Introduction
Of all the formats of fantasy baseball, the Dynasty variety has probably risen the most in popularity in recent years, mostly due to the additional strategy to appeal to the fantasy expert and the skyrocketing availability of prospect information.
When formulating an initial strategy for creating a Dynasty team, in either a draft or auction format, an owner usually faces two options: try to win now or play for the future. Should they go all-in for the first year, selecting players like Nelson Cruz or Edwin Encarnacion despite their advanced age? Or should they forego competing in the first year and instead select top prospects who ‘should’ become future stars?
Rather than commit to one approach, what most owners do – after all the usual suspects are selected in the first few rounds – is to take young (usually between 23 and 27 years old) players who should also contribute in the current year (e.g. Michael Conforto, Matt Olson, Max Kepler). In other words, they try to win…but with players who would be expected to produce over the next few years as well. These owners would still take a prospect with an early-to-middle round pick but only reluctantly because it is a pick being ‘lost’ that could have been used to help in the first year. It’s often only in the later rounds – after the player pool of current-year contributors becomes depleted – when most owners then go about filling out their prospect roster. Seems completely reasonable as an approach.
This strategy of aiming to both win now and win later, though, is becoming harder and harder to do – especially with the increased expertise of dynasty owners – and all too often ends up achieving neither. So, then…what is the right approach?
Are older players being under or overvalued? Are prospects being under or overvalued? Should you wait on pitching? These are the questions that probably every dynasty owner believes they have a general handle on – or at least has a feel for – but is there data to help us validate what the best approach is or should be??
To answer these questions, Walter McMichael (@realfakewalter) and I retrospectively analyzed four (4) Dynasty Leagues, each with their start-up draft occurring in a different year (spanning 2016 to 2019, inclusive). The Dynasty Leagues were specifically selected because (a) they had expert owners (to mitigate the influence of “novice” owners who may not be accurately representing ‘current’ player valuations), (b) they had consistent owners (to help mitigate owner-specific ‘subjectiveness’ of a player’s value), and, well, (c) because Walter was in all of them and had access to the data. Each of the leagues are 5 x 5 roto, employed a snake draft (ie no auctions), and are ‘keep forever’ leagues (where every year the roster is carried over from the year before, with an additional yearly “supplemental” draft taking place before the season).
What we were most interested in evaluating was whether there was a measurable change in perceived value of hitters and pitchers over time. The metric we used to measure this was Average Draft Position or ADP (the cardinal rank of the player selection). ADP was chosen because it essentially represents the market’s perception of a player’s (trade) value. If we could see how the value of certain players change over time, it could help us maximize the ‘future return’ of the players we draft.
The results of this analysis will be discussed in this three (3) part series. Part 1 here provides the boring but essential background on the data used and some high level observations of the player pool. Part 2 will discuss observations of players taken specifically in the Middle and Later Rounds of the drafts (where leagues are anecdotally ‘won’). Part 3 provides analysis of the Early Rounds (rounds 1 to 6) for completion and presents the conclusions of the analysis.
Caveat 1: Actual value (produced by the player during that baseball season) was not determined; merely ‘perceived’ value (as expressed via ADP in following years) was determined.
Caveat 2: Because some leagues in the data set are batting average leagues and some are OBP, the ‘accuracy’ of the analysis is slightly affected. However, although certain players’ values would vary depending on whether it’s a BA or OBP league, this influence was considered to be insufficient to invalidate the thrust of the analysis. Additionally, because aggregated data was used (which would smooth out the impact of individual outliers), the data, although somewhat dirty, could still be considered to generate reasonable conclusions.
Caveat 3: Because there were only four drafts analyzed (one per year), the conclusions should not be considered to be representative of all types of dynasty owners or leagues. It is the very definition of small sample size. However, because industry vets were involved in each of the Dynasty Leagues (and, as mentioned, appear consistently across multiple leagues), the analysis could be considered reasonably representative of ‘mainstream industry wisdom’. (Now that Brad Johnson (@BaseballATeam) is aggregating Dynasty ADP from Fantrax, his dynamic data should start to be used moving forward).
Caveat 4: Most leagues are 15-team formats, however, one of the leagues used has 30 teams. This, also, should have negligible impact on the analysis because irrespective of the format of the league, the players’ ordinal rank would still be a proxy for relative player value. (Although some players may get selected earlier or later due to positional scarcity, this is considered to have a relatively minor impact on the derived insights). Nonetheless, because the idea of draft ‘round’ is sometimes more intuitive than ‘pick’ to understand a player’s inherent value, when the term “round” is used, it is used to describe where in a 15 team draft a player was taken. Therefore, a “3rd round pick” is used to describe someone taken between the 31st and 45th pick of the draft.
One final note is that when performing the following analysis, it became clear that the “noise” of yearly value fluctuations was significantly reduced when focusing on player valuation changes observed over a period of two years rather than over adjacent years. Although the ‘one year’ analysis bears out in a similar way, for the purposes of clarity of illustration, only the ‘two year’ (or “Year+2”) analysis will be shown.
The Data
The dataset for the analysis consisted of every player selected who had either:
- A 2016 ADP and a corresponding 2018 ADP (denoted as ADPYEAR + 2) or
- A 2017 ADP and a corresponding 2019 ADP (denoted as ADPYEAR + 2)
- If there was no ADPYEAR + 2 (i.e. the player was not drafted), the player was considered a ‘bust’
Important Note: If a player was selected in 2016, was not drafted in 2017, and then was drafted in 2018, the player would still be included in the analysis (with a 2016 ADP –> 2018 ADPYEAR + 2); the ‘bust’ in the middle year was ignored.
The same player may therefore appear multiple times. (e.g. Mike Trout – selected in 2016, 2017, 2018 and 2019 – would have 2 data entries: one where he was selected in 2016 at the age of 24 (and then drafted in 2018); and the one where he was selected in 2017 at the age of 25 (and then drafted in 2019). Similarly, a player taken in 2016, not drafted in 2017, then drafted in 2018 and 2019 would have only one entry: ADP of 2016–>ADP of 2018.
The High Level Analysis of the Player Pool
To get a general feel of the data set, the overall breakdown of players selected is tabulated below:
For example, this shows that there were 21 players taken at age 18 and they account for 2% of all players taken. Of the 21 players, 2 weren’t selected two years later (“busted”) while 19 ‘survived’ and were taken two years later. In other words, 10% of age 18 players busted while 90% were selected two years later.
That’s a lot of data to parse so here’s a histogram of the players (based on their age at the time of their first selection):
To arrange the data into more bite-sized chunks of data, we collapsed the above data into four (4) equivalently sized (more or less) age quartiles: Age 17-22, age 23-25, age 26-29, and age 30+. The resultant data looks like this:
Here is the bust rate (by age quartile) of the entire data set:
The most striking conclusion from the above data is that the older the player, the more likely that they ‘bust’ and are essentially “out of the league” in two years. Although this is obviously intuitive, the fact that the bust rate is so high – players who are 30 years old or over have a bust rate over 1 in 4(!) – was very surprising to us. Owners whose competition windows are more than two years out should be aware that there is around a 25% chance that their older assets will become worthless within two years.
Survivor Analysis – Overall Player Pool
The performance of the 84% of the players who did not bust (referred to here as ‘survivors’) was plotted on a scatterplot with ‘original’ ADP on the x axis, and ADP YEAR + 2 on the y-axis. The COLOR and SIZE of the dot defines the player’s AGE at the time he was originally selected: younger = “Greener” and bigger; older = “Redder” and smaller. (The legend is listed on the graph). This was done to make it easier to discern if there were general age trends.
To make it easy to see if a player improved or worsened their ADP, we also plotted a y=x trend-line (which is equivalent to a player having an identical ADP in “year 0” and in “year + 2”). The vertical distance above or below the trend-line would therefore represent by how much the ADP improved or worsened two years later. Also, to be more intuitive, the y-axis was reversed so that dots ABOVE the line correspond to an improvement in ADP YEAR + 2; dots below the line indicate a worse ADP YEAR + 2. A dot found in the very top right of the graph would, therefore, represent a player selected with an original ADP of around 800 and then picked as the #1 selection two years later, with the color/size of the dot indicating how old the player was when he was taken in year 0.
Below is the overall plot of survivors (n=1055; excluding the 202 ‘busts’ who were not selected two years later):
It can be easily seen that most of the dots in the top right (which represent late round picks that improved their ADP two years later) are mostly GREEN. In other words, most of the players that improved their value two years later were younger than age 26. Similarly, there are very few dots above the line that are orange or red. This generally tracks with expectation that younger players would be expected to improve their value, and older players would be expected to have their value worsen.
Below is the same plot, separated into the four (4) age quartile layers (of approximately equal sample size):
This data seems to show that when it comes to player age, dynasty owners either intentionally apply a significant future discount value (i.e. emphasize the present over the (two-year) future) or are unintentionally over-valuing older players (and undervaluing younger players) in order to have better player performance in the current year at the expense of the near future.
In leagues that are expected to continue more than a few years, this may be important data.
So far, nothing here seems surprising. We generally know all this. Next time, in Part 2, we dive deeper – and get into the actual analysis – by separating the ‘draft’ into three (3) distinct sections and analyzing them separately for observable tendencies or inefficiencies in owner behavior depending on the section of the draft:
- Early Rounds (1st through 6th, corresponding to picks 1 to 90)
- Middle Rounds (7th through 30th, corresponding to picks 91 to 450)
- Later Rounds (31st onward, corresponding to picks 451 onward)
Leave a Reply
Friends don't let friends talk to themselves.
You must be logged in to post a comment.