Delinquency, Default and Roll Rates

Table of Contents

    Delinquency, default and roll rates describe different parts of a loan’s performance. Delinquency measures unpaid obligations; default follows a specified contractual, program or analytical definition; roll rates follow loans as their payment status changes. Reading them together helps an analyst distinguish a temporary payment problem from a growing pipeline of credit losses.

    A falling delinquency percentage can reflect genuine repayment, new lending that enlarges the denominator, or troubled loans leaving the books through charge-off. The useful question is what happened to the same loans. This guide explains the definitions, builds a monthly transition table and connects one deterioration path to expected loss. Numerical examples are hypothetical; U.S. rules and source definitions are described as read on November 14, 2026.

    Delinquency vs Default vs Charge-Off

    A Status, an Event and a Loss Are Different

    A past-due balance is not automatically a default, a default does not imply a complete loss, and a charge-off does not itself forgive the borrower’s debt. Identify the definition and measurement period before comparing percentages.

    Days past due (DPD) measures how long a required payment has remained unpaid under the applicable counting rules. An aging report groups loans into buckets such as 30–59, 60–89 and 90+ DPD. A 30+ measure includes the later buckets too; it is not the same as the exclusive 30–59 bucket.

    Default is the event specified by the relevant agreement, regulation or dataset. A research model may choose a particular DPD threshold, while another source uses additional conditions. Consequently, a reported 90+ delinquency rate cannot automatically be relabeled a default rate. The separate concepts of probability of default and loss given default distinguish the chance of the defined event from its financial severity.

    Nonaccrual identifies an accounting treatment of interest recognition. It is not simply another name for a DPD bucket. Charge-off removes an uncollectible amount from the lender’s recorded loan asset, generally against the allowance for credit losses. Recoveries can occur afterward. The Federal Reserve’s release distinguishes charge-offs from delinquent/nonaccrual loans; the FTC explains that a charge-off alone does not cancel the debt. [1] [7]

    The threshold depends on the product and source

    Selected U.S. definitions serve different purposes
    Product or source Relevant definition Interpretation limit
    Mortgages covered by Regulation X’s servicing subpart Delinquency begins when a periodic payment covering principal, interest and applicable escrow is due and unpaid; it continues until no periodic payment is due and unpaid. This definition is for that subpart. Its mortgage-loan definition excludes open-end home-equity plans; a statistical series may use a different cutoff.
    Bank retail credit under the interagency classification policy General charge-off thresholds are 120 cumulative DPD for closed-end retail loans and 180 for open-end retail loans. Known losses may require earlier recognition. Collateral, bankruptcy, fraud and documented collection circumstances affect treatment.
    Residential real estate under that retail policy At no later than 180 DPD, assess collateral value; charge off the loan balance exceeding property value less selling costs. The policy does not require treating every delinquent mortgage as a total unsecured loss.
    Federal Direct Loans, 34 CFR 685.102(b) The default definition includes failure to make due installments or meet other note terms, persisting for 270 days, with the Secretary’s specified finding about intent to repay. This is the Direct Loan program definition, not a universal student-loan or bank charge-off rule.

    These are scoped comparisons, not interchangeable clocks. The retail policy also provides qualified treatment for well-secured loans in the process of collection. Read the rule and the lender’s disclosures for the actual portfolio rather than applying a single threshold to every product. [3] [4] [6]

    How to Read a Published Delinquency Rate

    Start with the series notes. In the Federal Reserve’s commercial-bank release, delinquent loans include those 30 or more days past due and still accruing interest, plus loans in nonaccrual status. The delinquency ratio uses loan dollars, so a large troubled loan carries more weight than a small one. It does not report the percentage of borrowers who became delinquent during the quarter. [1] [2]

    Delinquency Stock Ratio
    Delinquency rate = Delinquent loan dollars at the reporting date / Total loan dollars outstanding at that date
    Multiply the ratio by 100 to express it as a percentage; use the source’s delinquency definition and loan universe.

    Suppose a portfolio has $3 million of delinquent loans and $100 million outstanding: $3 million / $100 million = 3%. Add $50 million of current loans while the delinquent dollars remain unchanged, and the ratio becomes $3 million / $150 million = 2%. The lower percentage alone does not demonstrate improvement among the original loans.

    Charge-off rates measure a flow

    The Fed’s charge-off measure instead takes quarterly net charge-offs, meaning gross charge-offs minus recoveries, and divides by average loans during that quarter. Its published rate is annualized. Delinquency is a reporting-date stock; charge-off is a period flow. [2]

    Hypothetical Quarterly Net Charge-Off Rate

    Suppose gross charge-offs are $600,000, recoveries are $100,000 and average quarterly loans are $100 million.

    Net charge-offs = $600,000 − $100,000 = $500,000.

    Quarterly ratio = $500,000 / $100 million = 0.005, or 0.5%.

    Annualized reported percentage = 0.005 × 400 = 2.0%.

    The annualization is simple scaling of that quarter’s pace, not compounding and not a forecast of next year’s realized loss. Before comparing two published rates, check the product, reporting population, dollars versus account counts, gross versus net losses, average versus ending balance, observation date, annualization and seasonal adjustment. Keep seasonally adjusted and unadjusted series separate.

    Roll Rates Follow the Same Loans

    A roll rate measures movement from one payment status to another over a defined interval. Rows identify the starting state; columns identify the ending state. The OCC describes roll rates in terms of either accounts or balances, which makes the weighting convention part of the metric’s definition. [5]

    One-Month Account Transition Rate
    pi,j = Accounts starting in state i and ending in state j / Accounts starting in state i
    Keep the original starting cohort in the denominator, including its paid and charged-off exits, unless an explicitly different conditional metric is intended.

    Consider a hypothetical fixed file of 1,200 loans observed at two month ends. “Under 30” means 0–29 DPD, not necessarily fully current. Every starting loan receives one ending classification. Full repayment and whole-account gross charge-off are separate terminal outcomes for this ledger; there are no sales, transfers or missing records in this example. New originations are excluded.

    Hypothetical one-month transitions: account counts
    Starting bucket (accounts) Under 30 DPD 30–59 DPD 60–89 DPD 90+ DPD Paid in full Charge-off
    Under 30 (1,000) 940 35 0 0 25 0
    30–59 (100) 20 25 50 0 5 0
    60–89 (60) 6 6 12 33 3 0
    90+ (40) 2 2 4 16 2 14

    Read the 30–59 row across. Of its original 100 accounts, 50 / 100 = 50% roll forward to 60–89; 25% remain; 20% move below 30 DPD; and 5% repay in full. These outcomes sum to 50% + 25% + 20% + 5% = 100%. Every other row also reconciles to its starting count.

    The 60–89 to 90+ rate is 33 / 60 = 55%. The 90+ to charge-off rate is 14 / 40 = 35%. These percentages describe different starting groups. They are not the proportion of the entire 1,200-loan portfolio reaching each outcome.

    Backward rolls, cures and exits

    A backward roll moves to a less delinquent bucket. A cure commonly means returning to current status, but reports must specify the actual cutoff. In the example, suppose 15 of the 20 accounts moving below 30 DPD have no overdue payments, while five remain 1–29 DPD. The zero-arrears cure rate is therefore 15 / 100 = 15%; the broader below-30 transition is 20%.

    The five paid-in-full accounts are exits, separate from the live accounts that cure. If they are removed from the denominator, the forward rate becomes 50 / 95 = 52.63%. That is a rate conditional on not repaying in full, rather than the original-cohort rate. Either can answer a defined question, but they cannot be compared as if they were the same measure.

    Also distinguish a payment-driven cure from a change caused by modification, deferral or re-aging. Reported status may improve even though some original arrears were not paid in cash. The retail policy calls for tracking these actions and their subsequent performance; a lower DPD label alone does not establish lasting recovery. [4]

    Account-weighted vs balance-weighted rolls

    For a balance-weighted transition, one consistent approach allocates each loan’s starting balance to its ending outcome, then divides by all starting balances in that row. Those shares can reconcile to 100%. By contrast, dividing closing balances in a destination bucket by opening balances in a starting bucket can be distorted by payments, new draws, interest and write-downs. A balance-flow report needs those movements reconciled separately.

    Two aggregate aging tables alone cannot reveal true transitions: the same ending totals can arise from different combinations of new delinquency, cures and exits. Match loan identifiers. In a real file, identify sales and transfers separately, resolve missing statuses and account for partial charge-offs rather than silently dropping observations.

    From Roll Rates to a Loss Contribution

    Return to the 100 loans starting in the 30–59 bucket. Follow just one path over the next three monthly transitions: 30–59 to 60–89, then to 90+, then to charge-off. For this illustration, assume the table’s row rates remain applicable in later months to this tracked subset, regardless of its prior history. This is a modeling assumption, not something established by one month’s observations.

    Also assume each loan has a fixed $10,000 exposure if it reaches the charge-off boundary, making the cohort’s reference exposure $1 million. Charge-off is the terminal event used for this calculation; it does not replace the loan product’s contractual definition of default.

    Hypothetical Consecutive-Deterioration Path

    Path probability = 50% × 55% × 35% = 9.625%.

    Expected gross exposure charged off through this path = $1 million × 9.625% = $96,250.

    Suppose eventual recoveries, after collection costs, equal 40% of that exposure. Ignoring discounting, loss given the event is 1 − 40% = 60%.

    Expected ultimate net loss from this path = $96,250 × 60% = $57,750.

    The multiplication uses conditional probabilities: the second rate applies to loans that reached the second bucket, and the third to those that reached the third. It does not require independent events; using pooled row rates for these particular arrivals requires the stated assumptions about history and stability.

    The $57,750 is this path’s loss contribution, equivalent to $57,750 / $1 million = 5.775% of the reference exposure. It includes assumed eventual recoveries on charge-offs reached by the third month. It is not the cohort’s total three-month loss, lifetime default probability or complete loss allowance: other paths, including staying delinquent, curing and later relapsing, can generate additional losses.

    Keep Frequency and Severity Separate

    The transition rates here lead to a gross charge-off event, so a separate recovery/LGD assumption is necessary. If an input already represents net loss per dollar of starting exposure, multiplying it by LGD again would count severity twice. Annualized net charge-off rates also cannot be inserted directly as monthly transition probabilities.

    Vintage Curves Compare Loans at the Same Age

    A vintage groups loans originated in the same period. A vintage curve then tracks a chosen performance measure by months on book. This helps distinguish the performance of a growing portfolio’s many new loans from that of older loans that have had more time to develop problems. The OCC specifically describes comparing origination cohorts at similar time on book. [5]

    Consider two fictional closed-end loan vintages, 2024 Q1 and 2025 Q1, each with $10 million of original principal and no later additions. Measure cumulative gross charge-offs divided by that fixed original principal. Repayment and prepayment do not shrink this denominator. These figures are an illustration, not historical portfolio data.

    Hypothetical cumulative gross charge-offs (% of original principal)
    Months on book 2024 Q1 vintage 2025 Q1 vintage
    3 0.2% 0.3%
    6 0.8% 1.1%
    9 1.6% 2.2%
    12 2.4% 3.2%

    At 12 months, the newer vintage has accumulated 3.2% − 2.4% = 0.8 percentage point more gross charge-offs. On the equal $10 million original balances, that is $320,000 − $240,000 = $80,000. Comparing the newer vintage at three months with the older one at twelve months would confound performance with age.

    The equal-age comparison still does not prove that underwriting deteriorated. The cohorts experienced different calendar conditions and may differ in borrower quality, term, collateral, pricing or collection treatment. A cumulative net-loss curve also differs from this gross-charge-off curve because recoveries affect its numerator. For revolving receivables, new draws and account additions require particular care; credit card ABS analysis adds pool and transaction considerations.

    Roll Rates vs Rating Migration and CECL

    Payment-State Transitions

    • States describe payment aging and defined exit events.
    • Often measured monthly for the same accounts.
    • Backward movement and cures can occur.
    • Useful for tracing how losses develop.

    Issuer-Rating Transitions

    • States describe credit ratings and the study’s default category.
    • The horizon and issuer population depend on the study.
    • An absorbing default state may be a model convention.
    • Useful for rating-change and portfolio-risk analysis.

    The matrix format is shared, but the states and meanings differ. A loan’s return from 60 DPD to current is not an issuer-rating upgrade. See credit-risk migration matrices for the separate rating-state framework; do not transfer an absorbing-default assumption to an ordinary delinquency bucket.

    Current expected credit losses (CECL) concerns the allowance measurement for applicable assets. The revised interagency policy permits methods including roll rates, PD/LGD and aging schedules. It requires estimating expected credit losses over the relevant contractual term while considering expected prepayments, historical experience, current conditions and reasonable and supportable forecasts, with reversion where required. A snapshot or single path does not supply that whole estimate. [8]

    Prepayment is an exit that removes future exposure; it is not a credit loss. Cures can leave exposure outstanding and can be followed by renewed delinquency. A complete model must handle both, rather than treating every nondefault outcome as the same permanent disappearance. Our guide to bad-debt allowances and CECL explains the accounting context separately from this loan-transition analysis.

    How to Analyze Loan Performance

    1. Write the metric definition first. Record the product, DPD/default boundary, reporting population, observation interval, account or dollar weighting, denominator and gross/net basis.
    2. Reconcile the same starting loans. Include staying, forward and backward rolls, full repayment, charge-off and other exits. Investigate missing records before estimating rates.
    3. Read stock and flow together. Check whether lower delinquency reflects cash repayment, charge-offs, sales, growth or changed reporting treatment.
    4. Compare like with like. Match months on book and relevant borrower/product segments, then consider the calendar environment and servicing changes.
    5. Test the forecast assumptions. Evaluate changing transitions, cure durability, prepayment, exposure and recovery severity over the intended horizon.

    Transitions show performance after lending; bank credit risk management addresses the related screening, monitoring and covenant decisions. Readers dealing with their own overdue bills need the contract and servicer’s explanation of their account, rather than a portfolio average; our consumer credit and debt-management guide covers that household perspective.

    Common Mistakes

    • Treating all overdue accounts as defaults. A reporting cutoff, contractual default and accounting charge-off can occur at different points.
    • Reading 30+ as 30–59. The inclusive measure contains later delinquency buckets; adding those buckets again double-counts loans.
    • Using the ending portfolio as a roll-rate denominator. A transition begins with a defined starting group, while a stock ratio uses the reporting-date population.
    • Dropping paid or sold loans without explanation. This changes the population and can make the surviving loans look worse or better for purely mechanical reasons.
    • Calling every backward roll a full cure. Less delinquent is not necessarily zero overdue, and re-aging is not the same as cash collection.
    • Multiplying incompatible rates. Match periods, conditional populations, weighting and event definitions before chaining probabilities.
    • Calling charge-off frequency an expected-loss percentage. Exposure and recoveries still matter unless severity is already embedded in the rate.

    Limitations of Roll-Rate Analysis

    A transition table summarizes a particular population and interval. Rates can change with unemployment, lending standards, loan age, payment scheduling, relief programs and collections. A single month can contain too few observations for reliable estimates, especially in late-stage buckets.

    Repeatedly applying the same matrix assumes stability and that the chosen state captures the history relevant to the next transition. Two loans with identical DPD can differ because one recently cured, one has repeatedly missed payments, or one received a modification. Segmenting or retaining additional history may improve the model, but it also leaves fewer observations in each group.

    Month-end observations can miss within-month delinquency and recovery. Sales, servicing transfers, partial charge-offs, reporting changes and unobserved exits complicate comparisons. Use a roll table to expose assumptions and reconcile outcomes; it cannot by itself establish a borrower’s legal status, the cause of portfolio deterioration or the adequacy of an accounting allowance.

    Frequently Asked Questions

    Delinquency describes an unpaid required payment under the applicable counting rules. Default is a separately defined event under the agreement, program, regulation or dataset. There is no single DPD threshold that applies to every loan and every reported series.

    Identify the accounts in a starting bucket, follow those same accounts to the next observation and divide the number reaching a specified ending state by the starting count. In the example, 50 of the 100 accounts starting 30–59 DPD end 60–89 DPD, giving a 50% forward roll. State explicitly if balances or a survivor-only denominator are used instead.

    Yes. New current lending can dilute the delinquency ratio, and charge-offs can remove troubled balances from the reported loan stock while increasing the period’s loss flow. Examine the delinquent dollars, charge-offs, recoveries and movements of the original cohort together.

    No. Charge-off is the lender’s accounting recognition of an uncollectible amount; it does not itself release the borrower’s obligation. Later collection or sale can occur. Any actual discharge, settlement or other change to the obligation is a separate matter.

    Older loans have had more time to develop delinquency and charge-offs. Equal-age comparisons reduce that timing distortion, although borrower mix, contractual terms, calendar conditions and servicing differences can still explain performance gaps. Always name the numerator and denominator of the vintage measure.

    They can be part of an appropriate supported estimation method. A complete estimate must address the relevant contractual term, expected prepayments, current conditions, reasonable and supportable forecasts and other applicable requirements. The article’s three-month single-path calculation is an educational component, not a complete CECL allowance.

    Sources

    1. Federal Reserve: Charge-Off and Delinquency Rates on Loans and Leases at Commercial Banks. Release definitions; displayed update August 25, 2026.
    2. Federal Reserve: About this Release. Stock/flow calculation and annualization; historical methodology page, updated May 2, 2011.
    3. CFPB: Regulation X, §1024.31 Definitions. Current displayed mortgage-servicing delinquency definition and scope.
    4. Federal Reserve: Uniform Retail-Credit Classification and Account-Management Policy. Product-specific classification, charge-off and re-aging guidance.
    5. OCC: Comptroller’s Handbook, Retail Lending. Version 2.0, October 2021, with March 2025 cover update; portfolio monitoring and roll-rate/vintage definitions.
    6. eCFR: 34 CFR 685.102(b), Direct Loan Program Definitions. Default definition; displayed Title 34 currency through November 12, 2026.
    7. FTC: How To Get Out of Debt. Charge-off and continuing debt obligation.
    8. Federal Reserve: Interagency Policy Statement on Allowances for Credit Losses. Revised April 2023; estimation methods, contractual term, prepayments and forward-looking information.

    Sources accessed November 14, 2026. All portfolio amounts, transition counts, probabilities and vintage results in the worked examples are hypothetical.

    Disclaimer

    This article is for educational and informational purposes only and does not constitute investment, accounting or legal advice. Loan terms, servicing rules, default definitions and reporting conventions vary by product and jurisdiction. Consult the applicable documents and qualified professionals for a specific loan, portfolio or allowance estimate.