Online Betting, Sports Forecasting, and Accuscore

One of my favorite quirky investment books is Calculated Bets, a comp sci professor’s journey through coming up with a working computation model for betting on the jai alai market. While it’s nominally about jai alai-related issues, much of the discussion is around coming up with functional models, real-world probabilities, transaction costs, and so on. Highly recommended for anyone with a quantitative finance bent — and it’s easy and fun reading.

I got to thinking about Steve Skiena’s five-year-old book today when mulling Accuscore, a former biologist’s sports-betting service that simulates U.S. football. He has been selling the service for more than a year, and he claims he is beating the Vegas two-point spread 56 per cent of the time, which is enough to be profitable. There are still lots of questions, not least of which is, If the model works, why would you ever sell it to outsiders rather than using it yourself?

Nevertheless, fascinating stuff — and sports simulation doesn’t stop here, as an article in the current New Scientist points out:

Can sports video games be used to predict the outcomes of real
matches? Accuscore’s simulations of American football games now have
some competition in the form of Madden Football ’07, a game from Los
Angeles-based Electronic Arts that is based on the playing habits of
real players and teams.

In 2001, Jon Robinson, a gaming enthusiast and columnist for bi-weekly sports digest ESPN The Magazine
started an online column to see how each year’s new version of the
Madden game stacked up against human experts in picking the winners of
real games. The computer would play itself and provide predictions. “So
far machine has never beaten man,” Robinson says. “This year it looks
like maybe it could happen.”

Why the
improvement? A lot may have to do with powerful new gaming consoles
such as Xbox 360 and PlayStation 3, which allow game producers to more
accurately mimic the strengths and styles of real National Football
League teams. Through 240 games this season, Madden had picked 140
winners correctly (58 per cent), six more than ESPN sports analyst
Lomas Brown (56 per cent) but six less than Robinson (61 per cent).
Accuscore predicted the winners correctly 62 per cent of the time. “Is
the sim finally catching up with man? I don’t know, but bookies might
be out some money by 2010,” says Robinson.


  1. I’m not a gambler – in fact, I think sports gambling ruins the enjoyment of sporting events, it quickly sucks out the fun of being a fan, and it consumes far too many people financially and emotionally.
    But this article from an Accuscore site holds a lot of good lessons for anyone interested in behavioral finance. Very analogous to investing.

  2. Hmm, how much sampling bias is there here? The model which is getting a good run of coin-flips is going to be the “newsworthy” one.