Markets that make decisions
I'm Kollan House, co-founder of MetaDAO. We're pioneering futarchy in the Solana ecosystem — we were the first to build a working futarchy mechanism.[2] Futarchy, if nobody knows, is Robin Hanson's idea from the late nineties:[3] there's all this information out there, and markets seem to distill it — so what if we used markets to drive decision-making in its place? I'm going to quickly walk through what a decision market is, and then get to the part that matters for today: information flow.
What we actually decide this way
A decision market is a market that directs an outcome. What outcomes? We use it for on-chain execution — token issuance, transfers of funds, updating parameters on the DAO. And we use it for executing legal agreements. I just had a proposal pass this year for market-making arrangements: it gave me the authority to go out on behalf of the Marshall Islands LLC and engage those parties.[4] Incorporation. A grant to Robin Hanson with GMU to research decision markets against traditional markets.[5] Approval of initiatives — budgets, quarterly plans, new business ventures. We use it for anything and everything, and we built a legal wrapper to make sure it holds up.
The mechanics: fork, trade, execute
Practically: we trade the price of yes and no — should we take this action or not? The canonical example is firing the CEO. Ginni Rometty at IBM presided over twenty-some consecutive quarters of revenue decline;[6] a decision market for IBM asks whether the stock would trade higher with her fired or kept, and if the fired-price is higher — that's what you do.
How we do it: split the token into two conditional versions and trade for three days — we started at ten, which is a long time to hold a market open. Then take the time-weighted average price of each market,[7] and if the delta clears the threshold, the on-chain instruction executes — which is what lets us do everything downstream. Think of it as a fork: only one side is canonical and redeemable for the underlying asset.
The information game
How does this relate to today's topic? It's all information flows. We want people to reveal what they know into these markets — that's how the decisions get better. But how do you get them to do it? Do you need deep liquidity to make it worth their while? Honest answer: we don't know. We've traded with really small liquidity and with larger liquidity, and we don't have a good answer.
“This is all information flows. We want people to reveal this information in these markets. Obviously, that’s how we get better decision. But how do we get them to do that?”
Counterintuitively, volume is not a good proxy for how good a decision is — outside of measuring opinion and conflict between parties. And unlike Polymarket or other prediction markets, there's no settlement event: the resolution is in the token price itself, and the payout might be a year or more away. So the design levers we have are things like different liquidity curves at different times, and paying for participation. The unglamorous question on the slide is the real one: who's paying for this?
When revealing hurts the revealer
The advantages are the disadvantages. These proposals are public information. What happens when that information feeds a competitor and damages your business — how do you get that decision made on a public market? And what happens when the revealed information creates an environment that no longer benefits the party who revealed it?
“What happens when this information could feed into a competitor and, you know, destroy your business? How do you get a decision to be made there on a public market?”
We hit this ourselves. When we went out to sell tokens and raise money, parties would show up and identify themselves — "hey, we're Pantera" — and the attention alone would drive the token price up through the cap they were willing to pay. The execution of the sale ended up outside the range of the proposal, and we couldn't take the money.[8] That was a troubled time. And on the legal side: we hired somebody through a decision market,[9] but firing on a pure decision market isn't really applicable — there are a lot of legal footguns there.
Known failure modes, candidate fixes
We know this isn't the final form. We use TWAP, and we know it's not great — there's a scenario where you simply can't impart your information into the market, because near the end of the window the price would have to move toward infinity to shift the average. Say the proposal transfers me ten million dollars and there's an hour left: your late trades can't move the time-weighted average enough to stop it.[10] We've said volume isn't a great proxy, but maybe volume plus a time-weighted price is. Robin suggested a statistical Brownian-motion approach; I have no idea how to get that on-chain and oracle-ready — if you do, I'd love to talk.[11]
“We use TWAP. We know it’s not great. There’s a scenario where you could not impart information in the environment because the price would have to move to infinity to impact it.”
Then there's structure. When a team wants to be exited, acquired, or liquidated, we currently trade those proposals over three days. Maybe what should be traded in twenty days shouldn't be traded in three — the information takes that long to arrive. And proposal crafting: for our OTC deals we no longer name the parties at all. We publish the minimum terms we'll accept and what we won't, and parties submit independently. The market has to trust us to execute accordingly — that's the price of not leaking the counterparty.
If any of this sounds interesting — that's the website and my contact. Happy to answer questions.
Q&A
audience questions, lightly condensed
Askers are unnamed on tape; one exchange involves the previous speaker. Reconstruction notes in the footnotes.[12]
audience → previous speaker
In your presentation you had some hesitation around objective functions — whether you can even define them correctly — and you talked about the difficulty of surfacing information. Markets in general, and decision markets in particular, are a mechanism for surfacing information, but they need some objective function to surface it against. How do you see the connection between your line of thinking and markets?
previous speaker
Prediction markets are your wheelhouse — I'd say they're one implementation of collective intelligence. What we're trying to do is the complementary side: provide an explanatory model for why you get the results you can measure with prediction markets. If we're making progress on our model, then in principle we'd be able to predict what would happen in the prediction markets for an actual company.[13]
kollan
Right — and note we don't do settlement. We don't ask whether the event is going to occur. We ask: if this event were to occur, would it drive the price of the asset up or down? It's distilling all that information and mapping it into a much broader market environment, with sophisticated participants whose models all derive into the price. Why does a company's share price go up? What drives it? There's a ton of information behind that.
audience
Have you done clustering studies on the people who screen companies?
kollan
It's so small right now. Our largest market traded with about fifteen hundred addresses; a typical market trades with forty. The liquidity and the incentive to drive profit isn't necessarily there yet. We'd like to get there.
audience
Hanson's mechanism does well where there's repeated expertise over time — people reliably feeding good information into markets get selected. But it's also meant to surface the best information at a single point in time. For something like a specific funding round,[14] the relevant expertise might not correlate with who accumulates resources over repeated markets. What does a mechanism look like that reconciles those two?
kollan
Very good question, and the simple answer is: we don't know. This is the first time it's ever been implemented and tried; we have opinions, and we use asset-based pricing. Robin's original work discusses KPI- and metrics-based markets, running multiple markets at once — even markets over what number should go into a proposal, so a twenty-percent allocation gets set by the sophisticated participants themselves. My view: if the markets get liquid enough and the industry's capitalization gets large enough, the experts show up naturally. If the DEX is going to buy a zoo with twenty zebras, somebody who knows the value of zebras says "I can influence this decision" — and profits by being right. If the incentives and alignment are there, these problems start solving themselves.
audience
Can you talk about adversarial outcomes you've seen? Proposals execute automatically on-chain — have you seen proposals designed to take advantage of that, and can the research community help prevent bad outcomes?
kollan
As the founder: I think the second or third proposal ever was to burn all the tokens in the treasury. It passed. I didn't necessarily agree with it, but we proceeded.[15] And right now there is an ongoing proposal to drain the treasury of one of our organizations — people can create these proposals; there's a cost to it. What I like about these markets is that attacking isn't a one-time expense like a fifty-one-percent attack: coming back tomorrow costs you the capital deployment and the time, again. The attackers stand to win about one and a half million dollars if they execute. But the market capitalization of the token is far larger, and we've seen some of these markets trade total volume several times that capitalization[16] — there are active participants who like defending these things. Spam, though — I wish spam were solved in crypto. It isn't. None of this guarantees great answers.
“I think the second or third proposal was to burn all the tokens in the treasury. So that passed. I didn’t necessarily agree with it.”
Glossary & further reading
Every dotted-underline term above, in full, with sources.
futarchy
See: Hanson, "Shall We Vote on Values, But Bet on Beliefs?".
decision market vs. prediction market
TWAP, and how it fails here
conditional tokens / the fork
structured proposals ("MetaDAO 2.0")