Cluster is also a sample design issue in survey analysis procs. For example students sampled within each School building of a School district. The STRATA could be School Districts and Clusters are the Schools. You would as a minimum need a totals data set and each strata should indicate the number of clusters. If your not sampling withing the Id_trader then it looks like you may be looking at more of a domain analysis then clusters. If you are looking for similarity between your ID_trader values then you would use one of the clustering procedures to identify the similar groups and likely want to merge that into the data as a model variable but not a sample cluster.
... View more
