United States · forecast accuracy pilot
Forecast comparison
A transparent pilot comparing published economic forecasts against actual outcomes, starting with US real GDP growth. Dates, sources and any corrections are tracked openly here. This is research and decision support, not individualized investment advice.
Pilot status
Planned — no prospective round has been scored on this page yet
This page previews how the comparison will read once real forecasts are admitted and scored. No prospective round has a registered protocol or calendar, and no institution has been named as a participant or asked to endorse this comparison. (Separate internal research has reproduced a retrospective comparison on already-published numbers; that is research validation, not a public, prospective tournament round, and nothing from it is shown here.) "Forecast availability" (has a forecast been published and admitted) and "results availability" (has it actually been scored against an outcome) are different questions — this page always answers both separately, never implying one from the other.
In preparation — the only metric currently being worked on
4 further metricsPlanned — definitions only, nothing running yet
Comparison
Illustrative example — not real data
To show the shape of a future row before any real one exists, here is one invented example, all targeting the same clearly fictional quarter: Q3 2027, invented for this example and not a real forecast target. The names and numbers are made up; no real institution, person or product is named or implied — including ForeGlass itself, which is not a historical entrant in this invented table.
| Forecaster | US real GDP growth, Q3 2027 (Q/Q SAAR, %) | Original issue | Information cutoff | Outcome |
|---|---|---|---|---|
| Example Forecaster A | 2.1% | 2027-07-02 | 2027-07-15 | Pending |
| Example Forecaster B | 1.6% | 2027-07-10 | 2027-07-15 | Pending |
| Example Forecaster C | 1.9% | 2027-07-13 | 2027-07-15 | Pending |
Invented for illustration only, with one common cutoff so the row shows what a fair comparison looks like; each forecaster's original issue date can differ, as long as it comes before that shared cutoff. A real row shows the forecast's original publication vintage, never a value recomputed afterward in this page's own code.
Upcoming rounds
No prospective round has a registered protocol or calendar yet. A later announcement will say exactly when scoring starts and which forecasts are eligible, and this page will link to it once it exists. Nothing is backdated: a forecast made after its own cutoff is not counted, no matter when scoring begins.
October 15 results
As of that same 2026-10-13 snapshot, at least 102 publicly visible ledger entries were due October 15, covering six distinct questions across three underlying data sources. (The ledger also holds additional sealed entries whose detail isn't public yet; this count reflected what was visible as of the snapshot date, not the full due cohort.)
Household Treasury-holdings share
One question: what share of US Treasury securities households directly hold, next quarter (Federal Reserve FWTW). Its grading rule has an unresolved internal definitional dispute — this question may stay held past October 15 even once the underlying data is published.
Household wealth shares
Two questions: the top-1% and bottom-50% shares of US household net worth, next quarter (Federal Reserve DFA). Of the six, these are the most likely to show a real resolved number first, pending a reviewed source-edition policy.
Payroll benchmark revision
Three questions, all about one BLS preliminary payroll benchmark revision: its size, its direction and whether it's material. The underlying data was effectively public before these forecasts were sealed — once resolved, treat these as a historical diagnostic, not evidence of genuine prospective forecasting skill.
No combined accuracy score is shown for this cohort. Even in the best case, at most two of the six questions (the wealth-share pair) could count as genuine prospective evidence — the payroll questions are historical once resolved, and the Treasury-holdings question is held pending its definitional dispute. Two is below the five-question minimum this project requires before showing any combined score; per-question results are still shown individually. Read the live sealed-forecast ledger →
Mock result row — demonstration only, not a real question or date
This is a different kind of "not real" than the Q3 2027 comparison example above: that one previews the GDP comparison's future shape, this one previews how an October-15-style ledger entry reads once it actually resolves. The metric name, prediction and outcome below are invented and match none of the six real questions.
| Question (invented) | Sealed prediction | Observed | Error vs. naive baseline | Status |
|---|---|---|---|---|
| Mock metric: widget-index quarterly change | 3.4% [2.9–3.9] | 3.6% | −0.2 pp better than baseline | Resolved (mock) |
Invented for illustration only. A real row never recomputes its prediction after the fact, and shows "held," "pending" or "unknown release" instead of a value whenever one of those actually applies.
External evaluation (in preparation)
Organizer and entrant disclosure. Comuvia organizes the comparison above, and intends to include separately identified ForeGlass forecasts in it once eligible, the same as any other entrant. External evaluation is different: it is independently administered by someone else, and there Comuvia would participate only as an entrant, not an organizer. No current house forecast or external acceptance is claimed by this page. No platform named below has accepted, reviewed, or endorsed Comuvia or ForeGlass, and nothing on this page implies otherwise.
Two independently-run platforms are being evaluated for fit, not yet joined:
- Good Judgment Open — Q4 2026 finance and economics challenge: the closer topical match (GDP, inflation, yields); automated/company participation terms still need clarifying with the platform before any entry.
- Metaculus FutureEval: built for automated forecasting entrants, with a broader subject range than this page's GDP focus.
Both are third-party platforms Comuvia does not control; their rules, availability and URLs can change independently of this page. An account, a disclosed automated/company method and a capped evaluation budget would all need to exist before any real entry — none of that has happened yet.
Methodology and evidence
Every scored forecast will carry its original source, its publication vintage and the exact resolution rule used to grade it — the same sealed-ledger approach ForeGlass already uses for its own forecasts. Read the public Verification Explorer →
A forecast's GDP annualization convention, its information cutoff, its vintage (the data release it is graded against), and its error are always shown next to a number, never hidden behind it.
Metric roadmap
US real GDP growth is the only metric actively being prepared for scoring in this pilot. The other four rows below are definition work only — nothing is running yet, and a row is never made to look functional before it is.
| Metric | Status | First intended definition |
|---|---|---|
| US real GDP growth | In preparation | Quarter-on-quarter, seasonally adjusted annual rate (%), exact quarter, originally published BEA advance edition |
| Headline PCE inflation | Planned | Total PCE price index; proposed Q4/Q4 percent change |
| Unemployment | Planned | Civilian U-3, seasonally adjusted; proposed Q4 quarterly mean (%) |
| Core PCE inflation | Planned | PCE price index excluding food/energy; proposed Q4/Q4 percent change |
| 10-year US Treasury yield | Planned | Exact reference series and quarterly averaging convention still to be fixed |
These are planning targets, not a frozen protocol — they may be refined before any of the four planned metrics is actually scored. Source institutions named in that research (survey panels, agency projections) have not agreed to participate and are not described as voluntary participants by this page.