Gambling odds best predictor of presidential elections?

You hear some people say that gambling odds are better at predicting elections than are polls, for various seemingly reasonable reasons. But how well do they do, really?

While I don’t know what the general pattern is, a notable failure was the Brexit vote. As the polls closed, the gambling odds were showing a 94% probability that “stay” would win. As this article notes, this was probably because the odds were tilted by big bets from London, which favored a stay vote.

Apparently gamblers are pretty good with regard to US presidential elections according to this.

“Gamblers’ success in this arena is nothing new. In presidential races beginning in 1896, the New York Times, Sun, and World provided daily betting quotes. The papers’ sources were bookies who had agents at every stump and whistle-stop to gather intel and quantify popular sentiment. Between 1884 and 1940, the bettors erred on just one of sixteen elections, Wilson’s 1916 upset of Hughes.”

Gambling markets tend to be more accurate the more total money is bet. That the Brexit vote was a complete surprise was shown on the world’s biggest gambling market — foreign exchange trading.

I check Rothschild’s summary page frequently. GOP White House is now shown as 29% at Betfair and by Rothschild. What is Hypermind? When I clicked their website it seemed to be subscriber-only. They show the highest GOP chance, though down from 45% recently.

Rothschild’s page shows Presidential chance by state (where does he get this data?) Here are the states where underdog has 11% chance or more:

Virginia        89 %            
Wisconsin       88 %
Pennsylvania    80 %
New Hampshire   74 %
Nevada          73 %
Iowa            72 %
Florida         66 %
Ohio            66 %
North Carolina  48 %
Arizona         16 %

GOP takes the White House if they get New Hampshire and all the states below it on the list. (Or, with Pennsylvania in which case they need NONE of the other 70%+ states.) This seems consistent with the overall 29% chance, though I didn’t do any statistical analysis.