The problem with Bayesian is that it's often over confident in the model. Error bars don't take this into account and people over rely on error bars - especially with complex domains.

Then there are the intentional and unintentional sampling biases in the polls that have to be taken into account. Then there are known human effects that add additional biases. E.g. the shy voter effect (e.g. Torres and Brexit), silent majorities (Nixon), momentum, dam breaking (Reagan), enthusiasm gap (Obama), the giant fk you to the establishment vote, pending indictments etc. In my view with the election being so emotional this year these effects are big enough to swing it to Trump making a poll based model wildly inaccurate.

I've taken an actual position in the election. $4k on Trump to win at 15% odds made just after the tapes. At the time I felt Trump had at least 50% chance so the bet was positive expected return for me. I figured the public would get over it. My Bayesian friends had his odds at 2%. Today I consider Trump to be at least 70% with the whole FBI inditement saga picking up steam.

In addition, the betting market behavior mirrors Brexit. A few very big bets on the status quo and many small bets against. It appears as if once again deep pocket punters are intentionality trying to manipulate the odds on the illiquid market in order to send a message that effects the vastly larger financial markets. So I think the odds are that there is some free money there.