“empirical studies have shown that such a simple equally weighted pooling of forecasts performs quite well in practice, relative to other approaches that rely on estimated combination weights, a phenomenon dubbed the forecast combination puzzle.”
“One of the puzzles for forecast combination is the documentation of simple average (or equally weighted combination) dominating more sophisticated forecast combinations (e.g. Huang and Lee (2010), Palm and Zellner (1992) and Stock and Watson (2004)).”