Forecasting newsletter #3/2025: Long march through the institutions

Highlights

• Manifold ending (a) cash markets, Kalshi slapped by regulators in Nevada

• Interactive Brokers now also offering election markets (a)

• Violet Hour looks at Where Would Good Forecasts Most Help AI Governance Efforts? (a).

Prediction markets and forecasting platforms

Metaculus is doing more things these days. They have a report on modelling AI by 2030 (a), a tutorial for setting up a Metaculus bot and a look at how bots are performing vs human forecasters (a).

Manifold is ending cash markets (a). Presumably due to regulatory pressure, though that's just a guess on my end. One way to model this is that by coming up with an alternative legal theory (sweepstakes are not regulated), they were able to do something that's illegal for a while until they came to the attention of regulators—or became afraid to.

manifold.love is back to life! Now disentangled from markets, so just a project from the adjacent community. Writeup by Sinclair here (a).

There is now a prediction market on Bitcoin. I don't really know how it works and don't vouch for its reliability. Markets are resolved by AI, lmfao.

ForecastEx (Interactive Brokers' prediction market offering) started offering (a) election markets. IB has over 2M users, so in some sense this is already one of the biggest prediction markets out there. A prediction market victory might in the end involve established actors adopting the technology instead of current upstarts becoming large actors.

Webull, a large, initially Chinese financial platform (a) is using Kalshi's markets

Polymarket is hiring for market creator, markets engineer and customer support roles. As shared in their Discord:

If you're interested, email jobs@polymarket.com with the subject line as the role you're applying for. In the body, tell us why you’re a great fit—share any relevant experience, projects, or work you’ve done related to the role, along with your resume.

Brown Forecasting (a) is running a college forecasting tournament with $2K in prices and "potential interviews with top firms" for the winners

Friend of the newsletter and widely read Spanish journalist Kiko Llaneras has a Metaculus tournament with $5K in prices.

Goldman Sachs looking at tariff risk on Polymarket

Research and articles

Jonny Spicer looks (a) at Kokotajlo's 2021 "What 2026 Looks Like", and finds it impressive how many things he got correctly.

The Violet Hour looks at Where Would Good Forecasts Most Help AI Governance Efforts? (a). This is a question I've pulled my own hairs about, and he seems to have some good thoughts.

Robin Hanson looks at (a) what appropriate spending on UFOs would be if they were real (very high), what the probability of them is according to readers (~10^-5), and what public odds one can find (1:150). He also gives his personal odds as >1%. However he isn't willing to bet about it (a), because he thinks correctly interpreting the evidence which we now have will take much longer.

My team at Sentinel writes weekly minutes covering possible precursors of large-scale risks. It's possible that at some point these would deliver immense value to you by letting you know about e.g., the next covid a bit beforehand. You can sign up here (a).

Richard Bruns condenses (a) lessons about cost-benefit analyses

Anthropic has (a) a pretty fake paper on forecasting rare behaviors. For large samples, the distribution might be different. For example, if an attacker uses the earlier inputs to tailor their outputs. The question "if such and such happens so and so often with X amount of compute, how often does it happen with 10X compute" is interesting and hard, and they don't answer it.

The Violence and Impacts Early Warning System (a) (VIEWS) is a competition of models to predict conflict . They have a neat dashboard (a)

Takes.fun (a) is a prediction market adjacent concept seeking to monetize takes. If people buy them enough, they eventually become "hot takes". The tokenomics don't work, but I thought it was a neat concept.

An overview (a) of studies looking at predicting replicability from the British Psychological Society.

Scotiabank does some geopolitical forecasting around Trump tariffs, which amounts to a somewhat sophisticated handwaving.

Advances in post-Bayesian methods (a). 1st Workshop 15–16 May 2025, London, UK.

Odds and ends

You can see the Biden presidency in the "market cap (a)" for USDC. It has now recovered, and is now at 56B. As a rough estimate, at ~4.5% interest rate, that pot is producing $2.5B flow in interest for Circle, the parent company. That's a lot of interest in keeping that ecosystem alive. There is power in numbers.

There is also some movement to regulate (a) stablecoins in Congress now.

Asterisk Magazine is paying $2K for articles forecasting effects of funding cuts.

Devans Mehta joins (a) Ethereum foundation as AI x Public Goods/Governance lead, will look into AI agents predicting whether a post will get community noted.

I've been going through Twitter fermi estimates as I write this newsletter, and commenting with my own estimates. The ones for this month start here. If any reader wants to take a moment and think about how many kilograms of potatoes the Heathrow airport uses each day, my answer is here.

Regulators

Can You Bet On the Super Bowl in California? Not Exactly—Try 'Trading' Instead. I feel that the title alone does carry the point across that prediction markets can be a shallow rebrand of sports betting, and that some platforms are pursuing that approach, since sports markets are not uniformly legal yet a profitable segment of the market.

The US Representative from Nevada—home of Las Vegas—is trying to get the CFTC to ban sports-related prediction markets (a) . The Nevada Gaming Control Board also issued a cease and desist order to Kalshi. I expect her to lose the fight at the federal level with Republicans in control, but win in Nevada . Here is Kalshi's answer, which takes a tale of entrepreneurship as regulatory compliance.

Malta court (a) rules it will not enforce Austrian player losses judgments, ruling that "EU Article 56 in the Treaty on the Functioning of the European Union (TFEU) gives provision for services (like online gambling) to be provided across EU states". This would be an interesting Trojan Horse mechanism to see betting legalized across the EU.

Election contracts are here to stay, says CFTC commissioner (a). And fancy law firm Wilmer Hale looks (a) at the first 30 days of the CFTC.

To me the high level story is that there is now a sympathetic administration in the US, but it’s still taking time to play moves such as regulating stablecoins, formalizing a regulatory framework for prediction markets, establishing whether and how the CFTC does have authority over these markets and how, and so on. There is a knob that you can turn to be friendly to prediction markets, but it takes time.

Overall I dislike the protagonism that regulators have been taking in the last several years of this newsletter; five years ago in the prediction market cyberpunk days of 2020 when I started this newsletter, regulators wasn’t even a section.

Belgium blocks (a) Polymarket over regulatory non-compliance.

News

Bloomberg has a tame hit piece over Pham, the new CFTC commissioner. Pham''s (or rather, the CFTC's) reply here (a)

Bloomberg on Polymarket bets on a new pope (a)

The histories of gambling and the papacy are long intertwined. Betting on papal elections was relatively common in the 16th century, until Pope Gregory XIV in 1591 formally banned Catholics from placing wagers on the election of a pope or on a pope’s tenure... (That edict was abrogated in 1918.

cf. Wikipedia on the history on gambling on papal conclaves (a).

A Swiss prediction market on a Federal Councillor position (a), covered in Swiss news here (a)

---

How do you identify small but fast-growing markets? Is it a good/bad idea to found a startup in a market that is no longer saturated but also has big players (such as social media?

sama on Sept 27, 2016 | parent | next [–]

Predicting the future is hard. I don't have a great general answer for this, other than to look for highly engaged users. IE, when the iPhone came out, users loved it and used it all the time. Even though the market was small in absolute size, you could be confident it was going to grow quickly.

It could be good or bad depending on details, but it's certainly not always bad.

---