The Lab · Reading the market · updated 2026-10-22
Does line movement mean anything?
We measured it across 122 games. Line movement genuinely predicts where the line goes next — and following it still loses money. Here's the number, the two traps we nearly fell into, and what's actually going on.
Yes — a line that's moving tends to keep moving the same way. But the effect is small, it doesn't grow with bigger moves, and it doesn't pay for the bookmaker's fee. Its one real use is timing: if you were going to bet a side anyway, earlier beats later.
Record figures marked as today's come live from the public record, refreshed as games are graded — as of Oct 31, 2026. Numbers we published on earlier dates stay as written, dated, so you can check them against today's.
A line opens at −140 and by first pitch it's −160. Somebody moved it. The folklore says that somebody was sharp, and you should follow.
We tested it on our own data. The answer has two halves, and almost everyone stops after the first one.
What line movement is
A betting line isn't a forecast handed down once. It's a price that gets adjusted every time money arrives or news lands — a lineup scratch, weather, a bettor the book respects. When the number drifts from open to close, that drift is the market changing its mind in public.
The question is whether the drift is information — whether the direction it's already travelling tells you where it's going next.
Yes. It genuinely predicts.
Our model watches lines all day and records when a game's consensus has moved by at least half a point. Then we ask a narrow question: from the moment that move was detected, did the market keep going that way through to the close?
Across 122 games:
- The market continued in the same direction 68.9% of the time
- When movement pointed up, the consensus went +0.71 points further
- When it pointed down, it went −0.28 points further
Those two buckets pulling apart is the finding. Pooled with a second, independent sample, the separation is 1.10 ± 0.37 points — about three standard errors from zero. This is not folklore. Line movement really does lead the close.
The first trap: is it just drift?
Here's what nearly fooled us, and it's worth understanding because the same mistake is everywhere in betting analysis.
Suppose lines simply tend to drift one direction over a day — public money piling onto favorites, say. Then a "movement" signal would usually point that way too, and the market would keep drifting that way, and you'd measure a beautiful correlation containing no information whatsoever. You'd just be rediscovering that the tide comes in.
The test that separates them: split the results by which way the movement pointed. If it's drift, both groups move the same direction. If it's real, they pull apart.
Ours pulled apart — up-moves led up, down-moves led down — and the signal fired almost perfectly evenly (60 up, 62 down), which matters because with balanced firing, drift cancels out entirely. So the effect survived the control.
But it loses money anyway
Here's the half nobody quotes.
Following that movement captures about 0.46 points of value. A sportsbook's fee on a two-way market runs roughly 1.5 to 2.25 points of implied probability per side.
+0.46 points of edge − 1.75 points of fee = −1.29 points per bet — at even money, about −2.6% of every stake
Corrected 2026-10-22. This line used to subtract the two as percentages ("+0.46% − 1.75% = −1.29% per bet"). They're points of implied probability, and at even money a point is worth about 2% of your stake, so the true loss is roughly twice what we printed. The conclusion stands; the arithmetic is now in the right units.
The strategy works and still bleeds. It's the same arithmetic as favorites versus underdogs: being right isn't the bar. Being right by more than the fee is the bar, and a half-point of edge doesn't come close.
You'd need roughly four times the effect to break even.
The second trap: does a bigger move mean more?
The obvious rescue: maybe small moves are noise and big moves carry real weight. Bet only the dramatic ones and clear the fee that way.
We tested that too — whether louder movement leads further. The slope came back at +0.07 with a z of +0.09: flat. Statistically indistinguishable from zero, and a second sample agreed.
A three-point move is worth no more than a half-point move. There's no threshold to raise, no premium subset hiding inside the data. The edge is a fixed, small offset that arrives the same size every time.
That single number is what killed the strategy for us — not the losing, the flatness. Losing tells you where you are; flatness tells you there's nowhere better to stand.
What's probably actually happening
Our best read is that this isn't foresight at all.
When our model detects "the line moved," what it has really detected is that some books moved. The lead we measure is largely the slower books catching up by close. That would explain everything: why it's directional, why it's small, and why it doesn't scale — a lagging book has a fixed amount of catching up to do regardless of how big the move was.
If that's right, the honest way to capture it isn't a signal at all. It's shopping every book before you bet — taking the best available number instead of the first one you see. That's a real, reliable edge available to anyone, and it requires no theory about sharp money.
What this doesn't tell you
Line movement predicting the line is not the same as predicting the winner. We tested that separately: applying this signal to our published probabilities made them no more accurate at all. It moves the price, not the result.
Update, September 5. With more games — 338 — it stopped being "no more accurate" and became measurably less accurate: a Brier score worse by 0.00136 ± 0.00068, about two standard errors. Both things are true at once: the signal leads the closing line, and it makes a closing probability worse. Why both can be true.
Our 122 games are a real sample but a single-sport, single-season one, and roughly half came from a stretch where some closing data was later lost to a storage cleanup — a flaw we found and fixed, and one we'd rather disclose than paper over.
Update, October 10. We later ran the overnight version of this idea on 26,000 games of purchased odds across four sports, building it on 2021–23 and checking it on 2024–25. It held up in both: the side the line moved toward won about 1.5 points more often than the morning price said. At the best US price that returned +1.0%, then +0.2% — roughly paying the fee back, and no more. Same verdict, far bigger sample. The full test, with six others.
And "it loses at these fees" is not "it can never work." A market with a thinner fee, or a bettor who consistently gets the best number across ten books, changes the arithmetic. What we can say is what we measured: on our data, at ordinary prices, following line movement is a real effect that is too small to pay for itself.
Our board said the same thing from every angle — 332 games graded in public when we wrote this, 54.5% against a claimed 55.9%, and 1,608 games today, 59.7% against 58.6%. The market is a little bit beatable and quite a lot expensive.
Model output is informational and entertainment content, not betting or financial advice. If you bet, bet what you can afford to lose.
Log what you actually bet — units and dollars — and see it graded beside a model that publishes every call it makes, wins and losses alike. Free, no card.
Start a free tracker