computer simulation and economics
TRANSCRIPT
Computer Simulationand Economics
Edmund ChattoeDepartment of Sociology
University of [email protected]
http://www.sociology.ox.ac.uk/people/chattoe.html
Plan of the Talk
• The ideology of simulation• The “agent based” perspective• A case study: pricing under oligopoly• What next?
The Ideology of Simulation• Agents are “fundamentally” heterogeneous• Agents are “really” cognitive but bounded• Agents are socially situated• Agents adapt in a complex environment of
other agents including organisations• There are systematic micro foundations to
macroscopic regularity• Environments are profoundly dynamic
The “Agent Based” Approach IAGENT 1
c1=a1y1
AGENT 2c2=a2y2
CSOC, Y
ECONOMISTC=aY
ABA: Implications I• Aggregability and absence of feedback: a
special case?• Where do agent models come from?• What is the “story of a”? (functional,
adaptive, other)• How do we find out about agent models?
(Problem of macro to macro inference.)
The “Agent Based” Approach IIAGENT 1MODEL
AGENT 2MODEL
INSTITUTIONRULES
“SOCIAL”SCIENTIST
REGULARITY
ABA: Implications II• How do we model if we relax
aggregability and absence of feedback?• What do we do about organisations?• Does “good science” needs to link levels to
avoid mere “data mining” ?• Are we modelling the “right stuff”?• What do we do about equilibrium?• Is economic theory “just another model?”
Common Concerns• What is the status of cognitivism?• Isn’t this all ad hoc? Yes, but ...• This is all very well but we can’t model it• We have good “social” processual reasons
to assume regularity in agent models• But, if we are wrong, we must all pack up
and go home or become novelists
Simulation: Provisional Definition• Computational (rather than verbal or
mathematical) representation of a social process
• Descriptive rather than instrumental use• An “explicit” representation?• Fundamental problem is not programming
but adequate data
Example: “Social” Market I• Economics aggregates to get D, S curves
and then solves for market clearing• S and D curves don’t exist in the minds of
buyers or (probably) sellers• What exists are inventories, shopping trips,
haggling, gossip …• When shoppers run out they go shopping• While shopping they search and gossip
Example: “Social” Market II• Shops produce on past sales but sell out• Shops may adapt prices through gossip or
direct “observation”.• Shops and customers may match “bids”
and “asks” to reach agreement• S and D curves can be “produced” from
such a simulated market but so can trade networks: effective falsification?
Example: “Social” Market III• Firms raise production level if they sell out or
lower it if they have unsold inventory• Firms lower price if too many customers walk
away or if they hear/observe too many lower prices in other shops
• Firms raise opening bid if nobody walks away• Consumers use 1 unit per period and go shopping
if they run out• Consumers pass/receive one message per period• And so on … this is a programme!
Example 1: Oligopoly Pricing• How do firms set prices in a complex
environment?• Simplify by making it a game or assuming
lots of common knowledge• Third approach is adaptive but this is “too
difficult” for simple adaptation• Possible solution is evolutionary learning:
firms adapt by trial and error and are selected
The Appeal of Evolution• Driven by heterogeneity• Open ended: actors don’t need to know
objective function (if there is one)• Works on minimally effective strategies
using relative success• Analogous to situation of firms?• Observed to produce stable self-organised
heterogeneity in ecosystems
A Brief History• Marshall and the representative firm• Alchian
– Outcomes not intentions– Genotype is firm practices– Phenotype is firm behaviour
• Nelson and Winter– Fixed decision rules
• Dosi et al. (1999)
The Dosi et al. Model• Candidate prices are small set of GP strings• Firms set price probabilistically based on
accumulated profits of candidates• Demand determined by “market” price and
allocated by current market share• Market share updated via set and market price• Profits are accrued to firms• Firms with losses/minimal market share replaced• New candidates may be generated
A Typical GP Price String+
3/
OP1
+ 2
OP2
The Operators• Crossover: Take two trees, identify “legal” cut
points and swop “tales”• Mutation: Take one tree and identify “legal” cut
point for new randomly generated tree.• Other possibilities• Some completely new trees• IF NOT AND OR > < = + - % *• OMP, OMD, OP x OUC, CUC, OS, integer• http://users.ox.ac.uk/~econec/thesis.html
Some Results• Main Dosi et al. result is evolution of price
following and “cost plus” pricing• This appears to be sensitive to assumptions made
about large variable unit costs• Profit maximisation doesn’t drive out other goals• Market share maximisation leads to monopoly• Fixed unit cost markets are speculative and can co-
ordinate using “salient” prices• Naïve expectations allow co-ordination and tacit
collusion
Monopoly Learning
Dosi et al. Replication
Cost Plus Pricing
Price Following
Much Lower Unit Costs
Speculative Market: Fixed Unit Cost
Co-ordination Through Salience
Stable But Uncoordinated Market
Expectation Formation Terminals
Tacit Collusion?
Sustainable Market Shares
Three Firms with Expectations
What Next?• Better data collection: ethnographic,
experimental, participatory• More effective sensitivity analysis• Much more “joined up” research mediated
by simulation• More “middle range” theory provoked by
new approach (dynamic decision)• More infrastructure (JASSS, CRESS, S3)
Example 2: Lifestyle Emergence• Based on qualitative data about money
management among pensioners• Importance of “practices” and “lifestyles”• Almost no explicit calculation: an
excellent corrective to economics• Abstraction but inductive abstraction• Linking sequence/narrative data to
individual choice
Lifestyle Emergence Simulation• Activity plans (444411122111) and budget
plans (1111000110111)• Distinguish plan and realisation• Adaptive rule for individual comparison of
(largely unobservable) budget plans• Adaptive rule for social comparison of
observable (communicated) activity plans• Improved wellbeing and emergent lifestyles
Example 3: Social Mobility• Paradigmatic statistical (GLR) sociology
linking highly theorised concepts• Dilemma with micro/macro link
– micro theory must be “anti social” (RCT) to guarantee transparent aggregation
– Plausible micro theories have uncertain macro consequences (Schelling example)
• Simulation as a tool for integration
MOBSIM: Work in Progress• Microsimulation: agents, attributes and
updating processes (environment)• Families, schools and jobs/classes• Families: demographics and social
practices• Schools: “epoints”• Jobs: Hiring by epoints, random firing• No social networks or “economics” yet
The Scope of Models
Labour Markets
Demography
Education
??
Implications of MOBSIM• Thought provoking surprises: identification
of lacunae• Integration of diverse research• Potential falsification using within
generation (labour market surveys), qualitative biographical and sequence data
• Exploring micro/macro relations: another possible mode of falsification
The Future?• Methods: Adapting methods to simulation
– Dynamic process data– Ethnographic decision elicitation– A sociological protocol for “experimentation”
• Data: Neglected approaches to sociality– Adaptive models: innovation diffusion (drugs)– Dynamic social networks and endogeneity– Time planning and lifestyles as sequences– Selectionism: Evolving social practices
Conclusions• A genuinely novel method of representing
social processes• Inspires new developments in methodology
(the agent based approach) and the possible return of falsifiability
• Suggests new kinds of theories and represents existing debates (micro/macro)
• Uses and generates data in novel ways: synthetic