Why Sportsbooks Don't Balance Their Books

Almost every guide to sports betting opens with the same premise. Books attract money on both sides of a game. They adjust the spread or line until the action balances. Then they collect the vig and pocket a guaranteed profit. This is the foundational story of how sportsbooks work.

The story is wrong.

In 2004, University of Chicago economist Steven Levitt published a paper in The Economic Journal that tested this assumption directly using a unique dataset of roughly 20,000 National Football League bets. The finding: sportsbooks do not try to balance their books. They take active positions against bettors, exploit systematic biases in how people bet, and profit by being better predictors than their customers. The paper estimated that books earn 20 to 30 percent higher profit margins by doing this versus the balanced-book approach.

That finding changes how you should think about betting markets, line movements, and where any real edge lives in 2026.

The Balanced-Book Myth

The balanced-book model makes intuitive sense. A book sets a line, bettors bet on both sides, the book adjusts until both sides are roughly equal, and the vig covers the payout regardless of which side wins. In this model, the book is a neutral intermediary taking a service fee. Your job as a bettor is to find pricing mistakes in a game before the crowd corrects them.

This model fails on the basic data. Levitt examined roughly 20,000 NFL bets and found that in nearly half of all games, two-thirds of the money wagered went to one side. The popular side lost against the spread more often than it won. Books did not move the line enough to balance the action. They absorbed the lopsided money and profited from the bias.

The most common biases Levitt identified: bettors systematically overback home team favorites and nationally prominent teams. Books know this. They set prices that shade slightly away from true probability on those teams, take the less-popular side, and let the bias produce the profit.

What the Paper Proved

Levitt's core thesis is specific: books are not neutral intermediaries. They are active market participants with their own position on each game. Their edge comes from two sources, not one.

The first is the standard vigorish, the cut built into every bet. The second, and larger, is the return from holding an asymmetric position against bettors whose systematic errors the book has quantified. Books are not primarily in the risk-management business. They are better-informed counterparties betting against the public.

The Economic Journal paper found that under this model, bookmakers achieve profit margins 20 to 30 percent higher than a balanced-book operation would produce. That is not noise. That is the structural reason sportsbooks are built the way they are.

Seven years later, Rodney Paul and Andrew Weinbach tested the same hypothesis on NFL data and confirmed the pattern held in American football markets. Their Applied Economics Letters paper documented the same price-setting dynamics: lines set to exploit known bettor biases rather than to balance action. A separate test of NBA markets by Paul, Weinbach, and Paul produced the same result.

Pinnacle as the Counterfactual

To see the Levitt model in contrast, look at Pinnacle. Pinnacle runs a different operation. They accept winning bettors, publish their policy on this explicitly, and operate on 2 to 3 percent margins rather than the 6 to 10 percent margins common at DraftKings or FanDuel. Pinnacle uses sharp money to improve their own lines. When a professional bettor wins consistently at Pinnacle, the book updates its model based on what the winning bettor is telling the market.

This is the market-maker model, the approach Levitt contrasted with the bias-exploitation approach most books use. Pinnacle's lines show the difference. Their NFL spreads frequently differ from DraftKings and FanDuel by half a point to a full point, particularly on games with high public interest. The deviation is not random. It is directional: Pinnacle prices toward the neutral probability estimate, while soft books shade toward the popular team.

Most US bettors cannot access Pinnacle directly. Pinnacle lines are available on free odds aggregators regardless. The gap between Pinnacle and DraftKings on a given game tells you something concrete about which direction the soft book has shaded its price and by how much.

What This Changes for Your Bets

The Levitt model has four direct implications for how you approach betting markets in 2026.

Line movements are not always what they look like

The standard framework treats every line move as information: the market is correcting toward a more accurate number. This is sometimes true. Under the Levitt model, a book also moves lines based on its own positioning goals, not only to balance incoming action. A book sitting heavily short on an underdog in a high-profile game might move the line toward the favorite not because sharp money came in, but because the book wants to attract offsetting action on a side where it is already exposed.

Reverse line movement, where the spread moves against the direction of public money, remains a real signal. The signal is cleaner on sharp books than on soft books. A move at Pinnacle is information-driven. A move at FanDuel carries more ambiguity.

Main-game spreads are the most refined Levitt markets

Books have accumulated decades of behavioral data on NFL and NBA main-game spreads. They know exactly how the public bets on nationally televised games, on home favorites, on playoff teams, on star quarterbacks. The Levitt model is most fully deployed in these markets. This is a core reason why simple fade-the-public strategies on NFL and NBA spreads have eroded. The books already priced in the public bias before you arrived.

For a deeper look at why that stopped working and what replaced it, the piece on why fade-the-public is dead covers the specific decay of that signal.

Player props, alternate lines, and markets with lower public volume are softer ground. Books have less behavioral data. They price more off internal models and less off the Levitt positioning playbook. This does not eliminate the vig problem, but it narrows the adversarial gap. The earlier piece on player props versus game lines documents where the softness shows up in 2026.

Pinnacle lines as your benchmark

Most US bettors cannot access Pinnacle. Pinnacle lines are available on free odds aggregators and should be checked before placing any bet on a high-volume US book. A spread that is Pinnacle -3 and DraftKings -3.5 on a popular team tells you the soft book has added a half-point of Levitt shading. You are paying for the public's bias, not only the standard vig.

Kalshi and Polymarket prediction markets serve a similar function for game outcomes. These markets aggregate crowd probability without Levitt-style bias exploitation built in. When a prediction market's implied probability differs significantly from a sportsbook's no-vig implied probability on the same event, the sportsbook is almost always the one deviating from true probability.

This is the structural comparison the piece on prediction markets versus sportsbooks covers in detail.

Your competition is the book's model, not other bettors

Every standard guide frames sports betting as a competition against the market, meaning other bettors pushing the line around. The Levitt model reframes the problem. You are competing primarily against the book's internal model of your behavior and the behavior of bettors like you. The book is not a neutral exchange. It holds a directional view on nearly every game it takes action on.

Beating the book requires finding spots where the book's bias model is wrong. Those spots are not evenly distributed. They cluster in markets with lower public volume, markets the book prices reactively rather than proactively, and markets where the book's behavioral data is thin. Knowing the book holds a view changes which markets are worth playing and which are not.

Three Questions Before You Bet

The Levitt model translates into three specific questions worth asking before placing any bet on a retail sportsbook.

First: is this a market where the Levitt effect is likely fully priced? High-profile NFL spreads on publicly popular teams are a strong yes. Alternate line options, niche props, and early-week totals on lower-profile games are weaker. The weaker the yes, the more likely you are finding a price that reflects neutral probability rather than bias exploitation.

Second: what does Pinnacle or the prediction market say? If the Pinnacle number differs from your retail book by more than a standard rounding difference, the direction of the deviation tells you which side the book shaded toward public money. That is the Levitt premium built into your bet price.

Third: am I betting with the public or against the book's model? When your bet lands on the same side as the majority of money at a soft book, you need extra conviction. The book already held the Levitt advantage on your side before you placed. Adding to the popular side is not always wrong, but it requires a sharper edge to break even against both the standard vig and the book's directional position.

The Structural Reality

Every piece of strategy built on the balanced-book assumption is built on the wrong foundation. Books take positions. They hold directional views. They profit from knowing more about bettor behavior than bettors know about themselves.

The data from Levitt, confirmed across NFL and NBA markets, describes the structure you are operating inside today. The major retail books, DraftKings, FanDuel, BetMGM, Caesars, run Levitt-model operations with two decades of behavioral data behind them. The exceptions, Pinnacle internationally and Circa in Nevada, are not. Prediction markets are another exception.

Understanding the structural difference does not eliminate the vig. It tells you which markets are worth playing, which lines carry an extra layer of adversarial pricing, and which benchmarks to check before treating a number as close to fair value. That is the starting point for finding genuine edge, not a shortcut around it.