A delinquency flag can disappear before the borrower has genuinely become lower risk. Cure is an event; recovery is a claim about what survives afterward.
Technical cure satisfies the approved operational condition for leaving delinquency or default. Behavioural cure adds improved subsequent payment behaviour. Sustainable cure requires the improvement to persist through a defined, decision-relevant observation horizon—without prescribing one universal horizon.
The Entimema Cure & Re-Default Architecture
- Delinquency
- Cure event
- Technical cure
- Post-cure behaviour
- Persistence / observation
- Re-default risk
- Sustainable cure
- Economic recovery
- LGD / collections outcome
- Monitoring
- Identify cure
- Assess prior severity
- Observe behaviour
- Estimate re-default risk
- Test persistence
- Classify sustainable cure
- Adjust monitoring / collections
- Feed LGD / strategy learning
Re-default means default after cure within a defined relevant horizon. Its event definition must align with the wider default architecture; operational cure, default cure, accounting staging and economic recovery are related but not interchangeable.
Cures have quality, not just count
| Borrower | Cure path | Post-cure evidence | Classification |
|---|---|---|---|
| A | Clears arrears | Current for 12 months; behaviour normalises | Strong sustainable-cure evidence |
| B | Clears arrears | Returns to delinquency after two months | Technical cure; re-default |
| C | Partially normalises | Current status not reached; behavioural risk remains high | Improvement, not binary cure |
Both numerator and denominator require governance. Report by starting state—early, mid-stage, severe delinquency or default/workout—because P(state → Current) differs materially. A cure from 5 DPD is not equivalent to one after 90 DPD.
Re-default turns cure into a survival problem
The re-default hazard hᵣ(t) asks for the probability of re-default at time t conditional on remaining cured until then. It can rise or fall with months since cure; no universal shape should be imposed. Time to re-default is T re-default − T cure.
| Month since cure | At risk at start | Re-defaults in interval | Surviving cured | Cumulative re-default |
|---|---|---|---|---|
| 1 | 1,000 | 90 | 910 | 9.0% |
| 3 | 910 | 80 | 830 | 17.0% |
| 6 | 830 | 70 | 760 | 24.0% |
| 12 | 760 | 50 | 710 | 29.0% |
Immediate failures suggest fragile technical cure; later failures may reflect renewed stress. Surveillance is most valuable where the post-cure hazard remains elevated and evidence can still affect a governed workflow.
Cure vintage separates durability through time
A borrower simultaneously belongs to an origination vintage, a delinquency-entry vintage and a cure vintage. Origination vintage describes booking conditions; delinquency vintage describes entry into stress; cure vintage describes the regime under which recovery appeared.
Compare cohorts at equal months since cure. A cure from last week cannot be classified as a sustainable 12-month cure. Recent cohorts are right-censored: they have not yet had enough time to demonstrate stability.
Cure quality can differ by macro environment, collections strategy, product and origination cohort. Record strategy version alongside cure date so changing composition is not mistaken for improving treatment.
Post-cure risk depends on the path, not only the current state
Maintain Episode₁, Episode₂, … rather than collapsing repeat delinquency into one event. Useful history includes N prior cures, maximum prior severity, delinquency duration, time to cure, treatment, payment path and whether cure was lump-sum, gradual or restructuring-driven.
A simple Markov transition P(Stateₜ₊₁ | Stateₜ) can miss duration and repeat episodes. A semi-Markov or richer multi-state view can incorporate time in state and prior path without requiring excessive complexity.
Partial cure also matters: Severe → Moderate → Early → Current is a different trajectory from an instantaneous jump to Current. Neither path is automatically superior; subsequent sustainability supplies the evidence.
Current status can conceal a post-cure risk premium
Behavioural Credit Scoring can reveal that a technically cured borrower still has high utilisation, weak payment ratios or elevated PD. Whether PD post-cure exceeds PD never-delinquent is an empirical question, not an assumption.
| Population | 12-month default | Interpretation |
|---|---|---|
| Never delinquent | 2% | Performing reference |
| Recently cured | 8% | Residual risk after technical cure |
| Repeatedly cured | 15% | History reveals structural instability |
Compare cured and continuously performing accounts on subsequent delinquency, default, utilisation and loss. A post-cure Early Warning window can track payment deterioration, liquidity pressure and repeated lateness without prescribing a universal monitoring duration.
Evidence to exit distress should differ from evidence to enter it
Cure hysteresis avoids oscillation when an account alternates around one status threshold. Operational delinquency cure does not automatically reverse SICR or Stage 2, nor does it necessarily establish Stage 3 exit or removal of a credit-impaired assessment. Each serves a different purpose and must follow its own governed definition.
Treatment should be judged on sustainable cure, not immediate status
Natural cure and treatment-assisted cure are counterfactual objects. Historical comparison is selected because stronger treatments often went to harder cases. If Cure Rate A is lower than Cure Rate B, that does not prove treatment A is worse.
| Treatment | 30-day cure | Re-default by month 6 | Sustainable cure among treated | Interpretation |
|---|---|---|---|---|
| A | 68% | 40% of cures | 40.8% | High immediate cure, fragile outcome |
| B | 57% | 16% of cures | 47.9% | Lower immediate cure, stronger durability |
Treatment B produces fewer immediate cures yet more sustainable cures. Rewarding teams only on Cure Rate 30d can optimise a temporary status rather than durable recovery.
Restructuring-driven cure should be analysed separately where material: reduced payments, extended tenor or other modifications change contractual cash flows and may have distinct P(Re-Default | Restructured Cure).
Cure durability changes economic loss
A rapid sustainable cure can restore cash flows and reduce loss. Cure followed by re-default can lengthen workout, add operational cost and delay recoveries. Status cure means the account becomes current; economic cure means expected contractual cash flows are restored sufficiently for the loss objective.
IFRS 9 LGD makes timing explicit through discounted recovery cash flows. Collections Prioritisation uses natural cure and treatment uplift to decide where scarce action can still improve that economic outcome.
Targets, models and episode data must describe the same process
Possible targets include P(Cure within h), P(Cure and no re-default through h), and P(Re-Default within h | Cure). Transition rates, logistic models, survival models, multi-state models and governed tree-based challengers answer different questions.
Build an episode-level ledger with account ID, episode ID, event date, state, payment, treatment, cure flag and re-default flag. Current status alone cannot reconstruct prior cures, duration or sequence.
A 54% headline cure rate becomes 40.5% durable recovery
A fictional non-bank consumer lender observes 20,000 accounts entering delinquency during one quarter. After six months, 10,800 have technically cured, 6,200 remain delinquent and 3,000 have defaulted. Among the cured group, 2,700 re-default and 8,100 remain current.
| Outcome | Accounts | Share of original 20,000 |
|---|---|---|
| Technical cure | 10,800 | 54.0% |
| Still delinquent | 6,200 | 31.0% |
| Defaulted without cure | 3,000 | 15.0% |
| Of technical cures: remain current | 8,100 | 40.5% |
| Of technical cures: re-default | 2,700 | 13.5% |
The headline cure rate is 54.0%; six-month sustainable cure is only 40.5%. Re-default consumes one quarter of technical cures.
| Starting state | Technical cure | Six-month sustainable cure |
|---|---|---|
| 1–15 DPD | 72% | 59% |
| 16–30 DPD | 58% | 44% |
| 31–60 DPD | 39% | 25% |
| 61–90 DPD | 21% | 11% |
| Observed treatment | Technical cure | Six-month re-default among cures |
|---|---|---|
| Digital outreach | 61% | 19% |
| Targeted call | 56% | 24% |
| Manual review / modification assessment | 43% | 17% |
The treatment table cannot identify effectiveness because account mix differs. The cure-vintage view, equal maturity and approved causal testing are needed before attributing the differences.
Non-bank portfolios compress the cure-learning cycle
Short tenors, frequent payment events, higher default incidence and repeat borrowing can produce multiple delinquency and cure cycles quickly. A single Current flag discards especially valuable history in these portfolios.
Post-cure risk may decay—or recur—within weeks rather than quarters. Monitoring cadence, maturity windows and workflow speed should match product velocity, while avoiding premature sustainable-cure classification.
Common failure modes
| Failure | Why it fails |
|---|---|
| Returning to current equals recovery | A state transition says nothing about whether improvement persists. |
| Technical cure equals sustainable cure | Operational exit criteria omit subsequent risk. |
| No re-default tracking | Fragile cures inflate performance. |
| One cure rate across states | Starting severity materially changes opportunity. |
| Delinquency duration ignored | Time under stress and cure speed disappear. |
| Prior cure count ignored | Repeated cycles can signal structural instability. |
| Path dependence ignored | Current status cannot describe how the borrower arrived there. |
| No cure vintage | Macro and strategy regimes are mixed. |
| Recent cures treated as mature | Censored cohorts have not had time to fail. |
| Immediate cure optimised alone | Short-term status can displace durable recovery. |
| Promise treated as cure | Commitment is not realised payment or stability. |
| Restructured cures pooled blindly | Modified cash flows and risk can differ. |
| Treatment rates compared causally | Selected populations confound borrower quality and treatment effect. |
| One model across products | Revolving, instalment and secured cure processes differ. |
| Current state without history | Episodes, durations and repeated cures cannot be reconstructed. |
| No episode ledger | Targets and outcomes cannot be reproduced. |
| No post-cure monitoring | Residual behavioural risk is ignored. |
| Cure disconnected from LGD | Timing, repeat default and cost are lost from economic loss. |
| Cure disconnected from priority | Natural cure and treatment uplift cannot inform scarce attention. |
| Headline KPI without durability | Top-line cure rewards temporary status. |
| No macro or strategy attribution | Changing environment and treatment policy are mistaken for model performance. |
A Cure & Re-Default Analytics Agent can monitor durability—not change treatment autonomously
A future Agent can reconstruct episodes, identify technical cures, track post-cure behaviour, calculate cure vintages, estimate sustainable cure and re-default risk, measure time to re-default, identify fragile or repeated cures, compare quality by strategy, feed risk into Early Warning and quantify LGD implications for human review.
Its role is cure durability analytics + re-default surveillance + post-cure monitoring + LGD support. It must not autonomously change customer treatment.
Credit Risk
Credit Risk for cure modelling, re-default analytics, collections strategy, LGD and portfolio monitoring.
Decision Automation
Decision Automation for post-cure surveillance, cure-quality classification, governed routing and recurring evidence.
Related research
Continue with Collections Prioritisation, Early Warning Systems, Behavioural Credit Scoring, Roll Rate Analysis, Credit Vintage Analysis, IFRS 9 LGD and Significant Increase in Credit Risk.
