Liquidity costs and tiering in large-value payment systems

This paper develops and simulates a model of the emergence of networks in an interbank, RTGS payment system. A number of banks, faced with random streams of payment orders, choose whether to link directly to the payment system, or to use a correspondent bank. Settling payments directly on the system imposes liquidity costs which depend on the maximum liquidity overdraft incurred during the day. On the other hand, using a correspondent entails paying a flat fee, charged by the correspondent to recoup liquidity costs and to extract a profit. We specify a protocol whereby one bank in each period can revisit its choice whether to link directly to the system, or to become clients of other banks, thus generating a dynamic client-correspondent network. We simulate this protocol, observing the emergence of different network structures. The liquidity pricing regime chosen by a central bank is found to affect the tiering process and the network structures it produces. A calibration exercise on data from the UK CHAPS system suggests that the model is able to generate realistic predictions, ie a network topology similar to that observed in reality, driven solely by the underlying pattern of payments and the structure of liquidity costs.