keyboard_arrow_uptop

It's my pleasure to inform you that we have started producing PECOTA updates for rest-of-season performance.

Before we get into the details, let's clarify what is being updated. The changes will affect:

For players who are in the depth charts, the PECOTA summaries at the top of the player cards will be updated as well. Players who are not in the Depth Charts but have preseason PECOTA projections will still have those at the top of the cards. Any other PECOTA data on the cards, including PECOTA percentiles, will be unaffected. In order to view the Depth Charts or use the Player Forecast Manager, you must have a monthly, fantasy, or premium subscription to Baseball Prospectus.

There are three principal changes to the way we're calculating in-season PECOTAs this year. One is how playing time is handled. We are still using the Depth Charts to estimate a player's playing time, but we are now using games remaining on the MLB schedule to determine how much playing time to allocate to each team (and from there to each player).

The next is the projections themselves. PECOTA works by forecasting a series of component stats (some easily recognizable, like walk rate and power on contact, some more obscure) and then building batting lines and eventually value stats. We assign each of those components a reliability score, based on our estimate of the margin of error of those component projections. (The major inputs are the amount of playing time the player has had, how long ago that playing time was, and the level—MLB, Triple-A, etc.—the playing time was at and the spread of talent in that population.) For season-to-date stats, we compute the same components and the same reliability scores. We then combine the two, using the reliability as the weight—in essence, we treat the preseason PECOTA as a Bayesian "prior." (You'll note I didn't mention regression to the mean—since the preseason forecast is already regressed, we don't need to regress the current season stats any—in essence, we're "regressing" the player's stats to his preseason projection, not the population mean.)

The third is that we are updating projections daily, rather than whenever the depth charts change. That way, changes in playing time due to the remaining games diminishing are captured, as well as changes to the PECOTAs based on player performance (although a single game's performance is highly unlikely to move a player's forecast in a noticeable way).

Again, as a reminder—only the PFM and DCs (and products that rely on them) and the PECOTA summaries at the top of the player cards of players in the DCs are affected. Every other PECOTA product is still using the preseason forecast. If you have any other questions, please leave a comment below.

You need to be logged in to comment. Login or Subscribe
rjblakel
5/08
Cool, but under the fantasy link, the PFM still shows as last updated on 05-07. Is that right?
cwyers
5/08
The article went up a little early, sorry about that. Everything is updated now and the right date is set in the Fantasy page.
jrbdmb
5/08
Is there any way to get updated full season projections, either in PFM or an updated PECOTA spreadsheet?
swarmee
5/08
What do you mean: A) New projections prorated to rest of season plus previous stats or B) New projections prorated out to 162 games, discounting already completed games?
jrbdmb
5/09
B. A full 162 game projection based on updated PECOTA projections. Which raises the question - have the underlying PECOTA projections themselves been updated, or just playing time forecasts?
cwyers
5/09
We are updating the projections themselves, I describe the process in the article. We are not publishing new full-season versions any more, as most users will find the prorated versions more useful.
doog7642
5/08
How do player comps fit into PECOTA projections? Are the same player comps used now that were used in the preseason projections? As I understand it, this is the calling card that makes PECOTA unique - that it is not simply a massaging of data, but data used to determine comps, from which further data was derived. Am I misunderstanding this?
cwyers
5/08
The comps are used to determine aging curves. Since we don't need to calculate additional aging for the RoS forecasts, we don't need to recompute the comps, and the comps are incorporated into the preseason forecasts that are used as an input to this.
doog7642
5/08
Thanks, Colin.
tballgame
5/08
I'm not sure PECOTA has accounted for players who started the season injured properly. Granderson's rest of the year prediction took a dive compared to the season prediction of a few days ago, despite the fact that he was not expected to have logged any time yet. I think if he were expected to play 70% of the season before the year started, PECOTA is now expecting him to play 70% of the remainder of the season.
cwyers
5/08
I've passed your comment on to Jason Martinez, who does our Depth Charts which PECOTA uses to project playing time.
mlbdepthcharts
5/09
I haven't touched that projection since there hadn't much of an update on Granderson. Now that he's set to start a rehab assignment, I'll bump it up into the 80-85% range when I do the next update in a day or two.
brokeslowly
5/08
Are the player forecasts in Team Tracker using the updated rest of the season PECOTA's?
cwyers
5/08
Yes.
newsense
5/08
Shouldn't it be "principAL changes" or are you changing the principles behind PECOTA?
bornyank1
5/08
Fixed.
rjblakel
5/09
I see PECOTA still has a love affair with Jeff Keppinger.... Ugh.
cwyers
5/09
We're forecasting him pretty much right at replacement, I'd hate to see what you'd call a hate affair.