At halftime a book has to put up a number for the second half. It has just watched two quarters, and so has everyone else. If those two quarters were quiet the number comes down; if they were wild it goes up.
It mostly shouldn't. The first half barely predicts the second. So when a live number swings hard at the break, that is usually the market reacting to noise — and the side it moved away from is the cheap one.
Correlation between a game's first-half total and its own second-half total: r = +0.06. A first half ten points busier moves the expected second half by 0.7 points.
| first half came in at | games | second half then averaged |
|---|---|---|
| under 14 | 184 | 21.0 |
| 14–19 | 205 | 21.6 |
| 20–25 | 305 | 22.6 |
| 26–31 | 231 | 21.9 |
| 32+ | 162 | 23.4 |
A 27-point swing in the first half moves the second by 2.4. Essentially flat.
Loading the board…
Each side's game-level first- and second-half totals from the last completed season, averaged across both teams, then scaled so the two halves add up to today's posted total.
It carries how big the game is expected to be. It does not carry a read on when the points arrive — see below.
Second half includes overtime, which is how a book grades a 2H market. Only the first half is measured directly; the second is the remainder.
It is not a team tendency. We checked whether "this team plays slow first halves" is a real trait. It is not. Split-half reliability across four seasons is r = −0.05 — a team's split in its odd-numbered games tells you nothing about the same team, same season, in its even-numbered ones. Year over year it is r = +0.05.
Team splits look spread out — one club at −9, another at +7 — but a single game swings with a standard deviation of 12.8 points. Over seventeen games that is a standard error of about 3.1, which is nearly the whole 3.4 spread you see. Sort teams by their half split and you get a table that looks meaningful and predicts nothing.
It is not a price. Nothing here knows what a book will actually hang at halftime. The thing that would make this pay is the live number over-reacting, and that is a property of the in-game market which we have not measured yet. This is the reference you hold it up against.