Is there any evidence these markets actually provide useful predictions?
Kalshi had Abbot solidly in the lead (polls were closer to final accuracy) until the minute enough votes came in for news orgs' forecasting models to tip the race to Lean D, and then plummeted to 0%.
On balance it seems like they provide little to no uniquely valuable information for anything that matters.
https://kalshi.com/markets/govpartytx/texas-governor/govpart...
Note that for the Texas governor, Kalshi had Abbott at 60+% on election day and he lost by a _million_ votes.
There will always be cases where everyone is surprised. If you have a useful predictor that means it incrementally helps improve your expectation. It doesn't eliminate surprises.
There are plenty of legit critiques of the prediction power of these markets, but they typically don't involve pointing at single observations. The overwhelming scholarly consensus is prediction markets are highly accurate relative to other forecasting methods, but have also significant biases. Such as long-shot bias, etc.
But those biases are small enough that they are only resolvable by looking at lots of independent at-bats. Real-world consensus was not that Abbott would get smoked to that extent, and so of course that consensus was exactly reflected in prices. If you read a lot of pre-election commentary (I was reading primarily Democrats) you would realize that people knew something was going to happen, but they really had no idea about the extent.
Furthermore, we don't learn a lot (anything really) about efficiency by looking at a large number of events that are highly correlated. And of course the whole blue wave was correlated events, so yes you will find a lot of "bad" prices, in terms of raw numbers. But taken together they are really one observation.
See this huge study on political bias: http://www.nber.org/papers/w35846.pdf (Spoiler alert, if there is one it's incredibly small.)