outline:

• this post describes how i think the prediction market ecosystem should set up its incentives to develop into a good epistemic institution.
• markets, in general, need uninformed flow
• prediction markets are a kind of market
• so, prediction markets need uninformed flow
• at present, prediction markets have found uninformed flow in providing entertainment value
• but this has substantial downsides:
  • this means that only prediction markets that are on entertaining subjects get lots of volume. many markets that are extremely important aren’t very engaging/fun — so this doesn’t solve finding uninformed flow for prediction markets on important topics.
  • often, the entertainment value is actually just “addiction,” and is net-bad for the world
  • there is much less total entertainment value to provide than, e.g. “hedging oil price volatility risks” — which means there’s a pretty hard cap on the total amount of uninformed flow that could come in
  • order flow from entertainment is not long-term sustainable.
  • the environment produced by incentives driven by providing entertainment value make me skeptical that government would (or should) continue to be as friendly to prediction markets as they’ve started to be over the last few years.
• however, i see two main possible routes out of the scarcity of uninformed flow we currently find prediction markets in:
  • direct subsidies for information value to a consumer
    • presently, directly subsidies on prediction markets are inordinately expensive, and so most institutions would just be better off doing something else.
    • that’s the case because uninformed flow provides some reward for informed, sharp betters to provide liquidity and thereby enable price discovery — and the time of informed, sharp betters is extremely expensive.
    • but if AIs become almost as good — or better! — than informed, sharp betters, then the cost of an informed, sharp better drops to the cost of inference (plus maybe some rent-seeking? though this should pretty soon drop to commodity levels)
    • in that case, the cost of directly subsidizing a prediction market to encourage a swarm of well-informed AIs to bet would become much, much cheaper.
    • downsides:
      • it’s possible AIs just won’t get good enough at forecasting
      • alternatively, it’s possible that a world in which they do implies that we’ve got ASI, and the world looks crazily different to the point where thinking about the prediction market landscape is senseless
  • more hedging, more insurance
    • this is the same fundamental case as the example case of airlines hedging risk of oil prices shooting up by entering oil futures markets expecting to lose money / pay some premium in exchange for eliminate a huge tail risk.
    • many individuals & institutions currently take on huge tail risks, often knowingly
    • at present, individuals generally can only do hedge those huge tail risks by buying insurance (e.g. fire insurance, car insurance)
    • but the insurance industry has a number of drawbacks to prediction markets:
      • they’re a bizarre and fucked up marketplace, so prices are probably not as efficient as they could be
      • insurance companies are a huge a hassle to deal with
      • most importantly: they can’t handle more idiosyncratic hedging needs (e.g. “i want to hedge my career against the electrical engineering jobs market going down 10-30 years from now”). so for many, many tail risks that people and institutions are in-fact taking on, they do not presently have the capacity to hedge that risk.
    • the main advantage insurance companies have is that the vast majority of consumers would be totally incapable of handling the complexity of hedging their risks adequately on prediction markets — so they present them with an also-complex-but-much-less-complex-than-a-prediction-market package deal.
    • importantly, the fact that this doesn’t already exist is mostly a product issue: some enterprising startup could develop e.g. “career insurance,” then provide liquidity between a prediction market on a variety of metrics of various careers over time and university students who want to specialize more deeply in some career without taking on excess risk.
    • downsides:
      • sure seems like the market would incentivize this. why doesn’t it exist already? (i talked to some higher-ups at kalshi about this a few years back. it was on their mind, and they seemed keen to try it — but i don’t know whether they gave it a serious shot (nor if they did take that shot, why it failed).)
      • again, this might just still not end up incentivizing prediction markets on important topics. (however, even if it doesn’t, spreading risk out more effectively for individuals and institutions is itself really, really great.)

---

draft:

I have a vision of prediction markets as a gleaming institution of canonical, trusted truth. One that’s used by influential institutions and by Bob and Jill down the street to improve the quality and outcomes of their decisions. And at the same time, opening up new ways of strategically betting on outcomes, one that’s easier for individuals to take stock of (no pun intended).

A few years ago, this was the dominant vision in the prediction market landscape. But prediction markets now look more like Vegas 2.0: the majority of volume is from sports betting or other

---

Related:

• that ACX article on AIs betting on prediction markets → the cost of subsidizing prediction markets goes way down

• haven’t read, but — the way that vitalik talks about prediction markets in this:
• haven’t read, but —
• from Nuno: