Open andrew-edwards opened 9 months ago
Can base the fit_models()
call on fit_mve
in there (with some tweaking; can use lags = mve_args
so we test the same lags (and valid subsets) as for mve.
Presumably try that first, then adapt sim_and_fit_realisations()
to actually store the output.
I added outputs for multivariate simplex, (mvs), with an option to not fit, as it takes a very long time to run it! Also, both EDM and MVS give negative forecasts sometimes.
Output from
multivariate_simplex()
may be slightly different to other ones. Also check what rho is being used in the selection of the best, but stick with Carrie's code for the comparisons. (best rho may be based on all variables, not just response, but that seems fine in keeping with Takens' Theorem).