The Baseball Prospectus PECOTA team projections are out

[Thrylos]

Here.

 

They project the Twins to tie with the Tigers for second in the AL Central with 77 wins, ahead of the White Sox and Royals and behind the 91 win Indians.

[Vanimal46]

65:1 Pennant odds.... Sounds like a value bet to me! 

The Royals IMO have the widest range of possibilities in the division. I could see them riding their existing core to another over .500 season and sniff at a WC berth. I could also see them tearing down the core by July and losing 90+ games. 

[Thrylos]

 

65:1 Pennant odds.... Sounds like a value bet to me! 

The Royals IMO have the widest range of possibilities in the division. I could see them riding their existing core to another over .500 season and sniff at a WC berth. I could also see them tearing down the core by July and losing 90+ games. 

 

 

The Pennant odds are not part of the PECOTA calculation.  These are Vegas odds and they are using them as a comparison point.

[biggentleben]

Baseball Prospectus...not Baseball Reference. Just wanting to ensure the right company gets credit! Nate Silver didn't work for Baseball Reference.

[markos]

The projected standings have changed a lot in the last couple of days. The Twins are now up to 80-82.

 

http://www.baseballprospectus.com/fantasy/dc/

 

<shrug/>

[Mike Sixel]

 

The projected standings have changed a lot in the last couple of days. The Twins are now up to 80-82.

 

http://www.baseballprospectus.com/fantasy/dc/

 

<shrug/>

 

interesting, you can see the dates of the updates, the Twins have the most recent, and some have not been updated in a month.....

[TheLeviathan]

So they think we'll be ok and our division will be terrible.  Interesting.

[Doomtints]

Hm. The win projections look like rubbish across the board to me. I just lost faith in PECOTA, whatever formula they are using clearly is regressing to the mean too much. 

[Mike Sixel]

 

Hm. The win projections look like rubbish across the board to me. I just lost faith in PECOTA, whatever formula they are using clearly is regressing to the mean too much. 

 

most have not been updated since early Jan......not sure you noticed that. And some are as of yesterday....

[Thrylos]

 

most have not been updated since early Jan......not sure you noticed that. And some are as of yesterday....

 

Wonder if that has to do with transactions.  + Belisle + Adrianze - Park - Light = 3 wins it seems

[Mike Sixel]

 

Wonder if that has to do with transactions.  + Belisle + Adrianze - Park - Light = 3 wins it seems

 

Hard to say what changed in what part of the model. I won't pay attn until all the teams are updated.

[tarheeltwinsfan]

Possibly Erroneously Calculated Organzational Team Averages   Hmmm.

[Taildragger8791]

 

Wonder if that has to do with transactions.  + Belisle + Adrianze - Park - Light = 3 wins it seems

 

I can't imagine these transactions impacted the assessment that much considering Light was expected to have a minimal role this year, and Park's production can't be that much worse than his main replacement in Vargas. Adrianze isn't going to supplant anyone without some serious injuries or total non-performance by Polanco & Escobar. There has to be something else driving these seemingly random predictions.

[Oldgoat_MN]

 

I can't imagine these transactions impacted the assessment that much considering Light was expected to have a minimal role this year, and Park's production can't be that much worse than his main replacement in Vargas. Adrianze isn't going to supplant anyone without some serious injuries or total non-performance by Polanco & Escobar. There has to be something else driving these seemingly random predictions.

 

Belisle is clearly more highly regarded by Baseball Prospectus than by TD boards!

[goulik]

Interesting stuff but I am not betting for anyone projected to lose more than they win just because Vegas thinks they will be more under 500 than the others predicting outcomes. (besides the fact that I don't gamble ever...)

[mazeville]

I think it's fair to say the Twins could improve. Projections are always questionable, and too many things can go right/wrong. But you can imagine improvement from Buxton and Sano and some better overall defense. And it's awfully hard to imagine the pitching staff getting any worse. 

[Oldgoat_MN]

To them it's all about regressing to the mean.

 

Does anyone else think only 3 teams will win >90 games?

Or only 2 teams will lose >90 games?

Last year 6 teams >90 wins, 7 teams >90 losses.

 

Doesn't occur as much like projections as covering your a**.

[Mike Sixel]

 

To them it's all about regressing to the mean.

 

Does anyone else think only 3 teams will win >90 games?

Or only 2 teams will lose >90 games?

Last year 6 teams >90 wins, 7 teams >90 losses.

 

Doesn't occur as much like projections as covering you a**.

 

No, it's about most likely outcome....and since people won't accept a range with error bars and uncertainty measures, they give one number. Of course everyone knows more than 3 teams will, but saying which is not likely to be any more correct than not saying.

[Doomtints]

 

No, it's about most likely outcome....and since people won't accept a range with error bars and uncertainty measures, they give one number. Of course everyone knows more than 3 teams will, but saying which is not likely to be any more correct than not saying.

 

The authors are clear that PECOTA purposely regresses to the mean. By these results it is obvious that their algorithm does this too much. Though yes it's most likely that the largest segment of teams will be within a standard deviation, there should still be some outliers.  

 

If you plot this on a bell curve (something BP should have done to check their work), with 81 as the mean (.500) and 10% of the season as a standard deviation (16 wins), you will find quickly that all but one of these teams are within 1 one standard deviation of one another. This means if you sum the good and bad teams in 2017, you get a grand total of one. Every other team is average.

 

If you cut the standard deviation in half (8 wins), all but 8 teams are within a standard deviation. Does this seem right to you? The sum of good + bad teams in 2017 will only equal 8, and only then if you bend the standard deviation to 5%?

[drivlikejehu]

 

The authors are clear that PECOTA purposely regresses to the mean. By these results it is obvious that their algorithm does this too much. Though yes it's most likely that the largest segment of teams will be within a standard deviation, there should still be some outliers.  

 

If you plot this on a bell curve (something BP should have done to check their work), with 81 as the mean (.500) and 10% of the season as a standard deviation (16 wins), you will find quickly that all but one of these teams are within 1 one standard deviation of one another. This means if you sum the good and bad teams in 2017, you get a grand total of one. Every other team is average.

 

If you cut the standard deviation in half (8 wins), all but 8 teams are within a standard deviation. Does this seem right to you? The sum of good + bad teams in 2017 will only equal 8, and only then if you bend the standard deviation to 5%?

 

Aside from the fact that standard deviations don't work that way (it's no longer "standard" once you invent arbitrary values), you first have to define what makes a projection successful.

 

If the goal is to reduce the root mean square error, then the PECOTA result is entirely plausible. If the goal is not to minimize error, then that alternate goal needs to be defined, and it should be explained why this alternate method is preferable despite the fact it will almost always be less statistically accurate.

[HitInAPinch]

I'm on the side that says: The Twins couldn't possibly suck as much this year as they did in 2016. 

 

I had a spreadsheet on that, but my dog ate my laptop.

[old nurse]

 

The Pennant odds are not part of the PECOTA calculation.  These are Vegas odds and they are using them as a comparison point.

Why would one use Vegas odds when Vegas is about extracting the most money they can from someone?

 

[gunnarthor]

80 wins seems high but it's not impossible if the pitching rotation stabilizes. After last years disaster it's easy to forget how much young talent the Twins have. I think 75 wins is more likely but they should be a fun team to watch this year.

[Doomtints]

 

Aside from the fact that standard deviations don't work that way (it's no longer "standard" once you invent arbitrary values), you first have to define what makes a projection successful.

 

That's the entire point. BP has not done this step in spite of having 115 years of history to compare to. These numbers are useless.

[drivlikejehu]

 

That's the entire point. BP has not done this step in spite of having 115 years of history to compare to. These numbers are useless.

 

Huh?

[drjim]

Hey, 20% chance the Twins can compete for a playoff spot is looking good!

[Thrylos]

Just noticed that they project the Twins to have the second best in the AL and third best in the majors team as far as defense goes, with 35.1 defensive runs above the average team.  Pretty sure that this drives that 80 wins projection.  Not sure where they came with that 35.1 number, and I could not found how the Twins did in that particular metric last year, but based on fangraphs' defense metric, they were the second worst team in baseball in 2016 with -45.7, so PECOTA really has then for a huge improvement for some reason