(359a) Process Systems Engineering Beyond Chemical Plants: Signed Digraph As a Modeling Tool for Analyzing Systemic Risk in Financial Networks

Process systems engineering (PSE) has been an active and important research area in chemical engineering for more than half a century. It encompasses the control, optimization, and risk management of large-scale chemical engineering systems. Complex sociotechnical systems, such as modern financial systems, are characterized by the same interdependencies and a large number of units one often observe in a chemical plant. Systemic disasters, such as â??08 financial crisis and â??10 BP Deepwater Horizon oil spill, share alarming similarities. The commonalities naturally lead to the question: Can we use PSE to understand and, more ambitiously, design complex sociotechnical systems?

Human factors such as self-interest and cognitive limitation are critical in the course of a systemic event. They are also the biggest challenges of applying PSE to such systems. Without a central planner and being subject to the limited information, each individual (person or institution) tends to maximize its own utility and oversee the local stability only. This myopia might lead to system-wide failures as we have seen in the financial crisis and other systemic disasters.

In this talk, we apply the signed digraph (SDG) technique used in fault diagnosis to modeling systemic risk in a financial system. This approach helps us identify whether and how a causal pathway in the financial network could potentially lead to a systemic risk outcome. The framework is capable of recreating known runaway situations and discovering potential vulnerabilities. More importantly, since each causal relationship is derived from underlying economic principles and rational individual behaviors, one would expect global stability for the overall system. Quite the contrary, we identified systemic runaway events such as fire sale and funding run from the diagnosis despite individual institutions being conservative. Actions that dampen risk on a local level can contribute positive feedback and cascades on the global level, leading to large-scale unintended consequences.

Automation plays an important role in such fault diagnosis. In the case study, one can perhaps manually identify and enumerate all the feedback loops. However, for a more realistic version of the network, where there are multiple interacting subsystems, it is virtually impossible to analyze such loops manually. The SDG framework can be automated to handle larger systems via artificial intelligence and optimization techniques. The end-goal of this framework is providing a systematic method to detect potential threats and design such complex systems.