Application Background: ConGestive Failure in Fedwire

• Goal/Aspiration for Project
• Develop modeling and analysis capability for Fedwire, the large transaction network in the U.S. financial system, which will allow us to identify system conditions and perturbations that could lead to network congestion. Identify conditions and network characteristics that lead to congestion and methods for reducing the likelihood and extent of congestion.
• Approach/Methods/Models
• Develop network model of Fedwire and conduct sensitivity analyses to identify conditions that increase the likelihood and extent of congestion in the payment system.  Use insights about the network topology and sensitivities to identify actions or controls that reduce the risks due to congestion.
• For network congestion applications, the group developed a comprehensive model, Polynet, which simulates cascading failure over a wide range of network topologies, interaction rules, and adaptive responses as well as multiple interacting and growing networks.  Polynet was tested by implementing the classical Bac, Tang, and Wiesenfeld (BTW) sand-pile in several network topologies and compared to the results from other models.  The interaction rules in Polynet were tailored to represent Fedwire, a Federal Reserve network service for sending large-value payments between banks and other large financial institutions.
• The Fedwire network model is defined by Fedwire transaction data: payments among more than 6500 large commercial banks, with typical daily traffic of more than 350,000 payments totaling more than $1 trillion.  The node degree and numbers of payments follow power-law distributions and bank behavior is controlled by system liquidity.  Payment activity is funded by initial account balances, incoming payments, and market transactions; payments are queued pending funding; and queued payments are submitted promptly when funding becomes available.
• Status, Accomplishments and Next Steps
• Completed development of Loki Fedwire and congestion analyses (2004-2008).  This set of studies found that payment flows follow a scale-free distribution and system performance is a function of both topology and behavior – neither alone can explain system robustness to disruptions (such as the loss of a bank). Liquidity limits can lead to congestion and limit throughput, but performance can be greatly improved by moving small amounts of liquidity to the places where it’s needed.  There are three time constants that control congestion: liquidity depletion time, net position return time and liquidity redistribution time through the market.
• CASoS Goals: General Capabilities
• Polynet
• Loki-Fedwire
• Network topology and condition effects on congestion
• CASoS Goals: Other Potential Applications
• Evaluate the potential for market transactions and congestion in other networks (energy, communications, transportation), expand the network analysis to condition/information dependent behaviors (confidence), and design controls that reduce congestion risks.
• Acknowledgements
• This application has been funded the Department of Homeland Security through the NISAC program and builds on capabilities and knowledge developed from working with the Federal Reserve Bank of New York on model design and problem description.