Arbitrage vs Middling in 2026: Which Bet Pays and Which One Can Lose

The quick answer

• Arbitrage: you cover every outcome at prices whose implied probabilities add up to less than 100%, so the result is a small profit whatever happens.

• Middle: you hold two sides of a game at different lines. Land in the gap and both legs win. Miss it and one leg wins, one loses, and you're out roughly the vig.

• Size of the payoff: the arb below returns 1.18% of stake. A middle that hits pays close to a full bet, but the NFL totals middle below missed in 91% of comparable games.

• How to judge each: arbs are judged with arithmetic, middles with expected value.

• Where the losing side costs least: a pre-game straight trade on Novig, which charges no fee to maker or taker, so a missed middle costs less there than on a pair of -110 lines.

An arbitrage and a middle both start with two venues disagreeing. The arb ends as a closed position with a known result, while a middle is a cheap ticket on a narrow band of final scores, worth buying only if games land in that band often enough.

What an arbitrage bet is

You back every possible outcome of one market at prices good enough that the combined return beats the combined stake. Convert each price to decimal odds, add up 1 divided by each, and call the total S. If S is under 1, there's an arb before fees.

Worked example. Team A is +110 at one sportsbook (decimal 2.10). Team B is -105 at another (decimal 1.9524).

• Implied probabilities: 1 / 2.10 = 47.62% and 1 / 1.9524 = 51.22%. S = 0.9884.

• Stakes on $1,000: $481.79 on Team A and $518.21 on Team B. Total $1,000.00.

• If Team A wins: $481.79 x 2.10 = $1,011.76 back. Profit $11.76.

• If Team B wins: $518.21 x 1.9524 = $1,011.74 back. Profit $11.74.

That's a 1.18% return whoever wins. The only risks are execution and settlement: a leg that doesn't fill, a price that moves before the second bet, or two venues that settle the same event differently.

What a middle is

A middle uses two different lines on the same game, classically a total that moved during the week. You took Over 44.5 early at one book, the market climbed, and another book now offers Under 47.5. Any final total of 45, 46 or 47 wins both bets.

Worked example, both legs at -110 (decimal 1.9091), $110 on each:

The miss costs $10, the vig on one -110 bet, and the hit pays $200. Break-even is the probability p where 200p = 10(1 - p), which works out to 4.76%. If games land in the window more often than that, the middle has positive expected value; less often, and it's negative.

How often does a middle hit?

We checked this against the nflverse NFL games file, which records the final score and closing line for 7,340 completed NFL games from the 1999 season through 5 October 2026.

Of the 745 games that closed with a total between 45.5 and 46.5, 64 finished on 45, 46 or 47, which is 8.59%. For the middle above:

• Hit: 8.59% x $200 = +$17.18.

• Miss: 91.41% x -$10 = -$9.14.

• Expected value: +$8.04 per $220 staked, or 3.65%.

That figure assumes you got both lines at -110 three points apart, which only happens when the market has moved, and it's an average across 28 seasons of scoring environments. A one-point middle (Over 45.5, Under 46.5) only wins on exactly 46, and just 19 of those 745 games landed there, 2.55%. At -110 on both sides that's well below the 4.76% break-even.

Key numbers: why 3 and 7 matter on spreads

NFL scoring comes in threes and sevens, so final margins cluster. In the same 7,340 games:

• Decided by exactly 3 points: 1,110 games, 15.1%.

• Decided by exactly 7: 663 games, 9.0%.

• Next most common margin, 6: 441 games, 6.0%.

Across a key number, a spread middle can clear break-even. With the favorite at -2.5 at one venue and the underdog at +3.5 at another, both at -110, a favorite win by exactly 3 wins both legs. Of the 2,395 games that closed with the favorite laying 2.5 to 3.5, the favorite won by exactly 3 in 216, or 9.0%. That's nearly double the 4.76% break-even. A middle across 7 is thinner: in the 1,132 games closing at 6.5 to 7.5, the favorite won by exactly 7 in 73, or 6.45%.

How a broken arb turns into a middle

An arb can break between the first leg and the second. You take Over 46.5 at +105 for $100, intending to take Under 46.5 at +100 somewhere else. Before you click, the second book moves its number.

If it moves up to Under 47.5 at -110 and you bet $100 there, you've got an accidental middle:

• 46 or fewer: Under returns $190.91, Over loses. Net -$9.09.

• Exactly 47: both win, $205.00 + $190.91. Net +$195.91.

• 48 or more: Over returns $205.00, Under loses. Net +$5.00.

A move down to Under 45.5 gives you the reverse: a final total of exactly 46 loses both bets, $200. Neither replacement carries the arb's guarantee. Before placing a second leg at a new number, write down what each final score does to the position.

Why middles are judged on expected value

Arb math only asks whether S comes in under 1. Two -110 legs put S at 1.0476, so by arb math a middle is always a losing bet. Some middles still carry positive expected value, because the three outcomes have different probabilities and the window can be priced.

The probability of landing in the window decides it, and that number comes from history or a model; the prices can't supply it. A single bet at 2.10 works the same way: a 50% chance makes it a +5% bet and a 46% chance a -3.4% bet, with only the probability estimate changing. A middle runs that calculation over three outcomes instead of two.

What the losing side costs on a prediction market

The cost of a middle is what you lose when it misses, and that's the fee or vig on the losing leg. Run the totals middle again with the Under 47.5 bought on Novig instead of at a second -110 book. Novig charges no fee to maker or taker on a pre-game straight trade, so at an illustrative price of 50c (decimal 2.00), $110 buys 220 contracts that pay $220 if the Under wins.

The Kalshi column uses Kalshi's published taker formula, 0.07 x contracts x price x (1 - price), which comes to $3.85 on 220 contracts at 50c and applies whichever way the game goes. At the 8.59% hit rate from the NFL data, with 339 of the 745 games finishing at 44 or fewer and 342 at 48 or more, the expected value per position is +$8.04 with both legs at -110, +$9.60 with the Under on Kalshi and +$13.45 with the Under on Novig.

Once the game starts, Novig takers pay a 0.03 coefficient on live straights, which is $0.75 per 100 contracts at even money, so the cheapest version of this trade is the one placed before kickoff.

What to look for

• Line gap: a middle needs at least two points of separation on a total, or a spread that straddles 3 or 7.

• Hit rate: check the window against historical data before trusting a scanner's label.

• Leg cost: the fee on the losing leg is the whole price of the trade, so look for a venue with no fee on pre-game straights and a published fee formula for everything else.

• Settlement match: both legs must grade on the same period, including overtime, or a middle can lose both sides.

Frequently asked questions

Where do I find good arbitrage bets?

Arbs appear where two venues price the same market independently and update at different speeds, such as a sportsbook and a prediction market quoting the same game. Scanners flag them, but each one still needs the stakes and settlement rules checked by hand.

Is middling the same as arbitrage?

No. An arb profits on every outcome, while a middle profits big on a narrow set of outcomes and loses a small amount on the rest.

Is middling profitable?

Only when the window hits more often than its break-even rate, which is 4.76% for a middle built from two -110 legs. Across 7,340 NFL games, favorites laying 2.5 to 3.5 won by exactly 3 in 9.0% of games.

What happens if a middle misses?

One leg wins and one loses, and you're out the vig on the losing leg: $10 on two $110 bets at -110. With one leg on a pre-game Novig trade, which carries no fee, one of the two miss outcomes costs nothing.

What are key numbers in NFL betting?

They're the margins games end on most often. In NFL games since 1999, 15.1% were decided by exactly 3 points and 9.0% by exactly 7.

Sources

All checked 7 October 2026.

• NFL scores, margins and closing lines: nflverse nfldata games file, github.com/nflverse/nfldata (7,340 completed games, 1999 season to 5 October 2026), analysis by the author

• Novig fee schedule: support.novig.com/en/articles/16195057-fees-on-novig

• Novig CFTC designation: cftc.gov, Ludlow Exchange, LLC, filing 59390, designated 16 June 2026

• Kalshi fee schedule: kalshi.com/docs/kalshi-fee-schedule.pdf, last updated and effective 7 July 2026