Chance Discovery with Self-Organizing Maps: Discovering Imbalances in Financial Networks

Abstract

In this chapter, we introduce the Self-Organizing Map (SOM) from the viewpoint of Chance Discovery. The SOM paradigm supports several principal parts of Chance Discovery: visualization of temporal multivariate data, discovering rare clusters bridging frequent ones, detecting the degree of event rarity or outliers, and dealing with continuously evolving structures of real world data. Here, we further enhance the standard SOM paradigm by combining it with network analysis. Thus, we enable a simultaneous view of the data topology of the SOM and a network topology of relationships between objects on the SOM. The usefulness of the Self-Organizing Network Map (SONM) for Chance Discovery is demonstrated on a dataset of macro-financial measures. While the standard SOM visualizes country-specific vulnerabilities by positions on the map, the SONM also includes bilateral financial exposures to show the size of linkages between economies and chances of shock propagation from one country to the rest of the world.