It's cool to see how the Pacers engineered the protection to be as close to a coin flip as possible. It seems like they would want to not land 5-9 because that means they get a top 4 pick (super valuable this year in particular) while only giving up a value-nebulous 2031 pick. I like Zubac to fit in their late-2020s window, but I think the deal could age poorly for them if the coin flips in LAC's favor.
The time discounting thing got me thinking of a possible way to do it empirically. Maybe you could blend the work you cited from a few months ago into this? Since in 5 years' time every team's standing prediction falls towards the middle (15), I'd imagine the value of picks over time would fall at a rate higher than 3% or 5% since the uncertainty seems to increase quite a bit in just a few years. Though like you mentioned, the liquidity of those picks could up their value as a result of all that, and I'm not sure how you could factor that in empirically. Maybe look at how freely teams have been willing to trade picks in the past based on year? E.g., how liberal do teams tend to be about trading picks in year n+4 vs year n+1, potentially factoring in the value involved in those trades (EPM?).
Lastly, this all got me thinking about the current lottery system. This piece serves as another reminder of the precipitous decline in value after the first handful of picks, which makes me wish the lottery wasn't as large as 14 teams. I feel like a system which includes only the bottom 10 at maximum (so no play-in teams) with flat odds could work. That way you're either fighting for the playoffs or you get a top 10 pick at random. Then since the lottery odds would be uniform, tanking wouldn't be as incentivized (any weight favoring top picks will always incentivize tanking). But a fan can dream.
In order to accurately estimate the value of a future pick, I don't think you can run a model that uses every team as the sample. I think you have to limit your sample to *teams that don't have control over their draft pick*
I think that sample would both change the expected placement in the standings and the distribution.
Yeah, I considered something similar. Teams without their own pick in a given year have different incentives than those that do — no reason to tank, more likely in win-now mode, etc. I think the hurdles there are (1) putting that historical data together accurately over a long enough time horizon, which would be an interesting project on its own, and (2) whether that sample ends up being big enough to give us sufficient signal over noise.
Right now the future estimation is trying to answer “where do teams at rank X end up N years later?” and I’d picture implementing the above as an adjustment on top of that, rather than limiting the entire sample itself. But curious if you’re thinking of it differently?
I would imagine that most unprotected picks are sufficient N years away where X current rank doesn't make much difference. So I think a better estimate at that point is what the expected pick rank given up from an unprotected pick traded 3+ years in the future. I don't know what that sample is though.
As well, I imagine that those traded unprotected picks are less likely to be at the very top end (where, as you noted, a disproportionate amount of value is) because it's very difficult to accidentally out-tank a team if they're trying to lose.
It's cool to see how the Pacers engineered the protection to be as close to a coin flip as possible. It seems like they would want to not land 5-9 because that means they get a top 4 pick (super valuable this year in particular) while only giving up a value-nebulous 2031 pick. I like Zubac to fit in their late-2020s window, but I think the deal could age poorly for them if the coin flips in LAC's favor.
The time discounting thing got me thinking of a possible way to do it empirically. Maybe you could blend the work you cited from a few months ago into this? Since in 5 years' time every team's standing prediction falls towards the middle (15), I'd imagine the value of picks over time would fall at a rate higher than 3% or 5% since the uncertainty seems to increase quite a bit in just a few years. Though like you mentioned, the liquidity of those picks could up their value as a result of all that, and I'm not sure how you could factor that in empirically. Maybe look at how freely teams have been willing to trade picks in the past based on year? E.g., how liberal do teams tend to be about trading picks in year n+4 vs year n+1, potentially factoring in the value involved in those trades (EPM?).
Lastly, this all got me thinking about the current lottery system. This piece serves as another reminder of the precipitous decline in value after the first handful of picks, which makes me wish the lottery wasn't as large as 14 teams. I feel like a system which includes only the bottom 10 at maximum (so no play-in teams) with flat odds could work. That way you're either fighting for the playoffs or you get a top 10 pick at random. Then since the lottery odds would be uniform, tanking wouldn't be as incentivized (any weight favoring top picks will always incentivize tanking). But a fan can dream.
Thanks, that’s a really good idea re: looking at past pick transactions. Added to my list.
In order to accurately estimate the value of a future pick, I don't think you can run a model that uses every team as the sample. I think you have to limit your sample to *teams that don't have control over their draft pick*
I think that sample would both change the expected placement in the standings and the distribution.
Yeah, I considered something similar. Teams without their own pick in a given year have different incentives than those that do — no reason to tank, more likely in win-now mode, etc. I think the hurdles there are (1) putting that historical data together accurately over a long enough time horizon, which would be an interesting project on its own, and (2) whether that sample ends up being big enough to give us sufficient signal over noise.
Right now the future estimation is trying to answer “where do teams at rank X end up N years later?” and I’d picture implementing the above as an adjustment on top of that, rather than limiting the entire sample itself. But curious if you’re thinking of it differently?
I would imagine that most unprotected picks are sufficient N years away where X current rank doesn't make much difference. So I think a better estimate at that point is what the expected pick rank given up from an unprotected pick traded 3+ years in the future. I don't know what that sample is though.
As well, I imagine that those traded unprotected picks are less likely to be at the very top end (where, as you noted, a disproportionate amount of value is) because it's very difficult to accidentally out-tank a team if they're trying to lose.