Scott Alexander

The result is they've got tons of them, for better or worse.

you're going to see 1 in 100 upsets on tiny calci markets for as long as this regulatory equilibrium holds, even if nothing unusual is going on, simply because they're publishing hundreds, thousands, of markets per day.

End quote.

Scott writes, There's a saying that you can't con an honest man.

This isn't exactly true, but it's easier to con people who are playing in a what-words-will-Brian-Armstrong-say-today market than people who are trying to do something useful, and I have trouble feeling sorry for these people when Brian Armstrong says silly words.

Conditional markets.

A modest proposal.

Conditional markets, decision markets, are the strongest case for prediction markets potentially being revolutionary.

The idea is, you may want to base a decision, like which candidate to elect, on an outcome, like how they'll affect the economy.

So you make two markets.

If the Democrat gets elected, will the economy be good four years later?

Or if the Republican gets elected, will the economy be good four years later?

And if one market is higher than the other, then you've successfully forced everyone to settle on a canonical probability of which candidate will be better for the economy.

The fatal flaw is confounding by non-causal pathways.

For example, betters might reason, suppose for some extrinsic reason, let's say someone struck oil, the economy is very good from 2026 to 2028.

Then in 2028, people will feel better about Trump and are more likely to elect Vance.

And if the economy is very good from 2026 to 2028, then it's more likely to be very good from 2028 to 2032.

The oil is still there.

Therefore, we should bet up the Republicans to good market and bet down the Democrats to good market before we even think about whether Republicans or Democrats will do a better job with the economy.

Therefore, this can't be a good way to determine whether Republicans or Democrats will do a better job with the economy.