When Do Sportsbooks Disagree?

In equity markets, when two exchanges quote different prices for the same security, the gap is closed in milliseconds by algorithms that exist for precisely this purpose. Sportsbooks are not equity exchanges. They have no consolidated tape, no regulatory mandate for price consistency, and their trading desks operate on human schedules. Cross-book pricing disagreements in sports betting are consequently not microsecond events. They are minute-to-hour events, and their duration follows a pattern that is not random.

We built a system that captures moneyline odds from 40+ sportsbooks every 60 seconds. When two books' combined implied probabilities on opposite sides of the same game sum to less than 100% (net of each book's vig), we record it as a pricing discrepancy and track its duration. Over one week, we catalogued 6,439 such discrepancies across 161 events in four sports.

The most notable feature of the data is the time-of-day dependence.

During peak hours — 8 AM to 3 PM Pacific, when the daily MLB, NBA, and NHL slates are being priced and every US trading desk is active — discrepancies appear in volume. The 2 PM hour generated 684 discrepancies, the highest of any hour, and the 10 AM hour produced 658. But these daytime discrepancies close quickly: average resolution times at 3 PM and 9 AM are 13 minutes. The market is doing what competitive markets do.

After 7 PM, the character changes. Volume drops — only 69 discrepancies at 9 PM, compared to 684 at 2 PM — but the ones that appear last far longer. The 9 PM hour shows an average discrepancy duration of 89.6 minutes, nearly an hour and a half. At 7 PM it is 67.5 minutes. Between 10 PM and midnight, the average remains above 45 minutes. We observed one discrepancy that persisted for 18.3 hours, detected at 7 PM when most US trading desks had closed.

The ratio is roughly 5:1. An average discrepancy detected during peak afternoon hours resolves in about 15 minutes. An average discrepancy detected between 7 PM and midnight takes about 60 minutes. Same market, same sportsbooks, same sport — the only variable that changed is the time on the clock.

The analogy to equity markets is imperfect but instructive. After-hours stock trading is thinner, spreads are wider, and prices can drift before the regular session corrects them. The same dynamic operates here, but the mechanism is different. It is not liquidity that dries up — bettors can wager around the clock. What dries up is human supervision. US sportsbooks run automated pricing engines that are monitored and adjusted by traders during business hours. When the traders leave, the engines continue quoting, but they respond more slowly to cross-market movements and overnight news.

A secondary factor is the geographical handoff. As US desks close, European and Asian books — serving their own customer bases — begin adjusting NBA and NHL lines based on their own flow. These adjustments sometimes diverge from the US consensus. Neither side is wrong, exactly. They are pricing the same event for different populations, with different information sets and different risk appetites. The resulting disagreements take longer to resolve because the natural counterparty — the US desk that would normally react — is closed.

Sport concentration

Of 6,439 discrepancies, 4,927 (76.5%) involved NBA games, 1,035 (16.1%) involved MLB, 340 (5.3%) involved NHL, and 137 (2.1%) involved NCAAB. The NBA concentration reflects the sport's combination of broad international coverage (many books pricing the same game creates more pairwise opportunities for disagreement), frequent line-moving information (injury reports, lineup decisions, load management announcements), and the timing of its schedule: most NBA games tip off between 7 and 10 PM Eastern, which places the initial pricing window squarely in the afternoon peak zone and the in-game adjustments in the overnight zone.

MLB's 16.1% share is worth noting. Baseball season barely overlapped with our observation window, and yet it generated more than a thousand discrepancies — likely because MLB moneylines have the tightest margins in US sports (see our companion study, "The Vig Tax"), and tight margins mean that even small absolute disagreements between books can produce pricing gaps.

Caveats

The limitations should be stated plainly. One week is a short observation window, and we would not claim robust seasonal conclusions from it. Our 60-second polling cadence means we miss sub-minute discrepancies, which are likely numerous during peak hours and would, if captured, skew the daytime count higher and the daytime average duration lower. We track only h2h (moneyline) markets; spread and totals markets, which are generally more liquid, may behave differently. And we filtered out discrepancies with implied gaps above 3%, on the assumption that these are more likely data errors than genuine pricing disagreements — a reasonable assumption, but one that could exclude real extreme mispricings in rare cases.

We also excluded prediction market platforms (Kalshi, Polymarket) from this analysis entirely. Cross-platform discrepancies between sportsbooks and prediction markets involve structurally different instruments with different fee models and are better analyzed separately.

What we can say with some confidence is that the overnight pattern is real and its magnitude is not trivial. The average discrepancy detected between 7 PM and midnight lasts roughly four to five times longer than one detected during the peak afternoon window. Whether that constitutes an inefficiency or merely a reflection of lower-urgency pricing during off-peak hours is a question we leave to the reader.