Auction $ value question - from VBD app (1 Viewer)

[Tiger Fan]

Running the VBD App for a standard scoring PPR league. 10 team, 16 round, 1QB/2RB/2WR/1TE/1K/1DEF - the following are the top 4 RBs, their # value, and their points projected (taken from the RB tab).

Player $ Pts Chris Johnson 79 315Maurice Jones-Drew 75 315 Ray Rice 72 312 Adrian Peterson 69 312Clearly, something is wrong:
[*]CJ & MJD have the same points projected, yet their $ values are $4 different

[*]RR & AP have the same points projected, yet their $ values are $3 different

[*]From 1-4, only 3 points separate them, yet a $10 difference exists

What gives? Anyone else run into similar problems?

 

[Maurile Tremblay]

I don't know if you can call it a bug (it's at least a quirk), but it's a consequence of how players are valued.

A player's value is determined by (a) how many points per game he's expected to score, and (b) how many fantasy starts he's expected to get. It's the product of those two things.

Part (a) is proportional to fantasy points, but part (b) is calculated based on a player's ranking at his position. The #1 RB is expected to get a slightly higher number of fantasy starts (because he's slightly less likely to bust) than the #2 RB.

In determining a player's ranking, the VBD app doesn't recognize ties. Because of that quirk, the VBD app is considering Chris Johnson the #1 RB and MJD the #2 RB even though they have the same projected points.

At least, that's what I think is going on.

 

[Tiger Fan]

I don't know if you can call it a bug (it's at least a quirk), but it's a consequence of how players are valued.A player's value is determined by (a) how many points per game he's expected to score, and (b) how many fantasy starts he's expected to get. It's the product of those two things.Part (a) is proportional to fantasy points, but part (b) is calculated based on a player's ranking at his position. The #1 RB is expected to get a slightly higher number of fantasy starts (because he's slightly less likely to bust) than the #2 RB.In determining a player's ranking, the VBD app doesn't recognize ties. Because of that quirk, the VBD app is considering Chris Johnson the #1 RB and MJD the #2 RB even though they have the same projected points.At least, that's what I think is going on.

Thanks for taking the time to respond to the quirk. Guess I'll have to make some adjustments.

 

[CalBear]

I don't know if you can call it a bug (it's at least a quirk), but it's a consequence of how players are valued.A player's value is determined by (a) how many points per game he's expected to score, and (b) how many fantasy starts he's expected to get. It's the product of those two things.Part (a) is proportional to fantasy points, but part (b) is calculated based on a player's ranking at his position. The #1 RB is expected to get a slightly higher number of fantasy starts (because he's slightly less likely to bust) than the #2 RB.In determining a player's ranking, the VBD app doesn't recognize ties. Because of that quirk, the VBD app is considering Chris Johnson the #1 RB and MJD the #2 RB even though they have the same projected points.At least, that's what I think is going on.

Thanks for taking the time to respond to the quirk. Guess I'll have to make some adjustments.

There is some magic going on in the VBD app (and the Draft Dominator) which is not always fully explained.Look at columns BW through CD. Column BZ is "VBD", CA is "Auction", which is some sort of value that's calculated by the spreadsheet macros. VBD and auction value should be directly related (in my opinion), but they do not appear to be; in the default configuration, Chris Johnson gets 127 VBD points and 2405 "Auction", while Peterson has 124 VBD points and 2269 "Auction". If they were linearly related, Peterson would have more like 2350 "Auction" value. That number is what's used to calculate the auction price.Probably what is happening is that the VBD app is using a multiplier of some kind based on the ranking by position, strength of schedule calculations, weekly/playoff projections, projections of number of games started, or all four. In DD you can modify the "auction.csv" file to remove or change the ranking-by-positiion multiplier, and you can turn off strength of schedule and weekly projections at different places in the interface. I wish there were a "give me clean auction values without all that cruft" selection you could make, but you really have to tweak it manually. I don't know how to do that in the VBD app.

 

[Maurile Tremblay]

VBD and auction value should be directly related (in my opinion), but they do not appear to be;

They shouldn't be proportional to each other, IMO, for reasons I state here.

 

[CalBear]

VBD and auction value should be directly related (in my opinion), but they do not appear to be;

They shouldn't be proportional to each other, IMO, for reasons I state here.

I think we've discussed this before, and I disagreed with you then, too. I think the kinds of factors you're considering are already priced into projections, and thus you're double-counting them by adding them into the auction prices. I certainly don't think that Chris Johnson should cost $4 more than Adrian Peterson just because Johnson is projected to score 3 more points for the year. You could add one TD to Peterson's projections and it would cause an $8 swing in value. That doesn't pass the smell test. I do not think that Chris Johnson is likely to start more games for a fantasy team than Adrian Peterson, so giving Johnson extra credit for being the #1 RB makes no sense.

Other things which don't pass the smell test:

In your article you give Matt Hasslebeck $2 more auction value than Tom Brady because his first name is earlier in the alphabet. Every stat for the two is identical, except Brady gets fewer projected starts because he slots in as QB4 instead of QB3. This is clearly an error and I think it speaks to the fact that the concept is fundamentally flawed.

 

[Maurile Tremblay]

VBD and auction value should be directly related (in my opinion), but they do not appear to be;

They shouldn't be proportional to each other, IMO, for reasons I state here.

I think we've discussed this before, and I disagreed with you then, too. I think the kinds of factors you're considering are already priced into projections, and thus you're double-counting them by adding them into the auction prices. I certainly don't think that Chris Johnson should cost $4 more than Adrian Peterson just because Johnson is projected to score 3 more points for the year. You could add one TD to Peterson's projections and it would cause an $8 swing in value. That doesn't pass the smell test. I do not think that Chris Johnson is likely to start more games for a fantasy team than Adrian Peterson, so giving Johnson extra credit for being the #1 RB makes no sense.

Other things which don't pass the smell test:

In your article you give Matt Hasslebeck $2 more auction value than Tom Brady because his first name is earlier in the alphabet. Every stat for the two is identical, except Brady gets fewer projected starts because he slots in as QB4 instead of QB3. This is clearly an error and I think it speaks to the fact that the concept is fundamentally flawed.

I think that's why Chris Johnson is being valued more highly than Maurice Jones-Drew in Tiger Fan's spreadsheet as well: because of his [last] name. I agree that that part of the calculation could stand some refinement, but positional rank is a better proxy for expected fantasy starts than any other easily available number I'm aware of, and as long as positional ranks don't recognize ties (and I agree that they should), name will be the tie-breaker.As for Johnson versus Peterson, this year's top four RBs are bunched more closely together than in the average year. That's another drawback of using positional rankings; in addition to not recognizing ties, rankings also don't indicate how close or how far apart the #2 and the #3 RB are. But still, positional rankings are a better proxy for expected fantasy starts than any other easily available number I'm aware of. (And whatever you think of Johnson and Peterson specifically, on average, the projected #2 RB will have more fantasy starts than the projected #3 RB.)

Auction values are not proportional to VBD values because bench points don't matter. Say that the #1 RB will score ten more points than the #2 RB, and the #47 RB will score ten more points than the #52 RB.

It should be evident that you'd prefer the #1 RB over the #2 RB by a greater amount than you'd prefer the #47 RB over the #52 RB. That's because the differential between #1 and #2 really matters, while the differential between #47 and #52 is largely illusory since they'll both be on your bench, so it doesn't matter what they score.

 

[renesauz]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

 

[Maurile Tremblay]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

I think Tiger Fan's question was legitimate. It's not so much that a couple dollars at the top of the auction is a big deal; but Tiger Fan saw something that seemed to be an error, which calls into question the VBD app's calculations on more important matters as well.The answer is that the VBD app makes certain simplifying assumptions that do produce small quirks here and there, but those quirks generally do not portend larger errors on more important matters (IMO).

 

[CalBear]

I think that's why Chris Johnson is being valued more highly than Maurice Jones-Drew in Tiger Fan's spreadsheet as well: because of his [last] name. I agree that that part of the calculation could stand some refinement, but positional rank is a better proxy for expected fantasy starts than any other easily available number I'm aware of, and as long as positional ranks don't recognize ties (and I agree that they should), name will be the tie-breaker.

As for Johnson versus Peterson, this year's top four RBs are bunched more closely together than in the average year. That's another drawback of using positional rankings; in addition to not recognizing ties, rankings also don't indicate how close or how far apart the #2 and the #3 RB are. But still, positional rankings are a better proxy for expected fantasy starts than any other easily available number I'm aware of. (And whatever you think of Johnson and Peterson specifically, on average, the projected #2 RB will have more fantasy starts than the projected #3 RB.)

Again, I think trying to predict the number of fantasy starts, and change the auction values based on that, is a fundamentally flawed concept. Essentially you're overlaying AVT on top of VBD in an attempt to improve the accuracy of the values--but AVT inherently contains less information than VBD, so you're actually reducing the amount of useful information in the values.The fact that the #1 drafted player on average started 11.1 games in the past, and the #3 started 10.7, says little or nothing about whether Chris Johnson will start more games than MJD, this year, for my team. The likelihood of a player starting on my team is partially based on the next highest-rated player at that position on my team, which I won't even know until later in the auction. If I already have chosen C.Johnson and R.Grant, J.Addai would have a relatively low number of projected starts. If I only have C.Johnson on the roster so far, Addai has a high number of projected starts. You have to project the number of starts based on the team being assembled, not on historical data for a draft position--otherwise you're throwing noise into the system and not describing anything meaningful.

Auction values are not proportional to VBD values because bench points don't matter. Say that the #1 RB will score ten more points than the #2 RB, and the #47 RB will score ten more points than the #52 RB.

It should be obvious that you'd prefer the #1 RB over the #2 RB by a greater amount than you'd prefer the #47 RB over the #52 RB. That's because the differential between #1 and #2 really matters, while the differential between #47 and #52 is largely illusory since they'll both be on your bench, so it doesn't matter what they score.

Whether a player will be a bench player is not determined by his draft position, it's determined by the other players on my team. During the auction, I will automatically discount auction values for clear backup players on my team. Furthermore, I will automatically increase auction values for clear starters or important backups (handcuffs, backups behind injured/suspended players, etc.) I won't pay the same amount for Addai-as-starter as I will for Addai-as-backup. The software shouldn't be trying to mastermind the auction values for me; if it does, then the adjustments I'll make during the auction will double-count the same factors. If you really want to do this calculation for the user, you should have "expected fantasy starts" as a separate stat column for each player, and let the user input values himself. Otherwise you're using historical data which are largely irrelevant to my team, this year.

As I said, you can turn most of this off in DD, but it's not straightforward how to do so.

 

[CalBear]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

You're not just being rude, you're being obtuse. It's not a question of interpreting the data reasonably--the data are not reasonable. How do you interpret two players with the same VBD value having different projected auction cost? The original poster is trying to figure out why the VBD app is putting out the numbers that way, so he can interpret the data reasonably.

 

[Ahmad Rashad]

VBD and auction value should be directly related (in my opinion), but they do not appear to be;

They shouldn't be proportional to each other, IMO, for reasons I state here.

Also, be aware that the prices from any list like the one above do not take into account team needs. A player's value is determined by how many more points your team would score with him than without him - and this depends on the quality of the other players on your roster at the same position. If you submit the winning bid on Peyton Manning, Tom Brady's value to your team will fall well below the value listed on your cheatsheet. The cheatsheet reflects how much better Brady is than the average backup QB . . . but his actual value to your team will be based on how much better he is than Manning. In other words, do not load up on high-priced backups who will score a lot of meaningless points on your bench.

This is the big problem I have with the Draft Dominator auction values (in addition to the others mentioned on this thread). The adjustment to auction values based on position rank assume that I'm going to draft like the average person in that historical league used for the calculations. IMO, the key to a good auction draft is NOT drafting like the average person. You need to take into account your league rules and how people in your league bid, and then use an unconventional strategy to gain an edge. Drafting like the average person guarantees you'll have an average team, barring unusual luck.For example, if I know that my league tends to overvalue the high receivers, leaving 3rd-tier and bottom-tier receivers undervalued, I might want to allocate more $ to other positions and wait until the 3rd-tier receivers to get my WR1. My WR1 might be the #20 WR, and my WR2 might be the #50 WR, so the #of fantasy starts for my WR1 would be much higher than assumed by the calculation that's based on that 2002-2005 league. My QB1 might be the #1 QB, and both my starting RB's might be top-8 RB's because of the extra dollars I've saved on WR. My team needs will be much different from an average team from some unassociated league with very different rules that played 8 years ago thousands of miles away.

For those of us who would rather make adjustments to proportional VBD-auction value figures based on our own leagues' history, rather than "the No Mercy - Great White league from 2002 through 2005," there should be an option to calculate auction values based on proportional VBD points. Because my league is so unusual, I end up having to export the DD player pool and make a bunch of adjustments to take out the calculations based on that league (as well as things like first name in alphabet extra points).

The powers that be at Footballguys may be convinced their way is the right way and the only way, but for those of us who disagree, can we at least have the option of having auction values based on proportional VBD points? (maybe with an "expected fantasy starts" column and a caveat that this option should be the basis of other draft strategies, and should not be used as is?)

 

[Ahmad Rashad]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

It's not just "a couple bucks at the uppermost end of the scales"... It's a problem throughout the player pool -- it's so bad, I completely ignore the DD auction values and recalculate my own. It's a pain, but necessary, IMO, until they fix it.Tiger Fan's example shows a $10 difference between 2 players whose projected points for the season is only 3 points. If you've participated in auction drafts before, you know that $10 in a $200 cap league can make a huge difference for your team. This is clearly a problem with the calculations that DD does based on the average player in some league that played 8 years ago. For those of us that don't like problems like this, Footballguys should give us an option to not use those calculations.

 

[Tiger Fan]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

I think Tiger Fan's question was legitimate. It's not so much that a couple dollars at the top of the auction is a big deal; but Tiger Fan saw something that seemed to be an error, which calls into question the VBD app's calculations on more important matters as well.The answer is that the VBD app makes certain simplifying assumptions that do produce small quirks here and there, but those quirks generally do not portend larger errors on more important matters (IMO).

Thanks again. I'll take this and make a few twerks to my final values.

 

[Tiger Fan]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

It's not just "a couple bucks at the uppermost end of the scales"... It's a problem throughout the player pool -- it's so bad, I completely ignore the DD auction values and recalculate my own. It's a pain, but necessary, IMO, until they fix it.Tiger Fan's example shows a $10 difference between 2 players whose projected points for the season is only 3 points. If you've participated in auction drafts before, you know that $10 in a $200 cap league can make a huge difference for your team. This is clearly a problem with the calculations that DD does based on the average player in some league that played 8 years ago. For those of us that don't like problems like this, Footballguys should give us an option to not use those calculations.

How do you go about recalculating your $ values? I brainstormed a few formulas and came up with this: (# of teams X salary cap/team) / (total # of points projected to be scored by starters & bench in league)

 

[Ahmad Rashad]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

It's not just "a couple bucks at the uppermost end of the scales"... It's a problem throughout the player pool -- it's so bad, I completely ignore the DD auction values and recalculate my own. It's a pain, but necessary, IMO, until they fix it.Tiger Fan's example shows a $10 difference between 2 players whose projected points for the season is only 3 points. If you've participated in auction drafts before, you know that $10 in a $200 cap league can make a huge difference for your team. This is clearly a problem with the calculations that DD does based on the average player in some league that played 8 years ago. For those of us that don't like problems like this, Footballguys should give us an option to not use those calculations.

How do you go about recalculating your $ values? I brainstormed a few formulas and came up with this: (# of teams X salary cap/team) / (total # of points projected to be scored by starters & bench in league)

You have to choose a VBD baseline and separate the above-baseline players from the below-baseline players. Then you take the available points (after keepers' costs are taken out) and allocate proportionally to the two pools of players, and then make adjustments based on your league. You have to redo it every time projections are updated. It would be much easier if Footballguys gave us the option of using proportional auction values instead of those calculations based on that historical league and the alphabet.

 

[GregR_2]

...Again, I think trying to predict the number of fantasy starts, and change the auction values based on that, is a fundamentally flawed concept. Essentially you're overlaying AVT on top of VBD in an attempt to improve the accuracy of the values--but AVT inherently contains less information than VBD, so you're actually reducing the amount of useful information in the values....

Just wanted to comment on this part. I don't think that's a correct statement.The problem (or one of the major ones) with AVT vs projections is that AVT takes the results of the players who ENDED up RB4 (or whatever) and uses that as a projection for the specific player (i.e. Ray Rice) PREDICTED to be RB4. The problem is that this overstates the likelihood that Ray Rice ends up RB4. If instead of using named players, we were just drafting "the player who will end up RB4 this year", then AVT would be a much more appropriate system. I point that out because MT is not mixing pre and post season like AVT. If I remember his article correctly he found how many games preseason RB4s played and is using that as a predictor for a preseason RB4. In that regard MT is appropriate where AVT isn't as MT is consistent in comparing preseason ranked players to preseason ranked players.Now you may (or may not) have some merit in saying it still isn't a good enough predictor (because of differences in the player pool, because of differences between the leagues MT took the data from compared to your league, because your backups tend to be better/worse than the leagues he took his data from) to want to use it. But I don't think it's right to invoke AVT as an argument. If MT's fantasy starts predictor has shortcomings, it isn't the same as AVT's shortcomings. Which leads to the question.... MT, now that you have 3 more years of data, have you had a chance to go back and taken a look to see how the model has done in predicting fantasy starts?I too would like to see Draft Dominator have the option to not include the fantasy starts. I actually didn't think they were used unless MT's Method was selected for the baseline... but testing just now I see it kicks in for other baselines too. If nothing else I'd want to be able to view both sets of values so I could make a judgment about which is better for the current crop of players.

 

[GregR_2]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

It's not just "a couple bucks at the uppermost end of the scales"... It's a problem throughout the player pool -- it's so bad, I completely ignore the DD auction values and recalculate my own. It's a pain, but necessary, IMO, until they fix it.Tiger Fan's example shows a $10 difference between 2 players whose projected points for the season is only 3 points. If you've participated in auction drafts before, you know that $10 in a $200 cap league can make a huge difference for your team. This is clearly a problem with the calculations that DD does based on the average player in some league that played 8 years ago. For those of us that don't like problems like this, Footballguys should give us an option to not use those calculations.

How do you go about recalculating your $ values? I brainstormed a few formulas and came up with this: (# of teams X salary cap/team) / (total # of points projected to be scored by starters & bench in league)

Check this thread, where some people who can't use laptops at their auction were asking both how to calculate auction values, and how to recalculate on the fly to account for what had actually been spent in the auction. If you're using a spreadsheet it's pretty easy to set up with formulas.

Near the bottom I gave a little more complex method that does a better job with backups than the first method. Though to really do a bang up job with it you probably need to do some work with backup values. I don't think you can use projections to do non-primary backups very accurately... since their value to your team is typically based on how much they'll score if they become NFL starters, and how likely it is that they'll become NFL starters.

 

[Tiger Fan]

If you're going to nitpick stuff like this...perhaps you should just let Dodds do your draft for you (assuming he would). Or you could pay me 20 bucks to draft for you, since apparently you can't interpret the data reasonably.

While I am being a little bit rude here, I simply can't see why anyone would complain about a couple bucks at the uppermost end of the scales...it should have virtually no baring on your strategy or draft.

It's not just "a couple bucks at the uppermost end of the scales"... It's a problem throughout the player pool -- it's so bad, I completely ignore the DD auction values and recalculate my own. It's a pain, but necessary, IMO, until they fix it.Tiger Fan's example shows a $10 difference between 2 players whose projected points for the season is only 3 points. If you've participated in auction drafts before, you know that $10 in a $200 cap league can make a huge difference for your team. This is clearly a problem with the calculations that DD does based on the average player in some league that played 8 years ago. For those of us that don't like problems like this, Footballguys should give us an option to not use those calculations.

How do you go about recalculating your $ values? I brainstormed a few formulas and came up with this: (# of teams X salary cap/team) / (total # of points projected to be scored by starters & bench in league)

Check this thread, where some people who can't use laptops at their auction were asking both how to calculate auction values, and how to recalculate on the fly to account for what had actually been spent in the auction. If you're using a spreadsheet it's pretty easy to set up with formulas.

Near the bottom I gave a little more complex method that does a better job with backups than the first method. Though to really do a bang up job with it you probably need to do some work with backup values. I don't think you can use projections to do non-primary backups very accurately... since their value to your team is typically based on how much they'll score if they become NFL starters, and how likely it is that they'll become NFL starters.

thx

 

[CalBear]

Just wanted to comment on this part. I don't think that's a correct statement.The problem (or one of the major ones) with AVT vs projections is that AVT takes the results of the players who ENDED up RB4 (or whatever) and uses that as a projection for the specific player (i.e. Ray Rice) PREDICTED to be RB4. The problem is that this overstates the likelihood that Ray Rice ends up RB4. If instead of using named players, we were just drafting "the player who will end up RB4 this year", then AVT would be a much more appropriate system.

That's a worthwhile distinction, but I don't think it makes much of a difference. The main problem with AVT is that it's using past results of a population to predict the future performance of an individual. That doesn't work. Maurile's population is slightly different than the AVT population, but the point is the same; the results of the last N years of 1.02 draft picks is not useful in predicting what Adrian Peterson will do this year. It doesn't even matter how large N is.

 

[Ahmad Rashad]

Any chance someone at Footballguys finally does something to fix Draft Dominator for auctions? It's clearly broken, as shown in the opening post, and the fix isn't that difficult (offer an option that removes the calculations weighting auction values based on average fantasy starts per position rank from an old irrelevant league and remove logic based on position rank that uses the alphabet to give players several $ more in value).