Favourite–Longshot Bias in Current Production

Favourite–longshot bias is one of the oldest regularities in betting markets. Horse-racing papers documented it decades ago; prediction markets show the same shape. What changed in production is why the curve bends — market makers run LMSR-style books under inventory pressure, and tail prices rise when they need to slow flow.

This page walks through how we measured that in live venues: from raw tape to calibration charts, then into the maker mechanics that generate the gap. Synthetic numbers stand in for venue-level data here; the pipeline and visuals match what we run internally.

How we built the dataset

We start with production fills joined to settlement. Each contract keeps the price quoted at trade time — not the closing line — because that is what bettors actually paid. Hover a stage to see what happens there.

Pull fills + quotes at trade time

After bucketing, the left tail is consistently rich: implied win rates run above realized outcomes. That pattern survives across sports and politics books once you control for selection on open interest.

Production calibration gap

The aggregate view buckets contracts by quoted price and compares market-implied win rates to realized outcomes. Hover a bucket to see the calibration gap in percentage points. The longshot bins are where bettors systematically overpay.

1–10¢10–20¢20–35¢35–50¢50–65¢65–80¢80–95¢

Contract-level view

Buckets hide dispersion. Plotting each settled contract shows a cloud above the diagonal in the tail — individual longshots quoted rich even when favourites sit near fair. Brighter dots are sub-15¢ contracts.

quoted price →realized

Each dot is a settled contract. Tail cluster sits above the diagonal — market-implied rate overshoots realized wins.

How makers produce the curve

The calibration gap is not purely behavioural. When a maker accumulates longshot inventory, they steepen tail prices to discourage further flow — the production curve lifts above fair LMSR. Drag inventory to see the markup appear.

41.8¢ quoted

Tail markup at current position: +11.2pp vs fair LMSR

Recalibrating tail quotes

For consumers of market prices — forecast dashboards, cross-venue aggregation — we apply log-odds shrinkage fit on held-out settlement data. The slider shows how much an 8¢ quote moves toward fair at different shrinkage levels.

8¢ bucket

Pulling tail quotes toward fair: 8¢ bucket moves down ~3.8pp at current setting

About this demo

Synthetic data and live visuals — a preview of the tooling we attach before publishing venue-level numbers. The formal write-up covers identification, cross-book robustness, and maker inventory inference in full.

← Back to research