When combining more than two individual models, changing the order of individual models in the combination_model() function will produce different combination results (variance). I give an example below for forecast combinations using the cafe.rds data in the NYCR talk by Rob.
It can be observed that, whether working with an equally or unequally weighted combination, the variances of the combined normal distributions with different order of individual models are different. The reason is that the combination_model() function combines the individual models in an iterative manner.
I think this should be modified using the mean and variance of the weighted sum/average of correlated variables (please refer to the link). And I wrote the following code to achieve the combinations of multiple models, resulting in the same combined distributions even when the order of individual models was changed.
When combining more than two individual models, changing the order of individual models in the
combination_model()
function will produce different combination results (variance). I give an example below for forecast combinations using thecafe.rds
data in the NYCR talk by Rob.It can be observed that, whether working with an equally or unequally weighted combination, the variances of the combined normal distributions with different order of individual models are different. The reason is that the
combination_model()
function combines the individual models in an iterative manner.I think this should be modified using the mean and variance of the weighted sum/average of correlated variables (please refer to the link). And I wrote the following code to achieve the combinations of multiple models, resulting in the same combined distributions even when the order of individual models was changed.