Rolling Marginal-Contribution Curve¶
Method — instantiates Aggregate–Marginal Trajectory Reconciliation
Estimates leading-edge direction by smoothing a declared sequence of entering units, cohorts, or periods.
The Rolling Marginal-Contribution Curve estimates the leading edge by taking a declared, ordered sequence of entering units — or cohorts, or periods — and smoothing their contribution over a moving window, producing a continuous curve whose level and slope represent "what is being added now." Its defining move is windowed smoothing of a declared sequence: it trades timeliness against stability through the window width, and it fixes the sequence order in advance so the estimate cannot be quietly re-sorted into a flattering shape. The output is the marginal member of the pair — a stable, current picture of the entering contribution, ready to be set against the aggregate.
Example¶
An online marketplace's gross merchandise value (GMV) is rising on the back of a large installed base of established sellers, and the question is what each new cohort of sellers actually contributes. The analyst orders new sellers by onboarding week and, for each week's cohort, computes its 90-day GMV contribution; a trailing eight-week rolling mean then traces a smoothed marginal-contribution curve. The curve shows new-seller contribution sliding for four straight months even as total GMV climbs — the entering direction that will eventually reweight the aggregate once these cohorts accumulate. A much wider window would have smoothed the slide into invisibility; a single week's figure would have been pure noise. The eight-week curve is the compromise that makes the leading edge both stable and current.
How it works¶
- Declare the sequence and its order. Entering units are ordered by their entry period or unit index, and that ordering is fixed before smoothing, so the curve cannot be re-sorted after the fact.
- Put each unit on a common maturity basis. Every entrant's contribution is measured at a comparable age so the curve reflects contribution, not differential aging.
- Apply a rolling or exponentially weighted window. The window smooths across the sequence to produce a continuous level curve, the leading-edge estimate.
- Preregister the window, show alternatives as sensitivity. One window governs; others are displayed to prove the shape is not an artifact of the choice.
Tuning parameters¶
- Window width — wide windows are stable but laggy and can hide the very transition being sought; narrow windows are timely but noisy.
- Weighting — flat (simple moving average) versus exponential decay; exponential responds faster but can be tuned to over-weight history.
- Sequence unit — individual units, cohorts, or periods; coarser units stabilize the curve but delay detection.
- Sensitivity set — how many alternative windows are shown alongside the governing one to demonstrate robustness.
When it helps, and when it misleads¶
Its strength is a stable yet timely leading-edge estimate that responds faster than full cohort summaries while resisting the whiplash of raw single-unit readings.
Its failure mode is over-smoothing: a window wide enough — or an exponential decay slow enough — to erase the transition it exists to catch, so a real change at the edge is averaged into invisibility[n1]. The paired hazard is window-shopping: quietly picking the window that tells the preferred story. The guarding discipline is to preregister the governing window, publish the sensitivity set, and keep the sequence order auditable so the curve cannot be reshaped by re-ordering the entrants.
How it implements the components¶
marginal_contribution_estimator— it is the estimator: a windowed smoother that converts a sequence of entering units into a leading-edge trajectory with a visible timeliness-versus-stability tradeoff.matched_aggregate_and_marginal_metric_pair— the curve is the marginal member of the pair, defined on the same construct as the aggregate and asserted to feed it through the contribution link.
It smooths levels into a leading-edge curve but does not report the raw period-to-period rate of change (opposite_sign_divergence_test — that is its nearest twin First-Difference or Derivative Estimate) or compare discrete like-aged groups (cohort_or_vintage_profile — that is Cohort or Vintage Analysis).
Related¶
- Instantiates: Aggregate–Marginal Trajectory Reconciliation — the curve produces the smoothed leading-edge trajectory the rest of the loop consumes.
- Sibling mechanisms: Aggregate–Marginal Sign-Divergence Alert · Cohort or Vintage Analysis · 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
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Rolling Marginal-Contribution Curve operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it estimates leading-edge direction by smoothing a declared sequence of entering units, cohorts, or periods.
Independent corroboration: The frozen evidence defines Rolling Marginal-Contribution Curve as 'Estimates leading-edge direction by smoothing a declared sequence of entering units, cohorts, or periods', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Smoothing contributions over an ordered moving sequence is a statistical trend-estimation technique.
Related originating lineages:
- Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: estimates leading-edge direction by smoothing a declared sequence of entering units, cohorts, or periods.
- Economics & Finance — Marginal contribution and leading-edge interpretation materially reflect economic analysis.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: estimates leading-edge direction by smoothing a declared sequence of entering units, cohorts, or periods.
Review resolution: Both blind reviewers agree that statistics_experimental_design is the primary historical origin. Explicit reconciliation of reported ambiguity, alternate origin disagreement, origin mode disagreement starts from reviewer_a’s mechanism-specific evidence: Smoothing contributions over an ordered moving sequence is a statistical trend-estimation technique. Reviewer A proposed alternates=economics_finance, origin_mode=convergent, domain_reach=multi_domain, and encyclopedia_synthesis=true; reviewer B proposed alternates=data_science, mathematics, origin_mode=cross_disciplinary_synthesis, domain_reach=multi_domain, and encyclopedia_synthesis=true. The final record retains every independently supported alternate from either review (economics_finance, data_science, mathematics) without an arbitrary cap, selects origin_mode=cross_disciplinary_synthesis to represent the combined lineage evidence, and keeps domain_reach=multi_domain and encyclopedia_synthesis=true from the more mechanism-specific assessment. Present-day transfer is recorded as reach and is not treated as proof of historical origin.
Attribution caveat: The generic entering-unit sequence does not identify a single named historical estimator.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
[n1] An exponentially weighted moving average weights recent observations more heavily than old ones by a decay factor; set the decay too slow and history dominates, so a real change at the leading edge is smoothed into invisibility — the central tuning hazard of any rolling estimator. ↩