Roll rates and cures: reading the movement behind delinquency

A stock measure cannot explain the flow

A delinquency rate describes a portfolio at a point in time. A roll rate describes how a defined starting population moves between states over a specified interval. The distinction matters because the same ending delinquency rate can arise from very different combinations of new arrears, cures, charge-offs, sales and portfolio growth. Decisions about collections or underwriting need those movements, not just the ending percentage.

The OCC’s Retail Lending handbook discusses portfolio monitoring and includes roll-rate reporting among the materials used to evaluate collections. [1] This is an analytical technique within a broader risk-management framework, not a regulatory formula that produces one universally correct forecast. The state definitions, observation interval and population exclusions are choices that must be documented.

Define the denominator before calculating the rate

For an account-count transition, divide the number of accounts moving from state i to state j by the number in state i at the start. States might include current, 1–29 days past due, 30–59, 60–89, later delinquency, closed and charged off. The rows should account for every member of the starting cohort, with separate treatment of accounts that disappear from the observation system.

A balance-weighted transition answers a different question. It can weight each account by its starting balance to measure migration of beginning exposure, or use ending balances to describe where dollars finish. Those methods are not interchangeable when customers pay down, borrow more or receive fee adjustments. Label the weighting explicitly and reconcile the residual change rather than forcing every movement into a credit-quality story.

Newly originated loans should not enter the denominator of a fixed starting cohort. They belong in a separate flow. Otherwise, rapid growth can dilute the observed problem and make performance look better without changing the behavior of existing borrowers. Similarly, removing charged-off accounts from the file without an absorbing charge-off state can make the worst outcomes vanish.

A hypothetical transition row

Assume 1,000 accounts begin a month 30–59 days past due. At the next observation, 300 are current, 100 are 1–29 days past due, 250 remain 30–59, 300 have moved to 60–89 and 50 have closed through verified payoff. The full-current cure rate is 30%; the backward-migration rate including partial improvement is 40%. Calling both measures cure without qualification obscures the difference.

The table shows a single hypothetical row, not a benchmark for any institution. Its percentages sum to 100% because every starting account is assigned one ending state. If an account cannot be located, create a data-exception state and investigate it. Do not assume that missing means paid or current.

Scroll horizontally to see all columns.

A cure should demonstrate more than a coding change

An account can become current through ordinary payments, a contractual modification or an administrative correction. Those paths have different predictive meaning. Track the cause of cure and subsequent redefault. A one-month cure followed by immediate delinquency may represent a temporary payment rather than durable recovery.

The interagency Uniform Retail Credit Classification and Account Management Policy addresses classification and account-management practices, including re-aging and extensions. [2] Its existence is a reminder to distinguish contractual relief from practices that conceal deterioration. This article does not replace the policy’s product-specific requirements or prescribe a universal charge-off date.

For analysis, retain both contractual delinquency under the current agreement and a history of assistance or modification events. That supports a fair comparison without treating borrowers receiving legitimate help as if their prior payment history never existed. A workout program can be beneficial even if assisted accounts perform differently; the right test considers sustainable repayment and the alternative outcome.

Forecasting requires more than matrix multiplication

A simple transition model applies a matrix repeatedly to estimate future state distributions. That assumes the transition behavior remains applicable across time and cohorts. In consumer credit, seasoning, seasonality, underwriting changes and household conditions can violate that assumption. A matrix calibrated during expansion may understate correlated deterioration during stress.

Build separate views by origination vintage, product, payment frequency and relevant risk segment where sample size permits. Avoid so many cells that random noise becomes a policy signal. Compare realized transitions with forecast intervals and investigate changes in data or servicing policy before attributing every deviation to borrower risk.

The Federal Reserve’s April 17, 2026 model-risk guidance provides the current supervisory reference for managing models and supersedes SR 11-7. [3] A roll-rate spreadsheet used only for descriptive reporting may carry different risk from an automated forecast driving material reserves or credit decisions. Governance should follow intended use and consequences, not whether the tool has an AI label.

Use the result to choose an intervention

Suppose early-stage forward migration rises while later-stage collections remain stable. The useful response may involve payment reminders, due-date alignment or investigation of a servicing defect. If late-stage cures deteriorate across mature vintages, a broader affordability or recovery problem may be present. These are diagnostic hypotheses, not causal conclusions established by the matrix alone.

Evaluate collection strategies with comparable starting populations. A team assigned easier accounts can produce a better cure rate without superior execution. A strategy that accelerates payments this month may increase redefault next month. Track durable cures, dollars recovered, customer complaints and treatment costs alongside immediate transition improvement.

What would change the interpretation

Reconcile the transition totals to the servicing ledger, confirm that aging definitions stayed constant and inspect missing-account exceptions. Evidence of changed due-date logic, portfolio sales or re-aging could explain an apparent improvement without better underlying credit. Conversely, stable definitions and repeated deterioration across comparable cohorts would strengthen the case for a real shift.

The value of roll rates is visibility into movement. Their limitation is that movement can reflect both borrower economics and institutional choices. A useful dashboard shows the two together and gives decision-makers enough history to distinguish a durable cure from a temporary or administrative change.

Sources

1. OCC, Retail Lending handbook, October 2021; current posted booklet reviewed September 27, 2026Official source · PDFBack to text: ↑1↑2
2. OCC Bulletin 2000-20, Uniform Retail Credit Classification and Account Management Policy, June 20, 2000Official sourceBack to text: ↑
3. Federal Reserve, SR 26-2, Supervisory Guidance on Model Risk Management, April 17, 2026Official sourceBack to text: ↑