Favorite-Longshot Pricing in NBA Markets on Polymarket: 61,981 Large Bets, Checked Against the Result

The previous NBA page claimed 16,382 large trades while one of its rendered chart datasets summed to 16,377, with no committed provider capture or reproducible query establishing which number, row set, or window was correct. That page now carries a correction withdrawing every volume, market, trader, timing, grade, consensus, strategy, and profitability claim it made, and it stays published at its own address so the links pointing at it resolve to that correction rather than disappearing.

This is the replacement the correction asked for: one retained row set, one defined universe, and every figure below traceable to the same query.

The universe is every row 0xinsider's `whale_alerts` table records with platform Polymarket, side BUY, and category NBA, the category the ingestion pipeline assigns at write time from the market's own listing, not inferred after the fact. The window runs from April 2 to July 24, 2026 in UTC: it starts April 2 because `outcome_index` was a defaulted 0 for a large share of buys before that day, the same defect the cross-sport calibration and sharp-money studies in this series work around, and it ends where NBA volume does, 47 trades in July and 2 in August as the season's board went quiet.

A trade counts only when its market has a recorded winning outcome and resolved after the trade was made, which rules out look-ahead and drops markets 0xinsider has no settlement record for rather than guessing a result for them. Price is bounded to 2 through 98 cents, cutting the few trades priced at the edges where a one-cent move changes the implied probability by an outsized amount. Each row is one recorded fill, deduplicated by the ingestion pipeline's own trade-event identity before it reaches this table, so no query-side dedup logic sits between the source and the count.

NBA carries no fixed alert floor. The 10th percentile and median notional by month (query 11 of the published SQL) sit near $3,000 across the window, so "large" here means roughly $3,000 and up, not the $1,000 floor the cross-sport sharp-money study uses or the $10,000 floor of the original grade study. Edge is the win rate minus the average price paid, in percentage points; dollar ROI is stake-weighted. Intervals resample markets, not trades, with 5,000 draws, because trades inside one market are not independent draws.

61,981 buys, 5,049 wallets, 1,818 settled markets. Buyers paid an average of 59.6 cents and won 59.4% of the time, minus 0.21 points against the price, with an interval from minus 1.75 to plus 1.26 that includes zero. The dollar-weighted return came in at plus 0.9%, a different number from the point-weighted edge because it weights by stake rather than giving every trade one vote; both describe the same sample.

That headline is a null result: at this size, this NBA sample shows no detectable gap between what large buyers paid and what the games were worth. The rest of this study looks for the gap in the places favorite-longshot bias is usually found and does not find it there either.

Split into longshots, the middle of the market, and favorites, none of the three intervals clears zero. Longshots under 30 cents ran plus 2.05 points ahead of their price over 3,122 buys, an interval from minus 5.30 to plus 10.40. The middle of the market, 30 to 70 cents, sat at minus 0.01, essentially flat, over 40,624 buys. Favorites at 70 cents and up came in minus 1.03 points, an interval from minus 6.69 to plus 4.22.

The classic favorite-longshot bias says longshots are systematically overpriced and favorites underpriced. This sample leans that way in its point estimates, longshots ahead and favorites behind, but every interval is wide enough to include no effect at all, and the finer five-cent buckets in the published output do not show a consistent slope from one end of the price range to the other.

Moneylines, the largest slice at 36,888 buys across 192 markets, came in minus 0.34 points, an interval from minus 2.73 to plus 1.96. Spreads, 13,314 buys, sat at plus 0.46, from minus 0.95 to plus 1.84. Totals, 7,805 buys, came in plus 1.08, from minus 0.61 to plus 2.76. All three intervals include zero, and none is more than a point and a half from it at the center.

The point spread study elsewhere in this series (0xinsider.com/research/point-spreads-polymarket) found large basketball bets laying the points ahead and taking them behind, in a sample that mixes NBA with a small amount of WNBA and college basketball. Restricted to NBA alone and to spreads specifically, this sample's edge sits close to flat instead; the two are not the same cut of the data, so treat this as a narrower, NBA-only check rather than a contradiction.

Pre-tip buys, 31,947 of them at an average implied probability of 55.1%, came in minus 0.25 points against the price. In-play buys, 26,380 of them at 63.5% implied, came in plus 0.31. Both are close to flat, unlike the cross-sport sharp-money study, where the gap between graded cohorts opens almost entirely after kickoff; here there is no overall gap for timing to open.

Polymarket listed 119 markets on this single Finals game. The moneyline alone carried $18.77 million in volume. Around it sat 41 spread lines ($3.67 million) and 25 total lines ($1.11 million), first-half moneyline, spread and total markets, an odd/even points market, a first-team-to-score market, and 30 player-prop markets across points, rebounds and assists.

That single game illustrates the scale behind the aggregate counts above: an NBA game is not one market but a family of them, and the moneyline usually carries most of the volume.

It does not test 0xinsider's wallet grade against NBA pricing specifically. Grade coverage widened in June 2026, and NBA's own trading has mostly stopped by then, so the graded subset from June 1 to July 24, 2026 holds only 196 S/A/B trades on 72 wallets and 380 D/F trades on 122 wallets, too few to draw a cohort edge from. The whole-sports sharp-money study at 0xinsider.com/research/sharp-money-polymarket-sports answers the grade question at a size that supports it; this study does not repeat that test for NBA alone.

It also does not separate playoffs from the regular season, or split by team, and it does not cover the WNBA or college basketball, which Polymarket lists under other categories. A fuller wallet census is out of scope here; the ten-wallet concentration check below is the only look this study takes at who is trading.

Ten wallets carried 22.8% of the notional in this sample. That is concentrated but not dominant: the other 5,039 wallets carried the rest, and the flat overall edge is not one large account's result.

Every figure above comes from one read-only run against the production database as `posthog_readonly` at 16:45 UTC on September 14, 2026. The query text, its raw output, the market-level export, and the clustered bootstrap script with its output are committed at docs/research-articles/verification/2026-09-14-sports-nba-pricing.sql.txt and the files beside it; the single-game breakdown for Knicks vs. Spurs is its own direct query on `markets.event_slug`, committed separately at 2026-09-14-sports-nba-example-game.sql.txt in the same directory.

Expect the counts to grow only slowly until the next NBA season resumes; the board itself stays withdrawn from discovery until games are on the schedule again, tracked separately from this historical study.

Every NBA game, once the season is running, lists its moneyline, spread and total families on its event page with each family's own volume, and the Top holders tab lists the wallets holding each side with the grade each has earned. The whale alerts feed at 0xinsider.com/whale-alerts shows every large trade as it fills, marked BUY or SELL.

The guide at 0xinsider.com/learn/how-to-bet-on-nba-polymarket covers every NBA market Polymarket lists and how each one settles. Since this study finds no consistent mispricing by price or market type, use it as a baseline: a price that moves against these patterns is the one worth a second look.