GSS preparatory meeting, Beijing, October 29, 2013
Notes on the parallel session about agent-based models for GSS
In general agent-based models have proven to be useful in modeling systems characterized by emergent properties at macro level, heterogeneous interacting agents, networks and bounded rationality. In economics agent-based models emerged from the criticism towards the representative agent and the perfect rationality assumption.
The group discussed the research area in which they are applying agent-based models, which are the following:
- Transportation/traffic issues
- Disaster response (shelter topology & impacts on economic activity)
- Climate change and sustainable economic system
- Housing markets
- Macroeconomic & financial system
- Integrated risk governance
Furthermore, it was discussed that agent-based modeling is one of several tools, whose usefulness depends on the specific system or problem to be studied.
Concerning the usefulness for different applications, an important question that was addressed was: Where have agent-based models been used successfully? Where have they improved the understanding of specific problems and related decision making. We identified:
- Pedestrian traffic (where heterogeneity is not important, but the topology and a large number of actors are important)
- Panics in large groups/herd behavior in financial markets
The first has been widely used to understand crowd behavior and design safer public structures such as subway stations and airports and possibly improve post-disaster management. The second have proven to replicate boundedly rational behavior of human beings, both in financial markets and in emergency situations. We regard the other topics as promising in terms of generating new insights.
An interesting question for the GSS conference would be how agent-based models can be used for analyzing global systems instead of for specific questions only. In this regard, we propose to invite agent-based modelers working on more elaborate models.
We discussed scale effects, in particular which number of agents are needed for specific applications. We discussed a proposals of aggregating agents (not using representative agents) and the possible levels of aggregation, such as micro or meso level modeling of agents.
Especially considering economic models we need to improve the communication with policy makers and mainstream economists. In this regard, a crucial aspect which should be addressed in the conference is how to estimate, calibrate and validate agent-based models with empirical data.
Another important question is which programming tools can be used to share code and models and to minimize programing errors. Due to lack of time we did not elaborate on this topic. However, it is an important issue and should be addressed in the conference.