I take it bookmakers don't just balance bets and take a slice, because it's more profitable to "bucket shop" (an old stock market crime term) and in effect, bet a little of their own money against their customers if they think the customers are wrong, using better models/predictive systems than their customers have to know when to do this. But the bookies don't take (so far, anyway!) reversion to the mean into account; therefore they don't partially discount their most extreme (compared to consensus) predictions, or not sufficiently. To profit from this error on their part you can find those edge cases and bet against the most extreme of the house's secret bets - which aren't really secret 'cause they skew they odds given (skew vs the odds created by bets actually received. But since the authors don't know those bets, they must be using consensus odds as a proxy, or comparing other betting services' odds.) Still thinking 'bout this though.