System Dynamics Methodology for

Policy Modeling: An IntroductionAsia Leds Forum 2014

Yogyakarta, Indonesia November 11-13, 2014

Muhammad Tasrif

Graduate Program in Development StudiesSchool of Architecture, Planning and Policy Development

Bandung Institute of TechnologyJl. Ganesha 10 Bandung 40132 – Indonesia

email: muhammadtasrif52@gmail.com

A. Introduction

(BOE/million Rp constant 2000)

Time

Energ

y_in

tens

ity

2.000 2.005 2.010 2.015 2.020 2.025

0,0

0,1

0,2

0,3

0,4

0,5

0,6

Historical(Why ?)

Future [-3%/year] ???

How?

A phenomenon: The behavior of Indonesian energy intensity 2000-2010

• The problem (an example)

• Two (2) aspects of a phenomenon:(1) Structure Behavior (2)

(elements constructing the phenomenon and the interactions among the elements)

(the change of a variable/element of the phenomenon over a certain period of time, quantitative or qualitative)

C

A

DB

Time

Energy intensity(A)

Social phenomenon: physical structure; and decision making structure.

Describing insights in terms of problem structure and its associated behavior is necessary to understand a problem.

• Question arising from the behavior

(a) Point prediction

(forecasting, prediction of the future)

(b) Behavior prediction- Why is the behavior following its current path?- Where is it going if we carry on as we are?- How can we design a robust strategy to radically

improve the behavior into the future?(designing strategy and formulating policy,

policy analysis/design)

Unknown process

Real world

decisionsReal world

history

Model

structureModel

behavior

• Policy modeling

Policy (intervention)? Phenomenon (Real world)

Model

Simulation

• The Requirements of A Model for Policy Design

• “It is important for policy design that the behavior of the model, which is a basis for design, and its empirical relevance are fully understood” *Saeed, 1994+.

• “The objective of a modeling exercise, therefore, may not be to obtain policies of change from the model but to understand the mechanisms of change existing in a system so that appropriate policies can be formulated” *Saeed, 1994].

• “…a useful model of a real system should be able to represent the nature of the system; it should show how changes in policies or structure will produce better or worse behavior” *Forrester 1961+.

• “System dynamics is a powerful tool with which to create realistic behavioral models for design of evolutionary change” [Saeed, 1994].

• “The system dynamics approach to complex problems focuses on feedback processes. It takes the philosophical position that feedback structures are responsible for the changes we experience over time. The premise is that dynamic behavior is consequence of system structure and will become meaningful and powerful. At this point, it may be treated as a postulate, or perhaps as a conjecture yet to be demonstrated” [Richardson, George P. & Alexander L. Pugh III , 1981].

• “…The feedback loop is a path coupling decision, action, level (or condition) of the system, and information, with the path returning to the decision point. Every decision is made within a feedback loop. The decision controls action which alters the system level which influence the decision” *Forrester, 2009+.

• There are two structures: physical and decision-making structure.

• The model structure has to be consistent with relevant descriptive knowledge of the system.

• The physical structure of the model should conform to basic physical laws.

• The decision-making structure of the model has to reflect the way decision is actually made in the system (the decision rules capture the behavior of the actors in the system).

• Desired states and actual states must be distinguished.

• The variables and relationships should have real world meanings; equations should balance dimensionally without the addition of scaling factors or parameters.

• All parameters must have real world counterparts.

B. The Premises

• Two (2) aspects of a phenomenon:(1) Structure Behavior (2)

(elements constructing the phenomenon and the interactions among the elements)

(the change of a variable/element of the phenomenon over a certain period of time, quantitative or qualitative)

C

A

DB

Time

Energy intensity(A)

Social phenomenon: physical structure; and decision making structure.

Describing insights in terms of problem structure and its associated behavior is necessary to understand a problem.

• There are two structures: physical and decision-making structure.

Decision making structure Decision making process

Decision ruleInformation Decision(action)

Decision making process(Decision making theory)

Physical structure Physical law

System state(actual state)

Desired state

Others

Decision making processes involve the dynamic phenomena. Those dynamic are resulted by theinteractions of the physical structures and decision-making structures.

• The physical structure is formed by the accumulation (stock) and flow network of people, goods, energy, and materials.

• Desicion making structure is formed by the accumulation (stock) and information flow network used by actors (human) in the system those describe the rules of their decision making processes.

• Physical structure and desicion making structure

Ability to intervene (create changes)

Events

Patterns

Structure

• The iceberg (Source: Innovation Associates)Because systemic structures generate patterns and events—but are

very difficult to see—we can imagine these three levels as a kind of

iceberg, of which events are only the tip. Because we only see the

tip of the iceberg, the events, we often let those drive our decision-

making. In reality, however, the events are the results of deeper

patterns and systemic structure.

C. Logical Framework (approach)

• Levels of understanding(Anderson, Virginia and Lauren Johnson, 1997: Systems Thinking Basics: From Concepts to Causal Loops, Pegasus Communications, Inc. MA USA)

React ! Present Witness event

“What’s the fastest way to react to this event now?”

Adapt !

Measure or track patterns of

events

“What kinds of trends or

patterns of events seem to be recurring?”

Create Change ! Future

Causal loop diagrams and other systems thinking tools

“What structures are in place that

are causing these patterns?”

Events

Structure

Patterns

Action Mode

Time Orientation

Way of Perceiving

Question You Would Ask

D. Modeling Methodology

Structure Behaviorelements,

interactions:

(1) feedback (causal loop)(2) stock (level) and flow (rate)(3) delay(4) nonlinearity

(ontological: the ways reality itself could be)

Structural ApproachSystems ThinkingSystem Dynamics

Systems Thinking and System Dynamics

E. System Dynamics Methodology

E.1 Source: System Dynamics Home Page.htm

E.1.1 System Dynamics Methodology

• System dynamics is a methodology for studying and managing complex feedback systems, such as one finds in business and other social systems.

• In fact it has been used to address practically every sort of feedback system.

• While the word system has been applied to all sorts of situations, feedback is differentiating descriptor here.

• Feedback refers to the situation of X affecting Y and Y in turn affecting X perhaps through a chain of causes and effects.

• One cannot study the link between X and Y and, independently, the link between Y and X and predict how the system behave. Only the study of the whole system as a feedback system will lead to correct results.

E.1.2 What is the relationship of Systems Thinking to System Dynamics?

• Systems thinking looks at the same type of problems from the same perspective as does system dynamics.

• The two techniques share the same causal loop mapping techniques.

• System dynamics takes the additional step of constructing computer simulation models to confirm that the structure hypothesized can lead to the observed behavior and to test the effects of alternative policies on key variables over time.

E.2 Richardson, George P. & Alexander L.

Pugh III (1981), Introduction to System

Dynamics Modeling with Dynamo,

MIT Press/Wright-Allen series in

system dynamics.

E.2.1 Overview of the System Dynamics Approach

• The system dynamics approach to complex problems focuses on feedback processes. It takes the philosophical position that feedback structures are responsible for the changes we experience over time. The premise is that dynamic behavior isconsequence of system structure and will become meaningful and powerful. At this point, it may be treated as a postulate, or perhaps as a conjecture yet to be demonstrated.

• As both a cause and a consequence of the feedback perspective, the system dynamics approach tends to look within a system for the sources of its problem behavior. Problems are not seen as being caused by external agents outside the system.

• Inventories are not assumed to oscillate merely because consumers periodically vary their orders. A ball does not bounce merely because someone drops it. A pendulum does not oscillate merely because it was displaced from the vertical. The system dynamicist prefers to take the point of view that these systems behave as they do for reasons internal to each system. A ball bounces and a pendulum oscillates because there is something about their internal structure that gives them the tendency to bounce or oscillate.

• In practice, this internal point of view results in models of feedback system that bring external agents inside the system. Customers orders become endogenous to a production system, part of the feedback structure of the system. Orders affect production; production affects orders. Part and parcel with the notion of feedback, the endogenous point of view helps to characterize the system dynamics approach.

• The are roughly seven stages in approaching a problem from the system dynamics perspective:

(1) problem identification and definition;

(2) system conceptualization;

(3) model formulation;

(4) analysis of model behavior;

(5) model evaluation;

(6) policy analysis; and

(7) model use or implementation.

• The process begins and ends with understandings of a system and its problems, so it forms a loop, not a linear progression. Figure E.2.1.1 shows these stages and the likely progression through them, together with some arrows that represent the cycling, iterative nature of the process. At a number of stages along the way one’s understanding of the system and the problem are enhanced by the modeling process, and that increased understanding further aids the modeling effort.

• Figure E.2.1.1 shows that final policy recommendations from a system dynamics study come not merely from manipulations with the formal model but also from the additional understandings one gains about the real system by iterations at a number of stages in the modeling process. Done properly, a system dynamics study should produce policy recommendations that can be presented, explained, and defended without resorting to the formal model. The model is a means to an end, and that end is understanding.

Figure E.2.1.1 Overview of the system dynamics modeling approach

Policy

implementation

Understanding

of a system

Policy

analysis

Simulation

Model

formulation

System

conceptualization

Problem

definition

E.2.2 Guidelines for Causal-loop Diagrams

1. Think of variables in causal-loop diagrams as quantities that can rise or fall, grow or decline, or be up or down. But do not worry if you can not readily think of existing measures of them. Corollaries:

a) Use nouns or noun phrases in causal-loop diagrams, not verbs. The actions are in the arrows (see Figure E.2.2.1).

b) be sure it is clear what is means to say a variable increases or decreases. (Not attitude toward crime”, but “tolerance for crime”.)

c) Do not use causal-links to mean “and then…..”

The apparent simplicity of causal-loop diagram is deceptive. It is easy for would-be modelers to go astray with them. The following suggestion may help to prevent the more common difficulties.

Figure E.2.2.1 Loops illustrating that the action in causal-loopdiagram is best left to the arrows

Rising

orders

Falling

inventory

Lengthening

delivery

delay

Shortening

delivery

delay

Rising

inventory

Falling

orders

Not:

Orders Inventory

Delivery

delay

But rather:-

-

-

2. Identify the units of the variables in causal-loop diagram, if possible. If necessary, invent some: some psychological variables might have to be thought of in “stress units” or “pressure units”, for example. Units help to focus the meaning of a phrase in a diagram.

3. Phrase most variables positively (“emotional state” rather than “depression”. It is hard to understand what it to say “depression increases” when testing link and loop polarities.

4. If a link needs explanation, disaggregate it – make it a sequence of links. For example, a study of heroin-related crime claimed a positive link from heroin price to heroin-related crime. The link is clear if disaggregated as in Figure E.2.2.2 into the sequence of positive links from heroin price to moneyrequired per addict, frequency of crimes per addict, and finally heroin-related crime. Some might feel a high price deters addict and so lowers the numberof addicts as it well might, but that is another link (see Figure E.2.2.2).

5. Beware of interpreting open loops as feedback loops. Figure E.2.2.2, for example, does not show a feedback loop.

Figure E.2.2.2 Links relating heroin price and crime

Money needed

to support habit

Frequency of

crimes per addict

Heroin-related

crime

Heroin

price

Addicts

+ +

+

+

-

F. Feedback Loop, Delay, and Nonlinearity

F.1 Causal relationships (Cause and Effect)

• A feedback structure must be established base on the causal relationship (cause and effect) between a pair of variables. In other words, a feedback structure is a causal loop (circular causality).

• The feedback structure is the model building block expressed by closed circles. The feedback loop represents the circulary link cause and effect of the variables, not the statistical correlation relationships.

• The cause-effect link of a pair of variables, in a phenomenon, should be viewed with a concept that the influence of other variables to the considered effect variable does not exist.

There are 2 types of cause and effect relationships, namely:

1) positive causal relationship, and2) negative causal relationship.

There are 2 types of feedback loops, namely:1) positive feedback loop (growth), and2) negative feedback loop (goal seeking).

While a statistical correlation between a pair of variables, in a phenomenon, derived from the data obtained through the condition that the whole variables are simultaneously changing.

F.2 Positive Causal Relationship

Birth Population

Population Birth

Births adds to the population

Adding

Population increases, birth will increase

or

Population decreases, birth will decrease

+

+

(cause)

(cause) (effect)

(effect)

Change aspect, same direction

A change in population produces a change in birth in

the same direction

F.3 Negative Causal Relationship

Death Population

Price Consumption

Deaths subtracts the population

Price increases, consumption will decrease

or

Price decreases, consumption will increase

Change aspect, opposite direction

-

-

A change in price produces a change in consumption

in the opposite direction

Substracting

F.4 Causal-loop diagram (CLD)

PopulationBirth Death

+

-+

(+) (-)

+

F.5 Level (Stock) and Rate (Flow)

• To represent activity in a feedback loop, twotypes of variables are used: level (stock) and rate(flow).

• The level variables describe the condition of the system at any particular time. The level variables accumulate the action within the system.

• The rate variables tell how fast the level are changing. The rate variables determine, not the present values of the variables, but the slope (change per time unit) of the level variables.

Level (Stock) and Rate (Flow) cont

• Levels and rates as loop sub-substructure A feedback loop consists of two distinctly different types of variables, the levels (states) and the rates (actions). Except for constants, these two are sufficient for represent a feedback loop. Both are necessary.

• Levels are integrations The level integrate (or accumulate) the result of action in a system. The level variables can not change instantaneously.

• Level are changed only by the rates A level variable is computed by the change, due the rate variables, that alters the previous value of the level. The earlier value of the level is carried forward from the previous period. It’s altered by rates that flow over the intervening time interval. The present value of a level variable can be computed without the present or previous values of any other level variables.

• Levels and rates not distinguised by units of measureThe units of measure of a variable do not distinguish between a level and a rate. The identification must recognize the difference between a variable created by integration and one that is a policy statement in the system.

Level (Stock) and Rate (Flow) cont

• Rates not instantaneously measurable No rate of flow can be measured except as an average over a period of time. No rate can, in priciple, control another rate without an intervening level variable.

• Rates dapend only on levels and constants No rate variable depends directly on any other rate variable. The rate equations (policy statements) of system are of simple algebraic form; they don’t involve time or the solution interval; they are not dependent on their own past value.

• Level variables and rate variables must alternateAny path through the structure of a system encounters alternating level and rate variables.

• Levels completely describe the system conditionOnly the values of the level variables are needed to fully describe the condition of a system. Rate variables are not needed because they can be computed from the levels.

F.6 Delay

Construction begins

Construction completed

Order Delivery

Price Price perception

Construction delay

Delivery delay

Perception delay

//

//

//

[nonlinier]

F.7 Nonlinearity

G. Energy Intensity Model(Developed base on Sterman J. David (1981). The Energy Transition and The

Economy: A System Dynamics Approach. PhD Dissertation, MIT.)

Depreciation Capital

Investment

Energy Intensityof Capital

Energy Req of Capital(ERoC)Decrease

of ERoCIncrease of ERoC

Desired Energy

Intensity

Perceived Relative Productivity

RelativeProductivity

Energy Intensity of Investment

Change

-

+

+

++

+

+

+

+

+

-

-

+

.

Marginal Revenue of Energy Marginal Cost of

Energy

+

+

-

Normal ERoC

+

Retrofit Rate

Normal EnergyIntensity of Capital

Potential Energy Intensity for Retrofit

Energy forProduction

Potential Output

Labor

Effective Capital

GDP

Capacity Utilization

Desired Production

Desired Export

Desired Labor

Population

Desired Capital

TechnologyCapital-labor

ratio

Price of Output

Wage

Inflation

Desired Price

of Output Energy Price Real

Interest Rate

Capital Charge Rate

Energy Consumption

Energy Intensity

-

+

+

+

+

+

-

+

+

+

+

- +

-

-

++

+

+

+ +

+

+

-

+

+

+ ++

+

-

+

+

+

+

-+

+

+

+

++

+

+

+

+

Unemployment

+-

+ -+

-

+

-

--

-

-+

G.1 Causal Loop Diagram of the Model

+

//

--

-

-//

+

Figure G .1 Causal loop diagram of the model

G.2 Energy Intensity (Energy Consumption)

Capital

Energy Intensityof Capital

Energy Req of Capital(ERoC)

.

Energy forProduction

Potential Output

GDP

Capacity Utilization

Desired Production

Energy Consumption

Energy Intensity

+

- +

+

+

-

+

+

+

-

Figure G.2 Determinants of Energy Consumption (Energy Intensity)

+

The short-run determinats of energy consumption (Energy Intensity) are shown

in Figure G.2.

• The energy required to operate a sector’s capital plant at normal utilization (Energy for Production) is given by the energy requirements of capital stock (ERoC).

• Variation in the planned utilization of capital (Capacity Utilization) produce proportional variations in the energy consumption to operate the sector’s capacity (Potential Output).

G.3 Energy Requirement of Capital (ERoC)

Figure G.3 Energy Requirements of Capital (ERoC)

• Figure G.3 shows the physical flows of capital and the energy requirements asosiated with that capital (ERoC).

Investment-

Depreciation Capital

Energy Intensityof Capital

Energy Req of Capital(ERoC)

Decrease of ERoC

Increase of ERoC

Energy Intensity of Investment-

+

++

+

-

Normal ERoC

+

Retrofit Rate

Normal EnergyIntensity of Capital

Potential Energy Intensity for Retrofit

-

+

+

+

++

-

+

+

+

-

+

+

-

-

+

Retrofit Potential

+

Retrofit Adjustment

Time

-

Desired Energy

Intensity

+

• Capital in the model represents both plant and equipment. Because the replacement of old, energy-intensive capital with energy-efficient capital is likely to be a major influence on the dynamics of the energy for production (energy consumption), a vintaging structure is used to represent the age distribution of capital.

• The energy required to operate each vintage of capital is determined endogenously, capturing the relative difficulty of retrofitting the oldest, least efficient capital.

• Moreover, each vintage of capital has two levels of energy requirements associated with it: the energy currently required to operate the capital, which includes the accumulated effects of retrofits, and the energy that would be required to operate the capital at its original design specifications.

• The energy requirements of the first vintage of capital are increased by the energy required to run new aquisitions (Investment). At the same time, old capital is filtering down through the aging chain and being discard, reducing the energy requirements of capital. If the new investment requires the same amount of energy to operate as the old, the average energy intensity for each age class will remain constant; if the energy required to operate new acquisitions exceeds that required for discarded capital, the average intensity will rise and vice-versa.

• Retrofits allow the energy required by existing capital to change thus allowing the energy intensity of sector to change faster than the rate at which it replace its capital stock. The decision to undertake retrofits is based on the same principles of economic optimization as decision to change the desired energy intensity of new investment, but the ability to vary the intensity of existing capital through retrofits is assumed to be lower than the ability to vary the intensity of new investment. The energy intensity achievable through retrofits is limited by the physical characteristics of existing capital, represented by the original energy requirements of the capital.

• The model allows the user to determine the degree to which physical characteristics constrain retrofits by specifying the “retrofit potential”. The retrofit potential expresses how close existing capital can be brought from its original energy intensity to the optimal intensity. If the existing capital so constrains the possibilities that no retrofits are economically feasible, the retrofit potential is zero, and energy intensity can only be changed through turnover of existing capital. If the existing capital offers no constrains to retrofits, the retrofit potential is 100% (though it still requires some time to make adjustments through retrofits, Retrofit Adjustment Time).

G.4 Desired Energy Intensity

Energy Intensityof Capital

Desired Energy

Intensity

Perceived Relative

Productivity

RelativeProductivity

+

+

+

.

Marginal Revenue of

Energy Marginal Cost of Energy

+

+

Energy forProduction

Potential Output

Price of Output

Wage

Inflation

Desired Price of Output

Energy Price

Real Interest Rate

Capital Charge Rate

+

-

-+

-

+

+

+

+

+

+

+

+

+

+

Unemployment

-

+

//

-

//Capital

+

Figure G.4 Determinants of the Desired Energy Intensity

Multibenefit Insentive

-

• The energy intensity firms would like to have (the desired energy intensity of capital) is based on explicit economic optimization: the ratio of the marginal revenue product of energy to its marginal cost (Energy Price) determines how much the current energy intensity of capital should be changed to move to the most efficient operating point.

• But the desired energy intensity is not based on the actual relative productivity of energy but a perceived value, capturing the delay in responding to change in energy prices, assessing the marginal productivity of energy, and planning for different energy intensities.

• The desired energy intensity corresponds to the energy intensity users of goods and capital request from their suppliers. The energy intensity actually available for new investment is further delayed to capture the time required for supplies to respond to changes in the requirements of their customers.

Depreciation Capital

Investment

Energy Intensityof Capital

+

+

+

+

.

Marginal Revenue of

Energy

-

Energy forProduction

Potential Output

Labor Effective Capital

GDP

Capacity Utilization

Desired Production

Desired Export

Desired Labor

Population

Desired Capital

Technology

Capital-labor ratio

Price of Output

Wage

Inflation

Desired Price of Output

Energy Price

Real Interest

RateCapital Charge

Rate+

-

+

+

+

+

+

+

+

+

-

+

+ ++

-

+

+

+

+

-

+

+

+

+

+

+

++

+

Unemployment

+

-+

-

-

-

-

-+

+

-

-

//

Figure G.5 GDP (Economy Sub Model)

-

G.5 GDP (Economy Sub Model)

+

+

• As shown in Figure G.5, potential output (production) is a two-level nested CES function. At the highest level, potential output is a Cobb-Douglas function of labor and “effective capital”.

• Effective capital is the capital stock modified by the influence of its energy requirements on its productivity. The Cobb-Douglas formulation implies an elasticity of substitution of unity between labor and effective capital. Effective capital is formed by CES function of capital and the energy requirement of capital stock.

• Note that since potential output reflects the productivity of the factors at normal utilization, the energy input to effective capital is not the actual energy consumed by the sector but the energy needed to operate the existing capital at its design specifications (Energy for Production).

• The price of output is determined by four basic factors: unit costs, the desired profit margin, supply and demand conditions, and expected inflation.

• The wage is determined by four basic factors: the relative of employment (unempoyment), expected inflation, expected productivity growth, and government wage controls.

H. Policy Analysis (an example)

Effectively current incentive & disincentive policies vsEffectively current policies+Increasing Energy Price vs

Increasing Energy Price Only

Compare curve 2 and curve 3 (relatively same behavior).

The simulation results show that the current policies is not significant to reduce energy intensity compared with increasing the energy prices approaching to the economic price.

The policy of increasing energy prices could reduce the energy intensity in the future .

Curve 1: Effectively current policiesCurve 2: Effectively current policies+Increasing energy price Curve 3: Incresing energy price only

56

Moderate increasing energy price Optimistic increasing energy price

Why the energy intensity was declining in the past, and then might increase in the future?

57

References

1. Duncan, Richard C. (1991), “The Life-Expectancy of Industrial Civilization”, SYSTEM DYNAMICS ’91 Proceedings of the 1991 International System Dynamics Conference, Bangkok-Thailand, August 27-30, 1991.

2. Forrester, Jay W. (1961), Industrial Dynamics, Cambridge, Mass.: MIT Press.3. Forrester, Jay W. (1969), Urban Dynamics, Cambridge, Mass.: MIT Press.4. Forrester, Jay W. (1971), World Dynamics, Cambridge, Mass.: Wright-Allen Press.5. Forrester, Jay W. and Peter M. Senge (1980), “Test for Building Confidence in System Dynamics

Models”, TIMS Studies in the Management Sciences.6. Forrester JW. 1990. Principles of Systems. Productivity Press, Portland, Oregon.7. Hamilton, H.R., et al. (1969), Systems Simulation for Regional Analysis, Cambridge, Mass.: MIT

Press.8. Kemeny, John G. (1959), A Philosopher Looks at Science, D.van Nostrand.9. Parkin, Michael (1996). Macroeconomics (third edition). Addison - Wesley Publishing Company, Inc..

10. Popper, Karl R. (1969), Conjectures and Refutations, Routledge and Kegan Paul.11. Richardson, G.P. & A.L. Pugh III (1981), Introduction to System Dynamics Modeling with Dynamo,

The MIT Press, Cambridge, Massachusetts.12. Saeed, K. (1984), Policy-Modelling and the Role of the Modeller, Research Paper, Industrial

Engineering & Management Division, Asian Institute of Technology, Bangkok.

13. Saeed K. 1994. Development Planning and Policy Design: A System Dynamics Approach. Avebury.14. Sasmojo, Saswinadi (2004), Sains, Teknologi, Masyarakat dan Pembangunan, Program

Pascasarjana Studi Pembangunan ITB.15. Senge, Peter M. (1990), The Fifth Discipline : the art and practice of the learning organization,

Doubleday/Currency, New York.16. Sterman, John D. (1981), The Energy Transition and The Economy: A System Dynamics Approach,

PhD Thesis, Cambridge : MIT.

Thanks