The closing line assigned the average home team 2.20 points from 1999 through 2025. The home team’s actual average margin was 2.28 points. The difference was only 0.07 points per game. Over 6,901 non-neutral regular-season games, the market’s home-side price and the final home margin nearly met.
The more useful change is across time. The average closing home spread was 2.50 points in 1999–2008, 2.37 in 2009–2018, and 1.58 in 2019–2025. That is a descriptive fall in the number attached to the home side. It is not proof that every stadium lost the same amount of value. The spread also prices team strength, so this post stays with what the closing line actually said.
What “home-field edge” means here
This article uses three separate quantities. The closing home spread is spread_line. Positive means the home team was favored. The actual home margin is home score minus away score. The home ATS margin is actual home margin minus closing home spread. Keeping those columns separate prevents one number from borrowing another number’s meaning.
When actual home margin exceeds the closing home spread, the average ATS gap is positive; when it falls short, the gap is negative. Neither quantity is a causal venue coefficient. Better teams may have hosted more often in the selected games, and the close also absorbed roster and matchup information. A causal estimate needs a different design.
The long-run closing-line ledger
Across the full sample, home teams went 3,286-3,426-189 ATS. That is a 49.0% cover rate over non-push decisions. The average home ATS margin was slightly positive, while the cover count was slightly negative. There is no contradiction. A few large covers can lift the average margin without changing how many individual games covered.
The average closing home spread peaked at 2.88 points in 2006. Home teams won by only 0.85 points on average that season, missing the closing number by 2.03 points per game. The lowest average closing home spread was 1.08 in 2020. Home teams won by 0.14 points on average and missed the close by 0.94.
The strongest actual average home margin was 3.63 points in 2003, against an average close of 2.49. The weakest was -0.05 points in 2019, meaning the average home team was outscored slightly, while the market still priced the home side at +1.78. These are season summaries, not venue ratings.
The recent seven seasons
| Season | Avg close | Avg result | Avg ATS gap | Home ATS |
|---|---|---|---|---|
| 2019 | +1.78 | -0.05 | -1.83 | 104-137-10 |
| 2020 | +1.08 | +0.14 | -0.94 | 126-127-0 |
| 2021 | +1.72 | +1.57 | -0.16 | 126-139-4 |
| 2022 | +1.67 | +2.12 | +0.45 | 126-129-10 |
| 2023 | +1.68 | +2.71 | +1.04 | 126-127-14 |
| 2024 | +1.51 | +1.72 | +0.21 | 132-131-4 |
| 2025 | +1.58 | +2.18 | +0.61 | 131-133-1 |
The recent table shows why “home field is worth X” is too tidy. The average close stayed between 1.51 and 1.78 in six of the seven seasons, apart from 2020 at 1.08. The actual result moved from -0.05 to +2.71. A static home number cannot explain that swing by itself.
It also shows why ATS record and average margin should travel together. In 2023 the home side beat the close by 1.04 points per game on average, yet went 126-127-14 ATS. The average was carried by the size of results, not by a majority of covers. A chart that showed only the average gap would hide the losing decision count.
What the market “paid” by era
The average closing home spread declined by 0.92 points from the first era to the last: 2.50 in 1999–2008 versus 1.58 in 2019–2025. The actual average home margin declined by 1.11 points over the same era split: 2.62 to 1.51. The price and result moved in the same direction at this broad level.
The middle era, 2009–2018, sat between them. Its average home close was 2.37 and its actual margin was 2.49. Across all three eras, the close was close to the observed average result. That is exactly what an efficient closing market should be judged on descriptively: not whether one side won every year, but whether systematic misses remained after the price was set.
This study does not estimate the value of travel, crowd, rest, or venue separately. Those fields would need an explicit model and out-of-sample test. Calling the average spread a pure “home-field adjustment” would make a team-strength mixture look like a controlled experiment. We are not doing that.
What the Number Says
The closing number says the market attached less average spread to the home side in the recent era than it did in the first decade of the file. The result says home margins also declined. The ATS ledger says blindly taking the home team still produced fewer covers than losses over the full period.
For a current game, the useful question is not “what is home field worth in general?” It is “what does this close charge for the home side, and what evidence says the price is wrong?” The Analytics surface is the right place to compare a declared slice. The NFL picks page is where current graded picks should carry their own record. Historical averages are context, not permission.
Three checks before using a home split
First, exclude neutral sites. This article removed 66 neutral-site regular-season games because a team labeled “home” at a neutral venue does not answer the same question. Second, name the line convention. In this snapshot positive spread means the home team is favored; other app tables use the opposite sign, so a silent join can reverse every conclusion. Third, retain pushes. A win by exactly the closing spread is not a cover.
Then ask whether the split survives a narrower, predeclared test. Venue, rest, distance, and team quality can all matter. Adding those filters after seeing the result is how a clean 6,901-game ledger becomes a fragile story. Start broad, register the next question, and test it on games that did not choose the question.
Why the average is not team-neutral
Every game contributes one home-side close, but the mix of teams is not held constant. A season with more strong teams hosting as favorites can raise the average even if the venue effect is unchanged. A season with more home underdogs can lower it. The mean therefore describes the scheduled market mix as well as any broad home premium.
A causal study would need to control team quality and other pregame information, then test its home coefficient outside the fitting sample. This article deliberately does less. It reports the stored market price, the observed margin, and the residual. That smaller claim is reproducible from the snapshot without pretending the schedule was randomized.
The same restraint should govern current displays. “Home edge” is too vague for a chip. “Home side closed as the favorite” or “home team beat the close” names the observed market fact. Any modeled home adjustment belongs in a separately labeled model output with its own version and record.
Where these numbers come from
How we counted: We joined the four CSV shards under data/training-snapshots/nfl/, kept played regular-season games from 1999–2025, and required location = Home. That left 6,901 games and excluded 66 neutral-site rows. The dictionary defines result as home score minus away score and spread_line as the closing spread with positive values meaning the home team is favored. Home ATS margin is result - spread_line. Season and era averages are arithmetic means of those game rows. Cover rate excludes pushes from its denominator. Shard hashes and the 7,548-row snapshot identity are recorded in manifest.json.
Model calibration: predicted vs observed
Predicted win probability bucket vs the empirical win rate inside that bucket on the test set. Points on the y=x reference line are perfectly calibrated; points below mean the model is overconfident in that bucket.
NFL ATS cover-margin distribution
Distribution of (final margin − closing spread) across an NFL season. Roughly normal with mean ≈ 0 and standard deviation ≈ 13 points, which is why most ATS edges live in the ±1.5 point window.
