Predicting Winners in MLB

[ImNew]

A Two-Stage Bayesian Model for Predicting Winners in Major League Baseball

http://www.sinica.edu.tw/~jds/JDS-142.pdf

Looks like they have a good system there but I have no idea what the forumula is, anyone cares to share their insights in that article?

[Thremp]

I just got to the ERA/BA part. Looks horrid off the cuff.

[rjp]

This isn't much use to a handicapper if they're not predicting odds.

[SonnyJay]

There's nothing wrong with using a Bayesian model, but the variables appear to be way too simplistic to actually be useful for handicapping. They're trying to capture way too many subtleties in three variables that don't always reflect what the authors are intending them to. For example, hitting statistics are subject to lots of short term luck. Even when you take that out, the use of batting average as a representation of a team's offensive ability is ridiculous. Same with ERA: subject to a lot of luck, and not necessarily an accurate representation of a team's pitching/defense output.

It looks fine as an academic paper, but not particularly useful for real-world application.

[TomG]

[ QUOTE ]
When our predictive probabilities differ from those of the bookmaker by a prescribed margin, this triggers a wagering situation. The natural question then is whether this betting strategy yields income sufficient to overcome the bookmaker’s vigorish. This is a topic of future research.

[/ QUOTE ]
Typical academia. The paper focuses only on the quantitative analysis itself. Whether the analysis is actually meaningful (i.e., profitable), well that's the beyond the scope of the paper. [img]/images/graemlins/confused.gif[/img]

[cboevey]

They are predicting probability of the home team winning. It is trivial to convert that to odds.

[rjp]

[ QUOTE ]
They are predicting probability of the home team winning. It is trivial to convert that to odds.

[/ QUOTE ] Correct me if I'm wrong, but after a brief skim it looked like they were predicting a long term winning percentage for the home team, not game by game winning percentages.

[cboevey]

[ QUOTE ]
[ QUOTE ]
They are predicting probability of the home team winning. It is trivial to convert that to odds.

[/ QUOTE ] Correct me if I'm wrong, but after a brief skim it looked like they were predicting a long term winning percentage for the home team, not game by game winning percentages.

[/ QUOTE ]
Both. The long term winning percentages was the second part. Models 1-3 was the first part and the output is a probability of home team win.

[newb411breaker19]

what i've noticed to work quite well in baseball is through making offensive power ratings and dividing that rating by the overall team ERA for the starting pitcher as well as the bullpen. To make the offensive rating I use something similar to the long drawn out version in bob mccune's "book education of a sports bettor". There's a couple different methods bob provides but the simple equation method is not good (the one that requires more work and comes later in the book is ideal). Instead of using batting average you may just want to use Slugging% but that's just preference

basically when working with pitcher ERA's I'm using XERA which I believe was invented by bill james.
So take the larger power number for team A let's say and divide it by Team B and adjust for home field and you have a pretty solid line that should beat the books over the long term. This does take a lot of time however but it seems to be worth the effort.. since may 24th (needed time to gather data) i'm up around 9 units

[rjp]

cboevey, I'll take anorher look then.