Favorites went 3,298-3,450-189 ATS from 1999 through 2025: a 48.9% win rate over 6,748 decisions. The label “favorite” did not beat the closing spread in the full window.
Played regular-season games, one declared closing-spread convention, seasons 1999–2025. The snapshot contains 6,967 games: 30 closed pick'em, 6,937 had a favorite, and 189 of those graded as pushes.
Every season, with the denominator attached
Each row is one regular season, W-L-P against the closing spread, followed by non-push sample and ATS win rate.
| Season | Favorite ATS | n | Win rate |
|---|---|---|---|
| 1999 | 110-125-9 | 235 | 46.8% |
| 2000 | 112-124-7 | 236 | 47.5% |
| 2001 | 110-124-11 | 234 | 47.0% |
| 2002 | 110-139-5 | 249 | 44.2% |
| 2003 | 123-121-9 | 244 | 50.4% |
| 2004 | 122-124-6 | 246 | 49.6% |
| 2005 | 142-99-10 | 241 | 58.9% |
| 2006 | 108-141-7 | 249 | 43.4% |
| 2007 | 130-121-5 | 251 | 51.8% |
| 2008 | 123-127-6 | 250 | 49.2% |
| 2009 | 121-127-8 | 248 | 48.8% |
| 2010 | 123-128-5 | 251 | 49.0% |
| 2011 | 121-124-11 | 245 | 49.4% |
| 2012 | 119-132-5 | 251 | 47.4% |
| 2013 | 135-115-6 | 250 | 54.0% |
| 2014 | 120-130-6 | 250 | 48.0% |
| 2015 | 114-134-8 | 248 | 46.0% |
| 2016 | 128-122-5 | 250 | 51.2% |
| 2017 | 132-113-8 | 245 | 53.9% |
| 2018 | 114-133-9 | 247 | 46.2% |
| 2019 | 113-133-10 | 246 | 45.9% |
| 2020 | 114-142-0 | 256 | 44.5% |
| 2021 | 125-143-4 | 268 | 46.6% |
| 2022 | 118-143-10 | 261 | 45.2% |
| 2023 | 137-121-14 | 258 | 53.1% |
| 2024 | 143-125-4 | 268 | 53.4% |
| 2025 | 131-140-1 | 271 | 48.3% |
The loudest warning came in consecutive seasons
Fixed annual windows from the same close query.
Favorites went 142-99-10 ATS in 2005, a 58.9% win rate over 241 decisions. One year later they went 108-141-7 ATS, a 43.4% win rate over 249 decisions. The best and worst annual rates in the file arrived back to back.
The rebound was immediate too: 2007 finished 130-121-5 ATS, a 51.8% win rate over 251 decisions. A rule built from the previous season would have chased the exact side that had just stopped working.
The recent run made the same point
Fixed annual windows, 2023–2025.
Favorites went 137-121-14 ATS in 2023, a 53.1% win rate over 258 decisions, then 143-125-4 ATS in 2024, a 53.4% win rate over 268 decisions. In 2025 they fell to 131-140-1 ATS, a 48.3% win rate over 271 decisions.
That sequence does not erase the first two years or crown the third. It shows why “favorites are hot” needs a declared start date, W-L-P, sample, and close source before it gets a vote.
What the long record says
Nineteen of the 27 seasons finished below a 50.0% favorite win rate. Eight finished above it: 2003, 2005, 2007, 2013, 2016, 2017, 2023, and 2024. That list is descriptive, not a signal available before those seasons began.
The chart is ATS, not straight-up. A favorite can win the game and miss the spread. Each point removes pushes from the rate denominator while the table keeps them in the audit trail.
What the record cannot say
The snapshot stores the final spread used for grading. It does not reconstruct when a team became the favorite or what number a bettor saw earlier in the week. Those are line-history questions.
The broad record also combines every favorite size. That answers the title and nothing narrower. A spread-band study would need fixed buckets, W-L-P, window, sample, and an out-of-sample test chosen before the results.
Use it as a baseline, not a system
A current favorite model should compare its selected games with this broad baseline only after naming its own selection rule. It cannot borrow the league-wide 48.9% rate as the expectation for a narrower population.
Use Analytics to inspect records with the market and window named, and the NFL picks page for current graded picks. A future update should append the next completed season to this ledger rather than moving the start date to flatter a new result.
Where these numbers come from
How we counted: These numbers come from a frozen historical snapshot. We concatenated the files matching data/training-snapshots/nfl/games-*.csv, kept played rows with game_type = REG, seasons 1999–2025, and a closing spread_line. Pick'em rows with spread_line = 0 were excluded because they have no favorite. Positive favorite ATS margin is a cover, zero a push, and negative a loss. Win rate is wins divided by wins plus losses. Snapshot identity and shard hashes are in data/training-snapshots/nfl/manifest.json.
NFL ATS cover-margin distribution
Bars count completed NFL schedule rows by closing-spread cover margin using the repository canonical home-margin grading convention.
