Upon examining the literature on whether real money prediction markets are more effective than play money markets there seems to be no clear cut answer. Some studies say play money is just as accurate as real money, while others say real money is significantly more accurate. I'm inclined to agree with the real money studies, as they're both more recent research and the incentives to "put your money where your mouth is" are much higher in real money markets.
However, I think this debate in the academic community of real vs fake money markets is largely irrelevant. All PM researchers agree that more liquid markets with higher volume produce more accurate forecasts.
Now, examining the prediction market field, the most popular ones using fake money (or in the case of Scicast, "votes") are http://www.foresightexchange.com/ and https://scicast.org/#!/. Look at their volumes and activity. Now compare this to something like InTrade, which did hundreds of millions of volume in election years.
In my opinion (and according to the trading volume evidence), the only way to get large volume in prediction markets is to use real money so people have a significant incentive to use them. If real money prediction markets have more volume (see Betfair) than even the most successful play money ones (Scicast), then real money PMs will provide more accurate forecasts. Since it's nigh impossible to get as much volume on a play money system as a real money one, I believe incorporating real money into a predictive market platform is a key feature to its success as a forecasting tool.
