Across decades of studies and markets, bets at long odds return worse than bets at short odds: recreational money loves lottery-shaped payouts, books shade longshot prices accordingly, and the sportsbook margin concentrates exactly where the dreams are.
The bias is one of the few market facts stable enough to build rules on: be suspicious of your own longshot enthusiasm, demand extra edge before taking 6.0+, and remember that most 'value' a naive model finds at big prices is the bias laughing at the model. Our own v1 went 0-for-8 on exactly this rake.
Typical finding: bets at 1.50 return ~97p per £1 staked; bets at 15.0 return ~85p. Same sport, same books, same day — the difference is where the margin hides. A model that ignores this will 'find' its biggest edges precisely where prices are worst.