What's your favorite projection system?

[Pauly]

Which projection system(s) do you tend to trust the most? I made this multiple choice, because often I try and average out several.

I was thinking about this last night when I was looking at Moncada's projections on Marcel, and he is only projected for 488 at-bats. How does one arrive at that low of a projection for a healthy 24-year-old?

Anyway, this should be interesting!

[revo]

I find the BBHQ projections through Rotolab to be just as confounding. They tend to lowball ABs and IP without reason.

The Marcel projections on Baseball Reference are pretty useless, IMO.

[Ken]

Marcel is not intended to be a good resource on a player by player basis.

Here's Nate Silver's posts about it after he created it. http://tangotiger.net/marcel/ http://www.tangotiger.net/archives/stud0346.shtml

it is the most basic forecasting system you can have, that uses as little intelligence as possible. So, that's the allusion to the monkey. It uses 3 years of MLB data, with the most recent data weighted heavier. It regresses towards the mean. And it has an age factor.

Note: I do not stand behind these forecasts. These forecasts are the minimum level of competence that you should expect from any forecaster. Do not attach my name to these forecasts in any kind of evaluation experiment. They should only be referred to as Marcel The Monkey Forecasting System, or simply The Marcels.

Determine the projected PA = 0.5 * 2004PA + 0.1 * 2003PA + 200. I take the result of #3, and prorate it to this projected PA.

For Moncada that's simply 0.5 * 559 + 0.1 * 650 + 200 = 545

Subtracting walks gets you to 488 AB.

Again, these are the projections a monkey would give you, based on what Nate was going for.

Marcel would today project Luis Severino to pitch 85 IP this year even though we know he'll pitch 0. There's no intelligence involved, it's just a formula based on past years.

[Pauly]

Marcel is not intended to be a good resource on a player by player basis.

Here's Nate Silver's posts about it after he created it. http://tangotiger.net/marcel/ http://www.tangotiger.net/archives/stud0346.shtml







For Moncada that's simply 0.5 * 559 + 0.1 * 650 + 200 = 545

Subtracting walks gets you to 488 AB.

Again, these are the projections a monkey would give you, based on what Nate was going for.

Marcel would today project Luis Severino to pitch 85 IP this year even though we know he'll pitch 0. There's no intelligence involved, it's just a formula based on past years.

Thanks Ken, I didnt know that (obviously)!

[Ken]

Thanks Ken, I didnt know that (obviously)!

No problem!

I have historically used some version of steamer. Steamer itself has projections, Rudy uses Steamer as his base. ATC uses portions of different projections systems and gives them weights. So some of these options presented are variations of the same data.

I like this post that you started. I think the more interesting question to start going into within the comments is why you use a certain projection system and then discussions of how each projection system is different.

I haven't finalized my platform this year. If I had to guess I'm probably going to use some combination of ATC and a playing time projection from Depth Charts or Razz. I definitely won't be using Shandler.

[joncarlos]

My suspicion is that it's supposed to be this way... but ATC always ends up mirroring the triple slash lines from Steamer/Depth Charts. Or at least coming very close. Then again, so does ZIPS.

So you're looking at using one of those, which basically have their own way of figuring out playing time but otherwise end up about the same... or using something totally different (PECOTA, BAT, HQ, Razz, etc).

I usually just end up using the Fangraphs depth chart weighted projections which is basically just Steamer with a human allocating the playing time a bit more finely.

[Ken]

How is there not a projection system in place yet using the statcast data?

[joncarlos]

How is there not a projection system in place yet using the statcast data?

It would be pretty easy to make a Marcel version of xBA/xSLG but figuring out the HR would be harder, and I have no idea how you figure out R/RBI. But there are a lot of smart people out there so maybe someone is working on it.

[GwynnInTheHall]

A six pack of craft Double IPAs and whatever fantasy mag is at the liquor store on draft day.

[bryanbutler]

I download as many as I can get my hands on and take medians and MADs (median and MAD are just robust versions of average and standard deviation). In some situations you might want someone with low variance - reliability. In some, you might be speculating more, so it's not a bad thing to go for someone with higher variance.

I've been having an email conversation with Merv at RotoLab to try to get him to allow import of other projections into the program. Like others, I'm not a huge fan of the BBHQ projections. Right now it's next to impossible to change stats because you have to hand-type them in. Each stat, separately, for each player. Ugh. He can't support automatic retrieval of other projection systems because he has a contract with BBHQ. But he could allow import if it's up to the user to get the data into some specified format. However, he's reluctant to do it because he's afraid that folks will muck it up, and user support will be a nightmare .

[mjl]

How is there not a projection system in place yet using the statcast data?

I think the Pod Projections (Mike Podhorzer) use that but I've never tried them. Maybe this year? https://www.projectingx.com/baseball...r-projections/

[The Dane]

I put CBS... for the irony.


I don't really trust any ONE source.

[cavebird]

A six pack of craft Double IPAs and whatever fantasy mag is at the liquor store on draft day.

Baseball is old school. Can't trust the new-fangled craft IPAs, no less the Double IPAs. Sheez.

[bryanbutler]

Baseball is old school. Can't trust the new-fangled craft IPAs, no less the Double IPAs. Sheez.

In Dennis Hopper voice (from Blue Velvet): "Heineken! Fuck that shit. Pabst Blue Ribbon!"

.