Estimating the Structure of the Payment Network in the LVTS: An Application of Estimating Communities in Network Data

In the Canadian large value payment system an important goal is to understand how liquidity is transferred through the system and hence how efficient the system is in settling payments. Understanding the structure of the underlying network of relationships between participants in the payment system is a crucial step in achieving the goal. The set of nodes in any given network can be partitioned into a number of groups (or “communities”). Usually, the partition is not directly observable and must be inferred from the observed data of interaction flows between all nodes. In this paper we use the statistical model of Čopič, Jackson, and Kirman (2007) to estimate the most likely partition in the network of business relationships in the LVTS. Specifically, we estimate from the LVTS transactions data different “communities” formed by the direct participants in the system. Using various measures of transaction intensity, we uncover communities of participants that are based on both transaction amount and their physical locations. More importantly these communities were not easily discernible in previous studies of LVTS data since previous studies did not take into account the network (or transitive) aspects of the data.