Nice LT and wtg with wv/x over...
the biggest mistake people make when capping baseball is looking at starting pitcher vs. starting pitcher, you must look at:
starter + bullpen vs. lineup
Thank you. Well, I could make a case for looking ONLY at bullpens and ignoring everythinbg else! Of course, that position could change if the angle falls below +76.2 units this year, because that would say that the oddsmakers are truly catching up to bullpens.
Still, you don't average 97 units in three years without having some juicy +200+ underdogs that are top 10 pen with a mediocre at best starter vs. a Cy Young candidate (or the overvalued Yankees last year) and a non-top 10 pen. If you considered the other conventional factors, you probably would have passed on some nice winners.
Excellent article. My data has estimates of bullpen strength (for each of various statistics, park and opponent-adjusted) and I went back to take a look and sure enough, betting bullpens was good over the last three years.
Looking farther back, though, I saw that bullpens were not at all mispriced (aggregately) in the 01-05 period. In my brief glance, I didn't look at the difference in pens, just the z-score of the bullpen overall, so the 3% return per game I found should be even greater using the difference, which you showed.
This made me wonder if it was simply that bullpens were not being properly evaluated the last three years or maybe that they are being less emphasized in the lines.
I modeled the point spread as a function of bullpen strength (I use normalized wOBA) and doing just a simple linear regression, I found that the relationship between bullpen strength and the betting line has definitely decreased in recent years. (coefficient estimates between the two models differ significantly). The same relationship holds when using my model's predicted bullpen score as a regressor.
But did bullpen strength factor into the game outcome more in recent years? Actually, its effect was smaller in the past three seasons.
However, running a probit model regressing team win on the betting line and bullpen strength, I find that bullpen strength is marginally significant in 05-07, but not so in 01-04.
Using my model's predicted bullpen performance instead of z-score for wOBA (and actually 05-07 is the out of sample period for the model, so if anything 01-04 would be expected to be better) I find that the predicted bullpen performance is insignificant from 01-04, but very significant in the 05-07 regression, especially for home teams (p=.01).
Looks like the market may be pricing bullpens worse now than they did before, but now that this is out, it should correct itself at some point.
I usually do a weighted formula of bullpen OBP, % inherited runners scored allowed and ERA to rank my bullpens. Its worked pretty good so far but I am always willing to refine it.
Love the article LT. Baseball is where I want to be most active this season. But is there a site to find the bullpen ERA stats, or is that a number one has to keep on his own?
Yes I shoud have mentioned covers was my source for all odds on past games, and yes they do have bullpen ERA. Also, this is not something that could be simply backtracked, as you would need to know the Top 10 pen ERAs GOING INTO each day. Full season bullpen ERA are useless for this angle.