Value Based Drafting Part I

The following article is a guest post by Frank DuPont, author of Game Plan: A Radical Approach to Decision Making in the NFL. You can also follow Frank on Twitter.

Fantasy football is all about value.  Value is at the heart of every discussion we have.  It's the idea behind sleepers (they're undervalued).  It's also the idea behind busts (they're overvalued).  Today I want to talk about an idea, Value Based Drafting (VBD), that is sort of at the core of all of the fantasy football related value discussion.  You may not think about your decisions as being related to VBD, but they probably are.  When you decide to wait on QB because you think you can get a good one later in the draft, that's a decision that is essentially based in VBD.  I'm going to spill some pixels talking about the idea of VBD, first covering conventional wisdom on the topic, and then I'm going to suggest a tweak that will hopefully improve the way you think about your draft.

A Brief Primer on Value Based Drafting

The idea of VBD is credited to Joe Bryant.  The theory is that each player's actual value isn't how many points he scores; it's how many points he scores relative to the other players at his position.  But in order to calculate that player value, you need to establish a baseline for comparison purposes.  For instance, Rob Gronkowski and Calvin Johnson were reasonably close in statistics last year.  But because they played different positions, how can we compare their relative values?  We need to come up with a baseline for WR and TE to make the comparison.

But coming up with the baseline is the thing that has perplexed fantasy owners pretty much since VBD came into existence.  Do you compare players to the average at their position?  Do you compare them to the last starter at their position?  Changing your baseline is going to change the results of the analysis.

The idea that seems to have become the dominant one is that in setting the baseline you should look at how many players from each position will be drafted in the top 100 players of your fantasy draft and then use that number to set the baseline.  I'll refer to this idea as the Top 100 Method.  As of this writing, here are how many players are going at each position in the top 100 of ADP.

So by the Top 100 Method, Arian Foster is worth however many points you can expect him to score more than RB39, while Aaron Rodgers is worth however many points we can expect him to score over QB13.

Potential Problems with the Top 100 Baseline

However, I think it's worth discussing some issues that using the Top 100 Method for setting the baseline could create and why it might be worth looking for another solution.

First, our entire goal when we draft fantasy players is to look for inefficiencies and mispricing of players.  We want to look for areas where the broad fantasy market doesn't correctly value a player or a position.  It's the only way to have exceptional results in the long term.  However, if we create our VBD baselines using the top 100 approach, we're using the market's valuation of players to set our own valuation of a position.  If the broad market is wrong, then we're going to be wrong.  The Top 100 Method of setting the baseline begs the following question:  If we believe that the market is right as to how many of each position should be in the top 100, what makes us think we'll find any market inefficiencies in terms of actual players?  If the market is right, then we might as well just draft off of ADP and hope to get lucky.  The Top 100 Method of setting the baseline gives up an arbitrage opportunity right out of the gate.

There is another issue with the Top 100 Method.  How do we know that "top 100" properly reflects the supply and demand related to fantasy player talent?  That's what we're talking about right?  Supply and demand is essentially what is at the heart of a fantasy draft.

Supply and Demand View of Value Based Drafting

If we're going to approach VBD from a supply/demand standpoint, let's just start on the demand side of the equation because it's the easiest.  The most common leagues today will usually be comprised of 12 teams that start the following positions: QB, RB, RB, WR, WR, Flex, TE.  The demand for each position is how many games we need out of each starter spot.  We need enough games out of each position to fill a starting slot for that position for every team, for every week of the NFL season.  We're going to assume that we basically need a starter for all 17 weeks of the NFL season.  This might seem odd because a lot of leagues actually wrap up in week 16, but since we're in part trying to address the impact that injuries have, and we have no way of predicting whether an injury might occur in week 3 or week 17, we need to assume that we need to fill every roster spot for every week of the season.

So the demand side of the equation in terms of the number of games we need is as follows:

Fantasy League Total Demand for Games by Position

That table is pretty much just 12 teams, times 17 weeks, times the number of starters at each position.  We now know what the demand side of the equation is based on that table.  We don't need to think about points yet, because points will take care of themselves after we set the baseline.  All we care about right now is figuring out how many games our league requires from each position, and then how many players it takes to satisfy that demand.

For the supply side of the equation, we need to know how many players at each position it will take to get to the required number of games for each position.  That means that we have to think about things like injuries and whether a player might be demoted or benched.  That might sound like we're creating a really difficult exercise in analysis, but it's actually not that difficult.

The easiest thing that we can do to get a sense as to the supply at each position is to look at past seasons and then see how many players at each position it took to get to the required number of games from the table above. 

For instance, in 2007 we needed to add all of the games for the top 14 QBs together in order to get to 204 games.  In 2008, the injury of Tom Brady in week one and the eventual benching of Vince Young meant that it took about the top 17 QBs to get to 204 games.  In 2009 it required 14 QBs, in 2010 it required 15 QBs, and in 2011 it required 16 QBs.  We can then use that information to set the baseline for QBs at QB15.  We know that on average the games required by our fantasy league will be about equal to the games played by the top 15 QBs.

I always think it's easier to think about stuff like this if you can visualize it, so let's do that.  The following graph shows the expectation for the number of games played for QB1-QB30 based on data for the last five years.

We can see that the games played expectation stays relatively high from QB1-QB11 or so.  Then it dips through about QB20, and then after that it falls off a cliff.  This is valuable information about how quarterbacks will actually fill the need for the 204 games that our league will require out of the position.

However complicated any of this sounds, it's not actually that complicated.  We're just taking the idea that different positions have different roster requirements and injury potential (or potential to get benched as is probably the case for QB20-QB30), and we're quantifying the issue.  You've no doubt thought to yourself "I need to draft a lot of running backs because they get hurt a lot."  All this exercise is doing is putting numbers to that idea. 

When we go through the same exercise that we just used on QB and apply it to WR, RB and Flex, we see that you have to go to about RB44 and WR44 to satisfy the game requirements of those positions.  Let's look at the graph that shows how many games we can expect to get out of each RB and WR based on ranking.

The following article is a guest post by Frank DuPont, author of Game Plan: A Radical Approach to Decision Making in the NFL. You can also follow Frank on Twitter.

Fantasy football is all about value.  Value is at the heart of every discussion we have.  It's the idea behind sleepers (they're undervalued).  It's also the idea behind busts (they're overvalued).  Today I want to talk about an idea, Value Based Drafting (VBD), that is sort of at the core of all of the fantasy football related value discussion.  You may not think about your decisions as being related to VBD, but they probably are.  When you decide to wait on QB because you think you can get a good one later in the draft, that's a decision that is essentially based in VBD.  I'm going to spill some pixels talking about the idea of VBD, first covering conventional wisdom on the topic, and then I'm going to suggest a tweak that will hopefully improve the way you think about your draft.

A Brief Primer on Value Based Drafting

The idea of VBD is credited to Joe Bryant.  The theory is that each player's actual value isn't how many points he scores; it's how many points he scores relative to the other players at his position.  But in order to calculate that player value, you need to establish a baseline for comparison purposes.  For instance, Rob Gronkowski and Calvin Johnson were reasonably close in statistics last year.  But because they played different positions, how can we compare their relative values?  We need to come up with a baseline for WR and TE to make the comparison.

But coming up with the baseline is the thing that has perplexed fantasy owners pretty much since VBD came into existence.  Do you compare players to the average at their position?  Do you compare them to the last starter at their position?  Changing your baseline is going to change the results of the analysis.

The idea that seems to have become the dominant one is that in setting the baseline you should look at how many players from each position will be drafted in the top 100 players of your fantasy draft and then use that number to set the baseline.  I'll refer to this idea as the Top 100 Method.  As of this writing, here are how many players are going at each position in the top 100 of ADP.

So by the Top 100 Method, Arian Foster is worth however many points you can expect him to score more than RB39, while Aaron Rodgers is worth however many points we can expect him to score over QB13.

Potential Problems with the Top 100 Baseline

However, I think it's worth discussing some issues that using the Top 100 Method for setting the baseline could create and why it might be worth looking for another solution.

First, our entire goal when we draft fantasy players is to look for inefficiencies and mispricing of players.  We want to look for areas where the broad fantasy market doesn't correctly value a player or a position.  It's the only way to have exceptional results in the long term.  However, if we create our VBD baselines using the top 100 approach, we're using the market's valuation of players to set our own valuation of a position.  If the broad market is wrong, then we're going to be wrong.  The Top 100 Method of setting the baseline begs the following question:  If we believe that the market is right as to how many of each position should be in the top 100, what makes us think we'll find any market inefficiencies in terms of actual players?  If the market is right, then we might as well just draft off of ADP and hope to get lucky.  The Top 100 Method of setting the baseline gives up an arbitrage opportunity right out of the gate.

There is another issue with the Top 100 Method.  How do we know that "top 100" properly reflects the supply and demand related to fantasy player talent?  That's what we're talking about right?  Supply and demand is essentially what is at the heart of a fantasy draft.

Supply and Demand View of Value Based Drafting

If we're going to approach VBD from a supply/demand standpoint, let's just start on the demand side of the equation because it's the easiest.  The most common leagues today will usually be comprised of 12 teams that start the following positions: QB, RB, RB, WR, WR, Flex, TE.  The demand for each position is how many games we need out of each starter spot.  We need enough games out of each position to fill a starting slot for that position for every team, for every week of the NFL season.  We're going to assume that we basically need a starter for all 17 weeks of the NFL season.  This might seem odd because a lot of leagues actually wrap up in week 16, but since we're in part trying to address the impact that injuries have, and we have no way of predicting whether an injury might occur in week 3 or week 17, we need to assume that we need to fill every roster spot for every week of the season.

So the demand side of the equation in terms of the number of games we need is as follows:

Fantasy League Total Demand for Games by Position

That table is pretty much just 12 teams, times 17 weeks, times the number of starters at each position.  We now know what the demand side of the equation is based on that table.  We don't need to think about points yet, because points will take care of themselves after we set the baseline.  All we care about right now is figuring out how many games our league requires from each position, and then how many players it takes to satisfy that demand.

For the supply side of the equation, we need to know how many players at each position it will take to get to the required number of games for each position.  That means that we have to think about things like injuries and whether a player might be demoted or benched.  That might sound like we're creating a really difficult exercise in analysis, but it's actually not that difficult.

The easiest thing that we can do to get a sense as to the supply at each position is to look at past seasons and then see how many players at each position it took to get to the required number of games from the table above. 

For instance, in 2007 we needed to add all of the games for the top 14 QBs together in order to get to 204 games.  In 2008, the injury of Tom Brady in week one and the eventual benching of Vince Young meant that it took about the top 17 QBs to get to 204 games.  In 2009 it required 14 QBs, in 2010 it required 15 QBs, and in 2011 it required 16 QBs.  We can then use that information to set the baseline for QBs at QB15.  We know that on average the games required by our fantasy league will be about equal to the games played by the top 15 QBs.

I always think it's easier to think about stuff like this if you can visualize it, so let's do that.  The following graph shows the expectation for the number of games played for QB1-QB30 based on data for the last five years.

We can see that the games played expectation stays relatively high from QB1-QB11 or so.  Then it dips through about QB20, and then after that it falls off a cliff.  This is valuable information about how quarterbacks will actually fill the need for the 204 games that our league will require out of the position.

However complicated any of this sounds, it's not actually that complicated.  We're just taking the idea that different positions have different roster requirements and injury potential (or potential to get benched as is probably the case for QB20-QB30), and we're quantifying the issue.  You've no doubt thought to yourself "I need to draft a lot of running backs because they get hurt a lot."  All this exercise is doing is putting numbers to that idea. 

When we go through the same exercise that we just used on QB and apply it to WR, RB and Flex, we see that you have to go to about RB44 and WR44 to satisfy the game requirements of those positions.  Let's look at the graph that shows how many games we can expect to get out of each RB and WR based on ranking.

Here's a graph that shows the games played expectation for RBs.

You can see that from RB1 through RB35 or so, the trend is gradually downward.  Then it falls off a cliff.  But it actually takes all of the running back games from RB1 through RB35 to satisfy our league's need for 408 running back games.  Then RB36 through RB44 are used to satisfy the Flex requirement.  In total that means that RB1-RB44 are equal to the demand for RBs.

Then here is the graph for WRs:

You can see that when it comes to WRs, the games played expectation is actually relatively flat until about WR20, at which point it starts to head south.  To satisfy the 408 games needed for just the WR position we end up using WR1-WR29.  The Flex position accounts for WR30-WR44.  It just happens to be coincidence that in order to account for all of the needed RB games and all of the needed WR games (including Flex) it takes about 44 players from each position.  The actual makeup of the Flex position is determined by which position yields more points in that range.  Basically, by that point WR is just a little more reliable for points that RB and that's why more WRs go into the Flex spot.  We'll discuss this issue a little further in Part II of this series because it's fairly important to your draft.

When you look at TE it ends up being about 15 players to satisfy the game requirements of our league.  Here is the graph that shows the games played expectation for tight ends.  This one is actually extremely interesting.  Because some of the top tight ends like Dallas Clark and Antonio Gates have missed significant parts of seasons, the games played expectation actually goes up as you go from TE1 to TE15.

The following table summarizes the results of the exercise that we've just done for each position.

Again, all this table is doing is quantifying how roster sizes, injuries, and the potential that a player might be benched, come together to create requirements for each position.  We've looked at how many games we might expect to get out of each player in order to do this.

We're then going to take that information and use it to set our baseline in VBD.  So instead of using TE9 as the baseline for TE (as the Top 100 Method of VBD would propose) we're going to use TE15 as the baseline.  That is going to increase the relative value of the top tight ends like Rob Gronkowski and Jimmy Graham because their basis of comparison is now a lower ranked TE.  Also, instead of using 38 or 39 spots for running back and receiver, we're going to use RB44 and WR44 as our baseline.  That is going to increase the value of the top RB and WRs.  The biggest difference between our supply/demand method of setting a baseline and the Top 100 Method of setting a baseline is in QB.  That position's baseline moved the least compared to the Top 100 Method, which means that QB might be undervalued by the Top 100 Method.

You could actually take the results of this research and plug in any projections that you might have.  You would just look at every player and compare their projected points to the baselines that I calculated above.  I'm actually going to do some of that in Part II of this series where I'll look at what these new VBD baselines will mean for this year's fantasy draft and which players might be undervalued by the broad market.