**5. Discussion**

In this paper we have introduced the biPCPG framework, a generalisation of the PCPG [30] algorithm to datasets with a multi-sample and multi-variable structure that allows a statistical significant and robust analysis, mainly by generating confidence bounds via an adapted bootstrapping procedure. We have then applied this new procedure to a recently introduced dataset that integrates the export of physical goods and services data. The proposed procedure allows the generation of a network of these economic sectors whose links represent the average influence in terms of temporal correlation. This can be seen as an an asymmetric formulation of relatedness [26,52]. The resulting network contains several hub nodes with high degree (namely Plastics, Pigments, Iron and steel articles, Preparations of cereals and milk and Aluminium) as well as distinct clusters of intuitively-related economic sectors (such as a food and plant cluster, a services cluster and manufacturing cluster). We find that, in this network, economic sectors display a relatively high assortativity according to their complexity rank and, to a lesser extent, their category.
