Content uploaded by Steven N. Durlauf
Author content
All content in this area was uploaded by Steven N. Durlauf on Dec 13, 2013
Content may be subject to copyright.
1
The Economy as an Evolving Complex System II.
W. Brian Arthur, Steven N. Durlauf, and David A. Lane, (Eds.)
Proceedings Volume XXVII,
Santa Fe Institute Studies in the Science of Complexity,
Reading, MA: Addison-Wesley, 1997.
Review by Gerald Silverberg, Maastricht
How does one instigate a scientific revolution, or
more modestly, a shift of scientific paradigm? This
must have been on the minds of the organizers of the
two conferences "The Economy as an Evolving Com-
plex System, I and II" and the research program in eco-
nomics at the Santa Fe Institute documented in the
present volume and its predecessor of ten years ago.(1)
Their strategy might be reconstructed as follows.
First, the stranglehold of neoclassical economics on
the Anglo-Saxon academic community since World
War II is at least partly due to the ascendancy of
mathematical rigor as the touchstone of serious eco-
nomic theorizing. Thus if one could beat the prevailing
paradigm at its own game one would immediately have
a better footing in the community than the heretics,
mostly from the left or one of the various `institu-
tional' camps, who had been sniping at it from the
sidelines all the while but were never above the suspi-
cion of not being mathematically up to comprehending
it in the first place. Second, one could enlist both
prominent representatives and path-breaking methods
from the natural sciences to legitimize the introduction
of (to economists) fresh and in some ways disturbing
approaches to the subject. This was particularly the
tack taken in 1987, where roughly equal numbers of
scientists and economists were brought together in an
extensive brain storming session. Physics has always
been the role model for other aspiring `hard' sciences,
and physicists seem to have succeeded in institutional-
izing a `permanent revolution' in their own methodol-
ogy, i.e., they are relatively less dogmatic and willing
to be more eclectic in the interests of getting results.
The fact that, with the exception of a brief chapter by
Philip Anderson in the present volume, physicists as
representatives of their discipline are no longer present,
presumably indicates that their services can now be
dispensed with in this enterprise.(2)
Finally, one should sponsor research of the highest
caliber, always laudable in itself, and make judicious
use of key personalities. Care should also be taken that
the work is of a form and style which, rather than ex-
plicitly provoking the profession, makes it appear as if
it were the natural generalization of previous main-
stream research and thus reasonably amenable to inclu-
sion in the canon. This while tacitly encouraging and
profiting from a wave of publicity in the popular me-
dia, a difficult line to tread if one does not want to ap-
pear frivolous and offend sensibilities. All of these
sometimes conflicting considerations are necessary, it
should be observed, because we are dealing with some-
thing about which the new `Santa Fe' theory itself
from its very inception has had something very rele-
vant to say, namely lock-in to an inferior technology.
For indeed the economics profession, particularly in
the United States, displays many aspects of a system
characterized by increasing returns to adoption, where it
becomes increasingly impossible, more for sociologi-
cal than for substantive reasons, to break out of a nar-
2
row mold enforced by a circular system of publication
and promotion procedures. This had already been borne
out by the experiences of both American and European
iconoclasts of the 50s to the 80s in attempting to es-
tablish bounded rationality, increasing returns, evolu-
tionary modeling, nonlinear dynamics, dissipative sys-
tems, synergetics and what have you in more than just
marginal niches in the economics research agenda.(3)
Has the Santa Fe Institute succeeded where others
have had only mixed success (but dared to tread)? This
is not as simple a question to answer as it would ap-
pear, since there are many different criteria one could
apply to measure success. In terms of popular con-
sciousness the answer is certainly yes. It has succeeded
in putting a certain style of reasoning into the media
limelight,(4) but as I have noted, this can be a double-
edged sword and thus a difficult act to balance with
professional considerations, as the example of catastro-
phe theory has amply demonstrated. In terms of the
mainstream, and in particular, the American main-
stream (which may be almost synonymous), it has
also succeeded in placing complexity onto some agen-
das, such as special sessions of the annual meetings of
the American Economic Association, without becom-
ing tainted by the patronizing odor previously associ-
ated with radical economics, Marxist economics, or
possibly feminist economics, in such forums. In the
forum that really counts, however--publication in lead-
ing mainstream journals--only now can one really iden-
tify a slow penetration, while more marginal journals
still play a major role. And the Santa Fe Institute’s
own publication and working paper series should be
recognized in themselves as significant contributions.
More remarkable is the fact that another evolutionary
school is already well established in the fast lane of the
journal superhighway, namely evolutionary game the-
ory. The reason may lie in its basically lesser radical-
ism, since it mainly resorts to radical new methods and
bounded rationality to address an immanent game theo-
retic problem, that of equilibrium selection, and return
the theoretical world to the comforts of uniqueness
while opening a new mathematical playground to the
suitably inclined.
The criteria I would finally like to apply are the in-
ternal and external substantive ones one would like to
believe should ultimately guide scientific research (and
in some long run probably do). Some of these are
enunciated with great lucidity in the editors' introduc-
tion and grouped under the following headings, which
for convenience I will separate into two categories: I.
dispersed interaction (plus: no global controller/cross-
cutting hierarchical organization); cognitive founda-
tions; structural foundations; II. continual adaptation;
perpetual novelty; out-of-equilibrium dynamics; what
counts as a problem and as a solution.
But before applying these substantive categories it is
useful to see what kinds of contributions make up this
volume. The first category consists of surveys and
research contributions of a more technical variety by
some of the principal people involved in or related to
the Santa Fe enterprise. These in turn can be classified
into primarily analytical vs. computational modeling
exercises, and form the core of what one would expect
from such a conference proceedings. The second con-
sists of the chapters by Lane and Maxfield and by
Geanakopolos, which are intriguing mixtures of busi-
ness history storytelling and verbal theorizing in the
former, heavier analytics in the latter. The third is
what, for lack of a better phrase, I would call `observa-
tions by distinguished outsiders': papers by North,
Leijonhufvud, and Anderson that discourse freestyle on
concrete economic topics related to the Santa Fe re-
search agenda, but neither themselves employ any of
the new methodological innovations nor show any
great readiness to connect to the relevant existing lit-
erature in order to suggest a step in that direction.
How do the contributions in the first category meas-
ure up to the ambitious goals outlined in the introduc-
tion? In terms of the structural considerations I think
there can be no doubt about the originality of the en-
terprise. With the exception of Arthur et al. (stock
market double auction, but with heterogeneous, bound-
edly rational agents), Shubik (game theory) and Geana-
koplos (general equilibrium), many authors do investi-
gate a wide range of (nonstandard, i.e., non star shaped,
non full strategic game theoretic, and in the meantime
non random matching) interaction structures. This is
particularly the case in the chapters by Durlauf, Ioan-
nides, Blume, and Kirman. Endogenously evolved net-
work structures seem to be the way to go for many
applications, and this book is an excellent resource to
find out which mathematical methods look promising
and what the literature already offers in terms of re-
sults.
3
Bounded rationality is another piece of the puzzle to
which more than lip service is being rendered, receiv-
ing considerable modeling attention and gradually find-
ing its place in the methodological puzzle. This is
mostly due to a loose consensus about the applicabil-
ity of certain computational methods, particularly ge-
netic algorithms and classifier systems, to the problem
of choice, behavior, and (social) learning. A number of
chapters have learning algorithms and social interaction
as their primary focus. To my mind the two most
valuable ones are Arthur et al. and Lindgren, for not
only do they address central economic and social prob-
lems (the stock market in the former, the evolution of
cooperation in the stylized form of the iterated Pris-
oner's Dilemma in the latter), but they also most ex-
plicitly take up the challenge of the second half of the
editors' catalog. First, boundedly rational behavior is
coded in high dimensional and sparely populated
spaces. In fact, one of Lindgren's models operates in an
unbounded space through the use of a clever genetic
operator: the `chromosomes' coding for the next move
on the basis of a finite history of previous moves can
double in length by self-duplication. This should be
contrasted with the analytic literature on evolutionary
games, where the strategy space is usually finite and
small (often consisting of only two strategies), and full
support is assumed (all strategies are initially repre-
sented in the population, i.e., there is no novelty). At
one fell swoop the promised land of continual adapta-
tion, perpetual novelty, and perennial out-of-
equilibrium dynamics comes into view.
Second, the emphasis on computational analysis
opens another fruitful can of worms--the question of
what constitutes an analysis and a `solution.' Take the
example of Arthur et al.'s artificial stock market. There
agents can formulate a huge variety of trading rules
which are activated with a probability proportional to
their previous success. The evolution of the market is
then the result of the continuing interaction of this
`ecology of rules' bidding and trading with each other,
against the background of another dynamic governing
the injection of new rules and the elimination of old
ones. What we get out of the model, varying with the
parameter determining these relative time scales, on the
one hand is an artificial time series which can be com-
pared with the predictions derived from a rational ex-
pectations equilibrium of an efficient markets model.
On the other it is the realization that agents' expecta-
tions remain heterogeneous and that the ecology of
rules, for high values of this parameter, does not ap-
pear to converge to some long-period statistical equi-
librium, i.e., it may not be stationary. While the time-
series properties of empirical asset prices have been
intensely studies and are still controversial (and the
authors could certainly go further in benchmarking
robust features of their model against such features as
fat tails, GARCH properties, long memory, self-
similarity, etc.), the conclusions about expectations
and nonstationarity (in terms of market `moods,' herd
effects, technical trading) are still only amenable to
`anecdotal' verification and thus may not make much of
an impression on believers in efficient markets. Seri-
ous thought still has to be given to how such models
can be tied to reality and what statistical data have to
be gathered to make their conclusions plausible, a
problem that plagues to an even greater extent the
more remote abstractions of other chapters.
What can we take away from the other computa-
tional contributions? Alas, much will be internal to
questions of modeling, I am afraid, and less about veri-
fiable aspects of the real world, but at this stage in the
game I do not think this is necessarily a damning criti-
cism of a young and rapidly developing field. Lind-
gren's treatment of the iterated Prisoner's Dilemma is
remarkable as well for its nonstationarity conclusions:
rules may come to dominant the system for long peri-
ods, but they are usually overthrown eventually. The
representation of the strategy space (e.g., memory
look-up rules vs. finite state machines) and the imposi-
tion of spatial structure can make a huge difference in
outcomes.
The contributions of Darley and Kauffman; Padgett;
Kollman, Miller, and Page; and Tesfatsion (who
should win a prize for the insolence of her title alone,
although the full connotations of the pun may be ob-
scure to non-Americans) take us into often surprising
applications and some technical thickets. What may be
an obstacle to the wider acceptance of this style of
modeling is that these applications, while dealing with
highly stylized and abstract versions of their subject,
are still too particular and idiosyncratic to lend them-
selves to easy generalization. Therefore the investment
in the modeling effort may appear excessive compared
to the robustness of the results, which lack the canoni-
4
cal appeal of such an obviously simplified representa-
tion of reality as Prisoner's Dilemma. This may be the
Modeler’s Dilemma: whether to go for a real-world
problem in all its specific gory detail, or find a highly
stylized problem which is transparent and can be ana-
lyzed in detail but remains only a suggestive metaphor
for the real world. The fate of anything in between may
simply be to get lost in the literature of high sunk
cost, one-off models.
Thus Darley and Kauffman propose replacing `ra-
tional' agents by adaptive agents, and show that com-
plex dynamics can result from their particular setup.
But will any of this structure be applicable to another
economic context, i.e., will it form part of a growing
evolutionary/adaptive toolkit? I doubt it, although the
conclusion is well taken. Padgett, and Kollman et al.
explore problems a bit far from the interests of a ma-
jority of economists, I suspect, and I am not in a posi-
tion to judge what lasting value these initiatives will
have in their domains, as technically interesting as
these models are. Tesfatsion certainly addresses those
interests more closely in her chapter on trading net-
works. The value of what she discusses, as she herself
recognizes, however, lies more in the demonstration
effect of showing how such problems can be attacked
and what difficulties arise in reaching specific results,
rather than in establishing robust conclusions. These
may well be in the offing with some combination of
simulational and analytical efforts, but it looks like a
substantial technological thicket still has to be trav-
ersed before we will come into that clearing.
I have already talked a bit about the more mathe-
matically formulated papers in terms of structural
foundations. As to specific methods, the volume pre-
sents a rich mixture of overviews and elaborations.
Lane returns to Santa Fe's inspirational source--the
Polya urn and increasing returns--to explore aspects of
the information contagion model and the tensions be-
tween individual and social choice. Durlauf and Krug-
man provide masterful overviews of statistical mechan-
ics approaches to interaction, and nonlinear dynamics
in economic geography, respectively. While Durlauf
displays a firm command of the technical jargon and
notational complexity, Krugman's chapter is a model
of nontechnical and lucid exposition, and it is a matter
of taste as to which style one prefers. In one respect
they both suffer from the `not invented here' syndrome,
however, where `here' refers perhaps less to the USA
than publication in mainstream economic journals.(5)
The chapters by Ioannides, Blume, and Kirman also
deliver high quality reviews of interaction models, both
evolutionary game theoretic and otherwise. What is
paradoxical about all these analytical approaches to
evolution, however, is that they always seem to come
back to a stochastic or systems dynamic concept of
equilibrium, and thus fall short in this respect of the
editors' vision of open endedness (although Blume does
devote a section to the question "How long is the long
run?" and the practical relevance of asymptotic results).
High theory naturally leads to high econometrics to
make the connection with data. This book provides
two examples. Brock contributes a comprehensive
guide to the time-series empirics of asset prices as well
as a tie-in to various complex dynamics models, and as
usual this review is about as definitively state-of-the-
art as we will probably ever get. Manski's chapter on
"Identification of Anonymous Endogenous Environ-
ments" is a highly abstract approach to this topic, but
I for one would have been well served by a concrete
example of an application to real data.
But this volume also treats us to two unusual--for
an economics treatise--chapters taken from the book of
life itself, so to speak, and among all of this high-level
theory, econometrics, and ALife they are a breath of
fresh air. At the same time, they illustrate in two
rather different ways just how problematic the scien-
tific enterprise may really be, and that for all the fun
and games of modeling a certain skepticism should be
brought to the question of its real, everyday decision-
making relevance. First there is Geanakoplos' reflec-
tions as an academic economic theorist on his five
years as a practitioner on Wall Street. This personal
memoir on what this turbulent experience meant
should be applauded for its self-critical openness, cul-
minating in the question (p. 291): what good did Kid-
der Peabody do, and how can it be that in a world of
rational investors it is possible to make money on the
sell side of a market, and then on the buy side of the
same market?
Yet Geanakoplos then proceeds to formulate a gen-
eral equilibrium model of his problem in which, at
least to my way of thinking, all of this turbulence and
out-of-equilibrium behavior (in the sense of the Arthur
et al. asset model) get thrown out the theoretical win-
5
dow. Evidently one cannot teach an old horse, however
brilliant, new tricks, although the model may be a
general equilibrium tour de force.
Lane and Maxfield's chapter is an equally valuable
first-hand excursion into the world of business, this
time the entrance of the Rolm Corporation into the
nascent PBX telecommunications market of the 1970s.
The first half of the chapter describes the evolution of
the PBX market as a user/producer learning process in
which no one had any entirely clear idea of what they
were doing, but where in retrospect certain strategies
proved eminently successful. This makes for a good
read, since Maxfield, like Geanakoplos, was a key
player with an inside view of mentalities and events.
The second half attempts to derive `lessons' from this
story, such as Lesson 4 (p. 186):
The "window of predictability" for the attributional
shifts and structural changes that characterize complex
foresight horizons are very short--and virtually nonex-
istent outside the particular generative relationship
from which they emerge.
Here, in their attempt at extracting some sort of
generality from their particular narrative, the authors
may have inadvertently hit on a deep problem in the
complex systems perspective when it comes up against
decision-making reality. For the danger may lie either
in its dissolving into high-flown inanities of little or
no practical utility (something I am afraid it probably
shares with much of the management literature). Or it
may wind up making such specialized predictions, un-
der such restrictive but in practice unverifiable condi-
tions, that one can never know when it will be appli-
cable. Can we ever step into the same Heraclitian river
twice?(6) And that is possibly why the two stories in
this volume seem to stand head and shoulders above
the theoretical exercises (formal or informal) their
authors seem compelled to pair them with. In the rec-
ognition that perpetual novelty and out-of-equilibrium
dynamics are the touchstones of the next modeling
revolution, may we not be forced, whether we like it or
not, to come full circle to the position of the German
Historical School in the long-buried Methodenstreit--
that each historical situation is unique, and thus only
narrative is possible and not axiomatic theory à la
physics?
I personally do not think the situation is quite as
hopeless as this (although I am always in favor of a
good narrative, and the editors should be commended
for the unusual experiment of including these two). On
the one hand there are statistical regularities, such as
Anderson mentions, which characterize large domains
of the social sciences, and are still begging for explana-
tion.(7) On the other there are a number of develop-
mental patterns and (ir)regularities in the historical
record of institutional and technological change crying
out for systematization and a basic theory, to which
North partly alludes. But it is remarkable to note that,
for a book purporting to be on the economy as an
evolving complex system, there is next to nothing on
technical change as one of the fundamental driving
forces of economic change. We have indeed come a
long way from Nelson and Winter.
In summary, this book sets out an ambitious pro-
gram to which it only partially lives up.(8) But even
in failing to achieve all of its goals it presents a spec-
trum of attempts of the very highest order. Thus the
first ten years of the Santa Fe Institute have amply
demonstrated that economists can get ALife if they
want to. The question that still remains to be an-
swered, however, is whether there is real life after
ALife.
Notes
1. Anderson, P.W., Arrow, K.J. and Pines, D.(eds.),
1988, The Economy as an Evolving Complex System,
Redwood City, CA: Addison-Wesley.
2. To be sure, computer scientists (including erst-
while physicists) and mathematical biologists have
become more actively involved in the modeling effort.
It should be noted that one of the greatest bones of
contention in the first meeting revolved around what
many physicists perceived as the economists' fetish of
mathematical rigor at the expense of empirical rele-
vance. Perhaps this was an additional argument against
a repetition of the original exercise.
3. Names worth recalling in this context are Herbert
Simon, Richard Day, Richard Nelson, Sidney Winter,
Ilya Prigogine, Herman Haken, and Wolfgang
Weidlich, to name only a few.
4. As evidenced by the success of Waldrop, M. M.,
1992, Complexity. The Emerging Science at the Edge
of Order and Chaos, New York: Simon & Schuster,
6
which, while a remarkably coherent popularization of
`complex' ideas, veers towards an almost romanticized
personalization of how science is done, and exposure in
other media.
5. Thus statistical mechanics methods and Ising-type
models of exactly the type Durlauf discusses were de-
veloped and extensively applied to the social sciences
starting in the 1970s not only by Föllmer--whom he
cites--but also by the Stuttgart school of synergetics
(Weidlich, W. and Haag, G., 1983, Concepts and Mod-
els of a Quantitative Sociology, Berlin: Springer Ver-
lag; Weidlich, W., 1991, "Physics and Social
Science— the Approach of Synergetics", Physics
Reports, 204: 1-163). And much of the complex dy-
namics in spatial models Krugman discusses was also
being intensively investigated in the 1970s by the
Brussels school (e.g., Allen, P. M. and Sanglier, M.,
1981, "Urban Evolution, Self-Organization, and Deci-
sion-making", Environment and Planning A, 13: 167-
183).
6. Cf. Winter, S.G., 1986, "Comments on Arrow
and on Lucas", in R.M. Hogarth and M.W. Reder
(eds.), The Behavioral Foundations of Economic The-
ory, special issue of The Journal of Business, 59:
S427-434.
7. The virtual neglect of fractality and self-organized
criticality (cf. Bak, P. and Chen, K., 1991, "Self-
Organized Criticality", Scientific American, January
1991: 26-33) in this volume is surprising, considering
how well established they have become in the complex
systems literature.
8. A more prosaic goal also needs some attending
to, namely, the quality of the copy editing.
