Page 1
1
A Co-evolutionary Simulation
of Multi-Branch Enterprises
Edmund ChattoeCentre for Research on Simulation in the Social
Sciences (CRESS),Department of Sociology
University of SurreyGuildford GU2 5XH
[email protected] ://www.soc.surrey.ac.uk/staff/edmund_chattoe.html
Page 2
2
Simulation
• Social processes as computer programs.
• Heterogeneous agents: firms, branches and consumers.
• Parallel autonomous processes.• Linking macro/micro and
mental state/action.• Integration of theories.• Integration of data: qualitative
and quantitative.
Page 3
3
Evolutionary Approaches to Social Action
• Not genetic determinism!• Population Approach: Mental
models leading to actions with differential success.
• Distinction between agent and environment.
• Robust self-organisation with minimal rationality.
• Concrete models based on Evolutionary Algorithms.
Page 4
4
Motivation of the Model
• Suitable market for evolutionary modelling.
• Intuitive environment for fair comparison of learning methods.
• Difficult learning task.• “Independent” specification of
agents in simulation.• Possibilities for emergent data.• High level view of relation
between agents and environment.
Page 5
5
Three Relations Between Theorist, Manager and Environment
Environment
Manager Theorist
Page 6
6
High Level Description
• Spatially distributed consumers and branches.
• Consumers seek suitable meals.• Firms collect sales/practice data.• Learning changes practices.• Consumers, branches and firms
linked by common representation of practices:
0000010010110
Page 7
7
Consumers
• Fixed chance of hunger.• Two behaviour modes (random
search and loyalty) and memory of “best” branch.
• Tolerance based on Hamming Distance.
• Changing tolerance.• Fatigue.• Consumer types.• Dis-satisfaction.
Page 8
8
Firms and Branches
• Set of practices.• Running and capital costs.• Income from sales.• Profits as surplus over costs.• Learning algorithm and
decision process.
Page 9
9
Typical Output 1: Location Graph
Page 10
10
Typical Output 2: Favourite Branch Graph
Page 11
11
Typical Output 3: Sales Graph
Page 12
12
Typical Output 4: Profit Graph
Page 13
13
Typical Output 5: Dis-satisfaction Graph
Page 14
14
Consumer Conditions
• Random practice change on loss.• Fixed preference, no tolerance, 4
types.• Fixed preference, tolerance 1, 4
types.• Changing preference, no
tolerance, 4 types.• No tolerance, fatigue, 4 types.• Fixed preference, tolerance 1, no
types.
Page 15
15
Results: Consumer Conditions
Page 16
16
Learning Conditions
• No adaptation.• Random change on loss.• Independent hill-climbing.• Loss threshold leading to
imitation/Lamarckian differentiation.
• Many more possibilities ...
Page 17
17
Results: Learning Conditions
Run Total Sales Dis-satisfaction Total ProfitsReference 6,689 213,439 12,155Reference 8,977 114,434 27,004Condition 6 4,321 317,728 -8,084Condition 6 5,964 245,496 -9,104Condition 7 7,123 195,608 929Condition 7 7,712 165,594 22,641Condition 8 7,588 172,319 19,893Condition 8 9,886 75,566 45,211
Page 18
18
Discussion
• Profit insulates from selection.• Satisficing is environmentally
stable while optimising is not.• Need to identify market niches
based on consumer knowledge.• Hill-climbing has difficulties in
non-stationary environments.• Even naive evolution is
surprisingly effective.
Page 19
19
IMAGES Project
• Payments for ecological activities.• Agent based models of innovation
diffusion.• Information transfer through
social networks.• Interaction of social and economic
reasoning.• Integration of qualitative
interviews, farm surveys and secondary data.
Page 20
20
Modelling Budgetary Decision of Pensioners
• Developing simulation inductively from interviews.
• Integrating economic and sociological theories.
• Importance of:o Time planning.o Durable scheduling.o Joint activities.o Budgeting strategies: multiple accounts.
Page 21
21
Evolution of Pricing Strategies in Oligopoly
• Firms use Genetic Programs to set prices.
• Programs are modified by profit induced mutation/imitation.
• Random generation of strategies still allows stable markets and responsive behaviour.
• Stylised behaviours like price following and “cost plus” pricing are observed to evolve.
Page 22
22
Research Plans
o Simulations linking actual firm micro practices with market outcomes.
o Extension of time planning approach beyond consumption choice.
o Memetic models of information transmission for establishment of “culture”.
o Social dimensions of electronic commerce and crime.
o Agent based models of innovation diffusion: a case based approach.
o Charity shops as alternative spaces of consumption.