On 7/2/26 2:36 PM, Salad via agora-discussion wrote:
On 01/07/2026 22:15, Mischief via agora-discussion wrote:
The price you end up actually paying is tedious to work out by hand -- and I've already automated the calculation so I don't have to -- but conceptually it's the generalization of a second price auction. In essence, you're paying the value you're denying others by winning

I've tried implementing these calculations myself to glean some understanding of the underlying auctioning model, but I fear that naively checking all possible combinations of bids for their combined utility does not scale to hundreds of bids.

I'd appreciate if you were able to provide a way for us lowly Agorans to experiment with, and perhaps even understand, the machinery of this auction.

Here's how I set it up; I imagine there are other ways to get to the solution.

I've done a lot with linear programming, so I view this as a mixed integer problem. If there are C combinations of possible baskets (e.g., 31 in this case) and B bidders, set up C*B binary variables. For each bidder, at most one of eir variables can be one. You also constrain it so that each asset is exactly sold. (For example, if there were 3 Ferret stamps, then each variable's coefficient for the Ferret stamp constraint would be 0, 1, 2, or 3, depending on the corresponding combination.) The objective function to maximize is the sum of bid value * bid binary variable.

The price paid by bidder X is the difference of two values. The first value is: what is the objective value if you rerun the auction without bidder X? The second value: what is the value that the non-X bidders received in the actual auction?

--
Mischief
Collar, Collector, Executor
Hat: steampunk hat

Reply via email to