ESPN's Accuscore: How Close Is This To Decent Modelling?

[lucciferg]

I hope this isn't a stupid question. I've been placing a few MLB bets and following the games on ESPN's website (I'm in the UK).

They have an "Accuscore" model on the preview page for each game that gives % favourite and underdog for each game. I've noticed that the % can sometimes be a fair bit different from what the odds at the bookies would suggest.

I wondered what people thought of the Accuscore model: is it missing lots of critical data or something? How close is it to the kinds of models that you guys are building and have built?

[twister]

I hope this isn't a stupid question. I've been placing a few MLB bets and following the games on ESPN's website (I'm in the UK).

They have an "Accuscore" model on the preview page for each game that gives % favourite and underdog for each game. I've noticed that the % can sometimes be a fair bit different from what the odds at the bookies would suggest.

I wondered what people thought of the Accuscore model: is it missing lots of critical data or something? How close is it to the kinds of models that you guys are building and have built?

Accuscore is actually a separate company, and they provide their service to ESPN's website. The Accuscore projections on the ESPN site seem to be a few hours outdated, but for general (non-betting) purposes they do their job.

Accuscore provide up to date projections on their site, but you have to be a paid member to access them. The fact that they charge, plus I wouldn't want to base my long-term betting on data from a company that could close up shop tomorrow, means I don't use them.

The way they recommend you bet, using their projections, is called "line value". Basically, you turn the closing (well, as close to possible) lines from Pinnacle, into %'s. Then compare the percentages to the Accuscore projections. Whichever side has the higher accuscore projection compared to the % derived from the odds, is the side you bet on. If you had done this with every single MLB game for 2007 season, betting 1 unit a game, you would be up something like 70 units. Their 2008 record was something like 80 units.

Example:

Houston are 2.08, Cincinnati are 1.86. Accuscore Projections are 49% Houston, 51% Cincinnati.

For Houston, you do 1.86/(1.86+2.08) = ~47%
For Cincinnati, you do 2.08/(1.86+2.08) = ~53%

As you can see, the projections say that Houston have a better chance of winning than their odds suggest, so you bet on Houston. That is basically how their "system" works.

Based on their results, I'd say their model is pretty damn accurate, and I think they get even more moneyline results (i.e who should be the fav) correct than the bookies themselves. Only by a 1-2%, but better than the books nonetheless.

[lucciferg]

Thanks for the detailed explanation, Twister. Unfortunately, I can't bet every game, but it looks like it'll be useful to double-check my instincts based on reading up elsewhere.