The discussion of football contests led me to consider what sort of math model might approximate the capper population and results in the Hilton SuperContest. The idea is to roughly assess the role of luck vs skill (long term capping ability and taking advantage of stale lines) in winning the contest.
Several data points about the Hilton SuperContest are known:
1. The number of contestants has been pretty stable, around 340 for the last 4 years (down from a peak of 505 in year 2005).
2. The overall cumulative record for the last 9 years is 126755-123461-6402, 50.66% win %.
3. A win % of 64-68% usually wins the contest.
Observations/assumptions:
The relative low cumulative win % and observing the mix of selections from week to week leads me to believe that a significant % of the population doesn't take advantage of many (if any) of the stale lines and are essentially losing bettors, i.e. handicapping ability is very low. Lets call this population "Dead Money" ("DM" - assume long-term 49.5% win %).
The remaining capper population I break into 3 groups, one called "Marginal" who would able to break against the stale lines ("M" - roughly long-term 52.4% win % cappers including partial use of the stale lines), one called "Good" ("G" - long -term 54% win % including the stale line advantage) who would win against the Hilton line, make good use of stale lines and perhaps employ some reasonable game theory at the end to maximize their $ win and one called Excellent ("E", 56% win % and all the attributes of the "G" group)
Now empirically, I have chosen 340 participants: 170 DM, 130 M, 32 G, and 8 E.
I've run some simulations on these parameters, but would first like to have comments on the set of assumptions. I would stress that this is still a very simplistic model, it doesn't account for the fact that the stale lines vary from week to week with Week 17 being particularly wild. It doesn't account for a few dropouts near the end of the contest who just quit in the final weeks. It doesn't account for various game theoretic strategies tha might be employed, but I doubt that any contestants have any sort of rigorous model to quantify/demonstate their choices as being anywhere near-optimal except in the near trivial cases of being very near only one or two contestants.
Suggestions on the problem's model/parameters would be welcome.
I'll give the results in a future post.
Several data points about the Hilton SuperContest are known:
1. The number of contestants has been pretty stable, around 340 for the last 4 years (down from a peak of 505 in year 2005).
2. The overall cumulative record for the last 9 years is 126755-123461-6402, 50.66% win %.
3. A win % of 64-68% usually wins the contest.
Observations/assumptions:
The relative low cumulative win % and observing the mix of selections from week to week leads me to believe that a significant % of the population doesn't take advantage of many (if any) of the stale lines and are essentially losing bettors, i.e. handicapping ability is very low. Lets call this population "Dead Money" ("DM" - assume long-term 49.5% win %).
The remaining capper population I break into 3 groups, one called "Marginal" who would able to break against the stale lines ("M" - roughly long-term 52.4% win % cappers including partial use of the stale lines), one called "Good" ("G" - long -term 54% win % including the stale line advantage) who would win against the Hilton line, make good use of stale lines and perhaps employ some reasonable game theory at the end to maximize their $ win and one called Excellent ("E", 56% win % and all the attributes of the "G" group)
Now empirically, I have chosen 340 participants: 170 DM, 130 M, 32 G, and 8 E.
I've run some simulations on these parameters, but would first like to have comments on the set of assumptions. I would stress that this is still a very simplistic model, it doesn't account for the fact that the stale lines vary from week to week with Week 17 being particularly wild. It doesn't account for a few dropouts near the end of the contest who just quit in the final weeks. It doesn't account for various game theoretic strategies tha might be employed, but I doubt that any contestants have any sort of rigorous model to quantify/demonstate their choices as being anywhere near-optimal except in the near trivial cases of being very near only one or two contestants.
Suggestions on the problem's model/parameters would be welcome.
I'll give the results in a future post.