Recipe · Apify Store · runs in a terminal
What 1,553 sportsbook quotes say about bookmaker margins
BetMGM had the lowest average overround in all three NFL markets in this sample. NBA leaders varied, and one game total differed by 1.5 points across books.
BetMGM’s New Jersey feed had the lowest average bookmaker margin in all three NFL markets in this sample. In the NBA, three different feeds led the three markets. For the same Knicks–Raptors game, one book posted a total of 216 and another posted 217.5 in the same capture. Those two quotes would settle differently on a combined score of 217.
I used version 1 of the public NFL and NBA Sportsbook Odds dataset. The archive contains 1,553 two-sided, full-game quotes across 33 games and 20 collection timestamps from November 11 to 17, 2026 UTC. I kept quotes marked scheduled and collected before the listed start, and excluded feeds marked as mirrors. For each game and market, I used the capture with the most distinct books, breaking ties in favor of the latest capture. Requiring at least two books left 235 quotes from 44 game-market snapshots across 17 games. Kaggle dataset · automatic/actionnetwork-odds
Which books had the lowest overround?
Overround is the sum of the two prices’ implied probabilities minus 100%. At −110 on both sides, each side implies 52.38%, for an overround of 4.76%. For positive American odds +A, implied probability is 100 / (A + 100). For negative odds −A, it is A / (A + 100), where A is the positive magnitude. Overround measures the margin built into the quoted pair of prices. Realized sportsbook profit depends on stakes, payouts and outcomes.
| League | Market | Lowest mean (quotes) | Second lowest (quotes) | Third lowest (quotes) |
|---|---|---|---|---|
| NFL | Moneyline | BetMGM 4.009% (6) | FanDuel 4.011% (6) | Caesars 4.241% (6) |
| NFL | Point spread | BetMGM 4.411% (6) | FanDuel 4.645% (6) | DraftKings 4.678% (6) |
| NFL | Game total | BetMGM 4.380% (6) | FanDuel 4.663% (6) | bet365 4.720% (6) |
| NBA | Moneyline | FanDuel 4.074% (10) | DraftKings 4.136% (10) | Caesars 4.395% (4) |
| NBA | Point spread | DraftKings 4.709% (8) | BetMGM 4.735% (4) | FanDuel 4.741% (8) |
| NBA | Game total | Unibet 4.404% (4) | BetMGM 4.735% (4) | DraftKings 4.753% (8) |
All feeds in the table carry New Jersey labels. I averaged the overrounds for each book, giving each game it quoted in that market equal weight, and ranked the unrounded means. BetMGM and FanDuel were nearly tied on NFL moneylines. In NBA totals, Unibet’s average came from four quotes and DraftKings’ from eight. Books can cover different games, so these rankings describe the observed quotes.
How far apart were the lines?
For the point spread, I compared the handicap listed for the home team. For totals, I compared the combined-points threshold. Each range is the highest line minus the lowest for the same game, market and capture. The medians include games where the books agreed.
| League | Market | Games compared | Games with differing lines | Median range (points) | Widest range (points) |
|---|---|---|---|---|---|
| NFL | Point spread | 6 | 4 (66.7%) | 0.5 | 1 |
| NFL | Game total | 6 | 4 (66.7%) | 0.5 | 1 |
| NBA | Point spread | 8 | 3 (37.5%) | 0 | 1 |
| NBA | Game total | 8 | 4 (50.0%) | 0.5 | 1.5 |
The widest gap came from the Knicks at the Raptors. At 03:09:13 UTC on November 17, Caesars showed over 216 at −110 and BetRivers showed over 217.5 at −109. At a hypothetical 217 combined points, the first over would win and the second would lose. The line changed which scores qualified for a payout. The odds changed the payout for a winning ticket.
What line shopping compares
On the Washington Commanders–New York Giants moneyline, bet365’s two-sided overround was 3.79%, compared with 3.96% at BetMGM in the same capture. Yet the Giants were priced at −140 at bet365 and −135 at BetMGM. The BetMGM quote required a smaller stake to return $100 in profit if the Giants won. On the Washington side, bet365 showed +120 and BetMGM showed +115. A book’s overround summarizes the whole quote. What a single ticket pays, and which scores it pays on, depend on the side, odds and any points line.
These are historical feed snapshots. The capture time records when the Actor retrieved a quote. Sportsbooks can update their prices at different times, and live availability can vary. Cypher Odds live comparison tool
Reproduce the tables
Download version 1 from Kaggle and extract the CSV.
curl -fsSL -o odds.zip \
'https://www.kaggle.com/api/v1/datasets/download/kirkins/nfl-nba-sportsbook-odds?datasetVersionNumber=1'
unzip -p odds.zip data.csv > data.csv
Save the Python below as odds.py and run python3 odds.py data.csv. The script uses only the standard library.
import csv
import sys
from collections import defaultdict
from statistics import mean, median
with open(sys.argv[1], encoding="utf-8", newline="") as file:
rows = list(csv.DictReader(file))
snapshots = defaultdict(lambda: defaultdict(dict))
for row in rows:
captured = row["collected_at_utc"]
# The archived UTC timestamps use a uniform ISO format.
if (row["event_status"] != "scheduled"
or captured >= row["event_start_utc"] or row["book_mirror_of"]):
continue
key = row["league"], row["event_id"], row["market"]
snapshots[key][captured][row["book"]] = row
selected = {}
for key, by_time in snapshots.items():
_, books = max(by_time.items(), key=lambda item: (len(item[1]), item[0]))
if len(books) >= 2:
selected[key] = books
games = {(league, event) for league, event, _ in selected}
print("Archive quotes:", len(rows))
print("Selected snapshots, games, quotes:",
len(selected), len(games), sum(map(len, selected.values())))
def implied(odds):
if odds > 0:
return 100 / (odds + 100)
return -odds / (100 - odds)
def overround(row):
sides = ("selection_a_american", "selection_b_american")
return 100 * (sum(implied(int(row[side])) for side in sides) - 1)
def line(row):
return float(row["line"])
margins = defaultdict(lambda: defaultdict(list))
ranges = defaultdict(list)
for (league, _, market), books in selected.items():
quotes = list(books.values())
for row in quotes:
margins[league, market][row["book"]].append(overround(row))
if market != "moneyline":
low = min(quotes, key=line)
high = max(quotes, key=line)
ranges[league, market].append((line(high) - line(low), low, high))
for (league, market), books in sorted(margins.items()):
ranked = sorted(books.items(), key=lambda item: (mean(item[1]), item[0]))
for book, values in ranked[:3]:
print(league, market, book, len(values), f"{mean(values):.3f}%")
for (league, market), comparisons in sorted(ranges.items()):
gaps = [gap for gap, _, _ in comparisons]
different = sum(gap > 0 for gap in gaps)
print(league, market, len(gaps), different, f"{different / len(gaps):.1%}",
median(gaps), max(gaps))
gap, low, high = max((item for group in ranges.values() for item in group),
key=lambda item: item[0])
print("Widest gap:", gap, low["away_team"], "at", low["home_team"],
low["collected_at_utc"])
for row in (low, high):
print(row["book_name"], row["line"], row["selection_a_american"])
for book in ("bet365_nj", "betmgm_nj"):
row = selected["NFL", "290970", "moneyline"][book]
print(row["book_name"], row["selection_a_american"],
row["selection_b_american"], f"{overround(row):.2f}%")
Data source
Every finding uses data.csv from version 1. The CSV’s SHA-256 is a384e612fb5c725589989d077a7bd58004c3c57aa8a7af46375924c80ed7250c. Odds were displayed publicly by sportsbooks via Action Network and compiled by kirkins. The archive’s README documents the fields, collection and coverage. Kaggle dataset
Affiliate disclosure: The Apify Store link carries our referral code. If you sign up for a paid plan through it, ssh.autos may earn a commission at no cost to you.
Responsible gambling: 21+. Gambling problem? Call 1-800-GAMBLER.
Sources in this recipe
-
US Sportsbook Odds: DraftKings, FanDuel, BetMGM, Caesars Lines
Moneyline, spread and total per sportsbook, relayed by Action Network; quotes can be stale or wrong.
Machine-readable: openapi.json · llms.txt