Transcript Agent-Based Modelling of Complex Social

Agent-Based Modelling of
Complex Socio-Economic
Systems
Tomasz Michalak, 05.10.2009
based on the talk of R. Axtell: Very LargeScale Multi-Agent Systems and Emergent
Macroeconomics
Plan

Agent computing in economics and other fields

Artificial economies of adaptive agents

The new, emergent macroeconomic approach

Architecture of an artificial macroeconomy

Conclusions
Historical Background
1.
Microsimulation (e.g., Orcutt)


Small # of households (e.g., 16K memory)
No strategic behavior
2. Aspen model (1996)



elements of evolutionary learning and
parallel computing
model had ability to predict business-like cycles
(which is in general difficult for all other models)
3. ‘Financial fragility’ models of Gallegati and
coworkers (2003)


Exogenous shocks
Firms as agents
Reductionist Principle

Mainstream economics still adopts the classical
mechanics approach of 19th century physics,
based upon the reductionist principle: since
the aggregate is simply the sum of its
components, in order to understand the working
of the aggregate it is sufficient to understand the
working of each single element.
3 country New-Keynesian Model
Setting:

Two countries in a monetary union

1 outside economy
Four sectors in each economy:
(1)
Final Tradable Goods Sector
(2)
Intermediate Tradable Goods Sector
(3)
Final Non-Tradable Goods Sector
(4)
Intermediate Non-Tradable Goods Sector
230 equations  still a representative agent in each
economy and an representative firm in each sector
Economy and Society
as a Complex Machine
Classical Economic Approach:
Macroeconomists and Economists in general use two main
abstractions:
(1)
(2)
Representative agent;
Representative firm.
Main problem: whereas it is possible to evaluate a policy
on a macro level it is not possible to check its effects
on medium or meso-, micro level.
Typical example: nominal interest rate of the central bank
Conventional Econometrics

If the economic system is populated by heterogeneous agents, the
microfoundations of macroeconometrics should be redefined;

This is because some standard procedures (e.g. cointegration,
Granger-causality, impulseresponse functions of structural VARs)
loose their significance (Forni and Lippi, 1997).

All in all, we may say that macroeconomics (and
macroeconometrics) still lack sound microfoundations, Gallegati
(2003).
Power of Interaction
Paradigm of non-interactive computing:

Data

Machine (e.g., Turing machine)

Machine turns data into the answer via algorithm
Multi-agent systems: interactive computing

P Wegner: systems of interacting agents at least as powerful as a
Turing machine

Movement to rework the foundations of computer science from
perspective of interaction
Multi-disciplinary research
Economics:
•
•
•
•
•
•
•
Macroeconomics
Microeconomics
Game Theory
Evolutionary Game Theory
Social Choice Theory
Institutional Economics
Experimental/Behavioural Economics
Computer Science:
•
•
•
Computational Game Theory
Evolutionary Game Theory
Multi-Agent Systems
Emergent (Macro)Economics
Dynamical models for all important components of an economy
as a multi-agent systems (MAS)
Flavors:
 Institutions as agents
 Individuals as agents
 institutions as MAS
1.
2.
3.
4.
5.
Explicitly specify interactions between agents
Spin the whole artificial economy forward in time
Dynamic equilibria emerge
Aggregates emerge
Emergent macrovariables influence individual agents
behavior
Agent-based modelling (1)

The quantum revolution of the last century radically changed the
perspective in contemporary physics. According to the holistic
approach, the aggregate is different from the sum of its components
because of the interaction of particles.

Instead of sticking to the older paradigm, social scientists should
derive inspiration from the latest revolution in physics.

A step in this direction is the agent-based modeling strategy,
which is increasingly applied also in economics. Agent based
models have been developed to study the interaction of many
heterogeneous agents.
Agent-Based Modelling (2)

Agents are the only way for economists to fully utilize modern
machines

Code a few classes of agents and replicate

‘Small-compile time, large run-time’ model

No way fill 1 GB RAM with equations!

Agent models can be considered as richer specifications than
typical econometrics
Agent Computing in Other Fields
1.
Computer science: AI →DAI →MAS
2.
Ecology: decade of work on ‘individual based models’
(IBMs)
3.
Epidemiology: ODE models now agents
4.
Traffic: Before 1990 all traffic models were linear
programming models realized on vector
supercomputers. Today agents have displaced these
5.
Military OR: Complete transition from PDEs to agents
over past decade.
Any Artificial Economy must have…




Artificial Agents…
…have preferences, are consumers
…earn wages in firms as workers, migrate between firms
…own shares of firms
Artificial Firms…
…make products to sell to consumers and firms
…pay wages to workers
…banks as special case  “credit crunch”
Artificial Markets…
…for consumption and capital goods, prices emerge
…for ownership of firms, share prices emerge
Vital institutions emerge…
…money, price level, exchange regimes, etc.
…social norms of contracts, work effort and so on
…informal social networks
What is Feasible Today with
Agent Computing?
Simple agents on modern workstation

10^6 – 10^7 agents in C/C++

10^5 – 10^6 agents in Java

Complex agents on good workstation

10^2 – 10^5 agents in C/C++

10^1 – 10^4 agents in Java

Bigger numbers on ‘big iron’, the grid

Main limitation today is software:
Some Immediate Problems

How to get realistic institutions into such a
model?

Let them emerge…
…or build them in?


Evidence of our limited knowledge of how
agents form institutions
Ostrom: Emergence of self-governance
institutions
Satisfactory execution of this research program will
take many decades!