Sageworks Advisory Services PRACTICAL CECL TRANSITION DISCOUNTED CASH FLOW (DCF)

Transcription

1 Sageworks Advisory Services PRACTICAL CECL TRANSITION DISCOUNTED CASH FLOW (DCF)

2 Use of this content constitutes acceptance of the license terms incorporated at This content license confers certain rights to you, the user, of this material, and reserves others to Sageworks, Inc. Further, this content license disclaims liabilities for use of the content. It should be carefully reviewed before transmitting or using this content. The following summary is not a substitute for, nor supercedes, the content license agreement referenced above: You may share this content in the exact form it was provided to you, so long as it is shared unaltered and in its entirety, including these terms, and our brands and marks -- any user of the content will be bound by the same Agreement by using the content. While we encourage sharing of this content, you are prohibited from profiting from the use of the content or any derivative works. We make no warranty of any kind for this content -- it is provided in good faith as a courtesy to the industry we are privileged to serve. As this content is provided to an industry, it may have absolutely no relevance or applicability to your particular institution or use case, and your critical thinking is required to successfully use (or not use) the content. We undertake no obligation to keep the content up to date, answer inquiries regarding the content, or support it in any other way. 2

3 EXECUTIVE SUMMARY Discounted Cash Flow (DCF) models, while not widely adopted as a means to account for allowance for loan and lease loss under ASC (current GAAP), have been accepted as best practice for adherence to other analogous accounting standard objectives. For example, fair value measurement (ASC 820) and purchased credit impaired (ASC ) both inherently require an understanding/estimate of lifetime credit loss. Forward-focused cash flow models are commonly deployed to accommodate both fair value and purchased credit impaired requirements resulting in an approach that has a precedent of successful audits and examinations. This document is intended to cover DCF outputs; inputs, assumptions, and sensitivities; forecast application; specific guidance; and conceptual soundness under the context of ASU (Topic 326). While there is no requirement to utilize a DCF within the new standard, this document should assist in determining when such an election would be appropriate and even practical. APPLICABLE GUIDANCE FASB Accounting Standards Update Topic 326 (CECL): If an entity estimates expected credit losses using methods that project future principal and interest cash flows (that is, a discounted cash flow method), the entity shall discount expected cash flows at the financial asset s effective interest rate. When a discounted cash flow method is applied, the allowance for credit losses shall reflect the difference between the amortized cost basis and the present value of the expected cash flows. If the financial asset s contractual interest rate varies based on subsequent changes in an independent factor, such as an index or rate, for example, the prime rate, the London Interbank Offered Rate (LIBOR), or the U.S. Treasury bill weekly average, that financial asset s effective interest rate (used to discount expected cash flows as described in this paragraph) shall be calculated based on the factor as it changes over the life of the financial asset. Projections of changes in the factor shall not be made for purposes of determining the effective interest rate or estimating expected future cash flows p , emphasis ours The Board decided to allow entities to determine the EIR and expected cash flows (including expected prepayments and defaults) using their own expectations (projections) of future interest rate environments when estimating credit losses on variable-rate financial assets using a DCF method provided those expectations are reasonable and supportable. However, the use of projections will not be required. To reflect this decision, an amendment to the FASB Codification will be drafted as a part of the ongoing Codification improvements project. FASB Memorandum, Minutes of December 13, 2017 Board Meeting, December 19, 2017, emphasis ours 3

4 INTRODUCTION All implementation and data considerations being equal, DCF is the method of choice for most portfolio concentrations. This is mainly due to the methodological alignment with the accounting standard s requirements to estimate lifetime losses while effectively giving proper consideration and support for prepayment behavior, timing, and forecasted conditions. Generating and interpreting a contractual schedule of cash flows for financial assets is not theoretically difficult. Nor is it conceptually challenging to generate and interpret a schedule of expected cash flows. However, the mechanics of generating and analyzing such schedules can prove cumbersome without the assistance of purpose-built software. When software is available, a DCF election may be the most practical approach to CECL. Alternatively, when software is not available, the long-term benefits provided by the capability of executing portfolio-wide cash flow projections may serve, by itself, as the justification for software acquisition or internal development. Instructions regarding how to perform portfolio-wide cash flow projections is one intended purpose of this document. However, information regarding why to perform portfolio-wide cash flow projections is of equal value and intention. Readers of this chapter should conclude with the following: An understanding of the difference between contractual cash flows and expected cash flows An awareness of the various inputs and assumptions required to generate expected cash flow projections A comprehension of the symmetry between forecasted conditions and the application within a DCF model The knowledge to adhere to the requirements within the standard specific to the utilization of a DCF method An appreciation for primary, secondary, and tertiary DCF benefits 4

5 CONTRACTUAL CASH FLOW VS. EXPECTED CASH FLOW Amortization schedules generated on the basis of contractually required principal and interest payments are used in the creation of contractual cash flow schedules. The format is structured in a manner that includes or allows for a period-level cash flow determination. Expected cash flow schedules, on the other hand, include estimates for periodic prepayments, defaults, losses (collateral/recovery shortfall), and recoveries (usually modeled with a delay). Like contractual schedules, expected schedules are also formatted for the purposes of period-level cash flow examination. Contractual Cash Flow A reasonable practitioner may be thinking: If CECL requires a comparison of the Net Present Value (NPV) of expected cash flows, why would an institution be interested in generating contractual cash flow schedules? In short, the ability to create a contractual schedule (data - payment amount, maturity date, coupon, etc.) is the only requirement to construct an expected cash flow schedule; true. However, in order to compute effective yield, a contractual schedule or modified contractual schedule (includes estimated prepayment) must be produced. Additionally, understanding the difference between contractual and expected returns is institutionally beneficial independent of CECL. For context, early versions of CECL included language requiring institutions to derive contractual cash flows not expected to be collected. Those familiar with acquired loan accounting will recognize this metric as Nonaccretable Difference. Below is an example of aggregated contractual cash flow schedules; note that no considerations have been made for estimated behavior: Date Beginning Balance Principal Interest Prepayment Defaulted Principal Estimated Loss Estimated Recovery Ending Balance Cash Flow Present Value 10/31/2017 $816,099, $5,273, $2,945, $0.00 $0.00 $0.00 $0.00 $810,825, $8,218, $8,184, /30/2017 $810,825, $5,607, $2,827, $0.00 $0.00 $0.00 $0.00 $805,217, $8,435, $8,373, /31/2017 $805,217, $11,480, $2,900, $0.00 $0.00 $0.00 $0.00 $793,737, $14,380, $14,224, /31/2018 $793,737, $3,012, $2,858, $0.00 $0.00 $0.00 $0.00 $790,724, $5,871, $5,788, /28/2018 $790, 724, $8,723, $2,572, $0.00 $0.00 $0.00 $0.00 $782,000, $11,296, $11,087, /31/2018 $782,000, $4,865, $2,813, $0.00 $0.00 $0.00 $0.00 $777,135, $7,679, $7,503, /30/2018 $777,1l5, $3,923, $2,703, $0.00 $0.00 $0.00 $0.00 $773,211, $6,627, $7,503, /31/2018 $773,211, $5,157, $2,778, $0.00 $0.00 $0.00 $0.00 $768,053, $7,936, $7,719, /30/2018 $768,053, $13,263, $2,671, $0.00 $0.00 $0.00 $0.00 $754,789, $15,935, $15,427, /31/2018 $ $3,066,63.96 $2,711, $0.00 $0.00 $0.00 $0.00 $751, 723, $5,778, $5,579, /31/2018 $751,723, $6,768, $2,701, $0.00 $0.00 $0.00 $0.00 $744,954, $9,469, $9,060, /30/2018 $744,954, $11.323, $2,586, $0.00 $0.00 $0.00 $0.00 $733,630, $13,909, $13,422, An awareness of the intention behind producing contractual cash flows and the requirement to subsequently leverage the outputs to derive effective yield is a fundamental learning objective. The specifics of computing effective yield, however, are described within the Inputs, Assumptions, and Key Considerations section of this document. 5

6 Expected Cash Flow Projecting future cash flows is an exercise in estimation. Analyzing past experiences (internal and external) pertaining to prepayments, defaults, chargeoffs, recoveries, and their relationship to external and internal influences provide institutions with support and a basis for estimation. Resulting estimates are applied to each period within an amortization schedule resulting in each subsequent period s behavior being impacted by the compounded effect of preceding estimates. In the example shown below, the first period s ending balance (10/31/17 in this example) is net of estimated prepayments and defaults. Subsequent periods beginning balance will reflect the prior period s ending position: Date Beginning Balance Principal Interest Prepayment Defaulted Principal Estimated Loss Estimated Recovery Ending Balance Cash Flow Present Value 10/31/2017 $482,530, $2,129, $1,748, $6,253, $69, $7, $0.00 $474,078, $10,130, $10,094, /30/2017 $474,078, $3,733, $1,662, $6,137, $68, $7, $0.00 $464,139, $11,533, $11,451, /31/2017 $464,139, $10,227, $1,681, $5,916, $65, $7, $0.00 $447,929, $17,826, $17,633, /31/2018 $447,929, $2,071, $1,622, $5,808, $93, $11, $0.00 $439,955, $9,502, $9,368, /28/2018 $439,955, $3,452, $1,440, $5,690, $91, $10, $0.00 $430,720, $10,583, $10,397, /31/2018 $430,720, $3,434, $1,560, $5,578, $89, $10, $0.00 $421,618, $10,572, $10,342, /30/2018 $421,618, $2,114, $1,476, $5,473, $104, $13, $0.00 $413,864, $9,125, $8,904, /31/2018 $413,864, $3,137, $1,498, $5,355, $102, $12, $0.00 $405,268, $9,991, $9,714, /30/2018 $405,268, $9,092, $1,420, $5.225, $99, $12, $0.00 $390,850, $15,738, $15,228, /31/2018 $390,850, $2,240, $1,413, $5,123, $144, $20, $0.00 $383,341, $8,778, $8,474, /31/2018 $383,341, $5,019, $1,386, $4,985, $140, $19, $0.00 $373,195, $11,391, $10,920, /30/2018 $373,195, $7,514, Sl $4,810, $135, $19, $0.00 $360,734, $13,628, $13,135, It is worth noting that the analysis methodology used to observe behavior should be symmetrical to the application. Therefore, an understanding of the methodology used to derive inputs and assumptions is paramount to accomplishing a conceptually sound expected cash flow schedule. INPUTS, ASSUMPTIONS, AND KEY CONSIDERATIONS Discounted Cash Flow analysis is reliant upon a variety of loan-level data, peripheral model outputs, and key assumptions. The data fields required to create the contractual portion of the forward-looking cash flow schedule relate to the terms of each loan and include information regarding payment amount, payment frequency, interest rate, interest type, maturity date, amortization term, etc. For this reason, DCF has been stereotyped as being data intensive. While such fields have not been leveraged for pooled loan analysis under ; this information is not required in arrears, is fundamental, and should be readily available. The information required to introduce expected or possible variances from contractual behavior fall into fundamental categories regardless of the specific methodology deployed - credit and timing. Such information may be sourced from new models, existing processes within the organization, or subjectively applied. Note that the manner in which inputs were derived will dictate the manner in which inputs should be applied. Regardless of the source, an understanding of the methods used in the calculation of any input will mitigate the risks of asymmetrical application. Likewise, an understanding of the application of such inputs will assist in calculating and supporting necessary assumptions. 6

7 The following is intended to provide a conceptual understanding of the various inputs, assumptions, and key considerations required to generate an expected cash flow schedule in accordance with the new standard. An awareness of the inputs themselves, how they are derived, and how they are deployed are essential learning objectives. Contractual Terms Analysis will begin with the creation of each loan s contractual amortization schedule. This requires point-in-time information on balance, interest rate, interest type, payment amount, payment type, payment frequency, and maturity. Variable rate loans dependent upon market rates were originally to be analyzed without projections of changes in the underlying factor. However, the FASB has since provided clarification and simply requires that projections of future interest rate environments are reasonable and supportable. If the financial asset s contractual interest rate varies based on subsequent changes in an independent factor, that financial asset s effective interest rate shall be calculated based on the factor as it changes over the life of the financial asset. Projections of changes in the factor shall not be made for purposes of determining the effective interest rate or estimating expected future cash flows p , emphasis ours The Board decided to allow entities to determine the EIR and expected cash flows (including expected prepayments and defaults) using their own expectations (projections) of future interest rate environments when estimating credit losses on variable-rate financial assets using a DCF method provided those expectations are reasonable and supportable. However, the use of projections will not be required. To reflect this decision, an amendment to the FASB Codification will be drafted as a part of the ongoing Codification improvements project. FASB Memorandum, Minutes of December 13, 2017 Board Meeting, December 19, 2017, emphasis ours Current term information is readily available from most core systems. Contractual term information as of a specified date in the past is helpful for prepayment rate calculation and particular back testing exercises. In the event that historical contractual term data is unavailable, the absence of such information would not preclude institutions from utilizing and/or electing a DCF approach. A critical function of the contractual cash flow schedule is the determination of the asset s effective interest rate, which serves as the discount rate applied to the expected periodic cash flows. The Effective Interest Rate (EIR) can be determined prior to, or after, prepayment assumptions are made. 7

8 Prepayment Rate Prepayment rate indicates the proportion of a loan pool s outstanding balance that is expected to be collected prematurely. This is sometimes referred to as Conditional Prepayment Rate (CPR) and is expressed as an annual percentage. Topic 326 requires institutions to explicitly or implicitly include prepayment behavior in the estimation of expected credit losses: An entity shall estimate expected credit losses over the contractual term of the financial asset(s) when using the methods in accordance with paragraph An entity shall consider prepayments as a separate input in the method or prepayments may be embedded in the credit loss information in accordance with paragraph An entity shall consider estimated prepayments in the future principal and interest cash flows when utilizing a method in accordance with paragraph An entity shall not extend the contractual term for expected extensions, renewals, and modifications unless it has a reasonable expectation at the reporting date that it will execute a troubled debt restructuring with the borrower p110, emphasis ours Prepayment rates are derived by comparing periodic contractual obligations to observed receipts. The variance between observed curtailments and the contractual principal obligation is divided by beginning outstanding balance resulting in a rate referred to as the Single Monthly Mortality (SMM). Note that loan-level aggregation is typically performed for weighting purposes prior to the SMM calculation. For example, suppose expected principal collections were $10 for a pool with $100 in outstanding balances. Once analyzed, $11 in principal collections were observed. The additional $1 in principal collections results in a 1% SMM or a 12.68% CPR (annualized). While the logic is relatively straightforward, there are computing/modeling constraints, methodology considerations for some loan types, and there can be extreme volatility in smaller loan pools. CPR assumptions can have a material impact on expected cash flows due to the diminishing interest obligation. Like its effect on interest, prepayment rates can also reduce the expectation of defaulted dollars. Because defaults are derived as a percentage of dollars outstanding for each period, the accelerated decline in balance has a direct impact on resulting loss expectations. Institutions utilizing a DCF for the computation of expected loss, as a matter of model risk management, should understand and quantify the impact CPR and similar assumptions have under certain conditions. Due to the material nature of prepayment assumptions, The Financial Accounting Standards Board (FASB) and the Transition Resource Group (TRG) have discussed the application of prepayments as it relates to the computation of Effective Interest Rate (EIR). Specifically, should entities include a prepayment assumption when calculating EIR? In a subsequent memorandum, the option to make an election at the class of financing receivables level was communicated. The staff believes that an entity should not be required to use an EIR adjusted for prepayment expectations Accordingly, an entity should be permitted to choose, through an accounting policy election at the class of financing receivable level for loan receivables, whether it will use an EIR adjusted for prepayment expectations when using a DCF method to determine the allowance for credit losses. FASB Memorandum No. 6 August 25, 2017, emphasis ours 8

9 The consensus of the TRG members at the June 2017 meeting was to support the staff s recommendation that would allow for an accounting policy election to use an EIR adjusted for prepayment expectations when using a DCF method to determine the allowance for credit losses. Upon making the accounting policy election, entities should update the adjusted EIR periodically based on changes in expected prepayments. FASB Memorandum No. 6c December 21, 2017, emphasis ours When utilizing prepayment assumptions within an expected cash flow setting, a full understanding of the methodology employed in the determination of the assumption is imperative. The application of this assumption should be symmetrical to the manner in which it was derived. For example, if observed prepayment behavior was determined by analyzing all loans within a particular class of financing receivables without further disaggregation for vintage, the application should follow suit. Likewise, if the observed prepayment behavior was derived with further disaggregation for vintage, the application would be dependent on each loans origination/renewal period. Conceptually, disaggregation is appealing. In practice, it can be impractical to derive a statistically sound conditional prepayment rate due to limited observations (loan counts), application modeling requirements, and the overall level of estimation elsewhere within the expected cash flow schedule. For this reason, analogous accounting standards, ASC 820 and , have been executed utilizing a more practical approach such as constant or pool average in the application of prepayment assumptions. Effective Interest Rate Topic 326 requires institutions estimating expected credit losses by way of the projection of future principal and interest to discount the resulting cash flows at the asset s effective interest rate (EIR): If an entity estimates expected credit losses using methods that project future principal and interest cash flows (that is, a discounted cash flow method), the entity shall discount expected cash flows at the financial asset s effective interest rate. When a discounted cash flow method is applied, the allowance for credit losses shall reflect the difference between the amortized cost basis and the present value of the expected cash flows p , emphasis ours The effective interest rate is a function of the basis relative to contractually obligated payments. Effective Interest Rate as defined by FASB is as follows: The rate of return implicit in the financial asset, that is, the contractual interest rate adjusted for any net deferred fees or costs, premium, or discount existing at the origination of acquisition of the financial asset. For purchased financial assets with credit deterioration, however, to decouple interest income from credit loss recognition, the premium or discount at acquisition excludes the discount embedded in the purchase price that is attributable to an acquirer s assessment of credit losses at the date of acquisition. FASB Accounting Standards Codification s Master Glossary In most instances today, the EIR used for interest income recognition does not consider estimated prepayments. However, when there are inconsistencies in the loan term used to calculate EIR and the term used to project future 9

10 cash flows, differences in the net amount expected to be collected will arise resulting in different allowance levels. Differences in allowance levels would remain present even where no expectations of credit loss exists. For this reason, it is important that practitioners understand the impact EIR can have on various asset classes and/or certain loan positions. Impact analysis examples are displayed within the Adjusted Effective Interest Rate section below. Adjusted Effective Interest Rate As mentioned in the Contractual Terms section above, institutions may choose, through an accounting policy election at the class of financing receivable level, whether it will use an EIR adjusted for prepayment. Prior to making this election, an impact analysis should be performed and support for the election documented. The adjusted effective interest rate is a function of the basis relative to contractually obligated principal, interest, and expected prepayments. The following examples illustrate the impact EIR and adjusted EIR can have on the net amount expected to be collected and ultimately reserve levels independent of loss expectation for assets in a discount position and premium position: Payment Type Principal & Interest Principal & Interest Principal & Interest Principal & Interest Maturity Date 9/30/2022 9/30/2022 9/30/2022 9/30/2022 Unpaid Principal 100, , , ,000 Net Deferred Costs / (Fees) / Premium / (Discount) 10,000 10,000 10,000 10,000 Interest Rate (Note Rate) 10.00% 10.00% 10.00% 10.00% Payment Amount 2,149 2,149 2,149 2,149 Effective Yield 15.28% 17.65% 4.51% 3.76% CPR 15.00% 15.00% 15.00% 15.00% Amortized Cost Basis 90,000 90, , ,000 Net Amount Expected to be Collected 92,865 90, , ,000 Excess / (Shortfall) 2,865 0 (1,308) 0 Due to the inherent variance between the cash flows used to determine effective yield and the cash flows used to derive the net amount expected to be collected, the inclusion of prepayments will result in a variance when compared to the basis. Even when expected losses, 5% Probability of Default and 15% Loss Given Default, are introduced; this phenomenon does not change: Payment Type Principal & Interest Principal & Interest Principal & Interest Principal & Interest Maturity Date 9/30/2022 9/30/2022 9/30/2022 9/30/2022 Unpaid Principal 100, , , ,000 Net Deferred Costs / (Fees) / Premium / (Discount) 10,000 10,000 10,000 10,000 Interest Rate (Note Rate) 10.00% 10.00% 10.00% 10.00% Payment Amount 2,149 2,149 2,149 2,149 Effective Yield 15.28% 17.65% 4.51% 3.76% CPR 15.00% 15.00% 15.00% 15.00% Amortized Cost Basis 90,000 90, , ,000 Net Amount Expected to be Collected 91,700 89, , ,703 Excess / (Shortfall) 1,700 1,000 (3,510) (2,297) 10

11 When losses are introduced, EIR will always result in offsetting of expected losses when the financial asset is in a net discount position and inversely impacting assets in a net premium position. This issue is present under current GAAP. ASC , originally issued as FASB 114, requires institutions to discount expected cash flows using EIR. However, the impact to the overall allowance levels are less material when compared to the overall loan portfolio reserved in accordance with ASC Institutions utilizing a Discounted Cash Flow approach as described in may find that adjusted EIR more accurately reflects the impact of credit losses on the net amount expected to be collected. If net deferred fees or remaining fair value adjustments are material, adjusted EIR is a more appropriate election. Regardless, executing parallel calculations and comparing the outcomes over time will ensure proper elections for each class of financing receivable. Probability of Default Probability of Default (PD) is a term utilized to indicate the likelihood that a financial asset will reach a default status over a certain range of time and is exhibited as a percentage. Prior to CECL, probability of default was a metric mainly used by analysts and portfolio managers. Today, the metric is more familiar but often misused when applied in an ALLL setting. There are many models and alternative methods used to derive PD. Some models utilize historical databases to observe actual defaults while some models utilize credit default swap and bond pricing in an effort to leverage market data. ASU does allow for both internal and external information in the development of its expected credit loss, in this case PD: When developing an estimate of expected credit losses on financial asset(s), an entity shall consider available information relevant to assessing the collectibility of cash flows. This information may include internal information, external information, or a combination of both relating to past events, current conditions, and reasonable and supportable forecasts. An entity shall consider relevant qualitative and quantitative factors that relate to the environment in which the entity operates and are specific to the borrower(s). When financial assets are evaluated on a collective or individual basis, an entity is not required to search all possible information that is not reasonably available without undue cost and effort. Furthermore, an entity is not required to develop a hypothetical pool of financial assets. An entity may find that using its internal information is sufficient in determining collectibility p , emphasis ours Historical credit loss experience of financial assets with similar risk characteristics generally provides a basis for an entity s assessment of expected credit losses. Historical loss information can be internal or external historical loss information (or a combination of both). An entity shall consider adjustments to historical loss information for differences in current asset specific risk characteristics, such as differences in underwriting standards, portfolio mix, or asset term within a pool at the reporting date or when an entity s historical loss information is not reflective of the contractual term of the financial asset or group of financial assets p111, emphasis ours 11

12 When deriving PD using historical databases, loan-level attribute files are analyzed for default criteria. The definitions for default status vary across institutions, asset classes, and models. Common criteria include days past due, accrual status, TDR, chargeoff, and credit quality indicator thresholds. The approaches used to determine PD are equally as diverse. However, it is imperative that the methodology used to determine PD is symmetrical to the application. Topic 326 requires that expected credit losses reflect losses for assets on balance sheet as of a specific point-in-time. Any expected losses related to off-balance sheet exposure (future funding) should be recorded as a liability. Therefore, the analysis should also attempt to determine the probability of default of on-balance sheet exposure as of a specific pointin-time. For this purpose, a proper PD model will utilize loan-level detail, track full loan populations over an appropriate range of time, and identify loans that entered into a default state independent of the eventual loss outcome. When utilizing external data to derive PD or when attempting to apply default rates derived by a third-party using external data, institutions must consider adjustments for differences in asset specific risk characteristics and term. Additionally, institutions must understand the methodology used in the determination of PD in order to ensure proper application. When the application is asymmetrical, adjustments should be made to accommodate such discrepancies. This is also true for internal analysis, however, internal data is often inherently more analogous. An entity shall consider adjustments to historical loss information for differences in current asset specific risk characteristics, such as differences in underwriting standards, portfolio mix, or asset term when an entity s historical loss information is not reflective of the contractual term of the financial asset or group of financial assets p111, emphasis ours One common area of PD misuse and/or area of modeling risk, is the range of time the probability of default expresses. For deployment in a Discounted Cash Flow model, PD should be calculated and expressed as an annual default rate. In the example shown below, the performance of a pool of like-kind assets are analyzed for annual default behavior. Note that periodic observations of future performance over a one-year period provides for multiple observations and trend evidence of annual default behavior: 12

13 The above example utilizes a count-based probability of default. When back-tested, a count-based PD will not result in an exact recreation of loss expectancy as a dollar-based PD would. However, count-based PD is inherently less volatile and carries less modeling risk. The following example illustrates how a dollar-based probability of default can introduce volatility and result in modeling risk when used as a basis for future default estimation: 1 750,000 Y 525,000 75, ,000 Y 450, , ,000 Y 200, ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A - Measurement Period As of 12/31/16, on balance sheet loans are analyzed over the subsequent 12-month period capturing default and loss activity 13

14 A. PD (Dollars) 1,500,000 2,025, % C. LGD Default) 175,000 1,175, % B. PD (Dollars) % D. LGD (Measurement Period Balance) 175,000 1,500, % PD and LGD Resulting PD (A. dollar vs. B. count) and LGD (C. balance at default vs. D. measurement period balance) 4 100,000 74,074 8, ,000 30,000 3, ,000 37,037 4, ,000 15,000 1, ,000 18,519 2, ,000 7, ,000 37,037 4, ,000 15,000 1, ,000 55,556 6, ,000 22,500 2, ,000 74,074 8, ,000 30,000 3, ,000 92,593 10, ,000 37,500 4, ,000 74,074 8, ,000 30,000 3, ,000 74,074 8, ,000 30,000 3, ,000 74,074 8, ,000 30,000 3,500 Total 825, ,111 71,296 Total 825, ,500 28,875 Total % 74.1% 9.6% Total % 39.9% 3.5% Estimation Period Comparison of dollar-based PD and count-based PD - A x D and B x D (see Loss Given Default section for LGD comparison) Utilizing prior period observed dollar-based default behavior as a basis for subsequent 12-month expectancy would result in an assumption requiring 7-8 of the remaining 10 loans to default. While this example highlights an extreme case, it certainly begs the question; does the fact that previous defaults occurred on larger or smaller loans indicate that more or less borrowers will default in the future? In our experience, the benefits of count-based PD measurements when utilized as a basis for loss estimation outweigh the benefits of dollar-based PD. In other words, dollar-based default rates that weight each default by outstanding balance are useful for benchmarking and backtesting but are less useful for estimating future defaults. Regardless of the preferred method, institutions should document the election and attempt to apply the results in a consistent and objective manner. When applied to a series of expected cash flows, periodic defaults will affect subsequent interest, prepayment, and defaults. As shown in the example below, Ending Balance is a function of principal collection, prepayment, and monthly defaults. The resulting balance serves as the beginning balance in the next period, which will drive interest, prepayment, and defaulted dollars: Date Beginning Balance Principal Interest Prepayment Defaulted Principal Estimated Loss Estimated Recovery Ending Balance Cash Flow Present Value 10/31/2017 $482,530, $2,204, $1,748, $6,253, $77, $13, $0.00 $473,995, $10,205, $10,169, /30/2017 $473,995, $3,796, $1,662, $6,137, $76, $13, $0.00 $463,985, $11,596, $11,513, /31/2017 $463,985, $10, $1,681, $5,916, $73, $12, $0.00 $447,700, $17,892, $17,699, /31/2018 $447,700, $2,139, $1,621, $5,808, $145, $29, $0.00 $439,607, $9,569, $9,434, /28/2018 $439,607, $3,511, $1,438, $5,689, $142, $29, $0.00 $430,264, $10,640, $10,453, /31/2018 $430,264, $3,475, $1,558, $5,576, $138, $28, $0.00 $421,073, $10,610, $10,379, /30/2018 $421,073, $2,236, $1,474, $5,471, $212, $48, $0.00 $413,153, $9,182, $8,960, /31/2018 $413,153, $3,198, $1,495, $5,352, $207, $47, $0.00 $404,394, $10,045, $9,767, /30/2018 $404,394, $8,966, $1,417, $5,220, $200, $45, $0.00 $390,007, $15,604, $15,098, /31/2018 $390,007, $2,282, $1,410, $5,117, $232, $54, $0.00 $382,374, $8,810, $8,506, /31/2018 $382,374, $5,057, $1,383, $4,978, $225, $53, $0.00 $372,112, $11,418, $10,946, /30/2018 $372,112, $7,544, $1,299, $4,802, $218, $51, $0.00 $359,547, $13,646, $13,152,

15 As previously mentioned, symmetrical application relative to the determination of PD is vital. If default rates were derived or disaggregated by months since origination or vintage, the default rates should be applied in a similar manner. Likewise, if default rates were derived independent of vintage, the application should follow suit. It is common practice when using a DCF in the accomplishment of analogous accounting standards to disaggregate by risk characteristics in lieu of vintage characteristics. Disaggregation by both vintage and risk can often lead to statistical population deficiencies when utilizing historical databases and/or create impractical modeling requirements. In this next example, each underlying loan exhibited earns a specific default and loss rate based on its risk rating at the time of analysis. Historical database analysis indicates prior performance (left), a reasonable and supportable forecast directs periodic future performance (right), and resulting cash flow analysis displays the effect (below): 15

16 Note how the various considerations required by ASU have been addressed; historical performance provides a basis, environmental adjustments to historical loss information are quantified, and consideration for prepayments as a separate input is accomplished. It should be acknowledged that annual Probability of Default metrics are of little or no use in the accommodation of Topic 326 without a process to apply such rates over the contractual lives of the various classes of financing receivables. If institutions are planning to utilize a stand-alone Probability of Default and Loss Given Default model, the range of time considered in the determination of historical PD must match that of the expected life of the instrument under consideration. For more information on the use case of Probability of Default and Loss Given Default models independent of a Discounted Cash Flow model, refer to Practical CECL Transition Probability of Default and Loss Given Default. Loss Given Default Probability of Default is a measurement of the likelihood that a certain percentage of borrowers will be unable or unwilling to meet contractual obligations. Probability of Default alone does not provide for an estimate of expected loss. Rather, expected loss is a product of Probability of Default and Loss Given Default (LGD) or PD x LGD = EL. Loss Given Default reflects the expected loss amount in the event of a default and is typically presented as a percent. The LGD calculation itself is theoretically simple; loss divided by balance. However, both the numerator and denominator require certain elections to be made. Such elections are wholly dependent upon how the outcome will be utilized. As with other inputs and assumptions, a fundamental understanding of Topic 326 and the modeling approach is vital to conceptual soundness both in the calculation and utilization of LGD. One key point of understanding is the relationship with Probability of Default. Because expected loss is the product of PD and LGD, if little is known about how PD was derived or asymmetry exists within the loans and/or loan types analyzed, the product yields little in the way of meaningful loss estimation. This principle is best communicated with a simplified back-testing exercise. In the example below, historical experience is analyzed in order to serve as a basis for PD and LGD. Applying observed PD and LGD to the same measurement period should theoretically result in the same level of losses. In other words, to predict future losses having a perfect knowledge of subsequent events. While the figures used are dramatic for visualization purposes, note how both numerator and denominator inclusion and exclusion elections will have a material impact on predicted outcomes: 1 100,000 Y 70,000 10, ,000 Y 90,000 20, ,000 Y 50, ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A ,000 N N/A - Measurement Period As of 12/31/16, on balance sheet loans are analyzed over the subsequent 12-month period capturing default and loss activity 16

17 A. PD (Dollars) 300,000 1,000, % C. LGD Default) 30, , % B. PD (Dollars) % D. LGD (Measurement Period Balance) 30, , % PD and LGD Resulting PD (A. Dollar vs. B. Count) and LGD (C. Default vs. D. Measurement Period Balance) rates are compared Expected Loss (AxC) 4.29% 1,000,000 42,857 30,000 12,857 Expected Loss (AxD) 3.00% 1,000,000 30,000 30,000 - Expected Loss (BxC) 4.29% 1,000,000 42,857 30,000 12,857 Expected Loss (BxD) 3.00% 1,000,000 30,000 30,000 - Expected Loss The various combinations of PD and LGD are applied to the measurement period exposure and compared to actual activity When planning to utilize LGD to determine future losses on current exposure or exposure as of a certain point-in-time, as Topic 326 requires, utilizing the beginning balance results in a more accurate estimation due to the inherent reliance on PD and measurement period balance. Because the allowance for credit losses is presented on financial statements as a valuation account deducted from the amortized cost basis, the allowance should reflect the net amount expected to be collected on the basis presented on the face of the balance sheet. Credit losses expected on off balance sheet exposure should be recorded as a liability. Therefore, institutions should take steps to prevent the perversion of allowance levels with losses on future funding: The allowance for credit losses is a valuation account that is deducted from the amortized cost basis of the financial asset(s) to present the net amount expected to be collected on the financial asset p109, emphasis ours In estimating expected credit losses for off-balance-sheet credit exposures, an entity shall estimate expected credit losses on the basis of the guidance in this Subtopic over the contractual period in which the entity is exposed to credit risk via a present contractual obligation to extend credit, unless that obligation is unconditionally cancellable by the issuer. At the reporting date, an entity shall record a liability for credit losses on off-balance-sheet credit exposures within the scope of this Subtopic p112, emphasis ours As with other inputs and assumptions, an institution s own data may be insufficient due to a shortage of loss observations, availability of historical loan-level data, the introduction of a new line of business, a new lending practice, or some other common yet limiting cause. In many cases, external data may provide for improved modeling accuracy or mitigate such limitations. External credit loss data can come in many forms, however, it is difficult to obtain analogous PDs and LGDs over a long range of historical periods for many loan types as loan-level detail is required in order to derive such baseline measurements. A significant benefit of a Discounted Cash Flow approach is the ability to deploy annualized inputs and assumptions to derive a lifetime estimate. When evaluating external sources, and in many cases expanding internal sources, the Federal Financial Institutions Examination Council (FFIEC) provides institutions with a valuable source of information. Unfortunately, the information is limited to pool-level data, which inhibits the ability to compute many key inputs and assumptions. However, quarterly or annual loss information is relevant provided it is used appropriately. 17

18 Suppose an institution employs the approaches for PD and LGD mentioned above utilizing their own internal database(s). Upon analyzing the results, it becomes apparent that certain classes of financing receivables with certain risk characteristics contain statistically limited default and/or loss observations. Rather than relying on limited observations, it may be more appropriate and/or practical to expand the analysis to include default and/or loss behavior experienced by like-kind institutions. Below is an example PD and LGD calculation utilizing internal information illustrating where limited default and loss observations may justify the introduction of external data. CRE Risk Grade 3 credits experienced few defaults over the analysis period, which serve as the population for LGD determination Upon aggregating 12 months of CRE defaults, 12 defaults and 3 chargeoffs serve as the entire population for LGD determination at 3/31/15 Using 3/31/15 as an example, 9 out of 12 observed defaults experienced no loss, 3 out of 12 experienced losses ranging from 9 percent to 36 percent. In total, LGD for that 12-month analysis period is roughly 2%. Taking into consideration all analysis periods, LGD results range from 0 percent to 40 percent, which is a conceptually unsatisfying and a common phenomenon when observations are limited. 18

19 Likewise, suppose an institution employs the approaches for PD and LGD mentioned above utilizing their own internal database(s). Upon analyzing the results, it becomes apparent that historical availability is limited. Rather than relying on a condensed timeframe and a singular economic environment, it may be more appropriate and/or practical to expand the analysis to include default and/or loss behavior experienced by like-kind institutions over an expanded time horizon. Below is an example PD and LGD calculation utilizing internal information illustrating where limited historical loan-level data may justify the introduction of external data. CRE Risk Grade 3 credits experienced few defaults over the analysis period; however, the periods analyzed were limited Due to historical data limitations, loss observations are few resulting in an inappropriate go-forward LGD basis Taking into consideration all analysis periods, LGD results range from 0 percent to 13 percent, which is a conceptually unsatisfying and a common phenomenon when observations are limited. Note that this is the same portfolio as the previous example with a notional limitation on available data. In the previous example, LGD ranged from 0 to 40 percent resulting in dramatically different outcomes. Institutions may mitigate the impact of common internal database limitations by introducing industry data. The utilization of peer, industry, or readily available pool-level internal information can take many forms. LGD is a key driver in expected loss estimation yet is often a bottleneck for modeling validity. Recognizing that LGD computation should be symmetrical to other estimates used by the institution. We know intuitively and by evidence that LGD rises as PD rises, but the exact nature of that relationship can be difficult to quantify and beyond the scope of most institutions interior capabilities or availability of data. Therefore, an institution might elect to measure, assume, or 19

20 incorporate static industry LGD when estimating expected losses. Alternatively, many methods exist for relating loss given default to probability of default. When relating PD and LGD, Sageworks Advisory Services is partial to the research performed by economists Frye and Jacobs, which is summarized in a 2013 paper 1. While technically complex, this model can be summarized as follows: LGD can be related to PD by a single parameter, k We assume that certain asset classes have a particular k value As k approaches 1.0, LGD is more insensitive to default rates (e.g. loans collateralized by first liens on real estate) As k approaches 0, LGD is more sensitive to default rates (e.g. a consumer signature loan) k can be measured in one environment (e.g. a largely performing economy) and applied in a different environment (e.g. a deteriorating economy). The use of the single parameter, k, conceptually simplifies the relationship between PD and LGD. While institutions may not be equipped to estimate this value based on their own data, Sageworks will provide studies of this parameter as observed in its proprietary data set on different asset classes for clients to use in their analysis. Thus, allowing institutions to determine PD and/or LGD where internal or external pool-level information or information made publicly available from the FFIEC is required or preferred. Suppose either, or both, of the preceding illustrated limitations are present when utilizing internal information for PD and LGD determination. Under such conditions, it may be more appropriate to incorporate FFIEC quarterly loss rate information in conjunction with an industry-level parameter, k. Such an approach would allow for the conversion from loss rate to PD and LGD. In the example shown below, if k as provided as an industry-level assumption for a particular class of financing receivable was equal to.75, and FFIEC information indicated an annualized loss rate of 75 bps for one (or a group of) institution(s), PD and LGD may be derived for deployment within an expected cash flow schedule:.05(pd) x.15(lgd) =.0075(Loss Rate) 2 1 The endorsement of the research authors or the Federal Reserve Bank of Chicago for this application of the work has neither been sought nor provided. 2 Graph from 20

21 The computations described above can be completed with Excel and other freely available tools, and would typically not be recomputed as a matter of policy barring the availability of additional data through the passage of time or a high-loss environment (in which case measurement, rather than estimation, may be preferred). The use of peer or industry data in performing analysis should not be taken lightly, but institutions ought to be prepared to consider and confront data limitation issues for some classes of financing receivables. Utilizing readily available pool-level and industry-level information can help overcome limitations, improve modeling accuracy, and mitigate volatility. Documentation of these considerations, as well as a long-run plan to remediate them, are important. For more information on the relationship between PD and LGD and the utilization of the parameter used to describe the relationship, refer to the document Practical Transition Loss Driver Analysis. Recovery/Foreclosure Delay Timing is inherently material within a discounted cash flow approach. Contractual obligations, prepayments, defaulted principal, loss, and eventual recovery/foreclosure require timing consideration. Recovery and/or foreclosure delay are typically assumptions unique to each class of financing receivable. The purpose this assumption is to allow for the impact of defaulted principal while simultaneously allowing for the recognition of secondary, tertiary, or mitigating sources of repayment such as collateral. To illustrate how this assumption is used within an expected cash flow model and the impact recovery/foreclosure delay can have on the net amount expected to be collected, two examples are presented. Both examples contain analogous inputs and assumptions with the exception of recovery/foreclosure delay. The first example assumes a 12- month delay while the second assumes an immediate recovery/foreclosure: Date Beginning Balance Principal Interest Prepayment Defaulted Principal Estimated Loss Estimated Recovery Ending Balance Cash Flow Present Value 10/31/2017 $482,530, $2,248, $1,748, $6,253, $75, $13, $0.00 $473,993, $10,249, $10,213, /30/2017 $473,953, $3,838, $1,662, $6,136, $73, $12, $0.00 $463,904, $11,636, $11,553, /31/2017 $463,904, $10,337, $1,680, $5,915, $71, $12, $0.00 $447,581, $17,933, $17,740, /31/2018 $447,581, $2,181, $1,621, $5,806, $140, $28, $0.00 $439,452, $9,609, $9,473, /28/2018 $439,452, $3,549, $1,438, $5,687, $137, $28, $0.00 $430,079, $10,675, $10,487, /31/2018 $430,079, $3,516, $1,557, $5,573, $134, $27, $0.00 $420,854, $10,648, $10,417, /30/2018 $420,854, $2,275, $1,474, $5,468, $205, $46, $0.00 $412,905, $9,218, $8,995, /31/2018 $412,905, $3,238, $1,494, $5,348, $201, $45, $0.00 $404,116, $10,082, $9,803, /30/2018 $404,116, $9,004, $1,416, $5,217, $193, $43, $0.00 $389,700, $15,638, $15,132, /31/2018 $389,700, $2,321, $1,409, $5,113, $224, $52, $0.00 $382,041, $8,844, $8,539, /31/2018 $382,041, $5,096, $1,382, $4,973, $218, $51, $0.00 $371,752, $11,452, $10,978, /30/2018 $371,752, $7,582, $1,298, $4,798, $210, $49, $0.00 $359,162, $13,678, $13,183, /31/2018 $359,162, $4,772, $1,302, $4,674, $236, $57, $62, $349,479, $10,811, $10,312, Estimated Loss is a function of LGD. The recoverable portion is delayed 12 months 21

22 Date Beginning Balance Principal Interest Prepayment Defaulted Principal Estimated Loss Estimated Recovery Ending Balance Cash Flow Present Value 10/31/2017 $482,530, $2,248, $1,748, $6,253, $75, $13, $62, $473,953, $10,311, $10,274, /30/2017 $473,953, $3,838, $1,662, $6,136, $73, $12, $60, $463,904, $11,697, $11,614, /31/2017 $463,904, $10,337, $1,680, $5,915, $71, $12, $58, $447,581, $17,992, $17,798, /31/2018 $447,581, $2,181, $1,621, $5,806, $140, $28, $111, $439,452, $9,720, $9,583, /28/2018 $439,452, $3,549, $1,438, $5,687, $137, $28, $109, $430,079, $10,784, $10,594, /31/2018 $430,079, $3,516, $1,557, $5,573, $134, $27, $106, $420,8854, $10,754, $10,521, /30/2018 $420,854, $2,275, $1,474, $5,468, $205, $46, $158, $412,905, $9,377, $9,150, /31/2018 $412,905, $3,238, $1,494, $5,348, $201, $45, $155, $404,116, $10,238, $9,954, /30/2018 $404,116, $9,004, $1,416, $5,217, $193, $43, $149, $389,700, $15,788, $15,277, /31/2018 $389,700, $2,321, $1,409, $5,113, $224, $52, $171, $382,041, $9,015, $8,704, /31/2018 $382,041, $5,096, $1,382, $4,973, $218, $51, $166, $371,752, $11,618, $11,139, /30/2018 $371,752, $7,582, $1,298, $4,798, $210, $49, $160, $359,162, $13,839, $13,337, Estimated Loss is a function of LGD. The recoverable portion is immediate In the first example where the estimated repayment via collateral is expected to take place 12 months following a default, the suggested reserve for the entire asset class is $6.7MM or 83 bps. However, in the second example no delay is modeled and an immediate recovery/foreclosure is assumed resulting in a suggested asset class reserve level of $5.9MM or 72 bps. Even though the recovery/foreclosure amounts are identical, the timing resulted in a 12.5 percent variance in required reserve. An analysis may be performed at the class of financing receivable level on past recoveries and foreclosures to determine a defensible assumption. In many cases, a simple 6, 12, or 24 month assumption is used which may be based on past experience or state-by-state legal considerations for certain asset classes. For this assumption, consistency takes precedent. This assumption is not one that would or should be updated with every reserve calculation. Rather, this assumption is better suited as a policy election made at the class of financing receivable level with a documented plan to calibrate the input periodically. 22

23 FORECAST APPLICATION The standard s requirement to apply forward-looking projections in allowance modeling is a major shift from the incurred parameters set forth in ASC (FAS 5). There is not a single, prescribed way to fulfill this requirement, and institutions of different sizes, natures, and complexities may take very different approaches. In application, the standard requires forward projections, or adjustments to historical losses, to include the impact of forecasted conditions and that the conditions be reasonable and supportable. When developing an estimate of expected credit losses on financial asset(s), an entity shall consider available information relevant to assessing the collectibility of cash flows. This information may include internal information, external information, or a combination of both relating to past events, current conditions, and reasonable and supportable forecasts p110, emphasis ours An entity shall not rely solely on past events to estimate expected credit losses. When an entity uses historical loss information, it shall consider the need to adjust historical information to reflect the extent to which management expects current conditions and reasonable and supportable forecasts to differ from the conditions that existed for the period over which historical information was evaluated p111, emphasis ours Because the standard requires that expected losses are reflective of the remaining contractual term, many have voiced concerns over their ability to forecast conditions accurately or even reasonably many years into the future. However, there is no requirement or expectation to forecast conditions over the contractual term. In fact, provision for this inherent limitation has been made within the standard itself. Some entities may be able to develop reasonable and supportable forecasts over the contractual term of the financial asset or a group of financial assets. However, an entity is not required to develop forecasts over the contractual term of the financial asset or group of financial assets. Rather, for periods beyond which the entity is able to make or obtain reasonable and supportable forecasts of expected credit losses, an entity shall revert to historical loss information determined in accordance with paragraph that is reflective of the contractual term of the financial asset or group of financial assets. An entity shall not adjust historical loss information for existing economic conditions or expectations of future economic conditions for periods that are beyond the reasonable and supportable period. An entity may revert to historical loss information at the input level or based on the entire estimate. An entity may revert to historical loss information immediately, on a straight-line basis, or using another rational and systematic basis p111, emphasis ours Some institutions may choose to decouple the impact of forecasted conditions from loss estimation. While this approach may be appealing due to familiarity with today s incurred loss practice involving qualitative and environmental factors, in an expected loss environment such an election may prove extremely difficult to maintain directional consistency while simultaneously mitigating volatility. 23

24 In application, it becomes rapidly apparent that many modeling approaches carry inherent weaknesses. In many cases, the most significant weakness relates to the application of a reasonable and supportable forecast with the exception being application within a Discounted Cash Flow model. DCF most appropriately aligns with the manner in which forecasts are prepared, allows for conditional impact quantification, and requires significantly less loan-level historical data to support such effects. The following illustrations outline a practical process by which an institution may quantify the impact of a given economic forecast. Infrequent analysis (bi-annual or greater) will identify relevancy and influence. Below is the historical application (left) of selected factors (right) Reasonable and supportable conditions are forecasted for 24 months reverting to historical averages thereafter The impact of forecasted conditions on periodic default rates (left) and periodic expected losses (right) 24