ALLL will also reduce total capital.
Since CECL represents an increase in complexity over the existing method for calculating the ALLL, regulators will be closely scrutinizing the risk management of models used to forecast losses under CECL. determines that the ALLL is inappropriate, accounting restatements, supervisory fines and sanctions, and potential litigation can result. A central component of determining the appropriateness of the ALLL is an assessment of the modeling methodologies used in the process. 2 Since CECL represents an increase in complexity over the existing method for calculating the ALLL, regulators will be closely scrutinizing the risk management of models used to forecast losses under CECL. Supervisors will expect strong model governance including independent model-validation activities that is proportionate to the risk and complexity of each model. Model Governance and Risk Management Under CECL The CECL standards require financial institutions to produce a statistically and economically sound forecast of credit losses. The models financial institutions use to implement the standard will fall under the Federal Reserve Board and Office of the Comptroller of the Currency s Supervisory Guidance on Model Risk Management. 3 A central theme of this guidance is the need for a function dedicated to model risk management (inclusive of model validation), which is independent of model developers and provides a strong effective challenge. Internal audit, in turn, must be independent of both, in accordance with long-standing practice and supervisory guidance. In practice, financial institutions typically employ a three-lines-of-defense framework for managing model risk. In this framework, model owners, developers, and users are the first line of defense; independent model risk management is the second; and internal audit is the third. Synergies and Differences Between CECL Estimates and Stress-Test Loss Estimates Under Baseline Conditions Frameworks for CECL compliance and stress testing are inherently similar, with overlap in the data used and estimates derived. For that reason, many financial institutions are leveraging capital stress-testing models such as those used for Dodd-Frank Act stress tests for CECL. Larger, 2 Consistent with supervisory guidance, this article uses the term model to refer to a quantitative method, system, or approach that applies statistical, economic, financial, or mathematical theories, techniques, and assumptions to process input data into quantitative estimates. The definition of model also covers quantitative approaches whose inputs are partially or wholly qualitative or based on expert judgment, provided that the output is quantitative in nature. 3 Supervisory Guidance on Model Risk Management, Federal Reserve Board and Office of the Comptroller of the Currency (April 4, 2011). The Federal Deposit Insurance Corp. adopted this guidance on June 7, 2017. PROMONTORY Sightlines InFocus OCTOBER 12, 2018 2
more complex financial institutions are anticipated to base much of their CECL loss forecasting on granular loan-level information and follow a construct similar to those used for their baseline forecasts in the annual and semiannual Comprehensive Capital Analysis and Review exercises. Midsize financial institutions are likely to base much of their CECL loss forecasting on pool-level information and to follow a construct similar to those used in their DFAST models. The CECL standard does not require any particular forecasting methodology. Multiple modeling constructs (including those for expected loss, rating transition, roll-rate, historical or empirical, charge-off, and discounted cash flows) will often be used at a single financial institution, with the choice of construct dependent on considerations such as data availability, segment materiality, differences in loan characteristics, supervisory expectations, and industry convergence. This is especially true for larger, more complex financial institutions. Both CECL estimates and stress-test results under baseline conditions produce estimates of expected losses. Moreover, CECL models should be used to estimate the ALLL under the various economic scenarios used for stress tests. As part of effective model governance, supervisors will expect institutions to be able to explain the drivers of these two alternative measures of expected losses, particularly when there are significant differences in loss rates across the two measures. Some important conceptual considerations for this analysis include: Principle of conservatism in stress testing: The design of stress-testing models emphasizes the quality of results under stressed conditions and includes a principle of conservatism. This can lead to reduced accuracy (i.e., overestimates) of baseline loss forecasts. Modestly higher baseline estimates than for CECL would not be unexpected. Inherent differences in forecast horizon between stress testing and CECL: Stresstesting models are constructed to produce losses over a nine-quarter horizon, while CECL requires lifetime losses. Financial institutions should consider comparing their stress-test results with CECL results that truncate losses at nine quarters. Static (CECL) vs. dynamic (stress test) portfolios: CECL only considers loss estimates for the existing portfolio of loans and securities, while stress testing considers future originations. Therefore, comparative analysis should focus on portfolio loss rates, rather than total losses, and consider the characteristics of new originations relative to the existing portfolio. Differences in economic scenarios: CECL forecasts must incorporate an internal assessment of the economic environment as of each quarter, while stress-test forecasts must use prescribed supervisory scenarios. 4 Financial institutions can use sensitivity analysis to determine the impact of the different scenarios on their estimated losses. Differences in reporting frequency: CECL calculations are quarterly, while stress-testing estimates are annual or semiannual. Differences in segmentation granularity: CECL requires segmenting portfolios into loans with similar risk characteristics, which may mean greater granularity than that offered by some DFAST models. Financial institutions can conduct sensitivity analysis to determine the impact of granularity on their loss estimates. 4 SR 13-1/CA 13-1: Supplemental Policy Statement on the Internal Audit Function and Its Outsourcing, Federal Reserve (Jan. 23, 2013). PROMONTORY Sightlines InFocus OCTOBER 12, 2018 3
In the years since the 2008 financial crisis, expectations for model risk management including independent model validation have increased substantially and continue to increase. Financial institutions, in turn, have made significant investment in model-governance and model-validation activities. CECL s additional accounting and audit requirements: Unlike the stress-testing application of loss forecasting, CECL results will be audited and Sarbanes-Oxley Act controls will be required now that the results directly affect public financial statements. The bar for internal controls and documentation will be even higher under CECL than the already high standards in stress testing. CECL-Specific Considerations in Model Risk Management In the years since the 2008 financial crisis, expectations for model risk management including independent model validation have increased substantially and continue to increase. Financial institutions, in turn, have made significant investment in model-governance and model-validation activities. CECL s fundamental shift in provisioning brings a unique set of challenges for managing model risk: Many financial institutions will employ multiple CECL approaches across loan segments, adding to the complexity of the overall validation exercise. Vintage-disclosure requirements will require enhanced data-collection and reporting requirements across all financial institutions. The depth and scope of MRM activities will need to be tailored to the supervisory expectations surrounding the size and complexity of the financial institution. The depth and scope of risk management activities will need to be proportionate to model complexity, as implied by the selected methodology, which may vary by segment. The lifetime-loss concept underpinning CECL requires additional conceptual considerations, including: The sensitivity of loss-rate estimates to the following calculations, assumptions, and calibrations: Seasoning effects Vintage effects Amortization effects Prepayment and effective-maturity effects Discount rates PROMONTORY Sightlines InFocus OCTOBER 12, 2018 4
Establishing a supportable approach for complying with the CECL requirement that forecasts of lifetime losses are reasonable and supportable. The different lifetimes associated with different loan types (e.g., revolving lines of credit and term loans). Approach to Model Validation Validation is a key component of effective model risk management. Validation processes that support effective model risk management emphasize three components which financial institutions can incorporate into their model validation and reviews ensuring that each is covered appropriately and aligns with the expectations for CECL: Developmental evidence and documentation: Developmental evidence should describe key decisions and highlight both strengths and potential weaknesses of the models and scenarios selected, in comparison to viable alternatives. Assessments of developmental evidence for CECL forecasting focus on data selection, dependent-variable definitions, variable selection, model selection, assumptions, and scenario design. Process verification and testing: Financial institutions will benefit from testing to determine whether models and data systems are functioning as designed, and whether performance is consistent with expectations. Such testing should be designed to confirm that internal processes are structured to operate in ways that promote the overall integrity of risk management. For example, processes should not be unduly reliant on manual intervention. Model-specific performance tracking (i.e., outcomes analysis): Financial institutions and their third-party advisers can review performance-tracking and -reporting frameworks to determine whether models are performing as expected and to ensure that appropriate actions are linked to potential or actual performance triggers. Back-testing may be used for certain components of CECL modeling frameworks; but data from historical ranges, including suitably severe periods, is rarely sufficient to compare predicted to actual outcomes. As a consequence, alternative methods are often used in lieu of traditional back-testing. Comprehensive model-validation exercises typically include a series of key documents and activities: A thoughtfully selected set of validation tests and review activities that are commensurate with the model s risks, including review of documentation, developmental evidence, implementation or process verification, benchmarking, and outcome-based performance analysis A model-validation report that summarizes all validation activities, evaluates the model with respect to these validation activities, and provides a well-organized summary of findings and issues Model grades and issue ratings that reflect a considered view of the impact of all issues identified during the validation that may affect a model s suitability for use Discussion of findings with model owners and/or developers PROMONTORY Sightlines InFocus OCTOBER 12, 2018 5
Building a Strong Foundation for CECL Compliance Compliance with CECL requires paying careful attention to new technical accounting issues; but it also requires careful attention to identifying suitable sources of data, choosing appropriate forecasting methodologies, and the governance of CECL estimates. A critical component of governance is independent model risk management that encompasses independent validation. Model validators must be attuned to the unique features of CECL models, understand how these models operate, and be prepared to bring the appropriate level of rigor to the validation regardless of whether a particular forecast uses discounted cash flows, loan-level or pool-level expected losses, cumulative charge-off rates, rating transitions, or roll rates. Validators will face the additional challenge of performing the necessary reviews of vendor-supplied models, which may lack the transparency of internally developed models. To meet such challenges, validation resources must understand what examiners are likely to expect of financial institutions implementation of CECL. Given the extensive changes that CECL requires for models to support an appropriate ALLL, financial institutions need to move quickly to ensure they have a strong approach to managing CECL model risk. This will mean engaging independent model risk management early in the process to establish the requirements for model developers, owners, and users. This effort will include establishing checkpoints for various issues including assessments of data adequacy, reviews of internal controls, evaluation of the conceptual soundness of modeling approaches, and assessment of the adequacy of model documentation. Undertaking these steps early on will enable financial institutions to successfully implement models that will form the basis of an appropriate calculation of the ALLL under CECL. PROMONTORY Sightlines InFocus OCTOBER 12, 2018 6
Contact Promontory For more information, please call or email your usual Promontory contact or: William Lang Managing Director, New York wlang@promontory.com +1 212 542 6790 Erik Larson Managing Director and Global Head for Quantitative Methodologies and Analytics, Washington, D.C. elarson@promontory.com +1 202 384 1200 Aaron Johnson Director and Chief of Staff, Quantitative Services, Washington, D.C. ajohnson@promontory.com +1 202 370 0561 To subscribe to Promontory s publications, please visit promontory.com/subscribe.aspx Follow Promontory on Twitter @PromontoryFG Promontory Financial Group, an IBM Company, excels at helping clients resolve critical issues, particularly those with a regulatory dimension. Promontory professionals have unparalleled regulatory experience and insight, and provide our clients with frank, proactive advice informed by best practices and regulatory expectations. Founded in 2001 by Chief Executive Officer Eugene A. Ludwig, former U.S. comptroller of the currency, Promontory became a wholly owned subsidiary of IBM in 2016. 801 17th Street, NW, Suite 1100, Washington, DC 20006 Telephone +1 202 384 1200 Fax +1 202 783 2924 promontory.com 2018 Promontory Financial Group, an IBM Company. All Rights Reserved.