Best Puck Manifesto - Volume 3

Welcome back for part three of the Best Puck Manifesto, focusing on the fantasy hockey season-long best ball tournament structure on Underdog Fantasy. If you haven’t yet, now’s a good time to check out Volume 1 and Volume 2, where I dove into roster construction and ADP/closing line value.

Be sure to subscribe to the newsletter for immediate updates whenever I post a new piece!

Let’s pick up where we left off, digging further into stacking and the correlative aspects of Best Puck.

The same two suggestions from V1 and V2 if you’re looking to improve your game:

1. Hop into the Morning Skate Podcast Discord, with lots of discussion and hundreds if not thousands more of the 15k Best Puck Classic entries hanging out in one spot. It’s the place to be to discuss the latest news, especially when we get into training camp, and is the best way to get a hold of us outside of streaming hours.

2. Which brings us to the streams - which will primarily be hosted on DJ’s YouTube channel, with a weekly audio version going up on the MSP podcast feed up through the beginning of the season. Sub to both so you don’t miss a second! There’s lots of relevant discussion about structure and strategy in last year’s content as well, helpfully labeled with UD or Underdog in the title. DJ recently pitched this year’s versions of last year’s league winners in a tightly produced video for those who don’t want to sit through 90-minute streams.

Best Puck Manifesto, pt. 3: Stacking

First, a caveat: I am thoroughly convinced that I could have spent from now until October looking at stacking and still left gaps in the analysis. This is an expansive topic, with several inputs and pathways, that will ultimately take an army to solve, if it ever is solved. I’m pleased to be a trailblazer in putting raw numbers behind the theories that I’ve held and shared over the years, but I am fully aware that my assumptions may not match those posited by others. I encourage you, the loyal reader and (presumably) sharp Best Puck drafter, to incorporate what I’ve put together and think about how it might fit into your current beliefs.

What is stacking?

Stacking is the colloquial term for selecting players from the same team. If you take Auston Matthews in the first round, it’s possible that Mitch Marner, his fellow Toronto Maple Leaf, will be available when it comes time to make your second-round selection. Taking Mitch Marner would constitute a stacked TOR pairing of Matthews and Marner.

Who is stacking?

Just about everyone. In 2023-24, across 14,100 drafts, only 68 drafters failed to pair a single player with a teammate on their fantasy roster. That’s less than half a percent (0.5%) of the field. This, and a few other aggregated datapoints dealing with stacking at a high level, is available in the massive Excel data dive I did, shared publicly here. There’s a very similar trend in 2022-23, with a similarly tiny percentage of non-stacked teams, showing that the vast majority of the field is intentionally stacking.

How does the field stack?

By reaching for players with ADP lower than the pick they are currently using. To speak nothing of the results of stacking, here’s where the first kernel of information enters the fray. When you stack a player, you tend to pay more for that player than other drafters do.

The first player you stack (meaning you are drafting the teammate of someone already on your roster) costs on average about one pick more than the unstacked variety of that same player. The average ADP capital difference (a concept covered in detail in volumes 1 and 2) totals to ~1.5 points.

The reason for this is the supply and demand dynamic of the draft board. When it is your turn to select a player, the players that are at, or after, ADP could be on one of 32 NHL teams. The odds of a player that fits with your existing roster being in supply and serving as an ADP value are rather low, whereas the demand for stacked players on your roster of course has increased, because most of the field is doing this for one reason or another that we’ll cover shortly. Thus, the field tends to follow basic economics and will reach and take players before ADP that fit in as a stacking partner. In the aggregate, the effect may seem minor, as 1.5 points of draft capital doesn’t mean much, but as you build out a team stack, layering on stacked partner after stacked partner, we find that the supply/demand dynamics accelerate further and result in the average team with a five-stack (a base player + four stacked players) sacrifices an average of 3 draft spots per pick and an overall ADP Capital amount of ~15 points.

Keep in mind that this is just for one stack! Most teams build between 3-4 stacks (62% in 23-24 and 60% in 22-23 were in this range of 3-4 stacks of varying sizes) into their drafts, and if you build them out in alignment with the field, the effects add up, potentially sliding you into a lower Capital Gained bucket.

And, well, this is problematic. One clear sign of a Good Draft is by accruing positive ADP value when drafting. Simply drafting players that slide past ADP and fall to you was a very predictive indicator of success, as last-newsletter Moods explains:

What is there to gain from stacking?

Great, in order to get players from the same team you’ve made your fantasy team fare significantly worse by ADP Capital Gained, which is a clear indicator of future success.

If stacking forces you to make sub-optimal selections, then why do it? Let’s dig in:

Reason #1 to stack: You get to double down on your best takes

Here’s a chart that I suggest zooming in on if you are able to blow it up a bit, but I wanted to be able to show all teams on a relatively even scale.

Each fantasy draft can be classified by the number of players from each NHL team found on that 16-man roster. When you remove the “0”s for obvious reasons and filter this dataset down to show only team + stack size combos that have at least 50 teams to work with (out of 14,100 drafts), this is how the regular season advancement looks.

Lastly, I wanted to show all teams but the bar sizing & coloration was bothering me, so I sorted the teams so that teams with four bars are up top and two bars are at the bottom, sorted by advancement rate within each range/bucket.

{It’s not a mistake in the graphic, Dallas truly does have n=54 teams that drafted a DAL 4-man stack. Zero of those teams advanced out of the regular season. Pour one out for the Jason Robertson drafters…}

One clear trend is that if you find a league winner, the effects of stacking said league winner appear to be amplified. Having a player rostered “naked” (stacked with zero teammates, for the uninitiated/particularly erotic) from Tampa, Minnesota, Florida, NY Rangers, Colorado, and Toronto led to above-expectation advancement rates, from the top group of teams with lots of stack sizes from 2-4. In every case except Florida, building out that correct take into a full-fledged stack led to huge advancement gains. In Florida’s case, it was nearly impossible to stack them given every useable player besides Matthew Tkachuk was a C. Barkov, Verhaeghe, Bennett, Reinhart, all Cs. I’m willing to bet that the failure of those stacks (largely buoyed by Samson’s incredible season) came from the sheer lack of projectability from any position other than C last year, and not some fatal flaw in stacking and/or the Florida Panthers.

This trend carries into the “max = 3” stacked teams and leaps off the page for the “max = 2” teams, specifically Nashville, Philly, and Montreal. When you nailed the league-winners in Filip Forsberg/Roman Josi, Travis Konecny/Owen Tippett, and Mike Matheson/Cole Caufield, pairings of players who each individually finished in the top 25 among all players (among n=189 players drafted at least n=500 times in 23-24) in Total Expected Value, it was a major plus to pair those players together.

That’s not necessarily surprising, but lastly take the time to consider things from the inverse perspective. Beyond just Dallas, there are several teams that on the aggregate led to poor outcomes when you drafted a naked player from said team. In nearly every circumstance, doubling down further on that bet by stacking the player with more teammates led to worse results. For example, if you whiffed on Tage Thompson, odds are you weren’t getting much help from drafting his underachieving teammates such as Alex Tuch or Jeff Skinner, either. If only anyone had seen that coming…

Based on this season-long data covering actual performances, it seems overwhelmingly obvious that good results for a player (relative to ADP, of course, as this is how all bets should be valued) beget good results for the team, and vice versa. Stacking simply sacrifices a bit of ADP value to earn a lot of equity… or lose a lot, if you stack the wrong team. In a game that pays out over half of the prize pool to the top 15 finishers out of 14,100, amplification of that sort is critical to unlock high-end outcomes.

Reason #2 to stack: Stacking increases your weekly upside

For those entering the Best Puck space from the DFS NHL side of things, it’s well-known that tournaments are often won by the player that scores a hat trick, stacked with his linemates that night who chipped in with an assist or two, or even managed to snipe a goal themselves.

The same can be said about a week of Best Puck… if there’s a week that has a game where Kaprizov-JEEK-Boldy go batshit crazy in a 10-8 game featuring three hat tricks and almost wins me, a noted 2023-24 loser, a GPP, you’re going to have a high weekly score if you stacked any two of those players together. For the record, in week 19, the week with that game, five of the top seven weekly scorers came from that game (with JT Miller entering from Vancouver’s side and Zuccarello sneaking in from Minnesota).

Setting aside once-in-a-generation nonsense, weekly upside does transfer from one player to his teammates. When breaking apart every player’s season, by week, with a divider of whether that player’s team had a top-five positional score somewhere on the roster, it is significantly more likely that said player posts a usable score or a ceiling result, as well as corresponding top-20/10 position scores as well.

To prove this statement, I want to cheat a little bit by looking only at players who had at least 10 weeks of 25+ fantasy points. This gives us a wonderfully round set of the “best “100 players last year, or at least the most consistent ones: 38 were C, 31 W, 15 D, and 16 G. The trends I noticed do appear to apply broadly to the set of NHL players I looked at (everyone with an Underdog ADP last year, or about 400 players), but it’s far cleaner to delve into those who are most certainly fantasy relevant.

Finally, note that this dataset does look for any teammate with a top-five positional score that week, not just from this set of 100.

To ensure we are not overvaluing certain players who performed better than the rest (…MacKinnon, or Matthews, or Kucherov), these metrics are grouped by player, and then averaged across all 100 players defined from above.

Let’s take a quick look at the formulation of the underlying math before diving in:

When a C from the list defined above has a top-five scoring teammate, they crack 40+ points 29.8% of the time. When they do not, they only crack 40 4.5% of the time.

Why 40? I found a use for this chart!

The odds that a weekly score of 40 UDPts or higher grants you nearly 100% certainty that it will wind up counting, regardless of position. This mark is slightly lower than 100% in the ballpark of 90% for goalies, but the point being you’re looking for as many 40+ weeks as possible to create separation on top of your already-locked-in top-5 positional scoring teammate, as indicated by the above table.

Accepting that there might be some signal lost in summarizing this data as simply as the +22.1% Impact for D might indicate (it seems rather noteworthy that not a single league-winning D put up a week of 40+ without they themselves or a teammate finishing top five at their respective position on the week!), this will allow for much quicker takeaways.

When sorting these players into ADP buckets along with position buckets, there’s not much of a difference in impact. One thing I do take away is that later round D were more reliant on their teammates having top-5 weeks to put up ceiling weeks (40+) than the self-sufficient D like Roman Josi who went earlier in drafts. This noticeably dissipates when focusing on just “usable” weeks (25+), which appeals to my sense that D are more often a beneficiary of situation than anything else. No similar phenomenon exists among this cohort of Cs or Ws.

Across the board, you are almost always between 20-40% likely to improve your results dramatically in a given week when your team has a top-five scorer on it. I posit that the lack of impact from later round Ws (round 6 and on) stems from the fact that most of the upside is gone from W after round 5, so there are simply less 40+ UDPt weeks to go around, limiting the ceiling potential regardless of how their teammates fared. With several Cs being placed in the W bucket this year with UD back-end updates (Kempe, Reinhart, Kyrou, Stamkos, Jarvis, Barzal, to name just a handful…), this misbalance could correct itself this year.

Reason #3 to stack: Mini-stacks dominated in 2023-24

Perhaps this is a combination of the above two reasons. When looking at Total Expected Value, accounting for both regular season advancement and playoff week upside and corresponding high placements when the big money counts, Total EV skyrockets across the board for ministacks (Two skaters from a team) and worsens as you further your bets on that player’s team.

• In aggregate, the median EV for the 198 players (all skaters this time) included in this analysis was $7.10. By doing nothing further than stacking with just one player onto your key player (the “_1” column on the right), you’ve boosted your expected results by $0.60, nearly 10%! That doesn’t sound like a lot, and the effects aren’t necessarily additive onto a given roster, but this suggests that if you do this across a full set of entries and implement all the other strategies discussed in this manifesto, you’ll generate a ton of equity by mini-stacking.

• Stacking the league-winners once was good, twice or more was not. When looking at those in the “rd2_adv_rate_category” of 1, they collectively advanced at a 30% rate and totaled $16.80 in equity as a result of their incredible performances. Stacking these players with exactly one teammate led to an additional $2.80 of equity, while stacking 2 or 3 teammates cost you $1.60 and $3.00 off of the unstacked player, respectively.

• Stacking D did not generate any additional EV. Even the ministacks were a net equal endeavor relative to the defender himself, something not found in any other position.

• In general, the massive stacks got completely bodied. Stacks of 4 (“_3”) or 5 (“_4”) were quite rare, of course, but when they did get created, the outcomes were horrific from a total_EV perspective.

Final Stacking Thoughts

This is where I tell you what I think, and conveniently tell you that not a ton has changed in my draft approach, from my point of view, having spent multiple weeks on this endeavor and learning a lot about the fantasy landscape in the meantime.

I am certainly going to prioritize mini-stacks (pairs of players) going forward, more than I have been. As I mentioned off the bat, there are several paths of exploration that I simply did not have the time to pursue. I encourage anyone with an interest to dig deeper and apply your own perspective to the data, you might uncover something that my methods failed to reveal.

I believe there are a few reasons that mega-stacks (3+ players from a team) tend to fail:

1. The cost of acquisition goes up: As covered in the early section about ADP Capital, the later additions to a stack tend to cost you more as you move down the draft board to scoop the perfect fit to your existing stack on the roster. The cumulative effects of this intention, and the mindset that originates it (i.e., doing this multiple times within one roster for two or more stacked NHL teams), probably trades out the value of the correlation for a greater penalty of the loss of ADP value. The field is not perfect, but continually swimming upstream with your selections is proven to be a negative-expectation endeavor for The Field.

2. There are 32 NHL teams, and 192 Best Puck draft selections: It’s a simple math equation. With 32 teams, there are exactly six players per team that will be drafted. Of course, this isn’t a clean one-to-one, as some players contribute more players to a given draft board than others, but it is worth considering what this means. Removing Goalies from the equation, there are now 156 skater slots to fill (192 - 36 drafted goalies, three per team) from across the NHL. If we had perfect knowledge of every NHL team, there are 160 PP1 options at any given time across the league. In theory, it is possible to draft a team full of players that are on the PP1. When stacking, especially later in drafts, the draw is to stack the good teams.. well, those PP1 guys are usually off the board, and while it is sometimes worthwhile to take a shot at a guy with a good top six and PP2 role and hope that it becomes a great role, there are options - known or unknown - who will be PP1 from day one. Think Edmonton with Connor Brown and Evander Kane last year, those bets on 5v5 production absolutely destroyed EDM stackers because the PP job never materialized with Zach Hyman scoring 50. Sure, Jeff Skinner can score 30 on McDavid’s wing at 5v5, just as Evander Kane could have last year had the opportunity arisen, but there are far more options on other teams who have a clear path to PP1 opportunities that Skinner does not have this year.

3. Only six of your sixteen players will score in any given week: As you add on players to your stack, you’ve placed a bet on the stack to succeed. If your stack succeeds, success likely looks like two scoring positions used up for the week. Having a third, fourth, and even fifth option to append to a mini-stack can increase your ceiling on paper, but you now have four scoring spots to formulate from your fourteen remaining players. Is the third-best option on your squad from an NHL team going to score enough goals and get enough points to beat out what in theory should be an otherwise competitive set of 14 players? After all, you should have at least some of the best skaters from other teams, too, as the above point outlines.

Similarly, here are some additional thoughts on why mini-stacks work:

1. The three reasons above, in the article you just read: I don’t feel the need to re-litigate the cases for correlation to league winners, enhanced weekly upside, and last season’s fantastic results for mini-stacks relative to unstacked and overstacked groupings. But here are a few more takes for the road…

2. You need to get less right: We know a lot about each team as Best Puck currently stands, but we have no guarantee that through injuries, line changes, and general development over the course of a season that our bets placed in August and September will come to pass in March. I can’t guarantee you anything other than this is going to be another crazy year. That’s sports. In no area is this point stronger than stacking, which generally implies not just team-level correlation, but line and PP-level correlation as well. Being right about two players being on the same PP unit is slightly easier than saying the same about three players (especially if one of those players is a later-round pick in fantasy), and nailing the 5v5 correlations of the entire first line is a much more difficult task than getting a key pairing right. Coaches even tend to think of their lines in the same way, and while line tinkering is common, the reshuffling of line pairings (think Matthews/Marner, Tavares/Nylander) is a more seismic event across each NHL dressing room. Why make things harder for yourself by needing to get more right?

3. The Schedule might actually be all that matters?: Anyone who has participated in a playoff best ball contest knows the pain of advancing a team to the next round, only to realize that only four players are still alive in the playoffs and that your ceiling is capped as a result. In Best Puck, the ideal scenario is to have all of your players playing four games in a week, every week. Last year, the Rangers managed to fit five games into a single week, and predictably went apeshit for teams with Rangers mixed in with their teams that advanced to the finals. More of that, please.

   Looking at the fantasy playoffs this year, there’s nothing that extreme, but it does illustrate this point all the same:

   

   A wonderful viz from Tart_Blueberry in MSP Discord, pinned in #dog-app

   If you make the finals with Auston Matthews, you’ll be rather disappointed if you are leading the contest after Finals Week 1 and go to battle in Finals Week 2 with Matthews, Vincent Trocheck, and Nico Hischier, three Cs who all play only two games in the final week.

   Although we can’t ignore better players in the interest of single-week outcomes (after all, there are ways to address this issue beyond “never taking Auston Matthews”), it is undeniable that some of the schedule issues are in our control. If you accept that this schedule is largely reflective of most Best Puck weeks (I have not analyzed the full season yet in this manner, nor do I anticipate doing so), it shows a general pattern, one where teams range between 2-4 games in a week. If some players get double the games and minutes, it follows that they’ll expect to get double the fantasy output as well.

   I think that some of the power of stacking has nothing to do with the correlative power of teammates loving each other, playing hard for one another, and directly helping each other put the puck in the back of the net. Rather, I think that the positive influence these mini-stacks have on our rosters is that they allow for a more natural fulfillment of this natural tenet of Best Ball I’ve been circling around: If you can’t have everyone projected at the max always, then project at least a full team for the max for every individual scoring period. After all, not every player you draft in the NFL Playoff Gauntlet can make it to the Super Bowl, but you need to field a team if you do manage to make it there to maximize your EV.

   While this, too, is mostly out of our control (although you may be able to quantify the cohesiveness of groups of team-level schedules…) across ~25 weeks of Best Puck, it does stand to reason that 6/16 score each week, not your full roster. In the playoffs, I count 66 total 4-game weeks across six weeks for 32 teams.. 66/192 is a few hundredths different from 6/16 and around 1/3.

   I count only 11 of 66 instances where these weeks happen back-to-back, about half of what you’d expect if this were truly a random distribution of games-per-week. Now, if you have a mini-stack that plays 4 games in one week, well that’s two spots accounted for that are maxed out. The following week, it’s very unlikely that the team plays four games again. So perhaps another mini-stack takes its place. With the yin-and-yang we see in 4-game weeks, there is a natural symmetry to certain teams in certain blocks of the schedule.

   Fill out five or six well-constructed mini-stacks of good players who serve as alphas for their teams, maybe mix in a goalie or two that fits these builds, and let them play off each other for certain pockets of the schedule and reap the benefits as they fill in the gaps for each other. Sure, you’ll need those teams to generally outperform their expectation to have true league-winners and finals equity, but the power of stacking carries through even to the worst teams (the lowest adv_rate bucket in reason #3 benefits the most from mini-stacking!).

   Why might that be?

   Maybe the real power of stacking is a culmination of every point raised in the first 4,000 words of this article… but not in relation to what happens on the ice, but rather reflective of the very paper that the games are famously not played upon… the schedule.

Thanks for reading! Subscribe to the newsletter and stay tuned to Discord and my Xwitter to make sure you don’t miss a thing!