We have created our own version of a customizable Mock Draft Simulator, which gives you the opportunity to be the general manager of any NFL team.
You can choose the team(s) you want to draft for and customize the mock with different settings. While the settings for the number of rounds and the speed of the simulation is reasonably self-explanatory, it might be worth discussing some of the other settings so that you can come away from your draft happy with your selections.
The tool features two boards: Our own PFF Big Board, compiled by our lead draft analyst Mike Renner, and a public board that uses the average draft position from Benjamin Robinson’s valuable resource, GrindingTheMocks, which collects mock draft data from the experts, media and fans.
You can choose either of the two boards, or you can decide to blend the two. Changing the board can change the behavior of the simulation, as there are more than a few players who we are higher or lower on than the public. Next week, we will present results of 100,000 simulations with various board settings, and among other interesting takeaways, we’ll illustrate the strongest discrepancies between our PFF Board and the public mock draft data.
Draft for needs?
The probability of a given player being selected at a given spot mainly depends on his position on the chosen board. However, there are some adjustments, one of which is team needs.
The algorithm uses two sorts of team needs: The needs presented in our latest article on team needs and Kevin Cole’s WAR projections for 2020, split up by position. The lower the WAR projection of a given position for a given team, the higher the need for this team. In particular, the adjustments for needs are not binary; they change continuously from team to team based on our own models.
In short, the need adjustment will make it so that the Minnesota Vikings, for example, are more likely to draft a cornerback than a safety, even if the cornerback is slightly lower on the board. How much should a position of need climb up the board? You decide with your setting!
Care for Positional Value?
Just like the team needs, positional value can be another factor that changes the behavior of the simulation. For example, if the setting is activated, safety Antoine Winfield Jr. will go off the board earlier on average than running back D’Andre Swift, even though Swift is four spots higher on the PFF Big Board.
With the exception of quarterbacks for QB-needy teams, the effect isn’t too large with the standard settings, and you can decide to either make it larger or remove it altogether. However, be careful! Lowering or removing the effect of positional value will treat the quarterback like any other position, so there would be a reasonably good chance that the Cincinnati Bengals pass on Joe Burrow for Chase Young, Jeffrey Okudah or another prospect who is high up on the board, which is not necessarily a realistic reflection of what we would expect to happen on draft night.
This might be the most mystifying setting. At each spot of the draft, the algorithm computes a list of candidates and assigns probabilities that each gets picked based on the settings explained above. The simulation will sample one of these candidates based on the respective probabilities. The randomness setting can skew these numbers toward the best candidates.
As an example, we use the No. 2 overall pick with Burrow already gone to the Bengals. Using the public board and removing needs and positional value, we get several candidates when using the standard “random” setting, with the first four being Chase Young (53.7%), Tua Tagovailoa (10.8%), Jeffrey Okudah (8.4%), Isaiah Simmons (7.6%). Even a prospect like Tristan Wirfs is a candidate, being selected 3.2% of the time.
If we use the “less random” setting, these probabilities are transformed, resulting in a higher probability for the best candidates: Chase Young (73.5%), Tua Tagovailoa (12.1%), Jeffrey Okudah (7.0%), Isaiah Simmons (4.6%). Tristan Wirfs has now vanished from the list of candidates.
If that’s still too randomized for you, you can use the “not random at all” setting, which will always select the candidate with the highest probability. In other words, every simulation will be the same, which somewhat limits the fun of using the Mock Draft Simulator.
After choosing the settings, one can start the simulation. From there, the functionality of the simulator is quite self-explanatory. Once the mock draft is finished, you will be presented with a summary of your picks and trades, both of which will be graded based on our projections. Feel free to share your results on social media!
Before we explain more about the math behind draft grades and the possibility of trading, we shed light on a somewhat hidden feature that is unique to our Mock Draft Simulator: After the mock is finished, you can choose to get a closer view into the simulation for each pick.
By clicking on the row of a pick in the final table, you can look up the probabilities that the simulation sampled through. This also works for your own picks, so you can essentially check what the simulation would have done with your pick. This is a crucial step towards understanding the draft grades, which we will explain in the following sections.
The simulator features two types of trades for the user: User-initiated trades and CPU-initiated trades.
As the name suggests, you can initiate the former type of trade at any point during the draft, and you are even asked whether you want to perform them pre-draft. By clicking on the “offer a trade!” button at any point after starting the draft, the simulation is paused and you can initiate a trade with any team (pre-draft) or with the teams between the current and your next pick.
To make these trades easier during the draft, it might make sense for you to slow down the simulation so you are able to react to the CPU picks. When you don’t want to use that feature anymore, you can speed it up again. After you’ve selected a team to call, you can make any trade offer. However, this should be mostly used for trading up.
Why? Because not every trade passes through. Based on our draft value charts, we will compute the percentage that the CPU wins the trade, and any trade the CPU wins more often than not will be accepted.
In the instance where you make a halfway reasonable offer to trade up but the team you want to trade with would select a QB more than 50% of the time with their next pick, your trade offer is analyzed with the QB draft value chart, which significantly shifts the trade math (as explained in this article).
When the CPU wins the trade less often than 50% of the time, it will accept the trade based on the trade details. A trade the CPU wins 45% of the time will still be accepted 48% of the time, but this probability decreases drastically the more lopsided the trade is. If the CPU wins 40% of the time, it will accept the trade only 21% of the time.
Since trades are mostly won by the downtrading team, trading down and getting a market-based return is kind of unlikely with this type of trade. Just like in the real draft, downtrading is mostly initiated by the team that wants to trade up. Whenever you have to make a selection, the simulation not only computes the candidates for your own selection, but also for any other team.
Based on the probability of the best candidate, the delta to the probability of the second-best candidate and the number of spots the team would have to trade up, any team is assigned a probability with which it’s interested in trading up to your spot. Based on these probabilities, you’ll see a selection of interested teams (or no team at all if you are unlucky, which often happens later in the draft).
In short, a team’s interest in trading up to your spot depends on the remaining board. It is chosen probabilistically, and each simulation is different. While the Dolphins will want to trade up to the second overall pick fairly often (as Tua is clearly the most likely candidate on their board with Burrow gone), there will be simulations in which they aren’t interested. A team with a need at offensive tackle is much more likely to be interested in trading when there is only one of the top offensive tackles left, as opposed to when Tristan Wirfs, Andrew Thomas and Jedrick Wills Jr. are all still on the board.
From time to time, surprising trades can happen. After all, who of us would have predicted that the New Orleans Saints would trade up 13 spots in the first round to grab edge rusher Marcus Davenport in 2018?
Once you’ve selected a team to trade with, you can then make a trade offer, which automatically includes your current pick. The advantage of a CPU-initiated trade down is that the CPU is much more likely to accept a trade offer, even if it loses the trade more often than not (the team wants to get their guy). If the trade is accepted, all involved picks are exchanged and the draft continues with the trading team selecting the best candidate on their board 100% of the time (e.g., if Miami trades up for Tua, it will select Tua. It won’t trade up and randomly select Chase Young).
With the trades being explained, always keep in mind that everything is probabilistic, and crazy trades that may leave you feeling like you’re completely whipping the other team might happen if you are lucky. This is more consistent with the real NFL trade market than you might think. After all, we’ve just recently witnessed DeAndre Hopkins getting traded for a second-round pick.
A unique feature of the PFF Mock Draft Simulator is that grades, in the common 0-100 style, are given for each pick the user makes.
The crucial ingredient in these grades is the PFF consensus WAR projection for each draft prospect over their first four years in the NFL. There are also some other aspects that go into these pick grades:
- Drafting a position of need yields a boost
- The grade is created by comparing the WAR projection of the selected player against the expected WAR projection the simulation would have gained at this spot with the current remaining players left on the board.
- This means that the same player at the same draft spot can yield different grades when there are different players remaining.
- The highest player on the board or the simulation's best candidate isn’t necessarily the player with the highest WAR projection. The boards aren’t optimizing for WAR projections, so the simulation isn’t optimizing for WAR projections.
- This means that selecting the highest remaining player on the board doesn’t guarantee a good grade.
While the first point is clear, the second point could need some clarification: Each prospect in the draft comes with a college-to-pro projection using the model created by our own Eric Eager with the help of AWS machine learning capabilities. The WAR of a player not only depends on how good a player has been on a per-snap basis, but it also depends on the number of snaps he plays. So, we use the big board to project the number of snaps the player will play and use the projected per-snap efficiency along with the projected snaps to come up with a WAR projection for each prospect. Since the projections not only depend on our models but also on the scouting work of our draft analysts, we call these the PFF consensus WAR projections.
The final WAR projection for the selection also depends on the team, as there are diminishing returns if the team already has a lot of good players at the same position.
After coming up with a WAR projection for each player for the team currently on the clock, we can compute the expected WAR added by the simulation, using the probabilities with which the simulation selects the different candidates. The final grade is obtained by comparing the WAR projection of the selected player to this expectation.
In particular, the Miami Dolphins would select Tua Tagovailoa over 90% of the time when he is on the board at No. 5 overall. Thus, picking him doesn’t yield a very high grade, as the expectation is already very high. While most people would pick Tua at this spot, doing it is not necessarily a strong achievement — it’s expected.
On the flip side, the Jacksonville Jaguars have a strong need at their interior defensive line, and with Derrick Brown and Javon Kinlaw both available, there would be a good chance the simulation selects one of those two (if drafting for needs is activated). However, these two players aren’t necessarily the most valuable selections. If you can overcome the will to fix the position and select a valuable wide receiver like Jerry Jeudy or Henry Ruggs III instead, you will almost certainly get graded in the 90s for that selection.
Since the variance of WAR projections among possible candidates is much smaller later in the draft, the actual selection and the expectation will be much closer to each other in the later rounds. Hence most of the later grades will hover around the average grade of 60.0, illustrating that the bottom of the board is mostly a crapshoot. Giving out really high or really low grades would inspire much more confidence than we should have in whether a late-round pick will be fruitful or not.
Grades for trades are much more straightforward, as they only depend on the rate at which you win the trade and the average WAR you added through the trade. If you select a QB with a pick you got via trade, it will be accounted for.
The overall grade is computed as a weighted mean of the pick grades, with earlier picks carrying stronger weights. Finally, each positive trade (with a grade better than 60.0) will increase the overall grade, while each negative grade (worse than 60.0) will decrease it. In particular, a one-round draft with one pick graded at 80.0 and one trade graded with a 70.0 will not yield an overall grade of 75.0, but an overall grade of 83.6, as you have not only selected a good player, but you've also added additional draft value by trading in a smart way.
Now that the burning questions about the usage and math of the NFL Mock Draft Simulator have (hopefully) been answered, we hope you enjoy using it until the wait for the real NFL draft is finally over. If any questions — especially about the underlying math — arise, feel free to reach out on Twitter.