Timestamped odds snapshots from 50 sportsbooks across 22 sports. Reconstruct what prices were available at any point in time. Validate your model before risking capital.
22
Sports archived
50
Books per snapshot
380K+
Odds rows collected
Historical data available on Business plan ($99/mo)
{
"sport_key": "basketball_nba",
"event": "Knicks vs Heat",
"commence_time": "2026-04-28T23:30:00Z",
"bookmakers": [
{
"key": "pinnacle",
"markets": [{
"key": "h2h",
"outcomes": [
{"name": "Knicks", "price": -135},
{"name": "Heat", "price": +122}
]
}]
},
{
"key": "draftkings",
"markets": [{
"key": "h2h",
"outcomes": [
{"name": "Knicks", "price": -140},
{"name": "Heat", "price": +120}
]
}]
}
],
"snapshot_time": "2026-04-28T20:15:00Z"
}Problem
Opening lines move 2–5 points before game time as sharp money arrives. Backtesting against openers overstates your edge because those prices were gone before your model triggered.
Problem
Testing against one book's history overfits to that book's pricing patterns. A model that "works" on DraftKings data may fail entirely on the broader market. Multi-book data exposes this immediately.
Solution
TheOddsAPI archives odds with timestamps from 50 books. Simulate execution at the price that was actually available when your model signaled. Realistic backtests, realistic expectations.
The single strongest predictor of long-term profitability. If you consistently beat the closing line, you have an edge. Period.
Example: NBA — Knicks vs Heat
Your Bet (3hrs pre-game)
Knicks -130
Implied: 56.5%
Closing Line
Knicks -145
Implied: 59.2%
Your CLV
+2.7%
You beat the close
You bet Knicks -130 three hours before game time. The line closed at -145. That means the market agreed your bet was underpriced — you got 2.7% more value than the final efficient price. Track this over 1,000+ bets. Positive CLV = edge confirmed. Negative CLV = your model is behind the market.
Multi-bookmaker historical odds. Not a single-source feed.
| Dimension | Coverage |
|---|---|
| Sports | 22 (NBA, MLB, NHL, NFL, EPL, La Liga, Bundesliga, Serie A, Champions League, and 13 more) |
| Markets | h2h (moneyline), spreads, totals |
| Sportsbooks | 50 per snapshot (DraftKings, FanDuel, Pinnacle, Betfair, BetMGM, and 45 more) |
| Data start | April 16, 2026 (expanding daily) |
| Snapshot frequency | Timestamped with each odds refresh |
| Credit cost | Zero additional — included in Business tier |
Pull historical snapshots and calculate CLV in a few lines.
import requests
# Pull historical odds for NBA
response = requests.get(
"https://api.theoddsapi.com/v4/historical/odds",
headers={"x-api-key": "YOUR_KEY"},
params={
"sport_key": "basketball_nba",
"date": "2026-04-28",
"bookmakers": "pinnacle,draftkings,fanduel"
}
)
events = response.json()["data"]
# Calculate CLV for each event
for event in events:
# Compare your bet price vs closing snapshot
my_bet_price = get_my_bet(event["event_id"]) # your records
closing_price = event["bookmakers"][0]["markets"][0]["outcomes"][0]["price"]
# Positive CLV = you beat the market
clv = implied_prob(closing_price) - implied_prob(my_bet_price)
print(f"{event['event']}: CLV {clv:+.1%}")const res = await fetch(
"https://api.theoddsapi.com/v4/historical/odds?sport_key=basketball_nba&date=2026-04-28",
{ headers: { "x-api-key": "YOUR_KEY" } }
);
const { data: events } = await res.json();
// Reconstruct what odds were available at any timestamp
events.forEach(event => {
const pinnacle = event.bookmakers.find(b => b.key === "pinnacle");
const dk = event.bookmakers.find(b => b.key === "draftkings");
console.log(`${event.event}: Pinnacle ${pinnacle?.markets[0].outcomes[0].price} vs DK ${dk?.markets[0].outcomes[0].price}`);
});A rigorous backtest answers one question: would your model have made money if executed at prices that were actually available?
ML model, consensus fade, sharp-follows, or any system that generates a bet decision with a timestamp. The timestamp is critical — it determines which historical snapshot to compare against.
Query /historical/odds for the snapshot closest to when
your model would have triggered. This gives you the actual price
available — not an opening or closing approximation.
For each signal, record the best available price across 50 books at that timestamp. Apply your staking strategy (flat, Kelly, fractional Kelly). Account for vig — don't backtest against mid-market prices you can't actually get.
Compare your execution price against the closing line. Calculate CLV across all bets. Positive CLV over 500+ bets = confirmed edge. Track ROI, drawdown, and Sharpe ratio to determine if it's capital-deployable.
Split your data. Train on one period, test on another. If CLV disappears out-of-sample, your model overfit the training data. This is where most "winning" systems fail.
Only testing on events that completed normally. Cancelled games, postponements, and voided markets get excluded — inflating your results. Include all events in your sample.
Your model signals at T=0, but you execute at T=30s. In that window, the line may have moved. Use the snapshot closest to your realistic execution time, not the signal time.
Testing against mid-market or vig-removed prices that you can't actually bet. Always backtest against the real offered price from specific books. Multi-book data lets you find the best available line — that's your realistic execution price.
Running 50 model variations and picking the one that worked best. Each parameter you tune increases overfitting risk. Out-of-sample testing on held-out data is mandatory, not optional.
Compare every bet placed against the closing line across all books. Automated CLV calculation tells you if your edge is real or if you're just getting lucky in small samples.
Before deploying a new model live, backtest against 50-book historical data. If it shows +CLV against Pinnacle closing lines over 500+ events, deploy. If not, iterate without burning capital.
Reconstruct historical cross-book disagreements to measure how often arbs appeared, how long they lasted, and what profit range was realistic. Size your arb operation based on empirical data, not assumptions.
Study how quickly soft books converge to Pinnacle's line. Measure the half-life of edges by sport, market, and time-to-kickoff. Build timing models that tell you when to bet for maximum CLV.
Historical odds are available on the Business plan ($99/mo). Zero additional credit cost. 22 sports. 50 books per snapshot.