My writings about baseball, with a strong statistical & machine learning slant.

Sunday, April 4, 2010

Team Pitching Projections (I: Intro & Adjustments)

It's opening day! I didn't get to produce reasonable projections for everything I set out to do before the season started, but I still got a lot done. Now that I published my VORP and ERA, I thought I'd break down a few teams' pitching staffs for 2010, and see how my projections compare to those of PECOTA (Baseball Prospectus's famous projection system) and CHONE (another famous projection system).

I didn't have the current updated teams for every player, so instead I chose a few teams, took their opening day rosters and DL'ed pitchers, and based my team projections based only on those 11-14 guys. The opening day rosters I used are from an article on BP for April 1st (AL rosters, NL rosters).

In the next few posts, I will present and discuss my team projection for the Yankees, Rays, White Sox, Rockies and Giants. Since these are all good pitching teams, I will also break down my projections for the Indians.

Most of the stats I show are pretty straightforward: take some projections & add 'em up. However, there are a few caveats. I explain those below. If you don't care, just skip forward to the team breakdowns, and maybe come back here if something doesn't make sense.

Missing Innings

In a 162-game season, each team's pitchers should throw 1450 innings, give or take a few. However, as I noted for my Yanks and Rays IP breakdowns, a system should not project 1450 total innings to any team's top 11 pitchers. Teams end up using guys not on the opening day roster to pitch anywhere from 100-600 innings every season. Not infrequently, some of those innings go to pitchers who are not even projected by PECOTA or CHONE, systems that project thousands of minor league players. Therefore, no matter how many pitchers we project, the team numbers should add up to less than 1450.

In order to have projections for team ERA and team VORP that do add up, I credit the "missing innings" to an unknown player. It would be nice to give that player a "replacement level" ERA and a VORP of 0.0, but that doesn't quite work out. PECOTA uses a replacement level of around 5.4 ERA (equivalent to 5.8 RA), league-neutral and park neutral. However both my system and the CHONE system project an average ERA of 5.0 for the bottom 1/6 of projected pitchers. In other words, both my system and CHONE starts a pitcher off at 5.0 ERA, if we know nothing about him, other than the fact that he's not on an MLB opening day roster, but ends up pitching in the majors.

Therefore, I credit the "missing innings" to a pitcher with league-neutral and park-neutral 5.0 ERA in all cases. I do the same thing for the "missing innings" for PECOTA and CHONE's projections, if there are any.

Without this adjustment, projected VORP totals would end up being low, compared to previous years' averages.

FWIW, I'm not claiming that the replacement level should be set to 5.0 ERA. I'm just saying that my system would project an unknown pitcher to a 5.0 ERA, so I wanted the VORP projection to reflect that. I don't know how PECOTA or CHONE would project such a player, so I fill out their team projections at the same level, for consistency.

Team Run Factors

My FIP/ERA projections are based off of previous years' QERA and FIP. Since these are not affected much by the league or park of the pitcher, I can use my VORP projections directly without team-based adjustment. I also use PECOTA's VORP projections directly. However, CHONE does not project VORP. It projects WAR, but I'm not project that as of yet. So I compute VORP projections from CHONE based on its ERA and IP projections. I'm assuming that CHONE's ERA figures take park and league into account. If so, I need a different "replacement level" RA for each team in order to compute VORP properly. For this, I use the 2009 VORP and RA/G season totals for each team from Baseball Prospectus's "Team Audit" pages.

As an added bonus, I can compute the team's "run factor," compared to the MLB average. My ERA projections are league-neutral and park-neutral, but using this factor, I can project ERA for each player adjusted to his current team's "run factor."

My adjusted ERA projections (based on that team's run factor) are listed for each pitcher. Also I included each team's 2009 run factor and team VORP, as well as implied replacement-level RA.

Hopefully you'll find my projections interesting. Can't wait to see they turn out...

No comments:

Post a Comment