Mysports.AIMysports.AI
Sign in
Mysports.AI Research

Mysports.AI Research: Sports Betting Odds and Match Data Studies

The Mysports.AI team keeps researching sports to push our models and data further.

We study player news, odds and prediction platforms to learn which data help predict games. Here we publish each study’s findings, progress and limitations.

A basketball player with a wrapped knee sitting courtside, next to an injury record board
R1AnalyzingPlayer news

When News About a Key NBA Player Breaks, Do the Odds Move Right Away?

A study of sudden player-status news: when a key player is suddenly ruled out or cleared to play, how fast do the odds on the prediction platform Polymarket change?

Adding up all the news: after the NBA posts injury news, most of the odds change after the report happens in the first 15 minutes. About 40% of the whole change happens before the official report (about half when re-measured trade by trade at exact seconds).

Limitation：This comes from adding up many news items, so not every item is the same. We only have one odds reading every 5 minutes. The tiny change after 15 minutes is no longer clear under a different data filter.

Where we are：Trade-by-trade re-measurement (plan timestamped first): about 90 seconds after the official PDF goes up, trade prices have completed half of the post-release adjustment; re-measured at exact seconds, the hour before release accounts for about half of the move (40% with 5-minute prices)

Next：Check again on new games from the 2026-27 season; we compare the answers in mid-February 2027

Start date Last updated
Read the full study
NBA odds on the Taiwan Sports Lottery website: away and home columns
R2AnalyzingOdds comparison

Same Game: How Do Taiwan Sports Lottery Odds Compare with an International Bookmaker?

Odds decide how much you get back when you are right. We compare the odds Taiwan Sports Lottery and the large international bookmaker Pinnacle offer on the same outcome of the same game.

Using the same estimated win chances, for every NT$100 put in, Taiwan Sports Lottery is expected to return about NT$83 on average and the international bookmaker Pinnacle about NT$96 — a gap of about NT$13. The gap is even bigger on underdogs.

Limitation：An average worked out from estimated win chances — not real betting records, and not a fixed deduction; the data only covers Sept–Oct 2026.

Where we are：Extended analysis (post-hoc): the higher the odds Taiwan Sports Lottery offers, the less you can expect back on average, and popular leagues get no better deal; a confirmation on games from Oct 23 onward is pre-registered

Next：Add new games every month; in February 2027, confirm whether the gap is still there

Start date Last updated
Read the full study
A receipt, coins and a calculator under stadium lights, representing trading costs
R3AnalyzingTrading costs

Polymarket Raised Its Fees — How Much More Does a Trade Cost?

We compare what one trade cost on Polymarket before and after its fee change, for the same kind of games, and also check whether the gap between buy and sell prices changed.

After the fee change, the cost of trading on Polymarket went up, mostly because of the fee itself; we could not see the gap between buy and sell prices widen.

Where we are：Removed numbers compared with another platform (data terms restrict it), keeping only Polymarket’s own before-and-after comparison; a forward comparison around the NBA fee return is collecting data

Next：Once NBA fees return, collect about two more weeks of data and compare costs before and after under the pre-registered rules

Start date Last updated
Read the full study
An ice hockey goalie standing in front of the net
R4AnalyzingPlayer news

After an NHL Starting Goalie Is Confirmed, How Long Until the Odds Move?

A study of “goalie confirmations”: after reporters confirm who will start in goal tonight, do the odds on Polymarket change, and how long does the change last? (The first stop of our starting-lineup research.)

After a goalie is confirmed, the odds keep changing a little, but only by a very small amount; we can’t yet say how long it takes for the odds to fully reflect the news.

Limitation：We only have one odds reading every 5 minutes; the changes are very small, and we did not calculate trading costs this time, so we can’t tell whether it is profitable; the “how long” estimate is not very precise.

Where we are：Extended analysis (post-hoc): confirmations of a backup goalie starting unexpectedly came with clearer odds moves, while a starter returning unexpectedly barely moved them; new-season data collection is under way

Next：Check again on new games from the 2026-27 season; we compare the answers in mid-February 2027

Start date Last updated
Read the full study
A phone showing a list of game predictions next to a clock, illustrating the tracking of public prediction records
R5AnalyzingPublic picks

After a Public Pick Is Posted, Do the Odds Move Its Way?

After free NBA picks from PTT, Yuncai Xiaowei (a Taiwanese tipster) and Predictem are posted, does the backed team become more favored on Polymarket before tip-off? Pick data are collected by betlead.win.

No sign of it: between posting and tip-off, the backed team’s odds fell (it became more favored) 49% of the time, about the same as 50% for comparable games. Net of the usual drift, its implied win chance barely changed on average (−0.002 percentage point).

Limitation：Three sources, two of them single authors, so results represent only these sources; last season’s posting times were self-reported by the sites.

Where we are：2025-26 main results are in and corrected after AI-tool code review; all five data checkpoints passed

Next：Re-test on the 2026-27 season with the same method (decision on 2027/2/15; until then we only count picks without looking at results)

Start date Last updated
Read the full study
Illustration of a starting formation on a soccer field, with one position marked with a question mark
R6PublishedPlayer news

When Soccer Lineups Come Out, Do the Odds Move Right Away?

A study of “full starting lineup announcements”: about 1 hour before kick-off, both teams name their 11 starters. How do the odds on Polymarket change after that, and for how long? (The second stop of our starting-lineup research.)

Yes, and fast: 15 minutes after a Premier League lineup is announced (timed from the rule-set release 75 minutes before kick-off, not measured game by game), the odds have made about 60% of their move. The size is small, though—about 0.40 percentage point in implied win chance in the 30 minutes around the announcement. Whether odds move before the announcement, or keep drifting after it, is not yet clear.

Limitation：Only the Premier League has a fixed announcement time, and we use that rule time rather than measuring each match; moves are small, and fees and spreads were not computed, so profitability is unknown.

Where we are：Extension 1 (pre-registered): in MLS and the Europa League, odds also moved toward the line-up news in the estimated line-up window, by more than in the top five leagues

Next：Study completed; no separate new-season forward test, and any future re-test will be planned and timestamped first

Start date Last updated
Read the full study
A basketball player having their ankle checked by a trainer courtside, waiting to find out if they can play
R7AnalyzingPlayer news

When an NBA Player Is Listed “Questionable,” Do They End Up Playing?

We count what share of players listed as “Questionable” actually end up playing. Whether the market gets this uncertainty right is the question for the next step.

About 56% of key players listed as questionable two hours before tip-off ended up playing; 47% of those first listed less than 12 hours out played, versus 60% of those listed earlier. We could not see that this status predicts how the odds move before tip-off.

Limitation：The official injury report only goes down to rough categories like “Questionable,” and Polymarket only has odds on “will the team win,” so this version does not judge how well the market guesses a player’s chance of playing.

Where we are：10/29: a shared pre-registered add-on (timestamped) for the player-news studies—the 2026-27 re-check will also report a version using official tip-off times

Next：Re-check in 2026-27 with the frozen rules (decision 2027-04-14); the original J1 is expected to stop at a data gate, in which case the clearly labelled “J1 alternative” is reported instead

Start date Last updated
Read the full study
A basketball, a glowing computing device and a court analysis panel, illustrating AI analyzing games
R8Collecting dataAI predictions

Do Teams Favored by AI Become More Favored Before Tip-Off?

Three AI models forecast win chances before games; we check whether they anticipated pre-game odds moves and compare their accuracy. Part 1 uses the new NBA season; Part 2 looks back at 2026 MLB forecasts already locked before games on aisportsinfo.com.

Part 2 (2026 MLB, look-back): no sign of it. From just after the AI forecasts were submitted until tip-off, the AI-favored team’s implied win chance changed by only +0.01 percentage point on average, and there is no sign the AI was more accurate than the market.

Where we are：Part 2 extension (post-hoc): re-checked with order books and trade prices, still no sign the AIs are a step ahead of the market; Part 1 (NBA) has been collecting daily since Oct 21

Next：Analyze Part 1 after the regular season ends in April 2027 to re-test whether AI can get ahead of the market

Start date Last updated
Read the full study
Illustration of a jersey, a star and an odds tag
R9AnalyzingOdds comparison

Are Popular Teams Overpriced?

Do popular teams like the Lakers, Warriors, Yankees and Dodgers win less often than other teams at the same odds? We also compare Pinnacle and Taiwan Sports Lottery.

No sign that popular teams are overpriced—or underpriced: at the same odds, they won 1.2 percentage points less on average, but the range runs from 4.7 less to 2.3 more, so luck cannot be ruled out.

Limitation：Only one popularity index and 10 popular team-seasons; the lists were computed after the season (using only the prior season), so the new-season re-test is the true out-of-sample check.

Where we are：Extension 1 (pre-registered): re-tested on the NHL and NFL, popular teams won 6.9 percentage points less often than other teams at the same odds, which meets our pre-set rule

Next：Re-test with the frozen lists in NBA and NHL 2026-27, NFL 2026 and MLB 2027 (both the original per-group checks and the combined four-league check are pre-registered)

Start date Last updated
Read the full study
Illustration of implied versus actual win rates
R10AnalyzingMarket accuracy

Are Underdogs Overrated?

How far do Polymarket odds at different price levels stray from actual results? We check the NBA, MLB, NHL and soccer.

No sign that underdogs are overrated: across 7,023 games in four sports, underdogs (implied win chance below 40%) had an average implied chance of 25.5% and actually won 25.1%—just 0.4 percentage point apart, indistinguishable from zero.

Limitation：Polymarket prices are displayed (mid) prices with no historical bid/ask or fees; this is not a blind test—the new-season re-test is.

Where we are：Main question closed: 4,000+ more MLB and NHL games still show no sign underdogs are overrated; only the registered new-season re-test and the NBA heavy-favourite forward test remain

Next：Re-test with the same rules on new-season games (NBA, NHL, soccer) from 2026-10-23 to 2027-02-28

Start date Last updated
Read the full study
Illustration of the international line moving first and Taiwan following
R11AnalyzingOdds comparison

When International Odds Move, How Long Until Taiwan Sports Lottery Follows?

Using only odds snapshots (no betting slips), we measure how long Taiwan Sports Lottery takes to move its odds the same way after Pinnacle’s odds change noticeably.

Where we are：Analysis completed; results withheld while permission to use the Taiwan Sports Lottery odds source is confirmed

Next：Publish after the data-source permission is confirmed or an allowed Taiwan Sports Lottery odds source is used; a new-season re-test plan is already timestamped

Start date Last updated
Read the full study
Illustration of an American football with a “Questionable” tag
R12AnalyzingPlayer news

When an NFL Player Is Listed “Questionable,” Do They Play—and Do the Odds Keep Up?

We count how often key players listed “Questionable” in 2016–2025 actually played, and check whether the odds the night before already reflected these public injury statuses.

In 2016–2025, about 72% of key NFL players listed “Questionable” on the final injury report played; only 48% after missing the last practice. The night-before odds did not appear to fully reflect these statuses, but the gap is small.

Where we are：Main results are in: 72% of Questionable key players played; the main odds test holds but the effect is small (about 0.1 points per player); the plan and analysis code were timestamped first, and results were reviewed with an AI tool

Next：Collecting the remaining ~178 games of the 2026 season; re-test with the frozen rules once 2026 participation data are released

Start date Last updated
Read the full study
Illustration of spread and moneyline price lines
R13AnalyzingOdds pricing

Do Polymarket’s spread and moneyline markets for the same game disagree — and which one moves first?

Comparing the spread and moneyline markets for the same NFL, NBA and NHL games: how far apart their implied win chances are before the game, whose odds move first, and which is closer when they disagree.

Where we are：Main results: a median 2.3-point gap one hour before NFL games; no detectable leader or accuracy difference; illogical prices absent in the NFL and rare in the NBA. Plan and code were timestamped first; results reviewed with an AI tool

Next：The remaining 165 regular-season games of the 2026 NFL season plus the playoffs are being collected; after the season we will re-check with the frozen rules

Start date Last updated
Read the full study
Illustration of odds jumping seconds after a score
R14AnalyzingIn-game

After an NFL Score, How Many Seconds Until Polymarket Odds Catch Up?

We line up the start time of every 2025 NFL scoring play (touchdowns, field goals, safeties) with every on-chain Polymarket trade to measure how many seconds until trades lean toward the scoring team, when the odds move most, and how much they catch up within a minute.

Where we are：Main results are in: odds barely move for 10 seconds after a score and complete half their move by second 25; within a minute they catch up only about 57% of the public model’s jump (main test supported); the plan and analysis code were timestamped first, and results were reviewed with an AI tool

Next：Collecting the remaining 165 games of the 2026 season; re-test with the frozen rules in mid-March 2027

Start date Last updated
Read the full study
Illustration of a clock split into four 15-minute slots with three highlighted
R16AnalyzingPlayer news

After the NBA Moved Its Injury Report to Every 15 Minutes, How Much News Lands Off the Hour?

We counted each player’s first status change on the official report during the 2025-26 regular season (from December 22, 2025) by the report it first appeared in (:00, :15, :30 or :45), to see how much news you would see late if you only checked on the hour.

Where we are：Main results are in: 71.9% of first status changes land off the hour, and the hour gets only slightly more than an even split; to learn news quickly, keep checking every 15 minutes. The plan and analysis code were timestamped first, and results were reviewed with an AI tool

Next：For 2026-27, recalculate with the same rules after R1’s forward cutoff

Start date Last updated
Read the full study

Want to see how the AI performed at a major tournament? Full 2026 World Cup AI prediction review (tournament over, record kept for verification)

How research helps our models

01

Find signals models miss

The timing of injury news, lineups and odds moves may hold information our models are not using yet.

02

Understand how odds and prices are set

Knowing where bookmaker odds and prediction-platform prices are accurate — and where they are off — tells a model what to rely on and what to avoid.

03

Checked before it goes into a model

Every finding is checked again on upcoming games and only used in a model if it still holds.

04

Track and refine

Markets keep changing, so studies are tracked over time and refined as new data arrives.

Where this research feeds into our models:NBA Model Explained, MLB Model Explained, NHL Model Explained, How Our AI Works

How a study makes it into our models

  1. 1Find a question worth studying in games and markets
  2. 2Collect data and write the study plan
  3. 3Analyze the data to find patterns
  4. 4Check again on upcoming games that the finding still holds
  5. 5Use validated findings to improve models and data, and keep tracking

FAQ

How does this research relate to Mysports.AI models?

Research is where model improvements come from. Validated findings improve the data our models use, bring estimated probabilities closer to real outcomes, and help control risk; dead ends save us from wasted effort. See AI Model Performance →

Is this betting advice?

No. The research is about understanding games and markets better and is not betting advice.

Why run so many studies at once?

Studies share data and tools — data cleaned for one can be reused by others — so everything moves faster and findings can be cross-checked from different angles.

How often is this updated?

Each card is updated as its study progresses; the latest progress and results appear directly on the card.

This page is a research log for reference only and is not betting advice or a guarantee of profit.