This post is going to be crazy because I am going to deep dive on fantasy football draft strategy and, in tribute to my loyal fan base, which is no doubt sick of this new football thread, include a reference or two to online marketing. It is going to be sick!
Let’s get started. Most people doing fantasy drafts simply make a list of who they think the best players in the league are and bang, bang, bang, draft.
Let’s call that “Old school media planning strategy”. Or let’s call it “People who evaluate campaigns based on CTR“. Or “people who don’t love math”. Or “people who suck!” Or “me”, most years.
Anyway, here is what I think you are supposed to do. First, recognize that there is data and there are algorithms. Algorithms take data and tell you what you should do, hopefully. If you feel as though what the algorithm is telling you to do is wrong, the problem could be that the algorithm is bad or the problem could be that the data is bad. Figuring this out is absolutely critical.
There are three algorithms in fantasy football draft strategy, one of which is simple and perfect, one being the most interesting algorithmic challenge, and one being shear poker madness. Here is the model I use when I think about optimizing the fantasy football draft process:
The algorithm that works right every time is the middle algorithm: Given an accurate prediction of future player performance and an understanding of the scoring rules for your fantasy league, you can predict with absolute accuracy the scoring of other players. Simple. The problem is everywhere else. Let us walk through the process sequentially.
The first thing we want to do is generate a prediction of how each player will perform in an upcoming year. The best indicator we have of this is past performance, however much like your favorite mutual funds, past performance is no guarantee of future success. Everyone who is remotely familiar or unfamiliar with sports recognizes this, so there is a degree of guesswork and other “psuedo-algorithms” injected into this process. One could take average performance over the last several years, the age of the player, and an algorithm that compares past performance over time to performance of other players at the position at a similar age to model performance in the upcoming year, plus you are really high on this rookie running back. But the point is that you want to model what you think they will do on the field next year. Not “where I think they should be in my fantasy rankings”. If you think Player X should be above Player Y in your rankings, you clearly think that Player X will have better statistical production than Player Y. If not, you are doing yourself a grave disservice. My point is that you don’t simply bump someone up in your fantasy rankings. What you do is model how their actual production will actually be different then run it through the algorithm to generate your fantasy rankings. Change their predicted production, not their ranking. Post algorithm “manipulation” indicates a flaw in an early step in the process that should be addressed at its root. So we take a bunch of data and generate a prediction of how each player will perform in the upcoming year. Then we apply league scoring rules to determine how many points each player will score that year.
A lot of fantasy football players would stop there. If you had a list that showed with absolute certainty how many points each player would score, you might think you would be in pretty good shape for your fantasy draft, but you can go further. A really excellent model then arbitrages the players, positions and league. What you would like to do is generate a model that arbitrages the value of players against each other. Let me give you a simple but preposterous example: You are in a 5 team league. You know with absolute certainty the point production of a variety of players and you are in a league that only allows you to start 2 players each week: A QB and a RB. You have the first pick in the draft and the highest scoring player in the league this year will be a QB. Simple, you pick the QB, right? Wrong. By analyzing the relative strength of each position, you realize that the top 10 QBs all score with 10 points of each other, yet the #1 RB scores 50 points more than the #2 RB. Knowing this, you realize that the relative value of the #1 RB is much higher than the relative value of any other RB in the draft (or league). Selecting this RB will ensure your fantasy victory this year because having the #9 QB (The last pick of the second round, worst case scenario) is not significantly different than the #1 QB. The arbitrage opportunity is typically more subtle (After the top 3 QBs, the next 6 QBs are basically the same), but identifying the relative value of players is critical to successful drafts. Recognizing where the “cliffs” are is key.
You may need to amp this model up by modeling production in the last weeks of the season (when fantasy playoffs are likely) and considering the impact (resting players, WRs vs. Darrell Revis, etc.) and factor in the draft tendencies of other players (True story: I drafted Tony Romo in the third round last year just to trade him to a guy that loved Tony Romo – the result: I got a Top 20 player for a Top 30 player.). This is the poker part of the equation. Knowing your cards and the cards on the table are one thing, reading other players counts for a lot – The #1 defense is worth a lot more than the #2 defense, but how long can you afford to wait to grab that defense?
This same logic is critical in online advertising. Recognizing the effective cost of media is huge, but identifying the best opportunities to exploit media value is how you take it to the next level.