Likelihood-based inference in temporal hierarchies


Jan Møller, Denmark Technical University, Denmark


The variance-covariance matrix of the base forecast prediction errors is of importance in hierarchical reconciliation. The main hypothesis in this talk is that the variance-covariance of the base forecast can be given a parametric formulation and hence estimation of the variance-covariance matrix can be subject to usual model identification and maximum likelihood estimation. The statistical model as well as testing procedures to obtain parsimonious models will be discussed in this talk, and further some examples will be presented.


Dario Azzimonti, IDSIA, Switzerland