Many institutions are struggling to apply the CECL standard as it pertains to credit cards, and in particular determining the lifetime value for credit card portfolios. In this paper, we explore the different approaches to evaluating lifetime estimates for the credit card portfolio.
Life changes; so does your allowance
In June 2016, the Financial Accounting Standards Board (FASB) issued ASU 2016-13, Measurement of Credit Losses on Financial Instruments, known as the current expected credit loss (CECL) standard. Intended to improve financial reporting, it requires earlier recognition of expected credit losses for assets measured at amortized cost. More granular disclosures of the potential risk in a bank’s portfolio are also expected.
CECL replaced the previous incurred loss impairment model. The incurred model was based on current and historical conditions only, recognized losses over a loss emergence period, and only after the probable threshold was met. CECL is also based on historical and current information, but removes the “probable” threshold. It introduces reasonable and supportable forecast conditions, and requires the recognition of lifetime expected credit losses at origination or date of purchase.
Credit card allowance process today
Under Generally Accepted Accounting Principles (GAAP) (FAS51), the standard practice for calculating allowances is to estimate projected losses over the next 12 months. It can be less if an issuer can reasonably document that the life of the portfolio is less than 12 months. According to the FDIC Credit Card Activities Manual2 the typical range of methodologies includes roll-rate, average charge-off methods, vintage analysis, regression analysis, and portfolio liquidation method. This longstanding practice is about to be impacted significantly. However, for smaller institutions (less than $1 billion), the old method can remain largely intact, modified to account for the incremental CECL requirements.
CECL requirements addressed
In this paper, we address the determination of the contractual lifetime value for credit card portfolios and similar assets that do not have a preset maturity date. ASU 2016-133 states: “… An entity shall estimate expected credit losses over the contractual term of the financial asset(s) …” – meaning that for assets without contractual maturities, the contractual life must be determined by using analytical approximation in some fashion.
The industry as a whole is still struggling with the application of the CECL standard as it pertains to credit cards. For institutions looking for a simple yet robust approach: there is no “easy” button! We dissect the problem and attempt to provide a low-ingredient, theoretical recipe that allows institutions to start thinking about their own approaches. For the initial implementation of the allowance on credit cards, let’s first review some of the key elements that can be considered when determining “contractual lifetime.”
Allowable lifetime estimation method4
The key lifetime metric is the time it takes to run off a current outstanding balance. It is based on expected payments after interest and fees have been paid. FASB has tried to take the appropriate steps to make the standard applicable on this issue by providing two main choices:
- FIFO5 (referred to by FASB as “View A”): determines lifetime based on the application of all future payments to the current balance until it is extinguished.
- Pro-rata (referred to by FASB as “View B”): determines lifetime based on the application of all future payments on a pro-rata basis between future draws and current balances until the current balance is extinguished.
Application of payments
Almost universally, banks apply credit card payments to interest and fees before applying them to principal balance. It is important to consider this notion. Whether the FIFO or Pro-rata approach is used, apply the future payment forecast first to interest and fees, then to the reduction in the principal balance.
FASB requires institutions to estimate the expected credit loss of off-balance sheet credit exposures6 unless the obligation is unconditionally cancelable by the issuer. Since most cards are unconditionally cancelable, future draws are not considered within the calculation of ECL under the GAAP view of reserving. It is unlike current processes for stress testing and capital planning where amounts funded in the future must be considered.
The CARD Act7 enacted in 2009 for the protection of consumers says to apply payments to the balances with the highest rates. Hence, an institution considering the Pro-rata approach, will forecast future balances and rates to determine which balances have the highest rates.
Reasonable and supportable forecast
The CECL guidance provides for the notion of a “reasonable and supportable” forecast horizon, after which an institution reverts to long-term historical averages. Paragraph 326-20-30-9 states: " . . . 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 can make or obtain reasonable and supportable forecasts of expected credit losses, an entity shall revert to historical loss information determined in accordance with paragraph 326-20-30-8 that is reflective of the contractual term of the financial asset or group of financial assets . . . "
Institutions thinking about the payment forecast for their portfolios must consider how the reasonable and supportable requirement would impact the segments. Relatively longer principal pay-down periods can involve some type of reversion to long-term average payments.
Justification and documentation
The burden of proof for all assumptions made in the determination of the lifetime estimates is subject to the “professional skepticism” of audit firms. All assumptions must correlate to institutions’ historical experience. It must represent management’s best interpretation of the impact of current conditions and a reasonable and supportable forecast. We expect the burden of proof for documentation of lifetime assumptions to be material on the teams delivering quarterly CECL estimates.
Before considering one approach or the other, there is a fundamental question: are future remits influenced by future spend? If not, logic holds that FIFO is the only rational answer. Consideration of anything past the balance sheet date is not in line with GAAP. Payments associated with an unfunded commitment are akin to reserving for these payments. Others feel that this approach might put institutions at regulatory risk except for the smallest of portfolios, for which it can still be an appropriate decision.
The industry agrees that the Pro-rata approach is conservative, while the FIFO approach can lead to more pro-cyclicality at the turn of the business cycle. The Pro-rata approach might lead to a larger capital hit initially but will be less pro-cyclical as we barrel towards the next downturn. The FIFO method ramps up reserves at the worst possible time, after customers start using their credit cards as a last resort. Whether you believe that either approach is GAAP-compliant – or not – the FASB Transition Resource Group (TRG) working group concludes that both approaches are acceptable.
Most institutions decided not to follow the CARD Act to the letter, since it has not been explicitly mandated for CECL purposes. There are also some operational considerations since some (if not most) card processors do not give individual loan-level payment information before default. Therefore it is difficult to apply CARD Act payments vs. new expected draws to get a non-FIFO life-of-loan balance. Based on these considerations, most institutions do not forecast using the CARD Act, and instead follow guidelines using approximation. For example, apply haircuts to future payments to account for the higher interest balances. CECL guidance is clear about using information that is obtainable without undue cost or effort. So approximation and judgment can prevail in this case for the initial implementation.
Following is a set of steps for consideration when evaluating lifetime estimate for your institution’s credit card portfolio:
Step 1: Segmentation
Step 2: Approach to lifetime estimation
Step 3: Qualitative adjustments to consider
Step 1: Segmentation
Pool the credit card portfolio by common risk characteristics for evaluation purposes8. For credit cards, the following segments can be considered, over and above the ASU 2016-13 recommended segmentation characteristics9:
- Borrower type: anticipate segmenting by borrower type and pay special attention to the percentage of borrowers that migrate between these two segments10:
- Revolvers: customers that typically carry a balance
- Transactors: customers that typically pay in full by the grace period
It is expected that the composition and level of migration between segments have a significant impact on lifetime determination and therefore be tracked closely. More transactors (higher-quality borrowers) means lower revolving balances and higher payments, and so a much shorter life and a lower CECL reserve. At the other end of the spectrum, companies that go downstream and make more on revolving fees have revolvers, with higher revolving balances, and lower payments as a percentage of balance. Effectively, the latter group has a longer lifetime average, driving a much higher CECL reserve than the former.
Other typical card segments also include:
- Prime borrowers
- Near-prime borrowers
- Subprime borrowers
- Super subprime borrowers
- Private label vs. general purpose
The more granular the segmentation, the more challenging historical data collection for each segment becomes. Institutions can also turn to data providers like Equifax to supplement their analysis of portfolios.11
Step 2: Approach to initial lifetime estimation
Regardless of the final approach chosen, a reasonable strategy is to first do a historical analysis for each segment, using both the FIFO and Pro-rata approaches. The goal is to understand the portfolio and segment payment behaviors, as well as the level of fees and interest based on different periods over an economic cycle.
Paragraph 326-20-30-7 of the CECL guidance specifically refers to " . . . relating to past events, current conditions, and reasonable and supportable forecasts . . . ” – mapping your data history to historical conditions and then relating it to today’s conditions for adjustment is critical. Forecasting economic conditions and the impact to payments, fees, and interest is a key determination which can either be modeled or estimated via Q-Factor adjustment12. All these items have an impact on the forecast of principal pay-downs, and therefore on the lifetime estimate as well.
A common sense method to solving for the reasonable and supportable forecast horizon13 is based on current practices used in budgeting and planning for credit card portfolios. It minimizes the differences in these two processes. The historical analysis can give us insights into the payment-level averages by segments to be used after the reversion period. Typically, length of reversion can be observed from historical data. For example, for a given segment, what is the long-term average payment rate? Then, based on observing historical deviation from the mean, what is the average length of time it takes to revert to the mean? This tactic is used industry-wide for determining the reversion period.
We also recommend that you look at potential funded commitment to understand the potential Q-factor adjustments. It might be required from a safety and soundness perspective regardless of the approach chosen, especially when the off balance sheet relative percentage is deemed to be high14.
Now, let us consider the importance of documenting the steps taken in the preceding section – and keying in on the differences between the two approaches. The documentation of this complex process will be necessary for future audits. The required analysis outlined in this section illustrates the difficulty and amount of work required to comply with CECL in the estimation of the lifetime metric for the approach chosen by the institution.
Step 3: Qualitative adjustments to consider
Based on the set of questions listed in the following section, institutions must think through the potential adjustments and document and analyze internal data for these factors. Institutions can also use external data analysis to help supplement any lack of internal data, which is deemed a CECL-compliant15 practice:
- Historical payment patterns:
- How different is the current total aggregate monthly payment from our historical patterns?
- Did the firm receive an unusual number of catch-up payments due to seasonality or other reasons?
- What is the long-term average payment trend of the borrowers?
- How does this trend compare to the industry?
- How large is the short-term volatility of payment patterns?
- Did the firm detect a larger than usual increase in line usage, which could indicate upcoming losses?
- Is our monthly average payment fairly stable? If not, how does the lender account for those changes?
- Future payment patterns:
- How does the firm think the payment pattern might differ over the reasonable and supportable forecast period (based on current understanding of where we are in the economic cycle)?
- Does the lender expect changes in terms and conditions that can affect future payments? For example, as the Federal Reserve increases interest rates, how fast does the firm adjust the rates?
- Do we expect line increases or decreases soon that might impact payment patterns?
- Are we seeing or anticipating a significant change in the percentage of transactors vs. revolvers within the portfolio that could adversely impact payment patterns?
The FASB adopted the principle-based guidance for the current expected credit loss standard such that it can be adaptable to institutions of all sizes. One of the most difficult aspects of loss forecasting under CECL for consumer credit portfolios is credit cards, especially the lifetime calculation. The credit card portfolio presents some unusual challenges that simply have not been part of any processes before. It likely requires changes in operations16 over time to understand and capture each data point required to determine the lifetime metric.
We believe that with the right balance of analysis, and starting with an acceptable, simplified approach, the credit card lifetime determination challenge can be met. The Moody’s Analytics ECCL framework uses Equifax industry performance data and Moody’s Analytics econometric models to deliver lifetime loss forecasts under the CECL standard using both the FIFO and Pro-rata approaches. Users can calibrate industry credit card models to their portfolio data and compare results with either approach to determine impact quickly during implementation. The strategy of using industry models still relies on forecasted loss rates that are easily digestible for management and incorporates both views permissible by the FASB for comparison.
The Moody’s Analytics ECCL framework also incorporates reasonable and supportable economic scenarios; the loss forecasting models use macroeconomic variables that incorporate both current and future economic conditions, with mean reversion built into the results through input variables. The additional work would then rest on understanding what bank-specific qualitative factors to apply – and their justification – and the evaluation of unfunded commitments. Unfunded commitments are not in scope for CECL. It might be an important factor to consider based on the business cycle and the relative percentage of unfunded to funded balances, since this analysis is likely to be required for regulatory exam purposes.
In our paper, “Modeling Credit Card Losses under CECL,” we provide details on our methodology to calculate ECL based on both FIFO and Pro-rata approaches, which merge the effective realities of managing a credit card portfolio. It also ensures that the lifetime metric is not too simple or too complex in producing the right level of allowance for the credit card portfolio. Over time, we expect that even the smallest institutions will want the ability to estimate the lifetime metric under both permissible approaches. Analyzing the ECL delta between the two approaches to make educated decisions on the right level of reserves for their institution is also critical.
1 FAS5 common denomination for the accounting standard codification ASC-450-20
2 FDIC – Credit Card Activities Manual: https://www.fdic.gov/regulations/examinations/credit_card/ch12.html#sec5
3 Financial Instruments—Credit Losses (Topic 326) – paragraph 326-20-30-6
4 Memo #5 June 12, 2017 TRG - Determining the “Estimated Life” of a Credit Card Receivable
5 FIFO in this context not to be confused with the FIFO – “First in First out” methodology for inventory valuation.
6 Unfunded commitments are classified as off-balance sheet exposures.
7 Credit Card Accountability and Responsibility Disclosure Act of 2009: https://www.ftc.gov/sites/default/files/documents/statutes/credit-card-accountability-responsibility-and-disclosure-act-2009-credit-card-act/credit-card-pub-l-111-24_0.pdf
8 Although pooling of assets is required under CECL for evaluation purposes (e.g., disclosures) it doesn’t preclude institutions from using loan-level models and aggregating results at pool levels for disclosures. In fact, loan-level models might be the best for complex portfolios such as credit cards.
9 ASU 2016-13 paragraph 326-20-55-5: “Pooling segmentation outline includes: Internal credit score, financial asset type, size, EIR, location, vintage, historical loss patterns.
10 At a minimum, historical analysis should be performed to evaluate the movement between transactors and revolvers since the migration could have a significant impact on the ECL outcome. Historical analysis might require modeling of transitions if not stable.
11 CreditForecast.com, which is a partnership of Equifax and Moody’s Analytics, has segmentation for Credit Card vs. Retail Card as well as vintage, score band, and state granularity.
12 Q-Factor adjustments are estimations of qualitative factors using expert judgment, and are covered in the next section.
13 Meaning at what time period should one start reverting to historical long-term averages (payments, in the case of lifetime estimation).
14 The relative funded vs. unfunded commitment percentage is a key metric to keep track of when using the FIFO approach given the potential level of pro-cyclicality that this method may engender.
15 ASU 2016-13 paragraph 326-20-30-7: “… This information may include internal information, external information, or a combination of both …”
16 Some if not most card payment processors do not give individual loan-level payment information prior to default.