THE BOOK--Playing The Percentages In Baseball

Maybe you remember this article?

http://www.news.iastate.edu/news/2010/feb/baseball

Kerry’s a smart guy, but isn’t very involved in the online saber community, and thus tends to reinvent the wheel.

I have no desire to verify his contest standings, although that’s mostly because I won the thing last year.

Wow, Baseball Prospectus was only slightly more accurate than predicting every team finishes 81-81 ...

Those numbers look very, very different from what I’m getting, presumably using the same dataset (I have the Vegas Watch numbers, which Kerry says he’s using). I don’t know - there have been eight days of baseball played since then, but they don’t seem to resemble the results I got on the 23rd, either.

Also, I’m curious what date those BP forecasts were pulled - they don’t line up with what I have from the fourth.

I’m going to wait for VegasWatch to do it.  Plus he also has MGL’s forecasts I think…

This is what I get, today, using the numbers Vegas Watch published at the start of the season:

PECOTA 4/4 is the numbers I have from the stat update on the morning of Opening Day; 81 is 81 wins across the board. Everything else comes from here:

http://vegaswatch.net/2010/04/2010-mlb-wins-predictions-summary.html

Sorting Colin’s data:

Forecaster     RMSE
Brown        8.82
Bukiet       8.99
Law          9.12
Henson       9.18
Average      9.21
Sheehan      9.22
Pinnacle     9.26

PECOTA 4/4   9.64
Neyer        9.67
CHONE        9.68
CAIRO        9.88
Marcel       9.97
PECOTA       10.00

Oliver        10.39
Passan        10.52
ZiPS          11.09
81            11.29

Pinnacle would be the Vegas line.  Is Henson our own John Eric Hanson (link is blocked at the office)?

We see that Marcel provides the typical “not best, not worst” baseline that it’s perfect for.

Pretty sure that it’s Steve Henson, from Yahoo! Sports.

Btw, you guys may not appreciate this, but Colin is showing both PECOTA forecasts, the last one prior to the start of the season, and, I presume, the book version.

After all, most fantasy players will draft in March, and they’d have the book forecasts and not necessarily the online forecasts.  Book sales of BPro is huge, I think over 100K.  The subscriber base is closer to 10K I think.  So, Colin definitely needs to report the book numbers if he’s going to report the online numbers.

At the same time, the other non-Marcel systems don’t have this kind of restriction.  So, it’s only fair that Colin shows both.

We don’t publish team forecasts in the book, just player forecasts. The numbers here are from wherever Vegas Watch got their figures (there’s no citation) and then what we had on the site on opening day.

(From the PECOTA player forecast tests I’ve done, yes, I’ve included the book forecasts as their own line item.)

I wasn’t able to get Vegas Watch my numbers this year.  For what it is worth, and after doing many years of team w/l projections, the uncertainty around them is huge, such that the one year results are literally next to meaningless.

Imagine that you know the exact forecast for every player on ever team and you are very good at forecasting playing time (which is more of an art than a science anyway).

Given injuries, trades, GM and manager decisions about playing time, and then the usual random fluctuations in player performance, pretty much every decent projection is going to be as good as another and none of them is going to be much better than using last year’s actual w/l or pythag or every team being 81-81.

Basically, rating team w/l projections is folly…