Transcript Eric Beinhocker

Modelling Economic Evolution
Eric Beinhocker
McKinsey Global Institute
EC Workshop on the Development of Agent
Based Models for the Global Economy and Its
Markets
Brussels, 1 October, 2010
Copyright © 2010 McKinsey & Company, Inc.
Today’s discussion
• Facts – five empirical observations to be explained
• Proposal – economic change as evolutionary search
through physical, social, and economic design spaces
• Implications for agent-based modelling
1
Today’s discussion
• Facts – five empirical observations to be explained
• Proposal – economic change as evolutionary search
through physical, social, and economic design spaces
• Implications for agent-based modelling
2
Fact no. 1 – discontinuous economic growth
World GDP per capita, constant 1992 US$
2.5m BC to 2000 AD
15,000 BC to 2000 AD
1750 to 2000
7000
7000
7000
6000
6000
6000
5000
5000
5000
4000
4000
4000
3000
3000
3000
2000
2000
2000
1000
1000
1000
0
-2500000
-1500000
-500000
Source: J. Bradford DeLong, U. Cal. Berkeley
0
-15000 -10000 -5000
0
5000
0
1700
1800
1900
2000
2100
3
Fact no. 2 – increased order and complexity
From . . .
102 SKU economy
To . . .
1010 SKU economy
• Wal-Mart 100,000 SKUs
• Cable TV 200+ channels
• 275 breakfast cereals
4
Fact no. 3: evolutionary patterns in technology
“Add successfully as many mail
coaches as you please, you will
never get a railway thereby”
Joseph Schumpeter
5
Fact no. 4: economies are physical systems
subject to the laws of thermodynamics
Low order inputs
Interacting agents
Ordered outputs –
goods and services
(entropy locally
decreased)
• Food calories
• Fossil fuels
• Raw
materials
• Information
Economic activity is fundamentally
an order creating process
(Georgescu-Roegen)
Disordered outputs – waste
products, heat, gases
(entropy exported – universally
increasing)
6
Fact no. 5 – no one is in charge
7
Today’s discussion
• Facts – five empirical observations to be explained
• Proposal – economic change as evolutionary search
through physical, social, and economic design spaces
• Implications for agent-based modelling
8
A paradigm shift
Neoclassical economics
Complexity economics
Economies are closed, static,
linear systems in equilibrium
Economies are open, dynamic,
non-linear systems far from
equilibrium
Homogeneous agents
• Only use rational deduction
• Make no mistakes/no biases
• Already perfect, so why learn?
Heterogeneous agents
• Mix deductive/inductive
decision-making
• Subject to errors and biases
• Learn and adapt over time
Networks
Assume agents only interact
indirectly through market
mechanisms
Explicitly account for agent-toagent interactions and
relationships
Emergence
Treats micro and
macroeconomics as separate
disciplines
Sees no distinction between
micro- and macroeconomics;
macro patterns emerge from
micro behaviors and interactions
Evolution
Contains no endogenous
mechanism for creating novelty
or growth in order and complexity
Evolutionary process creates
novelty and growing order and
complexity over time
Dynamics
Agents
9
Do we need evolution in agent-based models?
Complexity economics
Economies are open, dynamic,
non-linear systems far from
equilibrium
Dynamics
Agents
Agent-based
models typically
good at this
Networks
Explicitly account for agent-toagent interactions and
relationships
Sees no distinction between
micro- and macroeconomics;
macro patterns emerge from
micro behaviors and interactions
Emergence
Evolution
Heterogeneous agents
• Mix deductive/inductive
decision-making
• Subject to errors and biases
• Learn and adapt over time
Do we also
need this?
Evolutionary process creates
novelty and growing order and
complexity over time
10
Evolution as a form of computation
Algorithms
Search algorithms
Evolutionary search
algorithms
Biological
evolution
Physical
technologies
Human social
evolution
Social
technologies
Business
Plans
Coevolution
Other types of
algorithms
Non-evolutionary search
algorithms
Other
evolution
Culture?
Other?
11
Evolution is a search algorithm for ‘fit designs’
Create a variety of
experiments
Variation
Select designs that
are ‘fit’
Selection
Amplify fit designs,
de-amplify unfit
designs
Amplification
Repeat
12
A generic model of evolution
Design space
Schema
Schema
Reader – Builder
Environment
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
0
0
0
Interactor
0
0
13
Evolution creates complexity from simplicity
Information
World
Rendering
of design
Physical
World
Order,
complexity
1
0
1
1
Energy
0
Variation,
selection,
amplification
0
1
0
0
Feedback on
fitness
0
Design encoded in a schema
Interactor in an environment
14
Applying a computational view to social systems
Design space
Schema
Schema Reader – Builder
Design
A
BUSINESS
PLAN
MegaCorp
Physical artefacts
Social structures
Economic
designs
15
Who designed the modern bicycle?
16
The reality – evolution through ‘deductive-tinkering’
17
Technologies evolve
18
Economic evolution occurs in three ‘design spaces’
Physical
technologies
Business
plans
Social
technologies
19
Business plan evolution works at three levels
Individual minds
Organizations
A?
A?
B?
D?
C?
6?
Markets
A+C?
D? E?
B+D+E?
E?
Independent
booksellers
20
What would economic evolution predict?
• Periods of
stasis/bursts of
innovation
• Spontaneous self
organization
• Increasing
economic order
(non-monotonic),
increasing
pollution
21
Today’s discussion
• Facts – five empirical observations to be explained
• Proposal – economic change as evolutionary search
through physical, social, and economic design spaces
• Implications for agent-based modelling
22
Should we include innovation processes in agentbased models?
It depends…
• Stock market model testing options for institutional structure
– PROBABLY NO
• Macro model exploring short-term options for monetary and
fiscal policy – PROBABLY NO
• Model of the financial crisis – MAYBE
• Micro model of industry dynamics – YES
• Multi decade model of climate change mitigation – YES
• Macro model of long-term growth – YES
23
Options for modelling innovation
• Exogenous, stochastic process
–What kind of stochastic process?
–No feedback from economy to innovation process
• Endogenous, increasing returns to R&D (Romer)
–Does not account for variety, complexity
–No networks, inter-relationships between innovations
–No “bursts” of innovation
• Endogenous, evolutionary
–Genetic algorithms
–Grammar models? Other?
24
Can we incorporate economic evolution in agentbased modelling?
• Imagine agents searching a ‘design space’ (physical technology,
social technology, or business plans) for ‘fit designs’
–Finite set of primitives, coded in a schema
–‘Grammar’ for re-combination of primitives into modules and
architectures
• How to model the fitness function, how does it endogenously evolve?
• Who are the schema-reader/builders? (individuals, firms?)
• How to model processes for turning schema into interactors (new
products and services, new firms)?
• How can evolution in social technologies change the structure of the
model itself?
25
Remember . . .
“Evolution is cleverer than we are”
Orgels’s second rule
26