AVT vs. Projections (1 Viewer)

[BusterTBronco]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

If I'm going with AVT (and I believe him to be QB1) then I project him at 320. If I'm going with VBD then I project him at 330.

I believe that AVT has advantages in that it regresses projections toward historical averages and also provides for greater use of the collective wisdom of many(which is consistently shown to outperform the individual wisdom of one). There are dozens and dozens of expert rankings, mock drafts, etc. but very few expert statistical projections.

 

[Maurile Tremblay]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

No, we don't have a different understanding of AVT.What you're missing is that Tom Brady doesn't have a 100% chance of being QB1.If QB1 scores 320 and QB2 scores 310 and QB3 scores 300, and if Brady has a 40% chance being QB1, a 30% chance of being QB2, and a 30% chance of being QB3, then why would you project him to score 320? That's overprojecting him. That's why AVT is conceptually wrong.

 

[BusterTBronco]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

No, we don't have a different understanding of AVT.What you're missing is that Tom Brady doesn't have a 100% chance of being QB1.If QB1 scores 320 and QB2 scores 310 and QB3 scores 300, and if Brady has a 40% chance being QB1, a 30% chance of being QB2, and a 30% chance of being QB3, then why would you project him to score 320? That's overprojecting him. That's why AVT is conceptually wrong.

By that logic, you would say that AVT always overprojects the top player at each position. This hypothesis is easily disproven with the following example.Looking at the historical results in my WCOFF style league, the top QB for the last three years has been:1995 - Carson Palmer, 320 points1996 - Peyton Manning, 364 points1997 - Tom Brady, 454 pointsIf I go with a three year average for my AVT, the AVT projection for QB1 would be 379 fantasy points.Now suppose that I did an independent projection for Tom Brady and project that Brady will score 400 fantasy points in 2008. Furthermore, I believe that there is a 50% chance that Tom Brady will score more than 400 and a 50% chance that he will score less.In this case, my AVT (379 points) is actually UNDER-projecting my QB1.So, clearly, it is possible that AVT underprojects the top players at each position. I have yet to see any evidence that AVT is more likely to overproject or underproject.

 

[Maurile Tremblay]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

No, we don't have a different understanding of AVT.What you're missing is that Tom Brady doesn't have a 100% chance of being QB1.

If QB1 scores 320 and QB2 scores 310 and QB3 scores 300, and if Brady has a 40% chance being QB1, a 30% chance of being QB2, and a 30% chance of being QB3, then why would you project him to score 320? That's overprojecting him. That's why AVT is conceptually wrong.

By that logic, you would say that AVT always overprojects the top player at each position.

Given the information it is based on, it always overprojects (ex ante, not ex post). If you consider more information than AVT is based on (like when you do individual projections), there may be a reason to project a top player to score more than AVT would.

This hypothesis is easily disproven with the following example.

Looking at the historical results in my WCOFF style league, the top QB for the last three years has been:

1995 - Carson Palmer, 320 points

1996 - Peyton Manning, 364 points

1997 - Tom Brady, 454 points

If I go with a three year average for my AVT, the AVT projection for QB1 would be 379 fantasy points.

And Manning was everybody's QB1 last year. How many points did he score?

Now suppose that I did an independent projection for Tom Brady and project that Brady will score 400 fantasy points in 2008. Furthermore, I believe that there is a 50% chance that Tom Brady will score more than 400 and a 50% chance that he will score less.

Great. That's not AVT.

In this case, my AVT (379 points) is actually UNDER-projecting my QB1.

Yes, because you evidently took more information into account than AVT does.But if we're only taking into account the information that AVT considers (i.e., historical points by the EOY Nth-ranked player at each position, and a set of rankings), the question is whether AVT makes appropriate use of that information to arrive at a good set of projections. The answer is that it doesn't, because it implicitly gives a 100% chance that Manning will be QB1, when in fact there is at least a small chance that Tom Brady will be instead. So Manning's projection is too high (ex ante).

So, clearly, it is possible that AVT underprojects the top players at each position. I have yet to see any evidence that AVT is more likely to overproject or underproject.

The evidence would be very easy to come by. Figure out the average historical values of the players who ended up QB1, RB1, WR1, etc. Then figure out the average historical values of the players who were ranked (in preseason cheatsheets) as QB1, RB1, WR1, etc. Then confirm that the first set of numbers is higher than the second set of numbers. You can try this at home.

 

[Maurile Tremblay]

Looking at the historical results in my WCOFF style league, the top QB for the last three years has been:1995 - Carson Palmer, 320 points1996 - Peyton Manning, 364 points1997 - Tom Brady, 454 points

BTW, I did the exact same thing in a conversation yesterday -- being a decade off regarding those same years.

 

[GregR_2]

I think it is worth noting that someone can make the exact same mistake in doing individual projections that MT is saying that AVT is making. I think everyone will agree our goal is to make as few mistakes as possible and get to the best prediction to use. So even when doing individual projections, a good approach will consciously take the uncertainty in the range of outcomes for each player into account. Even GUT drafters do this though they leave it a less-defined "feeling" about how comfortable they are drafting the player.

With the base AVT methodology, you'll always have this issue though. You would need to do something like MT alluded to in his post about what the odds are that Peyton ends up the QB1, QB2, QB3, QB4, etc, and find what his expectation is based on that. And once you're at the point of doing that, I'm not sure what advantage AVT is giving you that a good individual projection methodology that compares to historical results won't give... and taking the time to do the projections will give the kind of extra benefits MT cross-posted from his article.

 

[John Mamula]

I'm debating between Average Value Theory and using Projections to base my VBD board on. And I'm wondering what everyone likes to use and why. Thanks for all your help.

Projections, but look at your projections to see where they come in to relation to AVT. They don't need to mirror AVT, but should be in the ballpark.Otherwise if you overinflate projections for the 2nd-3rd tier of guys your tier dropoffs will be completely goofed and your VBD values will point you in a very wrong direction.

 

[cmigs]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

No, we don't have a different understanding of AVT.What you're missing is that Tom Brady doesn't have a 100% chance of being QB1.

If QB1 scores 320 and QB2 scores 310 and QB3 scores 300, and if Brady has a 40% chance being QB1, a 30% chance of being QB2, and a 30% chance of being QB3, then why would you project him to score 320? That's overprojecting him. That's why AVT is conceptually wrong.

Isn't this solved by prorating the values? I.e. combine 40% of the QB1 points, 30% QB2, 30% QB3.I've seen an article that suggests basically this:

http://www.fantasyfootball.com/pagebank/index.html?id=192

 

[Maurile Tremblay]

Perhaps you and I have a different understanding of AVT. To me, AVT is based on formulas (i.e. a three year average) and is completely backward looking. There is no room for personal conjecture or intepretation of the number. If QB1 has a three year average of 320 fantasy points then I plug that in as my "projection" for whoever I think QB1 will be this year. I could, independently, project that an individual player would score more the AVT number. For example, I might project that Tom Brady would score 330 fantasy points next season.

No, we don't have a different understanding of AVT.What you're missing is that Tom Brady doesn't have a 100% chance of being QB1.

If QB1 scores 320 and QB2 scores 310 and QB3 scores 300, and if Brady has a 40% chance being QB1, a 30% chance of being QB2, and a 30% chance of being QB3, then why would you project him to score 320? That's overprojecting him. That's why AVT is conceptually wrong.

Isn't this solved by prorating the values? I.e. combine 40% of the QB1 points, 30% QB2, 30% QB3.

Yes, that would be an easy improvement over AVT.

I've seen an article that suggests basically this:

http://www.fantasyfootball.com/pagebank/index.html?id=192

Before it was called PAVT, it was known as APVT. (From the original Shark Pool thread discussing AVT, the day after the AVT article was released.) I didn't name it, but I came up with the idea.In any case, I favor APVT only as an improvement over AVT -- not as an improvement over individual projections.

 

[BusterTBronco]

Both are fundamentally flawed.

I wouldn't say that doing projections is flawed. I would say that drafting strictly off a set of projections without taking other things into account is flawed. But doing projections is still worthwhile.AVT . . . not worthwhile, IMO. (Look for an FBG article on the subject in the near future.)

Where can I find the FBG article you are referring to?

 

[gheemony]

 

[JerseyPaul]

One interesting note to the FBG staff. I agree that projections are really the foundation of your drafting strategy. For those interested in Auctions, however, the FBG Excel App assigns the same number of Auction points (equiv. to auction $ when applied to your leagues auction set-up) to RB1, WR1, etc. independant of their points, VBD, or any other valuation. In essence, this auction value is the equivalent of an AVT auction value.

I guess it's a simplification used because the demand is not there to do something more complex. But the user should know what he/she is getting.

 

[gheemony]

One interesting note to the FBG staff. I agree that projections are really the foundation of your drafting strategy. For those interested in Auctions, however, the FBG Excel App assigns the same number of Auction points (equiv. to auction $ when applied to your leagues auction set-up) to RB1, WR1, etc. independant of their points, VBD, or any other valuation. In essence, this auction value is the equivalent of an AVT auction value.I guess it's a simplification used because the demand is not there to do something more complex. But the user should know what he/she is getting.

I'm a huge FBG fan, but one thing lacking is good auction values. All of the values are really bad (Excel app, Dominator), except I think a staffer (Maurile) wrote a great article on how to set auction values. I basically use FBG projections and/or rankings and use them to come up with auction values. The key is accurately converting a VBD value into a $$ value.

 

[Muahahaha]

In practice, there is very little difference in AVT vs. VBD based drafting. I use both methods concurrently during drafts. They basically come to the same conclusion as to who and what position to draft throughout the draft.

Focusing on the differences between these two methods is probably less useful than focusing on what baselines to use, what round to stop using VBD or AVT, how to change drafting strategy during runs, to what extent should you "modify" value for RBs once you have already taken three of them vs. only one WR, etc. There are so many other elements of drafting that are so much more important than the differences between AVT and VBD. Focusing on the differences between these two methods is really losing the forest for the trees.

Having said all that, IMHO, projections make for much better rankings and appear to be very useful. This site's rankings always seem to outperform other sites for that reason, among many.

 

[Doug Drinen]

One interesting note to the FBG staff. I agree that projections are really the foundation of your drafting strategy. For those interested in Auctions, however, the FBG Excel App assigns the same number of Auction points (equiv. to auction $ when applied to your leagues auction set-up) to RB1, WR1, etc. independant of their points, VBD, or any other valuation. In essence, this auction value is the equivalent of an AVT auction value.I guess it's a simplification used because the demand is not there to do something more complex. But the user should know what he/she is getting.

I'm a huge FBG fan, but one thing lacking is good auction values. All of the values are really bad (Excel app, Dominator), except I think a staffer (Maurile) wrote a great article on how to set auction values. I basically use FBG projections and/or rankings and use them to come up with auction values. The key is accurately converting a VBD value into a $$ value.

I can't speak to the Excel app, but the Draft Dominator last year introduced the option of using a number of different auction valuation methods, one of which was the one outlined in Maurile's article. MyFBG's auction values are also based on Maurile's article.