The Method to My Rankings Madness, Part 1
The Method to My Rankings Madness, Part 1
Our new staff dynasty rankings released over the weekend, and if you’ve seen them, then you probably noticed that my contributions are, in places, radically different from those of Bryan and Chad. The root of many of those differences is my personal leaning toward perceived safety over potential longevity, and that is not idle philosophy. I based my rankings heavily on a framework that values players with a statistical model, and so much of the leg work is done before I’ve even considered my opinions. Here is how I got there.
I explained the components of my valuation methodology in great detail last summer, and that goes into greater depth than I will in this summary.
1. Create 2013 projections
Previously, I had always relied on projections from other sources I trust, such as those Pro Football Focus will release closer to the season. Since I am starting in February and will need to repeat this process periodically, I decided to create my own. Without the time to really dive into this—and I fully intend to someday—I used a simple 3-year weighted average of fantasy points regressed to the replacement level whenever time was missed.
I wrote about that entire process and published the projections here.
2. Calculate points above the replacement level for each position and season
Whenever I prepare for a draft, I always start as if it were an auction. I find it easier to operate in terms of dollars, and when I finish, it is a simple process to sort prices to create ordered rankings.
Well, consider in an auction: the players that are below the replacement level are those that should not be drafted. By definition, they are worth $0. The same is true with respect to fantasy points. In absolute terms, a player such as Blaine Gabbert will score points. However, in all but the deepest of leagues, there will be better alternatives to Gabbert on the waiver wire. The fantasy points that matter, then, are those scored above a baseline of those replacement-level options.
At a broad level, positions have different volatilities in their value, often because of talent cycles, and sometimes because of the evolution of the game. For example, ten years ago, no quarterbacks threw for 5,000 yards in a season. Now, several do every year. As the combined value of all players at a position change relative to the baseline player at that position, the value of each position changes relative to other positions.
Think of it like exchange rates. If I had a dynasty team that was strong at quarterback but weak at wide receiver, I would want to trade that surplus of quarterbacks to fill the deficit at receiver. Ten years ago, that would have been difficult because most quarterbacks had little differentiation from the replacement-level options. Today, if I had a pair of those 5,000-yard passers, I could buy a better receiver with one of them. Because elite quarterbacks have distanced themselves from the baseline quarterback to a greater extent than other positions have done the same, quarterbacks, in general, have become more valuable.
We all make some effort to capture our expectations for those trends in our dynasty prices, but changes in the relative value of positions are still unpredictable. What I can estimate, however, is the volatility of positions. The less volatile the position, the less likely the players I draft at that position will lose value over time by holding currency, so to speak, in that position.
Here are the combined points scored above replacement in a typical 10-team league for each position since 2008:
And here is the same as a percent of points scored by all positions above replacement:
Finally, here is the standard deviation of the value of each position above replacement, which is an approximation of the risk of volatility:
3. Assign categories to players
This was the biggest hurdle for me. Previously, I did this manually. In order to automate the process, I had to define each category in a way that a script could determine from a statistical profile. Here is what I came up with for each category.
All ages are as of September 1 of each year.
Retired: 39 years old for QBs, 33 for RBs, and 37 for WRs and TEs, as discussed previously.
Rookie prospects: Drafted in the first two rounds for QBs, and in the first four rounds for RBs, WRs, and TEs.
Potential stars: Ranked outside of the top n of players, previously listed as a rookie prospect, and not yet 26 years old, where n is 75 percent of the rank of the baseline player.
Star in prime: Ranked in the top n of players in two of the previous three seasons, based on fantasy points regressed to the replacement level in the event of missed games, where n is 75 percent of the rank of the baseline player.
Ranked in the top n of players after being designated a rookie prospect or potential star the previous season.
RORA (Revelation or Aberration): Ranked in the top n of players but unqualified for star in prime.
Star in decline: Ranked outside of the top n of players, listed as either a star in prime or star in decline in either of the previous two seasons and age within 82.5 percent of expected retirement age, where n is 75 percent of the rank of the baseline player.
Non-star: Everyone else.
Previously, I had two additional categories that were star with reservations and P&Q (Potential and Question marks), both of which I eliminated because of their lack of statistical clarity and necessity given the other categories.
4. Create a category matrix
This describes the chances I will go from a category in year y to each other category in year y + 1, based on the historical trends from 2008-2012:
|Rookie prospect||Potential star||Star in prime||RORA||Star in decline||Non-star||Retired|
|Star in prime||0.00||0.03||0.71||0.00||0.08||0.10||0.01|
|Star in decline||0.00||0.00||0.00||0.00||0.27||0.00||0.09|
And this is the expected points per game of a player of each position and category:
|Potential star||Star in prime||RORA||Star in decline||Non-star|
Finally, here is the expected rate of change, an approximation of the rate of attrition, for each position-category, calculated as (Expected Value After – Expected Value Before) / Expected Value Before:
|QB||Star in prime||19.82||15.92||-0.20|
|RB||Star in decline||8.23||2.61||-0.68|
|RB||Star in prime||15.06||12.03||-0.20|
|TE||Star in decline||8.31||2.24||-0.73|
|TE||Star in prime||13.94||11.25||-0.19|
|WR||Star in decline||5.88||1.88||-0.68|
|WR||Star in prime||13.77||11.04||-0.20|
5. Combine every risk rate into one discount rate
In addition to the positional risk rate and the rate of attrition, which I calculated in the previous three steps, I also included a risk-free rate of 0.009 for all positions, as well rates for risks of opportunity and efficiency, which are split by position:
|Opportunity Risk||Efficiency Risk|
Find more of our Dynasty Content here.