How much value in projections?
dechols8
Junior Member
I'm looking for opinions on the projected fantasy points from sites like RotoWire, numberFire and RotoGrinders (and others?). Are some more useful that others?
Basically, I'm trying to find out what these site's projections factor in. For example, do some include the Vegas line, or BvP stats?
I was thinking about writing a program to grab the projections from these sites, then evaluate later which were the most accurate.
Basically, I'm trying to find out what these site's projections factor in. For example, do some include the Vegas line, or BvP stats?
I was thinking about writing a program to grab the projections from these sites, then evaluate later which were the most accurate.
Comments
I'd be very interested to see the results of your program. If you're doing it, i'd recommend testing all the sites against each other, and then also adding one that's just the average of all the sites you're testing. My guess is that the average would probably perform the best, but I'd have to see the results.
Yes, among other the other info that these sites provide, they have daily point projections. Some are even accessible to non-subscribers for free. For example on numberFire.com, click on the "Projections" at the top.
The first question I want to answer for myself is to find out how much these projections change from day to day. If they don't change much, then my guess is that they are simply 'rest of season' projections and could be enhanced by factoring in the Vegas line, batting order, pitching info, etc. My hope is that this leads me to creating my own, enhanced projections.
If anyone knows of other projections that I could include in this experiment, I would greatly appreciate it.
nudder: I've also been viewing your other thread's strategy based on Vegas lines and batting order and I think that is an excellent approach also. I totally agree with your suggestion about including an average projection in the analysis as well.
I just looked at the numberfire ones for today however and noticed that ALL of the top five (in proj. pts, NOT "value") were in the LAD-COL matchup, so it's at least taking ballpark into consideration. it's probably fair to assume that batting order is not a consideration since lineups aren't announced yet, but i'd be curious to know how much other stuff like opponent pitcher, Vegas over/under, BvP (or even just batter splits vs lefty/righty) are being used.
I don't know of other projection sites (i didn't even know of these LOL), but if i find any i'll make sure to include them in here. Very excited to see where you could take this.
(And glad you're following my simple approach. i'm starting to wonder if it's more ideal for 50/50s than GPPs, but i think it's still way too early to conclude one way or the other on that.)
Here is an example of what I'm capturing for each pitcher:
Name: Stephen Strasburg
Team: WAS
Salary: 9600
RazzBall projection: 13.1
RotoGrinders projection: 14.2
numberFire projection: 13.09
RotoWire projection: 13
Average projection: 13.35
Opponent Expected Runs: 2.63
FIP: 2.89 (I calculated this number myself from the player's season stats to date)
Actual Points: 14
And, an example for each hitter:
Position: OF
Name: Michael Trout
Team: LAA
Salary: 5700
RazzBall projection: 3.9
RotoGrinders projection: 4.34
numberFire projection: 4.49
RotoWire projection: 3.8
Average projection: 4.13
Team Expected Runs: 5.22
wOBA: 0.428 (I calculated this number myself from the player's season stats to date)
Actual Points: 2
I heard back from a representative at one of the site's projections that I'm using and they said that they do include "batting trends, pitching matchups, our editors' weekly rankings, and a few other factors...we also include ballpark factors".
A second site said "...we factor weather into our projections as well as optimal lineups. With that said, we go on FACTUAL information..."
Both of these sites also stated that they do NOT use the Vegas line in their projections. I'm still waiting to hear back from the others as to what goes into their projections (if they will share that).
I think that's good info to know in terms of how the sites are basing their projections. Very interesting that one site also factors weather into their projections. I don't know of too many models out there that can do this.
Is it possible they don't offer the projection until the lineup is announced? Maybe they take consideration of where a guy is batting in the order? I noticed the Tigers lineup wasn't announced yet.
Can you currently view any projections from players in the late games with no announced lineups yet?
nudder: I think you made a valuable observation here. NumberFire's projections may be heavily weighted on a player's last game. That is something I hadn't considered. My assumption was that they already had determined that the player wouldn't play that day.
Another thing I noticed about numberFire's projections is that they change throughout the day. The other sites don't do this. So, numberFire may have a good reason to suspect that Cabrera won't play tonight -- let's see if he is in the lineup later.
First, I looked to see how each of the 4 projections (RotoWire, NumberFire, RotoGrinders and Razzball) were correlated to the actual results. The answer: Not very correlated at all -- 8.7%, 11.1%, 14.7% and 14.3% respectively. And, the average of all 4 (when all 4 had a projection for a player) was 12.9%. The best combination I could find was by averaging just Razzball and RotoGrinders -- that was 16.4%.
Second, I wanted to see how closely correlated each projection was to the season average points for each player. They were 67.3%, 78.0%, 67.0% and 63.4%. The average of all 4 projections was the best at 81.7% correlated to the season average. That tells me that just taking the season average and using that as the projection gets you most of the way there. Each site probably starts with that, then applies it own adjustments to it. Interestingly, the season average was only 7.9% correlated to the actual results. So, one might say that each of the 4 sites does something to improve on that.
One other correlation that jumped out at me was that Razzball's projection was 53.4% correlated to the team's expected runs (as I calculated it from the Vegas Odds and Over/under). The other sites were 26.0%, 18.2% and 11.9%. So, perhaps Razzball is the only one factoring that into their projection?
I may try to capture more data in the weeks ahead after the All-Star break. So far, in addition to the 4 site projections, I'm capturing:
-- Season Avg fantasy points
-- Standard Deviation (of the season avg fantasy points) -- my hope was that this would indicate who the most consistent hitters are
-- Expected team runs (base on Vegas lines)
-- Batting Order
-- wOBA
RotoGrinders to Result: 60.4%
RotoWire to Result: 56.3%
Razzball to Result: 60.4%
NumberFire to Result: 58.6%
Also, for comparison, here are the season averages and last 5 game average correlations:
Season Avg to Result: 57.9%
Last 5 Avg to Result: 51.5%
Anyway, from all of this, I made a web-based tool that I've been using for a little while now. I just got a web domain for it yesterday. So, if anyone wants to try it out and let me know what they think, feel free: dailydraftwizard.com
I'm looking for feedback to find out how it might become even more useful.
One question, when you're calculating the Pitcher's projected points, how are you calculating their % chance to pick up the W in the game?
One more question, where are you getting the strikeout/9 innings stats? I was using mine from the Steamers ROS projections on FanGraphs but I notice yours differ on some of the players.
Thanks for the feedback. I'm glad to hear this might be a time-saver.
For the pitchers projected runs and win %, I am just taking the Vegas line for the team and using that. Nothing too fancy. The projected runs for the pitchers are the runs that the opposing team is expected to score (so less is better) for the game. Same with the win % -- it is really just the expected chance that the pitchers team will win the game (derived from current Vegas lines). I'd be interested in other ways to do this if you use a different method.
The raw stats like "Strikeouts/9 innings" and "xFIP" are taken directly from Fangraphs. Specifically, this location: http://www.fangraphs.com/leaders.aspx?pos=all&stats=pit&lg=all&qual=20&type=8&season=2014&month=0&season1=2014&ind=0&team=&rost=&age=&filter=&players=
I'm not sure why we would be seeing different values.
Also, I could also add more fields/columns to these tables -- I'd be interested to hear what other stats people use to evaluate players.
As far as using the team win % to project the pitcher's win %, you're going to get a result that's always higher than it should be since the SP's don't always get the decision. From my research, they get a decision about 72% of the time on average. I'm sure that number varies for big favs or big underdogs. In other words a -200 pitcher might get a decision 80% of the time instead of 72%. But still, even then you'd only expect them to win .66*.8 = ~53% instead of 66%
You have a good point about the pitcher win % -- these numbers will be high for both winning and losing pitchers. Other than to use a flat scale (like just take 72% of the teams win expected %), I'm not sure what else to do. Have you seen other sites that have a true 'starting pitcher expected win%'?
Maybe I should just be more clear about what these columns actually represent.
Yeah the ROS stats are rest of season projections. I think they take into account career and season data. I don't know all the specifics of how the projection system works but I've read it's one of the most accurate. That said I'm sure there are a number of people out there that could derive more accurate numbers themselves particularly when accounting for the opponent and park in a specific outing.
I actually haven't seen any site that calculates the starting pitcher's expected win %. It'd be nice if there were props somewhere out there on it but I don't know of any site that offers them. The only way I can really think to figure it out would be to go back and log all the teams that were -110, -120, -150, etc. and see how often the starting pitcher recorded the win. I do know on all games, they get a decision around 72%. So if the game's a pick their chances of picking up a win should be very close to .5*.72 or 36%. I don't think it matters a whole lot when it comes to fantasy projections. If you consider a site where the points for a Win are 3 points, then let's assume a -200 pitcher gets a decision as high as 80% of the time. Now they're recording a win .66*.8 or about 53%. So it's a 17%ish difference on 3 points which is roughly .5 points and that's one of the more extreme examples.
Does anyone know what happened to this site? The last change to the site was made in Oct 2014.
This is all it says on the Baseball tab: Baseball done for the season. Back in 2015.
DailyDratWizard is cross-linked to drewby's blog, maybe he'll know if they moved or just stopped.
Looks like it's just for season-long fantasy and doesn't offer daily help. Great looking site though.