MLB DFS FLOOR AND CEILING PROJECTIONS
One of the new features of the DailyRoto Premium MLB Product that we’re excited to launch is MLB DFS floor and ceiling projections that help quantify a player’s range of outcomes. For both hitters and pitchers. we will have projections for their 10th, 25th, 50th, 75th, and 90th percentile outcomes.
In addition to providing percentile projections, we’re committed to making this information as actionable as possible. As a result, our percentile projections for hitters and pitchers are fully integrated with our optimizer. This means that:
- In addition to optimizing off of our standard projections, you can now optimize directly off of our percentile projections for hitters and/or pitchers, which can be useful if you’re setting cash game lineups (may focus on floor/median for pitchers) or tournament lineups (may focus on ceiling for hitters and pitchers).
- If you enjoy the fact that our projections are customizable, this applies to the percentile projections as well. If you think our baseline wOBA for Mike Trout is too low, you can change it. Not only will his standard projection immediately update to reflect the new wOBA baseline, but his percentile projections will immediately update as well.
What does that mean? It’s pretty similar to how Baseball Prospectus does their seasonal percentile forecasts: https://legacy.baseballprospectus.com/glossary/index.php?mode=viewstat&stat=2. Let’s use Mike Trout as an example. Here are his projected percentile outcomes for Opening Day:
A 25th percentile projection of 2.9 DraftKings points means that Mike Trout is expected to score that many points or less 25 percent of the time. Put another way, 75 percent of the time we’d expect Mike Trout to score more than his 25th percentile projection. Conversely, we’d expect Trout to score 23.46 DraftKings points or more, his 90th percentile projection, just 10 percent of the time.
A couple of baseball specific notes:
- You’ll notice that Mike Trout‘s 10th percentile projection is 0. There is actually no combination of hitter and context where a hitter’s 10th percentile projection isn’t 0. More on this later.
- Of course, it’s actually impossible for Mike Trout to score a fractional amount of points given the DraftKings scoring system. There’s some probabilistic math occurring that results in the fractional projections (same as for our standard projections).
It then follows that a player’s 50th percentile outcome is their median projection, since we’d expect them to score less than that amount half of the time and more than it the other half of the time. Generally speaking, a hitter’s median projection in MLB DFS is lower than their mean projected outcome (~ our standard projection). This occurs because the standard MLB hitter has a fantasy distribution that is skewed to the right (right-tailed distribution). This is not necessarily the case for pitchers, whose outcomes more closely resemble a normal distribution.
Generating Percentile Projections From Player Baselines
How did we come up with these projections? Again, the Baseball Prospectus explanation is a great starting point. The explanation linked above states, “Pecota [their projection system] runs a series of regressions within the set of comparable data in order to estimate how changes in peripheral statistics are related to changes in equivalent runs.”
Let’s try and apply that to what we’re doing. We can use the past two years of data, where we have not only all of the DailyRoto fantasy point projections saved and lined up with their actual results, but we have all of the individual baseline projections saved. In the Baseball Prospectus quote, think of “peripheral statistics” as “DailyRoto baselines” and “equivalent runs” as fantasy point projections.
Using this methodology, at each percentile outcome for each position we’re able to weigh the importance of our different baselines. We’ll get into this more in a little bit, but a quick example is this. For a hitter’s floor, our projected wOBA is one of the more important metrics. However, for a hitter’s ceiling, our projected ISO plays a much more important role than projected wOBA. This is somewhat intuitive. Hitters who get on base consistently are more likely to have a higher wOBA and more games where they reach base at least once. However, the ceiling for hitters is driven by big events and, most importantly, home runs. ISO is a better indicator of that than wOBA.
If you’re unsure what is meant by baselines, it refers to most of the factors that go into or directly make up our standard projections. Since we’ve tracked these baselines historically, we’re able to train a model to understand which of these baselines has a larger impact on floor/median/ceiling. So, things like batting order, team total, projected wOBA, projected ISO, projected stolen bases, etc. will all effect the percentile projections in different ways and have a higher or lower relative importance as the percentiles change.
Why Not Use Player Specific Distributions
Why do it this way, rather than using player specific distributions? For NFL, the answer to this question is simple: sample size. For MLB, though, the sample size is rather large for most players. So, why not use a player’s past distribution as a factor, perhaps the primary factor, in determining their range of outcomes?
Despite the larger sample size, we think a player’s past distribution of fantasy points can be noisy or irrelevant. For starter’s, we’d rather focus on the general distribution that we see for a player’s skill type (their wOBA, ISO, stolen base combination for example) than narrow the scope to a single player’s distribution and assume that it carries more weight than the distribution of all similar players.
Most importantly, though, we want to know what a player’s range of outcomes is based on their current context. Using a specific player’s past distribution often times does not reflect this context. A player’s role, context, and matchups are always changing. If Buster Posey is playing a game in Coors Field (best hitter’s parks in MLB), using his distribution of outcomes that includes roughly half of his games in San Francisco (one of the worst hitter’s parks in MLB) doesn’t help us arrive at the right answer. Consider some other scenarios:
-a player who generally hits 7th in the order for a team that is low scoring, suddenly finds himself hitting leadoff on a night that team has a 5.0 implied run total
-a player on a brutal cold streak who is reportedly playing through injury and his skill baselines have decreased a result
We’ve mostly focused on the hitter point of view so far, but you can easily think of examples for pitchers as well where the opposing lineup, recent velocity and pitch type data (could affect skill baselines), the assigned umpire, the ballpark, a change in usage (projected outs may vary differently than in the past) etc. can all meaningfully change their expected distribution for the current day dramatically from their historical distribution.
Categorical Importance and Actionable Takeaways by Position
One of the fun aspects of going through the exercise of creating the percentile projections was seeing the impact of our baselines on certain percentile outcomes by position. Here are some of the more interesting observations, some of which are pretty intuitive, and others may be a little bit surprising.
-As mentioned earlier in this article, there is no caliber of hitter or quality of context that results in a hitter having a 10th percentile projection greater than 0. Along these lines, very rarely do any hitters even have a 25th percentile projection greater than 3. The variance in baseball is very real. Even the best hitters in the best of circumstances, *completely* fail 10 percent or more of the time and have a “bust” game at least 25 percent of the time. This knowledge should help give you the courage you need to make some tough fades in tournaments when you see ownership on any individual hitter getting out of hand.
-Hitter ceiling in cash games is probably underrated. Because hitters are right-tailed (discussed above), players hitting their ceiling have a disproportionate impact on your score for the night. Meanwhile, the 10th and 25th percentile outcomes for all hitters are pretty congested. It might make more sense to increase your probabilities of getting a ceiling performance, even if it comes at some expense to overall mean/median projection. Our standard projections do a pretty good job of that (see the note earlier on why our hitter mean projections exceed our hitter median projections).
-A hitter’s 25th percentile projection is dominated by their projected wOBA.
-The impact of projected stolen bases takes large leaps between a player’s 25th and 50th percentile outcomes and their 50th and 75th percentile outcomes before flattening out and actually losing some relative importance. This is also true for projected plate appearances.
-Projected ISO has minimal impact on a hitter’s floor or even median projection, but it is one of the most important characteristics for a player’s ceiling.
-Along with projected ISO, projected team total gains in relative importance as you increase percentile outcomes, although.
Here are some examples from Opening Day. Notice that in our standard projections, Khris Davis is our 10th highest projected Outfielder (also LOL Mike Trout is good):
Here are our median outfielder rankings/projections:
Here are our 90th percentile outfielder rankings/projections:
As you can see, Khris Davis‘ projection changes drastically when rotating through different percentile outcomes. In our median projections, he ranks outside our Top 20 outfielders. In our 90th percentile projections, he ranks eighth overall. This makes sense given what we know about the relationship between projected ISO and percentile outcomes. Davis has our third highest overall ISO baseline among outfielders on the slate.
Meanwhile, you’ll notice a lot more leadoff and two hole hitters climbing up the median outfield rankings due to a combination of higher plate appearance expectations and generally these guys possessing some stolen base ability.
-Projected strikeouts are a huge driver of a pitcher’s low-end floor (10th percentile outcome). They seem to have lower relative importance in regards to 25th and 50th percentile outcomes as run and hit prevention becomes more important (projected wOBA, implied run total against) before gaining in importance again at the 75th and 90th percentile outcomes. The math on this aside, our general philosophy at DailyRoto is Strikeouts Are King!
-Projected outs surprisingly has minimal impact on a pitcher’s floor. At first, this seems odd. You’d expect a pitcher who pitches deeper to have a higher floor, but if a pitcher is getting hit hard and struggling – it doesn’t really matter how deep we’d normally expect them to pitch. The only positive benefit is facing more batters to accumulate more strikeouts, but strikeouts viewed in isolation covers this aspect. The importance of projected outs rises steadily as the percentile outcomes increase.
-Floor projections (10th and 25th percentile outcomes) are more meaningful when making cash game decisions on pitchers than they are on hitters.
-There are less stark differences for pitchers since their distributions more closely reflect a normal distribution.
Here are some examples from Opening Day, starting with our 25th percentile projections:
Now, let’s look at the 90th percentile projections:
You’ll immediately notice that pitchers with higher projected outs shoot up the 90th percentile rankings. Zack Greinke goes from our eighth projected SP in 25th percentile projections to our fifth projected SP in 90th percentile projections.
Meanwhile, Eric Lauer is our fifth projected SP in 25th percentile outcomes as a home favorite with a low implied run total against (3.5), but he falls behind Zack Greinke and Carlos Rodon in 90th percentile projections. Those two pitchers have strikeout projections than Lauer, and in Greinke’s case, he has a much higher outs baseline as well.