Welcome all to the first edition of my new article series, ‘The Fantasy Detective’, where I will be analyzing different fantasy football strategies, trying to come to the conclusion if they are useful or not.
One of the biggest questions going into the fantasy football season every year is what round you should draft a QB in. Prevailing sentiments tell us to grab our RBs and WRs (maybe even TEs) and save the QB for later rounds. This idea is formed on the knowledge that there are generally more good QBs to go around (relatively) than there are for the FLEX positions, as you only need 1 good QB for your team, but need at least 2 good RBs and WRs each if you want to be successful. Regardless, we all have players in our fantasy leagues who insist taking a QB early (1st to 3rd rounds) is the best way to go. While there will always be variation during the fantasy season and any team could go on to win it all, there is a way for us to settle this argument through artificial intelligence (AI) simulation techniques.
Performing the Simulation
My method to complete this task was simple. I first programmed 8 different AI fantasy players (5 main and 3 experimental), each with different drafting strategies. I then simulated 1,000 drafts between these AIs (I made it a 10-player league–yes, some AIs were used twice to fill the league), randomizing the draft order each time. Following each simulation, my program took down the total WAR and average player WAR for each team the AIs drafted. WAR (wins above replacement), tells us how many wins a given player would give you in a full season (on average) more than a replacement level player (a player on waivers), assuming the rest of your team is average and the other teams you are playing are average. While this stat does make a lot of assumptions, it allows us to gauge how much better players are in comparison to others and is a mathematically strong metric to base player rankings on. After the 1,000 draft simulations, we can look at each team’s average WAR to see which strategy worked the best to gather the best players (WAR calculated using PPR fantasy points).
Before we evaluate the results, let’s take a look at the 5 main AI strategies. Each strategy follows 4 steps (in order) when drafting players and never waivers off these tracks.
Strategy 1: QB1 or Bust
- Draft the best QB available
- Draft the best 4 RBs and 3 WRs available
- Draft the best remaining TE or QB
- Draft the best player available
Strategy 2: Heavy WR FLEX
- Draft the best 3 RBs and 4 WRs available
- Draft the best TE or QB available
- Draft the best TE or QB available (whatever one not taken before)
- Draft the best player available
Strategy 3: One of Each
- Draft the best RB and WR available
- Draft the best TE or QB available
- Draft the best TE or QB available (whatever one not taken before)
- Draft the best player available
Strategy 4: Heavy RB FLEX
- Draft the best 4 RBs and 3 WRs available
- Draft the best TE or QB available
- Draft the best TE or QB available (whatever one not taken before)
- Draft the best player available
Strategy 5: Normal Flex
- Draft the best 2 RBs and 2 WRs available
- Draft the best TE or QB available
- Draft the best TE or QB available (whatever one not taken before)
- Draft the best player available
Evaluation of Results
Below we can see the first set of results. I plotted each strategy’s distribution of their average drafted player WAR for the 1,000 draft simulations. Strategy 1 is red, 2 is light blue, 3 is yellow, 4 is magenta, and 5 is mint green.
Right away, it is stunningly clear that strategy 1 is the worst (by far). With a mean player WAR of 5.4, it lags behind the top strategies by around 0.5 WAR. That is the equivalent to half of a win per player compared to strategies 3 and 5 (yellow and mint), which have mean player WARs of 5.9 and 5.85, respectively. Strategies 2 and 4 don’t look too great either (blue and magenta). They have average player WARs of 5.7 and 5.6, respectively, setting them in the middle of the pack.
Another look below shows each strategy’s average total team WAR following each simulated draft. These graphs should be merely identical (barring the x axis).
We can also take a look at each team’s average positional ranking–that being the positional ranking of the average player that the strategy drafts (the lower the better). We see that strategy 2, Heavy WR FLEX gives you the best average positional rank. This makes sense, for this strategy targets the best ranked RBs and WRs–while it is getting highly ranked players, since there is larger drop-off in WAR for these positions, their player WAR average is much lower relative to the others. With this in mind, we realize that positional rank doesn’t necessarily matter, at least not close to as much as the true value of a player’s WAR is.
Wow. That’s a lot. To break down what we’ve learned so far, we can break our 5 strategies into 3 tiers.
- (Best) One of Each and Normal Flex
- (Mediocre) Heavy WR FLEX and Heavy RB FLEX
- (Worst) QB1 or Bust
Now that we have our tiers, let’s break them down a bit further to see what the best time to draft a QB is. Below you will find the proportions for each strategy of how many times they drafted their first QB in each given round (data used from another experimental trial hence being out of 2,000, zoom in for a better view).
Since most strategies vary on which rounds they take their first QB (based on their draft position and other selections before them), it is important to see which round for selecting the first QB yields the highest total player WAR for each strategy.
QB1 or Bust
Round of first QB selected: 1
- Avg. total team WAR: 80.6
- Avg. player WAR: 5.4
Heavy WR FLEX
Round of first QB selected: 8
- Avg. total team WAR: 86.2
- Avg. player WAR: 5.7
Round of first QB selected: 9
- Avg. total team WAR: 83.1
- Avg. player WAR: 5.5
One of Each
Round of first QB selected: 3
- Avg. total team WAR: 88.4
- Avg. player WAR: 5.9
Heavy RB FLEX
Round of first QB selected: 8
- Avg. total team WAR: 83.9
- Avg. player WAR: 5.6
Round of first QB selected: 9
- Avg. total team WAR: 81.0
- Avg. player WAR: 5.4
Normal FLEX
Round of first QB selected: 5
- Avg. total team WAR: 86.6
- Avg. player WAR: 5.8
Round of first QB selected: 6
- Avg. total team WAR: 88.9
- Avg. player WAR: 5.9
Looking at our top two strategies, Normal FLEX and One of Each, we can see that our target range for drafting a QB should be rounds 3, 5, or 6. As these strategies are fairly similar, it is reasonable to go down any avenue as they should give you similar results. However, if you are looking for the best mathematical strategy, your best bet is to take your first QB in the 6th round.
Experimental Strategies
Now that we have answered our initial question, let’s take a look at some alternative strategies, and how they performed against our 5 main ones. This will help us see which strategy is best in larger leagues where there are more players with different drafting strategies. I devised 3 new AIs, which are outlined below.
Strategy 6: TE+QB Combo
- Draft the best TE available
- Draft the best QB available
- Draft the best 3 RBs and 4 WRs available
- Draft the best player available
Strategy 7: TE+FLEX Combo
- Draft the best TE available
- Draft the best RB or WR available
- Draft the best QB available
- Draft the best 3 RBs and 4 WRs available
- Draft the best player available
Strategy 8: Late Round QB
- Draft the best TE available
- Draft the best 5 RBs and 5 WRs available
- Draft the best QB available
- Draft the best player available
After running through the same process as the first simulations, now with the 3 new strategies included, let’s take a look at how they performed (still 10 teams, just different strategies instead of the duplicates). TE+QB Combo is dark blue, TE+FLEX Combo is turquoise, and Late Round QB is pink. Notice the light blue curve in the fourth grid, Heavy WR FLEX, which has become one of the most effective strategies.
The results were disappointing. All three strategies seemed to fall into tier 2 as mediocre options. Even the seemingly best strategy of the 3, TE+FLEX Combo, falls behind One of Each and Normal FLEX. For reference, TE+QB Combo took a QB in the 2nd round every time, TE+FLEX Combo took a QB in the 3rd round every time, and Late Round QB took a QB in the 12th round every time. Below find the averages for each strategy throughout the 1,000 experimental simulations.
QB1 or Bust
Round of first QB selected: 1
- Avg. total team WAR: 82.9
- Avg. player WAR: 5.5
Heavy WR FLEX
Round of first QB selected: 8
- Avg. total team WAR: 88.6
- Avg. player WAR: 5.9
Round of first QB selected: 9
- Avg. total team WAR: 85.1
- Avg. player WAR: 5.7
One of Each
Round of first QB selected: 3
- Avg. total team WAR: 87.1
- Avg. player WAR: 5.8
Heavy RB FLEX
Round of first QB selected: 8
- Avg. total team WAR: 86.9
- Avg. player WAR: 5.8
Round of first QB selected: 9
- Avg. total team WAR: 82.6
- Avg. player WAR: 5.5
Normal FLEX
Round of first QB selected: 5
- Avg. total team WAR: 87.4
- Avg. player WAR: 5.8
TE+QB Combo
Round of first QB selected: 2
- Avg. total team WAR: 84.2
- Avg. player WAR: 5.6
TE+FLEX Combo
Round of first QB selected: 3
- Avg. total team WAR: 84.7
- Avg. player WAR: 5.7
Late Round QB
Round of first QB selected: 12
- Avg. total team WAR: 84.0
- Avg. player WAR: 5.6
There are some notable differences–First, the Heavy WR FLEX strategy (when taking its first QB in the 8th round) yielded the best results. With this in mind, while risky, this strategy could be effective in big leagues especially, where players have many different draft strategies (as was so in this second simulation). Our other two ‘best’ strategies, One of Each, and Normal Flex regressed a bit, but still are at the top. This gives us a multitude of different options for where we decide to take our first QB in larger, more unpredictable leagues–we can effectively select a QB in the 3rd, 5th, or 8th round if we want to be risky. Again, the most mathematically sound/reliable option is the 5th or 6th round, but you could make do with any decision, especially depending on the people you are drafting with.
As always, feel free to reach out to me on Twitter @analytacist with any questions, or if you are interested in seeing how your drafting strategy would do in the simulation!
