Session Report: Methodology: Alternative Approaches-Hybrid Modeling

This session opened to a standing-room-only crowd. First up was Jim Duggan describing 'A Simulator for Continuous Agent-Based Modelling'. The paper presented an approach to modeling agent-based dynamics using standard stock-and-flow structures for the agents and aggregation calculations to determine population variables, all within a standard SD simulation package. After laying out a general approach to handling the agent-based and aggregate levels, an example of market-share dynamics was developed in some detail and results were presented succinctly. An interesting feature of the paper was the demonstration of an aggregate, oscillatory response to a step-change in competitors' capacities and a visualization of the underlying behavior of each agent (the latter making use of Excel graphics). The presentation wrapped up with a brief description of future work to be done and a lively question period.

The second presentation by Ignacio Martinez-Moyano was a 'Design for a Multi-Layer Model of Financial Stability: Exploring the Integration of System Dynamics and Agent-based Models' authored by Ignacio and David Sallach, Mark Bragen and Prakash Thimmapuram. The presentation concentrated on the important aspects of integrating agent-based models (representing detailed domain actions in financial markets) and SD models representing the aggregate evolution of the financial system's context. Two approaches to the integration problem were explored: 1) an SD-centric approach in which the SD model controlled the overall simulation process and the agent-based model provided an aggregated view of agent details; 2) a controller-centric approach in which the details of interaction and communication between the SD and agent-based models was handled by an intermediary software component, the Controller. A good deal of the paper described the implementation of Java- and J-based implementations of the necessary communications between the SD model written in Vensim and agent-based models written in either Java or J. Since the purpose of the study was to provide a proof-of-concept for the integration of standard SD models and general agent-based models, no specific dynamic insights were provided but are expected from further work.

The final paper, 'Qualitative, Quantitative or both?: An Experimental Investigation' was presented by Rajat Dhawan who authored the paper along with Marcus O'Connor and Mark Borman. After an initial discussion of the strengths and weaknesses of Qualitative-SD (such as the use of Causal Loop Diagrams) and Quantitative-SD (modeling and simulation and analysis activities), the paper focuses on tests of various hypotheses about the contribution of one or the other or both kinds of activities compared to a No Intervention situation. An important consideration in the specification of hypotheses was the characterization of problem situations as 'simple' or 'complex', with different hypotheses tested accordingly. The paper presents the details of the experimental procedure very well. In the words of the authors, the results may be summarized by "A combination of qualitative and quantitative tools resulted in the best performance in simple as well as complex tasks. Qualitative tools help in understanding the components of the system that interact over time and subsequent modelling of the system in software further enhances understanding and ability to tackle simple and complex tasks."

Jim Duggan Ignacio Martinez-Moyano Rajat Dhawan

Joel Rahn