Markets
The Information-Lead Board
One question, answered permanently and in public: does Weather Trader's issue-time fair value
carry information the market's price does not? Both probabilities are frozen at issue time — our
calibrated P(Tmax ≥ K) and the ForecastEx quote for the same contract — and the settled observed
high is the only judge. Proper scores, a block-bootstrap confidence interval, no retrospective edits.
When the evidence is thin, this board says so.
The headline accumulating
Loading the proof record…
Cumulative skill difference
running mean of (market Brier − WT Brier) over settled fair-lead contracts · above the zero line = Weather Trader ahead
Accumulating — the chart draws as fair-lead contracts settle.
By lead
days between the frozen quote and the valid day — the market prices Day 1; the longer leads are where an information edge would live
By station
same contest, split by city
Every joined contract — the frozen WT probability, the frozen market price, the outcome and both
proper scores — one row per contract, pending rows included so the freeze is auditable before settlement.
Method, in one paragraph. Our fair value P(Tmax ≥ K) is produced by the calibrated
station pipeline and
frozen at issue time; the market side is the ForecastEx daily close for the same
contract, frozen the same day (the feed carries no bid/ask, so the close stands in for the mid). Only
settled contracts are scored, with
proper scores (Brier and log) that reward honesty, not luck.
The headline is the difference in mean Brier (market − WT, positive = we lead) with a
95%
block-bootstrap CI blocked by date, because same-day contracts share one weather outcome and are not
independent samples. The fair contest uses only quotes frozen
≥1 day before the valid day — the
same-day close is quasi-settled (the high is nearly realized) and is excluded from the headline, shown only
as a reference. Nothing is re-scored after the fact. Full methodology:
/method.