[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
THE BOOK cover
The Unwritten Book
is Finally Written!

Read Excerpts & Reviews
E-Book available
as Amazon Kindle or
at iTunes for $9.99.

Hardcopy available at Amazon
SABR101 required reading if you enter this site. Check out the Sabermetric Wiki. And interesting baseball books.
Shop Amazon & Support This Blog
RECENT FORUM TOPICS
Jul 12 15:22 Marcels
Apr 16 14:31 Pitch Count Estimators
Mar 12 16:30 Appendix to THE BOOK - THE GORY DETAILS
Jan 29 09:41 NFL Overtime Idea
Jan 22 14:48 Weighting Years for NFL Player Projections
Jan 21 09:18 positional runs in pythagenpat
Oct 20 15:57 DRS: FG vs. BB-Ref

Advanced

Tangotiger Blog

A blog about baseball, hockey, life, and whatever else there is.

Tuesday, November 01, 2016

WAR Marcels ... WARcels?

?The Marcels is a simple forecasting system.  It is not only simple, but transparent.  It is so simple and transparent that multiple people have implemented it on their sites or provided the code.  I introduced it about 12 or 13 years ago, and have not made a change to it.  It has taken on all-comers, and held its own.  There are better ones.  But those are better in the way that an 82, maybe 83 win team is better than an 81 win team.  The goal of Marcel was always to be the minimum acceptable baseline, to be so simple and transparent that anyone can do it, and hopefully use it as their own core, where they would improve it on the periphery.  There are many systems, or WERE anyway, that would be like a 73 or 75 win team, clearly below Marcel, not able to luck their way into being better. Those guys were always the pollution to forecasting, getting in the way, and I needed a way to mute those systems.  That's what Marcel was, is, and will-be.  That's why you see ZiPS and Steamer and MGL providing forecasts.  They are not pollution.

Now, how about something EVEN SIMPLER?  WAR already encapsulates a player's season.  And we have multiple years for a player.  Why not come up with something that ONLY uses a player's WAR?  None of his components, none of his playing time, nothing except his WAR.  And, I have to make it simple and transparent.  And, not only forecast the upcoming year, but future years.  The WAR Marcels... WARcels?

As usual, when it comes to creating a forecasting system, you go down that rabbit hole.  You go down far enough and you are tempted to look at every little variable, improving it on the periphery, maybe making inroads for 1% of the players. But there's a reason that The Marcels has staying power: simple, transparent. That's the goal, that's the constraint.

Forecasting Year T+1: 

Step 1: Take 60% of year T, 30% of year T-1, 10% of year T-2.  Let's look at Edwin Encarnacion.  For this example, I'm going to use the Baseball Reference version of WAR (rWAR).  Later this week, I will do this for Fangraphs (fWAR) to confirm that this methodology will hold, and how the results will differ, if at all.  His rWAR the last three years is: 3.7, 4.7, 3.6.  That gives us a weighted average of 4.0.

Explanation: now you may think that the weights  are too aggressive for the current time period, given that Marcel follows a 5/4/3 for hitters and 4/3/2 for pitchers model.  However, that weighting scheme is for rate stats.  For playing time, it uses a more aggressive 5/1/0 scheme.  And since WAR is a combination of rate and playing time, we need a weighting scheme somewhere between the two.  And a 6/3/1 fits the bill.

Step 2: Regression.  Simply take  80% of the weighted WAR.  Encarnacion is now at 3.2.

Explanation: now you may think we need playing time.  And you'd be right, sort of.  But given the constraints here of simply focusing on WAR, and given that WAR itself purports to represent itself as an overall metric, using playing time would undermine WAR.  Indeed, what you'd want instead is WAR/PA and WAR/IP, which if you do that, you may as well do WAA/PA and WAA/IP.  Which if you do that, you may as well rely on wRC+ and ERA+.  Which if you do that, you may as well use The Marcels.  (And eventually I will create something more granular, more based on components, more based on Statcast.) The idea for this metric is to NOT use The Marcels, but come up with something simpler than the most simple system.  You have WAR in hand, let's just use that.

Step 3: compare the player's age in year T to the age of 30, where age is simply year T minus birth year.  For every year away from age 30, add or subtract 0.1 wins.  Obviously, add if he's under 30 and subtract if he's over 30. EE was born in 1983, which makes his calculated age 33 for the 2016 season, or 3 years beyond the peak of 30, or another 0.3 wins.

Explanation: A player who has a weighted WAR at age 28 of 4.0 and another player who had a weighted WAR at age 38 of 4.0 have historically shown to be around 3.2 the following year if 28 years old and 2.2 if 38 years old.  Age makes a big difference.

So, for the 2017 season, Encarnacion gets a forecasted WAR of 2.9.

You may be thinking "darn, that is LOW!  We started at 4.0 and we're down to 2.9?"  There were 59 nonpitchers born since 1931 with a weighted WAR of between 3.5 and 4.5.  In the following year, they averaged 2.8.  This goes from a near high of his mate Bautista who at age 34 got a 6.1 WAR down to Nick Swisher of negative 1.2 WAR.  Don't like Swisher as a comp?  That's ok, other negative WAR at age 34: George Foster, Willie McCovey, Bobby Bonds.

Forecasting Year T+2 through T+5: 

Year T+2: Start with your forecast of year T+1, and then subtract 0.4 wins.  Then apply a further adjustment based on age.  Compare his year T age to 30 and add or subtract 0.08 wins. EE gives us 2.89 minus 0.4 minus 0.24, or 2.25.

Year T+3: Take Year T+2, subtract 0.4.  Compare his age to 30, and add or subtract 0.03 wins for each year away.  EE gives us 2.25 - 0.4 - 0.09, or 1.76.

Year T+4: Take Year T+3, subtract 0.4.  Compare his age to 30, and add or subtract 0.03 wins.  EE gives us 1.76 - .4 - .09, or 1.27.

Year T+5: Take Year T+4, subtract 0.4.  Compare his age to 30, and add or subtract 0.03 wins.  EE gives us 1.27 - .4 - .09, or 0.78.

Encarnacion Comps:

So, over the next five years, his WAR forecast totals 9 wins.  How does that compare to his comp group of 55 players (it was 59, but we lost guys who are too recent to give us 5 year forecasts)?  Their 5-year actual WAR was 10 wins.  That's on average.  His best-case among recent players includes David Ortiz, Manny Ramirez, and Chipper Jones. His top 25th percentile averaged 18 wins.  His worst-case scenarios includes: Bobby Bonds, Jim Rice, Albert Belle. His bottom 25th percentile averaged 2 wins.  As you can see, forecasting is very difficult, since anything can happen.

So, there you have it... The WAR Marcels.

***

With this forecasting model as a framework, look for a deeper dive as they relate to this year’s free agent class on MLB.com in the coming days and weeks.

(9) Comments • 2018/03/20 • Forecasting

Latest...

COMMENTS

Nov 23 14:15
Layered wOBAcon

Nov 22 22:15
Cy Young Predictor 2024

Oct 28 17:25
Layered Hit Probability breakdown

Oct 15 13:42
Binomial fun: Best-of-3-all-home is equivalent to traditional Best-of-X where X is

Oct 14 14:31
NaiveWAR and VictoryShares

Oct 02 21:23
Component Run Values: TTO and BIP

Oct 02 11:06
FRV v DRS

Sep 28 22:34
Runs Above Average

Sep 16 16:46
Skenes v Webb: Illustrating Replacement Level in WAR

Sep 16 16:43
Sacrifice Steal Attempt

Sep 09 14:47
Can Wheeler win the Cy Young in 2024?

Sep 08 13:39
Small choices, big implications, in WAR

Sep 07 09:00
Why does Baseball Reference love Erick Fedde?

Sep 03 19:42
Re-Leveraging Aaron Judge

Aug 24 14:10
Science of baseball in 1957

THREADS

November 01, 2016
WAR Marcels ... WARcels?