定例研究会報告
第16回ナショナルモデル研究会は以下の通りでした。
報告テーマ：「『韓国のSD研究』について」
金度勲他著『システムダイナミックス（原題ハングル）』を中心として
報告者：Edward
Kim（ハワイ大学大学院）
日時：1999年9月18日（土）13:30〜17:00
出席者数：
10名
発表概要：ハワイ大学大学院JAIMSのMBAコース院生Edward
Kim氏（現在修士論文作成のためCITI
BANKにて研修中）によりハングルで書かれた最新のSDテキストの紹介がなされました（発表は英語）。Kim氏が用意された詳細なレジュメをお送りいただいたのでそのまま添付します。
編集後記：今回はいつも有意義なニューズレターを書いて下さっている末武氏がバングラディシュへご出張中のため本格的なお知らせにいたりませんでした。無事お戻りになったので次回より連載を続けます。乞期待。＜事務局＞
お願い
SD学会の会員名簿を添付しました。誤記・変更等ございましたら、当事務局までお知らせ下さい。なお、ニュースレター等のご連絡をe-mailにて配信したいと思いますので、アドレスをいまだ記載されていない方はkobaken0@fps.chuo-u.ac.jpまでお知らせ下さい。
System
Dynamics
This
book was written by three authors below in order to introduce SD to Korea.The
presentation on this book is to summarize the content.Due to the lack of
presenter’s Japanese, English was used for presentation which might cause
some degree of miscommunication.
For
some parts of the book, presenter’s personal opinion was added, mainly
in the last chapter, to illustrate the possiblity of SD application other
than policy analysis.Though the content was in the same order, chapter
numbering was redone for the sake of simplicity.
Most
of summary remains the same as in the presentation except 4.2 and 4.3,
in which a bit of detail description based on presenter’s interpretation
is inserted to clarify the original content.
Authors
lKIM
Do-Hoon (dhkim@sookmyung.ac.kr)
lMOON
Tae-Hoon (thmoon@naeri.cc2.cau.ac.kr)
lKIM
Dong-Hwan (sddhkim@cau.ac.kr)
Preface
lConcise
and practical text book writing in order to introduce SD
lSD
is a proper tool for policy analysis, especially in the context of IT era
1. System
Thinking
1.1 Tragedy
of Laundry thinking
lReasons
of policy failures due to laundry thinking
?Example
- Taxation policy to suppress land speculation ? Short term effects and
long term side-effects such as inefficient land usage, rapid urbanization
1.Temporary
response ? Static understanding of a problem
2.Short
time horizon & Partial view of a situation - At most 1 or 2 years of
time horizon
3.Conflict
of interest among stakeholder bureaucrats ? Departmentalization of bureaucracy
through functional speialization
lTypical
cases of laundry thinking summarized by Donella H. Meadows
1.One
cause and one result
2.Every
growth is good and achievable ? Growth and Size complex among policy makers
3.Myth
of a garbage can ? Then where does the garbage can go?
4.Myopia
on technology ? Can technology solve every problem in a society?
5.Future
is to forecast not to choose or to create
6.Measurability
is the proof of existence ? Number, Figure, and Number
7.Economic
feasibility centered approach ? Then where is public interest?
8.Linear,
instant, and continuous relation ? Nonlinearity with time delay and discontinuity
9.More
investment for more output ? More guns for better security?
10.Standalone
system ? Every system is networked
11.Current
system is good enough to endure and it will sustain ? Every system changes.
1.2 System
thinking and Laundry thinking
lLaundry
thinking ? Unilateral casual effect relationship, No relationship among
independent variables
lSystem
thinking ? Feedback, Dynamic, and Operational thinking
1.Feedback
loop and circular casualty relation
2.Dynamic
behaviour of a system
3.Holistic
view from a distance & Detail observation at proximity
?Bilateral
casual effect relationship
?Mutual
inter-dependence among independent variables
?External
factor is a noise rather than a cause ? Endogenous attitude, Self-Responsibility
?Operational
thinking ? Factual, realistic modelling
2. Conceptual
tools for SD
2.1 SD
approach
lHeritage
of System Dynamics from Cybernetics and Servomechanism
?Cybernetics
? Feedback loop in Communication and Control
?Servomechanism
? Feedback loop in Dynamic behaviour
?Computer
Simulation
lFeedback
Structure oriented thinking
?Endogenous
rather than a exogenous
?Beyond
event driven thinning ? Historical event and its structural cause
?Beyond
a parametric understanding of system behaviour
lComparison
between statistical approach and SD
|
|
Statistical
Approach
|
System
Dynamics
|
|
Inference
method
|
Data
set from experience
|
Casual
relation among variables
|
|
Object
for analysis
|
Static
|
Dynamic
behaviour
|
|
Focus
for analysis
|
Bi-variable
relation
|
Multi-variable
circular feedback relation
|
|
Objective
for analysis
|
Numerical
accuracy
|
Structural
accuracy
|
|
Forecasting
|
Short
term
|
Long
term
|
|
Experiment
of a Policy
|
Difficult
|
Easy
|
|
Example:
Korean ETRI
Study on IT network
development in the future, 1995
|
More
accuracy on market size for short term time horizon
|
Market
structure, Consumer demand analysis, Government support policy
|
lComparison
between econometrics and SD approaches
|
|
Econometrics
Approach
|
System
Dynamics
|
|
System
& Environment
|
Open
System & Divided
|
Closed
System like an Amoeba & Closed interaction
|
|
Strength
in
|
Short
term forecast
|
Long
term forecast
|
|
Research
on
|
Equilibrium
of a system
|
Evolution
of a system structure
|
|
Knowledge
of
|
Observable
facts
|
Invisible
feedback structure
|
|
Structure
& Parameter
|
Parameter
oriented
|
Structure
oriented
|
2.2 Feedback
structure and Causal diagram
lConcept
of Casual map
?Positive
feedback ? Self reinforcing feedback, Deviation amplifying feedback
?Negative
feedback ? Goal seeking feedback, Stabilizing feedback, Self restraining
feedback
lCase
of Feedback Thinking
?Dynamics
of epidemics in Borneo revealed unanticipated surprise
lFeedback
theory in social science
1.Epidemics
paradox ? Liar paradox
2.Dialectics
? Contradiction between thesis and antithesis becomes the error for a feedback
loop
3.Self-fulfilling
prophecy ? Social belief fathers social reality.Bandwagon effect due to
psychological resonance.
lUrban
dynamics
?Urban
growth dynamics = f (Population, Natural environment, Technology development,
Expansion of social organization)
?Political
factor is regarded as an external because it tries to solve urban problems
through policy, which is exogenous.
?Conclusion
for urban dynamics analysis using SD ? There is no policy to satisfy all
aspects of urban dynamics.Policy development is to allocate priority and
to set limits on each aspect based on long-term and holistic perspective
of urban dynamics.
lArchetypes
of feedback structure
1.Accidental
Adversaries
2.Balancing
Loop
3.Drifting
Goals
4.Escalation
5.Fixes
That Fail
6.Growth
and Under-investment
7.Limits
to Success
8.Reinforcing
Loop
9.Shifting
the Burden
10.Success
to the Successful
11.Tragedy
of the Commons
?All
these archetypes are downloadable from (http://www.outsights.com/systems/arch/arch.htm)
in a zip file for iThink or Vensim format.
3. Modeling
tools for SD
3.1 Basics
of SD modelling tools
lStock/
Flowvariable
?Visible
/ Invisible on a snapshot
?Example:
A snapshot of a river - Water height is visible but in and out flow to
the river is not visible.
lFlow
variable categorization
?Growth
rate
?Decrease
rate = Stock / Average Life
?Adjustment
rate = (Goal ? Current) / Adjustment time
?Normalized
look-up variable
lTime
step
?The
smaller, the better
?Convergence
problem with Euler Integration because it is a simple linear extrapolation
method, it tends to overshoot the turning point of a curve Runge-Kutta
gives provide higher order extrapolation, looking at both the trajectory
and how the trajectory is changing to give a better solution.If accuracy
is below an acceptable tolerance, the integration interval is decreased
further until the desired accuracy is obtained.
3.2 Casual
Map and SD model
lModelling
approach
?Top-down:
Big picture, Casual loop then individual variable
?Bottom-up:
Operational thinking, Individual variable
?Current:
Graphic user interface with causal loop tracing capability
1.System
structure modelling
2.Parameter
measurement
3.Validation
- Calibration against a reference model and a surprising value of a model
?Parameter
and Unit
?Model
structure ? Feedback loop
?Model
behaviour ? Material and Information delay
?Model
boundary
4.Simulation
and Interpretation
5.Policy
implication
3.3 Modelling
of Material delay and Information delay
lHow
can multiple players fill a cup with water to a target level with the constraint
of material and information delay?
3.4 Policy
leverage and its application
lPolicy
leverage and policy interruption point
?Is
a policy feedback loop blocked by a bottleneck?
?Is
a critical mass for network externality obtained?
?Is
there any difficulty in adaptation due to material and information delay?
lCommunicability
of a SD model to policy developers
4. Application
of SD
4.1 Fluctuations
in Agricultural products
lHog
cycle
?Cobweb
theorem
?Cyclic
changes in supply quantity due to time delay of production level to price
fluctuation
?Price
elasticity of demand for hog (In Korea, hSupply
/ hDemand=
1.52 in 1982)
?Because hSupply
> hDemand,
it is necessary to stabilize price to prevent instability of hog market
?Path
dependency of demand and supply curve forms a hysterisis cycle, which means
net loss to an economic system.
lVarious
policies
1.Import
?Short
term effect and long term side effect
?Buffering
between supply and demand
2.Government
purchases in low demand and sells in high has
?Significant
effect to reduce the amplitude and frequency of fluctuation
3.Price
elasticity of demand for hog
?Most
critical factor
4.2 Ecological
dynamics
In
real situation, competition occurs with other competitors.There are several
ways to model growth under competition.One method is simply to introduce
crowding effect in the growth equation as below.However, logistic growth
curve derived from the following equation does not clearly demonstrate
the aspect of competition.
p=
Customer Population
a=
Growth coefficient
b=
Crowding effect coefficient
Logistic
growth model
Limit
of growth & S-Curve
Growth
is proportional to population P, while stress due to crowding limits infinite
growth.
Another
way is to introduce prey-predator metaphor into the analysis of competition
in a market.
x= Population of prey, y = Population of predator
a=
growth coefficient of prey
b=
coefficient of prey to become a food for predator
c=
growth rate of prey
d=
crowding effect coefficient
Equilibriumoccurs
when both groups reach zero growth rates simultaneously.
?(x,y)=
(o,o) ? unstable equilibrium
?(x,y)=
(d/c, a/b) ? Stable equilibrium
?Othercase
? Cyclic fluctuation
Prey
? Predator model
Using
a built in model of rabbit-fox population, some lessons for building up
a competitive strategy were extracted such as the importance of taking
initative in competition, sensitivity of growth with respect to word-of-mouth
conversion rate, the expression of customer loyalty resulted from satisfaction
with quality.
The
archetype diagram on the left illustrates two reinforcing feedback loops
of resoruce allocation mechanism under compeitition.When
the initial equilibrium is broken, the gap between the two will widen more
and more due to competition.
4.3 Modelling
of a game theory situation
In
a real market, there is always a certain degree of uncertainty in competitor’s
behavior as well as in environment conditions.Competitive strategy in this
kind of situation is well analyzed as a case of Probabilistic Nash Equilibrium
by Mixed strategy game using game theory by G. Tshbelis in 1989 and the
result is reviewed compared to the analysis based on SD.In general, comparison
of both methods can be summarized as in the following table.
|
|
Game
Theory
|
SD
|
|
Mechanism
|
Mutually
dependent decision
|
Feedback
loop, stock & flow
|
|
Situation
|
Game
(Player, Preference, Strategic alternatives)
|
Decision
making (Object variable, Control variable)
|
|
Alternatives
|
Discrete
|
Continuous
|
|
Focus
|
Equilibrium
status
|
Dynamic
behaviour
|
For
example, in a baseball game, pitcher will try to check a runner on the
first base while the runner will make every effort to steal the second
base.The decision of checking is depend on the pitcher’s expectation of
the runner’s possible trial to steal.Extending this to the case of Police-Driver
behaviour, following conclusions were obtained from the analysis. Payoff
matrix for this situation is given in the following table.
|
1
for Driver & 2 for Police
c1 > a1, b1>d1,
a2>b2, d2>c2
|
Police
|
|
Patrol
|
Rest
|
|
Driver
|
Speeding
|
(a1,a2)
|
(b1,b2)
|
|
No
Speeding
|
(c1,c2)
|
(d1,d2)
|
Conclusions
obtained through game theory analysis are summarized as following.Notice
that the probability of committing a certain action is in proprotion to
the attractiveness of that action, which is expressed in terms of utility.
lIncreasing
fine will not change the probability of speeding, which is expressed as
P in the following.
lIncreasing
fine will only reduce the probability of patrol, which is expressed as
Q in the following.
lProbabilistic
equilibrium as below will be reached and it will be the only one equilibrium
status.
Probability
of speeding P is equal to
lP
= (d2 ? c2) / [(d2 - c2) + (a2 ? b2)] = net utility of police’ rest when
no speeding occurs / net utility of police’ rest = function of police
utility
lTherefore
probability of speeding is proportional to the normalized intensity of
temptation for police to take a rest rather than to patrol.
Probability
of patrol Q is equal to
lQ
= (b1 ? d1) / [(b1 ? a1) +(c1 - d1)] = net utility of drivers’ speeding
when no patrol occurs / net utility of drivers’ speeding = function of
driver utility
lTherefore
probability of patroling is proportional to the normalized intensity of
temptation for drivers to do speeding.
Further
analysis by SD shows that
lThough
it reaches an equlibrium, it takes long time.
lIncreasing
fine will have enough short-term effect for policy maker to adopt.
lWhen
there is information delay, no equilibrium can be reached.
One
of the lessons confirmed from this analysis is that competitive strategy
should incorporate the expected response from competitors.Second point
is that determination and capacity to reduce a competitor’s utility in
terms of raising entry barrier and effective retaliation will be one of
the major factors for a competitor to decide its action.Thirdly, though
raising a price war would be a way to decrease the attractiveness seen
from a competitior’s viewpoint, the effect will not last because sooner
or later the competitor will develop a way to match.However, policy makers
and managers tend to take this method when they are driven by visible performance
outputs.Fourthly, retarding effect of competitor’s chasing by information
lag is still an effective measure to maintain competitive advantage in
the long run. However, with the environment of networked society, information
gap is getting narrower so that it becomes easier to overcome.
What
is confirmed from the analysis of competition under uncertain market conditions,
which is the most realistic case in this article, is the fact that it is
the knowledge gap that enables an orgainzaiton to maintain competitve advantage
over its competitors.Therefore, the gap should come from highly tacitized
knowledge base, which is thoroughly internalized in an organization, so
that the high degree of tacitness of the knowledge can successfully prevent
a competitor from immitating easily.
4.4 Urban
growth and decline
4.5 Innovative
eduation for System Thinking
lFrom
teacher oriented to student oriented education
?Teacher
oriented = Learning is the process of assimilation of knowledge
?Student
oriented = Learning is constitutive dynamic process
1.Application
Group dynamics in Kwang-Un University
2.Simulation
by SimCity, Beer game, Strategem, SimEarth, Management Flight Simulator
