The network of corporate clients: customer attrition at commercial banks

Abstract

Commercial banks might profit from the adoption of methods widely used in networktheory. A decision making process might become biased if one disregards network effectswithin the corporate client portfolio. This paper models the phenomenon of customerattrition by generating a weighted and directed network of corporate clients linked byfinancial transactions. During the numerical study of the agent-based toy model wedemonstrate that multiple steady states may exist. The statistical properties of thedistinct steady states show similarities. We show that most companies of the samecommunity choose the same bank in the steady state. In contrast to the casefor the steady state of the Barabási–Albert network, market shares in this modelequalize by network size. When modeling customer attrition in the network of3 × 105 corporate clients, none of the companies followed the behavior of the initial switcher inthree quarters of the simulations. The number of switchers exceeded 20 in 1% of the cases.In the worst-case scenario a total of 688 companies chose a competitor bank.Significant network effects have been discovered; high correlation prevailed between thedegree of the initial switcher and the severity of the avalanche effect. This suggeststhat the position of the corporate client in the network might be much moreimportant than the underlying properties (industry, size, profitability, etc) of thecompany.