A snapshot misses the movement between states
follow accounts or balances from one state to another over a defined interval. They explain how many become worse, return toward current status, remain delinquent or leave the population. Two portfolios can end with the same delinquency percentage after very different amounts of new distress and successful resolution.
The OCC’s Retail Lending handbook includes roll-rate reporting in its portfolio and collection analysis. This is a useful analytical method, not a universal forecast formula. For readers outside portfolio management, the practical value is seeing whether payment problems are spreading, persisting or being resolved, and what that means for customer support and cash collection. [1]
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 , 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 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.
| Ending state | Accounts | Share of starting cohort |
|---|---|---|
| Current | 300 | 30% |
| 1–29 days past due | 100 | 10% |
| 30–59 days past due | 250 | 25% |
| 60–89 days past due | 300 | 30% |
| Verified payoff / closed | 50 | 5% |
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 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 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 , 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 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 , 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.
A cure needs a defined endpoint
Analysis: an account moving from 60 days late to 30 days late has improved but is not fully current. An account paid off, sold or charged off has left a particular queue through a different route. Report those exits separately and follow cured accounts long enough to see whether improvement lasts.
A temporary accommodation or payment adjustment may help a customer without immediately fitting the same state definition as an ordinary payment cure. Keep the contractual treatment and reporting basis visible. Otherwise a policy change can look like improved borrower cash flow when it mainly changes classification or timing.
Transitions help size the work
Hypothetical service example: 1,000 newly accounts require an average of 15 minutes of initial case work, or 250 hours. If 300 need an additional 20-minute review, that adds 100 hours. The resulting 350 hours is a capacity estimate before quality review and other work, not a required contact strategy or an assurance that every account should receive the same treatment.
Compare resolution per completed case with repeat delinquency, customer effort and the cost of assistance. A faster queue may hide unresolved accounts returned to collections next month. For finance teams, separating these flows also improves forecasts of cash receipts and future loss rather than merely reacting to the latest stock of overdue balances.
What would establish a real improvement
Look for durable cures in comparable groups, fewer worsening transitions and consistent reporting of exits. Relate those results to changes in customer circumstances, assistance and portfolio mix before attributing them to a new process.
are most useful when they connect what customers are experiencing with the resources and financial outcomes of servicing. The transition table is the starting point, not the conclusion.
Sources
- OCC, Retail Lending handbook, October 2021; current posted booklet reviewed September 27, 2026Official source · PDFBack to text: ↑
- OCC Bulletin 2000-20, Uniform Retail Credit Classification and Account Management Policy, June 20, 2000Official sourceBack to text: ↑
- Federal Reserve, SR 26-2, Supervisory Guidance on Model Risk Management, April 17, 2026Official sourceBack to text: ↑