Product questions

What are we trying to conduct futarchy on?

AI markets: Timelines. Opus release date. Sb1047 results.

Upcoming election: who will win mayor? (just a better UX might be good). Then: #housing for each mayor? #crimes? Etc

Broader US governance? (GDP, etc)

all should be answerable independent of the probability that the things happens (person elected, bill passes, etc)

Motivation

SB1047 — trying to get an understanding of what happens if this thing passes

Good for elections, but goes beyond — most things happen outside of election

Ballot props: laws you can vote on

Joe Biden stepping down as a great example of prediction markets helping Nate and his friends

• Information quality
• The more local you get, the drastically lower information quality you get
• Federal senators have access to great quality
• State senators have pretty mediocre quality; and many are not clear thinkers themselves.
• Problem is not info quality, but determining which to believe
• Even at state level, it’s tricky if not smart
• hard to get them to

What’s the goal?

• Fun viral app that showcases futarchy markets, in advance of the upcoming (local or US) elections?
• Serious dashboard that informs govt decisionmakers on the projected impacts of their proposed policies?
• Tool that helps orgs (startups, communities, eventually governments) decide what policies to implement?

Who’s the target audience?

• Uninformed voters vs decisionmakers
• Local/SF vs national govt vs global
• Traders vs readers vs creators

Technical questions

What is the market mechanism?

• Numeric markets, futures, perp swaps
• Date markets
• Markets denominated in other coins
• Full exchange capabilities (leverage, margin)?
• Crypto might have the scaffolding for this now? Idk
• Incremental payouts (dividends) for continued alignment? Self resolving? Think of stocks...
• Hook: "GDP stocks" - trade on GDP. Then: counterfactual stuff

What is the financial layer?

(is it real money?)

1. Manifold
1. mana
2. sweeps?
2. Roll our own database
3. Crypto…?

What’s our tech stack?

My default tech stack for a new web app today looks like:

• NextJS
• Vercel
• Tailwind
• Supabase
• Drizzle?
• bun

Though I’m also down to play with other things (Supabase is probably the thing I’d most like to replace, also NextJS App Router)

Existing problems with futarchy

• “Not enough traders” problem
• (aka: not enough liquidity, fragmentation)
• Could solve with consolidation; arbitrage ability; more bots/LLM trading agents
• “Insufficient bits to steer by” problem
• Resolution risk/”Creating resolution criteria is hard” problem
• Condensing your worldview down to specific numeric quantities is hard
• bet sizing is hard (?)
• “long term markets” problem. Solutions:
• pay out shares in eg S&P 500 or some synthetic thing, instead of USD
• provide interest
• provide leverage
• pay out dividends
• I like this because it provides more dopamine hits, “wins”
• encourage exits before the market resolves — aka more day trading, less buy-and-hold?
• More liquidity helps with this. Maybe also UX/UI tweaks, eg limit sells

Discussion

• Kind of want to just trade the gap
• Eg gap on ceasefires is 21% but nate thinks it should be 30%
• Want to express sentiment on the difference
• 75 vs 78 is not that large, most of this is a bet of what inflation will be. Just want to bet on variance
• [a] Can solve with leverage?
• Hardest part is setting up financial product

• Eg gap on ceasefires is 21% but nate thinks it should be 30%
• So Nate buys
• default
• lower for Harris 39% ⇒ 35%
• higher for Trump. 60% ⇒ 65%
• Then Trump wins
• Harris trade unwinds
• Trump market became the

liquidity hack

are the markets even accurate enough to be meaningful?

probably!*†‡

this is the range outside which you can make a >5.41% return in expectation, if you believe the true probability is outside that range

notice that lots of the ranges don’t even overlap! the current probabilities only need to be about 10 points apart to be quite meaningful*

things that increase/decrease this range:

• fees are absolutely brutal. if they were 1% or 0% it would be way, way tighter. basically linear relationship with uncertainty window size
• liquidity hack helps (but out of scope), see below for more notes
• extreme P(A), the event being conditioned on, is brutal because you probably get no returns for being right on the unlikely half. fortunate that this election is close, but I’m not sure this will generalize well to other questions without a better market structure

*what about risk though

the risk-free rate is not the actual hurdle rate — there’s still plenty of risk even if you’re right

I did not try to compute the risk-adjusted returns or Sharpe ratios or anything, but that would be the proper way to do this

some reasons to be less than massively worried about this:

• the risk can be hedged a lot using the delta neutral or P(A)-normalized strategies in the spreadsheet. still not perfect but similar returns to the pure bets without nearly as much exposure to P(X) and P(A)
• you can sell your shares for profit as soon as A resolves, you don’t have to wait for X to resolve (much later), so lost interest is less concerning

†these are mana markets

extremely bad thing for accuracy, because the manipulation incentive is the same while the profit incentive to fix it is way weaker

might be a dealbreaker but is very easy to fix. already messaged Steven, sounds likely to happen

‡also manifold’s liquidity sucks

a $2 (200 mana) bet moves the market a whole percentage point

oof. very hard to make any real claims given that

however:

• with arbing between markets you actually have 4+ pools of liquidity all tied together, which helps some between trades (but still bad UX for each trader)
• markets with the most attention do get some limit orders in practice, which helps
• we should set some limit orders ourselves before launching, I’m sure the launch will cause a lot of noisy variance we can profit from while keeping the numbers more reasonable

but it’s still pretty bad. I would love to be able to say “you can make 5% annualized returns on $1k if you think the true number is outside this range”, and I don’t think we’ll be able to say that for more than $10

conclusion

I think we should do this! The “make %5 annualized returns if you think it’s outside this range” claim is compelling enough even if it’s on small $ amts, and this way of visualizing uncertainty is novel and useful. There actually are many issues where the ranges don’t overlap, and the ones where they do are informative as well

should probably include some big disclaimers though:

• ‘this is an alpha release and much less accurate than the main markets bc low liquidity (for now)’
• we should maybe not update the page automatically, so we can double check for anything too crazy before we publish it as fact (e.g. large moves driven by one whale)

side notes (out of scope)

swaps/perps based system

might be more accurate / better in some ways. didn’t finish modeling it. out of scope

liquidity hack basically cuts uncertainty range in half

but see section above to make sure it actually works?

either way, seems unlikely that manifold will implement it, and neither will we in the next two weeks

although, wait a second, is this the same as just having a Dependent Multiple Choice type question? I think so?? TODO: update thoughts based on this, although doesn’t change the conclusion

I’ll probably write a guest blog post about this after the election

it was pretty interesting. also want to explore the swaps/perps setup

next steps

I think this is worth doing an MVP of. It’s interesting and useful and shouldn’t be too much work

• top pri: nice electionbettingodds-but-pretty site summarizing the conditional markets
• I think this is all we need
• spent some more time thinking about the UI options and I think a bunch of cards like this is probably the way to go

• stretch: arb bot to keep the markets in line
• stretch but I think high pri because a) we assume they’re always in line, and b) this is a great arb opportunity

Exploring visualizations

• 🪡futarchy.dev + yield.sh