-
[...] Outperforming Taiwan’s polls shouldn’t be hard. They’re notoriously bad as a forecast of election outcomes. In late 2006, for example, many media polls underrated the pro-independence party’s support – a recurring problem. Taiwan’s prediction markets did a much better job of estimating vote shares (the island’s two markets both called the Kaohsiung mayoral election wrong, but that contest was a statistical dead heat). “Most opinion polls usually have 20 to 30% ‘no answer’,” said Lin Jih-wen, director of the Center for Prediction Markets. “We don’t have missing data or a sampling bias, that’s our strength.”
The market has now picked the strong likelihood of victory by the China-friendly candidate Ma Ying-jeou. Time will tell if it’s got the right guy. But even if it doesn’t, the markets’ enthusiastic reception shows how Asia – like the US and Europe – has embraced such markets as a powerful fortune-telling tool.
A “powerful fortune-telling tool”? Jesus. 
“Outperforming” the Taiwanese advanced indicators which they are feeding on? Humm…
[Mike Giberson will write a comment, below, reminding me of Prof Koleman Strumpf's work (PDF file) showing that the historical prediction markets were accurate enough, even though the scientific polls were not invented yet. Yes, I know of that, Mike, but I still don't get whether it's a puzzle or a mystery. ... Do you?]
I don’t think it is puzzling that a prediction market could outperform polls. Polls provide useful data, certainly. But some traders will take the data and analyze it, read more deeply the news reports, talk to friends and neighbors, and then trade based upon some complicated mix of all of the information.
Polls are part of the data that informed traders will feed on, but they aren’t the whole meal.
I’d say scientific polls are 90% of their "mix". Is there a way to quantify that scientifically?… other than with a poll, of course.