Tuesday, June 12, 2007


A reader just alerted me to the presence of Accuscore, which estimates the probability of one team winning a game by simulating the contest 10,000 times.

This is a great idea in practice, something I would probably include in my handicapping if I had the computer power and the programming skills to do so. In practice, Accuscore is probably horribly flawed, and I'm not just saying that because the Ari/NYY forecast recommends betting against my 5 unit play.

Why do I believe this? Garbage in, garbage out. Here are some bits of wisdom from the Accuscore website:

(Note: all spelling and grammar errors are theirs, and only serve to make the product look even more unprofessional)

"Overall, road teams this year are winning .479 percent of games while home teams are winning .528 percent of the time."

Isn't there some constant those numbers are supposed to sum up to? I keep forgetting these things.

"I had some “research” on the number of gentleman’s clubs per city. In cities where the club per capita is higher than average, Lugo is hitting 194 points below his season average? As of 5/16 that was 0-19 in 2007.

The question is, are there other players who also exhibit a decline in performance in big cities, cities with a high “SCI” (Strip Club Index), or both?

Throughout the year, we’ll provide statistical analysis investigating the effects of Krystal, Destiny, and Jasmine on Major League Baseball players’ bating average. "

You read that correctly: Strip Club Index. You can't make this kind of stuff up. I wonder if ESPN read these articles before signing the partnership agreement?

"It is important to know that predictions with bigger differences between Simulation Percentages and Gambling Odds Percentages DID NOT perform better than predictions with smaller differences.

[B]ecause MLB scoring dynamics differ significantly from other sports we cover, we strongly advise not reading too much into how much higher or how much lower our simulation winning percentages are from the Vegas Odds generated percentages."

Take a look around the site. They're clearly marketing themselves to gamblers. Imagine a stock analyst confessing to you that his hottest picks don't actually beat the market. (This is true for most pundits, but they wouldn't dare admit it.) That's basically what StatShark is doing here.

Found a game where the Vegas line is 50 cents off? Apparently that's no better than a 1 cent difference.


I could go on, but do I need to? You might as well make picks by flipping a coin; at least when you're done, you can still spend the coin on a gumball.

No comments: