(@ Manifold founders: if you think this is worth chasing, I will happily spend at least 20 hours working for/with you to come up with a good market design, and trying to get subsidy-funding.)
At Manifest, Robin Hanson asserted:
There are trillions of dollars of value to be unlocked by better institutional decision-making, e.g. decision markets, e.g. markets on which of various job applicants would be best to hire.
As far as he knows, nobody has even tried.
Manifold is very open to experimentation and very ideologically aligned; I would like to convince them to try this at least once.
Clarifications:
To satisfy this market's condition "Manifold runs a hiring-decision market," the market doesn't need to determine the hiring decision, as with futarchy; it just needs to happen, and the PDF needs to be looked at by the people making the hiring decision.
The decision market does not need to be on Manifold.
On 2025-10-01, I'll ask one of the Manifold big-shots how many new-hire decision markets they ran. This market resolves:
N/A if Manifold ran 0 new-hire decision markets (or if I don't get an answer)
NO if they ran 1-2
YES if they ran 3+
1,000
1.00Hi, yes, I think we do want to try this for our next hire. (Which probably won't be until next year.)
I haven't thought that hard about this, but was thinking of several markets of the form, "If we hire this person, what will be their evaluation in 1 year?"
And maybe it's multiple choice with a 3 point scale: 0 (fired), 1 (great), 2 (outstanding). We'd resolve N/A for any candidates we didn't hire. And we'd add a large amount of liquidity. Probably we'd make the markets unlisted, but allow public bets. (It's possible we could make it fully public, but I wonder if the candidates would mind that haha.)
Multiple-choice sounds like a good plan!
The major difficulty I see is conflict-of-interest, since the people doing the year-end evaluation surely have stakes in the market. Two possible solutions I see:
1. Restrict bets to be small enough that people's honor will greatly outweigh their financial interests. (Though obviously this also means various biases like "entertainment" or "social desirability" might also outweigh their financial interests.)
2. Make bets blind, so bettors don't know their stakes. (You can never fully blind the bettors -- e.g. if Austin thinks the candidate will suck but everybody else thinks the candidate will be great and thinks Austin's being irrational, then Austin knows he'll end up with lots of shares of Fired -- but you might be able to mitigate this. E.g.: make a black-box centralized mechanism with an AMM; privately construct your individual order-books and feed them into the box; ask the box whether the AMM's final PDF "looks good" according to some predicate (e.g. "P(Fired) < 0.1 and P(Excellent) > 0.3"). Or, to retain more human decision-making power at the cost of leaking more info, look at the AMM's final PDF.) (There's... probably?... prior art designing systems like this, but I can't find it.)