competitive innovation and the emergence of technological epochs/adaptive agent modeling in a policy...
TRANSCRIPT
Competitive Innovation and the
Emergence of Technological Epochs
Jeremy Pesner's Excellent Lecture SeriesPart 1
Overview Simulates evolution of economic goods (innovation) Model takes cues from:
Brian Arthur (recombination of existing technologies) Eric Beinhocker (utility of technologies defined on a
fitness landscape)
Mirrors some interesting aspects of technological evolution:
Creative destruction Punctuated disequilibria (innovation waves) Human involvement (lack of technology agency)
Technological Evolution vs Biological Evolution
Technologies “inherit” properties from previous technologies, but don't reproduce
Humans drive technological evolution “Lamarckian” characteristics – Technologies change
their form over time, can “reproduce” in new context Ex: Laser was initially a measurement tool Has “reproduced” as metal cutting tool and disc-
reading mechanism
“Unnatural” selection – What people want, not what they need
Setup
Some number of agents & some number of total goods with randomly distributed fitnesses
Agents can hold and use goods with defined fitnesses. Goods have finite lifetimes
Agents invent a new good by combining 2 existing ones – new fitness value f = [0, g1 + g2]. Some probability that it is marketable
Agents evaluate their own fitness of good = [0, f]. If new good's fitness is greater than least useful good agent owns, will adopt it
Default Parameters Used In Paper (Not in Model)
Size of population (A) – 10000 Max number of goods of each agent (G) – 20 Initial number of goods (N) – 30 Probability of marketability (p) - .0001 Length of time model is run (T) – 1000 Agents are networked in small world network with p
= .01 Can talk to their neighbors about their goods
Adoption of Goods
Which goods are most used? How quickly are goods (un)adopted?
At first blush this figure looks somewhat like a colorful city at night, reflected across icy still water.
Exponential Total (Average) Fitness Growth
Implication: Newer technologies offer increasing amounts of fitness
New classes of technology
Technology fitness reliably declines (relative to average fitness)
Absolute fitness increasesexponentially with every new
technology class
Conclusions These trends persist regardless of distinct types
of technology Guns vs butter
Can enable quantification of qualitative ideas Creative destruction – ratio of goods displaced
to total goods Difficult to measure outside of model
Model has technological epochs (innovation waves)
Increasing avg fitness = more capabilities & better standard of living
A New Look at Policy Tim Gulden's (CSS prof) dissertation at UMD
People on committee: Thomas Schelling, some guy from Brookings
In 2004, ABMs not as well-known/numerous Had mostly been used to demonstrate social science
concepts But can we actually derive policy implications? Gulden says yes, and proves it (mostly) Looks at 3 different, unconnected cases. ABMs used
differently in each
ABMs: An Evolution in Models Static models: state of a system at single point in time
Price of a house today
Comparative static models: State of a system at multiple points in time
Most economic equations. Tend to make many assumptions
Systems dynamics models Trace evolution of system over time through
differential equations and software
And now, ABMs
Case 1: Ricardo Theory of Comparative Advantage
Theory: All nations have ideal equilibrium of goods they can produce & trade
Even if others nations can produce goods themselves
Gomory & Baumol: Multiple equilibria for trading arrangements
Countries produce what they do because of histories, polices
Industries may complement each other, providing further advantage to a country
A nation can produce too many goods – there should be a certain distribution
Let's see what a model can say about this...
Details of Model* Two nations, developed & developing, with agents
living in each Different employment distributions & production
functions
Barter wine & cloth, with variable exchange rate Agents get paid a certain amount, can adjust Model is robust – Compared to Paul Samuelson under
traditional economic assumptions If technology enables more efficient production for one
country, can lead to extreme disadvantage for other
*I am far from an expert on the economics of international trade
Details of Model, Part II Model has startup costs, increasing returns Agents may (with small prob.) choose to leave their
nation and work in the other When developing nation gets efficiency boost...
...nothing happens Cannot suddenly shift into high-tech industry
When nation closes borders, develops industry When borders reopen, can trade on par with
developed nation
Policy Implications Ricardo Theory: Trade, trade, trade!
Can always trade no matter production arrangements Should always trade, even if bananas
Samuelson: Significant technological advantage can harm other nations
ABM: Ehhhhhh.... Closing borders not an unreasonable idea
economically Technological efficiency is meaningless without
training/infrastructure
Case 2: Zipf Distribution of Cities Within Nation
Zipf distribution: Size of city is proportional to ordinal rank
10th largest city has roughly a tenth the population of largest city
Has defied empirical explanation
Significant exceptions: United States (New York half of what Zipf says it
should be) France (Paris way higher than Zipf says it should be) Russia (nothing matches the Zipf model)
Jar and Beans Model Jars... with beans in them
Jars face off with each other and “wager” half the beans of the smaller jar
Larger jars at more of an advantage Floor assumption for smallest jars – will not lose last
beans
Zipf distribution among jars is produced No matter initial configuration Cities “churn” and switch places often Doesn't really represent city migration
So Let's Make Some Changes Size of the bet is now a parameter of the model Some growth in smaller jars to offset design of model Some beans will never leave their jars A few other changes to reflect political situations (i.e.
Russia's limits on city migration) Cities still grow and change places unrealistically Policy implications: In countries with fewer cities and
more people, people will concentrate in “megacities” Management of middle-tier cities, broad ideas for
urban policy
Case 3: Guatemalan Civil Violence 1977-1986
Different than the previous two – this is more micro Also far, far more specific Data collected by AAAS and CIIDH Analyzed, compared to existing ABM of civil
disobedience Epstein, Steinbruner & Parker (2001) at Brookings Citizens & Cops: Citizens have parameters which
can incite them to violence Cops will arrest one random citizen within their view
per turn Can be broken into red and blue groups
Guatemalan Data Intensity and frequency of violence weakly correlated Killings most prevalent in area with high percentages
of Mayan and Ladino ethnic groups A large spike in monthly killings in 1982
A punctuated disequilibrium
Killings in both conflicts and genocide are Zipf distributed
Cause not clear, likely not the same as Zipf distribution from before
Comparison of Data to Model Brookings – Elimination of leaders = effective
repression technique Logic for weak correlation between frequency and
intensity of violence
Violence spikes in model Consistent with punctuated disequilibrium from data
Model does not produce Zipf-distributions of violence, but does produce heavy-tailed distribution
When red and blue are adversarial, there is “ethnic cleansing”
Conclusions Demonstrates insights that ABMs can generate into
policy matters In first two, enables an understanding between inputs
and outputs In third, suggests that models are in fact useful for
the subject
ABMs allow for history, bounded rationality, etc. A mix of a quantitative and qualitative perspective
Actual policy derivations not given much attention Three cases and ABM applicability very different &
disconnected