• A MUST WATCH Co-Sponsored Webinar with CU Sol:

    Strategic Planning in a Pandemic
    August 27, 2020

  • COVID-19 Crisis:

    Stress Test and Peer Comparision Report

  • A MUST WATCH Webinar:

    When the Government Lifeline Ends
    July 16, 2020

  • Prepare for COVID-19

    Learn More:
    Stress Testing for Pandemics

In Regards to Proposed Rule 10-17-19 Interagency Policy Statement on Allowances for Credit Losses

Written by: Dr. Joseph Breeden | Posted on: | Category:

The proposed rule provides helpful comments on how the new Current Expected Credit Loss (CECL) standard should be managed, validated, and monitored. One seemingly simple statement, however, has significant implications.

In the sections “Analyzing and Validating the Overall Measurement of ACLs”, “Responsibilities of Management “, and “Examiner Review of ACLs”, versions of the following statement are used:

“…comparing estimates of expected credit losses to actual write-offs in aggregate, and by portfolio, may enable management to assess whether the institution’s loss estimation process is sufficiently designed…”

Comparing expected credit losses to actual write-offs seems sensible as a general principle, but unfortunately does not work for CECL. CECL is not a loss forecast. First, CECL is a “lifetime” loss forecast where lifetime is not contractually assured. Since a comparison of CECL estimates to actual losses will generally be for a period that is less than the full life of the loan, the timing of the loss expectation will be critical for such a comparison, and yet many CECL-compliant methods make no attempt to correctly time future losses. Therefore, comparing over any time frame less that the full lifetime can be problematic.

Several points in the CECL guidelines will also cause it to deviate dramatically from future loss experience over any chosen fixed horizon, even when considering only existing loans. Some known examples where CECL will not align with actual loss experience are:

• Defaults after a loan renewal or extension where CECL estimates stop at the extension. • Reduced losses because of term extensions or loan rewrites. • Increased utilization of lines of credit or credit card, either normal balance growth by the consumer or in response to a credit line increase, leading to increased losses. • Consumer activation of inactive accounts leading to increased losses.

The greatest difficulty is in measuring a CECL-compliant historic loss experience given the chosen payment allocation rule for lines of credit. This is an accounting election that dramatically affects model performance and can even lead to changing how the models are created, even though actual losses are not being changed.

To tell developers, managers, examiners, and the board that CECL estimates should be compared to actual write-offs is certain to create a false expectation of what CECL is and how it can be monitored. In each section where this is mentioned, it is the first such item suggested, which will further increase the emphasis among those subject to the rule. Particularly alarming is the idea that examiners will judge a CECL model based upon how closely the CECL estimates align with actuals given all of the above reasons that the CECL estimate will, in fact, never align with actual write-offs regardless of the accuracy of the underlying model, i.e. even a perfect CECL model would fail this test.

The rule should be modified to make clear that CECL will NOT agree with actual write-off experience and can only be compared to a retrospective CECL loss in which actual write-offs have been adjusted for all of the accounting rules present within CECL. This means that the institutions must estimate a historic CECL loss against which to compare the CECL estimates, both of which are different from the actual historic write-offs.

Comparable to this is a historic contribution analysis where past losses are split into components of CECL loss and non-CECL loss, perhaps even splitting the non-CECL loss by some of the causes listed above.

Obviously, both of these approaches are additional work, but they are the only way to perform the recommended comparison. If this is viewed as too difficult for some lenders, then the statements about comparing CECL loss estimates to actual write-offs need to be deleted and left to the kind of detailed model performance review conducted during model validation, not as part of high-level oversight.

As CEO of Prescient Models LLC and Deep Future Analytics LLC, I have already seen this confusion in action. We are a CECL provider to nearly 200 regional banks, community banks, credit unions, and finance companies. We also provide validation services for many in-house CECL models. Prior to CECL, I have worked in credit risk modeling for 27 years and written many articles and books on the subject, including the Living with CECL series. During my CECL experiences, clients, validators, and auditors have already demonstrated confusion over the simple point that CECL is an accounting calculation, not a loss forecast.

As a long-time model developer, I sympathize with this confusion, because CECL is the first time in my career where I have needed to create a model to predict a quantity that itself is subject to how I as a developer interpret the CECL guidelines. I hope that in writing rules on model validation and monitoring, the language can be written more carefully to avoid reinforcing this confusion.

Joseph L. Breeden, PhD CEO, Prescient Models LLC and Deep Future Analytics LLC breeden@prescientmodels.com


SCHEDULE A DISCOVERY CALL

Learn More About How DFA's Software Can Help Your Institution?


Recent Publications

All publications below were authored by DFA's own, Dr. Joseph Breeden

  • Image

    Sample Chart

    Reserves: All Loans vs. RE Loans

    Image

    CECL (Current Expected Credit Loss) is the new accounting standard for estimating loss reserves on loan portfolios. The CECL guidance provides a great amount of flexibility in which models are used and a range of other choices that may impact the calculations. This book provides details of a study on how to apply CECL to US mortgage data. It seeks to disclose as many modeling details, results, and validation tests as possible so as to provide a reference for comparison and best practices. Because CECL is so similar to IFRS 9 Stage 2, this can also serve as a benchmark for implementing the new international account standards. The book is organized into three parts. Part I: Study Summary provides an overview of CECL, the design of the mortgage study, and the key comparative results across the models tested. Part II: Model Details provides in-depth discussions of how the models were designed and estimated, the coefficients, and the validation. Part III: Background provides additional conceptual material. Chapters 11 and 12 may be particularly useful to those new to modeling, and Chapter 13 puts CECL modeling in the context of lending analytics overall.

    • Date // May 2018
    Image

    Sample Chart

    Vintage Performance

    Image

    Building on the solid foundation of the previous bestselling first impression, this extended updated impression walks through the various issues of retail lending and develops approaches to address the interaction between economic cycles and retail lending. The complexity of time is extensively explored: vintages, current time and maturity. Reinventing Retail Lending Analytics, Second Impression covers complex issues such as scenario based forecasting, stress testing, volatility analysis, economic capital and portfolio optimisation, credit scoring and last, but not least, model risk.

    The book ends by providing examples of the application of nonlinear decomposition. These examples will provide you with rich data sets for exploring portfolio dynamics and improving portfolio management using nonlinear decomposition techniques.

    • Date // March 2019
    Image

    Sample Page

    Preface

    Image

    The new loan loss accounting rules for CECL and IFRS 9 require thousands of organizations to learn about modeling. Likewise, accountants and others in finance are now required to learn about statistical modeling concepts. This book is intended to define terms in a manner consistent with decades of academic literature on statistical modeling and hopefully reduce some of the noise and confusion just around definition of terms. It may also serve as a useful guide to analysts new to the field tasked with IFRS 9 compliance, the international loss accounting rules, and credit risk modeling in general.

    Each chapter of this book is a term that one might encounter when discussing creating lifetime loss forecasting models for CECL or IFRS 9. Not every term is a model, and some models listed are being mentioned only to explain why they are not likely to be used for loss forecasting. The CECL guidelines and subsequent FAQs have given examples of modeling techniques. Some people new to loss forecasting have assumed that those are all the available or applicable methods. This book is meant in
    part to dispel that misconception.

    The definitions and descriptions provided here are meant to provide an intuitive understanding across a range of modeling techniques. Mathematical derivations are kept to a minimum. The references listed will provide all the necessary details for an eager analyst.

    • Date // June 2018
© 2020 Deep Furture Analytics Contact Us