The combination of forecasts may be superior (on MSFE, say) to each of the constituents. However, forecast combination runs counter to the concept of encompassing viewed as part of a progressive research strategy. The latter would suggest that a better approach is to refine a model when it is found to be wanting in some dimensions: here, the inability to predict some aspects of a process as well as a rival. The goal is a model which encompasses its competitors by forecasting the process better than its rivals. A test for forecast encompassing is the same as that for whether there is any benefit to combination, although conceptually the encompassing approach is different from the ethos behind the combination of forecasts. When models do not draw on a common information pool, and are essentially of a different nature or type, or when models are differentially susceptible to structural breaks, then the case for combination is more persuasive.