A Strategic Agent-Based Analysis of Economic and Technological Changes in Financial Networks

Abstract

Economic events and advancements in technology have drastically transformed the financial system over the past 20 years. This includes the implementation of new policies post-2008 financial crisis, as well as new methods for customers to engage with the system from cryptocurrency to faster processing payments. This system is a complex network, vital for connecting individuals, businesses, and financial institutions for the purposes of monetary transactions. Thus, it is important to carefully consider how changes may impact and transform the system in the future. This dissertation applies a computational approach to study strategic decisions that arise from economic and technological changes in financial networks. I introduce the extended financial credit network model capable of expressing diverse financial scenarios and enabling the creation of agent-based models for studying strategic interactions. To study the strategic decisions of agents within the model, I apply methods from empirical game-theoretic analysis to identify equilibrium behavior.

I apply these methods to four case studies of strategic behavior in financial networks. First, I analyze portfolio compression, a method for eliminating cycles of debt, as a strategic decision. Compressing cycles offers both potential benefit and harm to the network, and banks must weigh both possibilities carefully. I find that the compression decision is best made using simple, local information available to banks. Further, I note the importance of the recovery rate of insolvent banks in the perceived affect of cycles on systemic risk. Second, I investigate the strategic use of costly fraud detection systems by banks. I observe the behavior of banks is subject to the relative, not specific, capabilities of their fraud detectors. In particular, those with strong detectors are better able to adjust their behavior in response to rising costs due to the existence of weaker banks in the system. The third study addresses bank allowance of real-time payments by customers. Customer overdrafts pose a potential credit risk to banks, who assume short-term liability, and lead to strategic decisions regarding which customers should be allowed use of these new payments. My analysis shows banks choose to allow most, though not all, customers to send real-time payments. I discover the strategic choice of banks will not lead to the socially optimal outcome in this scenario. Lastly, I extend the previous two works to study banks' strategic mitigation of fraud risk in real-time payments. In particular, I focus on the strategic trade-off banks make between restricting customer access to these payments and investing in costly fraud detection. I find banks value the ability to control customer access, though they never invoke overly strict restrictions. Instead, they balance some constraints with the use of fraud detection. I observe these strategic measures limit the negative effects from fraudulent actors with minimal disruption to customers.

Broadly, this dissertation demonstrates the effectiveness of agent-based modeling and empirical game-theoretic analysis to gain insight into strategic interactions in financial networks. It also illustrates the flexibility of the extended financial credit network model, which is employed in all four studies. Finally, I establish the new strategic feature gains assessment, a method for assessing the benefit of strategies in the strategy space. The assessment uncovers valuable insights into strategic behavior in the studies on portfolio compression and fraud risk in real-time payments.