Introduced on day 26
Forecast combination
A weighted or simple average of forecasts from different methods, which often has lower error than the best single member when members' errors are not perfectly correlated.
In detail
The classic result (Bates and Granger, 1969) is that a combination can beat each of its members. Equal weights are a strong default because weights estimated on short samples are noisy. Combinations also tend to be more robust to a single method failing at a regime change. They are a cheap alternative to building a model-selection layer.