Input Volatility, Week 2

It's early, and everything is tough to trust

This is the first edition of a new series I’m going to do weekly, and I’m still not really sure how it’s going to look. For the last couple years, I’ve been asked pretty regularly for more content here at Stealing Signals. Each week, those Monday and Tuesday posts tend to recap the past week’s action with a goal of recalibrating big-picture player value, but I have more or less stayed away from looking ahead at the upcoming schedule and matchups in part because it would just be another layer of analysis and the scope of those posts is already insanely broad.

It’s kind of funny that maybe the only way I’ve tried to focus what it is I do in those posts is to avoid looking at upcoming matchups, and I can’t really explain that decision other than to say that when I started my post years ago, there already were great longform pieces covering that stuff like Evan Silva’s Matchups and Rich Hribar’s Worksheet. Those still exist and I highly recommend them, and the scope of those pieces helps indicate that if I were to try to look ahead at matchups more in my Stealing Signals writeups, that wouldn’t be a small addition. That’d be a whole can of worms because there’s a ton that goes into that.

But late last season, I asked for some ways I could add more value here at Stealing Signals with content that spun ideas forward for the next week but wouldn’t necessarily be a super heavy lift. I got some suggestions about streamers or a generic start/sit post, but we as fantasy football content producers always run into situations where there are leagues of various sizes, and something like a “QB Streamer” article is both probably too focused to be of use to everyone and also still might not be of particular use to some people who are looking for QB advice, because their league might be shallower or deeper than the advice given.

In terms of trying to help the most people while also not spending a ton of time on elements that aren’t great bang for my buck time-wise, that type of article isn’t a good fit. There’s a reason those types of pieces are often written by people trying to break into the industry, and I won’t knock that at all — in 2015, my first in-season column was a weekly TE report at RotoViz, and I loved how focused that was and that I could sink my teeth into that one position. I absolutely crushed it, too — I remember being early on both Jordan Reed and Tyler Eifert, who were big late-round hits at the position that year. (If I’m not mistaken, I think I also contributed to a WR report throughout my second year in 2016.)

Anyway, the best suggestion I got for what this column could be came from a discussion with some of my industry buddies, and I think specifically from Ship Chasing’s Peter Overzet. The idea is about this concept of “input volatility,” which focuses on the nuts and bolts of projections a bit, and I want to walk through where it comes from and why it’s relevant.

The first thing to address is an appreciation for the accuracy of weekly projections in-season. In the DFS world, where users are putting a lot of money up on a weekly game, good projections are king. From its advent, DFS has been marketed as your way to win by finding that massive week from a player no one else was on. The reality of course is someone might win the massive contests with a million dollars to first with those types of plays, but that someone isn’t likely to be you, because we’re talking about more or less a lottery. The other reality is even when there are these absolute out-of-nowhere stars, they aren’t a lock to help anyone win, because you still have to have a really good team built around that player.

Even if you build 150 lineups (which most casual players aren’t doing) and have some complete unknown who turns out to be a weekly star in half of them (which would be a massively aggressive play), you’re not a lock to win a contest with hundreds of thousands of participants, because 75 teams isn’t enough. Let’s take this further — say you limited your lineups to only three possible quarterbacks, and had 50 lineups of each, and you hit on that too(!) because one of those QBs was the clear best QB to play. You’d essentially have 25 lineups with the best QB and your sleeper unknown star, and of those 25 lineups with that quarterback and the sleeper you nailed, you probably have a couple iterations of that quarterback’s pass-catchers and the “bring backs” on the other team in terms of how you stacked that game — it could have been one or two or three pass-catchers that was the right move, based on how the game went and it could have been zero or one or two bring backs that made the most sense. But let’s mostly set that aside and say you got 10 lineups through with the correct QB stack that nailed two of his pass-catchers and also nailed one bring back from the other team, plus you nailed your unknown sleeper on those rosters.

Those 10 teams would all do very well, but you still have four lineup spots left, and we’ve made a ton of assumptions to even get to a point where you had 10 teams that performed to this level. There will be thousands of other teams in a huge tournament like this who have the QB stack correct, and while they won’t have your unknown sleeper, the other four picks you have in those 10 lineups will have to have the right configuration of what were the more popular plays from around the league for you to win. In other words, you have one dud in some of those lineups who scored only 5 points, or you don’t have the correct DST play (which is a high-variance slot), and you might still lose to someone who had the right DST and had Justin Jefferson or whichever star went for 35 that you maybe didn’t get into your lineup, even if your unknown sleeper star put up 30 points of his own.

Because of how this all works out, the best DFS players tend to play a lot of contests with smaller fields, but at higher buy-in amounts. In a contest like that, the game is different. The correct QB stack with the correct bring backs might only be in a handful of rosters (or fewer) in a contest with only a couple hundred (or fewer) entrants. Rather than needing to “hit the nuts” with the perfect lineup, you’re playing the concepts of leverage and field ownership more. There’s a ton of game theory that goes into it, but it’s contests like these where projections are so heavily relied upon. (For what it’s worth, I firmly believe even more casual players should focus their play in these smaller-field contests at lower buy-in amounts.)

If you consume a lot of DFS content, you’ll hear concepts like “the optimal lineup” thrown around, and there’s a ton of focus given to the players who project like clearly strong options at their given salaries for that week. And it works; over time, it’s become clear that while football is a high-variance sport, good projections should be a hallmark of most small-field DFS strategies, because more often than not they are going to account for enough factors to be directionally accurate. That doesn’t mean you win in DFS by just playing the highest-projected “optimal” lineup, because of those game theory considerations you have to weigh. Finding out how to balance smart pivots that still project well but might still give you some leverage on the field you’re playing against is how to build lineups that consistently maximize your chance to win; it’s not this idea of figuring out that sleeper that no one but you realizes is going to have a monster game.

But because DFS is a weekly game, it iterates fast. There are new ideas each week and new results that confirm or deny them. Compared to seasonal, where each season is more or less its own data point — and then the conclusions from it are drawn all offseason and often that one data point is given way too much weight — in DFS, you get new information each week. The advances in how the game is played don’t take years, but months. I’ve frankly fallen behind some, and I don’t pretend to be an expert in the area, but I’m merely trying to push forward why you should care about DFS if you’re serious about seasonal, and particularly if you’re looking for help with start/sits.

And DFS is where this concept of “input volatility” became a popular term last season. The idea is fairly simple. We should care about good projections, and trust that they are seeing factors we might not be. The biggest ones that I tend to miss when I’m considering projections — that the projection is capturing but I’m not — tend to relate to the overall team situation, meaning the way the game is being projected by Vegas lines like the spread and over/under, which give a good proxy from an efficient market of how much offensive production we might expect from a given team. I call this an “efficient” market because it’s very difficult to consistently win while betting sides and totals. If you could consistently predict the spread and game totals better than those betting lines, you wouldn’t need to play DFS, because you could just win through those bets. (Similarly, if you can consistently identify sleepers that hit on a weekly level, go find their player prop markets and bet those instead of hoping to get them into a DFS lineup where you need to get a ton of other stuff correct around that decision to win.)

So it’s important to understand that good projections take into account a ton of factors that we don’t always immediately think about, and they are probably better than our gut feels for that reason. At the same time, they aren’t perfect, and our seasonal start/sit decisions are a little different than DFS.

Some of the ways they aren’t perfect overlap with ways I’ve talked about seasonal projections and my lack of trust for them. The biggest issue with seasonal projections is the constraints aren’t known for the entire timeline. We have a good idea of what to expect from these rosters for the first month, but the chaos of NFL seasons shifts the decisions we were considering in August to often massive degrees, and it does it for more teams than you expect (my past research has me believing there are significant impacts to roughly half the teams, meaning whether or not to trust the constraints of a team’s seasonal projection is something of a coinflip; we already got our first major victim in Week 1 when Dak Prescott’s injury immediately changed the way anyone would view the Cowboys’ offense).

But another part of seasonal projections that is always difficult is how things piece together. The overall targets need to match up with the pass attempts. The rushing shares need to match up with the projected rushes. For the Cowboys, the size of their target/rush attempt pies will shrink, and that’s a problem for the accuracy of those seasonal projections. It’s not an issue on the weekly level.

But while good weekly projections take in a lot of objective data like those betting lines to get a good indication of player opportunity — and also objective data on the individual player level to get a good indication of what each player’s potential outcomes might be — there’s an art to them, too. Target shares and efficiency and those elements are all inputs. For the best projections, these inputs are heavily researched and they are probably as accurate as possible. But the element of how things piece together is still a challenge on the weekly level, and that’s where some art comes in.

Take Thursday night’s game this week, and specifically the Chargers’ receiving projections without Keenan Allen. As I looked over those projections, I noticed two things I didn’t believe, and they were highly correlated, and they were driven largely by the reality that projecting the Chargers’ target rates with Allen out of the lineup was a task that couldn’t be easily done with objective data. There aren’t enough players on the Chargers that have strong target-earning data, and so decisions have to be made.

Like I said, I trust that projections were as close to as accurate as they could be, but the point is there were simply wide ranges of potential outcomes for this team. This is where the concept of input volatility comes in. We can trust that weekly projections are as accurate as possible, but still recognize there is value in identifying where there is more volatility on the inputs that drive the projections than usual.

The two things I didn’t believe related to Josh Palmer’s and Gerald Everett’s projections. Co-managers of mine can attest I didn’t like Palmer’s projection. Palmer’s past data is limited, and while he had a very solid finish to the 2021 season that including a few games with very solid target shares, he also had some duds. In a long view sense, he’s a player I’ve been hesitant to buy into. The idea behind Palmer and his ranking and draft positions all offseason was such that in a situation where Allen was out, he was expected to run a number of routes and should have been in line for a substantial uptick in opportunity. As it turned out, that probably paid off, because he caught a touchdown on the final drive that pushed him to a serviceable 8-4-30-1 receiving line.

The flip side of that is I started Everett in a ton of leagues, a lot of which I co-manage with my Stealing Bananas co-host and perhaps the world’s biggest Everett fan, Shawn Siegele, but also places like home leagues. My toughest decision was in a deep-bench home auction league where I didn’t land a top TE but managed to grab three upside options I really like. I decided to play Everett over both T.J. Hockenson and Albert Okwuegbunam this week, despite him projecting worse than both, and I might still live to regret that, but after watching Thursday night, I’m very content with the process. He beat his projection en route to a 10-6-71 line.

My logic on the Everett decision was driven almost entirely by this idea of “input volatility.” I knew that even if Mike Williams had a strong bounceback game, there were going to be targets available in this offense in a matchup that looked tight and with a high over/under, and featuring two offenses that were in most scenarios going to wind up combining for a ton of pass attempts. There was some possibility that Williams would have another bad game, as well, or that Williams and also someone like DeAndre Carter were the big target-earners, but most of the outcomes featured a lot of potential opportunity for some combination of Palmer and Everett, both of whom looked likely to run a lot of routes.

Interestingly, that volume did wind up split fairly evenly. But part of my opinion in this specific situation was driven by a long-term belief on Everett, as well. While with the Rams, Everett was a solid part-time player who through a combination of some injuries and just general usage, never really put up numbers I thought he was capable of. Last year with Seattle was supposed to be that opportunity for him, but Seattle was kind of a mess, and Everett himself had a few very memorable miscues. But you also look at how Seattle treated the TE position in Week 1 of this season — relegating a former first-round pick and seemingly a key part of their Russell Wilson trade in Noah Fant, a guy who is arguably a top-10 TE in terms of pure skill in the entire NFL, into a committee of sorts — and you’re free to wonder if ways we viewed Everett’s poor 2021 might have been driven somewhat by Seattle. So with Everett, the idea was that this offense might be the one where all that delayed hype finally hits, and that was the idea on the seasonal level, of course. After Week 1, it felt a little like he might not just be able to replace Jared Cook, but also might be able to do more, because Cook himself — who was somewhat productive — was limited to a rotational role, mostly due to age.

All of that is very uncertain, and I want to go back to trusting the projections in most cases. But there are these situations on a weekly basis where input volatility does exist, and I’d argue in those instances it’s OK to let your mind wander a bit and have a stance. What you’re essentially saying is, “I trust most of the inputs for these projections, but for these specific inputs that require a little bit of an art — a little more projection of role — I have a different opinion about the ranges for those inputs.” It’s a disagreement on a specific subset of the projection where you’ve identified that even the best projector can’t actually know the answer, because of the volatility that is present.

I chose to focus this week’s article on the theory behind input volatility rather than specific situations, because there’s one last element to understand. Projections get more accurate with more data, and because there is a ton of turnover in the offseasons — and especially this year — that means that the early-season portion is going to have far more of these “input volatility” situations than anything into November or December. This is reflected in DFS results, by the way, where the players who have an incredibly strong idea of how to leverage projections tend to start gobbling up tournaments the further along we get. I myself saw a significant dropoff in DFS results after the first month or two into the later part of seasons for each of the first few years that I played DFS regularly, and I have heard similar from other more casual players who might be more prone to player takes as opposed to an almost robotic process that features a thorough understanding of the week’s projections (which, if I haven’t harped on it enough, is the far more sustainable way to play).

But in the early season, these Palmer/Everett decisions are everywhere. With Ty Montgomery out, after he played 9 of 10 third downs last week, the RB receiving inputs in New England are impossible, and likely to be flattened out. But there’s volatility there, and upside for a guy like Rhamondre Stevenson.

With Elijah Mitchell out for San Francisco and little data there, we could see Jeff Wilson get 20 carries, or we could see a hot-hand approach that allows Jordan Mason or Tyrion Davis-Price to work into enough touches to be viable, or we could see Deebo Samuel get every high-value touch and wind up with a massive weekly line.

The injury situation are the obvious ones, but there are still others that have volatility. The Carolina Panthers as a team ran just 50 plays last week, and the market has them as underdogs to the Giants in a game with a low total. I’m pretty frustrated with them and really can’t believe they didn’t fire Matt Rhule last year, because they looked unprepared last week and look to be in for another frustrating season on a team level that wastes yet another year of D.J. Moore, but even I can admit that I might be overreacting — and the market might be overreacting — and it’s possible Cleveland’s defense is strong and that played into last week, and also that Baker Mayfield will settle into this offense after he’s played it in for more than one game. The input volatility here is actually with the markets, where it’s possible Carolina has a much better game than their overall expectations, which is the type of thing that we’ll also have a much stronger feel for as the season goes along. For Moore’s and Christian McCaffrey’s sake, I really hope this is the case and the Panthers put up 30, but this one might be more wish-casting than a great example of input volatility, so take it as an example that we don’t even know what these teams are yet. At this time last year, Cincinnati had gone very run heavy in their first game and over the following weeks they looked a bit like an offense to avoid because Zac Taylor might just not be a good enough coach. It turned out to be easing Joe Burrow back in, and the Bengals went on to decide fantasy championships by the end of the year.

I mentioned before that seasonal decisions are different than DFS ones, and the biggest factor is you have a limited number of options in seasonal. This concept of weekly volatility impacted the TE decision I described above largely because it was one of three players I needed to go with, and they all had somewhat similar projections. In a case like that, my inclination is to lean into the volatility, because in seasonal, we win through big weeks, not the avoidance of duds (in fairness, DFS is the same way in this regard).

But in some deep leagues, like some of my Main Events, I have decisions this week that include some seriously gross RB2 options, which, look, it happens. In one league, my preference is to go with Jordan Mason over some options that clearly have better cases to score at least a few points. My reasoning is there’s enough input volatility in San Francisco that I want to lean into, because trying to decide between who has the best shot at 5 points and who is the most likely to take a complete zero isn’t a discussion that should matter; what should matter is who has the best shot at 10 and then 15 and then 20 points (which Mason probably doesn’t have, but I guess I can see outcomes with a likelihood probably in the fraction of a percent range where he does).

We all remember situations where we took a dud in our starting lineup and lost by just a couple of points, but that’s just an emotional response to an outcome that could have been decided by hundreds of plays across any of the players in either lineup, and didn’t actually “come down to” that one start/sit. (Just as an aside, over a long enough timeline, you’ll win more weekly fantasy games seeking upside and ignoring the emotional response to duds in your lineups, and I don’t really have data on that other than saying there’s not much I believe in more as a result of my own experience, and you’ll have to trust me on this one.)

So the lesson for this week is in your start/sit decisions, you should definitely consult projections, but you should also be aware of any circumstances where input volatility exists. Just the presence of input volatility doesn’t indicate one option is better than another, but it should be an important factor in your decision, especially here early in the season where there is so much unknown. For these early-season decisions, ask yourself if you can see a situation where you’re looking back in a few weeks and the decision that feels a little crazy this week would make a lot more sense down the line. That was a major driver of my Everett and Palmer thoughts, because I do think this could be the year things really click for Everett and his “lookback projections” to this game would have us all thinking they should have been stronger back then, and also that Palmer could struggle enough that we’d look back on him and project him differently as well, even given the context of Allen being out for this one.

Alright, that’s all for today. I’ll see you guys Monday with Stealing Signals, as usual.