‘-Are Political Markets Really Superior to Polls as Election Predictors’- is a paper by Robert Erikson and Christopher Wlezian that calls into question whether prediction markets, and specifically IEM, is as accurate as proponents of prediction markets claim. The paper was highlighted here where it was suggested that advocates of prediction markets were turning a blind eye to it.
In the paper the authors show that while when you compare IEM with raw polls, IEM outperforms the polls however when you manipulate the poll data the polls are more accurate. The generate the manipulation the authors looked at poll data from elections from 1952 onwards which show that over time the early leader tends to lose that lead. They then used that relationship to manipulate polls for elections from 1988 onwards and compared the result with the IEM forecasts. The manipulated polls showed a higher level of forecast accuracy.
I think this is an interesting piece of research but it is a stretch to use this to claim that polls are more accurate than prediction markets. The fundamental problem is that when newspapers (or anyone else for that matter) quotes polls, they don’-t refine them using historical data, they quote the actual poll result. If anything the authors have shown a small bias in IEM that one would now expect to get traded out (like the longshot/favorite bias in betting markets or the January effect in financial markets).
Fundamentally, the thing to note is that while polls make prediction markets more accurate, the converse does not hold.