Interbank Market Formation through Reinforcement Learning and Risk Aversion

In this study, we propose a multi-agent model to examine bank lending and borrowing risk behaviors and their implications to interbank market dynamics. Using data from 2001 to 2014 that covers around U.S. 6600 banks, we model individual bank decisions using the temporal difference reinforcement learning algorithm based on banks' lending preferences and environment, and we then generate the interbank market dynamics from the empirical data. This dynamic model allows us to construct interbank networks as they change with bank risk preferences, and thus facilitates the analysis of the banking systems stability. The model successfully replicates the key characteristics of interbank lending and borrowing relationships that have been documented in the recent literature. A key finding of this study is that the risk aversion choice of individual bank leads to unique interbank market structures that suggest the macro risk preference of the market. Combined with the use of balance sheet data, this modeling framework helps central banks and regulators building more functional models for examining the interbank market stability problems