Saturday, July 28, 2012

Today's Games - Relaunched

Coffee cup iconThe Today's Games feature of this site has now been relaunched for Major League Baseball.  I had originally published this during the 2011-2012 NBA season.

Purpose

The purpose of this feature is to, effectively, "predict a prediction".  The main feature of this site is the betting market power rankings, which are my attempt to reverse engineer what the betting market thinks about team and pitcher strength.  What Todays' Games does is then take those rankings and try to predict how the betting market will set the moneylines and run totals for each day's games.

The goal here is to look for discrepancies between my prediction of the betting numbers and what they actually turned out to be.  If my modeling perfectly captured how the betting market "thinks", then any discrepancy in the predictions would be due to new, late breaking information.  During the NBA season, I found that these discrepancies were usually good indicators of key players either sitting out due to injury, or a key player returning from injury after a long absence.


Supplementing Stat Based Models

The Achilles Heel of most stat based models is that they do not handle injuries very well.  One of my goals in developing this feature was to provide an automated way to address this weakness.  For games with a big discrepancy in the predicted betting lines, it may be wise to "lay off" those games if you're using a stat based model to try to beat Vegas.  Time permitting, I plan on back-testing this idea to see if a simple stat based model, when supplemented by this information, generates positive performance against the betting lines.

One additional note: the numbers shown in Today's Games reflects some methodology changes in my ranking methodology that I haven't rolled out to the other rankings on this site just yet (but I plan to do so soon).

Glossary

Here is a description of the fields (see my previous posts for more detail on the methodology):

  • win prob miss - The difference in the win probability implied by the moneylines and what my rankings predicted.  I convert the moneylines into a "vig-neutral" win probability.  The bar graph points in the direction of the team that is favored by the miss.
  • home line - the moneyline for the home team
  • away line - the moneyline for the away team
  • win prob act - the win probability for the home team, as implied by the moneylines
  • win prob pred - the win probability for the home team, as predicted by my rankings
  • predicted total - the run total, as predicted by my rankings
  • actual total - the actual betting run total for that game

2 comments:

  1. One nice thing about baseball though is that an injury to a position player is unlikely to significantly affect his team's run production on game by game basis (unless he's Barry Bonds). I don't know if Vegas agrees but this has been noted by the sabermetrics community.

    ReplyDelete
  2. A downgrade of 100 OPS points in a single lineup spot is worth ~2.5% in win probability, so injuries aren't usually to be ignored.

    ReplyDelete