Complexity and Business Studies: Empirical Patterns, Models, Simulations, and Visualization

1
Complexity and Business Studies Empirical
Patterns, Models, Simulations, and Visualization

• Bruce Kogut, INSEAD and Columbia University
• Presentation to CETRA, The management of
  complexity, the complexity of management.
  Complexity, Management and Education

July 2007
2
An inter-organizational network Death of Swiss
Air
1995
1997
1996
1998
3
Outline

• Decomposing Complexity Networks
• Big and Small Science Explaining Patterns
• Watts and Strogatz Small Worlds
• Globalization Case of Corporate Germany
• Barabasi and Alberts Scale-Free
• Venture Capital The movie
• Growing Power of Corporate Women Simulations by
  Machine
• Comments on the future

4
The Long Tail, Edge of Chaos, Design Rules

• Business studies are filled with ideas of
  complexity.
• The 80/20 rule, criticality and creativity, rules
  generating design and structures

5
Science has found long-tail regularities in many
areas
Samples from the work of Geoff West and
collaborators see David Lane (ed), Complexity
Perspectives on Innovation and Social Change
6
Similar Findings in Business Studies
Stanley et al argue that this pattern is the
product of small correlations among
businesses. John Sutton shows that even
independence across businesses can generate the
pattern
7
Some times the value of the pattern investigation
is to give simply insight by visualization

• The graph shows that within 10 years of the
  collapse communism, the size distribution of
  Russian banks has the familiar long tail of size
  distribution found in capitalist countries.
• The insight massive economic power is quickly
  generated stochastically in a society PRIOR to
  the consolidation of political power.

Log-log plot of assets and bank rank in 1998 and
2000 Russian Banks
8
Importance of taking baseline stochasticity
seriously. Is it true that Asian economies are
corrupt because rich people are proportionally
more dominant?

• The World Bank says yes bad governance leads to
  theft by powerful people.
• However, in economies that are growing at high
  levels with lots of variance, we can expect very
  extreme tails.

9
Science and Business Studies The search for
empirical patterns and for models

• A major contribution of complexity studies has
  been to stimulate the search for empirical
  patterns and then to propose models.
• Negative findings are important.
• Often, toy models precede many of the important
  empirical patterns.
• This is the case in the small world literature.

10
Measuring Networks Large-Scale Models
Small World Networks

• Small Worlds Getting a letter to a stranger
  depends on two parameters
• The Clustering Coefficient (c) average
  proportion of closed triangles
• The average distance (L) separating nodes in the
  network

11
Small World Max Difference between Short Paths
and High Clustering Values
Less like Random Graph
Ratio average clustering coefficient of actual
average clustering coefficient to the of a random
graph of similar size and density
More Like Random Graph
No nodes rewired
Many to all nodes rewired
12
Globalization in Germany Foreign Capital and
Anglo-Saxon Governance Puts an End to Clustered
Networks

• Populist politicians say private equity investors
  are locustsbut
• SPD lowered capital gains tax on restructuring to
  0.
• Big banks and Allianz are selling their shares.
• Is this the end of the German corporatist model?

13
Using Simulations

• First, find data on who owns whom in Germany and
  build an ownership network
• Second, simulate what happens when firms sell
  their links
• Determine a rule they sell shares randomly to an
  owners already within the given network
• Then see if the topological properties change
  much.

Mexico and Brazil
14
Simulating Robustness to Globalization
Why the decline here?
We can show that German firms are more likely to
be acquired by friends than by foreign strangers
and that the corporate structure of Germany is
very robust.
Kogut and Walker, American Sociological Review,
2001.
15
Problems with Watts and Strogatz Small Worlds

• S-W based on regular graphs all nodes have the
  same degrees (edges)
• But empirical graphs are rarely regular.
• S-W model is static
• But empirical graphs evolve over time.
• Barabasi and Alberts enter and say lets grow
  graphs that are not regular.

16
Growing Networks Large-Scale Models
Scale-Free Networks
Barabási shows that the macro network structure
(i.e. degree) is highly skewed by a power law
and this skewness can be explained by a micro
behavioral rule called preferential attachment.
17
The Gotland study Sexual Promiscuity
Partner during last 12 months
Liljeros F, Edling CR, Amaral LAN Sexual
networks implications for the transmission of
sexually transmitted infections,
18
A Causal Strategy the no barking dogThis is an
example of a useful negative finding.

• Preferential Attachment gt power law
• no power law, no preferential attachment.
• Power law does not necessarily imply preferential
  attachment
• No preferential attachment, maybe power law.
• So one way inference no barking power law means
  no preferential attachment.

19
Consider other ways to grow networks by many
rules Start with 18 nodes and grow the ties
Random
Preferential
Transitivity
Propensity
20
Lets try this out to see if we can understand a
business setting.
The dataset VC transactions

• We have a log of each investment made by
  venture capital firms into target companies (like
  startups)
• It comprises historical US data, 1960 2006,
  about 240k transactions.

Along with the transaction data we have
geography, industry sector, size and quite a few
other attributes available.
21
Where to start syndications as a network

• We characterize the network as an evolving
  bipartite graph. We have two types of nodes, VC
  Firms and Target companies, where each investment
  represents a tie.

Goldman Sachs
Sequoia Capital
Google
Amazon
Yahoo
However, for now we concentrate on the VC Firms
co-investment one mode projection.
Goldman Sachs
Sequoia Capital
Published in Management Science, special issue on
complexity Kogut, Urso, Walker Emergence of the
Venture Capital Finance Market, 2007.
22
Oh No, theres no power law in degrees!!!! The
dog doesnt bark!!!
The log-log plot is not linear and there is no
power law in degrees VC syndications dont seem
to be the product of rich gets richer type
explanations.
23
A different view on network growth
New entrants deal with new entrants. Old entrants
deal with trusted friends unless they need new
expertise. If right, we should find a power law
in WEIGHTS that means repeated ties.
24
What if Investors like to co-invest with trusted
Partners
A Weighted Network? A large number of deals are
repeated ties investors tend to invest with the
same people over and over.
From Kogut, Urso, Walker (2007)
Now things start making some sense, although
can we really say something about repeated ties
at this point?
From Barrat, Barthelemy, and Vespignani (2004)
25
Generative Social ScienceFind the mechanism

• Build the network using real data.
• Estimate the Rules, such as do deals with past
  partners unless you need new skills
• Simulate and compare against the real graph

26
Complexity is also useful for simulating the
futureIs the World Changing? Can the rich be
displaced? Can economic clubs be demolished?

• Do boys clubs still dominate?
• Are national systems still strong?
• What happens when the rules of the game change?

27
Consider the case of male domination of boards of
directors

• Some countries are imposing requirements for
  diversity, e.g. Norway.
• Will rules that impose diversity increase the
  importance of women in the network relative to
  men?

28
Measuring the Macro Structural Changes
Centrality
Betweenness Centrality Model based on
communication flow A person who lies on
communication paths can control communication
flow, and is thus important. Betweenness
centrality counts the number of shortest paths
between i and k that actor j resides on.
b
a
C d e f g h
Eigenvector centrality measures if powerful
directors are tied to powerful directors related
to Googles page rank algorithm.
29
Male versus Female Centrality Simulations
30
Thinking about Change The Importance of
Visualizations and Conceptualization

• Many economic models say firms get stuck at low
  peaks of performance, e.g. women will never
  become powerful in business.
• We need therefore better ways to conceptualize
  the possibility of radical innovation as arising
  from firms that are often stuck.
• One model is called exaptive bootstrapping but
  I will let Roberto Serra tell this story.

• This simple graph is from a NK simulation and
  makes a simple point
• It is probable that most firms get stuck at low
  level peaks, where a peak represents better
  performance to nearby alternatives.

31
Making Sense of Patterns, Visualizing Alternative
Futures, Conceptualizing the Problem

• Complexity offers new metaphors for business
  long tails, edge at chaos, tipping points.
• It legitimates the coupling of the search for
  empirical patterns and models.
• And it provides powerful tools of simulation and
  visualization to help conceptualize and explain.
• And sometimes it offers a better future than the
  present.