Three recent polls have come out in regard to the Republican Presidential Nomination. These polls by CNN, Fox, and NBC have Donald Trump in first, third, and second respectively. Further, they have Mitt Romney in fourth, second, and first respectively.
Imagine that there is three types of information about the current field of Republican candidates: first is known to the average Republican voter, second is known to political informed people, and third is unknown. An example of the first type of information for Trump is that he is a popular TV personality and for Romney that he is a former governor of Massachusetts. An example of the second type of information for Trump is that his business affairs and personal life are very rocky and for Romney that he was an extremely successful businessman. The third type of information is all of those things that will happen between now and Election Day, for which is unknown.
In short, differences between poll-based and prediction market-based forecast of who will ultimately win the Republican nomination hinge heavily on the second type of information: information that political informed people in the prediction market know and the average Republican voter does not know now, but will know on Election Day. In this situation, that type of information is generally neutral or positive for Romney, Tim Pawlenty, and Mitch Daniels, but negative for Trump. That is why although Trump is trending second in the polls of today, the self-selected politically informed people who gamble in the prediction markets have Trump as the fourth most likely person to get the Republican nomination. They are betting that the public will learn what they know about Trump and then become less likely to vote for him, relative to the other candidates.