Agent-based modelling
in social sciences
Andreas Krause
School of Management
What is ABM?
Agent-Based Modeling (ABM) is the computational
study of dynamic systems of interacting agents. Here
"agent" refers broadly to a bundle of data and
behavioral methods representing an entity constituting
part of a computationally constructed world.
Examples of agents
Individuals: consumers, producers, investors
Social groupings: families, firms, communities, government
agencies
Institutions: markets, regulatory systems
Biological entities: crops, livestock, fish, insects, forests
Physical entities: infrastructure, weather, geographical regions.
Examples of applications
Investment and trading decisions in financial markets
Product market competition
Marketing
Macroeconomics
Traffic flow and road pricing
Ethnic conflicts
Spread of diseases
Opinion dynamics
Adoption of new technologies, languages
My work
Simulating stock markets
Formation and evolution of social networks
Using simple behavioural rules for traders
With realistic market structure rules
Investigation of the aggregate behaviour, i.e. stock prices
Networks evolve locally following exogenous rules
Networks evolving in response to dynamics on the network
(Stock) Market Design (started with PhD student)
Optimization of markets structures with GAs (maybe GP)
Methods used
Simulations, usually generation of long time
series for a large range of parameter settings
Optimization of behavioural rules or
institutions, often using Genetic Algorithms
(GA) or Probability-Based Incremental
Learning (PBIL)
Scale of computations
Mostly a large number (>1,000) of time series (each easily >
1,000,000 time steps)
GAs/PBILs often require more simulations (optimization in
high-dimensional spaces means slow convergence)
Computing speed is important (simulations often take 2 weeks or
longer, GAs could take months)
Relatively low computational complexity, but large number of
computations
Limitations of research
Computing power limited access to up-to-date
computers (speed, memory, computer lab for parallel
computing)
Access to software relevant software (MATLAB)
not available for desktop, specialist software exists but
time consuming to learn
Programming skills need to have proficient
programmers, e.g. PhD students
Funding issues
ABM is a new methodology in economics/ finance
(about 10 years old), not mainstream or yet generally
accepted
Funding is difficult to obtain as it falls between areas
Most publications are in Physics journals and IEEE
Transactions, not truly recognized for
promotions/RAE etc.
Research at other universities
Research Centre in Essex: CCFEA (Economics +
Computer Science)
Broader Research Centres which include ABM: Santa
Fe, Carnegie-Mellon, AI-ECON (Taiwan)
Other places have smaller groups, often focussing on
special areas, usually centred around a small number of
individuals: Cranfield, Cambridge, Oxford, Kiel
(Germany), Genoa (Italy), NDA (Japan), …
Research in Bath
Interests are in various departments mainly in
Management and Computer Science, relatively
isolated
No common forum for exchange of ideas
Limited outside and inside visibility of our
research
Key issues
Adequate hardware/software
Attraction of funding
Visibility within the university and outside
Prospects for ABM research
Emerging field with a wide range of applications
Fast growing community, the main conference covering
economics and finance applications started in 1995 and
has grown from about 60 participants (1999) to 200
(2006).
At present limited competition, chance of taking a
pioneering role
Interdisciplinary approach
Potential key areas of application
Market design
Carbon trading
Road pricing
Electricity and gas markets, in future: water markets?
Procurement markets
Rail franchises
Airwave spectrum auctions
Modelling of ecosystems and their reaction to
environmental/climate change
Modelling of ethnic and religious conflicts