Lula won Brazil's election, despite prediction markets giving him as little as a 16.5% chance of victory. According to @Dan_Zmann's analysis, that was the lowest pre-result price for an eventual winner across 34 presidential elections on Polymarket since 2020, using the same comparison method. But there's another side to this story that deserves more attention. Who was actually allowed to trade? In April, Brazil banned political event markets, restricting access to @Polymarket and @Kalshi. 27 prediction-market websites had been blocked following government requests by September. And @RPeTeixeira at @Folha reported that average daily transactions on Polymarket's Brazilian election market had fallen by 89.6% following the restrictions. Both companies told Folha that excluding Brazilian participants was affecting the quality of their forecasts. It's an interesting problem for the supposed wisdom of crowds. How useful is that wisdom when people closest to an election face restrictions on taking part? And although more than $200 million had been traded on the election by Sunday evening, that cumulative figure doesn't tell us how much money was actually available to buy or sell at any given price. A market can have enormous trading volume and still be moved sharply by comparatively small amounts of money. There's also the question of how these odds were being used politically. Jeniffer Mendonça at @nucleojor uncovered at least ten Brazilian election candidates, including Flávio Bolsonaro, sharing prediction-market odds as evidence of electoral support. One politician even paid R$1,500-R$2,000 to promote an advert featuring Polymarket and Kalshi probabilities. It reportedly generated between 250,000 and 300,000 impressions. An 80% chance of winning doesn't mean 80% of voters support you, but that distinction can get rather lost in a campaign graphic. And there's more. @analuizaalbuq at @Folha uncovered at least eleven foreign X accounts publishing paid promotional posts about Bolsonaro's rising Polymarket odds. These included former US congressman @mattgaetz and influencer @BehizyTweets. The posts were labelled as paid partnerships, but didn't identify who was financing them. Bolsonaro's campaign denied knowing about the adverts. @tainalon at @aosfatos subsequently explored the wider network behind this activity. Her reporting identified 220 pieces from the media outlet @laderechadiario that used prediction-market odds to support political narratives across Latin America, including in Brazil, Colombia and Peru. That's quite an interesting development. We're reaching a point where prediction-market probabilities are being used as political marketing material, sometimes by people who have nothing to do with the platforms themselves. None of this proves that anybody manipulated the market prices, let alone the election results. But it raises questions about who influences market prices, who can participate, and what happens when those prices are presented to millions of voters as though they're opinion polls. And we haven't got long before the next test. On 3 November, Americans vote in the US midterms. Millions will be watching @Polymarket and @Kalshi again. I'd like to see the market prices, polling forecasts, and timestamps saved beforehand, then compared properly once the results come in. I'd also like to see more discussion of liquidity, who was trading, and whether prices moved before or after new information became public. There's already some excellent work being done by traders, researchers, journalists, and sceptics asking difficult questions. I'd like to see more of it. I want prediction markets to succeed. I want to understand when their prices tell us something useful, when they're too confident, and whether the people with the best information are actually able to participate. I also want critics to recognise when markets respond quickly to new information and demonstrate their value. And when we get to the US midterms, I'd rather compare what people actually predicted before the results than spend another week watching everyone cherry-pick the charts that support their existing beliefs. Prediction markets are far too interesting to reduce to an endless argument between people who think they're infallible and people who think they're nonsense. Daddy chill... 4/4

The PNAS study made me more optimistic about collective forecasting. But I’m still curious which ingredients really do the work: money at risk, the people involved, the questions, or how quickly forecasts change. @PTetlock, @robinhanson, what would you test next? 16/16