THE BOOK--Playing The Percentages In Baseball

do ever get any actual work done?

tango, what are your thoughts on the PFM tool in general? I stopped using it a couple years ago, in favor of the Last Player Picked tool.

I didn’t like the fact that PFM had so many options and was a bit of a black box. never could figure out why there should valuation options at all, really.

Tango,
When you say you don’t do an SP/RP split, does that mean that all pitchers are lumped together and you don’t use a different “replacement level” for starters and relievers? 

I value players the way you’ve described on this site and the way Last Player Picked does it (I believe they are pretty much the same), by creating z-scores, and I’ve always determined a starting pitchers value above replacement based on the replacement level starting pitcher and relief pitcher value above replacement based on the replacement level reliever.  Would you just throw all pitchers into a group and use a replacement level pitcher as the baseline for both starters and relievers?

I’ve never used it other than when prompted by specific queries.

I’ve posted my valuation system and process.  I don’t think it should be black-boxed, especially since I consider it pretty simple to do to begin with.

I mean, I can put out yet another challenge and construct some really weird league structure, and have forecasters submit their draft order and see how they do.  Heck, maybe I will do that next year.  My contention is that a Marcel-like valuation system gets the job done.

Bobby/3: Unless you are specifically limited to a set number of starters and relievers, you absolutely, definitely, and emphatically CANNOT have two different baseline points.

Suppose you were told to draft 14 hitters, with no positional quota.  Would you STILL value catchers more, all other things equal?

***

anon: actually, I just performed some SQL testing this morning, and fixed something that was causing major problems here.  I also did some side non-office work for someone as well.  I spoke to my wife too.  No bathroom break yet.  I have to call my mom soon (like now).

But, things at the office goes in waves, so in downturns, I try to do my best to do everything else.

Tango, thanks for the response.  I play in one league where every team must have 5 starting pitchers and 2 relievers active in their lineup and there is no bench, so I take it that in this league I should split out the baselines for starters and relievers.  But in a standard Yahoo league (2 SP, 2 RP, 4 P), I wouldn’t.

How about a league that has 4 active starting pitchers and 2 active relief pitchers each week, with any number of bench pitchers?

Right, the 5/2 is akin to the OF/C split.

As for your weekly league, there’s an extra component where the starting pitcher would be pitching once or twice in that week.  It’s harder to say without knowing more details.

"I mean, I can put out yet another challenge and construct some really weird league structure, and have forecasters submit their draft order and see how they do.  Heck, maybe I will do that next year.  My contention is that a Marcel-like valuation system gets the job done. “

It’s not too late to ask for a second list for this year.  I believe you set the challenge up to model a 5x5 league so the Marcel-like (z-score valuation process) is effectively the default scoring system (or it was designed to be, perhaps the scoring values would need to be adjusted slightly).  A second ordered list, to be evaluated based on 5x5 scoring, would be easy enough.  I would guess that most of the participants are pro z-scores and would use the same list for both evaluation metrics.

I was thinking of challenging on the league structure. Say: you can select a maximum of 2 SP, you must select at least 4 catchers, and so on.  Really push the boundaries beyond what their models were designed for.

And have everyone use the same player forecasts.  So, I’d give everyone the Marcel forecasts let’s, and they give them the league rules, and have them submit a dollar value for each player.  It’s a challenge purely on valuation, not on forecasting.

you could almost ask for valuation models based off of last years projected Marcels (or any common data set). have the “draft” then compare it to 2010 actual MLB stats.

or even do it kind in reverse, where you ask people how they value the 2010 actual MLB results, and THEN hold the draft. and then have a draft and see who ends up best.

since you are working with common stats, whether or not use you 2010 actuals or 2011 actuals or 2010 CHONE wouldn’t really matter, right?

#9/#10

I agree testing a valuation system separate from a projection system is ideal.

I think the optimal way to do that is as Ken lays out in #10 (run the test on a set of stats known in advance).  You could use the Marcel projections for that and judge the outcome based on the Marcel projected stats for 2011 (and not the actual stats for 2011).  Under those constraints, a “challenge” can be held any time.

Ken: right, I can use the 2010 stats, and ask everyone to give us $ valuation on 2010 stats (but use a very weird league structure to challenge their system, like say having a category triples x homeruns, or 5 catchers).  And right, once that happens, I can hold a draft based on their valuations, and then simply use the actual 2010 stats to see the results.

***

Actually, one place it would matter is this: say that wins was ALL random.  Then the value you place on wins should be 0.  But, if we are using OBSERVED wins, and we are valuing wins, then those will NOT have a value of 0.

So, you do need to be careful about that.

more food for thought: I’m not sure this would answer the old 70/30 question, or “what % of my budget should I spend on pitching?”. running valuations after the fact, I think, will generally tell you that the best SP were worth more than they went for in the draft. but, they are discounted on draft day (I’m guessing/surmising) b/c of greater injury risk and greater projection uncertainty.

Meant to add: so, a system which artificially limits the amount a user can spend on pitching (Last Player Picked has this option, as do most valuation tools) might fare worse than a system which doesn’t, but in reality this *may* not be the optimal way to draft? So you might want to control for this? Not sure.

Alright, hopefully you won’t mind this question, which is somewhat related to the topic at hand.
A friend of mine is joining this league where there are 10 hitters and 9 pitchers. The pitchers can be anything, and the 5 roto categories are QS, K, ERA, WHIP, and S. The league does not have an IP limit; it’s on ESPN so it has a 200 GS limit for pitchers. I put his league into LPP and it says that there should be like 5 relievers per team.

I’m not questioning the results, but does anyone have any idea why they (counterintuitively, I think) suggest that? You’d need at least 6 SP to reach the GS limit. If you choose to not attempt to reach the GS limit, you can’t really expect to be competitive in K and QS I would think. Would that really be the optimum roster composition?

I tell him that it’s like the chapters on bunting and stealing in The Book. If other owners are behaving suboptimally, you need to adjust to take advantage of them. So my thought is that he should have 1 more RP than everyone else. But I’m not 100% sold on my thinking.

@13 / Ken

The split in value between pitching and hitting with known stats, no replacement (i.e., the stats for the players we acquire in the draft are the only stats we get all year) and a fixed draft pool should mirror the value in points allocated to pitching and hitting.  For the familiar NxN league configurations that means a 50-50 split.

Since the draft pool is unlikely to be fixed there can be some variation. 

The “discount” you see on pitching in an auction has, at least, 4 contributing factors:

1.  Uncertainty (there is generally more in pitcher stats as well more injury risk)
2.  Starter/Reliever Dichotomy (Pitchers aren’t homogeneous (starters and relievers are actually two separate pools of players.  In other words, not all players contribute in all categories.))
3.  Saturation (because of 1 and 2, a smaller percentage of the stats available are actually bought at auction)
4.  Precedent (valuing pitching at 50% against 11 teams that value pitching at 30% will result in a massive hitting deficit).

JEH: That is not correct for a league with a different number of hitters and pitchers.

Right, you need to know number of hitters and pitchers… IF YOU DON’T KNOW THE STATS. Imagine you have a league with 24 hitters and 1 pitcher. 

However, with known stats (which is the key point here), are you necessarily going to have a 50/50 split, regardless of the number of hitter/pitcher split?  I think that’s possible.

#17/18

uber-

Tom (18) clarifies nicely for his process.

The number of players does not matter; as long as enough granularity exists in the stats.

Post 15 in this thread:

http://www.insidethebook.com/ee/index.php/site/comments/economics_of_fantasy_valuation/

touches on it.

Tom-

I understand your point about needing to know the number of hitters and pitchers if you don’t know the stats (i.e., you are working off of projections) so you can calculate the standard deviation for the category, but I consider that a meta-issue. 

It’s separate from the valuation tied to the stats (and, I believe, you are even calculating the variance on the actual stats of the preceding season or some other collection of data different from the projections being used).

I won’t guarantee there is no impact on the discretionary pricing split based on the number of players selected from each pool but it would surprise me.  I do expect the number of players in each pool to have some impact on the value of players from with-in that pool.

@19

I just want to clarify my comment:

“I won’t guarantee there is no impact on the discretionary pricing split based on the number of players selected from each pool but it would surprise me. “

I was referring to my process for valuing players.

Anyone using the z-scores method Tom outlines on this site ( http://www.insidethebook.com/ee/index.php/site/comments/the_worth_of_sb_hr_and_all_other_categories_in_fantasy_baseball/ ) will, of course, get different splits depending on the number of players.

JEH / Tango: I always enjoy reading your thoughts on this, tx.

since the Last Player Picked model is easy to play around with, here are some optimal spending allocations it gives me in different scenarios. All use 2010 actual stats, all %’s are allocation to hitting budget.

5x5, 9 hitters + 9 pitchers: 53%
5x5, 13 hitters + 9 pitchers: 66%
5x5, 9 hitters + 13 pitchers: 42%
10x5, 9 hitters + 9 pitchers: 27%
5x10, 9 hitters + 9 pitchers: 53% (???)
5x5, 9 hitters + 63 pitchers: 24% (???)

those results in the wacky scenarios intuitively don’t seem right to me.

Perhaps the extra 5 categories in pitching are pretty random?

***

I’m not sure why the 9/63 is “???” at 24% for hitters.  What did you expect?

Don’t forget, all the extra pitches beyond a certain number are random pitchers as well: they literally will add almost nothing to the results because they only pitched 3 innings.

***

The questionable one to me is this:

10x5, 9 hitters + 9 pitchers: 27%

Again, it all depends on the extra categories.  Are they random and/or “opposite” of the other 5 categories?  Are we talking about strikeouts and hit batters and sac flies or something?

ah, yeah, you are just adding a bunch of garbage pitchers really, so most aren’t worth much.

the extra hitting cats I used were:

OBP
SLG
OPS
H
BB

so a lot of overlap within those.

the extra pitching cats were:

BB
H
HR
W-L
2W - L

(I just used the next 5 on the tick box the site has, I wasn’t thinking about whether that was smart or not)

Right, you have to be careful about the non-independence of it, and whether it’s a rate stat or counting stats.

I don’t know how that site works, but the split may be calculated on total dollars while the discretionary dollars may remain close.  I.e., 63 pitchers results in a lot of forced $1 players, assuming a $1 minimum bid.

As far as categories go, they generally have little impact on the split.  One scenario under which they do have an impact is when the categories effectively split a pool of players.  For example, if there were three pitching categories and they were Wins, Saves and Holds you are essentially drawing from three separate player pools, not one.  It would be interesting to see how the site handles those three categories compared to, say, Wins, IP and ERA.

JEH:

3x3 league, 2010 actual stats. 9 hitters plus any 9 pitchers:

R, HR, SB x Wins, ERA, IP

Hitting/Pitching Split: 34/66

R, HR, SB x Wins, Saves, Holds

Hitting/Pitching Split: 70/30

#15, Zac:

“I’m not questioning the results, but does anyone have any idea why they (counterintuitively, I think) suggest that?”

The easy answer is that it doesn’t know anything about a GS minimum. So, basically, even though you can start whatever combination of SP and RP you want, the 200 GS restriction means that teams are really forced to start 6+ SP. In which case I recommend just entering 6 SP/3 RP to actually get a realistic team, instead of 9 P.

Ken/#26

Thanks.

That’s actually a very instructive pair of splits.

The hitting categories you selected have (I am assuming) a lower correlation than the first set of pitching categories so the pool of potential players is deeper which leads to fewer resources being allocated to hitting in that case.  If you change the hitting categories to, for example, HR, SLG, RBI then hitting would likely (I am guessing again) get over 50% compared to ERA, W, IP. 

It looks like that site does a good job of determining the player pool to pick from based on the categories in use.

JEH/#28:

You’re a pretty good guesser. In that scenario the split goes to 48/52.

Also, I’m sure Mays/#27 will be happy to hear your thoughts, since he runs that site!