> There's no guarantee either of them have a utility function anywhere near your own.
Yep.
> Which leads me to conclude that the net information content of a stock trade is, fundamentally, zero.
It might not be if you have a bunch of trades, averaging over lots of people with different utility functions. Maybe. Depending on how average your utility function is.
In practice people end up with heuristics like "100 - age" and diversification out of stocks or hedging of their stocks to deal with the imperfect matchup between their utility function and the averaged one.
It's hard not to think of the whole thing as a house of cards sometimes.
> And yet prediction markets work
Sometimes they do. As long as everyone involved has broadly the same utility function: that of maximizing their money above all else. If enough people, or more properly enough monetary units, come in with a weird utility function (e.g. valuing a particular prediction more than their money), you get prediction market failures.