The pay-off for the traders is contingent upon the real event. That means they have to wait until the actual event occurs, before they know what their prognosis is worth and what the pay-off for each contract is.

What about long-term events?

For long-term prediction, this doesn’t seem to be a practicable way. Waiting till 2020 when e.g. the real unemployment rate is published does not provide an incentive to trade shares for this.

But what could be a way to construe a reasonable pay-off function?

One possible approach by Prof. Spann at the University of Passau is to run two separate prediction markets concurrently. All information about prices, traders etc. of market A would not available for traders of market B and contrary.

The pay-off for Market A would be the final price of Market B and vice versa.

So it is possible to run an virtual market for e.g. 2 months predicting the outcome of an event in 2020.Another way is to let an expert(s) assess the event an use his (their) opinions as the pay-off function.

Furthermore the own end-price of the market could be the pay-off. But then manipulation is to be expected.
I’m looking forward to discuss this challenge with you!

P.S. www.Ideosphere.com seems to be a prediction market for long term predictions, but they don’t provide a solution for the problem.