Pooling and estimation methods under CECL

• Segment a portfolio into pools with similar risk characteristics
• Decide when an asset is evaluated individually rather than in a pool
• Compare the estimation methods permitted by 326-20-30-3
• Build qualitative adjustments that link to observable data

Pool first

Under 326-20-30-2 an entity measures expected credit losses on a collective (pool) basis when similar risk characteristics exist. An asset that does not share risk characteristics with others is evaluated individually, and an asset evaluated individually is not also included in a pool. No double counting.

Risk characteristics that support pooling include internal or external credit score, risk rating, financial asset type, collateral type, size, term, geography, industry of the borrower, and vintage. A practical small-entity segmentation for trade receivables might be: government customers, large commercial customers, small commercial customers, and international customers.

Pools are reassessed. Under 326-20-35-2 an entity evaluates whether an asset in a pool still shares risk characteristics with the pool. A customer that files for bankruptcy is moved out and evaluated individually.

Methods

The Codification does not prescribe a method. Under 326-20-30-3 an entity may use discounted cash flow, loss-rate, roll-rate, probability-of-default or aging schedule methods. It is not required to use discounted cash flows or reconcile its method to one.

• Loss-rate. Historical lifetime losses divided by the balance of a closed pool, applied to the current balance. Simple and common for receivables and small loan books. The historical periods must reflect the contractual life of the assets.
• Aging schedule. Loss rates by delinquency bucket, built from history. Natural for trade receivables (Lesson 12).
• Roll-rate. Migration rates between delinquency buckets, compounded to estimate how much rolls to charge-off.
• Probability of default. PD x loss given default x exposure at default. Needs default and recovery data, so banks and finance companies use it.
• Discounted cash flow. Expected cash flows discounted at the effective interest rate; the allowance is amortized cost less present value (326-20-30-4). Used for individually evaluated loans.

Method choice should match data. A company with five years of write-off history and stable customers does not need a PD model. A lender with thin history needs peer data, which the standard allows as external information (326-20-30-8).

Qualitative adjustments

History rarely matches current conditions. Under 326-20-30-9 adjustments may be qualitative and should reflect changes in relevant data such as unemployment, property values, commodity prices or delinquency. Good adjustments have three features:

1. They name the factor and its direction (for example, customer industry downturn).
2. They tie to observable data (a sector index, delinquency trend or the client's own credit memos).
3. They are sized by a documented method, not a round number added at the end.

An adjustment that stays at the same number for three years while conditions change is a bias indicator.

Forecast period and reversion

326-20-30-9 does not require a forecast over the full contractual term. The entity forecasts for the period it can support, then reverts to historical loss information that reflects the contractual term. It shall not adjust history for current or forecast conditions beyond the reasonable and supportable period. Reversion may be immediate, straight-line, or another rational and systematic basis, applied at the input level or to the whole estimate. For a 90-day receivable, the forecast period covers the whole life and reversion never comes into play; for a five-year loan, the memo states the forecast length, the reversion method and why.

Data quality

Every method is only as good as the history behind it. Common data problems in small-entity files:

• Write-offs recorded in a single year-end entry, so the loss cannot be traced to the aging bucket or vintage it came from.
• Credit memos for pricing disputes mixed with credit losses, which inflates the loss rate.
• A lookback period that includes no downturn, which understates loss rates for a forecast period that does.
• Recoveries ignored, which overstates net losses.

Fix the data before tuning the method. If write-off detail cannot be rebuilt, document the limitation and consider external data on similar customers, which 326-20-30-8 allows.

Documentation

The CECL memo should name the pools and why, the method per pool, the data source and period, the forecast and reversion approach, each qualitative factor, and the individually evaluated assets. The methodology memo template and the loss-rate calculator support this.

Allowance on equipment notes at Crestline Equipment Finance, LLC