Alpha DFS Model - Understanding the Data

Explaining the model's outputs.

This article explains how to read the data output by the model and some theories surrounding potential arbitrage opportunities. This article will be updated frequently as we create new ways to visualize the data. If you haven’t already, please see the Alpha DFS Model post which describes the logic surrounding our methodology, it will also be linked at the bottom of this article.

First, let’s quickly summarize the process surrounding our model:

1. Player prop bets odds are gathered for each statistical category relevant to DFS scoring.

2. Odds are converted to probabilities, creating a probability distribution for each category and player.

3. Slates/games are simulated by making a probabilistic choice for each category, for each player, and converted to fantasy points.

4. Each simulated slate/game is run through a line-up generator to create the Optimal Lineup, or what lineup would have won in that unique simulation.

For smaller slates, like single game, we typically run 5,000 to 10,000 simulations.

Output Example

Let’s check out an example of our standard output below. This is for a FanDuel, MLB, Single Game slate for CLE @ SD on 08/24/2022. In FanDuel’s MLB Single Game slates, you create a lineup of 5 batters, consisting of an MVP, STAR and 3x UTIL players. These fantasy point scores are multiplied by 2x, 1.5x, and 1.0x, respectively.

The OPT column stands for Optimal Percentage. This is the percentage of time, out of all of our simulations, that player ended up on what would have been the winning lineup. The MVP, STAR and UTIL columns provide the same information, only filtered through their positions in the lineup; the percentage of time that player was an MVP, STAR or UTIL on the Optimal Lineup.

Player Ownership

Player ownership is one of the most important aspects of playing DFS successfully. Anyone that’s played DFS knows how often those GPP winning lineups have 1% or 2% owned players. And it makes sense, imagine how smaller the pool of potential lineups you have to beat is if you have the unanimous MVP on your team. Generally speaking, you want to differentiate your team more the larger, or skewed, the contest is. It would make sense to avoid chalk in the Millimaker, and chalk it up in a 50/50 contest. What I find most interesting, though, is how much the DFS userbase over and underestimates the potential for a player to be in a winning lineup.

Ownership vs. Optimal

I do not personally create ownership projections, however, if you play enough DFS and/or know enough about a sport, you probably have a good idea of what the ownership of players will look like, within some margin of error. Someone like Manny Machado or Jose Ramirez in the above example are likely to be way over owned relative to their actual chance of being an MVP or STAR in the optimal lineup.

Obviously, this does not mean that it won’t happen, in fact in this particular game the model is suggesting it’s most likely to happen. But if Manny Machado is going to be owned as MVP 18% to 25% of the time, and he’s only going to be an Optimal MVP 14.8% of the time, you’re much better of fading him, or reducing his ownership in the case of multiple entries. Or you may even take him as a UTIL player, where he will likely be under owned relative to his 15.6% Optimal UTIL percentage, as most of his ownership will be concentrated in the MVP and STAR positions.

Players that immediately pop out as GPP MVP and STAR contenders are Oscar Gonzalez and Amed Rosario. Amed Rosario shows up as MVP in 9.7% of lineups, yet, because of his $6,000 cost, he will likely be significantly under owned relative to his Optimal MVP potential. Players like this often see as low as 2% to 5% MVP ownership (NOTE: this is just an educated guess I am making about his ownership, but low-cost players typically have low MVP and STAR ownership). In fact, most of those players under $7,000 will likely have MVP and STAR ownership percentages far below their Optimal percentage.

Arbitrage?

This is the logic that formulates much of my decision making with this data. If odds-makers and the prop betting market are better predictors of player performance than fellow DFS contestants, then we should be able to find be able to find DFS arbitrage opportunities. GPP players may find success in skewing their own lineups’ player ownership to take advantage of under owned players and minimize ownership of over-owned players.

If you haven’t already, for information on the logic surrounding methodology, see here:

Alpha DFS Model