Optimal Draft Strategy (1 Viewer)

[Ryan99]

Hey all, I wanted to post some results from a drafting simulation I've been running. The next couple paragraphs will be explanation. You can find the results at the end, followed by some additional comments.

I originally started this project with the hopes of checking various draft strategies from different draft positions. To do this, I needed a way of comparing rosters, which means I needed a way of assigning fantasy points to draft positions. I wrote previously about one of my attempts at this. To make a long story short, it involved doing a linear regression on average fantasy points as a function of ADP (by position) over the past 5 years. Basically, I got an equation (for each of QB, RB, TE and WR) that gives an expected number of fantasy points based on where each player is drafted (by position: 1st RB, 5th WR, that sort of thing). I wanted an additional way of estimating player output, so I also compiled player projections from some different sources and took an average. I did the simulations twice, using both of these point assignment methods.

For the draft simulations, I used a 'default' drafting method by looking at ADP. Basically, every owner would draft the highest remaining player (by ADP) assuming their team still needed that position. Then for each of the draft positions (I used 12 owners), I did a draft simulation using every possible draft strategy and recorded the one that produced the highest fantasy points (by average projection and also my linear regression ADP method) for that draft slot. So what this gives you is the optimal draft strategy by draft position, assuming players score according to their average projection (or my linear ADP method) and all the other owners draft by ADP. Here are the results:

Fantasy Points details: 1 QB, 2 RB, 2 WR, 1 TE, 1 RB/TE/WR flex

6 points per passing TD, -0.5 points per sack, standard scoring otherwise

12 owners, just starters (so 7 rounds)

Key: 0 QB, 1 RB, 2 TE, 3 WR

Using Average Projected Stats

Slot: Picks

1: 1112303

2: 1112303

3: 1113203

4: 1113203

5: 3112301

6: 3112301

7: 1012313

8: 0132311

9: 0112133

10:1032113

11:1123013

12:1123013

Using Historic Points by ADP

1: 1331102

2: 1331201

3: 1313120

4: 1311203

5: 1313021

6: 1133021

7: 1313021

8: 1133201

9: 1113023

10:1133201

11:1133021

12:1133021

Conclusions: I'm certainly not suggesting that if you're drafting 1st overall you go 1112303 (RB, RB, RB, TE, WR, QB, WR), although I don't think that's a terrible draft order. But I do think it's instructive to look at the overall trends. The biggest one that stands out to me is the RBs. Almost every slot has 2 RBs by the 3rd round. This indicates to me that ADP undervalues RB according to both experts' projections and historic point totals. The second is that for most draft slots, the best time to draft your starting QB is in the 5th or 6th round. Accordingly, this indicates that the tier 2 QBs like Stafford, Newton and Vick are being drafted too soon (and this is with 6 points per passing TD). Also, if you buy the experts projections, this actually indicates that it makes sense to take Megatron with the 5th or 6th pick, over Brees or Brady or a second tier RB.

Anyway, there you go. Some questions: did my explanation of the methodology make sense, and do you think this type of analysis is useful? Any improvements / corrections? For people who have done a lot of mock drafts, do these results make sense as good strategies? Other comments are obviously welcome.

 

[silentcoach]

interesting thread. Any way to come up with the optimal strategy adding a wr AND eliminating the wr/te/rb flex?

Thanks.

 

[MASSHOLE]

+1

 

[ShaHBucks]

I couldn't understand what you were hinting last time but this was interesting. When you did the linear regression for each position when did you see the biggest drop-off for each position? And anything on defense?

 

[Ryan99]

I couldn't understand what you were hinting last time but this was interesting. When you did the linear regression for each position when did you see the biggest drop-off for each position? And anything on defense?

I haven't done anything with defenses. There seems to be two main schools of thought on team D, and I'm in the one that thinks its a total crapshoot and you shouldn't take one until the end of the draft. Also, I don't play with IDP, so its never interested me.There are a few dropoffs that stand out when you look at average fantasy points by ADP. For QB, it happens after the 4th QB by ADP, which this year is the difference between having a Newton/Stafford or a Vick/Romo/Manning. In RB, around 20 and around 40 there are dropoffs; I like to make sure I get 2 RBs in the top 20 and 4 total in the top 40. The top 2 TEs tend to do much better than all the others, which everyone thinks will happen again this year, so no big revelations there. For WR, after #5 there's a significant dropoff and again around #35. This year that #5 is the difference between Greg Jennings/Andre Johnson and Wes Welker/Brandon Marshall.Interestingly, the 16th WR by ADP has been the best slot in the last 5 years. Those 16s would be Wes Welker (2011), Michael Crabtree (2010), Vincent Jackson (2009), Anquan Boldin (2008) and Randy Moss of course (2007). But the 16th is an anomaly; the 15th and the 17th are nothing special.As I wrote in a previous post the best slot for QBs has been #4. They've been Rivers (2011), Brady (2010), Rodgers (2009), Brees (2008), and Brady (2007). It seems like people draft the first 3 QBs as 'safe' picks, and the #4 is the breakout guy. This year Newton and Stafford have been trading #4.

 

[chiefs2929]

That is pretty cool. Good information.

 

[buck naked]

Excellent discussion thread. Thanks.

 

[Ryan99]

interesting thread. Any way to come up with the optimal strategy adding a wr AND eliminating the wr/te/rb flex?Thanks.

Unfortunately due to lack of both foresight and programming experience I hard coded the positional requirements. Sorry.

 

[silentcoach]

Sorry to hear. It seems that maybe the possibility of a rb flex may show why ADPs have undervalued rbs?

 

[RavenLunatic]

Nice read. Thanks!

 

[LargeMouthBass]

Hmmmmmmm...

 

[-jb-]

Dot

 

[Deamon]

Very interesting indeed. What do you do for a living? haha.

Anyways I actually think going RB RB RB is a good strategy this year. Not even touching a WR for the first few rounds (obviously that changes if there's great value when I'm about to pick), but I'm not in Calvin range anywhere, so picking at #3 in one league, I think the RB RB RB the formula suggests is what I was planning on anyways.

 

[ShaHBucks]

Since the elite QB(top 4) and TE(top 2) position are so scarce why wouldn't you advocate going QB-TE to start a draft? I think this strategy is becoming more mainstream. The more I read about it, the more I'm buying it because I totally agree with the theory of VBD now. Based on consistency from year-to-year, QB and TE seem to retain most of their value. Based on the historic ADP draft I still have a shot at a top 20 rb in round 3, though that would be likely vs people.

What stuck out to me the most was what you think of the second tier QB's and the RB position in general. Have you tried to simulate any other positional draft strategies?

 

[Ryan99]

Sorry to hear. It seems that maybe the possibility of a rb flex may show why ADPs have undervalued rbs?

Yeah I think you're right. All the optimal draft strategies used a flex RB in this sim. This might be partly because the ADP is an average over lots of different draft types; some might have no flex, some might have 3 WR (making WR more valuable), etc.Does anyone know if rules-specific ADP exists somewhere?

 

[Ryan99]

Very interesting indeed. What do you do for a living? haha.

Went to school for physics but I'm trying to get a job as an actuary.

Since the elite QB(top 4) and TE(top 2) position are so scarce why wouldn't you advocate going QB-TE to start a draft? I think this strategy is becoming more mainstream. The more I read about it, the more I'm buying it because I totally agree with the theory of VBD now. Based on consistency from year-to-year, QB and TE seem to retain most of their value. Based on the historic ADP draft I still have a shot at a top 20 rb in round 3, though that would be likely vs people.

A problem I see with QB-TE is the danger that your guys won't be there when your pick comes around. If you're drafting late in a 12 team league, there's a real chance Brady and Brees are both gone, and at that point it seems crazy to me to go QB. If you're drafting in the middle and get Brees/Brady you could miss out on Gronk/Graham.Let's say your guys are there and you're drafting 5th (the first spot with a real decision in my opinion) in a 12 team league with 6 points per passing TD. Your picks are 5,20,29,44,53,68,77. QB,TE,RB,RB,WR,WR,RB might net you Brady, Gronkowski, Darren Sproles, Reggie Bush, Miles Austin, Stevie Johnson, Donald Brown. Since you went QB-TE, you were drafting RB when the good WRs came off the board, so your RBs and WRs are not the greatest. Going RB,RB,WR,WR,QB,TE,RB gets you Chris Johnson, Steven Jackson, Hakeem Nicks, Marques Colston, Tony Romo, Vernon Davis, Donald Brown. Pretty solid at all positions. I'd rather have this second team, but some might prefer the first. I don't think either is a bad teamI think there's a lot of value in 3rd tier QBs (Romo, Matt Ryan, the Mannings) and these guys can be had in the 5th or 6th round. I also like the 3rd tier WRs like Brandon Marshall, AJ Green, Roddy White, Victor Cruz, etc. in the 3rd or 4th round. And you can get a very solid TE in round 6 (Witten if he's there, Finley/Vernon Davis otherwise). To take advantage of these you need at least 1 RB in the first 2 rounds.Despite all of this however, the thing that I most dislike about QB-TE is that is completely handcuffs you for the rest of the draft. I always try to keep myself in a position where I can take the best player available on the board. If you fill up QB and TE right away, it totally kills your flexibility. And that's what good drafting is all about in my opinion; you need to be able to take advantage of other owners' mistakes. To do that you need to keep your options open.

What stuck out to me the most was what you think of the second tier QB's and the RB position in general. Have you tried to simulate any other positional draft strategies?

I'm not sure what you mean.

 

[VaTerp]

Good stuff.

I have to say that I'm still trying to understand the math but I like the concepts.

In the mocks I've done recently from the 3 spot I think I've liked my teams the best going RB-RB-RB, WR, TE, QB. But that's with getting Ryan as my QB, who lately seems to be EVERybody's wait and take a QB choice. And Gates or Witten at TE. Or taking a WR in the 4th and 5th if no TE I like.

Either way, I really like the options if I go RB heavy early this year.

I feel like there is the most depth and least variance at the WR spot this year and am comfortable waiting on one. Which, it seems to me that your data suggest. Then again, I really don't understand the math right now.

 

[moleculo]

excellent thread. Love this type of stuff. Thanks!

 

[We Tigers]

Seriously cool analysis. The part I find most interesting is the divergence between experts' projections and historic points as relates to QBs. By projections, slots 7-10 should pick up a QB in round 1 or 2, but by historic points no team should grab a QB till round 5. Obviously the numbers aren't perfect, but it does suggest to me a sizable discrepancy between perception of elite QB success and their actual performance.

I think this is particularly notable after last year's QB scoring explosion. Would love to see this analysis refried for 2013 after we see if 2011's astronomical numbers hold up this season.

 

[moleculo]

one concern I have with this is the linear regression for points vs ADP. I don't think this should be a linear function (Y=mX+b), rather it should be some sort of decay function - exponential, logarithmic, or something like that.

 

[Ryan99]

Seriously cool analysis. The part I find most interesting is the divergence between experts' projections and historic points as relates to QBs. By projections, slots 7-10 should pick up a QB in round 1 or 2, but by historic points no team should grab a QB till round 5. Obviously the numbers aren't perfect, but it does suggest to me a sizable discrepancy between perception of elite QB success and their actual performance.I think this is particularly notable after last year's QB scoring explosion. Would love to see this analysis refried for 2013 after we see if 2011's astronomical numbers hold up this season.

Yeah. Keep in mind, these are optimal draft strategies assuming every other owner drafts by ADP, so it doesn't mean that NO QB should be drafted before round 5, it just means that the ones already going prior to round 5 shouldn't be drafted even earlier. If Rodgers is still there in the second round, take him! It is interesting that QBs show up in the projections but not in the historic data. That means that ADP this year is drafting QBs higher than they historically should have been, but the experts are projecting that the top QBs should go even higher than ADP. There seems to be a major push for early QBs this year, and I'm just not convinced that the data supports it.

 

[moleculo]

Seriously cool analysis. The part I find most interesting is the divergence between experts' projections and historic points as relates to QBs. By projections, slots 7-10 should pick up a QB in round 1 or 2, but by historic points no team should grab a QB till round 5. Obviously the numbers aren't perfect, but it does suggest to me a sizable discrepancy between perception of elite QB success and their actual performance.I think this is particularly notable after last year's QB scoring explosion. Would love to see this analysis refried for 2013 after we see if 2011's astronomical numbers hold up this season.

Yeah. Keep in mind, these are optimal draft strategies assuming every other owner drafts by ADP, so it doesn't mean that NO QB should be drafted before round 5, it just means that the ones already going prior to round 5 shouldn't be drafted even earlier. If Rodgers is still there in the second round, take him! It is interesting that QBs show up in the projections but not in the historic data. That means that ADP this year is drafting QBs higher than they historically should have been, but the experts are projecting that the top QBs should go even higher than ADP. There seems to be a major push for early QBs this year, and I'm just not convinced that the data supports it.

as I understand what you have here, ADP is a backwards looking estimate, whereas experts is a forwards looking analysis. If the experts are forcasting an increased passing relevance based on rule changes and an ever increasing trend towards the passing game, it makes sense to rank QB's higher.

 

[-jb-]

one concern I have with this is the linear regression for points vs ADP. I don't think this should be a linear function (Y=mX+b), rather it should be some sort of decay function - exponential, logarithmic, or something like that.

:headexplode:

 

[Ryan99]

as I understand what you have here, ADP is a backwards looking estimate, whereas experts is a forwards looking analysis. If the experts are forcasting an increased passing relevance based on rule changes and an ever increasing trend towards the passing game, it makes sense to rank QB's higher.

Sure, but its an issue of how much higher. If you look at the top 4 QBs over the past 5 years, fantasy points are fairly steady from 2007 to 2010 and then explode in 2011. Expert projections have the top 4 guys coming down from that high, but still much higher than in 2007-2010. Was 2011 an aberration, or did the league fundamentally change last year? I've no doubt there's an upward trend, but I think the trend will continue closer to its historic rate of increase than people seem to think. This difference is much larger when looking at how the top 4 QBs by ADP have done historically (I've excluded Brady's 2008). There's definitely a trend upward, but the 2012 projections make a huge leap above and beyond this.Something I've been thinking about recently is how people use projections. Since the projections assume everyone stays healthy or injury risk is spread out evenly, by assuming all RBs play 14 games and all QBs play 15 games for instance (other than for super high risk players like Vick), the projections don't match historic averages, even after accounting for trends. This is fine and it doesn't mean that the projections are wrong. What I've found though is that while the projections indicate that it makes sense to take a top QB in the first or second round, historic points-by-ADP indicates that it doesn't. Since looking at historic ADP accounts for injury, and RBs are more likely to get injured than QBs, you'd think QBs would be valued higher, but they aren't.I don't yet fully understand what's going on, but I think it involves at least these two factors.1) You're taking a risk by drafting a high injury (just because of his position) RB early, but you're taking a much larger risk by not drafting one because of the massive VAR for top RBs. By contrast, QBs are more consistent but you're not giving up as much by passing on one in the first few rounds.2) Since most leagues only start one QB, you're paying a big opportunity cost by filling that position right away. You've lost the ability to gain later by drafting a good QB that unexpectedly falls into a round you didn't expect him to be available in.

 

[Ryan99]

one concern I have with this is the linear regression for points vs ADP. I don't think this should be a linear function (Y=mX+b), rather it should be some sort of decay function - exponential, logarithmic, or something like that.

A linear function is undoubtedly sub-optimal, but the data are all over the place. I just needed something to force the data to be monotonically decreasing, and a straight line fit no worse than anything else (although I was doing this in open office and the options for fits are limited). This may be able to be improved by having more than just 5 years worth of data, although the more years you have, the more significant trend effects are going to be, so you'd have to preprocess the data to get it in line with current averages. There's also other cleaning that could be done, like trimming outliers and such.Do you have good data on fantasy points by ADP that indicate that it should be a decaying exponential, n-th order polynomial, or whatever?

 

[-jb-]

as I understand what you have here, ADP is a backwards looking estimate, whereas experts is a forwards looking analysis. If the experts are forcasting an increased passing relevance based on rule changes and an ever increasing trend towards the passing game, it makes sense to rank QB's higher.

Sure, but its an issue of how much higher. If you look at the top 4 QBs over the past 5 years, fantasy points are fairly steady from 2007 to 2010 and then explode in 2011. Expert projections have the top 4 guys coming down from that high, but still much higher than in 2007-2010. Was 2011 an aberration, or did the league fundamentally change last year? I've no doubt there's an upward trend, but I think the trend will continue closer to its historic rate of increase than people seem to think. This difference is much larger when looking at how the top 4 QBs by ADP have done historically (I've excluded Brady's 2008). There's definitely a trend upward, but the 2012 projections make a huge leap above and beyond this.Something I've been thinking about recently is how people use projections. Since the projections assume everyone stays healthy or injury risk is spread out evenly, by assuming all RBs play 14 games and all QBs play 15 games for instance (other than for super high risk players like Vick), the projections don't match historic averages, even after accounting for trends. This is fine and it doesn't mean that the projections are wrong. What I've found though is that while the projections indicate that it makes sense to take a top QB in the first or second round, historic points-by-ADP indicates that it doesn't. Since looking at historic ADP accounts for injury, and RBs are more likely to get injured than QBs, you'd think QBs would be valued higher, but they aren't.I don't yet fully understand what's going on, but I think it involves at least these two factors.1) You're taking a risk by drafting a high injury (just because of his position) RB early, but you're taking a much larger risk by not drafting one because of the massive VAR for top RBs. By contrast, QBs are more consistent but you're not giving up as much by passing on one in the first few rounds.2) Since most leagues only start one QB, you're paying a big opportunity cost by filling that position right away. You've lost the ability to gain later by drafting a good QB that unexpectedly falls into a round you didn't expect him to be available in.

This is pretty much exactly my perspective on things this year.

 

[Captain Hook]

First off - Very Good effort, thanks for the work and the presentation. One very important delineation is the league setup and I don't see 1,2,2,1,Flex as being the norm for most leagues and certainly not for the FBGPC or higher stakes leagues where the norm is 1,2,3,1,Flex or 1,2,2,1,2Flex)And the other very important factor is PPR or not (assuming since not mentioned your trials are not)

Fantasy Points details: 1 QB, 2 RB, 2 WR, 1 TE, 1 RB/TE/WR flex6 points per passing TD, -0.5 points per sack, standard scoring otherwise12 owners, just starters (so 7 rounds)

 

[moleculo]

one concern I have with this is the linear regression for points vs ADP. I don't think this should be a linear function (Y=mX+b), rather it should be some sort of decay function - exponential, logarithmic, or something like that.

A linear function is undoubtedly sub-optimal, but the data are all over the place. I just needed something to force the data to be monotonically decreasing, and a straight line fit no worse than anything else (although I was doing this in open office and the options for fits are limited). This may be able to be improved by having more than just 5 years worth of data, although the more years you have, the more significant trend effects are going to be, so you'd have to preprocess the data to get it in line with current averages. There's also other cleaning that could be done, like trimming outliers and such.Do you have good data on fantasy points by ADP that indicate that it should be a decaying exponential, n-th order polynomial, or whatever?

for starters, plot your projected points vs positional rank at any position. You should see a decaying function. Looking at my current projections, QB and TE both seem to fit best with a logarythmic function, and RB and WR both seem to be exponential, although WR could be linear. What this all means, essentially, is that there is a premium for the top picks, which should not be a surprise. Think about it - the difference between RB1 and RB5 should be a lot more than RB35 and RB40, whereas a linear approximation says the difference should be the same.Another argument point: find a pick value calculator, and plot pick value vs pick. you should see a decaying trend line. The difference between picks 1 and 2 should be much more than the difference between picks 20 and 21, for example. Per the NFL trade value chart, the difference between 1 and 2 is 400 points. The difference between 20 and 20 is only 50.

 

[moleculo]

as I understand what you have here, ADP is a backwards looking estimate, whereas experts is a forwards looking analysis. If the experts are forcasting an increased passing relevance based on rule changes and an ever increasing trend towards the passing game, it makes sense to rank QB's higher.

Sure, but its an issue of how much higher. If you look at the top 4 QBs over the past 5 years, fantasy points are fairly steady from 2007 to 2010 and then explode in 2011. Expert projections have the top 4 guys coming down from that high, but still much higher than in 2007-2010. Was 2011 an aberration, or did the league fundamentally change last year? I've no doubt there's an upward trend, but I think the trend will continue closer to its historic rate of increase than people seem to think. This difference is much larger when looking at how the top 4 QBs by ADP have done historically (I've excluded Brady's 2008). There's definitely a trend upward, but the 2012 projections make a huge leap above and beyond this.Something I've been thinking about recently is how people use projections. Since the projections assume everyone stays healthy or injury risk is spread out evenly, by assuming all RBs play 14 games and all QBs play 15 games for instance (other than for super high risk players like Vick), the projections don't match historic averages, even after accounting for trends. This is fine and it doesn't mean that the projections are wrong. What I've found though is that while the projections indicate that it makes sense to take a top QB in the first or second round, historic points-by-ADP indicates that it doesn't. Since looking at historic ADP accounts for injury, and RBs are more likely to get injured than QBs, you'd think QBs would be valued higher, but they aren't.I don't yet fully understand what's going on, but I think it involves at least these two factors.1) You're taking a risk by drafting a high injury (just because of his position) RB early, but you're taking a much larger risk by not drafting one because of the massive VAR for top RBs. By contrast, QBs are more consistent but you're not giving up as much by passing on one in the first few rounds.2) Since most leagues only start one QB, you're paying a big opportunity cost by filling that position right away. You've lost the ability to gain later by drafting a good QB that unexpectedly falls into a round you didn't expect him to be available in.

I like where your head is at. Again, excellent thread.

 

[Dinsy Ejotuz]

Which of the two methods produced the most points for a team? Was there a difference? IOW... are the projections better than ADP data?

 

[Ryan99]

for starters, plot your projected points vs positional rank at any position. You should see a decaying function. Looking at my current projections, QB and TE both seem to fit best with a logarythmic function, and RB and WR both seem to be exponential, although WR could be linear. What this all means, essentially, is that there is a premium for the top picks, which should not be a surprise. Think about it - the difference between RB1 and RB5 should be a lot more than RB35 and RB40, whereas a linear approximation says the difference should be the same.Another argument point: find a pick value calculator, and plot pick value vs pick. you should see a decaying trend line. The difference between picks 1 and 2 should be much more than the difference between picks 20 and 21, for example. Per the NFL trade value chart, the difference between 1 and 2 is 400 points. The difference between 20 and 20 is only 50.

It makes sense that projections would have a nice, smooth trend line, but the ADP data really is all over the place. Linear, exponential, power, and log all has very similar goodness-of-fit scores, so I went with the simplest one. One complicating factor is that the trend line is points versus positional ADP, rather than absolute ADP. So for RB for instance, the drop off per spot was calculated to be .2 points per game. Since the first few RBs are typically drafted consecutively, this means that the difference between 1 and 2 is about .2 points per game. But when you get beyond the first round, there's going to be a larger average draft difference between consecutive RBs, so that -.2 PPG might represent 2, 3 or 4 (or more) draft spots. So there is a kind of built in decrease in difference per draft spot.

Which of the two methods produced the most points for a team? Was there a difference? IOW... are the projections better than ADP data?

I think in general the projections produce higher overall points, because the projections don't account for injury, but of course the historic data does. As to which is better, that's a question I don't have the data right now to answer (also its a really good question). You'd need to compare historic projections against even more historic game data and see which one did a better job of predicting player performance. This certainly sounds like something people would do ... if you know of any articles or posts on this please send them along.

 

[Dinsy Ejotuz]

I think in general the projections produce higher overall points, because the projections don't account for injury, but of course the historic data does. As to which is better, that's a question I don't have the data right now to answer (also its a really good question). You'd need to compare historic projections against even more historic game data and see which one did a better job of predicting player performance. This certainly sounds like something people would do ... if you know of any articles or posts on this please send them along.

Sorry... I meant if you conducted one draft each year using the optimal draft strategy suggested by the projections data, and then ran another using the optimal draft strategy suggested by the historic ADP data which method actually worked better for each of the drafts covered by your data? IOW... have you run your optimal strategy versus what actually happened each year?

 

[Ryan99]

I think in general the projections produce higher overall points, because the projections don't account for injury, but of course the historic data does. As to which is better, that's a question I don't have the data right now to answer (also its a really good question). You'd need to compare historic projections against even more historic game data and see which one did a better job of predicting player performance. This certainly sounds like something people would do ... if you know of any articles or posts on this please send them along.

Sorry... I meant if you conducted one draft each year using the optimal draft strategy suggested by the projections data, and then ran another using the optimal draft strategy suggested by the historic ADP data which method actually worked better for each of the drafts covered by your data? IOW... have you run your optimal strategy versus what actually happened each year?

I haven't. These strategies have been optimized using this year's projections, so for a previous year I'd need experts' historic projections from that year, which I haven't assembled (are these generally available?).

 

[Imagine]

I think in general the projections produce higher overall points, because the projections don't account for injury, but of course the historic data does. As to which is better, that's a question I don't have the data right now to answer (also its a really good question). You'd need to compare historic projections against even more historic game data and see which one did a better job of predicting player performance. This certainly sounds like something people would do ... if you know of any articles or posts on this please send them along.

Sorry... I meant if you conducted one draft each year using the optimal draft strategy suggested by the projections data, and then ran another using the optimal draft strategy suggested by the historic ADP data which method actually worked better for each of the drafts covered by your data? IOW... have you run your optimal strategy versus what actually happened each year?

I haven't. These strategies have been optimized using this year's projections, so for a previous year I'd need experts' historic projections from that year, which I haven't assembled (are these generally available?).

Hey Ryan99! I like this, even though I don’t think I quite understand. Can you help me?For “Using Average Projected Stats”: What is this? Could you give me an example? I’m assuming it is if you look “back” on the last five year’s results for example, that you’d take all the players each year, and rank them according to how many points they scored each year. The top scorer would go to owner #1, the 12th to owner #12. So this method looks back and indicates for example that the 5th best scorer over those five years was a WR (3 in the key). So “3” goes to the 1.05 owner. Surprising to me, 1.06 is also a WR, so that’s why owner #6 also has a 3 assigned to his draft slot.But how did you decide on the "position" over those five years? For example, if in the last five years, the 5th best scores each year belonged to 2 WRs, 1QB, 1RB, and 1TE, who gets the “assignment”? The WR because there were two of them compared to only one of the others? 1. So that’s my first question…am I understanding this “Using Average Projected Stats”, and how are the assignments made?Then looking at your second category, “Using Historic Points by ADP”: I assume this is similar, but that you went back over the last five years, and listed the players each of those five years by their ADP when drafted. Then you determined how many points each of those players scored that year. Then you ranked each of those players by the points they scored, and determined on average that the top 12 were always RBs. That’s why each of the owners in the second category has a “1” for their first draft pick. So if I understand this correctly (and I must not be), what happened to the QBs, because they will always have the highest scoring each year? What in the calculation am I missing that bumps the RBs to the top slot?2. In summary, my second question is whether I have correctly stated what you do in the second analysis, and if so, how come the QBs don’t rank higher?I apologize for the long questions, but I love this stuff, and really appreciate the analysis you have done. I only want to understand it better, and if my questions help in some fashion, that’s a bonus!Thanks

 

[Ryan99]

Hey Ryan99! I like this, even though I don’t think I quite understand. Can you help me?For “Using Average Projected Stats”: What is this? Could you give me an example? I’m assuming it is if you look “back” on the last five year’s results for example, that you’d take all the players each year, and rank them according to how many points they scored each year.

Its much simpler than that. Its just the average of a bunch of experts' predictions that I've compiled.

Then looking at your second category, “Using Historic Points by ADP”: I assume this is similar, but that you went back over the last five years, and listed the players each of those five years by their ADP when drafted. Then you determined how many points each of those players scored that year. Then you ranked each of those players by the points they scored, and determined on average that the top 12 were always RBs. That’s why each of the owners in the second category has a “1” for their first draft pick.

The first part is right, the second part isn't. I looked at historic ADP and calculated average (median, actually) scoring for that ADP, and used that number in the simulation. It was positional ADP, because otherwise you'd be averaging points from different positions which doesn't make sense. I then used these point values in the simulations.A thing I think you're missing here is that the optimal draft strategies wouldn't make sense if every owner followed them. They are only 'optimal' assuming every other owner drafts just based on ADP (they draft the highest available player by ADP, assuming they need that position). So, let's say you have the 8th pick (experts: 0132311, Historic ADP: 1133201). By ADP, the players already gone are Foster, Rodgers, McCoy, Rice, Brady, Chris Johnson and Calvin Johnson. If you believe the experts' projections, you should draft QB (Brees probably). If you believe the historic ADP method, you should draft RB (McFadden according to ADP).Anytime you draft, your strategy is going to change in response to the other owners' strategies. There's no such thing as an overall optimal strategy, it can only be optimal against some other strategy (although you may be able to find a strategy that no other strategy can exploit, the so-called 'Nash equilibrium', look up 'game theory' if this sounds interesting). So to optimize, I had to assume a strategy for the other players, and drafting by ADP seemed like the most common sense thing to do. If every owner followed these 'optimal' strategies, they would in fact be highly sub-optimal, and you could draft Aaron Rodgers in the 4th round and laugh in everyone's face.Just to reiterate, I am NOT suggesting that if you have the 8th pick you should go 0132311 (or 1133201) and that it will be optimal. First, the other owners will not perfectly follow ADP. Second, all this was done assuming a particular point structure and set of positional requirements, which differ wildly between leagues. Third, you might do you own research and not agree with the experts or the positional ADP and be totally right. There's probably more problems I'm not thinking of. But I do think this type of analysis is useful as a jumping off point for a more detailed analysis or to look at general trends.All that being said, I think 0132311 and 1133201 are pretty solid draft orders from the 8th slot

 

[BassNBrew]

One major issue I have with your approach is that I don't see you taking into account the accuracy of the projections. While the 5th RB (end of year) historically might be a better point producer then the 3rd QB, the 5th projected RB might be a worst producer than the 3rd projected QB.

Working off the top of my head last years QB rankings, finishing position, and points.

Vick 1/11/292

Rodgers 2/2/489

Brees 3/1/490

Romo 4/7/355

Rivers 5/9/336

Last year the RBs held up well (end of season ranking)

Rice/McCoy/MJD/Foster/Lynch - Lynch was the only outlier.

 

[Johnny Rock]

Love this stuff. Appreciate the work.

 

[KillerKlown]

Wow..this is interesting.

I did my first fgpc draft last week and drafted from the 1 spot.

I just saw this post and was wondering what strategy I should have used.

It just so happens that this is it...or very close to it anyway..

Nobody really likes my team..but I think Ive got a really good shot.Of course,I could be WAY off base..like I said,this is my first team in this format..but Ive been playing this game for quite some time(almost 20 years)..and my gut tells me that I at least HAVE A SHOT..

do I have the best team?Probably not.

The worst?probably not.

Will I manage it to the best of my ability?definatly.

I went into the draft knowing that id go rb heavy early..Id have to hit on some wr late..I really didnt know how fast the TE would go,being that they get an extra half a point per catch(they actually went alot quicker than I even expected).

Anyway here's how it went for me:

1)Rice(I feel more comfortable with him over Foster,and Ill be chastised for this,Im sure)

2)Sproles(Love this guy in ppr,we all know he was top 5 last year)

3)sjax(this is a bit early for him,but I wanted him in this format,and he wasnt coming back to me)

OK..so,Ive committed myself to a strategy..I gotta hope that some good wrs come back to me 20 plus picks later

4)Colston(Missed Cam by 1 pick.this deviates from the OPs strategy a bit,but I love Colston this year and think he's top 10)

5)F.Davis(I really thought that Gates or Hernadez would be available here..GOD was I wrong.LOL.rookie mistake,but I knew I had to get a TE here,or be looking at Jared Cook and company when it came back)

Ill get another wr when it comes back

6)Hillis(I committed to a plan..Im going 4 rbs DAMMITT..but actually,there were no good wr left..I drafted for value,and really like Hillis this year)

7)Romo(He was the best QB available,and with so many teams needing them I figured I should get one now,Ive had Romo in the past and hes helped me win leagues)

sooo...Now I have my QB.RB.RB.FLEX.FLEX.WR1.TE..Ive got a HUGE whole at WR2..but theres guys I like that should make it back to me.

NOPE

8)Celek(I may have made a mistake here,but I like Davis' upside,but dont trust his attitude.I like Celek as a more reliable TE2 with less upside)

9)P.manning(I have been targeting Manning as my QB1 in almost every draft.I feel WAY more comfortable with him as a QB1B..but basically,I dont like any of the wr here

DAMN..The 10th round is coming up and I have 1 friggin wide receiver!!!I HAVE to take wrs with these next two picks

10)Collie(I LOVE his consistency and think hes massivly undervalued.i think Luck will lock on early.I left Hillman on the board here,which may have been a mistake)

11)Tampa Mike(Im hoping he can get into the endzone this year)

OK..Im not gonna go through the rest of the draft...I took rbs and wrs the rest of the way..here's my team:

Romo,Manning

Rice,Sproles,Sjaz,hillis,LMiller,Powell

Colston,mWilliams,Collie,Amedola,Hankerson,Burleson,Criner,JJones

FDavis,Celek

Crosby

Lions

This is the type of team you can expect using this stratagy..or a similar one..I think you can do well,hopefully things fall a little bit better for you,and you can do better at wr2..but all-in-all I think youll be happy.

I hope that ryan99 dosnt feel like Ive hijacked his thread at all..I just wanted to give a real-life example of this strategy.Hope this helps.

 

[koreansteve]

Very interesting stuff.

Does the draft strategy for each draft position change for bigger leagues (14 or 16 team leagues)? I have the 11th pick in a 14 team league drafting Sunday so I'm curious about your findings.

 

[joey]

Love this kind of stuff.

I'm drafting from the 12 in my only redraft league and the mocks I've done have me leaning RB/RB/WR/WR since I love the depth and quality at WR this year, as well as the QBs that fall to my 5th pick in mocks. And now that I've seen this thread, I realize my strategy is a stone cold lock to win it all for me

 

[Ryan99]

One major issue I have with your approach is that I don't see you taking into account the accuracy of the projections. While the 5th RB (end of year) historically might be a better point producer then the 3rd QB, the 5th projected RB might be a worst producer than the 3rd projected QB.

That's why I ran the sim with both the projections and the historic ADP data, so that I could compare the results. I would love to study projection accuracy in detail, but experts' historic projections seem to not generally be available. Do you know of a good source for them?

Does the draft strategy for each draft position change for bigger leagues (14 or 16 team leagues)? I have the 11th pick in a 14 team league drafting Sunday so I'm curious about your findings.

I would assume the details would change, but I expect the general trends would remain, and they're what's important. The biggest one that stands out to me is the RBs. Almost every slot has 2 RBs by the 3rd round. This indicates to me that ADP undervalues RB according to both experts' projections and historic point totals. In a 14 team league, I would guess that its even more important to get 2 RBs in the first 3 rounds (assuming you have similar positional requirements). The second is that for most draft slots, the best time to draft your starting QB is in the 5th or 6th round. Accordingly, this indicates that the tier 2 QBs like Stafford, Newton and Vick are being drafted too soon (and this is with 6 points per passing TD). In a larger league though you might not be able to wait until round 5 to pick up your QB, especially if you're near the end of the draft.

 

[moleculo]

Using Historic Points by ADP1: 13311022: 13312013: 13131204: 13112035: 13130216: 11330217: 13130218: 11332019: 111302310:113320111:113302112:1133021

so this indicates that, historically speaking, the first 17 picks should be RB's? I'm not sure I buy that.

 

[Ryan99]

so this indicates that, historically speaking, the first 17 picks should be RB's? I'm not sure I buy that.

No, those are the optimal strategies for each position assuming everyone else drafts based on ADP. Its more an indication of how the historic data compares to current ADP than anything else. In this case, current ADP undervalues RBs compared to the historic data.

 

[moleculo]

so this indicates that, historically speaking, the first 17 picks should be RB's? I'm not sure I buy that.

No, those are the optimal strategies for each position assuming everyone else drafts based on ADP. Its more an indication of how the historic data compares to current ADP than anything else. In this case, current ADP undervalues RBs compared to the historic data.

all right then, historically speaking, assuming the rest of your league closely resembles ADP, you should be taking a RB in the first, regardless of draft position then...correct? And the second pick should always be a second RB or a WR.

 

[Ryan99]

all right then, historically speaking, assuming the rest of your league closely resembles ADP, you should be taking a RB in the first, regardless of draft position then...correct? And the second pick should always be a second RB or a WR.

Yup, that's what the data says. Keep in mind this is for 1 QB, 2 RB, 2 WR, 1 TE, Flex with 6 point passing TDs. I like the historic ADP method for valuing picks, but there's a lot of work that could be done on it. More data, adjustment for positional trends over time, a better fit curve, elimination of outliers, etc.If I were to redo this analysis using a more common set of fantasy scoring rules and positional requirements, what would you suggest I use?

 

[moleculo]

all right then, historically speaking, assuming the rest of your league closely resembles ADP, you should be taking a RB in the first, regardless of draft position then...correct? And the second pick should always be a second RB or a WR.

Yup, that's what the data says. Keep in mind this is for 1 QB, 2 RB, 2 WR, 1 TE, Flex with 6 point passing TDs. I like the historic ADP method for valuing picks, but there's a lot of work that could be done on it. More data, adjustment for positional trends over time, a better fit curve, elimination of outliers, etc.If I were to redo this analysis using a more common set of fantasy scoring rules and positional requirements, what would you suggest I use?

I'm fine with that...frankly, I doubt it matters a whole lot. If anything, what might be interesting is to see how much it varies - some sort of sensitivity study, depending on how easy it is to run your simulation. Drop it down to 10 teams, what happens? Try PPR, what happens? try 4 points/passing TD, what happens? 2 points/reception for a TE, what happens?maybe some areas for further research:

[*]compare expert drafts to a completely retrospective draft - that is, given actual points scored and with full benefit of hindsight, how should an optimal draft gone? This is along the lines of what BassnBrew was talking about earlier.

[*]this assumes teams completely draft starters @ all 4 positions before back-ups. Any way to allow some teams to draft depth @ RB/WR before TE/QB? Could that be optimal?

 

[gheemony]

all right then, historically speaking, assuming the rest of your league closely resembles ADP, you should be taking a RB in the first, regardless of draft position then...correct? And the second pick should always be a second RB or a WR.

Yup, that's what the data says. Keep in mind this is for 1 QB, 2 RB, 2 WR, 1 TE, Flex with 6 point passing TDs. I like the historic ADP method for valuing picks, but there's a lot of work that could be done on it. More data, adjustment for positional trends over time, a better fit curve, elimination of outliers, etc.If I were to redo this analysis using a more common set of fantasy scoring rules and positional requirements, what would you suggest I use?

I think many leagues are shifting to less emphasis on RBs, so I would try 1 QB, 2 RB, 3 WR, 1 TE, No Flex. Not being able to start 3 RBs should (obviously) eliminate RB/RB/RB as optimal, but I think you'll still see RB/RB as the predominant strategy to follow.

 

[Ryan99]

I'm fine with that...frankly, I doubt it matters a whole lot. If anything, what might be interesting is to see how much it varies - some sort of sensitivity study, depending on how easy it is to run your simulation. Drop it down to 10 teams, what happens? Try PPR, what happens? try 4 points/passing TD, what happens? 2 points/reception for a TE, what happens?

maybe some areas for further research:

[*]compare expert drafts to a completely retrospective draft - that is, given actual points scored and with full benefit of hindsight, how should an optimal draft gone? This is along the lines of what BassnBrew was talking about earlier.

[*]this assumes teams completely draft starters @ all 4 positions before back-ups. Any way to allow some teams to draft depth @ RB/WR before TE/QB? Could that be optimal?

These are good ideas. One problem with drafting backups is that I don't know how to evaluate them. One thing I've been thinking about is trying to get an average proportion of games played by position for starters, and use that to scale the starters numbers. Say top-of-the-depth-chart RBs play in an average of 90% of their team's games, and your projected points for that player is X, you have a backup with projected points Y, and the best free agent RB you expect to score Z. The average point total will be .9*X + .1*.9*Y + .1*.1*Z (there's a .1*.1 chance that both RBs will be hurt and you'll have to get a free agent).This is somewhat off topic, but another thing I've been thinking about is pairs of players that have complementary schedules, meaning one plays against weak defenses on weeks when the other plays against strong defenses and vice versa. You wouldn't always start one or the other, you'd alternate based on schedule. This is a situation where the whole is greater than the sum of the parts. You could have two mediocre RBs that combine to equal the output of a great RB, assuming you pick which one to start properly. Valuing this would be really hard, you'd need strength of schedule information and you'd need week by week projections.

I think many leagues are shifting to less emphasis on RBs, so I would try 1 QB, 2 RB, 3 WR, 1 TE, No Flex. Not being able to start 3 RBs should (obviously) eliminate RB/RB/RB as optimal, but I think you'll still see RB/RB as the predominant strategy to follow.

Ok, thanks for the info.

 

[silentcoach]

I think many leagues are shifting to less emphasis on RBs, so I would try 1 QB, 2 RB, 3 WR, 1 TE, No Flex. Not being able to start 3 RBs should (obviously) eliminate RB/RB/RB as optimal, but I think you'll still see RB/RB as the predominant strategy to follow.

+1

 

[sn0mm1s]

One major issue I have with your approach is that I don't see you taking into account the accuracy of the projections. While the 5th RB (end of year) historically might be a better point producer then the 3rd QB, the 5th projected RB might be a worst producer than the 3rd projected QB.

That's why I ran the sim with both the projections and the historic ADP data, so that I could compare the results. I would love to study projection accuracy in detail, but experts' historic projections seem to not generally be available. Do you know of a good source for them?

Does the draft strategy for each draft position change for bigger leagues (14 or 16 team leagues)? I have the 11th pick in a 14 team league drafting Sunday so I'm curious about your findings.

I would assume the details would change, but I expect the general trends would remain, and they're what's important. The biggest one that stands out to me is the RBs. Almost every slot has 2 RBs by the 3rd round. This indicates to me that ADP undervalues RB according to both experts' projections and historic point totals. In a 14 team league, I would guess that its even more important to get 2 RBs in the first 3 rounds (assuming you have similar positional requirements). The second is that for most draft slots, the best time to draft your starting QB is in the 5th or 6th round. Accordingly, this indicates that the tier 2 QBs like Stafford, Newton and Vick are being drafted too soon (and this is with 6 points per passing TD). In a larger league though you might not be able to wait until round 5 to pick up your QB, especially if you're near the end of the draft.

I think I have projections from FootballOutsiders for 2006 and 2008-2011.