Analysis Accurate points curve and table

[Suma Magic]

This is really on the nerdy end of trading/draft stuff.

I get pretty frustrated with the AFL's points index as it doesn't reflect how clubs value picks. This is because the AFL based on it salary, which doesn't line up with the pick-for-pick trading market. It's basically garbage, and yet the entire industry still refers to it.

I took all the pick-for-pick trades over the years and calculated a points curve which more accurately reflects the value of picks relative to higher-lower picks. It's not perfect, but it's 100 times more accurate than the AFL's curve. If anything, it could probably be slightly steeper.

Feel free to use it or to provide feedback.




Edit: I think correct formula is actually
=(-850*LN(pick number))+(3000*(pick number^1.002725))

 

[Lore]

This is really on the nerdy end of trading/draft stuff.

I get pretty frustrated with the AFL's points index as it doesn't reflect how clubs value picks. This is because the AFL based on it salary, which doesn't line up with the pick-for-pick trading market. It's basically garbage, and yet the entire industry still refers to it.

I took all the pick-for-pick trades over the years and calculated a points curve which more accurately reflects the value of picks relative to higher-lower picks. It's not perfect, but it's 100 times more accurate than the AFL's curve. If anything, it could probably be slightly steeper.

Feel free to use it or to provide feedback.


View attachment 1528842


View attachment 1528844
View attachment 1528847

How many examples were you working from?

I was just thinking it’d be interesting if we had a list of all the trades involving e.g. pick 12 or whatever over the years and what it was traded for.

The value of the picks would be different year on year depending on the players available at that pick, but I don’t know how much that comes out in the data, maybe the sample size isn’t big enough.

If there were enough data to do it, perhaps a points curve for each year or perhaps group of years (reflecting the increasing depth of the draft pool over time).

 

[Vivy]

This is really on the nerdy end of trading/draft stuff.

I get pretty frustrated with the AFL's points index as it doesn't reflect how clubs value picks. This is because the AFL based on it salary, which doesn't line up with the pick-for-pick trading market. It's basically garbage, and yet the entire industry still refers to it.

I took all the pick-for-pick trades over the years and calculated a points curve which more accurately reflects the value of picks relative to higher-lower picks. It's not perfect, but it's 100 times more accurate than the AFL's curve. If anything, it could probably be slightly steeper.

Feel free to use it or to provide feedback.


View attachment 1528842


View attachment 1528844
View attachment 1528847

Great stuff.

Interesting to see how drastic the fall off is

 

[Suma Magic]

How many examples were you working from?

I was just thinking it’d be interesting if we had a list of all the trades involving e.g. pick 12 or whatever over the years and what it was traded for.

The value of the picks would be different year on year depending on the players available at that pick, but I don’t know how much that comes out in the data, maybe the sample size isn’t big enough.

If there were enough data to do it, perhaps a points curve for each year or perhaps group of years (reflecting the increasing depth of the draft pool over time).

I’ve shut it down now. But about 50 trades I think.

I looked at the overall variation of all trades with different formulas. The afl formula consistently favoured the side of the trade with the highest pick. My formula has it roughly even.

 

[The Passenger]

My general rule of thumb has been to add 20% in cases where there is only 1 pick difference in quantity, 40% when there are 2 picks quantity difference and 60% when there is 3 picks quantity difference.

I.e. according to the offical points 20 + 30 = 8... But no club is trading 8 for those. Adding 20% to 8 brings its value to 1860 points which is equivalent to 15 and 25 which feels like a much more realistic trade. Funnily enough, that is almost bang-on value from the OP's table. 15 + 25 = 1281 and pick 8 is 1290 points.

If you made it 15 + 25 + 35, the official system gives you a further 520 points but using my formula above only brings you up to between pick 6 or 7.

To be exact pick 6 + 40% is 2451 points, which makes it a little more valuable than 15 + 25 + 35 (2390 points). Again, this is pretty on par with the OP where 15 + 25 + 35 in his system gets you 1550 points which makes those 3 picks slightly more valuable than pick 6.

Mine has not been statistically tested or anything but seems to pass the test more or less, and is pretty in line with what the OP has put together.

IMO The OP is on the mark. The highest picks definitely aren't rated high enough, with the curve not steep enough...S/he does provides a much more rigorous analysis than my back-of-a-cigarette packet calculations above.

 

[Thetrader15]

I’ve shut it down now. But about 50 trades I think.

I looked at the overall variation of all trades with different formulas. The afl formula consistently favoured the side of the trade with the highest pick. My formula has it roughly even.

I think if you include pick trade for points to match a bid, you will bias the data as clubs are targeting points on the AFL curve, plus the 20% discount in some cases.

If you just have trades for picks which the club took to the draft or traded for players, that would be interesting.

 

[Suma Magic]

I think if you include pick trade for points to match a bid, you will bias the data as clubs are targeting points on the AFL curve, plus the 20% discount in some cases.

If you just have trades for picks which the club took to the draft or traded for players, that would be interesting.

Yeah that’s what I did (excluded trades done purely for points).

 

[Sector 7G]

This is really on the nerdy end of trading/draft stuff.

I get pretty frustrated with the AFL's points index as it doesn't reflect how clubs value picks. This is because the AFL based on it salary, which doesn't line up with the pick-for-pick trading market. It's basically garbage, and yet the entire industry still refers to it.

I took all the pick-for-pick trades over the years and calculated a points curve which more accurately reflects the value of picks relative to higher-lower picks. It's not perfect, but it's 100 times more accurate than the AFL's curve. If anything, it could probably be slightly steeper.

Feel free to use it or to provide feedback.


View attachment 1528842


View attachment 1528844
View attachment 1528846

I think the problem with the AFL curve is that it doesn't take into account how many player you can have on a list and develop at the one time. Sure if you have 6 picks in the 30s you have as much chance of getting a player as good as the no1 pick but most teams don't have the space for that many players

 

[Andyy]

I really like this. Great work mate.

 

[Suma Magic]

I think the problem with the AFL curve is that it doesn't take into account how many player you can have on a list and develop at the one time. Sure if you have 6 picks in the 30s you have as much chance of getting a player as good as the no1 pick but most teams don't have the space for that many players

Yeah I think that’s part of it.

Whatever the problem is, basing the curve on trade data is the solution I reckon.

 

[Ants]

An interesting question of course is whether clubs are rating pick values accurately. The whole moneyball concept was that baseball clubs weren’t rating certain characteristics correctly. The same could be happening in Australia meaning clubs perceptions of value doesn’t match real value.

A second possibility is that the AFL’s data is starting to be out of date. Especially if drafting accuracy has improved.

 

[Suma Magic]

An interesting question of course is whether clubs are rating pick values accurately. The whole moneyball concept was that baseball clubs weren’t rating certain characteristics correctly. The same could be happening in Australia meaning clubs perceptions of value doesn’t match real value.

A second possibility is that the AFL’s data is starting to be out of date. Especially if drafting accuracy has improved.

If the aim of the points curve is so that clubs can match bids with whatever picks they want, then all that matters is that the points values equal market values. It’s just supposed to stop clubs from being able to produce points out of thin air.

If you are just talking about a guide for points values, then yes maybe intrinsic values don’t match market values. But I reckon it’s pretty close.

 

[Ants]

If the aim of the points curve is so that clubs can match bids with whatever picks they want, then all that matters is that the points values equal market values. It’s just supposed to stop clubs from being able to produce points out of thin air.

If you are just talking about a guide for points values, then yes maybe intrinsic values don’t match market values. But I reckon it’s pretty close.

Not really. If it’s so they match bids with accurate values then market value may be irrelevant.

 

[sydney_gws]

My general rule of thumb has been to add 20% in cases where there is only 1 pick difference in quantity, 40% when there are 2 picks quantity difference and 60% when there is 3 picks quantity difference.

I.e. according to the offical points 20 + 30 = 8... But no club is trading 8 for those. Adding 20% to 8 brings its value to 1860 points which is equivalent to 15 and 25 which feels like a much more realistic trade. Funnily enough, that is almost bang-on value from the OP's table. 15 + 25 = 1281 and pick 8 is 1290 points.

If you made it 15 + 25 + 35, the official system gives you a further 520 points but using my formula above only brings you up to between pick 6 or 7.

To be exact pick 6 + 40% is 2451 points, which makes it a little more valuable than 15 + 25 + 35 (2390 points). Again, this is pretty on par with the OP where 15 + 25 + 35 in his system gets you 1550 points which makes those 3 picks slightly more valuable than pick 6.

Mine has not been statistically tested or anything but seems to pass the test more or less, and is pretty in line with what the OP has put together.

IMO The OP is on the mark. The highest picks definitely aren't rated high enough, with the curve not steep enough...S/he does provides a much more rigorous analysis than my back-of-a-cigarette packet calculations above.

Can also use rejected trades. Didn't Melbourne reject 2 for 8,9 from Port, also Richmond rejected 3 for 6,14 to keep 3 for Dustin Martin.

 

[Suma Magic]

Not really. If it’s so they match bids with accurate values then market value may be irrelevant.

I'll explain what I mean.

Let's assume for the sake of the example that the AFL index actually correctly measures intrinsic value.

And let's say that my curve measures market value.

A comparison of using the two indexes shows how the intrinsic index leaves the academy team better off than they should be.

	Intrinsic index	Market index
Original pick 10 value	1,395​	1,117​
Trade for 28, 29, 30 - market rate
Points of 28, 29, 30	1,959​	1,126​
Bid comes at pick 12
Value of pick 12	1,268​	979​
12 after 20% discount	1014​	783.2​
Value 28	677​	396​
Points left to pay after using pick 28	337​	387​
Value pick 29	653​	375​
Points left to pay after using pick 29	-316​	12​
Value pick 30		355​
Points left to pay after using pick 30		-343​

In the intrinsic scenario, the team is left with picks 29 and 47 after matching. In the market scenario, they are left with pick 31.

If the curve is guided by the market value, then teams trading to obtain points will exploit this to improve their points situation. This doesn't create value, it just shifts it around. It's negative sum (edit: meant zero sum). Meaning that there is an unintended transfer of value from other teams with pick 29-47 to the academy team.

 

[Ants]

But if the market value is flawed there is no negative sum as the teams trading the picks have picked that up. Your system would actually create a negative sum.

 

[Keystone agony]

A graph showing past history the value of pick 1

 

[Suma Magic]

But if the market value is flawed there is no negative sum as the teams trading the picks have picked that up. Your system would actually create a negative sum.

I don’t understand what you’re trying to say sorry. Perhaps you could try to explain with an example.

Edit: I had meant to write zero sum instead of negative sum. All trades and points and matching is zero sum, as they don’t change the total value of players picked.

 

[Ants]

What I’m saying is if market value is flawed, the imbalance is neutralised by the teams trading the points, as one is then actually doing a bad deal. To then apply a market-based curve would create an imbalance. Whereas the value-based curve would not.

 

[Suma Magic]

What I’m saying is if market value is flawed, the imbalance is neutralised by the teams trading the points, as one is then actually doing a bad deal. To then apply a market-based curve would create an imbalance. Whereas the value-based curve would not.

The imbalance isn't actually neutralised though. If we continue to assume a gap between market and intrinsic value, for the sake of the example:

Team(s)	Intrinsic (AFL) curve	Market curve
Academy team trading down	Gain significant advantage as able to turn a pick into picks with higher points values.	Unable to gain significant advantage. In the example, gain a tiny advantage.
Team trading up	Lost value as intrinsic value was less than what they paid.	Lost value as intrinsic value was less than what they paid.	Value lost is exactly the same in both scenarios.
Other teams with picks in 30s and 40s.	Disadvantaged, with the total disadvantage of all clubs equal to the value gained by academy team.	In example, tiny disadvantage equal to tiny advantage enjoyed by academy team.

So while the team trading up has lost value, this would happen regardless of the curve used.

The market curve is just stopping the transfer of value from the other teams with picks in the 30s and 40s to the Academy team trading down. It's fixing an issue without creating any new ones.



I think in reality, while the AFL's curve is supposed to be based off the intrinsic value, it wouldn't be anything close. The market value and the intrinsic value would likely be very similar.

Let's look at the more likely scenario, that market value equals intrinsic value.

Team(s)	(AFL) curve which over-values lower picks	Market curve
Academy team trading down	Gain significant advantage as able to turn a pick into picks with higher points values.	Unable to gain significant advantage. In the example, gain a tiny advantage.
Team trading up	Come out even.	Come out even.	Come out even in both scenarios.
Other teams with picks in 30s and 40s.	Disadvantaged, with the total disadvantage of all clubs equal to the value gained by academy team.	In example, tiny disadvantage equal to tiny advantage enjoyed by academy team.

Again, all the market curve is doing is stopping the transfer of value from the other teams with picks in the 30s and 40s to the Academy team trading down.

 

[eastfreo75]

This is really on the nerdy end of trading/draft stuff.

I get pretty frustrated with the AFL's points index as it doesn't reflect how clubs value picks. This is because the AFL based on it salary, which doesn't line up with the pick-for-pick trading market. It's basically garbage, and yet the entire industry still refers to it.

I took all the pick-for-pick trades over the years and calculated a points curve which more accurately reflects the value of picks relative to higher-lower picks. It's not perfect, but it's 100 times more accurate than the AFL's curve. If anything, it could probably be slightly steeper.

Feel free to use it or to provide feedback.


View attachment 1528842


View attachment 1528844
View attachment 1528846

You are awesome.

 

[Flarn Rarn]

Unless I'm reading it incorrectly it still has the same major flaw that the AFL's system does, which is that the points are additive (albeit the AFL created the system to value FS and academy bids, not trades). That means that you can very easily create examples where you trade a bunch of garbage picks for gold, which is not how clubs value trades. Clearly there needs to be a discount applied based on the number of picks you use. My thoughts are that you should have some scaling factor on the total points of the form a^(n-1) where 0 < a < 1 and n is the number of picks you use. The exponent is n-1 so that in the base case where you use one pick, it still has the same value as its value in the table.

For example, arbitrarily pick a to be 0.85 (this could be tuned, I just picked something sensible), here are some examples using your salary based curve that show the value of discounting applied to cases where you would otherwise be able to trade (in some cases junk picks) to get pick 1.

# Picks Used	Pick Combination	Total	Discounted Value	Discounted Pick Value
2	2+32	3,000	2,550	2+leftover points
3	16+17+18	3,077	2,223	3
5	27+28+29+30+31	3,268	1,706	6/7
10	42 : 52	3,101	718	26

Comments:

1. The scale factor of 0.85 is arbitrary but I'm looking to show that qualitatively it is better to do it this way.
2. There is clearly no trading crap and expecting to get pick 1 here. Even Trading pick 2 and a late second round sweetener only has marginally more value than pick 2 itself, albeit with the leftover change being less than pick 32 is worth.
3. The combination of 3 and 5 roughly equal picks look about right to my eye. Certainly doesn't get you pick 1, but I can see clubs potentially agreeing to both sides of the discounted value.
4. The combination of 10 third round picks is just the logical extreme - clearly a ridiculous example and maybe you could make a case that it'd be worth more, but a mid second rounder for a bag of junk picks is certainly more realistic than pick 1.

 

[Suma Magic]

Unless I'm reading it incorrectly it still has the same major flaw that the AFL's system does, which is that the points are additive (albeit the AFL created the system to value FS and academy bids, not trades). That means that you can very easily create examples where you trade a bunch of garbage picks for gold, which is not how clubs value trades. Clearly there needs to be a discount applied based on the number of picks you use. My thoughts are that you should have some scaling factor on the total points of the form a^(n-1) where 0 < a < 1 and n is the number of picks you use. The exponent is n-1 so that in the base case where you use one pick, it still has the same value as its value in the table.

For example, arbitrarily pick a to be 0.85 (this could be tuned, I just picked something sensible), here are some examples using your salary based curve that show the value of discounting applied to cases where you would otherwise be able to trade (in some cases junk picks) to get pick 1.

# Picks Used	Pick Combination	Total	Discounted Value	Discounted Pick Value
2	2+32	3,000	2,550	2+leftover points
3	16+17+18	3,077	2,223	3
5	27+28+29+30+31	3,268	1,706	6/7
10	42 : 52	3,101	718	26

Comments:

1. The scale factor of 0.85 is arbitrary but I'm looking to show that qualitatively it is better to do it this way.
2. There is clearly no trading crap and expecting to get pick 1 here. Even Trading pick 2 and a late second round sweetener only has marginally more value than pick 2 itself, albeit with the leftover change being less than pick 32 is worth.
3. The combination of 3 and 5 roughly equal picks look about right to my eye. Certainly doesn't get you pick 1, but I can see clubs potentially agreeing to both sides of the discounted value.
4. The combination of 10 third round picks is just the logical extreme - clearly a ridiculous example and maybe you could make a case that it'd be worth more, but a mid second rounder for a bag of junk picks is certainly more realistic than pick 1.

Or you could just make the curve steeper? Wouldn’t that achieve the same thing?

My argument is if the curve accurately represents pick for pick value then it’s the best curve to use.

 

[Flarn Rarn]

Or you could just make the curve steeper? Wouldn’t that achieve the same thing?

My argument is if the curve accurately represents pick for pick value then it’s the best curve to use.

I'm not sure there is such a curve - assuming there is pick for pick value you will always run into trouble when you weigh a premium pick versus a handful of crap. It might value intrisics about players picked at those spots fairly well, generally speaking of course, but it doesn't correctly value how clubs would perceive their trade value, which is a separate problem and the one you're trying to model.

Nevertheless, my own proposal is not perfect - you can quite easily run into the problem where you add sweeteners to a deal and the whole package loses value, which isn't correct either.

 

[Suma Magic]

I'm not sure there is such a curve - assuming there is pick for pick value you will always run into trouble when you weigh a premium pick versus a handful of crap. It might value intrisics about players picked at those spots fairly well, generally speaking of course, but it doesn't correctly value how clubs would perceive their trade value, which is a separate problem and the one you're trying to model.

Nevertheless, my own proposal is not perfect - you can quite easily run into the problem where you add sweeteners to a deal and the whole package loses value, which isn't correct either.

It’s easy to make a curve that overvalues high picks. So surely you can make one that accurately values them?