Comments on Advanced NFL Stats: Game Theory and Fantasy Draft Strategy

Partially inspired by this post, I have created an...

2009-08-23T12:21:21.726-04:00

Partially inspired by this post, I have created an ADP-adjusted Dynamic VBD spreadsheet and made it freely available. It's similar to VONA/dVBD in principle, but accounts for the fact that players are not likely to be taken in the order which _you_ have them projected. Details here:

http://yocto.wordpress.com/2009/08/22/adp-adjusted-dynamic-value-based-drafting/

Well, I'm not sure what else I can spell out. Ther...

2008-12-03T16:52:00.000-05:00

Well, I'm not sure what else I can spell out. There's no equation or algorithm, just the method I described. If you go to the 'example draft' link in the article, you'll see how the method is applied.

If you're referring to my game prediction model, the full description is here.

Hi Brian,

Would you mind sending out your c...

2008-12-03T10:23:00.000-05:00

Hi Brian,

Would you mind sending out your complete model?

"Value over next available" is also referred to as...

2008-09-15T11:10:00.000-04:00

"Value over next available" is also referred to as "Dynamic VBD." There is an article or two on the web about it.

I use a combo of similar ideas. The pieces I use i...

2008-09-10T01:47:00.000-04:00

I use a combo of similar ideas. The pieces I use in our modeling here at Protrade (each is relative to the specific position in question):

1) value over starter, which is essentially what you are calling VORP above
2) value over last roster - there is value in having bench players - for bye week and injury purposes. For example, a 3 RB can be very important, even if you only start 2.
3) VONA

For each of the "value" components, we take a measure of expected points scored, as well as consistency (beta).

The key is - how do you weight these three things together? The weighting should slide based on what round you are in the draft (are you late, and depth is all that matters), as well as what draft slot you are (for example VONA for the 9th pick in a 10 team draft in round 1 is not as relevant as for the 5th team).

I find the game theory elements extremely interesting for fantasy football. Even if you all have perfect, similar projections for every player, you may not win because of game theory. Throw in the variability of projections, and it is a very fun game!

-Mark (mkamal@protrade.com)

Just found this website. Super series of articles...

2008-08-26T21:34:00.000-04:00

Just found this website. Super series of articles! I just wanted to add one extra twist, which is probably obvious. After the first 6-10 rounds (depending on the size/format of the league), there isn't much VORP or VONA to be found, at which point one should switch to players with very high variance (i.e., could be busts, or could be nuggets of gold, like rookies or Leinart/Warner for example).

This VONA concept is interesting... I think you're...

2008-08-26T20:46:00.000-04:00

This VONA concept is interesting... I think you're on the right track but one adjustment is needed.

See, VONA fails to consider the relative value of a prospective player with respect to playing against those players already taken. If you select someone much worse than those already taken, that's a major opportunity cost (OC).

Value should be estimated in terms of how a player compares to those whom he will actually be playing against day in and day out. For a 8 team league, I'd compare to the value of the top 8 (for QB).

In your spreadsheet, let V be the sum of the difference between the projected points of your prospective pick and the projected points of each of the top 8 QBs.

Once you have a V for each player, THEN apply your VONA logic. OC is the difference between V_available_now and V_available_next_round.

Great stuff! I started using a VONA-like approach ...

2008-08-22T06:54:00.000-04:00

Great stuff! I started using a VONA-like approach last season after struggling with the exact VORP problem you describe (TE too early and QB too late). My teams all were much stronger as a result and I got lots more "you stole my next pick!" comments, which is always a good sign. It's good to get some confirmation that what I am doing makes sense. I'm wondering though if there is a way to make the analysis a MULTI-ROUND one. For example, if VONA analysis recommends a WR in round 1, that will impact how others will pick in round 2, and then ultimately what my picks will be. Since the goal is to maximize my total pick values, not just one round, is there a way to account for this multi-round impact? Or is this getting too late in the causation chain? Is the variation in what others pick so large that it makes pointless any effort to think 2-3 rounds ahead? Again, great work here. Thanks.

Ken-I think that's a good point. That can happen. ...

2008-08-21T19:51:00.000-04:00

Ken-I think that's a good point. That can happen. But if someone does that, they're leaving a lot of differential value at another position for you. It's a net gain for you. Plus, realistically, you're not going to be stuck with the #20 QB, maybe the 11th or 12th worst case. And there isn't much difference between QBs when you get that deep.

I think I disagree with your definition of replace...

2008-08-21T17:04:00.000-04:00

I think I disagree with your definition of replacement value. If there are 10 teams and everyone gets 2 qb's, the replacement level is the 20th, the absolute worst case scenario for you and a production level you can be guaranteed to exceed if your rankings are correct.

I feel you are suggesting that no team would take their second qb before all other slots are filled, but if you have say the 9th qb, you might take the 10th with a later pick as well if available since you aren't to confident in your 9th ranked quarterback and et cetera.

Sky-
I've tested and compared both strategies, ...

2008-08-21T10:37:00.000-04:00

Sky-
I've tested and compared both strategies, and for the most part, against live opponents VORP and VONA often recommend similar picks. If everyone were playing VORP, then I'd guess, yes, they'd come out the same. The big difference is that VORP will often recommend picking a TE too early, and a QB too late.

They would give identical recommendations if in VONA you assumed that all players above replacement will be taken in every round. VONA is really just a round-by-round VORP. And VORP is really just a worst-case-scenario version of VONA.

I guess VONA is a little like Nate Silver's probability distribution, except that I'm not doing a full distribution, just a central best-guess.

Good stuff, Brian. VONA is definitely a theoretic...

2008-08-20T15:33:00.000-04:00

Good stuff, Brian. VONA is definitely a theoretical improvement on VORP, but I wonder how well we can really predict who will be drafted? If everyone else was picking off your personal VORP ratings, your VONA picks would be exactly the same as your VORP picks right? So it comes out to predicting how your own list is different from the other teams' lists.

One way to do that is to compare your own list to an ESPN or Yahoo average-draft position list, noting the biggest differences. For anyone you think is underrated, you can wait longer.

Nate Silver discussed a fantasy baseball drafting spreadsheet one time where he assigned each player a probability distribution for where they'd be selected. You could then combine those distributions (by position, etc.) to pick the player you need now, while letting others slide that provide the greatest expected value for your next pick (or for all future picks). Sounds nice, but difficult.

You just gave away my exact drafting strategy! I ...

2008-08-20T14:01:00.000-04:00

You just gave away my exact drafting strategy! I mean to a tea, this is how I draft in fantasy sports...plug the projections into an Excel SS and find the VORP. Then use average draft position to pick the best available of the guys who will probably be gone by the time it's my turn to pick again. Like you said, it's still real tough to win...a lot of great looking teams just don't produce. So much of it depends on the player projections...that's the real meat and potatoes of fantasy. Good job.

Brian - Excellent analysis. It is surprising how ...

2008-08-20T11:32:00.000-04:00

Brian - Excellent analysis. It is surprising how many people still do not use the concept of "opportunity cost" to inform their decisions. However, if your league allows costless trading, I think that the players you have already drafted and any given point should not have a large effect on which players you subsequently pick. By that I mean if you have drafted Tom Brady, and the opportunity cost of Payton Manning is higher than that of any TE, WR, RB then you should take Manning. After the draft you can try to trade him for a TE, WR, RB of similar value.

Michael- Agreed. But one great thing about expecte...

2008-08-20T09:04:00.000-04:00

Michael- Agreed. But one great thing about expected value is that can account for risk. If you think McNabb has a 50% chance of making it through the season uninjured, then his value would be = 0.50 * (McNabb full season value), or however you choose to weight his value. Usually, injury concerns are already factored in to published values and rankings. And keep in mind, everyone is an injury risk.

Thanks for this, I really enjoyed the series. I u...

2008-08-20T07:52:00.000-04:00

Thanks for this, I really enjoyed the series. I use "value over average" for drafting which basically yields the same thing.

This is problematic in that ranking a players performance tends to be a very difficult thing to do year-in and year-out. Players coming off of career years tend to get penciled in for those type of numbers (which rarely ever happens as there is usually equal parts luck and skill involved in a Madden-cover worthy season). There is not a surefire method for removing this subjectivity.

Also, I think there are players that are inherently conservative and risky. Donovan McNabb is injured every year. When he plays, he's a top-shelf QB. Then there are guys who are the model of consistency, like Reggie Wayne. In drafting, I personally tend to opt for safer bets early and take big risks later on. Looking down my value-over-average sheet doesn't tell the whole story in this way. I might pass over a player because he's risky or I might target a player a little early because, even though he doesn't project as high as some, his ceiling is far higher than the rest of the pack. From my experience, fantasy drafts are won and lost on those types of picks.

This type of analysis that you have adeptly articulated is ultimately the starting point for me, though. Great post.

Great job man...I spend hundreds of hours every ye...

2008-08-20T04:37:00.000-04:00

Great job man...I spend hundreds of hours every year preparing my draft charts. You have a lot of awesome work here. I really appreciate you taking the time to share that information. I'll definitely be incorporating this information in al of my future drafts.

Just in time for my first draft, tomorrow night. G...

2008-08-20T00:12:00.000-04:00

Just in time for my first draft, tomorrow night. Good series.