• Home  >  
  • Perspectives  >  
  • CECL in Loss Forecasting – Challenges and Opportunities  
Blog June 11, 2020
6 min read

CECL in Loss Forecasting – Challenges and Opportunities

Authors: Nidhi Chadha, Karthik Gandhi Introduction One of the key lessons the financial world took away from the Great Recession […]


Authors: Nidhi Chadha, Karthik Gandhi


One of the key lessons the financial world took away from the Great Recession of 2008 was that the existing and traditional GAAP approach of incurred loss methodology to recognize credit losses needed a rehaul. This helps Financial Institutions (FIs) maintain a sufficient loss reserve to ensure smooth sailing (or, at the very least, avoid sinking) in times of financial crises.

In this two-part series, we will explore various aspects of Current Expected Credit Loss (or “CECL”) – a recently implemented vital component of the current Loss Forecasting approach for FIs in the US.

In the first part, we will introduce the basics of CECL, the challenges it poses – especially for unsecured revolving portfolios such as credit cards, and discuss approaches for implementing CECL.

The second part will focus on how we developed and implemented CECL into the loss forecasting model for the credit card portfolio in a mid-sized US bank.

The American Bankers Association has called CECL “the most sweeping change to bank accounting ever” with good reason – the implications are huge. This change is expected to lead to an increase of 35-50% in loss reserve levels. While CECL was introduced by the FASB in 2016, it was to be implemented from January 2020 for large publicly traded banks and January 2023 for smaller reporting companies. However, in light of the COVID19 pandemic, the stimulus package signed by US President Donald Trump on 27th March, allows lenders to delay implementing CECL till 31st December or till health officials declare the end of the national emergency (whichever comes first).

The road to CECL

Even as the broader economic conditions started worsening in 2006, the loss forecasts by lenders raised no red flags. Banks are no strangers to risks or losses, and all of them had established loss forecasting models in place. Why, then, did their balance sheets fail to show any indication of the upcoming calamity? This can be attributed to the delayed identification of credit losses. Under the traditional method, FIs delayed recognition until it was probable that a loss had been incurred. This model restricted FIs ability to record credit losses that are expected, but do not yet meet the “probable” threshold. Under the new paradigm, FIs could then make estimates of “expected credit losses” using forward-looking information.

This led to the Comprehensive Capital Analysis and Review (CCAR) test being introduced by the Federal Reserve in 2011 for Bank Holding Companies with total assets of $50 billion or more. Composed of multiple stress tests, CCAR serves to ensure that banks have enough reserve capital in place to protect themselves in adverse financial conditions or market downturns.

However, this still was based on incurred loss. Cue: CECL. Introduced by the Financial Accounting Standards Board (FASB) in 2016, CECL, or Current Expected Credit Loss, transforms how FIs estimate losses. Under CECL, the expected lifetime credit loss (potential losses over the entire life of the loan) needs to be estimated at the time of origination. The value of the reserve needs to be based on the past, existing, and future information, taking into account the inherent risk for each loan as well as the economic factors.


While CECL does promise to provide a more accurate estimate of expected loss, it comes with a set of implementation challenges:

• CECL requires the credit loss allowance to be established for the entire foreseeable life of the loan, at the time of loan origination. This necessitates changes to the underlying credit loss models that currently estimate potential losses over a 12-month horizon.

• The Life of Loan forecast is not straightforward, especially for revolving lines of credit, like credit cards – where borrowers have the option to make minimum, partial or full payments at any given point in the monthly billing cycle. For products like mortgages, banks must keep in mind the possibilities of loan modifications and prepayments.

• Additional analysis is needed for economic risk factors – to assess their potential impact on credit losses better. The macroeconomic forecasts cannot merely be based on deviations in the past. CECL requires additional data granularity for macroeconomic variables, like real estate prices and unemployment rate.

• Since the FASB has not specified guidelines on the calculation methodology, banks need increased transparency with regard to the modeling assumptions. The justifications provided will have to be quantitatively supportable.

• The models will be audited internally and by the federal supervisory bodies for CECL standards, and the level of disclosure will increase substantially. This necessitates comprehensive documentation that captures the entire process.

• In addition to this come the structural challenges – the CECL compliant forecasts need to be integrated with the representation of losses in the general ledger, the firm’s prevalent accounting practices, and financial reporting systems.

CECL modeling approaches

Since CECL poses unique challenges for portfolios such as credit cards, we examine here the different approaches one can take to tackle this head-on. Both bottom-up and top-down regression approaches will help meet the CECL requirements.

Let’s discuss these approaches in detail:

I. Bottom-Up Approach:

This is used for building models for the most relevant segments. The credit loss model is built at the loan-level portfolio, using granular, geography-specific macroeconomic indicators. The granular data provides a better estimate for more complex portfolios, evaluating the changes over time on loan attributes and the performance for each account.

II. Top-Down Approach:

A vintage based approach can be utilized for portfolios or segments based on well-defined delinquency behavior. Typically, these segments (such as deceased, transactors, etc.) have a small life of loan. Hence loan-level data may not possess discriminatory power. Overall segment-level balanced information data can be used for these segments.

Bottom-up loss forecasting approaches have gained increased favor in recent years. Our experience with both methods inclines us to believe that the added value from the extra complexity outweighs the drawbacks.

Segmentation analysis, a crucial part of CECL models, needs to be performed separately for each of the dependent events, i.e., charge-off and prepay. The initial segments may reflect the delinquency status, transition status, payment activity, and tenure at observation points. The segment distribution can be examined at different snapshots to ensure that they are stable across time. A loan-level “bottom-up” modeling approach can be utilized for the major segments

Life of Loan Estimation

CECL standard defines lifetime as the life of the balance as of evaluation date (not the account opening date). According to the Transition Resource Group for Credit Losses (TRG), unused credit lines for a credit card can be treated as off-balance-sheet credit exposures that are unconditionally cancellable by the issuer. Since CECL does not require an allowance for available credit, it is not to be evaluated; determining the life of loan, thus, is the critical factor in the expected loss estimation. TRG discussed two views to determine the ‘estimated life’ of a credit card receivable, first-in-first-out (FIFO), and CARD Act method. Apart from this, there is a sophisticated hybrid approach combining both. Subsequently, TRG concluded that both methods are applicable, though FIFO is easier to implement. The amount of expected future payments estimated can be either of the following:

View A: All payments expected to be collected from the borrower

View B: Only a portion of payments expected to be collected from the borrower.

After reviewing various scenarios, considering consistency, simplicity, and justifiability, we are inclined towards using View A and FIFO methodology for future payments. This approach sets the path for effective loss calculation. A detailed monthly paydown curve is created to study the life of loan for an account.

Components of Loss Forecasting Models

CECL is calculated using a multiplicative framework. This requires the estimation of the following parameters describing the risk exposure.

i) Probability of Default (PD) – gives the average percentage of accounts, that experience a default event.

The performance duration is defined as the status of the account until one of the following “terminal events” happens after any given observation month.
1. Charged-off – Where the default is 180+ days past due
2. Charged-off due to bankruptcy
3. The balance is paid in full – the account has prepaid the outstanding balance at the time of evaluation

The terminal events are modeled using a hazard competing risk framework. We advocate using calculation of polychotomous logistic regression parameters using individualized regressions.

ii) Loss given default (LGD) – the percentage of monetary exposure if the borrower defaults

For an unsecured portfolio, the LGD can be safely taken as 100% for all the accounts, or other values based upon historical loss experience.

iii) Exposure at Default (EAD)- the outstanding amount (drawn amounts) in case the borrower defaults.

EAD can be defined based on balance paydown curves using FIFO payment hierarchy.

To obtain the final Expected Credit Loss (ECL) for an account, the following equation is used:



On balance, we have observed that the implementation of CECL standards for credit card portfolios is very complex compared to the previously incurred loss methodology. At the same time, the added value from the extra complexity outweighs the drawbacks. This new regulation prepares banks for any adverse financial situations which could pop-up in the future. This is best observed in the early adopters of CECL, who appear to be better positioned to brace the headwinds brought on by the global slowdown due to the pandemic.

This is the first part of a two-part series on CECL. In the next part of the blog [CECL Modelling & Implementation – Practical approaches], we will discuss how we assisted a midsize bank in the US to comply with CECL regulatory requirements.

To know more about how we can support your CECL journey and to schedule a 1:1 discussion with our specialists, send an email to info@tigeranalytics.com


Copyright © 2023 Tiger Analytics | All Rights Reserved