Operational Risk Modeling with ABM

Summary

Bonabeau (2002) describes ABM applications in operational risk management, where traditional statistical approaches fail because risk events are rare, interdependent, and arise from complex interactions between human behavior and organizational processes. ABM enables bottom-up simulation of risk scenarios that can produce fat-tailed loss distributions matching observed operational risk data.

Overview

Operational risk — the risk of loss from failed internal processes, people, systems, or external events — became a major focus after Basel II required banks to quantify and hold capital against operational risks. ABM offers an alternative to historical data analysis and scenario modeling by simulating the interactions that produce risk events.

Main Content

Why ABM for Operational Risk

Traditional operational risk modeling faces challenges:

  • Rare events: Operational risk losses are infrequent but can be catastrophic
  • Interdependencies: Risk events are not independent — one failure can cascade through interconnected processes
  • Human factors: Operational risk often involves human error, fraud, or failure to perform in a timely manner
  • Fat tails: Loss distributions have heavier tails than standard parametric distributions predict

ABM Advantage for Risk

ABM provides “a bottom-up enterprise simulation” that can be used to “construct the interactions of the bank to be computerized in some other way, for example by its dealers, lending officers, or other staff executing their authority or conducting business in an unethical or risky manner” (Bonabeau 2002).

The SCGAM Approach

Bonabeau describes an ABM approach to operational risk based on the Society of Generative Agent Models (SGAM) approach:

  1. Bottom-up construction: Model the organization as a collection of agents (employees, systems, processes)
  2. Interaction rules: Define how agents interact, including normal operations and failure modes
  3. Scenario generation: Run simulations to generate a distribution of outcomes
  4. Loss distribution: The simulated loss distribution can exhibit fat tails and correlations that emerge from the interaction structure
  5. Capital calculation: Use the simulated distribution for Basel II capital requirements

ABM vs Traditional Risk Models

FeatureTraditionalABM
Data requirementHistorical loss dataProcess knowledge + behavioral rules
CorrelationsAssumed or estimatedEmerge from interactions
Rare eventsExtrapolated from dataGenerated from scenarios
Fat tailsParametric fitEmergent from cascading failures
Process changesRequires new dataCan be simulated immediately

Examples of Operational Risk Events Suited to ABM

Bonabeau lists examples of large operational losses:

  • Daiwa, Sumitomo, Barings, Kidder Peabody
  • Orange County, Jardine Fleming
  • Common Fund, Yamaichi
  • These involved complex interactions between individual actors, organizational processes, and system failures

Connections

See Also