How to calculate consensus of probabilistic forecasts? Suppose a group of forecasters each made a probabilistic forecast, for example:
Forecaster 1:


*

*40% probability that company ABC will add 2 - 3 million subscribers

*20% probability 3 - 4 million subscribers

*20% probability 1 - 2 million subs

*10% < 1 million subs

*10% > 4 million subs


Forecaster 2:


*

*20% probability > 2 million subs

*50% 1 - 2 million subs

*30% < 1 million subs


Etc...
How can I find the consensus forecast? Something like:


*

*Consensus expected sub addition is X

*Consensus probability of X% that sub addition will be between  A and B, C and D

 A: If you can ensure that your forecasters use the same bins ("<1", "1-2", "2-3", "3-4", ">4"), you can simply average probabilities within each bin between the forecasters. These average probabilities will still add up to 100%, so they are bona fide probabilities, and you can report the expectation or the median of the averaged probability distribution as a consensus point forecast.
If your forecasters have different bins, like in your example (forecaster 2 uses a bin ">2", which forecaster 1 subdivides into three bins "2-3", "3-4", ">4"), you will need to map the forecasts to a common grid. In your example, you could subdivide forecaster 2's bin ">2" into forecaster 1's three bins and divide forecaster 2's assigned probability of 20% equally into 6.67% for each one of the smaller bins, then proceed as above.
If you are only interested in certain fixed quantiles, you can work ad hoc. For instance, the median for forecaster 1 is "3-4" and for forecaster 2 it is "1-2", so you could reasonably report a consensus median of "1-4" or "2-3" - either one would be defensible, and to be honest, with the coarse grids your forecasters use and the probabilities that are always multiples of 10%, I wouldn't worry too much about getting fancy in combining the forecasts.
