Funded by the Defense Advanced Research Projects Agency (DARPA), Adam Rose is contributing to a project studying resilient supply-demand networks (SDNs), which are essential to US national security and military sustainment. Current analysis of SDNs focuses on local-level networks, simple scenarios, and metrics that favor myopic optimization over broad resilience. Moreover, no robust global SDN stress test framework exists to understand risk implications, discover unforeseen potential risks, and develop mitigations. One core challenge is the scale and complexity of simulating SDN counterfactuals on networks with thousands of nodes and links. Whereas top down-simulations, such as computable general equilibrium (CGE) models are typically used to model how policy influences global SDNs, these models aggregate elements of the network to the sector or population segment level. This aggregation makes it impossible to model endogenous feedback effects at the micro level (e.g., bullwhips) and network evolution due to stress tests. Bottom-up simulation, such as agent-based models (ABMs), do model individual decisions and interactions, providing a way to understand endogenous network reactions and evolutions.
However, ABMs require specific data to calibrate behavioral rules (e.g., how firms choose suppliers), can be slow to run due to modeling thousands of agents, are often opaque (i.e., it is difficult to understand what drives decision-making), and lack the solid theoretical basis of CGE models. The understanding of risks and mitigations in SDNs requires a method that combines the advantages of both types of simulation while addressing their limitations.
To address this need, Raytheon BBN (BBN) with teammates Clarkson University (CU), University of Southern California (USC), and The Ohio State University (OSU) is focused on Forecasting Risk in Supply-Demand Networks (FRSN). FRSN combines a bottom-up micro model and top-down macro model to provide DoD planners and analysts a capability for forward and reverse stress testing of global SDNs. FRSN helps users identify scenarios that can cause critical SDN disruptions and develop mitigations to these scenarios. Central to our approach is a novel micro/macro counterfactual simulation engine. Our micro model is a fine-grained simulation of an SDN, leveraging BBN’s established research in this area and an agent-based model (ABM). The ABM augments simulation by modeling the decision logic and entity-to-entity interactions that produce surprising responses to stress scenarios, including SDN network evolution. We form a hybrid, detailed ABM model to understand how areas—sectors, geographic regions, etc.—of an SDN may respond to shocks over time, including how the network may evolve.
Central to the approach is a novel micro/macro counterfactual simulation engine. Our micro model is a fine-grained simulation of an SDN, leveraging BBN’s established research in this area and an agent-based model (ABM). The ABM augments simulation by modeling the decision logic and entity-to-entity interactions that produce surprising responses to stress scenarios, including SDN network evolution. We form a hybrid, detailed ABM model to understand how areas—sectors, geographic regions, etc.—of an SDN may respond to shocks over time, including how the network may evolve and how behaviors change. The ABM model enables FRSN to capture endogenous feed-backs, such as bullwhips (i.e., a phenomenon whereby variance in orders is greater than variance in sales and this variability propagates and increases through the SDN) and ripples (i.e., propagation of SDN disruptions throughout a network).
We use the output of our micro model to configure a computable general equilibrium (CGE) model to analyze the global impacts of SDN disruptions. CGEs are the state-of-the-art economic tool to study spatial, temporal, and sectoral implications of major shocks. These models determine equilibrium outcomes, such as quantity and price, of commodity flows in the economy at an aggregated level (economic sectors, each based on a representative decision-maker). We build on the growing literature describing how to link fine-grained detailed simulation models, such as ABMs and with CGEs for a thorough and accurate counterfactual analysis.
Firm surveys augment our understanding of how firms may behave under stress. FRSN team member OSU in collaboration with USC has developed and administered large-area online economic surveys to understand both historical and potential future firm behaviors under stress. These surveys are crucially different from other data because they help us understand how firms might behave given counterfactual scenarios, which may be very different from normal economic times. While we do leverage data from contracts provided by other performers that describe requirements, even under stress, this data does not describe firms’ decision-making processes, such as how they choose suppliers (e.g., is it lowest cost or some other reason?), firms’ ability to substitute, or impacts associated with utilizing microeconomic resilience tactics (e.g., inventory buffers, alternative suppliers).