Rugby handicaps and the bonus point

Rugby prices a handicap in a sport where a converted try is seven points and a side can score four in twenty minutes. The lines are wide, the variance is high, and the thing that most reliably confuses a model is not either of those. It is that the teams are frequently not playing for the scoreline.

#The incentive that is not the result

A bonus point is awarded for scoring four tries. A losing bonus point is awarded for finishing within seven. Both are league points, and in a tight table both are worth more than the difference between a narrow loss and a slightly narrower one.

So consider a side leading by twenty with ten minutes left and three tries scored. Every model says they are protecting a lead. In fact they are throwing the ball wide from their own twenty-two, because the fourth try is worth something and the margin is not.

#What it does to the handicap

The observable effect is a distribution of final margins with more weight in the tails than a symmetric model expects. Blowouts get bigger because the leading side keeps attacking. Narrow games stay narrow because the trailing side within seven stops taking risks.

A handicap priced from an expected margin and a symmetric spread around it will misprice both ends, and it will do so systematically rather than randomly, which is the kind of error worth finding. The general point is the same one that applies to any market derived off a headline number.

#Which competitions it matters in

Only the ones that award them. Knockout rugby has no bonus points and behaves much more like a standard handicap sport. Reading a league model onto a cup tie is a straightforward error and a common one.

#What this does not tell you

The size of the effect is hard to measure, because the situations that trigger it are rare and the sample is small. Anybody quoting a precise number for what a bonus point chase is worth is extrapolating further than the data allows.