Heuristic assessment of the economic effects of pandemic control

Xiang Niu, Christopher Brissette, Chunheng Jiang, Jianxi Gao, György Korniss, Boleslaw K. SzymanskiarXiv 2020 · arXiv

Introduction

Published in late 2020, this research establishes a mathematical framework to analyze how governments can optimize interventions within the "cascading failures" of global risk networks. Using the COVID-19 pandemic as a primary case study, the authors demonstrate a method for converting qualitative expert-driven data into continuous mathematical models, enabling the application of optimal control theory to global crises.

The Problem of Interconnected Risks

Global risks—ranging from pandemics and climate change to financial collapses—rarely occur in isolation. Instead, they form a "risk network" where a single event can trigger a catastrophic cascade. Traditional control theory typically relies on precise differential equations derived from physical laws. However, risk networks are often "informally defined," relying on expert consensus rather than hard physics.

Furthermore, controlling these networks involves a complex trade-off: the cost of the intervention (e.g., economic stimulus) versus the "state cost" of the risk itself (e.g., lost GDP due to unemployment). Most existing models focus solely on the cost of the intervention signal, neglecting the compounding damage of allowing a "bad state" to persist.

Methodology: Bridging Social Science and Engineering

The researchers bridge the gap between social science data and engineering mathematics through a multi-step process:

  1. CARP Modeling: Risks are treated as nodes in a network using the Cascading Alternating Renewal Processes (CARP) model, transitioning between active and inactive states based on internal probabilities and "infections" from neighbors.
  2. Continuous Mapping: Discrete, stochastic transitions are converted into continuous dynamics, representing the state of a risk as an "expected value" of activity.
  3. Modified Linear-Quadratic Regulator (LQR): A staple of control theory, the LQR is modified to account for both control energy (intervention effort) and state cost (economic damage).
  4. Two-Phase Control: The model distinguishes between Reactive Control (stabilizing a system already in crisis) and Proactive Control (maintaining stability to prevent future slides into crisis).

The study utilizes the 2017 World Economic Forum (WEF) Global Risk Network, a dataset built from hundreds of experts defining the topology and interaction strengths of global risks.

Figure 1: The WEF Global Risk Network topology and cost distributions for driver node selection.

Key Results and Insights

The 2017 WEF network was found to be dense and "near regular," with a mean vertex degree of 18.27, making it highly susceptible to cascades. The authors identified a specific set of seven "driver risks" commonly targeted by governments during COVID-19: deflation, failure of financial institutions, unemployment, failure of national governance, failure of global governance, failure of urban planning, and profound social instability.

Topological Leverage and Cost Efficiency

The research highlights that not all risks are equal; certain "hubs" provide significant leverage for stabilizing the network. While the government-selected driver set performed better than the median of random samples, mathematical optimization identified alternative combinations of seven nodes that would have yielded significantly lower total costs.

The Cost of Delay

A critical finding is the negative correlation between control cost and total cost during the reactive phase. Investing more in the correct driver nodes early on drastically reduces the total economic damage. Furthermore, "proactive" control—applying constant, low-level pressure—is often more cost-effective in the long run than reactive measures taken after a cascade has begun.

Figure 2: Total costs decrease as the number of driver nodes increases in reactive control scenarios.

Limitations and Future Outlook

While the framework provides a powerful tool for policy assessment, it faces several limitations:

  • Expert Subjectivity: The model relies on the accuracy of WEF expert consensus regarding network topology and connection strengths.
  • Unit Cost Assumptions: For simplicity, the authors treated the cost of every risk as equal, whereas in reality, risks like infectious diseases carry much higher economic weights than others.
  • Static Topology: The model assumes a fixed network structure, though real-world crises may forge new links between risks.
  • Policy Translation: Identifying a mathematical "driver node" is only the first step; translating that signal into specific, effective legislation remains a challenge for policymakers.

Summary updated

Abstract

Data-driven risk networks describe many complex system dynamics arising in fields such as epidemiology and ecology. They lack explicit dynamics and have multiple sources of cost, both of which are beyond the current scope of traditional control theory. We construct the global risk network by combining the consensus of experts from the World Economic Forum with risk activation data to define its topology and interactions. Many of these risks, including extreme weather, pose significant economic costs when active. We introduce a method for converting network interaction data into continuous dynamics to which we apply optimal control. We contribute the first method for constructing and controlling risk network dynamics based on empirically collected data. We identify seven risks commonly used by governments to control COVID-19 spread and show that many alternative driver risk sets exist with potentially lower cost of control.
BibTeX
@misc{niu2020heuristic,
  author = {Xiang Niu and Christopher Brissette and Chunheng Jiang and Jianxi Gao and György Korniss and Boleslaw K. Szymanski},
  title = {{Heuristic assessment of the economic effects of pandemic control}},
  year = {2020},
  eprint = {2010.14452},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2010.14452},
}
Figure