You don’t need bookmakers to bet. On a betting exchange, you place bets directly against other people, with the exchange matching buyers and sellers and charging a small fee. This paper is the first to set out a model of how odds get set on exchanges and compare it with evidence.
The paper starts by explaining how betting exchanges work. Betting exchanges look like financial markets. You can be a Maker (posting odds that someone else can accept) or a Taker (accepting an existing offer).
The model assumes people disagree about probabilities but are, on average, correct. Based on their beliefs, people decide to act as Makers or Takers or not to bet. Takers are confident in their beliefs and pay for certainty: they accept slightly worse odds to guarantee execution. Makers are not as confident in their beliefs. They look for better odds accepting that they might not get matched.
That structure delivers a set of predictions you can take straight to the data. Odds should, on average, be about right but Makers should do better than Takers. And losses for Takers get worse as the probability of winning falls.
The fun part is that exchanges let you test this properly, because they make data available so you can identify for each trade the offer made by the Maker and accepted by the Taker. Using Betfair transaction-level data covering over 200,000 soccer matches, the results for pre-match and early in-play line up well with the model: Takers lose more and more heavily on longshots, while Makers generally break even.
But then something weird happens as matches progress. Late in games, odds don’t line up with the true underlying probabilities. Longshot bets lose an increasing amount of money as the game nears its end, even for Makers.
Key research findings: Prior to play, a “rational disagreement” model does a good job of explaining the prices on betting exchanges. But once the game goes in-play, the odds on longshots imply bigger losses on those bets.
Practical advice: Pre-play odds on betting exchanges are really good estimates of the probability that a bet can win. You will do better making offers on exchanges than taking them. And really don’t bet on longshots on exchanges once the game has started.