Herding in Prediction and Betting Markets

Herding in prediction and betting markets is the tendency for large numbers of bettors to follow visible price movements—moving odds, visible line shifts—even when those movements carry no new information about true probabilities. A prediction market might show 60% odds for a political candidate; as odds rise to 65%, more bettors interpret the shift as a signal that the candidate’s chances have improved, and they add money on that side. Their new bets push odds higher still, creating a self-reinforcing cycle that disconnects price from reality. Prediction markets are supposed to aggregate dispersed information; instead, herding distorts that aggregation.

The Herding Mechanism in Betting Markets

Prediction and betting markets—political betting, sports betting, crypto prediction markets—are designed to aggregate private information and reveal consensus probabilities. In theory, a bettor who believes a candidate has 65% odds to win, when market odds are 60%, should bet on that candidate. If many informed bettors do this, market odds rise to near 65%, and the market has discovered a truth.

In practice, the visible movement itself becomes the information. A bettor logging into a sports betting exchange sees that Team A’s odds have shifted from 2.5 to 2.8 (in decimal odds, implying a decline in the team’s perceived win probability). This shift need not reflect new game footage, injury news, or injury reports. It may simply be that one large gambler bet against Team A, or that the odds drifted due to algorithmic repricing. But the visible shift signals to other bettors: “Something has changed; Team A is less likely to win.”

This creates a cascade. Other bettors, unwilling to do deep independent analysis or lacking real information, see the odds move and assume someone knowledgeable acted first. They follow—betting against Team A—which pushes odds further against Team A. Each new bettor sees the worsening odds and interprets them as new evidence, betting again against Team A. The cycle becomes self-reinforcing.

Importantly, no new information about the team’s quality has arrived. The shift is pure herding: bettors responding to price movements as if those movements contain information, when in fact they contain only the echoes of prior bettors’ behavior.

Why Prediction Markets Are Vulnerable

Betting markets are particularly susceptible to herding for three reasons:

Opacity of individual rationales. Unlike stock markets, where company earnings reports and analyst notes are public, betting markets often hide the reasoning behind bets. A stock price falls because of a bad earnings report; everyone sees the report and understands the cause. A sports bet changes because the odds moved; no one knows whether the original bettor had injury information, inside knowledge, or simply flipped a coin. The lack of visible causal information forces followers to infer, and inference biases them toward following trends.

Speed and liquidity imbalance. In liquid stock markets, if prices rise on no information, many traders can simultaneously profit by shorting overvalued stocks, pushing prices back down. In smaller prediction markets, if odds on a candidate move sharply on a herding cascade, few well-capitalized contrarians exist to bet against the move. The crowd’s money overwhelms rational correction.

Psychological accessibility of odds. An odds display is simple and salient. “69% odds” is easy to compare to “65% yesterday.” An improvement from 65% to 69% looks like meaningful movement, even if it reflects only a single large bet or random walk. Complex fundamental analysis is invisible; odds movement is not. Bettors anchor to the visible number.

A Worked Example: The 2024 Prediction Market Bubble

In the months before a major election, a prediction market showed a candidate X at 45% odds to win. A respected political analyst publishes a long essay arguing the candidate is undervalued and will likely win with 55% probability. The analyst bets $100,000 on candidate X. This is real information; the market reprices to 50% odds.

But now the visible shift—45% to 50%—attracts attention. Other bettors see the 5-point move and assume new information has arrived. They do not read the analyst’s essay; they see the odds and assume the odds moved because someone smart acted. These bettors place $50,000 more on candidate X.

Odds now shift to 54%. The shift is larger than before (4 points), and it catches more attention. Smaller bettors see 54% and update their beliefs: “If odds are at 54%, candidate X must have a strong case.” They bet another $75,000 on candidate X.

Odds move to 58%. The cascade accelerates. By now, the original analyst’s $100,000 bet is long forgotten or unknown; bettors only see that odds have moved steadily upward. They infer that large, informed capital is backing candidate X, and they follow.

Odds eventually reach 70% or 75%, driven entirely by crowd herding, not new information. The analyst’s original $100,000 was justified by a real 55% probability. But the subsequent $200,000+ is not; it’s herding. Odds are now disconnected from reality.

After the election, if candidate X underperforms—winning with 52% actual probability, not 70%—the late bettors who bought at 70% odds lose heavily. They were not wrong about the candidate; they were wrong about the odds. They paid too much, chasing a herding crowd.

When Herding Reverses: The Stampede Out

Herding is symmetric. If a cascade can push odds up on no new information, it can push them down just as violently.

Suppose a single media report—speculative, not fact-checked—suggests candidate X has a campaign finance problem. One large bettor interprets this as material and bets against candidate X, pushing odds from 70% to 68%. Other bettors see the move and assume new negative information has arrived. They bet against candidate X as well. Odds fall to 65%.

The original report was debunked hours later. But by then, the visible 5-point decline has created its own momentum. Bettors who missed the herding into candidate X see the recent move out and assume they should exit as well. Odds collapse to 60%, then 55%, then 50%—despite no change in the candidate’s actual chances.

These reversals can be violent. Markets can overshoot in both directions, creating a boom-and-bust dynamic that has no relationship to underlying probabilities.

Measuring Herding: Betting Flow Correlation

Economists identify herding by looking for correlated betting flows that precede price moves, unaccompanied by new information. If a betting exchange sees:

• A spike in bets on candidate X at the same time market odds on candidate X shift sharply, but
• No contemporaneous news release, earnings report, or external event that could justify the shift, and
• The spike persists even as later bettors do not add new information,

then the price move is likely driven by herding, not information.

Another measure is the speed of odds adjustment relative to news events. Stock markets reprice within milliseconds of a news release. Betting markets, if information-driven, should show rapid repricing after genuine new information (an injury, a dropped debate). But if odds drift slowly even in the absence of news, or spike suddenly with no news catalyst, herding is likely at work.

The Paradox for Market Efficiency

Prediction markets are supposed to be more efficient than traditional polls or forecasts because they aggregate dispersed information and create financial incentives for accuracy. A bettor who believes a candidate will win with 65% odds, when market odds are 60%, stands to profit by betting. This profit motive should drive market odds toward the “true” probability.

Herding breaks this mechanism. Herding bettors do not have superior information; they are simply following trends. Their bets do not correct mispricing; they amplify it. Late herding bettors buy at the peak, when odds are 70% or 75%, and lose when the candidate wins with true 55% probability.

The result: prediction markets become less efficient, not more. They aggregate not information, but collective delusion. The early herding bettors profit at the expense of the later followers—a transfer of money, not production of knowledge.

Contrarian Profit from Herding

The flip side: If herding is predictable, contrarian bettors can exploit it. A trader who recognizes that candidate X’s odds have surged 10 points with no new information can bet against the herding crowd, confident that odds will eventually revert. This is a profitable trade, but it is not trading on superior information about the candidate. It is trading on the behavioral pattern of the crowd.

Over time, if enough sophisticated traders exploit herding, markets can self-correct. The availability of real-money contracts and algorithmic trading has made prediction markets somewhat more resistant to extreme herding. But the vulnerability persists, particularly in smaller, less liquid markets with fewer professional traders to provide contrarian pressure.

See also