Diversified Forecast Pool¶
Method — instantiates Ensemble Decision Aggregation
Combines forecasts from multiple forecasters, methods, horizons, or data feeds to support planning under uncertainty.
A Diversified Forecast Pool is a standing arrangement for combining forecasts that were produced independently by different people, methods, and data feeds. Its defining idea — the one false of its pooling siblings — is that it preserves the spread and the outlier's rationale rather than collapsing them: the pool publishes a combined forecast, but it keeps the dissenting forecasts and their reasons alive as first-class output, because the lone forecaster who sees a surge is often the point of the exercise. It is an operational, recurring planning input, not a one-off artifact. Where model averaging spends its effort deciding the weights, a forecast pool spends its effort ensuring the contributors are genuinely diverse and independent — and then usually just averages them.
Example¶
A public-health agency runs a forecasting hub for weekly hospital admissions during flu season. It does not build one model; it invites many independent teams — some running mechanistic epidemic models, some purely statistical, some human-judgment forecasters — to submit a forecast each week in a common format. The hub combines them, typically with a simple median, into an ensemble forecast that planners use to stage staff and beds.
One week, most teams project a gentle plateau, but two teams — reading an early signal in a new variant's growth rate — forecast a sharp surge. The hub does not average that warning away. It publishes the ensemble median and flags the high tail with the two teams' rationale attached. Planners hold surge capacity in reserve on the strength of the minority. When admissions spike two weeks later, the preserved outlier is what let the system prepare — a result no single "best" model would have delivered.
How it works¶
- Recruit diverse, independent producers. The value is in the mix — different methods, data feeds, and horizons — and in their independence, not in headcount.
- Align to a common target and format. Every contributor forecasts the same quantity in the same shape so the submissions are combinable.
- Combine simply and robustly. Usually a mean, median, or trimmed mean — simple combinations are famously hard to beat with clever weights.
- Preserve the spread. Publish the combined forecast with its dispersion and the rationale of dissenting forecasts, and refresh the pool each cycle as new forecasts arrive.
Tuning parameters¶
- Breadth of contributors — how many genuinely different methods and feeds feed the pool. More real diversity lowers correlated error; more of the same only adds volume.
- Combination rule — mean, median, or trimmed mean. The median and trimming resist a single wild submission; the mean keeps every voice.
- Outlier-retention policy — how aggressively dissenting forecasts and their reasons are surfaced versus smoothed. Retaining more catches rare warnings but complicates the headline.
- Refresh cadence — how often the pool re-collects and re-combines, trading responsiveness against churn.
When it helps, and when it misleads¶
Its strength is the aggregation dividend from independent, diverse forecasters — a simple combination routinely beats most of its individual members, and preserving the spread turns disagreement into an early-warning system for planning.
Its failure mode is that independence quietly erodes. When forecasters read each other's submissions or anchor on the published consensus, they herd, and the pool's apparent agreement becomes correlated echo rather than corroboration. Collapsing everything to one clean number — the temptation every planner feels — buries exactly the outlier that carries the actionable signal. And over-tuned weights tend to lose to the plain average.[n1] The discipline is to protect contributor independence, keep the spread and dissent visible, and resist the urge to over-engineer the combination.
How it implements the components¶
ensemble_member_set— the roster of independent forecast producers and data feeds whose submissions are pooled.diversity_criterion— values genuinely different methods, horizons, and data over more forecasts of the same kind.minority_signal_preservation— its defining component: it keeps dissenting forecasts and their rationale visible instead of averaging them away.
It does not decide the members' relative say through a statistical weight scheme — that is Model Averaging, whose separating component is member_weighting_rule; and it produces no single trained composite predictor with an out-of-sample calibration_feedback_loop, which is the job of Ensemble Model.
Related¶
- Instantiates: Ensemble Decision Aggregation — the operational instance where independent forecasts are pooled and their spread preserved.
- Consumes: Ensemble Model and Simulation Ensemble — individual contributors may themselves produce their forecasts with these before submitting to the pool.
- Sibling mechanisms: Ensemble Model · Model Averaging · Simulation Ensemble · Expert Panel · Scenario Ensemble · Committee Scoring · Multi-Source Intelligence Synthesis
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Diversified Forecast Pool operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it combines forecasts from multiple forecasters, methods, horizons, or data feeds to support planning under uncertainty.
Independent corroboration: The frozen evidence defines Diversified Forecast Pool as 'Combines forecasts from multiple forecasters, methods, horizons, or data feeds to support planning under uncertainty', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Forecast-combination research cohered pooling predictions from distinct models or forecasters, often finding robust gains from simple averaging.
Related originating lineages:
- Economics & Finance — Econometric forecasting developed a major independent literature on combining model forecasts.
- Futurism & Strategic Foresight — Superforecasting and judgmental-forecast practice supplied diverse human forecast aggregation.
Review resolution: Both current reviews place diversified_forecast_pool primarily in statistics_experimental_design; the reconciled classification retains only lineages that materially shaped the mechanism and keeps breadth of origin separate from reach.
Attribution caveat: Statistical, econometric, and judgmental forecasting produced convergent pooling methods.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The forecast-combination puzzle is the repeated empirical finding that a simple equal-weighted average of forecasts tends to match or outperform combinations using optimally estimated weights, because the weights must themselves be estimated from limited data and carry their own error. It is a strong argument for keeping combination rules simple. ↩