Assuming a user has a number of forecast models available: How can these be usefully combined to improve forecast accuracy on unseen data compared to choosing individual models? Which models should be selected and should the learned weights be corrected?
If this topic sounds interesting to you, we would like to refer you to the just published research article "Non-Linear Hybrid Shrinkage of Weights for Forecast Selection and Combination", authored by Felix Schulz.
This paper is an extension of the recently presented article "Linear Hybrid Shrinkage of Weights for Forecast Selection and Combination" at HICSS (2022) and presents the nonlinear shrinkage of optimal weights towards zero or the mean depending on the selection status as well as a new selection criterion based on forward selection. Among benchmark algorithms such as simple forecast averaging, estimated optimal weights, or linear and LASSO-based shrinkage methods, the presented method shows advantages especially for a larger number of forecast models, as shown in simulation-based experiments.