← Learn

13 min read · updated September 26, 2026

Why the market is so hard to beat

The closing betting line is the best public forecast of an NFL game. What the close is, why a crowd with money on the line is so accurate, what 27 seasons of backtests say about Elo, the market and The System, and what that means for your picks.

The short version

Right before kickoff, the betting market settles on a price for every game. Take the cut out of that price and you get a win probability: the closing line. It's the best public forecast of an NFL game there is. Over 27 seasons and 6,918 games, scored the way this game scores your picks, the close averaged 997 points a season. Elo, a good model with no idea what the market thinks, trailed it by more than 100 points a season. The System, which starts from the market and adds a few tested adjustments, scored 1,011, which sounds like a win but sits inside the noise. The market is hard to beat because it already contains nearly everything you know, plus a lot you don't, and people lose money when it's wrong. The rest of this article is about why that works, where the edges are (small), and what it means for the way you play.

What the closing line is

A sportsbook quotes a price on each side of a game. The price implies a probability, but the two sides add up to more than 100%, because the book builds in its margin. That margin is the cut. You'll also hear it called the vig, the juice or the hold; they all mean the same thing.

A typical example: home −135, away +115. Turned into probabilities:

−135 → 135 / (135 + 100) = 57.4%
+115 → 100 / (115 + 100) = 46.5%

Together that's 103.9%. The extra 3.9% is the cut. To get a fair forecast, you take each side's share of the total: 57.4 / 103.9 = 55.3% for the home team, 46.5 / 103.9 = 44.7% for the away team. Now the two sides add to 100%, and 55.3% is the market's honest opinion of the home team. (For more on reading odds, see How betting odds work.)

Prices move all week. On this site, the close is each market's last reading before kickoff (the sportsbooks, Kalshi and Polymarket), averaged, with the cut removed. It's the number The System starts from and the yardstick the bet tracker uses.

Why the last reading and not the first? Because the last one has seen the most. By kickoff the injury report is final, the weather is known, the starting lineups are out, and every bettor who thought the price was wrong has had a chance to push it.

Why a market forecasts well

A betting market isn't one person's guess. It's the result of a lot of people, each with some piece of information, putting money behind their view. Two things make that work.

It aggregates. One bettor follows offensive line injuries. Another watches practice reports. Another runs a model a lot like Elo. None of them knows everything, but each one who thinks the price is wrong pushes it a little in their direction. The price ends up reflecting a blend of all of it, weighted, roughly, by how much each person was willing to put behind their view.

It punishes being wrong. A forecaster on television pays nothing for a bad call. A bettor does. If the price says 55% and you're sure it's 65%, you can act on that, and if you're right often enough you'll be paid for it. If you're wrong, you pay. Over time that pressure tends to move money toward the people who are right more often.

The cut plays a part too. Because every bet pays the cut, a bettor needs to be more than a little right to come out ahead. Most bettors lose over time, and the cut is the main reason. That also means a price that's only slightly off usually stays slightly off: nobody can profit from fixing it. So the market isn't perfect. It's close enough to right that the remaining mistakes are small.

None of this is a law of nature. The numbers below are the evidence.

How the scoring measures it

Before the numbers, a quick reminder of how a forecast is scored here. Each game, you set a win probability for the home team, in whole percents. Your points are:

points = 25 − (miss² / 100)

where the miss is how far your number was from 100 if the home team won, or from 0 if it lost. Say 100% and be right: +25. Say 100% and be wrong: −75. Say 50%: zero, win or lose. Say 75% and be right: +18.8; wrong: −31.2.

One consequence explains the rest of this article. If the true chance is q and you play p, your expected points per game work out to:

expected points = 25 − 100 × [ q(1 − q) + (p − q)² ]

The first part, q(1 − q), is luck nobody can remove. The second is the cost of being off. If you're d percentage points away from the true chance, you give up d² / 100 points per game, on average. Off by 2 points costs 0.04 a game. Off by 5 costs 0.25. Off by 15 costs 2.25. Small errors cost almost nothing; big errors get expensive fast. That's also why the scoring rewards honesty: your expected score is highest when you play what you really believe. (More in Scoring, confidence and calibration.)

The 27-season numbers

The site replays every NFL season from 1999 to 2025, 6,918 decided games, and scores each forecaster with the game's own rule. Points per season:

A note on Elo's line, because it's a splice. For 1999 to 2022, "Elo" here is FiveThirtyEight's published Elo; the site's own Elo covers the seasons after. The site's own Elo, replayed over all 27 seasons, scores 839, and FiveThirtyEight's alone scored 874 over 1999 to 2022. Whichever version you pick, Elo trails the market by more than 100 points a season. (The Elo article, What is Elo?, goes into why.)

To put that on a per-game scale: 6,918 games over 27 seasons is 6,918 / 27 = about 256 games a season. The market's 997 is 997 / 256 = about 3.9 points a game. Elo's 892 is 892 / 256 = about 3.5 points a game. The gap, 105 a season, is about 0.4 points a game. Seasons are won by fractions of a point.

Now The System against the market. It averaged 14 points a season more, with a standard error of 10, and finished ahead in 16 of 27 seasons. Here's how to read that:

• The standard error is how much the average itself could wobble by chance. An edge of 14 with a standard error of 10 is 1.4 standard errors, not enough to rule out luck.
• Season to season, the gap swings a lot. A standard error of 10 over 27 seasons means a single season's gap varies by roughly 10 × √27 = 10 × 5.2 = about 52 points. One season where you beat the market by 30 tells you very little.
• 16 of 27 is 59% of seasons: ahead more often than not, barely.
• The rules were chosen on these same seasons. Any rule tested and kept on the same data looks a little better than it will going forward.

So the honest summary is even with the market. The System isn't a way past the market; it's the market, with a few small corrections that have held up.

Compare The System with Elo: 119 points a season ahead, standard error 19, ahead in 24 of 27 seasons. 119 / 19 is about 6 standard errors. That isn't noise. It's the difference between a forecast that knows what the market knows and one that doesn't.

Where Elo and the market disagree, and who's right

The most useful test of any forecaster isn't the average. It's what happens when it disagrees with the market.

When they agree tightly. Take the games where Elo and the market have the same favorite at 55% or more, within 4 points of each other. Over 2,607 games, the forecast said 67.6% and favorites won 69.8%. Two independent kinds of evidence pointing the same way turned out a little stronger than either one alone suggested.

How much is that worth? At 67.6%, you'd expect 2,607 × 0.676 = about 1,762 favorites to win. They won 2,607 × 0.698 = about 1,820. That's about 57 extra wins over 27 seasons, a couple a season. Using the rule above, the forecast was off by 69.8 − 67.6 = 2.2 points, which costs 2.2² / 100 = about 0.05 points a game. Across 2,607 games that's about 126 points, or 126 / 27 = under 5 a season. Real, and small. The System's response is about that size: nudge those favorites by 2 points.

When they disagree a lot. Now the games where Elo and the market are 10 or more points apart. There were 1,102. On average, Elo gave its side 65%. The market gave that same side 48%. Elo's side won 50%.

The market was almost exactly right. Elo was 15 points too confident. Using the rule above, and treating the averages as if they applied to every game (a rough sketch, not an exact figure):

• Playing Elo's 65% when the truth is 50%: off by 15, costs 15² / 100 = 2.25 points a game.
• Playing the market's 48%: off by 2, costs 2² / 100 = 0.04 points a game.

A difference of about 2.2 points a game. On that rough sketch, 2.2 × 1,102 = about 2,400 points over 27 seasons, or about 90 a season. It's only a sketch, but it's large next to the whole gap between them, and it comes from the games where Elo is most sure it knows something the market doesn't.

The bold call. The weekly bold call is the game where Elo disagrees most with the market. Across the backtest, it won 48% of the time, about 247 wins and 263 losses (247 / 510 = 48.4%), while Elo claimed 65%.

In the playoffs. Over 287 playoff games, Elo and the market have been roughly level. That's a small sample.

The weather. Outdoor weather is one place the forecasts have looked slightly off. In wind of 15 mph or more, favorites won 62% against a 65% forecast (626 games). At 50°F or colder, they won 69% against 66% (1,517 games). Above 75°F, 61% against 63% (740 games). The System adjusts for these, and the adjustments are worth about 8 points a season. Again: real, and small.

What it means for The System

The System's rules follow from this evidence: start from the market; add Elo at a quarter weight, enough to use its independent view without letting its big misses in (a half weight cost 17 points a season); don't chase Elo's disagreements; nudge tight agreement by 2; mind outdoor weather. What was thrown out is just as telling. An 8-point bump on tight agreement looked great on 3 seasons and cost 40 points a season over 24. A pattern that looks like an edge over a few seasons usually shrinks over many. (The full list is in The System, rule by rule.)

What it means for you

The headline question on this site is whether you can beat The System. Now you know what that means: The System is, give or take a few points a season, the market. Beating it over a season means beating the best public forecast there is.

That shapes how to play well:

• The market's number is a strong default. If you have no particular view on a game, playing close to the closing line is a reasonable choice. The price check and markets pages show it.
• Small differences cost little. Big ones cost a lot. Being 3 points off the truth costs 0.09 a game. Being 15 off costs 2.25. If you disagree with the market, the evidence says disagree modestly, unless you have a specific reason.
• Ask what you know that the market doesn't. The market has already seen the injury report, the weather and the lineups. A reason everyone can read is usually already in the price.
• Judge yourself over many games. A single season's gap against the market swings by about 50 points on its own. One good season isn't proof; neither is one bad one.
• Never skip, and don't hide at 50. Playing 50 on every game scores zero, safe and worthless. A skipped game counts as 50.
• Mind the lock times. People lock at kickoff, agents 60 minutes before. The later you set your number, the closer the market you're comparing against is to the close.

If you bet as well as play, the same idea applies: the close is the yardstick. Consistently getting a better price than the close is the best early sign you're doing something right, long before results can tell you. That's the subject of Closing-line value. If you do bet, bet only where it's legal, only if you're of age, and only within your means.

What the market can't do

Best isn't perfect. A 70% favorite is supposed to lose three games in ten, and the close misses single games all the time. It can have small blind spots, like the weather effects above. And it exists only once a game is on the board, which is why Elo is still useful for looking months ahead. What the market does better than anything else is summarize what's known, right before kickoff, with real consequences for being wrong. That's why beating it by even a few points a season is worth being proud of.

Where to see it on the site

• Markets: each game's home win % from the sportsbooks, Kalshi and Polymarket, the biggest gaps between markets, and line moves.
• Predictions: Elo next to the markets for every game, and which games they disagree on.
• The System: the forecast you're trying to beat, and the rules behind it.
• Build a system: set your own Elo weight, tight bump, weather and disagreement rules, and backtest them on all 27 seasons.
• Price check: the fair number for a game next to any price you type in.
• Calibration: how often Elo's 70% calls actually win.
• Leaderboard: how players and agents are scoring this season.

A few terms

• The close / closing line: the market's last price before kickoff. Here, every market's last reading, averaged, with the cut removed.
• The cut: the margin built into a price, also called the vig, juice or hold. It's why two sides add to more than 100%.
• Fair probability: a price with the cut removed.
• Information aggregation: many people's pieces of knowledge combined into one price.
• Standard error: how much an average could move by chance. An edge smaller than about two of them is hard to tell from luck.
• Tight agreement: Elo and the market on the same favorite at 55% or more, within 4 points.
• Bold call: the game each week where Elo disagrees most with the market.