When I started writing about dynasty leagues five years ago, my first goal was to create something that could serve as a dynasty analogue to VBD. Unfortunately, five years ago, the challenge was beyond my abilities, so I shelved the project and moved on to other things.
From time to time, I’d revisit the puzzle with the benefit of experience and try to chip away at it again. This offseason, after a long stalemate, I began to make some headway. But I’m getting ahead of myself…
What’s So Great About VBD, Anyway?
To understand why I wanted so badly to create a dynasty equivalent to VBD, we must first understand what VBD is. VBD stands for “Value Based Drafting”, and was coined by Joe Bryant roughly fifteen years ago.
Bryant’s observation was that, in terms of fantasy football, it didn’t matter how many points a player scored. Instead, his value was measured by how much he outscored his peers. VBD was a way to measure that difference. Essentially, a player’s VBD was the points he scored minus some baseline, usually one representing the “worst starter” at his position.
And just why is VBD such a powerful tool? It has three really big things going for it.
1. VBD is precise.
Rankings can tell us where players fall in relation to each other, but say nothing about the size of those gaps. If Andrew Luck is ranked first, Aaron Rodgers second, and Russell Wilson third, we don’t know whether Rodgers falls closer to Luck or to Wilson. Tiered rankings can help to some extent, but they’re still just a step in the right direction. VBD gives us precise differences. If Luck is projected to 100 VBD, Rodgers to 80 VBD, and Wilson to 40 VBD, we know that the gap between Rodgers and Wilson is precisely twice as big as the gap between Rodgers and Luck.
2. VBD is directly comparable across positions.
If a quarterback scores 500 points and a running back scores 300 points, which one is more valuable? We don’t know, because the two values aren’t directly comparable. If a quarterback produces 75 VBD and a running back produces 150 VBD, which one is more valuable? The running back is. VBD already has positional considerations baked in. It is position-agnostic.
3. VBD is universally applicable.
How much is C.J. Spiller worth? Well, that’s going to depend a lot. Does your league award a point per reception to running backs? How many backs do you start every week? Is there a flex position? Do you perhaps give additional points for return yards? Is this an 8-team league or a 16-team league? The answer to each of these questions will impact Spiller’s value. VBD, remember, is points scored over baseline. Points scored varies by scoring system, and baseline varies by roster settings. This means that VBD can be tailored to specific league settings, giving different advice to owners in different circumstances.
Now, VBD is not a perfect tool. It’s terrific as a descriptive statistic, telling us after the fact which players were most valuable, but for predictive applications it will only be as useful as the projections on which it’s based. Footballguys’ projections are among the best in the industry, but even the best projections aren’t going to be perfect.
The question, though, isn’t “is VBD perfect”. It’s “does VBD allow a quality of analysis better than what existed before”, and the answer is absolutely yes. VBD gave us a perfectly customized, universally applicable method to precisely measure value differences across positions. It was a big step forward for redraft analysis.
Challenges With Creating VBD for Dynasty
First and foremost, why can’t we calculate VBD the same way in dynasty as we do in redraft? Because predicting VBD involves projecting player performance, not just for the individual player, but for the entire league. It’s a fraught enterprise strictly when dealing with a single year at a time.
Predicting precisely what’s going to happen in 2019, though? Forget about it. It’s a non-starter. For dynasty leagues, we need to preserve the idea of a player’s “points above baseline”, but we need a way to estimate them in future seasons without having to resort to the impossible.
Originally, I tried to get around this problem by playing with aging curves. I could estimate where a player was on his career arc, and nudge future projections up or down to match. But this approach never produced acceptable results, and I couldn’t figure out why. It turns out it’s because aging curves don’t work on an individual level, but I didn’t know that at the time, and I wouldn’t have had a good alternative even if I did.
The other big challenge facing dynasty leagues is that not only do teams play under different scoring systems and lineup requirements, but they also play with different windows and goals. “Universally applicable” means a dynasty VBD needs not just to apply to various league setups, but it must also be as useful to a contender as it is to a rebuilder.
Obviously there are a lot of obstacles, which is why there isn’t anything out there yet to fill that need.
Overcoming Those Obstacles
In case you haven’t guessed, I now feel like I have a workable solution to the problem. While the challenges were too much to solve them all in one sitting, over the past five years I’ve been able to chip away at them a little at a time, often without even realizing that I was doing so.
To start with, while I’ve sung the praises of VBD so far, I believe that by taking a couple of extra steps, we can make it a better and more accurate measure of player value. To that end, I wrote earlier this season about calculating value per-game instead of per-season, and comparing players not just to a worst starter, but to an average starter.
This gave rise to two new statistics, EVoB and EVoS. EVoB, or Estimated Value over Baseline, represents a player’s true value over a “worst case scenario” baseline at the position. EVoS, or Estimated Value over Starters, represents the advantage he provides against an average starter. Remember the terms, as you'll see them a lot in the rest of this article.
After fine-tuning the way I calculated value over baseline, I wanted to improve on the way I calculated the baselines themselves. To that end, I went through real fantasy leagues and noted the actual “worst starter” and “average starter” values in real-world scenarios.
While EVoB and EVoS were still not a useful predictive tool for dynasty, they now gave me a phenomenal descriptive tool that I could use to measure what had actually happened in the past. So I built up databases of every fantasy season from 1985-2014, (30 years in total), and started looking for patterns and trends in the data.
The first and most important thing I noticed is that we were thinking about age all wrong. In short, the best model for aging wasn’t a curve at all, but a mortality table. This was a huge discovery.
Under this new “mortality table” model of aging, player performance tended to hold steady around a stable plateau representing their “true performance level” for some indeterminate amount of time until the player suddenly and unexpectedly fell from fantasy relevance, (or “died”). In other words, estimating future value didn’t require specific projections at all. Instead, it was simply a matter of estimating a player’s “true production level” and his estimated time remaining!
Based on my databases, I was able to use historical results to create mortality tables for each position, allowing me to calculate estimated years remaining at quarterback, at running back, at wide receiver, and at tight end.
All that was left was to figure out how to estimate “true production level”. Thankfully, I happen to work at Footballguys, home to a series of award-winning player projections. If I those projections are reliable enough to generate redraft VBD values, they should serve just fine for estimated “true production levels” for dynasty, too.
Using redraft projections had another very crucial advantage. Since the heart of my “dynasty VBD” was based on specific projections, it is possible to tailor it to different leagues settings and scoring systems. I could calculate value in a 2-quarterback league as easily as in a 1-quarterback league, in PPR as easily as in standard.
After feeding those projections into my EVoB/EVoS calculations, I have an estimated production level for every player, but with one important caveat: while I have a good way to model the decline stage of a player’s career, I have no way to model the ascent stage of his career. Players who have reached their “true production level” tend to stay there, but it would be silly to assume that how rookies are this year represents how they’ll remain for the rest of their careers.
So on top of the calculated EVoB/EVoS projections, I built in the ability to estimate peaks for ascendent players such as rookies. I also added the ability to reduce peaks for players in unstable situations. For example, Davante Adams’ projection for this year will be less indicative of expectations for next year after Jordy Nelson returns.
Each of these assumptions on my part introduces potential error into the value calculations, but remember, VBD was never perfect, either. VBD produced a player value that was only as good as the underlying assumptions and projections. My goal was never to eliminate the need for assumptions, it was to simplify the assumptions required to increase the chances that they were reliable.
So, after all that, I had a formula that could estimate remaining career value. The last hurdle was altering it so that it was applicable to contenders as well as rebuilders. For that, I needed a fair accounting of the preference differences between contenders and rebuilders. The key difference is one of time preference— how each values the current season relative to future seasons. So a simple solution was variable time discounts.
Putting It All Together
So, to recap, I took 2015 redraft projections and used them to calculate EVoB and EVoS based on specific league settings. I assumed that those EVoB / EVoS figures represented a player’s “true talent level”, except in instances where I had compelling reason to believe otherwise, in which case I entered improved estimates. I then estimated a player’s remaining career.
A player’s projected EVoB/EVoS, then, was their 2015 EVoB/EVoS, plus their estimated “true EVoB/EVoS” times their expected remaining career after 2015. To that projected future production, I applied a time discount. With a 10% time discount, I valued 2016 production at 90% of 2015 production, 2017 production as 81% of 2015 production, 2018 production as 73%, and so on.
The end result of this was a projected career EVoB/EVoS for every player, tailored to specific league settings and owner preferences. Or, in short… a dynasty equivalent to VBD.
What Does This Mean
In theory this means that, (assuming you are comfortable with my processes), we now have access to all of the advantages VBD confers in redraft. Projected EVoB/EVoS totals are precise, pointing out not only which players are more valuable, but by how much.
Projected EVoB/EVoS totals are directly comparable across positions. Instead of having to rely on simple heuristics like “wide receivers are more valuable than running backs in PPR”, someone can run the numbers and actually see where Le’Veon Bell winds up in relation to the top receivers.
Finally, with the ability to tweak settings at will, projected EVoB/EVoS totals can be customized to fit any owner in any league at any time.
Just like VBD, this doesn’t represent a panacea. Projections will still only be as good as the underlying assumptions. But by building most of the assumptions on observed historical data and the most accurate redraft projections in the industry, I feel we increase the reliability of our dynasty projections as much as is possible.
I think that this is a very exciting new development that has been five years in the making. In just the few weeks since I’ve gotten my early-stage algorithms up and running, it has quickly become my “favorite toy”. I’ve already used it as the framework to build several blockbuster trades in my leagues.
I’m so excited about it that I’m going to be bringing it to you this year, too. I will be writing a new weekly column that provides updated value charts every week for contenders, rebuilders, and balanced teams in both PPR and standard scoring.
Projected EVoB / EVoS is still a new tool, and there is still room for me to improve it. If you have any questions, comments, or suggestions, please don’t hesitate to contact me and let me know. In the meantime, I hope that you find it as fun and as helpful this coming season as I have.