ИСТИНА

Information about trade deals from USAIS is a new data source – analysis methods have not been developed, its’ potential is not clear. The information about all deals between enterprises makes it possible to study economic relations at the most fractional level. The key issue of our study is to understand whether enterprises break up into separate groups or not. A method of graph clustering was developed. It is a combination of Leiden algorithm, Fruchterman-Reingold layout and meanshift algorithm. The resulting clusters have various sizes and content. The average share of intra-cluster trade is 93 %. The modularity score of our partition is 0,863. We defined 5 morphological types of clusters: vertical monocentric, vertical polycentric, horizontal, dendritic and simple. There are three main factors of clustering in the wood industry of Russia: production chains, demand chains, common holder. Finally, the map of defined clusters was plotted. Most of the clusters form compact areas – wood trade regions. We have shown, that existing approaches of cluster detection, based on the location quotient often reflect economic networks in the wrong way. Proximity often doesn’t mean connectivity.