Complying with CECL We assess five ways to implement the new regulations September 2017
Analytical contacts Manish Kumar Director, Risk & Analytics, India manish.kumar@crisil.com Manish Malhotra Lead Analyst, Risk & Analytics, India manoj.malhotra@crisil.com Stephen Knights Director, Risk & Analytics, London stephen.knights@crisil.com Rahul Kumar Lead Analyst, Risk & Analytics, India rahul.kumar1@crisil.com Prithivinath Prabhunath Lead Analyst, Risk & Analytics, India prithivinath.prabhunath@crisil.com 2
Abstract On June 16, 2016, the Financial Accounting Standards Board (FASB) issued the Accounting Standards Update (ASU) No 2016-13, Topic 326, Financial Instruments Credit Losses. It introduced the concept of Current Expected Credit Loss (CECL) that takes into account reasonable and supportable forecasts, and life of loan. This marks a paradigm shift in Allowance for Loan and Lease Losses (ALLL) from the traditional, historically-focussed incurred loss approach to the forward-looking CECL. The ASU has suggested several methods to estimate expected credit loss without underlining a preferred one. Management have the discretion to select and develop methodologies that faithfully reflect expected credit losses for financial assets and also can be applied consistently over time. Potentially, different estimation methods can be applied to different groups of financial assets. This white paper examines merits, demerits and key aspects of these methods, and suitable modelling techniques that can be applied to estimate credit losses under each method. Background In 2009, FASB codified the accounting standards for recognition of credit losses and issued ASC 450-20 titled Loss Contingencies and ASC 310-10 titled Receivables. ASC 450-20 laid down conditions for which losses will be accrued and ASC 310-10 defined the impairment conditions and treatment of impairment losses. However, during the global financial crisis in 2008, concerns were raised about the effectiveness and timeliness of credit loss recognition under the current guidance. The procedure was also criticised for preventing banks from provisioning for an impaired asset until a triggering event occurred, thereby delaying the recognition of credit losses. In 2008, the FASB teamed up with the International Accounting Standards Board (IASB) and established a Financial Crisis Advisory Group to advise the boards on improvements to financial reporting in response to the financial crisis. The group identified delayed recognition of credit losses as a weakness in the current Generally Accepted Accounting Principles that results in the potential overstatement of assets. As a result, the group recommended exploration of more forward-looking alternatives to the incurred loss approach. The FASB initially worked with the IASB to develop common guiding principles; however, the two boards ultimately reached different conclusions on certain significant issues. In December 2012, the FASB published the exposure draft Proposed Accounting Standards Update, Financial Instruments Credit Losses and introduced the concept of CECL. Finally, the FASB issued the Accounting Standards Update (ASU) No. 2016-13 in June 2016, of which sub-topic 326-20 deals with credit losses. Timeline for implementation of CECL CECL Jun 2016 Dec 2019* CECL standards final version Issued CECL to go live for banks Note: * Dec 2019 for public companies and Dec 2020 for others. 3
Applicability The new accounting standard applies to all banks, savings associations, credit unions, and financial institution holding companies, regardless of asset size. Further, ASU 2016-13 applies to all financial instruments carried at an amortised cost. This includes investment instruments and off-balance-sheet credit exposures such as loan commitments, standby letters of credit and financial guarantees. Decoding the operational aspects of ASU 326-20 The new accounting standard for estimating loan loss reserves offers general guidelines and a list of possibilities, but provides no specific recommendations on the implementation of those rules. The standard is silent on many aspects such as definition of default and methodologies. Due to the absence of these details and precise guidelines, CRISIL GR&A have analysed the standard and formulated specific measures to deal with the operational aspects, especially with regard to the method. By applying these measures, we have outlined a broad class of estimation techniques that estimate the collectability of the financial assets and can be applied consistently over time. In particular, based on our experience in the area, we have set out operational aspects of the methods under ASU326-20 in terms of their respective merits, demerits and key aspects. Common methods under CECL The standard lists several common credit loss methods that will continue to be acceptable under the new guidance: 1. Discounted cash flow method 2. Aging schedule method 3. Roll-rate method 4. Loss-rate method 5. Probability of default-based method CECL estimation under all these methods can be done at the instrument level (the bottom-up approach), or in aggregate pools (cohort) using weighted average assumptions. A brief discussion on merits, demerits and key aspects for each of these methods, and suitable modelling techniques that can be applied to estimate credit losses under each method, is given below. 4
1. Discounted cash flow method According to paragraph ASC 326-20-30-3 of the standard, the allowance for credit loss may be determined using the discounted cash flow method. As described in paragraph ASC 326-20-30-4 of the standard, discounted cash flow analysis is based on the present value of expected future cash flows discounted at the loan s effective interest rate. 1 When the method is applied, the allowance for credit losses is the difference between the amortised cost basis valuation and the present value of expected cash flows. This is one of the methods prescribed currently to determine the FAS 114 component of ALLL for an individually impaired loan, 2 in which the expected future cash flow is based on each bank s best estimate of reasonable and supportable assumptions and projections. FAS 114 requires consideration of the likelihood of the possible outcomes, to determine the best estimate of expected future cash flows. For assets that have unique risk characteristics, the method is suitable for expected credit loss estimation. Merits Incorporates component loss assumptions when taking into account the distribution of cash flows relative to loanspecific attributes. Helps directly factor in additional assumptions such as maturity and prepayment during modelling. Preferred for the case of purchased assets, because it requires limited historical information. Demerits Sensitive to a large number of assumptions. For example, assumptions are required for each loan-specific attribute such as principal, repayment terms, coupon, maturity dates, collateral and prepayment. When used at the aggregate pool level, losses are projected against a static default or prepayment curve and so accuracy of projected losses relies on the accuracy of estimation for these curves. Not preferred for the case of revolving products, because it is time-consuming and computationally intensive to estimate cash flows at the individual loan level. Key aspects to be considered when using the discounted cash flow method The effective interest rate may not be readily available in the loan accounting system. In the case of variable-rate loans, based on an underlying factor such as the prime rate or the US treasury bill weekly average, the standard does not allow the estimates of future cash flows to be affected by various projections of the underlying factor. However, in the case of fixed-rate loans, changes in the interest rate forecasts may affect prepayment assumptions for estimates of future cash flows. There is no explicit requirement to consider probability-weighted outcomes. However, institutions are required to use available information that is relevant to assess the cash flows. Suitable modelling techniques For cash flow forecasting, modelling techniques such as linear regression are recommended. Other modelling techniques such as Cointegration and Error-Correction Modelling may also be used. The literature supports the use of simulation to project cash flows; however, simulation is not recommended for CECL-compliant models, because macroeconomic and borrower-specific variables are not commonly used during the simulation process. 1 The effective interest rate is the contractual interest rate adjusted for any net deferred fees or costs, premium, or discount existing at origination or acquisition of the financial asset. 2 Statement of Financial Accounting Standards (FAS) No. 114, Accounting by Creditors for Impairment of Loan, is a principal source of guidance on accounting for ALLL under Generally Accepted Accounting Principles. 5
2. Aging schedule method According to paragraph ASC 326-20-30-3 of the standard, the allowance for credit loss may be determined using a method that utilises an aging schedule. Through this method, the historical credit loss rate is provided in each age bucket. For example, the 5 buckets of Current, 1-30 Days Past Due (DPD), 31-60 DPD, 61-90 DPD and more than 90 DPD could be used. After compiling the historical credit loss rate under the respective age buckets, the adjustment in the loss percentage is made for current conditions, using reasonable and supportable forecasts. Based on this aging categorisation and the adjusted loss rate in each age category, CECL is computed by multiplication of the adjusted loss rate with the amortised cost. Merits Straightforward to implement and easy to comprehend. Demerits Relies on extensive actual loss data as an input when estimating credit losses. Fails to incorporate the time value of money. Ignores seasoning of loans. Key aspects to be considered when using the aging schedule method Disclosures are required under paragraph ASC 326-20-50-14 for provision of aging analysis on an amortised cost basis, for financial assets that are past due. These disclosures must be provided for each class of financing receivables and major security types. A model developed using this method and the underlying data thereof can be leveraged for these disclosures. Data requirements are intensive because the credit loss rate is modeled at a granular level (i.e for each DPD category). Institutions have to consider how to adjust the historical loss experience, not only for current conditions, but also for reasonable and supportable forecasts that affect the expected collectability of financial assets. Suitable modelling techniques For estimating the loss rate using the aging schedule method, modelling techniques such as linear regression, Logit and Tobit can be used. In addition, it is recommended that this method be used for loans that are currently monitored based on their DPD category. 6
3. Roll-rate method According to paragraph ASC 326-20-30-3 of the standard, the allowance for credit loss may be determined using the roll-rate method. This method uses historical analysis based on segmentation, by delinquency or risk rating, of a portfolio of financial assets. In particular, assessment is made of the roll-rate ; namely, the percentage of balances or the number of accounts which move from one delinquency stage to the next. 3 Financial institutions may follow the averaging technique over time to develop an average roll-rate for each segment. After the roll-rate is determined for each segment, the respective roll-rate percentage is applied to the balance in each category to estimate the amount that will migrate to the next category. Finally, the total migrations are aggregated across all categories to determine the allowance for credit loss. Merits Uses granular analysis, given that loan tracking is done of an individual account over time to assess its performance. Hence, estimation based on this method will be more accurate. Generates joint probability of loan migration from one bucket to another, when used with a separate probability of default method. Accommodates expected changes to transition rates as a result of macroeconomic conditions or defaults. Demerits Requires large quantities of data, because the movement of segments over time is modelled at a granular level (e.g. by delinquency or risk rating). Results in weak predictive power beyond the near term, particularly if projections are not properly conditioned on scenario variables. Selection of period, or the starting point of analysis, has a significant impact on model performance. Key aspects to be considered while using roll-rate method Requires longitudinal account-level performance data starting from the origination date, to capture the roll-rate of accounts over the time period. Assumes portfolio segmentation is static; hence, portfolio segregation should remain the same for the entire lifetime of a portfolio. Otherwise, changes in segment definition will impact model performance. Suitable modelling techniques To derive expected credit loss using the roll-rate method, the probability of default method may be used with the rollrate method. Within the CECL framework, the historical roll-rate can be derived. Forward-looking information can also be included using techniques such as the ARIMAX model, to forecast lifetime loss for each bucket. 3 The method is also known as migration analysis or the flow model. 7
4. Loss-rate method According to paragraph ASC 326-20-30-3 of the standard, the allowance for credit loss may be determined using the loss-rate method. This method begins with computation of the historical loss of the portfolio. The cumulative loss over the life of the loan is then divided by the current outstanding balance, to obtain the historical cumulative loss rate. This historical loss rate is adjusted to reflect the current conditions and forecasted changes in the risk drivers of the portfolio. Finally, the allowance for credit loss is determined by multiplication of the adjusted loss rate with the amortised cost of the loan portfolio. Merits Easy to comprehend. The use of historical loss rate, with qualitative adjustments to reflect the current conditions and forecasted changes in the risk drivers, is intuitive. Straightforward to implement. In particular, data for unadjusted cumulative historical losses at the aggregate level are relatively easy to obtain. Demerits Ignores the dynamic nature of items such as product mix, prepayment, collateral and terms to maturity. Loss estimates may become unreliable as the level of segmentation and complexity of products increase. Fails to capture idiosyncratic risk, as the method uses aggregate level averages of the historical loss rate. Key aspects to be considered while using the loss-rate method Cumulative losses are sensitive to business scenarios used. The loss-rate method is primarily based on past experience at the aggregate level. Therefore, an appropriate historical period needs to be chosen. The period should be long enough to take into account the life of a loan. For example, for a loan portfolio with a contractual tenure of 10 years, it would be prudent to use historical data from at least the last 10 years. However, the historical period should also be short enough to support a strong meaningful relationship between the risk drivers and the loss-rate. Suitable modelling techniques Modelling techniques such as linear regression and logistic regression can be used. 8
5. Probability-of-default method According to paragraph ASC 326-20-30-3 of the standard, the allowance for credit loss may be determined using the probability-of-default method. This method estimates credit losses by the calculation and aggregation of three components: namely, the probability of default, loss given default and exposure at default. The method is similar to that used for loss estimation under the Comprehensive Capital Analysis and Review (CCAR) guidelines issued by the Federal Reserve Board. The three components used in the probability-of-default method are defined as follows: Probability of default (PD) the likelihood of default of an obligor over a given time period. Loss given default (LGD) the percentage of exposure at default that will not be recovered in case of default. Exposure at default (EAD) this is typically taken to be the outstanding balance net of any specific provisions. Merits Transparency of results is provided, through segregation of credit losses into the three components described above. High granularity of results, as the probability of default can be estimated on a per-loan basis. Demerits Specialised techniques and competencies are required to apply and interpret findings effectively. Requires large quantities of data. Key aspects to be considered when using the probability-of-default method The standard does not define default for estimation of expected credit losses. Therefore, banks need to define this; for example, by use of current industry practices such as the Basel definition for default. Since CECL guidelines focus on the life of loan concept, it is imperative for banks to track accounts from origination to default and then until the recovery window. For estimating lifetime PDs, data on how defaults arise over time are required. These would ideally be recent historical information, over the life of similar loans. Segmentation of EADs should ideally be carried out, based upon relevant characteristics. Suitable modelling techniques Statistical techniques such as linear regression, logistic regression and hazard modelling are used to estimate PDs and LGDs. In particular, the estimation of LGDs needs to consider the collection procedure, charge-off and costs involved in the recovery process. EAD uses the default balance as the final exposure for non-revolving products (e.g. term loans), while final exposure utilises a credit conversion factor and loan-equivalent amount for revolving products (e.g. overdraft loans). 9
Summary and conclusion The CECL model signifies a transition from the current incurred loss model to the expected credit loss model, taking the life of loan and reasonable and supportable forecasts into consideration. To address this fundamental change, banks are required to develop loss estimates that are forward looking. They will be required to update their method of estimating credit losses to comply with the CECL approach. Selection and use of a methodology suitable for each set of products requires a comprehensive approach to data collection, portfolio analysis, model selection and development, model testing and ongoing review. CRISIL GR&A provides teams of dedicated professionals and experts to help clients meet CECL compliance requirements by providing gap analysis, model development and documentation, business analysis and testing services. We also understand that different methods will be suitable for different products. We apply our expertise and experience in development of credit risk models, process re-design, data warehouse design, implementation and model validation to meet CECL compliance requirements and maintain consistency with industry standards. We also provide experienced professionals to enhance the overall IT infrastructure and streamline data collection processes, while ensuring proper governance over changes to required data. CECL- CRISIL GR&A s global footprint can be leveraged by banks with an international exposure, to implement compliant methodologies across geographies. 10
About CRISIL Limited CRISIL is an agile and innovative, global analytics company driven by its mission of making markets function better. We are India s foremost provider of ratings, data, research, analytics and solutions. A strong track record of growth, culture of innovation and global footprint sets us apart. We have delivered independent opinions, actionable insights, and efficient solutions to over 100,000 customers. We are majority owned by S&P Global Inc., a leading provider of transparent and independent ratings, benchmarks, analytics and data to the capital and commodity markets worldwide. About CRISIL Global Research & Analytics (GR&A) CRISIL Global Research & Analytics (GR&A) is the world s largest and top-ranked provider of high-end research, risk and analytics services. We are the world s largest provider of equity and fixed-income research support to banks and buy-side firms. We are also the foremost provider of end-to-end risk and analytics services that include quantitative support, front and middle office support, and regulatory and business process change management support to trading, risk management, regulatory and CFO functions at world s leading financial institutions. We also provide extensive support to banks in financial crime and compliance analytics. We are leaders in research support, and risk and analytics support, providing it to more than 75 global banks, 50 buy-side firms covering hedge funds, private equity, and asset management firms. Our research support enables coverage of over 3,300 stocks and 3,400 corporates and financial institutions globally. We support more than 15 bank holding companies in their regulatory requirements and submissions. We operate from 7 research centers in Argentina, China, India, and Poland, and across several time zones and languages. CRISIL Privacy Notice CRISIL respects your privacy. We use your contact information, such as your name, address, and email id, to fulfil your request and service your account and to provide you with additional information from CRISIL and other parts of S&P Global Inc. and its subsidiaries (collectively, the Company ) you may find of interest. For further information, or to let us know your preferences with respect to receiving marketing materials, please visit www.crisil.com/privacy. You can view the Company s Customer Privacy at https://www.spglobal.com/privacy Last updated: April 2016 Argentina China Hong Kong India Poland Singapore UK USA CRISIL Limited: CRISIL House, Central Avenue, Hiranandani Business Park, Powai, Mumbai 400076. India Phone: + 91 22 3342 3000 Fax: + 91 22 3342 3001 www.crisil.com