I have nine independent series of financial numbers that are part of an overall budget. I want to build models and forecast each series. I was told by an expert that since each individual series is very "noisy", instead of building univariate models, it would be best to build multivariate models where the explanatory variable was just the linear combination of each series (less the series I want to model).

Has anyone ever heard of doing this before? I can't find any literature references for this approach.