This article is Part III in a three-part series. Here are links to the first and second installments.

In those two articles, I examined the question: if you draft a particular combination of four running backs, how many points can you reasonably expect to get out of the position? And similarly for wide receivers, quarterbacks, and tight ends (except, of course, at quarterback and tight end, we draft two and start one). This quantifies how much you can expect to lose by waiting a few rounds to draft a particular position. But of course, what you lost by waiting on one position you ought to gain back at another position.

This is the article where we put it all together and weigh the costs and benefits of any particular drafting strategy. For instance, if you draft receivers in the first two rounds, you will almost certainly get a lot of points out of the WR position. But the problem, as you well know, is that you will suffer at the running back position. You also missed out on your chance to get the top-ranked quarterback, if you're into that kind of thing. Is the benefit (of the great WR group) worth the cost (of the comparatively weak RB group)?

With a healthy dose of computing power, we can examine all possible orders in which you might draft players at the different positions, compute the expected points from each order, and thereby determine an "optimal" draft strategy. Of course, what we come up with will only be optimal in theory and will probably differ from real-world optimality in many ways that we'll discuss at the end of the article. But let's get started with an example.

Let's say you're drafting in the #4 position in a 12-team league. And let's say that you choose to draft your players in this order: rb, wr, rb, qb, rb, te, wr, qb, wr, rb, te, wr. (I'm assuming you start two 1 QB, 2 RB, 2 WR, and 1 TE, and I'm assuming you'll draft 2 QB, 4 RB, 4 WR, and 2 TE in the first 12 rounds.) That's an order that feels about right, doesn't it?

OK, if you take a running backs in rounds 1, 3, 5, and 10, you will end up with something like RB4, RB18, RB25, and RB44. This is based roughly on an "average" of the last six years' worth of Average Draft Position data. That particular combination of running backs can be expected to get you about 468 points. Drafting receivers in rounds 2, 7, 9, and 12 would land you something like WR5, WR29, WR37, and WR48, which has an expectation of about 313 points. Your quarterbacks would be roughly QB9 and QB16, which is worth about 301 points, and your tight ends --- TE5 and TE11 --- can be expected to get you 99 points. Just to make this a little less abstract, this particular order might this year land you a team that looks something like this:

QB - Jake Delhomme, Aaron Brooks
RB - Clinton Portis, Jamal Lewis, DeShaun Foster, Mike Anderson
WR - Larry Fitzgerald, Eddie Kennison, Terry Glenn, Joe Jurevicius
TE - Alge Crumpler, Heath Miller
Expected points:  1181

Compare that to what you could expect if you did something really wacky, like te, te, qb, qb, wr, wr, wr, wr, rb, rb, rb, rb. Here is the kind of team you could expect.

QB - Eli Manning, Jake Delhomme
RB - Laurence Maroney, Mike Anderson, T.J. Duckett, Ryan Moats
WR - Andre Johnson, Michael Clayton, Eddie Kennison, Koren Robinson
TE - Antonio Gates, Jeremy Shockey
Expected points:  1014

Now all we have to do is check the other 207,898 possible orders in which you can draft your team and see which yields the highest point total. As it turns out, the theoretically optimal combination of players is this one: rb, wr, te, qb, rb, wr, rb, wr, qb, te, rb, wr. It is worth 1211 points.

There are, as we would expect, several combinations that are nearly as good:

rb,rb,te,qb,rb,wr,wr,wr,qb,te,rb,wr    1208
rb,wr,te,qb,rb,wr,qb,wr,te,rb,rb,wr    1207   
rb,wr,rb,te,rb,wr,qb,wr,qb,rb,te,wr    1206   
rb,rb,te,qb,rb,wr,wr,wr,qb,rb,te,wr    1205   
rb,wr,rb,te,rb,wr,qb,wr,qb,te,rb,wr    1205   
rb,wr,te,rb,rb,wr,qb,wr,qb,te,rb,wr    1205   
rb,wr,te,rb,rb,wr,qb,wr,qb,rb,te,wr    1205   

Let's take a look at the top combinations for each slot.

Slot      "Optimal" draft order             Expected points
===========================================================
  1      rb,wr,te,rb,qb,rb,wr,qb,wr,rb,te,wr     1265
  2      rb,wr,te,qb,rb,wr,rb,qb,wr,te,rb,wr     1235
  3      rb,wr,te,qb,rb,wr,rb,wr,qb,te,rb,wr     1225
  4      rb,wr,te,qb,rb,wr,rb,wr,qb,te,rb,wr     1211
  5      rb,wr,te,rb,rb,qb,wr,wr,qb,te,rb,wr     1221
  6      rb,wr,te,rb,rb,qb,wr,wr,qb,te,rb,wr     1209
  7      rb,wr,te,rb,rb,qb,wr,wr,qb,te,rb,wr     1182
  8      rb,wr,rb,qb,te,te,wr,wr,qb,rb,rb,wr     1188
  9      rb,wr,rb,qb,rb,te,wr,wr,qb,te,rb,wr     1210
 10      wr,rb,rb,qb,rb,te,wr,wr,qb,te,rb,wr     1221
 11      wr,rb,qb,rb,rb,te,wr,wr,qb,te,wr,rb     1216
 12      wr,rb,rb,rb,te,qb,wr,wr,te,rb,wr,qb     1204

Warnings

• All of the above is based on a 12-team league with starting lineup requirements of 1QB / 2RB / 2WR / 1TE. It is also based on roster requirements of 2/4/4/2. It also assumes that the player pool is "typical." That is, the drop from QB1 to QB2, for example, is assumed to be about as big as it has been (on average) during the past six years. The very real tiers that exist in actual fantasy drafts are not present here.

• The optimal order from each slot assumes a "typical" draft in the rest of the slots. In other words, if you're picking in slot #1, the chart above recommends a tight end in the 3rd round. This recommendation is based on the assumption that TE1 is available. If three or four guys go beserk and draft tight ends in the second round, then that obviously changes things.

• More generally, the model is not very flexible. If you are on the clock at pick #43, then the model assumes that the players available to you are QB7, RB21, WR14, and TE2. That's it. No exceptions. The chart above says that the recommended pick is RB, but if WR9 is still on the board, that recommendation might be dead wrong, even under the rest of the assumptions in the article.

That's a lot of warnings. Is this worth anything?

It is interesting that the optimal strategy in every slot calls for a running back and a wide receiver (in some order) in the first two rounds. The fact that the optimal strategies call for RB in the first nine picks indicates that the drafting strategy that has evolved over the years in most leagues is indeed justified. That's not terribly surprising. But the fact that WRs, and not RBs, dominate the second round strategies is an indication that, generally, receivers are slightly undervalued in a lot of drafts. The strategies above suggest that you should start thinking WR at around 1.10. This means that building your team around WR1 or WR2 is generally more effective than building it around RB10 or RB11. The fact that WRs are prevalent in the middle and the end of the second round indicates that WRs 4 through 6 (roughly) are more valuable than RBs 13 through 17.

It's also interesting to see the tight end position recommended in the third round. This is a consequence of the fact that, over the past six years, the tight end position has been among the easiest to forecast, at least at the top. Compared to other positions, top-ranked TEs just bust less. Maybe that's worth an article on its own.

Another interesting but off-topic observation is that there does appear to be a real advantage to picking at the top of a serpentine draft. Every year, many people run mocks, look at the projected numbers, and claim that the #1 spot has an unfair advantage. Many other people dispute this, noting that top picks bust and/or lose their seasons to injury all the time, that things don't ever play around as we expect them to, and so on. This analysis takes into account the bust and injury rates of players available at various points in the draft, and still it concludes that the #1 hole has a nontrivial advantage. Even more interesting is the dip in the middle of the snake. I would have expected the worst slots to be 10, 11, and 12, but they turned out to be 6 through 9.

But let me be clear: do not print this out and take it to your draft. I do think that the ideas of the previous two articles (here and here) are of some use in playing around with different scenarios, but this one was a fun (for me) theoretical exercise and little more. We may or may not have learned some peripheral lessons here, but the literal strategies above should not even be assumed to be close to being optimal, even with the league parameters we've assumed. And if you've got 10 teams, start two QBs, can play a flex, have point-per-reception scoring, or have some other twists, then this article even less relevant.

For me, the value of this article, and the programming that went along with it, was that I learned how complex this little hobby of ours is. This exercise probably took a total of 20 minutes of running time on a pretty snazzy machine --- I don't want to make too big a deal out of that, but that's a lot of electronic huffing and puffing for what seems like a pretty straightforward problem --- and it just barely scratches the surface of the factors that need to be considered.

Enjoy your drafts, everyone.