DEVELOPING DYNAMIC STRESS TEST FOR BANKING SYSTEM WITH ADAPTIVE LEARNING AGENTS: AN AGENT BASED APPROACH

• Main Author: Raissa Raswan, Tara
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78603
• Institution: Institut Teknologi Bandung

Nowadays, world economic development cannot be parted from the financial system because financial growth closely linked to economic growth. The system which is comprises of all parties that involved in money flow, i.e. financial markets, instruments, and institution. A financial system which categorized as a sophisticated one is involving a higher degree of intermediation with a higher degree of dependency, hence it possesses a higher risk of failure. The importance of managing financial system stability mainly caused by the high cost of financial system failure. Based on several studies, the financial system failure is one of the costliest failure compared to other. This study proposes a relatively new approach for a banking system stress test. As the current generation of risk models which mostly employed in banking’s risk management are unable to model financial vulnerabilities comprehensively. The typical stress testing used in risk management does not consider few important thing, such as bank’s response to dynamically changed economic condition, including system feedback and other banks response. The stress test in this study is developed using agent-based modeling. In current study each agent is created as learning-agent, which has the ability adjust their response using reinforcement learning method. The result of this research revealed that: first, high net worth ratio is lowering the systemic risk of the system, and as net worth ratio decreasing the effect is nonlinear. Second, Size of interbank liability increases the risk of default, the bigger the exposure the bigger the risk. The relationship is also non-linear. Third, the connection number of the bank has non-monotonic relation to contagion effect. The effect is peaking at a number and the effect is lower elsewhere. In addition, the result showed that preferred action is mostly to increase the net worth and lower the interbank exposure. The magnitude of change determined by the initial condition of the system. In Addition, The model showed that the dynamic approach in banking stress test that taking into account the dynamic interaction among banks shows the important impacts. One of the main results is that the approach can avoid the underestimation of the dynamic effects the hat was not included in the static one.