Community / Research Note

Finding Repricing Windows in the France World Cup Market

Will France win the 2026 FIFA World Cup?

This post shows how regime modelling can be used to turn a raw trade export into a cleaner narrative. Trades are bucketed and labeled into a small number of recurring states, then the rest of the analysis is organized around regime transitions.

PolymarketSample export

Outcomes

Yes / No

Model

HMM-style regimes

Extra

Traditional ML baselines

What this demonstrates

Regime labels separate quiet drift from repricing windows, which makes market behavior easier to describe and compare.
Price and volume charts become more actionable once the “interesting windows” are clearly isolated.
The same workflow generalizes across markets, so users can build repeatable analysis rather than one-off charts.

Regimes (market moods)

A regime is a recurring market environment. The model buckets trades into fixed time windows and uses simple features like returns, volatility, and activity to assign each bucket to a hidden state. After fitting, states can be interpreted as “quiet drift” versus “fast repricing”.

In this market, the most useful output is the separation between long calm stretches and shorter windows where pricing moves faster and participation increases. Those higher-activity windows are where a trader would focus attention and where execution conditions can change quickly.

HMM regimes — Regime detection (HMM-style). Price colored by inferred state. Click to zoom.

Traditional ML baseline: simple next-step model

A simple linear model was also fit on the bucket features (returns, volatility, activity) to predict the next bucket’s return. This is not presented as a production signal. It is a baseline to test whether the feature set contains any short-horizon structure.

In this export, the baseline shows a small but non-zero out-of-sample fit, which suggests there is a weak relationship between “current bucket conditions” and the direction of the next move. For trading, the right takeaway is not “this predicts the market”, but that volatility and activity features help identify windows where repricing risk is higher.

PCA projection of bucket features. Useful for seeing whether market behavior forms distinct clusters. Click to zoom.

k-means clusters over price — k-means clustering of bucket features overlaid on price. Clusters often align with drift vs repricing phases. Click to zoom.

Volume clusters

Volume tends to arrive in bursts. Those bursts are often where execution conditions change and where the most informative trades happen.

Volume over time — Notional volume over time. Click to zoom.

Repricing moments

The price-jump view is a fast way to find the timestamps that matter. It highlights large single-step moves that are typically tied to information shocks or aggressive positioning.

Largest price jumps — Price jump detector (largest single-step repricings). Click to zoom.

Whale prints

Large prints help explain how the market got from one probability level to another. They are also useful for sanity checking whether a move looks broad-based or driven by a small set of participants.

Whale trades — Biggest trades by notional volume. Click to zoom.

← Back to Community