Videogames Make Traditional Super Bowl Predictions

Thanks to Reuters for its article discussing videogame-based predictions for this weekend's Super Bowl. The piece explains: "Days before the real football championship is contested, Carolina Panthers wide receiver Steve Smith has beaten the New England Patriots' wide-out Troy Brown 29-21 in a head-to-head video game matchup." The match was played on Sony's NFL GameDay 2004, and it's noted: "In the first eight years of the event, the winner of the electronic showdown went on to hoist the Vince Lombardi Trophy as the Super Bowl champion." Elsewhere, 1UP has done its own Super Bowl predictions on four different football videogames, and the final results also favor the Panthers.

Comments

Re:Mangled English Hurt Brain

[Acidic_Diarrhea]

Bill Simmons wrote an article on the game, as he was there. It can be read here. In it, he says that they were playing 3 minute quarters.

Editors? We don't need no stinkin' editors!

Re:Results weren't supported by tests

[leviramsey]

I am a professional handicapper and can confirm and elaborate.

Bookmaking in the US is based on -110 odds. Whatever side you want to back against the spread, you're risking $110 if you lose to win $100 if you win (there are discount books offshore/online; Pinnacle out of Curacao generally offers -105). Thus, as long as the betting on both teams breaks down in the 47.5% to 52.5% area, the bookmaker is guaranteed a profit whatever happens. At anything outside that range, the bookmaker is betting on the outcome, as they stand to lose money if the heavily bet side wins.

There's more to it than simple balanced action. The line opened in Vegas and offshore at NE -6.5 and got steamed up to -7. It's highly unlikely though, that any additional action would push it to 7.5 (at least at reputable books [see below]).

The reason is key numbers. You can generate pretty much every realistic number of points scored by an NFL team with a function like s(x,y,z)=3x+7y+z, where x, y, and z are all integers greater than or equal to zero, their sum is less than 8, and z is less than x+y. The realistic score differences are thus differences between s-values for two ordered triples satisfying those constraints. Analysis of this will confirm that 3, 4, 6, 7, 8, 10, 13, and 14 are commonly encountered score differences, with 7 and 3 being especially common.

The result is that bookmakers will require massively unbalanced action to move off of a key number (3, 4, 6, and 7) and to a lesser extent to move onto a key number. At those numbers, bookies are much more willing to gamble, as the balanced action around those numbers is a huge risk.

This risk is called "getting sided" or, even worse, "getting middled".

Imagine that you bet the Patriots -6.5, along with enough of the betting public, to push the line to 7. You buyback Panthers +7 at the same stake level. Let's say you're risking $1100 on both ends. If the Patriots win by 8 or more, Pats -6.5 wins, Cats+7 loses, for a net loss of $100. If the Patriots lose or win by less than 7, the reverse happens. But if the Patriots win by exactly 7 (which is common), your Panthers bet pushes (nothing gained, nothing lost) and the Patriots bet wins. Essentially you've hedged your way into betting at +1000 that the game lands on 7. This is "siding".

It's not difficult to see that the book gets mauled in this situation if everybody's doing it.

Now imagine that Patriots action pushed the line to 7.5. Now, both bets hit if the Patriots win by exactly 7. This is "middling".

You may recall a Monday Night Football game between the Giants and Bucs earlier this year. The line was Tampa -5.5 and steam on Monday pushed it to 6. Late in the game, Giants trailing by 4, they take a safety, giving Tampa two points. Al Michaels (who almost definitely bets on the games; it's not difficult to tell when he has OVER or UNDER) said, "That noise you just heard in Vegas wasn't an earthquake."