Cohort or Vintage Analysis¶
Method — instantiates Aggregate–Marginal Trajectory Reconciliation
Compares entering groups by common start period, design, supplier, policy, or exposure at equivalent maturity.
Cohort or Vintage Analysis sorts every entering unit into the group it belongs to — defined by when it started or under what design, supplier, policy, or exposure it entered — and then compares those groups only at equal maturity, so a young cohort is never judged against an old cohort's fully played-out outcome. Its defining move is like-for-like aging: it holds the maturity clock fixed so that differences between cohorts reflect the cohorts themselves, not the calendar. Where a raw comparison would blame a fresh vintage for outcomes it has not had time to express, this method waits until each group has aged the same distance and then reads the difference.
Example¶
An auto lender's portfolio 90-plus-day delinquency looks steady — the number is dominated by hundreds of thousands of seasoned loans. The question is whether new originations are deteriorating. The analyst groups loans by origination quarter (the 2024-Q1, 2024-Q3, and 2025-Q1 vintages) and plots delinquency not by calendar date but by months-on-book. At month-on-book 4 — the furthest the youngest vintage has aged — the 2025-Q1 vintage is meaningfully worse than the older vintages were at the same age. Because the newest vintage is right-censored (only four months observed), the analyst compares only at month four rather than against the older vintages' 24-month totals. The like-aged view reveals underwriting drift at the leading edge that the seasoned-dominated portfolio total completely hides.
How it works¶
- Assign membership by entry. Each unit joins the cohort defined by its start period or shared design/supplier/policy/exposure — the profile that everything downstream keys on.
- Freeze the maturity clock. Age is measured in the outcome's own units — months-on-book, operating hours, credits earned — not calendar time, so seasonality and reporting date cannot masquerade as cohort quality.
- Compare at identical age. Read each cohort's outcome at the same maturity, truncating to the youngest cohort's furthest observed age to keep the comparison honest.
- Version the definitions. The cohort boundaries and maturity rule are fixed before the sign is examined, so no one can re-slice the window after seeing the answer.
Tuning parameters¶
- Cohort granularity — monthly, quarterly, or by design/supplier; finer cohorts localize the change but shrink each group and destabilize the estimate.
- Maturity metric — calendar time versus exposure (operating hours, credits, months-on-book); the wrong clock manufactures or hides divergence.
- Censoring treatment — truncate to a shared maximum age, or model the unobserved tail with survival methods; the latter uses more data but adds assumptions.
- Minimum cohort size — the floor below which a vintage is shown as provisional rather than compared.
When it helps, and when it misleads¶
Its strength is that it kills immature-cohort bias and calendar confounds at a stroke; it is the workhorse of the archetype wherever outcomes mature slowly — lending, subscriptions, education, equipment reliability, policy rollouts.
Its failure mode is manufactured trajectories: shifting the cohort window or maturity cutoff after seeing the sign is a form of rebaseline laundering, and small cohorts produce rankings that jump around on noise alone. The classic misuse is comparing a four-month vintage against a two-year vintage's final delinquency and declaring the newcomer either a triumph or a disaster. The guarding discipline is to freeze cohort definitions in advance and to handle unequal follow-up with explicit right-censoring[1] — the concern that survival methods such as the Kaplan–Meier estimator were built to address.
How it implements the components¶
cohort_or_vintage_profile— defines and characterizes each entering group by its shared start, design, supplier, or policy; this profile is the object every comparison joins to.aligned_time_denominator_and_vintage_frame— fixes the maturity clock, censoring rule, and denominator so cohorts are compared at equal age rather than equal calendar date.
It profiles and ages groups but does not smooth them into a continuous leading-edge estimate (marginal_contribution_estimator — that is its nearest twin Rolling Marginal-Contribution Curve), nor project the horizon forward (masking_or_crossover_horizon_estimate — that is Crossover Scenario Projection).
Related¶
- Instantiates: Aggregate–Marginal Trajectory Reconciliation — cohort analysis supplies the maturity-controlled, like-aged view of the entering measure.
- Sibling mechanisms: Aggregate–Marginal Sign-Divergence Alert · Contribution Waterfall Decomposition · Crossover Scenario Projection · Cumulative-versus-Incremental Dashboard · Dual-Metric Decision Memo · First-Difference or Derivative Estimate · Mix-Shift and Base-Effect Audit · Paired Confidence-Band Review · Rolling Marginal-Contribution Curve
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Compares entering groups by common start period, design, supplier, policy, or exposure at equivalent maturity, making its operative form a computation, comparison, model, or analytic representation used to infer, estimate, or choose.
Independent corroboration: The frozen evidence defines Cohort or Vintage Analysis as 'Compares entering groups by common start period, design, supplier, policy, or exposure at equivalent maturity', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Economics & Finance
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Credit, investment, and insurance analysis established comparing vintages at equal maturity so calendar age does not confound cohort quality.
Related originating lineages:
- Statistics & Experimental Design — Longitudinal cohort methods supply like-for-like aging and composition control.
Review resolution: Both reviewers agree on economics_finance as primary. Reading the source mechanism confirms that its defining operation belongs to that lineage; the final record retains statistics_experimental_design only where it materially formed the mechanism and keeps present-day application breadth separate from provenance.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Kaplan, E. L., & Meier, P. "Nonparametric Estimation from Incomplete Observations". Journal of the American Statistical Association 53(282), 457–481 (1958). Provides a survival estimator for incomplete and unequal follow-up by incorporating right-censored observations. registry ↩