Forecasting competitions should provide empirical evidence about the relative performance of various forecasting techniques/methods in the framework of a (particular) controlled – fair – experimental design. Essentially, they should either help maintaining best practices, by confirming past evid… ences, or improve forecasting practices by revealing new evidences. Also, they may stimulate research by pointing towards potentially interesting developments.
- I'm convinced that the `controlled experimental design' of past forecasting competitions – including the famous M competitions – does not match expectations and needs of many practitioners `out there' (symptomatically most competitions are organized by academics).
Let me digress on this topic here and let me propose some ideas/thoughts/directions for setting-up alternative designs.