Hybrid forecasting (combining models and markets for example) is a very promising approach. Two ways to do it, one could use market prices as a model input or you could have trading bots representing models active on markets. Of course markets do absorb model forecasts in any case but it would be interesting to see if experimental interventions using models result in more accurate predictions.
One of the principles in building model ensembles is that you should incorporate models that encode diverse assumptions regarding the X --> Y mapping. Using similar logic, consolidating information from statistical models and prediction markets should help us make better predictions on average unless one takes what is, in my estimation, the radical view that election forecasting is a lost cause. Decision Desk used an ensemble of machine learning algorithms to generate probabilistic forecasts, but the ensemble I'm thinking of is taken across all statistical and market predictions. Did someone do a prospective aggregation of this sort that you know of?
Hybrid forecasting (combining models and markets for example) is a very promising approach. Two ways to do it, one could use market prices as a model input or you could have trading bots representing models active on markets. Of course markets do absorb model forecasts in any case but it would be interesting to see if experimental interventions using models result in more accurate predictions.
One of the principles in building model ensembles is that you should incorporate models that encode diverse assumptions regarding the X --> Y mapping. Using similar logic, consolidating information from statistical models and prediction markets should help us make better predictions on average unless one takes what is, in my estimation, the radical view that election forecasting is a lost cause. Decision Desk used an ensemble of machine learning algorithms to generate probabilistic forecasts, but the ensemble I'm thinking of is taken across all statistical and market predictions. Did someone do a prospective aggregation of this sort that you know of?