PECOTA player projections (1 Viewer)

[Mr. Pickles]

I've never seen these before and how they are formatted. Does anyone have access to an old spreadsheet that I can take a look at? I like the idea of performance distributions and am mostly curious at how BP handles these.

 

[Frostillicus]

Nerds who live in mom's basement, imo.

 

[Mr. Pickles]

Good info here. Best forum on the internet.

 

[Eephus]

I use Dan Szymborski's ZiPS projections because they're free. I couldn't tell you how Dan's methodology differs from PECOTA or the Bill James ones on Fangraphs. I think the lot of 'em tend to be overoptimistic. But that's just me. When TRE kicks off rate-a-player, my projections usually fall below the mean.

 

[Mr. Pickles]

I use Dan Szymborski's ZiPS projections because they're free. I couldn't tell you how Dan's methodology differs from PECOTA or the Bill James ones on Fangraphs. I think the lot of 'em tend to be overoptimistic. But that's just me. When TRE kicks off rate-a-player, my projections usually fall below the mean.

I'm not looking for specific values, I'm curious about any system that uses distributions. PECOTA purportedly does this, but I've never seen the data.

 

[Eephus]

The PECOTA article on Wikipedia goes into some detail about the various factors considered in the projections but they don't reveal the secret sauce recipe. This BP article gets more into the math.

This article from Fangraphs is a summary level comparison of some of the many projection systems out there. At some level, they're all weighted averages of past seasons with some adjustments for regression, park factors, aging, etc.

All of them are useful but I think it's kind of risky to aggregate their importance when evaluating players. Because their methodologies are somewhat similar, they tend toward groupthink. The rare cases when one system's projection is way off from the field are more interesting to me. Fangraphs lists three projections: Bill James, Steamer and Oliver. The latter is generally the most pessimistic with greater negative variance compared to the other two. This isn't true for all players though. Usually the three projections are pretty tightly clustered. But when there is an outlier, it's usually Oliver guessing on the low side.

 

[Mr. Pickles]

The PECOTA article on Wikipedia goes into some detail about the various factors considered in the projections but they don't reveal the secret sauce recipe. This BP article gets more into the math. This article from Fangraphs is a summary level comparison of some of the many projection systems out there. At some level, they're all weighted averages of past seasons with some adjustments for regression, park factors, aging, etc. All of them are useful but I think it's kind of risky to aggregate their importance when evaluating players. Because their methodologies are somewhat similar, they tend toward groupthink. The rare cases when one system's projection is way off from the field are more interesting to me. Fangraphs lists three projections: Bill James, Steamer and Oliver. The latter is generally the most pessimistic with greater negative variance compared to the other two. This isn't true for all players though. Usually the three projections are pretty tightly clustered. But when there is an outlier, it's usually Oliver guessing on the low side.

I must not be making myself clear.PECOTA purportedly generates a distribution of stats for individual players rather than point estimates. Well, maybe they boil this down to an average expected set of stats, but it appears to be the only system that gives you ranges for stats. I would like to see an example of this, even if it's older data.Ya dig?

 

[Eephus]

The PECOTA article on Wikipedia goes into some detail about the various factors considered in the projections but they don't reveal the secret sauce recipe. This BP article gets more into the math. This article from Fangraphs is a summary level comparison of some of the many projection systems out there. At some level, they're all weighted averages of past seasons with some adjustments for regression, park factors, aging, etc. All of them are useful but I think it's kind of risky to aggregate their importance when evaluating players. Because their methodologies are somewhat similar, they tend toward groupthink. The rare cases when one system's projection is way off from the field are more interesting to me. Fangraphs lists three projections: Bill James, Steamer and Oliver. The latter is generally the most pessimistic with greater negative variance compared to the other two. This isn't true for all players though. Usually the three projections are pretty tightly clustered. But when there is an outlier, it's usually Oliver guessing on the low side.

I must not be making myself clear.PECOTA purportedly generates a distribution of stats for individual players rather than point estimates. Well, maybe they boil this down to an average expected set of stats, but it appears to be the only system that gives you ranges for stats. I would like to see an example of this, even if it's older data.Ya dig?

The Prospectus book only gives a single stat line projection. I don't recall the website giving ranges either but it's been a few years since I had access to a subscription.In addition to the base projection, PECOTA adds percentage factors for collapse, breakout and improvement. These are kind of obvious most of the time.