HELSINKI UNIVERSITY OF TECHNOLOGY Systems Analysis Laboratory

Sampsa Ruutu

National sea transport demand and capacity forecasting with system dynamics

Master’s Thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Technology. Espoo, October 29, 2008

Supervisor: Instructor:

Professor Ahti Salo Dr. Tech. Jean-Peter Ylén

HELSINKI UNIVERSITY OF TECHNOLOGY Author: Title: Date: Pages: Professorship: Supervisor: Instructor:

ABSTRACT OF THE MASTER’S THESIS

Sampsa Ruutu National sea transport demand and capacity forecasting with system dynamics October 29, 2008 71 Mat-2 Applied Mathematics Professor Ahti Salo Dr. Tech. Jean-Peter Ylén

The majority of Finnish foreign trade uses maritime transport as a transport unit, and this is why ports are a major part of the Finnish national transport system. A recent trend has also been rising volumes of transit traffic in Finnish ports that are further transported to Russia. In this thesis a system dynamics model is developed for the long term forecasting of the Finnish port system. The purpose of the model is to understand the causal mechanisms that determine the behavior of the entire system. The model can be used to study the effects of changes in the macro economy to the transport system. The model focuses on unit load traffic (i.e. containers, trailers and trucks) and includes the main ports through which most unit load transports pass. The proposed system dynamics model can be divided into three submodules: a) the capacity acquisition process of each port, b) the demand trend estimation process that port owners go through in order to determine the level of desired capacity in a given port and c) the port market share determination process. The total transport demand in the system dynamics model is an exogenous variable which can follow different paths of development corresponding to various macroeconomic scenarios.

Keywords:

System dynamics, logistics, transport, forecasting

i

TEKNILLINEN KORKEAKOULU Tekijä: Otsikko: Päiväys: Sivumäärä: Professuuri: Valvoja: Ohjaaja:

DIPLOMITYÖN TIIVISTELMÄ

Sampsa Ruutu Valtakunnallisten merikuljetusten kysynnän ja kapasiteetin ennustaminen systeemidynamiikan avulla 29. lokakuuta 2008 71 Mat-2 Sovellettu matematiikka Professori Ahti Salo TkT Jean-Peter Ylén

Suurin osa Suomen ulkomaankaupasta käyttää merikuljetuksia. Tämän vuoksi satamilla on merkittävä osuus valtakunnallisessa kuljetusjärjestelmässä. Viimeaikoina satamissa on ollut myös kasvava määrä transitokuljetuksia, jotka kulkevat Suomen kautta Venäjälle. Tässä diplomityössä rakennetaan systeemidynamiikkamalli valtakunnallisen satamajärjestelmän pitkän ajan ennustamista varten. Mallin tarkoituksena on parantaa ymmärrystä kausaalimekanismeista, jotka määrittävät kokonaisjärjestelmän käyttäytymisen. Mallia voidaan käyttää makrotaloudellisten muutosten vaikutusten tutkimiseen liikennejärjestelmässä. Mallissa käsitellään yksikkökuljetuksia (kontteja, perävaunuja ja rekkoja), ja mallissa on kuvattu tärkeimmät satamat, joiden kautta yksikkökuljetukset kulkevat. Työssä esitetty systeemidynamiikkamalli voidaan jakaa kolmeen osaan: a) satamakohtaiseen kapasiteetin muodostumisprosessiin, b) kysynnän ennustusprosessiin, jonka perusteella satamien omistajat päättävät halutun kapasiteetin määrän satamassaan sekä c) satamien markkinaosuuksien muodostusprosessiin. Kokonaiskuljetuskysyntä on systeemidynamiikkamallissa ulkoinen muuttuja, joka voi noudattaa eri makrotaloudellisia skenaarioita vastaavia kehityskulkuja.

Avainsanat:

Systeemidynamiikka, logistiikka, kuljetus, ennustaminen

ii

Acknowledgements This Master’s thesis has been done in the System Dynamics team at VTT Technical Research Centre of Finland as part of the LOGSUO project. The client of the project has been the Finnish Defence Forces, whom I wish to thank for their cooperation. I want to thank my instructor, Dr. Tech. Jean-Peter Ylén for his guidance and insight. I wish to thank my colleagues at VTT for their comments and for creating a good and inspiring working environment. I would also like to thank the supervisor of this thesis, professor Ahti Salo, for his guidance. Finally, I would like to thank my family and friends for their support during my studies.

Espoo, October 29, 2008

Sampsa Ruutu

iii

Contents Abbreviations

vii

List of Figures

ix

List of Tables

x

1

Introduction

1

1.1

Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.2

Research Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

1.2.1

Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

1.2.2

Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

1.3

Methods and Data Acquisition . . . . . . . . . . . . . . . . . . . . . .

3

1.4

Scientific contribution of thesis . . . . . . . . . . . . . . . . . . . . . .

4

2

System dynamics and forecasting

5

2.1

Statistical forecasting methods . . . . . . . . . . . . . . . . . . . . . .

5

2.1.1

Time series models . . . . . . . . . . . . . . . . . . . . . . . . .

5

2.1.2

Regression models . . . . . . . . . . . . . . . . . . . . . . . . .

6

2.1.3

Neural network models . . . . . . . . . . . . . . . . . . . . . .

6

System dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

2.2.1

Dynamic vs. static models . . . . . . . . . . . . . . . . . . . . .

7

2.2.2

Feedback and causal loop diagrams . . . . . . . . . . . . . . .

8

2.2.3

Stocks and flows . . . . . . . . . . . . . . . . . . . . . . . . . .

8

2.2.4

Time delays . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9

2.2.5

Mental models and modeling human decision making . . . .

9

2.2

iv

2.3

3

4

Structure and behavior of dynamic systems . . . . . . . . . . . . . . .

11

2.3.1

Analyzing loop dominance . . . . . . . . . . . . . . . . . . . .

12

2.4

System dynamics modeling of capacity and demand . . . . . . . . . .

14

2.5

System dynamics in economics . . . . . . . . . . . . . . . . . . . . . .

15

2.6

System dynamics in supply chain management . . . . . . . . . . . . .

15

2.6.1

International supply chain management . . . . . . . . . . . . .

16

2.6.2

Inventory management . . . . . . . . . . . . . . . . . . . . . .

16

2.6.3

Demand amplification . . . . . . . . . . . . . . . . . . . . . . .

17

2.6.4

Supply chain re-engineering and design . . . . . . . . . . . . .

17

2.6.5

System dynamics combined with other disciplines . . . . . . .

17

2.7

Transport models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.8

Modeling uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.8.1

19

Scenario analysis . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis of the operating environment

20

3.1

Finnish foreign trade traffic . . . . . . . . . . . . . . . . . . . . . . . .

20

3.2

Finnish economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

3.2.1

Gross domestic product . . . . . . . . . . . . . . . . . . . . . .

22

3.2.2

Industry GDP shares . . . . . . . . . . . . . . . . . . . . . . . .

24

3.3

Transit traffic to Russia . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

3.4

Transport capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

Model of port demand and capacity

27

4.1

Total transport demand . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

4.1.1

Foreign trade sea transportation . . . . . . . . . . . . . . . . .

28

4.1.2

Transit traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

Port demand and capacity . . . . . . . . . . . . . . . . . . . . . . . . .

33

4.2.1

Port market shares . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.2.2

Port demand forecasting . . . . . . . . . . . . . . . . . . . . . .

37

4.2.3

Capacity acquisition . . . . . . . . . . . . . . . . . . . . . . . .

39

Model validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

4.2

4.3

v

5

4.3.1

Structural validation . . . . . . . . . . . . . . . . . . . . . . . .

42

4.3.2

Model calibration . . . . . . . . . . . . . . . . . . . . . . . . . .

42

Simulation results

45

5.1

Sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.1

Sensitivity simulation S1: Years 1992-2025 . . . . . . . . . . . .

50

5.1.2

Sensitivity simulation S2: Effect of port model variables . . .

52

5.1.3

Sensitivity simulation S3: Effect of port model variables and total traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

Sensitivity simulation S4: Effect of total traffic . . . . . . . . .

54

5.1.4 6

Discussion and conclusions

62

6.1

Summary and review of objectives . . . . . . . . . . . . . . . . . . . .

62

6.2

Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

63

Bibliography

69

A Selected systems archetypes

70

A.1 Success to the successful . . . . . . . . . . . . . . . . . . . . . . . . . .

70

A.2 Growth and underinvestment . . . . . . . . . . . . . . . . . . . . . . .

71

vi

Abbreviations and nomenclature BSR

Baltic Sea Region

SCM

Supply Chain Management

SD

System Dynamics

GDP

Gross domestic product. The value of goods and services produced in a country per year.

Disturbance

A short term event which causes difficulties in the functioning of the system

Unit load

A container, trailer or truck

Transport unit

In this study refers only to unit load shipments (containers, trucks, trailers)

Scenario

A possible path of development in the forecasting model

Base/ BAU (Businessas-usual) scenario

The likely future development, assuming the continuation of historical trends.

Robustness

The ability of the system to maintain its performance when disturbances occur.

vii

List of Figures 1.1

Interrelation of the different models . . . . . . . . . . . . . . . . . . .

2

2.1

Causal loop diagrams of positive and negative feedback . . . . . . .

8

2.2

A stock variable with an inflow and outflow . . . . . . . . . . . . . .

9

2.3

Impulse responses of delays of different order . . . . . . . . . . . . .

10

2.4

TREND function [Sterman, 2000] . . . . . . . . . . . . . . . . . . . . .

11

2.5

Fundamental modes of dynamic behavior [Sterman, 2000] . . . . . .

13

3.1

Traffic in ports (Source: Finnish Maritime Administration) . . . . . .

21

3.2

Factors affecting GDP [Ministry of Labour, 2007] . . . . . . . . . . . .

23

3.3

GDP index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

3.4

Industry shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

3.5

Transit traffic

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

4.1

Submodules of the model . . . . . . . . . . . . . . . . . . . . . . . . .

27

4.2

Transport unit shares . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

4.3

Forecast of total foreign trade traffic . . . . . . . . . . . . . . . . . . .

31

4.4

Forecast of transit traffic . . . . . . . . . . . . . . . . . . . . . . . . . .

33

4.5

Variation of exogenous inputs . . . . . . . . . . . . . . . . . . . . . . .

33

4.6

Causal diagram of the port demand and capacity model . . . . . . .

34

4.7

Port market shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.8

Port attractiveness

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

4.9

Port demand forecast with a TREND function . . . . . . . . . . . . .

38

4.10 Port capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

4.11 Total foreign trade calibration . . . . . . . . . . . . . . . . . . . . . . .

43

viii

4.12 Port traffic calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . .

44

5.1

Simulation results of port capacity . . . . . . . . . . . . . . . . . . . .

46

5.2

Simulation results of port traffic

. . . . . . . . . . . . . . . . . . . . .

47

5.3

Simulation results of port capacity usage . . . . . . . . . . . . . . . .

48

5.4

Simulation results of port market shares . . . . . . . . . . . . . . . . .

49

5.5

Port traffic in sensitivity simulation S1 . . . . . . . . . . . . . . . . . .

52

5.6

Port market share in sensitivity simulation S1

. . . . . . . . . . . . .

52

5.7

Helsinki traffic correlations in sensitivity simulation S1 . . . . . . . .

53

5.8

Kotka traffic correlations in sensitivity simulation S1 . . . . . . . . . .

53

5.9

Port capacity in sensitivity simulation S2 . . . . . . . . . . . . . . . .

55

5.10 Port traffic in sensitivity simulation S2 . . . . . . . . . . . . . . . . . .

55

5.11 Port market share in sensitivity simulation S2 . . . . . . . . . . . . . .

56

5.12 Port capacity in sensitivity simulation S3 . . . . . . . . . . . . . . . .

57

5.13 Port traffic in sensitivity simulation S3 . . . . . . . . . . . . . . . . . .

57

5.14 Port market share in sensitivity simulation S3 . . . . . . . . . . . . . .

58

5.15 Helsinki traffic correlations in sensitivity simulation S3 . . . . . . . .

58

5.16 Helsinki market share correlations in sensitivity simulation S3 . . . .

59

5.17 Port market share in sensitivity simulation S4

60

. . . . . . . . . . . . .

5.18 Histogram of Helsinki’s capacity at the end of the simulation

. . . .

60

5.19 Histogram of Helsinki’s traffic at the end of the simulation . . . . . .

61

5.20 Histogram of Helsinki’s market share at the end of the simulation

.

61

A.1 Success to the successful . . . . . . . . . . . . . . . . . . . . . . . . . .

70

A.2 Growth and underinvestment

71

. . . . . . . . . . . . . . . . . . . . . .

ix

List of Tables 1.1

Model boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

2.1

A taxonomy of research and development on SD modeling in SCM [Angerhofer and Angelides, 2000] . . . . . . . . . . . . . . . . . . . .

16

5.1

Variables in sensitivity simulation S1

. . . . . . . . . . . . . . . . . .

51

5.2

Variables in sensitivity simulation S2

. . . . . . . . . . . . . . . . . .

54

5.3

Variables in sensitivity simulation S3

. . . . . . . . . . . . . . . . . .

56

5.4

Variables in sensitivity simulation S4

. . . . . . . . . . . . . . . . . .

59

x

Chapter 1

Introduction 1.1

Background

The development process of logistics systems in the Finnish Defense Forces is slow, which makes it difficult to cope with rapid changes in the operational environment. The current development trend in logistics is toward an adaptive mode of operation (sense and response logistics). However, the transition into an adaptive mode of operation requires improvement in the ability to anticipate long term trends and changes in the operational environment. A goal for the Finnish Defense Forces is therefore to improve the ability to anticipate and monitor the development of logistics capacity of the Finnish society, industry and commerce. To achieve this, a framework for forecasting and monitoring is needed. To address this need, computational forecasting methods and software solutions for their implementation are developed at VTT Technical Research Centre of Finland. The users of these tools will be the Finnish Defense Forces and the Finnish National Emergency Supply Agency (NESA), which is a government body under the Ministry of Trade and Industry. The model developed in this thesis is a part of a larger model of the logistics system. The entire model can be divided into the following parts: • Forecasting model • Disturbance model • Robustness model The interrelation of the different models can be seen in Figure 1.1. This thesis has been done as a part of the LOGSUO project at VTT Technical Research Centre of Finland and focuses on the forecasting model (section 1.2). In the disturbance model the impact of various disturbances is modeled. Here the term disturbance refers to short term events that cause difficulties in the functioning of 1

CHAPTER 1. INTRODUCTION

2

Scenario

Initial data

Forecasting model

Disturbance

Disturbance model

Capacity forecast

Robustness model

Robustness forecast

Figure 1.1: Interrelation of the different models the system. In the robustness model the focus is on analyzing the ability of the system to maintain its performance in the event of sudden disturbances.

1.2 1.2.1

Research Problem Objectives

The aim of this this thesis is to construct the capacity forecasting model in Figure 1.1. The purpose of the forecasting model is to provide the disturbance and robustness models with a long term forecast of logistics capacity. The capacity forecast is used as initial data for the disturbance and robustness models when they are used to evaluate the impacts of an external short-term disturbance on the logistics system and how well the system can overcome this disturbance. There is no feedback between the forecast model and the other two models, i.e., it is assumed that an external disturbance does not influence the capacity forecast. This is because of the different time scales of the models. The time scale of the forecasting model is 15–20 years, while the length of an external disturbance is measured in days. The system is also assumed to return to its prior state after the disturbance has been overcome. The inclusion of the aforementioned feedback is irrelevant. Furthermore, it would unnecessarily complicate the modeling procedure since very long and very short time constants would have to be included in the simulation model.

CHAPTER 1. INTRODUCTION

1.2.2

3

Scope

The model includes sea transports that pass through Finnish ports. Road, railway and air shipments are not included in the model. The transport units that are included are containers, trucks and trailers. These are the transport units that are mainly used for products the availability of which is important in possible disturbance situations. Major ports in Finland are modeled separately, but ports in other countries are not. Shipments in the Finnish ports are not divided based on their origin or destination, but Russian transit traffic and Finnish foreign trade traffic are treated separately. Furthermore, the capacities of vessels that operate routes in the Baltic Sea are not included in this study. While they are of interest to the client of the project, these are analyzed in a separate study. There boundary of the model is shown in Table 1.1. In the model, port capacities and demands are modeled endogenously, but the total transport demand for all ports in Finland is modeled exogenously. Table 1.1: Model boundary

1.3

Endogenous

Exogenous

Excluded

◦ Port capacity ◦ Port investments ◦ Utilization rates of ports ◦ Port foreign trade traffic ◦ Port transit traffic ◦ Port demand estimation ◦ Port market shares

◦ Total transport demand

◦ Road, railway and air transportation

Methods and Data Acquisition

The author of this thesis has been in the role of the knowledge engineer [Ljung and Glad, 1994], in charge of building the model based on the facts about the system. Domain experts in this research project are researchers from the Centre for Maritime Studies at the University of Turku and Technical University of Tampere. Data acquisition and preprocessing has been done in collaboration with the domain experts. Numerical data is only one type of information source in model building. In addition to numerical data, much information can be found in qualitative written data about the system. A third type of data is expert knowledge, which can be elicited through interviews with the domain experts.

CHAPTER 1. INTRODUCTION

4

The chosen method for the model is system dynamics, which allows the inclusion of “soft” variables to the model. Soft variables are often difficult to quantify and measure, and include aspects such as human decision making. System dynamics also allows the modeler to include knowledge about the structure of the system, unlike non-structural methods such as time-series models. In the LOGSUO project the focus is on developing an understanding of the entire logistics system, and this is why it is not appropriate to analyze the logistics network in detail with methods such as network equilibrium models.

1.4

Scientific contribution of thesis

System dynamics has been widely used to model different types of capacity investments and endogenous demand (see section 2.4). However, in many cases the capacity studied in the model is treated as a single entity. A difference in the model developed in this thesis is that the interactions of the capacities of different ports is modeled as well as the interaction of capacity and demand. In recent years several formal methods have been developed to analyze the structure and behavior of system dynamics models (see subsection 2.3.1) and have been tested on simple models. However, the use of these methods is rare in larger and more complex modeling projects. In this thesis one such method, “statistical screening” [Ford and Flynn, 2005], is used to analyze the structure and behavior of the implemented system dynamics model.

Chapter 2

System dynamics and forecasting One way to classify mathematical models is to divide them in terms of the known structure that is included in the model. In structural models the structure is known, whereas in non-structural (statistical) models the structure of the system is not known and the model is constructed based on measurements of the system. Methods for structural forecasting include system dynamics and econometrics. In econometrics the model can be based on economic theory.

2.1

Statistical forecasting methods

Traditional statistical forecasting methods include time series models and regression models. In recent years, more developed methods such as neural networks have also become increasingly popular since they are well suited to non-linear systems. Statistical (non-structural models) are constructed using statistical techniques and thus do not require understanding about the structure and causal relationships of the system. However, known relationships between variables can be used when choosing the independent variables of a regression model or the time lags of a time series model. Statistical methods can provide good forecasts if historical trends and seasonal occurrences that are found in the data continue also in the future. However, statistical models are unable to predict structural changes in the system which lead to a completely new kind of behavior. Long-term statistical forecasts will thus eventually be unsatisfactory, as pointed out by Granger and Jeon [2007].

2.1.1

Time series models

Time series models are one type of a statistical model. In time series forecasting, a forecast for a particular variable is made based on a time series of its earlier values. Commonly used forms of linear time series models are ARMA (autoregressive

5

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

6

moving average) models and their generalizations.

2.1.2

Regression models

Regression models are also statistical models. In regression analysis a dependent variable is modeled as a function of one or more independent variables. Regression models can be divided into linear and nonlinear regression models. Knowledge about the system can be used in the modeling when choosing the appropriate independent variables. For example, Fite et al. [2002] use a linear multivariate regression model to forecast freight demand where the independent variables correspond to various economic indices. Another example of regression analysis applied to traffic forecasting is presented by Garrido and Mahmassani [2000] who use a multinomial probit model to forecast freight transport demand.

2.1.3

Neural network models

Neural networks are statistical models that are well suited to model non-linear relationships. In recent research, neural networks have been used extensively in logistics and transport demand forecasting (eg. Zhou and Clausen [2007], Lam et al. [2004], Bowersox et al. [2003], Rodrigues et al. [2005], Celik [2004]). Nijkamp et al. [2004] compare neural network models with models of discrete choice (logit and probit). Aminian et al. [2006] present a brief review of recent research on neural networks in economic forecasting. Dougherty [1995] present a review of research where neural networks have been applied to transport, including transport demand forecasting among other research areas. According to Dougherty [1995] neural networks are a promising tool in analyzing transport problems, which are often highly non-linear. The disadvantages of neural network models according to Nijkamp et al. [2004] are the following: Firstly, it is not certain that in the learning procedure (i.e. estimation of the parameters) a global minimum is reached. Secondly, it is not not easy to understand what the structure of the system is from the weights of the neural network model (in contrast with a traditional regression model).

2.2

System dynamics

Structural models, such as system dynamics, incorporate understanding of the system into the model. In system dynamics feedbacks between variables are modeled. Lyneis [2000] claims that one way in which system dynamics can be useful in market forecasting and structural analysis is that it helps to understand the causes of industry behavior. Lyneis [2000] also claims that the behavior in many industries

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

7

in determined to a greater extent by the industry structure than by exogenous economic factors. In systems thinking [Senge, 1990], systems are analyzed as a whole rather than reducing them them into their components since the interactions and feedbacks between the parts of the system determine the behavior of the system. Sterman [2000] calls the complexity arising from the interactions of the parts of the system dynamic complexity in contrast with combinatorial complexity that has to do with systems with many components. System dynamics combines systems thinking with formal computer simulations, and was originally developed by Jay Forrester [Forrester, 1961] at MIT. Computer simulations can provide a means of understanding the structure and behavior of a system. This is because complex systems can behave in counterintuitive ways where actions result in a different kind of behavior than intended. Humans always have certain mental models (assumptions of how the system functions), and simulations can improve these mental models. System dynamics has its roots in control theory. However, system dynamics models are usually nonlinear and cannot be linearized to a particular operating point since they cannot be assumed to be in equilibrium. Well developed analytical methods exist in linear control theory, but not for all non-linear systems, which is why computer simulations are used in system dynamics modeling instead. Another contrast with system dynamics with control engineering is the wide range of applications of system dynamics models, which are often used to model nontechnical systems with “soft” variables which are hard to quantify and measure.

2.2.1

Dynamic vs. static models

A static relationship between two variables is one in which a change of one variable corresponds to an simultaneous change in the other. A static relationship can be modeled with an algebraic equation, eg. pV = N RT for the relationship between pressure p, volume V and temperature T . A dynamic relationship on the other hand must be modeled with a differential equation (or difference equation in the discrete case) since the system has a “memory” and a change in one variable does not change the value of the other instantaneously. Static relationships are much simpler to model, and this is why in many cases the dynamics are omitted completely. For example, static economic supply and demand curves only take into account the equilibrium price and not how long it takes to reach the given equilibrium [Smith and van Ackere, 2002]. The behavior of the system can also be different in the short and long term, which is the case with “Worse-before-better”-systems. “Worse-before-better”-behavior can be problematic if decision makers are unwilling to make decisions where short term performance has to be sacrificed to achieve better performance in the long term.

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

2.2.2

8

Feedback and causal loop diagrams

In an open loop system, the inputs influence the outputs of the system, but the outputs do not influence the inputs. This is not the case in feedback systems. Figure 2.1 shows causal loop diagrams of positive and negative feedback. In positive feedback, a change in one variable (here A1) causes a change in the same direction in another variable (here B1), which in turn causes a change in the same direction of the first variable. There is also a positive feedback loop when a change in variable A1 causes a change in the opposite direction in variable B1, and this causes a change in the opposite direction in A1. In negative feedback a change in variable B2 causes a change in the opposite direction in variable A2. The sign + is used for change in the same direction and the sign − for change in the opposite direction. The feedback loops are marked with R for reinforcing (positive) feedback and B for balancing (negative) feedback.

+

+ A1

B1

R

A2

B

B2

-

+

(a) Positive feedback

(b) Negative feedback

Figure 2.1: Causal loop diagrams of positive and negative feedback

2.2.3

Stocks and flows

Stock variables can only change through inflows and outflows, and this gives the system inertia and causes delays. Figure 2.2 shows a stock with an inflow and an outflow. The relation between stock and flow variables are given by the following integral and differential equations: Z

t

[Inf low(s) − Outf low(s)] ds + Stock(t0 )

Stock(t) =

(2.1)

t0

d(Stock(t)) =Inf low(t) − Outf low(t) dt

(2.2)

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

9

Stock Inflow

Outflow

Figure 2.2: A stock variable with an inflow and outflow

2.2.4

Time delays

A delay is a conversion process between an inflow and an outflow [Forrester, 1961] which causes the output of the system to lag behind the input. A delay can be characterized by its average delay time D and its transient response which defines how the time shape of the outflow is related to that of the inflow. In SD models level variables are a source of delays. Delays can be further divided into material and information delays [Sterman, 2000]. Material delays are used to model material flows in the system, where the amount of material in the system stays constant. Information delays do not have this property, and are used to model things like expectation formation (eg. with exponential smoothing techniques). Exponential delays of different orders are commonly used in SD models. In Figure 2.3 the impulse responses of different exponential material delays can be seen. The delay time in Figure 2.3 is the same for each delay process, but the orders of the delays vary. As the order of the delay n → ∞, the delay approaches a pipeline delay, where the delay time is constant. For a pipeline delay: Outf low(t) = Inf low(t − Delay)

(2.3)

Delays within feedback loops are often the cause of oscillation in systems (see Figure 2.5). Oscillation can occur in negative feedback loops when decision makers do not take into account relevant delays.

2.2.5

Mental models and modeling human decision making

Rather than having a perfect understanding of the world, each person has some interpretation of how the world functions. These interpretations are called mental models and influence the way we think and make decisions. However, our mental models for complex systems are incomplete and flawed. Modeling can raise our understanding of some of our mental models which we did not even know existed. Also, as our understanding of the system improves via modeling and simulation, we can update our existing mental models.

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

10 Order 1 Order 2 Order 3 Order 4 Order 5 Order 10

0.04

0.035

0.03

0.025

0.02

0.015

0.01

0.005

0

200

400

600

800

1000

1200

1400

1600

1800

Figure 2.3: Impulse responses of delays of different order Bounded rationality When human behavior is modeled, humans cannot be assumed to behave rationally because of their limited understanding of the system. Instead, humans operate based on bounded rationality [Simon, 1982], which is limited by their lack of understanding of the system. Even if more information could lead to better decisions, processing new information takes time and in practice decisions cannot be postponed indefinitely but have to made under limited knowledge. Expectation formation People update their assumptions of a system with a delay when something in the system changes. In SD models a TREND function (Figure 2.4) is often used to model human expectation formation. According to Sterman [2000], the TREND function represents a behavioral theory of how people form expectations. It takes into account three delays in the process: • Time required to collect and analyze data (Time to perceive present condition) • Historic time horizon (Time horizon for reference condition) • Time to react to changes in the growth rate (Time to perceive trend)

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

11

A causal diagram for the TREND function is shown in Figure 2.4.

Change in RC

Reference Condition (RC)

Time Horizon for Reference Condition (THRC)

INPUT

Change in PPC

Time to Perceive Trend (TPT)

Indicated Trend (ITREND)

Change in TREND

Perceived Trend (TREND)

OUTPUT

Perceived Present Condition (PPC)

Time to Perceive Present Condition (TPPC)

Figure 2.4: TREND function [Sterman, 2000]

2.3

Structure and behavior of dynamic systems

Sterman [2000] identifies six fundamental modes of dynamic behavior. The causal diagrams and the shape of the resulting behaviors can be seen in Figure 2.5. • Exponential growth is caused by positive feedback. The net increase rate increases as the state of the system increases. • Goal seeking is caused by negative feedback. A corrective action is taken based on the discrepancy between the state of the system and the desired state. • S-Shaped growth results from the combination of a positive and a negative feedback loop. The positive feedback loop dominates when the state of the system is not close to its carrying capacity resulting in accelerating growth. When the system approaches its carrying capacity, the dominance shifts to the negative feedback loop resulting in goal seeking behavior. • Oscillation is caused by a negative feedback with a delay. Due to the delay too much corrective action is taken and the state of the system overshoots the desired state, leading to corrective actions in the opposite direction. • Growth with overshoot is similar to s-shaped growth but there is a delay in the negative feedback loop causing the overshoot and the subsequent oscillation.

CHAPTER 2. SYSTEM DYNAMICS AND FORECASTING

12

• Overshoot and collapse is similar to s-shaped growth but now the carrying capacity decreases as the state of the system grows, adding another negative feedback loop to the system. Senge [1990] also presents several system archetypes. Each archetype corresponds to a generic system where the behavior is determined by one or more feedback loops. Some of these are identical to Sterman’s fundamental modes of dynamic behavior.

2.3.1

Analyzing loop dominance

In nonlinear systems the gains of feedback loops can be different depending on the state of the system, and this is why the feedback dominance can also change. Several methods for analyzing the the structure and behavior have been developed. However, many of the methods still lack automated tools which would make their implementation easy. Different methods to analyze loop dominance are presented below. Ford [1999] presents a behavioral approach to analyzing loop dominance. He identifies three atomic behaviors which correspond to different values (=0, >0 and 2007

(4.13)

ω2

∈ [0, 1]

(4.14)

d

∈ {Import, Export}

(4.15)

Transport direction: Share of transport units

The shares of the different transport units of interest to the model must also be taken into account. The total amount of foreign trade traffic tonnes are allocated to the different transport units (containers, trailers and trucks). The historical development the these transport unit shares is shown in Figure 4.2. In the model these are assumed to stay constant in the future, since no reliable forecast of their development was available. Transport unit shares of foreign trade 0.6

4 4

1

4

4

0.45

4

4 4

1

4

4

4

4

4

4

1 1

1 2

23

2

5 6

5

3

3 6

2

2

2

0.3

5

3 6

5

3

2

3

3

2 3

3

3

3

2

2

5 5

6

1

1

5

3 6

1

1 2

5

3 6

6

1 2

2

5

5

1

1

1

56

6

6

6

5 6

5

6

0.15

0 1993

1994

1995

1996

1997

1998

1999

Transport unit shares of foreign trade[import,container] : Current Transport unit shares of foreign trade[import,trailer] : Current Transport unit shares of foreign trade[import,truck] : Current Transport unit shares of foreign trade[export,container] : Current Transport unit shares of foreign trade[export,trailer] : Current Transport unit shares of foreign trade[export,truck] : Current

2000 2001 Time (Year) 1

2

1 2

3

4 5

6

Figure 4.2: Transport unit shares

2 3

4 5

6

5 6

2006 1

2 3

4 5

6

2005

1 2

3 4

5 6

2004

1 2

3 4

5

2003

1 2

3 4

2002

6

2007

Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

31

Economic scenario Output shares by industry Lookup

Transport intensities

w2 GDP index Lookup

Output shares by industry by scenario w1

GDP index by scenario

Transport intensity init

Change in transport intensity

Fractional growth rate of transport intensity

Transport intensity

Output shares by industry GDP index Output by industry (index)

Share of transport units



Foreign trade exports

Foreign trade by industry

Transport unit shares of foreign trade Lookup

Foreign trade imports

stime

Transport unit shares of foreign trade

Total foreign trade

Figure 4.3: Forecast of total foreign trade traffic

Share of transport units of foreign trade:

ρu,d (t) P

= Lookup table

ρu,d (t) = 1 ∀t, d

(4.16) (4.17)

u

Transport units:

u

∈ {Container, T railer, T ruck} (4.18)

The total amount of foreign trade traffic for each transport unit u is obtained as a function of the different factors above.

Total foreign trade trafficu (t, ω1 , ω2 ) = GDP(t, ω1 ) ·

XX i

[si (t, ω1 ) · ρu,d (t) · Ii,d (t)]

d

(4.19) The causal diagram total amount of foreign trade traffic is shown in Figure 4.3.

4.1.2

Transit traffic

Transit traffic through Finland is mainly imports. There is is also a small amount of exports but since their amount is not expected to grow in the future, the model assumes a constant amount of transit exports for the period 2008–2025. In the unit load transportations, transit traffic uses almost exclusively containers and the amount of transit trailer and truck traffic is small and their amounts in the model are assumed to remain constant for the period 2008–2025. Import transit traffic is expected to grow, but the growth rate is expected to slow down in the future. In system dynamics models S-shaped growth, where there is

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

32

accelerating growth in the beginning and where the growth slows down later on can be modeled as a combination of a positive and a negative feedback loop. The positive feedback dominates in the beginning, but as the system approaches the fixed carrying capacity of the environment, the negative feedback starts to dominate leading to slower growth. The notions of a population and a carrying capacity can be applied to any growing quantity, and similar behavior patterns can result from different fields, including the adoption of new products or the spread of diseases. If there are significant delays in the change of the population’s growth rate, there can also be overshoot (Figure 2.5). As discussed in section 3.3, there are several factors determining the development of transit traffic through Finland. For the purposes of the model these different factors can be aggregated to those that contribute to an accelerating development and to those that contribute to a decelerating development. One example of s-shaped growth is the logistic growth model, where the fractional growth rate declines linearly as a function of the population. However, a more flexible model is the Richards growth model [Richards, 1959] (cited in Sterman [2000]), which is used here. The amount of import transit traffic P (t) is a level variable in the SD model and is analogous with the population in the Richards model. The change in transit traffic P˙ (t) is obtained from:   

  m−1  P (t) 1− C dP (t) = P˙ (t) =    dt  −g ∗ P (t) · ln P (t) C g ∗ P (t) (m−1)

if m 6= 1

(4.20)

if m = 1

where C is the maximum capacity of the system, g ∗ is the maximum fractional growth rate and m is a parameter which determines the strength of the nonlinearity in the model. A causal diagram for transit development is shown in Figure 4.4. The parameters of the Richards growth model were calibrated such that the output of the system replicates historical behavior. A rough expert estimate of the growth of the transit traffic was also taken into account. Here the purpose of the transit traffic model for the whole model is to provide an exogenous input for the port demand and capacity model. For the understanding of the whole system it is more important to obtain the right sort of behavior (s-shaped growth, which is in accordance with expert opinion) than the exact magnitudes of the variables. To study the sensitivity of other variables to changes in the total amount of transit traffic, an alternative way to model transit traffic growth was also included in the model, namely a linear growth (ramp) where the change in transit P˙ is constant for the period 2007-2025. In the model different scenarios of transit traffic development can also be studied by assigning the value of transit traffic to be a weighted

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

Ramp slope

Transit traffic Ramp init



Transit traffic init (P(0)) Change in transit traffic (dP/dt) + +

m

Switch: drop in transit

w3 Import transit 0

Import transit

R Import transit RM Lookup

Fractional growth rate of transit traffic +

Transit traffic Ramp

Transit traffic (P)

Transit traffic capacity (C)

Max growth rate of transit traffic (g*)

33

Import transit RM

Total transit



B

-

Export transit RM Lookup

Export transit RM



Figure 4.4: Forecast of transit traffic average of the s-growth and ramp paths of development. In Figure 4.5 the development of the two exogenous variables, foreign trade and transit is shown with changes in parameters ω1 and ω2 accounting for the uncertainty in the development of foreign trade traffic and ω3 for the uncertainty in the development of transit traffic. Current ReferenceMode 50% 75% 95% 99% Total foreign trade[container] 60 M

Current ReferenceMode 50% 75% 95% Total transit[container] 8M

100%

45 M

6M

30 M

4M

15 M

2M

0 1992

2000

2009 Time (Year)

2017

2025

(a) Foreign trade

0 1992

99%

100%

2000

2009 Time (Year)

2017

2025

(b) Transit

Figure 4.5: Variation of exogenous inputs

4.2

Port demand and capacity

Finland’s total transport demand is divided among various ports. In the model the ports under study have been Hamina, Hanko, Helsinki, Kotka, Naantali, Rauma, Turku, and Others. A causal diagram for one port is shown in Figure 4.6. The model includes modules for the development of port capacity and the port market shares for foreign trade and transit traffic. The total transport amount of

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

34

foreign trade and transit are exogenous variables in the model, which together with the port market shares determine the traffic of a particular port. Based on the traffic in a particular port a demand forecast is made to determine how much capacity is needed in the future. Foreign trade market share

+

+ R: "Success to the successfull"

+

+

Net change in port attractiveness (foreign trade)

capasity usage influence factor (foreign trade)

Time Horizon for Reference Condition (THRC)

seed + Effect of port capacity usage on attractiveness (foreign trade)

+

market share influence factor (foreign trade)

port seed stdev

Change in PPC

+



+

+ Port foreign trade traffic

Forecast of traffic volumes per year in port +

Port traffic

R: "Capacity growth due to PPC"

+ Port capacity usage rate

+

Port cap usage deviation

Ref port capacity use

-

-

Port transit traffic

+ Port capacity acquisition delay



+

Expected Acquisition Delay Port capacity init

market share influence factor (transit) + +

Transit market share

+

+

Forecast time

Port market share + (transit)

Effect of market share on attractiveness (transit)

Port capacity on order

Order rate Effect of port capacity usage on attractiveness (transit) +

B: "Congestion weakens transit market share"

R: "Capacity growth due to Perceived trend"



+

Fractional growth rate of port attractiveness (transit) + +

TREND init

Perceived Present Condition (PPC)

Time to Perceive Present Condition (TPPC)

White noise

+

Port attractiveness (transit)

R: "Success to the successfull"

Perceived Trend (TREND) Change in TREND

+

Total traffic

Port attractiveness init (transit)



Indicated Trend (ITREND)



Port market share (foreign trade)

Port attractiveness (foreign trade) Port attractiveness init (foreign trade)

Net change in port attractiveness (transit)

Demand estimation

Time to Perceive Trend (TPT)

Effect of market share on attractiveness (foreign trade)

+ Fractional growth rate of port attractiveness (foreign trade)

Reference Condition (RC)

Change in RC

B: "Congestion weakens foreign trade market share"

Average life of port capacity

+

-

+

Desired port capacity

-

Port capacity

Discard rate

Acquisition rate

+

+ B

B

Supply line control Indicated orders +

Stock control Adjustment for Supply Line +

Supply Line Adjustment Time

capasity usage influence factor (transit)

Switch: Always at least discard rate?

Desired Supply Line + Desired + + Acquisition rate

Capacity adjustment time

+ Expected Discard Rate

Capacity acquisition

Figure 4.6: Causal diagram of the port demand and capacity model In Figure 4.7 the interaction of two ports is shown. In the entire model there are eight ports interacting with each other, but for simplicity only two ports are shown in the diagram. • Reinforcing feedback loops R1a, R1b, R2a and R2b: Reinforcing feedback loops that strengthen the market share of ports that already have a large market share. • Reinforcing feedback loops R4 and R5: A change in the market share of port A affects the market share of port B in the opposite direction, which affects the change of attractiveness and market share of port B. This in turn affects the change in attractiveness and market share of port A, thus creating a reinforcing feedback loop. • Balancing feedback loops B1a, B1b, B2a and B2b: A change in port traffic affects capacity use in the same direction, which affects the attractiveness and market share of the port in the opposite direction, which affects port traffic.

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

Change of foreign trade market share (port A) +

-

-

R 1b

R4

+

B 1a

+

Foreign trade market share (port A)

Capacity (port A)

+

Foreign trade market share (port B) +

-

R 1a Capacity use (port A) +

-

B 1b

Change of foreign trade market share (port B) -

+ Traffic (port B)

+ +

+

R 3a

+ Traffic (port A)

++

+

Change of transit market share (port A)

R 3b

Transit market share (port B)

R 2a

R5

Capacity (port B)

Total foreign trade traffic

Capacity use (port B)

R 2b

+

+

+ +

+

B 2a

35

B 2b

Total transit traffic

+

Transit market share (port A)

-

Change of transit market share (port B)

-

Figure 4.7: Port market shares • Reinforcing feedback loops R3a and R3b: An increase in traffic in a port increases the capacity of the port (with delay), which makes the port more attractive due to less capacity usage. This will then increase the market share and the traffic in the port. The structure of the system can be analyzed using Senge’s systems archetypes [Senge, 1990]. The structure created by feedback loops R1a, R1b, R4 (and R2a, R2b, R5) is similar to the “success to the successful”-archetype (Figure A.1), where several agents compete for a limited resource. The structure created by feedback loops R1a, B1a, R3a (and similarly by R2a, B2a, R3a / R1b, B1b, R3b / R2b, B2b, R3b) is similar to the “growth and underinvestment” -archetype (Figure A.2). In this archetype growth decreases the performance of the system (here performance decreases when capacity use rises), but performance can be improved by investments (here in investments for new port capacity).

4.2.1

Port market shares

The market shares of ports are divided into foreign trade and transit market shares, since these can vary from port to port and because the total amount of foreign trade traffic and transit traffic are of different magnitudes and develop in different ways. However, the formation of the two market shares are assumed to follow the same system structure. A causal diagram of the development of the attractiveness of a

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

36

particular port is shown in Figure 4.8.

Foreign trade market share

B: "Congestion weakens foreign trade market share"

+ Fractional growth rate of port attractiveness (foreign trade)

+



Effect of market share on attractiveness (foreign trade) +

+ R: "Success to the successfull"

Effect of port capacity usage on attractiveness (foreign trade)



-

+

Net change in port attractiveness (foreign trade)

Port market share (foreign trade)

Port attractiveness (foreign trade)

+

Port attractiveness init (foreign trade)

+

+ Port foreign trade traffic

Figure 4.8: Port attractiveness In SD models there are are two common ways to model the effect of different variables on a particular variable, the additive and multiplicative formulations:

Additive: y = y0 +

P

fi (xi )

(4.21)

i

Multiplicative:

y = y0 ·

Q

fi (xi )

(4.22)

i

where fi are functions that determine the effect of xi on y. Here a multiplicative formulation is used to model the effect of both market share and capacity usage on port attractiveness. Market share positive feedback There is a positive feedback loop that influences the market share of a port. If other factors are constant, a large market share of a particular port will tend to increase over time due to economies of scale. The customers of a particular port will also tend to favor the port that they already use when they expand their operations.

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

37

Capacity usage negative feedback In addition to the market share positive feedback loop, there is also a balancing feedback affecting the market share of a port. An increasing market share and increasing traffic volumes in a port increase the capacity usage of the port. As a port becomes congested, it will become less attractive compared to other ports and will thus decrease the market share of the port. The port attractiveness in addition to the attractiveness of the other ports determines the market share of the port. The port attractiveness of an individual port is calculated as follows:

Effect of market share:

fθM (·) =

1 + MSFθ · Mp,θ (t)

Effect of capacity utilization:

fθU (·) =

1 − tanh (CUFθ · (Up (t) − U ∗ )) (4.24)

Market share factor:

MSFθ =

Constant

(4.25)

Capacity utilization factor:

CUFθ =

Constant

(4.26)

fθM (·) · fθU (·) − 1

(4.27)

Fractional growth rate:

gp,θ (t) =

(4.23)

Net change rate: A˙ p,θ (t) = Ap,θ (t) · gp,θ (t) (4.28) t Z Attractiveness: Ap,θ (t) = Ap,θ (t0 ) + Ap,θ (ξ) · gp,θ (ξ) dξ(4.29) t0

The market share for port p and for transport type θ is calculated by normalizing the port attractiveness: Market share: Mp,θ (t) =

A (t) P p,θ Ap,θ (t)

(4.30)

p

p ∈ {Hamina, Hanko, Helsinki, Kotka, N aantali, Rauma, T urku, Others} correspond to the different ports, θ ∈ {F oreign trade, T ransit} are the different transport types. Up (t) is the capacity usage rate, U ∗ is the reference capacity usage and Mp,θ (t) is the market share.

4.2.2

Port demand forecasting

In the model the desired port capacity is determined by a forecast of traffic demand in the port. A TREND function is used to calculate the perceived fractional growth rate of traffic in the port. The TREND function takes into account delays in human expectation formation. A forecast of traffic in the port is then calculated using the fractional growth rate. Forecast of traffic(t) = PPC(t) · e(TREND(t)·Forecast time)

(4.31)

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

Change in RC

38

Reference Condition (RC)

Demand estimation



Change in PPC

Indicated Trend (ITREND)

Perceived Trend (TREND) Change in TREND

TREND init R: "Capacity growth due to Perceived trend"

Perceived Present Condition (PPC)

+

+

+

Forecast time

Forecast of traffic volumes per year in port



+

Figure 4.9: Port demand forecast with a TREND function A causal diagram of the port demand forecast is shown in Figure 4.9. Zt Perceived Trend:

TREND(t)

= TREND(t0 ) +

˙ TREND(t) dξ (4.32)

t0

Initial Trend: TREND(t0 ) = TRENDt0 ITREND(t) − TREND(t) ˙ Change in Trend: TREND(t) = TPT (PPC(t) − RC(t))/RC(t) Indicated Trend: ITREND(t) = THRC Zt ˙ Reference condition: RC(t) = RC(t0 ) + RC(ξ) dξ

(4.33) (4.34) (4.35) (4.36)

t0

Initial RC:

RC(t0 )

Change in RC:

˙ RC(t)

PPC(t0 ) 1 + THRC · TREND(t0 ) (PPC(t) − RC(t)) = THRC

=

(4.37) (4.38)

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

Zt Perceived present condition:

PPC(t)

= PPC(t0 ) +

39

˙ PPC(ξ) dξ

(4.39)

t0

INPUT(t0 ) Initial PPC: PPC(t0 ) = 1 + TPPC · TREND(t0 ) (INPUT − PPC(t)) ˙ Change in PPC: PPC(t) = TPPC

4.2.3

(4.40) (4.41)

Capacity acquisition

When the transport volumes in ports increases, investments are made to new machinery and technology that will lead to increased transport capacity in the port. The land available to the port can be assumed constant. Port capacity can be divided into different components including field space, lifting capacity, pier capacity and terminal capacity. For the purposes of the model this distinction is not necessary as the total capacity can be thought of as depending on the “weakest link” of the different factors and decision makers in ports can be assumed to direct investments accordingly. Various assumptions have to be made when modeling the investments in ports. First of all, the desired capacity, upon which investment decisions are made in the model is assumed to be based on the development of traffic in the port. Investments are not analyzed in monetary terms but rather on how much they contribute to capacity growth. For example, the calculation of net present values of the investments are not included in the model. Things that can affect the investments to a port in reality can include: • Net present value of investments: How much revenue investments in additional capacity would yield in the future. • Economies of scale: The same value of investment could yield a larger increase in capacity in a larger port than in a small port. • Limits to growth: Ports cannot grow forever since eventually land space will pose constraints. The capacity of road infrastructure can also create a bottleneck and limit the amount of traffic that can pass through a port. • Politics: Ports are mainly municipally owned in Finland and do not operate under free market conditions. Interest groups that are dependent on ports in the form of employment can influence decision making. A large port can also bring image status to the municipality. • Legislation: The process leading to an eventual investment decision for new capacity can be slow due to legislation in community planning. The envi-

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

40

ronmental effects of the decisions must also be assessed which can delay the investment process. • Investments cannot be adjusted smoothly but have to be taken in larger units. This would require quantization of the flow of orders in the model. The port capacity model here is based on a generic model of production capacity Sterman [2000] (p.806), and can be seen in Figure 4.10.

Port capacity acquisition delay



+

Expected Acquisition Delay Port capacity init

Port capacity on order

Order rate

+

- +

Desired port Average life of port capacity capacity -

Port capacity

Discard rate

Acquisition rate

+

+ B



B

Supply line control Indicated orders +

Stock control Adjustment for Supply Line

- +

Switch: Always at least discard rate?

Desired Supply Line +

+ Adjustment for Capacity

+

Desired + Acquisition rate

+ Expected Discard Rate

Capacity acquisition

Figure 4.10: Port capacity Some assumptions in the capacity acquisition model include: • Orders cannot be canceled once placed, so the amount of orders for new capacity is restricted to non-negative values. • The port capacity acquisition delay and discard rate are assumed to be known accurately, i.e. the expected acquisition delay and the expected discard rate correspond to the values of these variables, respectively. • The average life of port capacity, capacity adjustment time and port capacity acquisition delay are constants. For simplicity it is also assumed that the values of these variables are the same for all ports. • The supply line and stock are adjusted with first order proportional controllers, i.e. proportional to the difference between the value of the variable and its desired value.

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

41

The variables are calculated based on Equation 4.42–4.53.

Expected Acquisition Delay = Capacity acquisition delay Expected Discard Rate = Discard Rate Indicated Orders(t) =

(4.42) (4.43)

Desired Acquisition rate(t) (4.44) +Adjustment for Supply Line(t)

Order rate(t) = max [Indicated orders(t), 0] (4.45) t R Capacity on order(t0 ) + [Order rate(ξ) Capacity on order(t) = (4.46) t0 −Acquisition rate(ξ)] dξ Capacity(t) =

Capacity(t0 ) +

Rt

[Acquisition rate(ξ)

t0

(4.47)

−Discard rate(s)] dξ Discard rate(t) = Adjustment for capacity(t) =

Adjustment for supply line(t) =

Acquisition rate(t) = Desired Acquisition rate(t) = Desired Supply Line(t) =

4.3

Capacity(t) Average life of Capacity Desired capacity(t) − Capacity(t) Capacity adjustment time

(4.48) (4.49)

(Desired Supply Line(t) (4.50) −Capacity on order(t)) /Supply Line Adjustment Time DELAY3 [Order rate(t), Capacity acquisition delay]

(4.51)

Expected Discard Rate(t) +Adjustment for Capacity(t)

(4.52)

Expected Acquisition Delay(t) ·Desired Acquisition rate(t)

(4.53)

Model validation

Validation refers to the assessment of the model to suit its purposes. The validation of system dynamics models can be divided into two different tasks: structural validation and calibration. Structural validation refers to the ability of the model to capture the relevant structure of the system, while calibration refers to the process of matching the model output to historical behavior of the system. In this study validation has been done together with the domain experts, who are also part of the project. In practice the modeling has progressed iteratively. A preliminary version of the model was first developed based on the descriptive knowledge of the system obtained from the domain experts, which was then updated based on

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

42

comments and the effectiveness of the model to capture the dynamics of the real system.

4.3.1

Structural validation

The purpose of structural validation is to ensure that the model structure is consistent with descriptive knowledge of the system. For a system dynamics model to be structurally valid, causal relationships in the model have to be valid and the assumptions and simplifications that are made in the modeling process are justified. The validity of the dynamic hypotheses must thus be analyzed. If the structure of the model is not correct, there is a risk that the model might replicate historical behavior for the wrong reasons. The boundary of the model determines the endogenous and exogenous variables of the model as well as variables that are omitted. In this study, the development of GDP and that of the different industries were modeled exogenously. This seems justified when considering the model purpose. The model is used to make decisions on how to develop the transport system. While in reality the effectiveness of the entire transport system might have an impact on the development of the whole economy, this feedback process was not considered important enough to justify its modeling. In the model only unit load transportations (ie. containers, trucks, trailers) were included. It was assumed that other transportations like oil shipments do not influence the amount of the unit load transportations. The different transport units were modeled separately, and it was also assumed that port capacities can be independently calculated for each of the different transport units. Structural validation also includes assessment of the level of aggregation in the model. In the model, each of the major ports in Finland are analyzed separately. However, the port capacity acquisition delay and average life of port capacity are assumed to be same for all ports. In reality these are not the same for all ports since different ports make investments to target different capacity bottlenecks in each particular port. Also, ports can make different sorts of investments at different times, so in reality these parameters are not constant over time. Nonetheless, due to limitations in statistics of port capacity development and the scope of this research project, the disaggregated identification of these port parameters was not feasible.

4.3.2

Model calibration

Model calibration is the process of estimating the model parameters to match available data of the system. This can either be done by hand by iteratively varying one parameter at a time. One method for automated calibration is model reference optimization (MRO) [Lyneis and Pugh, 1996] (cited in Oliva [2003]), where

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

43

the calibration problem is formulated as an optimization problem. An error function depending on the differences between the model output and a data time series is minimized by varying the model model parameters. Schade and Krail [2006] present an example of stepwise automatic calibration of a large scale system dynamics model with various submodels. The calibration proceeds in steps, and in each step the results of the calibrations of previous submodels are used to calibrate the next parts of the model. Here the model was calibrated using the optimizer tool in Vensim. The purpose of the first calibration was to match the model output of the amount of total foreign trade imports and exports to historical values by varying the transport intensities and their growth rates. The model can capture the trend in the variables, but there are short term fluctuations not explained by the model, as shown in Figure 4.11. The error term has a zero mean but is autocorrelated. However, for the purpose of the model it is more important to predict the long term behavior of the system rather than seasonal trends. 14 M 4 1 4 1 1 4

10.5 M 1 4

2

1 4 1

5 2

4

4

2 5

1

7M

4 4 4

1

5

2

1

4

1

5 2 5

5 2

1

1

2

4

1

5 5

3.5 M

5

2

2

5

2

2

5

2

2 3

3

6

3

6

3

6

0 1992 1993 1994 1995 1996 1997 1 Total foreign trade[container] : Current 1 2 Foreign trade exports[container] : Current 3 Foreign trade imports[container] : Current 3 Total foreign trade[container] : ReferenceMode Foreign trade exports[container] : ReferenceMode Foreign trade imports[container] : ReferenceMode

1998

1999

1

6

6

6

3

6

3 4

5 6

2 3

4 5

4 5

6

2006

1 2

4 5

2005

1

6

3

6

2004

2 3

4 5

2003 1

2 3

4 5

2002

3

6

3

6

1 2

3 4

2001 1

2 3

5

2000

1 2

4

3

6

3

6

3

3

6

3

6

5 6

6

3

2007 tonnes tonnes tonnes tonnes tonnes tonnes

Figure 4.11: Total foreign trade calibration In the second calibration the port demand and capacity model was calibrated. 28 parameters (the capacity utilization and market share factors for the different transport types, port attractiveness for each port and for both transport types and the initial port capacities for each port) were varied to match the model output of

CHAPTER 4. MODEL OF PORT DEMAND AND CAPACITY

44

port traffic to the real system. The results of the calibration for three ports can be seen in Figure 4.12. While the model’s point-to-point accuracy is not very good, the model can capture the essential trends of the system. For the purpose of the model it is also more important to explain long term trends than individual occurrences, which can be caused by random events. A major difficulty in the modeling process was the lack of data for the historical development of port capacities. This means that the capacities and capacity usages calculated by the model are only rough estimates of the real values. However, an interview among port representatives was conducted during the course of the study, which gave a rough estimate of each port’s capacity during the year 2007. In the calibration process these were used to set some constraints for the decision variables in the calibration optimization problem. 6M

0.8

5

2

5

5 5

4.5 M

2

2

2

2

6

2 3

2 5

5 2

5

2

5 2 5

2 2

5

5

3

2

2

2

6

2

2

2

5

2

5

5

3

5 5

0.6

3

2

5

5 5

6

3M

5

2

0.4

3

5

2 2

5

6

2

6

2

3 3

4 6 3 4

3 1

1

4

1

1

0.2

1

3

4

3 1

4

1998

1

5

3

2 3

4

(a) Port traffic calibration

3 4

5 6

4 5

6

2006

1 2

5 6

2005

1 2

4 5

6

2004

1 3

4 5

6

2003

2 3

4 5

6

2002 1

2 3

4

2001 1

2 3

4

2000

1 2

3

6

1999

1 2

5 6

6

2007 tonnes tonnes tonnes tonnes tonnes tonnes

1

1

1 4

1997

6

6

6 4

4

1

1

4 0 1992 1993 1994 1995 1996 Port traffic[Hamina,container] : Current Port traffic[Helsinki,container] : Current 2 3 Port traffic[Kotka,container] : Current Port traffic[Hamina,container] : ReferenceMode Port traffic[Helsinki,container] : ReferenceMode Port traffic[Kotka,container] : ReferenceMode

3

3

3 3

5

3

4

4

3

3

6

6

6

3

6

6

1 1

4

4

3

3

6

6

4

1

4

1

6

3 3

3

1

3

6 6

1

6

3

6

1.5 M

2

5 6 6

4

1

4

1

1

1

1

4

1

1

4

4

4

1

4

1

4

4

0 1992 1993 1994 1995 1996 1997 1 Port market share[Hamina,container] : Current 2 Port market share[Helsinki,container] : Current Port market share[Kotka,container] : Current Port market share[Hamina,container] : ReferenceMode Port market share[Helsinki,container] : ReferenceMode Port market share[Kotka,container] : ReferenceMode

1998

1999

1

5 6

3 4

5 6

(b) Port market share calibration

Figure 4.12: Port traffic calibration

2 3

4

4 5

6

2007 1

2 3

4

2006

1 2

3 5

6

2005

1 2

4 5

6

2004

1 3

4 5

6

2003

2 3

4 5

6

2002 1

2 3

4 5

2001 1

2 3

4

2000 1

2 3

6

Chapter 5

Simulation results Once the model was calibrated, a simulation was performed for the time period 1992-2025 to see how the system would behave in the future. Historical data are available for port traffic and market shares, and these are depicted in the graphs by the “ReferenceMode” time series. The simulation results for four variables, port capacity, port traffic, market shares and capacity usage rate for the different ports and for container traffic are shown below. Figure 5.1 shows the port capacity forecast. The model predicts that the port capacity of Kotka will increase rapidly. The simulation results suggest Kotka’s capacity to have been smaller than that of Helsinki at the start of the simulation but that by 2005 it had already exceeded Helsinki’s capacity. The traffic forecasts for the different ports can be seen in Figure 5.2. According to the model the transport capacity in Kotka has risen rapidly, and it corresponds with the real historical trend. The model predicts that this trend will continue in the future. The capacity usage rates of different ports are the ratios of the amounts of traffic in a port and the port’s capacity. The model results for the capacity usage rates can be seen in Figure 5.3. Figure 5.4 shows the port market share forecast. The market shares portrayed by the model are in accordance with historical trends, namely the decrease in Helsinki’s market share and the rise of that of Kotka. According to the model these market shares will stabilize as the simulation proceeds. The future development of the port market shares is also in accordance with qualitative forecasts of the domain experts.

5.1

Sensitivity analysis

To study the effect of various uncertain parameters in the model, a sensitivity analysis was done using the sensitivity tool in Vensim. For each sensitivity run a list 45

CHAPTER 5. SIMULATION RESULTS

46

30 M

22.5 M 4

4

15 M 4

3

3 1

3

7.5 M 4 3 1

6

3 1

3

8

6 6

4

1 6 8

8

4

8

1 6

1

8

2

0 1992

5

1995

7

8

6

1 2 5

1998

2 2

7

7

7

5

2001

Port capacity[Hamina,container] : Current Port capacity[Hanko,container] : Current Port capacity[Helsinki,container] : Current Port capacity[Kotka,container] : Current Port capacity[Naantali,container] : Current Port capacity[Rauma,container] : Current Port capacity[Turku,container] : Current Port capacity[Others,container] : Current

2004

2010

2

7

4 5 6

7 8

3 5

6

6 7

8

1

2 4

5

2019

1

3 4

5

2016

2 3

6

5

2013

1

7

7

5

2007

1

2

2

7 8

Figure 5.1: Simulation results of port capacity

8

2022

2025

tonnes 2 tonnes 3 tonnes 4 tonnes 5 tonnes tonnes tonnes tonnes

CHAPTER 5. SIMULATION RESULTS

47

20 M

15 M

4

10 M

3

1

5M

3 4 B

3

6 1 C 4

0 12 1992

6 5

1995

2 2

2

8 7

8

6

41

9 A

1998

1

8

3

2001

2 5

2004

7

7

2007

5

2010

2013

2016

2019

2022

2025

1 1 Port traffic[Hamina,container] : Current tonnes 2 2 Port traffic[Hanko,container] : Current tonnes 3 3 Port traffic[Helsinki,container] : Current tonnes 4 4 tonnes Port traffic[Kotka,container] : Current 4 5 5 tonnes Port traffic[Naantali,container] : Current 5 6 6 Port traffic[Rauma,container] : Current tonnes 7 7 Port traffic[Turku,container] : Current tonnes 8 8 Port traffic[Others,container] : Current tonnes 9 9 Port traffic[Hamina,container] : ReferenceMode tonnes A A Port traffic[Hanko,container] : ReferenceMode tonnes B B Port traffic[Helsinki,container] : ReferenceMode tonnes C C Port traffic[Kotka,container] : ReferenceMode tonnes 1 1 Port traffic[Naantali,container] : ReferenceMode tonnes 2 2 Port traffic[Rauma,container] : ReferenceMode tonnes 3 3 Port traffic[Turku,container] : ReferenceMode tonnes 4 4 Port traffic[Others,container] : ReferenceMode tonnes

Figure 5.2: Simulation results of port traffic

CHAPTER 5. SIMULATION RESULTS

48

1

3 3 7 8

7 6

2 3

8

3

1

7 4

0.75

6

8 2

3

6 1

3

1 2

1

2

8

4

4

7 4

7 6

3

1

8 4

6 7

0.5

2 8

1 2

0.25

4

6

1

2

0 1992

5

1995

5

1998

5

2001

2004

5

2007

2010

5

2013

2016

5

2019

2022

2025

1 1 1 Dimensionless Port capacity usage rate[Hamina,container] : Current 2 2 Port capacity usage rate[Hanko,container] : Current 2 Dimensionless 3 3 Port capacity usage rate[Helsinki,container] : Current 3 Dimensionless 4 4 4 Port capacity usage rate[Kotka,container] : Current Dimensionless 5 5 5 Port capacity usage rate[Naantali,container] : Current Dimensionless 6 6 6 Port capacity usage rate[Rauma,container] : Current Dimensionless 7 7 7 Port capacity usage rate[Turku,container] : Current Dimensionless 8 8 8 Port capacity usage rate[Others,container] : Current Dimensionless

Figure 5.3: Simulation results of port capacity usage

CHAPTER 5. SIMULATION RESULTS

49

0.8

3

0.6

B

0.4

4 3

4 3 3

0.2 C 4 4

9

1 6

1 8

1

8

6

2

8

6

1

7

2

0 1992

3

1998

2

7

1

5

1995

2

2 A

7

5

2001

2004

Port market share[Hamina,container] : Current Port market share[Hanko,container] : Current Port market share[Helsinki,container] : Current Port market share[Kotka,container] : Current Port market share[Naantali,container] : Current Port market share[Rauma,container] : Current Port market share[Turku,container] : Current 7 8 Port market share[Others,container] : Current Port market share[Hamina,container] : ReferenceMode Port market share[Hanko,container] : ReferenceMode Port market share[Helsinki,container] : ReferenceMode Port market share[Kotka,container] : ReferenceMode Port market share[Naantali,container] : ReferenceMode Port market share[Rauma,container] : ReferenceMode Port market share[Turku,container] : ReferenceMode Port market share[Others,container] : ReferenceMode

5

2007

2010

2013

1

2016

2019

2022

2 3 4 5 6 7 8 9 A B C

C 1

1 2

2 3

3 4

Figure 5.4: Simulation results of port market shares

2025

1 Dimensionless

4

Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless Dimensionless

CHAPTER 5. SIMULATION RESULTS

50

of input variables were selected and a probability distribution assigned to them. Next, 1000 simulations were run using Latin hypercube sampling. After the sensitivity run was completed, a graph could be drawn with percentiles to show how much changes in the input parameters affect the output of the model. In addition, a correlation analysis was done to find the parameters which cause the greatest changes in the output variables. This was done using the method developed by Ford and Flynn [2005] called “statistical screening”. The method consists of calculating the correlations between input parameters and the model outputs at different times in the simulation once a sensitivity simulation has been run. While there are uncertainties concerning the value of each input variable, it is more important to focus on obtaining reliable estimates for those parameters that affect the behavior of the system the most. The correlation coefficient between two variables is obtained from the following equation: P ¯ i − Y¯ ) (Xi − X)(Y r = pP ¯ 2 P(Yi − Y¯ )2 (Xi − X)

(5.1)

The correlation coefficients are calculated for the various inputs and for the different years of the simulation. The implementation of the correlation analysis was done using Matlab by importing the sensitivity data from Vensim. In the sensitivity simulations the parameters were varied using a uniform distribution. The minimum and maximum values for the distributions were obtained by allowing the variable to vary ±30% from the value obtained in the model calibration phase. An exception were the parameters corresponding to the development of the exogenous amounts of the total transport volumes (ω1 , ω2 and ω3 ) which were varied in the range [0,1].

5.1.1

Sensitivity simulation S1: Years 1992-2025

The first sensitivity run was used to test the importance of the different variables in the timescale used for model calibration (years 1992–2007). The variables can be seen in Table 5.1. The results of the sensitivity simulations for two ports (Helsinki and Kotka) can be seen in Figure 5.5 and Figure 5.6. From the sensitivity simulations it can be seen that the variation of the selected parameters gives very wide confidence intervals due to the wide range (±30%) of variation in each input variable. The benefit of a wide range of variation, even if unrealistic, is that it serves as an extreme value test which could reveal formulation flaws in the model. Figure 5.7 and Figure 5.8 show the correlations of the input variables and the traffic in two chosen ports (Helsinki and Kotka).

CHAPTER 5. SIMULATION RESULTS

51

Table 5.1: Variables in sensitivity simulation S1

Variable

Distribution

Min

Max

Port capacity init[Hamina,container] Port capacity init[Hanko,container] Port capacity init[Helsinki,container] Port capacity init[Kotka,container] Port capacity init[Naantali,container] Port capacity init[Rauma,container] Port capacity init[Turku,container] Port capacity init[Others,container] CUF[container,foreign trade] CUF[container,transit] MSF[container,foreign trade] MSF[container,transit] TPPC THRC TPT Supply Line AT Capacity AT Port capacity acquisition delay[container] Average life of port capacity[container]

Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif

930783 542683.4 3521749 2248988 0 1238440 275473.1 1233764 0.3293185 0.1229263 0.0865844 0.5936364 0.7 0.7 0.7 1.4 1.4 3.5 17.5

1728597 1007840.6 6540391 4176692 0 2299960 511592.9 2291276 0.6115915 0.2282917 0.1607996 1.1024676 1.3 1.3 1.3 2.6 2.6 6.5 32.5

The correlation plot gives an indication of the most influential parameters of the model. The initial capacity of Helsinki has a strong influence on the amounts of traffic in both Helsinki and Kotka. A larger initial capacity in Helsinki increases its market share, hence the positive correlation. At the same time it also decreases the market share of Kotka explaining the negative correlation. As the simulation proceeds, the importance of the initial capacities decreases. In addition to the initial capacities of Kotka and Helsinki, the capacity usage factor for foreign trade is also an important parameter which affects the behavior of the system. Its correlation with the traffic in Helsinki is negative and positive with the traffic in Kotka. This makes sense, since the port capacity usage rate of Helsinki is higher during the period of the simulation.

CHAPTER 5. SIMULATION RESULTS s1 ReferenceMode 50% 75% 95% 99% Port traffic[Helsinki,container] 60 M

52

100%

45 M

30 M

15 M

0 Port traffic[Kotka,container] 60 M

45 M

30 M

15 M

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.5: Port traffic in sensitivity simulation S1 s1 ReferenceMode 50% 75% 95% 99% Port market share[Helsinki,container] 1

100%

0.75

0.5

0.25

0 Port market share[Kotka,container] 1

0.75

0.5

0.25

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.6: Port market share in sensitivity simulation S1

5.1.2

Sensitivity simulation S2: Effect of port model variables

The values of the parameters in the model were calibrated using historical data. However, they do not necessarily stay constant over time due to structural changes

CHAPTER 5. SIMULATION RESULTS

53

Figure 5.7: Helsinki traffic correlations in sensitivity simulation S1

Figure 5.8: Kotka traffic correlations in sensitivity simulation S1 or because of the fact that the parameters correspond only to a particular state of the system. As the state of the entire system changes, the values of the parameters might also change.

CHAPTER 5. SIMULATION RESULTS

54

A second sensitivity run was done to study the effect of various uncertain parameters in the future. The values of the parameters were held constant until 2007, after which a stepwise change was made in the parameters. The list of parameters (Table 5.2) is the same as for the previous sensitivity run except for the omission of the parameters corresponding to the initial values of the port capacities. Table 5.2: Variables in sensitivity simulation S2

Variable

Distribution

Min

Max

CUF[container,foreign trade] CUF[container,transit] MSF[container,foreign trade] MSF[container,transit] TPPC THRC TPT Supply Line AT Capacity AT Port capacity acquisition delay[container] Average life of port capacity[container]

Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif

0.3293185 0.1229263 0.0865844 0.5936364 0.7 0.7 0.7 1.4 1.4 3.5 17.5

0.6115915 0.2282917 0.1607996 1.1024676 1.3 1.3 1.3 2.6 2.6 6.5 32.5

The results of the sensitivity simulation can be seen in Figure 5.9, Figure 5.10 and Figure 5.11.

5.1.3

Sensitivity simulation S3: Effect of port model variables and total traffic

In the third sensitivity simulation the effect of uncertainties concerning the amounts of total foreign trade and transit traffic in the model were also taken into account as well as the model parameters in the previous sensitivity simulation. The results of the sensitivity simulation can be seen in Figure 5.12, Figure 5.13 and Figure 5.14. Results of the correlation analyses can be seen in Figure 5.15 and Figure 5.16.

5.1.4

Sensitivity simulation S4: Effect of total traffic

A final sensitivity simulation was done to analyze how the changes in only the total foreign trade and transit traffic affect the behavior of the system. A sensitivity analysis showed that changes in the total amount of transit traffic had a negligible effect on the market shares of ports. This can be seen in Figure 5.17.

CHAPTER 5. SIMULATION RESULTS s2 ReferenceMode 50% 75% 95% 99% Port capacity[Helsinki,container] 40 M

55

100%

30 M

20 M

10 M

0 Port capacity[Kotka,container] 60 M

45 M

30 M

15 M

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.9: Port capacity in sensitivity simulation S2 s2 ReferenceMode 50% 75% 95% 99% Port traffic[Helsinki,container] 20 M

100%

15 M

10 M

5M

0 Port traffic[Kotka,container] 40 M

30 M

20 M

10 M

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.10: Port traffic in sensitivity simulation S2 Since the share of transit traffic of all of the port’s traffic was the largest in Kotka, changes in the total transit traffic had also the greatest influences in Kotka’s traffic. A similar effect was obtained with a sensitivity analysis of the effect of the other exogenous variable, the total amount of foreign trade. While the total amount of

CHAPTER 5. SIMULATION RESULTS s2 ReferenceMode 50% 75% 95% 99% Port market share[Helsinki,container] 0.8

56

100%

0.6

0.4

0.2

0 Port market share[Kotka,container] 0.8

0.6

0.4

0.2

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.11: Port market share in sensitivity simulation S2 Table 5.3: Variables in sensitivity simulation S3

Variable

Distribution

Min

Max

CUF[container,foreign trade] CUF[container,transit] MSF[container,foreign trade] MSF[container,transit] TPPC THRC TPT Supply Line AT Capacity AT Port capacity acquisition delay[container] Average life of port capacity[container] w1 w2 w3

Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif Unif

0.3293185 0.1229263 0.0865844 0.5936364 0.7 0.7 0.7 1.4 1.4 3.5 17.5 0 0 0

0.6115915 0.2282917 0.1607996 1.1024676 1.3 1.3 1.3 2.6 2.6 6.5 32.5 1 1 1

traffic in each port varied, the respective market shares were virtually unaltered. Figure 5.18, Figure 5.19 and Figure 5.20 show histograms of the model output at

CHAPTER 5. SIMULATION RESULTS s3 ReferenceMode 50% 75% 95% 99% Port capacity[Helsinki,container] 40 M

57

100%

30 M

20 M

10 M

0 Port capacity[Kotka,container] 60 M

45 M

30 M

15 M

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.12: Port capacity in sensitivity simulation S3 s3 ReferenceMode 50% 75% 95% 99% Port traffic[Helsinki,container] 20 M

100%

15 M

10 M

5M

0 Port traffic[Kotka,container] 40 M

30 M

20 M

10 M

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.13: Port traffic in sensitivity simulation S3 the end of the simulation (year 2025) for the different sensitivity simulations. The histograms depict the number of simulations out of the total number of 1000 sensitivity simulations where the output variable falls within different ranges. There is a sharp contrast when the histograms of the port capacity and port traffic are

CHAPTER 5. SIMULATION RESULTS s3 ReferenceMode 50% 75% 95% 99% Port market share[Helsinki,container] 0.8

58

100%

0.6

0.4

0.2

0 Port market share[Kotka,container] 0.8

0.6

0.4

0.2

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.14: Port market share in sensitivity simulation S3

Figure 5.15: Helsinki traffic correlations in sensitivity simulation S3 compared with that of port market share. In the market share histogram there is very little variation for the sensitivity simulation S4 (in green) in which only the total traffic amounts were varied. This means that the market shares of ports are mainly determined by endogenous factors in the model. The total amounts of traf-

CHAPTER 5. SIMULATION RESULTS

59

Figure 5.16: Helsinki market share correlations in sensitivity simulation S3 Table 5.4: Variables in sensitivity simulation S4

Variable

Distribution

Min

Max

w1 w2 w3

Unif Unif Unif

0 0 0

1 1 1

fic influence each port’s traffic and capacity, but an increase in the total amounts of traffic gets distributed mostly based on the endogenously formed market shares.

CHAPTER 5. SIMULATION RESULTS

s4 ReferenceMode 50% 75% 95% 99% Port market share[Helsinki,container] 0.8

60

100%

0.6

0.4

0.2

0 Port market share[Kotka,container] 0.6

0.45

0.3

0.15

0 1992

2000

2009 Time (Year)

2017

2025

Figure 5.17: Port market share in sensitivity simulation S4

150 S2 S3 S4

Frequency

100

50

0 0.4

0.6

0.8

1 1.2 1.4 1.6 Port capacity[Helsinki,container] 2025 (tonnes)

1.8

2

2.2 7

x 10

Figure 5.18: Histogram of Helsinki’s capacity at the end of the simulation

CHAPTER 5. SIMULATION RESULTS

61

140 S2 S3 S4 120

100

Frequency

80

60

40

20

0 0.4

0.6

0.8

1 1.2 Port traffic[Helsinki,container] 2025 (tonnes)

1.4

1.6 7

x 10

Figure 5.19: Histogram of Helsinki’s traffic at the end of the simulation

1000 S2 S3 S4

900

800

700

Frequency

600

500

400

300

200

100

0 0.1

0.15

0.2 0.25 Port market share[Helsinki,container] 2025

0.3

0.35

Figure 5.20: Histogram of Helsinki’s market share at the end of the simulation

Chapter 6

Discussion and conclusions 6.1

Summary and review of objectives

The objective of this thesis was to develop a system dynamics model for the forecasting of the port system in Finland. The implemented model includes three main parts: a) the capacity acquisition process of each port, b) the demand trend estimation process that port owners go through in order to determine the level of desired capacity in a given port and c) the port market share determination process. The total transport demand in the system dynamics model is an exogenous variable which can follow different paths of development corresponding to various macroeconomic scenarios. There are different factors affecting the development of transit and foreign trade traffic as well as their allocation to different ports. This is why the development of transit and foreign trade market shares were treated distinctly in the model. However, as both transit and foreign trade traffic in a given port use the same shared port capacity, there are also feedback processes linking these two. The availability of historical data of the system was partly insufficient, and this caused difficulties in the validation of the model. However, it is good to remember that modeling is an iterative process. One important role of the model developed in this thesis is to bring insight about the data that is needed for a more complete model. Before the modeling process started it was difficult to know exactly what data is needed to validate the model. One policy change that this thesis suggests is the need to gather statistics concerning relevant variables of the system, such as the port capacities. One conclusion that could be drawn from the simulations is that much of the behavior in the system is governed by the internal feedback processes. While the total amounts of transport, which are exogenous factors, are important in determining the level of traffic in each port, they are far less important in explaining the development of port market shares. 62

CHAPTER 6. DISCUSSION AND CONCLUSIONS

63

Since there are so many feedback loops in the system, it can be helpful to study the system in parts by identifying certain “system archetypes” [Senge, 1990]. One of the system archetypes presented by Senge [1990] is the “limits to growth” archetype, which results from an interplay between a positive feedback and a negative feedback influenced by a limiting condition. Here, the market shares increase the future market shares in a positive feedback loop and the capacity of the port is the limiting condition which affects the negative feedback. However, since capacity is also allowed to grow in the model, the limiting condition is not constant and there is another positive feedback in the system. This is similar to Senge’s “growth and underinvestment” archetype. The “success to the successful” archetype [Senge, 1990] is a structure where two agents compete for a shared resource, and increased resources for either player will lead to more resources for that player. Here, the different ports compete for a fixed amount of traffic demand, and a large market share will lead to an even bigger market share if other factors are constant. Similarly, a small market share will tend to diminish even smaller. One solution for the "success to the successful" archetype is to modify the system in a way that the different agents do not compete for the same resource [Senge, 1990]. Here, the different ports can specialize in different types of transports. This also happens in real life. In addition to analyzing what different ports can do to increase their market shares, an important question is how to modify the system in order to improve the functioning of the whole national transport system. The purpose of government agencies is to design policies in order for this to happen. A problem here is that the different port decision makers in the system have different objectives which can lead to part-optimizing. In this study changes in the total transport volumes correspond to different macroeconomic scenarios. However, according to Sterman [2000] “the real value of modeling is not to anticipate and react to problems in the environment, but to eliminate the problems by changing the underlying structure of the system.” Once decision makers have a better understanding of the system, better policy decisions of how to achieve a desired future state can also be made. To alter the behavior of the system, it is important to know which are its “leverage points” or parameters that impact the output of the system the most. In this study a statistical screening analysis was done to find the key input parameter in the model.

6.2

Future research

One key difficulty so far in this project was the fact that no statistical data was available for important variables in the model. The port capacities are one example.

CHAPTER 6. DISCUSSION AND CONCLUSIONS

64

Statistics are currently available for the transport volumes in different ports, but not for the actual capacities of the ports. Therefore one future research area is further validation of the model once more statistical data becomes available. Another future area of research concerns the further disaggregation of some of the model’s variables. According to the domain experts there are different factors affecting the import and export traffic volumes in various ports, so the model could yield better results if these were disaggregated. The model developed in this thesis does not include road transportation. In the future these will also be included since they are of interest to the users of the model. Also, results of the performed statistical screening analysis could be complemented by an analysis of scatter plots between the inputs and outputs. Since the screening relied on calculating the correlation coefficients between the inputs and outputs, the results only showed the degree of the linear relationship between the variables. If there is a causal but highly nonlinear relationship between the variables, the screening does not give a reliable estimate of the importance of the given input variable. This was also pointed out by Ford and Flynn [2005].

Bibliography [1] H. Akkermans. Developing a logistics strategy through participative business modelling. International Journal of Operations & Production Management, 15(11): 100–112, 1995. 17 [2] H. Akkermans and N. Dellaert. The rediscovery of industrial dynamics: the combination of system dynamics to supply chain management in a dynamic and fragmented world. System Dynamics Review, 21(3):173–186, 2005. 15, 16, 17 [3] H. Akkermans, P. Bogerd, and B. Vos. Virtuous and vicious cycles on the road towards international supply chain management. International Journal of Operations & Production Management, 19(5/6):565–581, 1999. 16 [4] F. Aminian, E. D. Suarez, M. Aminian, and D. T. Walz. Forecasting Economic Data with Neural Networks. Computational Economics, 28(1):71–88, 2006. 6 [5] B. J. Angerhofer and M. C. Angelides. System dynamics modelling in supply chain management: research review. In Proceedings of the 2000 Winter Simulation Conference, pages 342–351. Society for Computer Simulation International, San Diego, CA, USA, 2000. x, 15, 16, 17 [6] U. M. Bhushi and C. M. Javalagi. System Dynamics Application to Supply Chain Management: A Review. In Engineering Management Conference, 2004. Proceedings. 2004 IEEE International, 2004. 17 [7] D. J. Bowersox, R. J. Calantone, and A. M. Rodrigues. Estimation of Global Logistics Expenditures using Neural Networks. Journal of Business Logistics, 24, 2003. 6 [8] L. Börjeson, M. Höjer, K.-H. Dreborg, T. Ekvall, and G. Finnveden. Scenario types and techniques: Towards a user’s guide. Futures, 38:723–739, 2006. 19 [9] H. M. Celik. Modeling freight distribution using artificial neural networks. Journal of Transport Geography, 12(2):141–148, 2004. 6

65

BIBLIOGRAPHY

66

[10] G. De Jong, H. Gunn, and W. Walker. National and International Freight Transport Models: An Overview and Ideas for Future Development. Transport Reviews, 24(1):103–124, 2004. 18 [11] M. Dougherty. A review of neural networks applied to transport. Transportation Research Part C: Emerging Technologies, 3(4):247–260, 1995. 6 [12] J. T. Fite, G. D. Taylor, J. S. Usher, J. R. English, and J. N. Roberts. Forecasting freight demand using economic indices. International Journal of Physical Distribution and Logistics Management, 32(4):299–308, 2002. 6 [13] A. Ford. System Dynamics and the Electric Power Industry. System Dynamics Review, 13(1):57–85, 1997. 14 [14] A. Ford and H. Flynn. Statistical screening of system dynamics models. System Dynamics Review, 21(4):273–303, 2005. 4, 12, 50, 64 [15] D. N. Ford. A behavioral approach to feedback loop dominance analysis. System Dynamics Review, 15(1):3–36, 1999. 12 [16] J. W. Forrester. Industrial Dynamics. Portland, Productivity Press, 1961. 7, 9 [17] N. B. Forrester. A dynamic synthesis of basic macroeconomic theory : implications for stabilization policy analysis. PhD thesis, Massachusetts Institute of Technology, Sloan School of Management, 1982. 12, 15 [18] R. A. Garrido and H. S. Mahmassani. Forecasting freight transportation demand with the space-time multinomial probit model. Transportation Research Part B, 34:403–418, 2000. 6 [19] B. Güneralp. Towards coherent loop dominance analysis: progress in eigenvalue elasticity analysis. System Dynamics Review, 22(3):263–289, 2006. 12 [20] C. W. J. Granger and Y. Jeon. Long-term forecasting and evaluation. International Journal of Forecasting, 23(4):539–551, 2007. 5 [21] O.-P. Hilmola, U. Tapaninen, E. Terk, and V.-V. Savolainen. Container transit in Finland and Estonia - Current status, future demand, and implications on infrastructure investments in transportation chain. Technical report, Centre for Maritime Studies, University of Turku, 2007. 15, 25 [22] Institute of Shipping Analysis, BMT Transport Solutions GmbH, and Centre for Maritime Studies. Baltic Maritime Outlook, 2006. 20 [23] Y. V. Joshi. Information Visibility And Its Effect On Supply Chain Dynamics. Master’s thesis, Massachusetts Institute of Technology, 2000. 17

BIBLIOGRAPHY

67

[24] T. Kadoya, T. Sasaki, S. Ihara, E. Larose, M. Sanford, A. K. Graham, C. A. Stephens, and C. K. Eubanks. Utilizing System Dynamics Modeling to Examine Impact of Deregulation on Generation Capacity Growth. Proceedings of the IEEE, 93(11):2060–2069, 2005. 15 [25] C. E. Kampmann and R. Oliva. Loop eigenvalue elasticity analysis: three case studies. System Dynamics Review, 22(2):141–162, 2006. 12 [26] T. Karvonen. Investoinnit Suomen satamiin 2001-2010 (Investments in Finnish ports 2001-2010). Finnish Maritime Administration, 2007. 26 [27] W. H. K. Lam, P. L. P. Ng, W. Seabrooke, and E. C. M. Hui. Forecasts and Reliability Analysis of Port Cargo Throughput in Hong Kong. Journal of Urban Planning and Development, 130(3):133–144, 2004. 6 [28] L. Ljung and T. Glad. Modeling of dynamic systems. Prentice-Hall, 1994. 3 [29] J. M. Lyneis. System dynamics for market forecasting and structural analysis. System Dynamics Review, 16(1):3–25, 2000. 6 [30] J. M. Lyneis and A. L. Pugh. Automated vs. ‘hand’ calibration of system dynamics models: An experiment with a simple project model. In Proceedings of the 1996 International System Dynamics Conference, 1996. 42 [31] B. Mezhepoglu and L. Sherry. Portfolio analysis of air transport infrastructure investment. Technical report, CATSR Working Paper, CATSR, George Mason University, 2006. 14 [32] Ministry of Labour. Työvoima 2025 - Työpoliittinen tutkimus, 2007. viii, 22, 23 [33] M. Mojtahedzadeh, D. Andersen, and G. P. Richardson. Using Digest to implement the pathway participation method for detecting influential system structure. System Dynamics Review, 20(1):1–20, 2004. 12 [34] M. Moorthy, F. E. Cellier, and J. T. LaFrance. Predicting U.S. food demand in the 20th century: A new look at System Dynamics. In Proceedings of SPIE – Volume 3369, 1998. 15 [35] M. G. Morgan and M. Henrion. Uncertainty: A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis. Cambridge University Press, 1990. 18 [36] P. Nijkamp, A. Reggiani, and W. F. Tsang. Comparative modelling of interregional transport flows: Applications to multimodal European freight transport. European Journal of Operational Research, 155(3):584–602, 2004. 6

BIBLIOGRAPHY

68

[37] R. Oliva. Model calibration as a testing strategy for system dynamics models. European Journal of Operational Research, 151(3):552–568, 2003. 42 [38] F. Olsina, F. Garcés, and H. J. Haubrich. Modeling long-term dynamics of electricity markets. Energy Policy, 34:1411–1433, 2006. 14 [39] J. Randers and U. Göluke. Forecasting turning points in shipping freight rates: lessons from 30 years of practical effort. System Dynamics Review, 23(2/3):253– 284, 2007. 14 [40] J. Richards. A flexible growth function for empirical use. Journal of Experimental Botany, 10:290–300, 1959. 32 [41] A. M. Rodrigues, D. J. Bowersox, and R. J. Calantone. Estimation of Global and National Logistics Expenditures: 2002 Data Update. Journal of Business Logistics, 26, 2005. 6 [42] W. Schade and M. Krail. Modeling and calibration of large scale system dynamics models: the case of the ASTRA model. In Proceedings of the 24th International Conference of the System Dynamics Society, 2006. 18, 43 [43] P. M. Senge. The Fifth Discipline: The Art and Practice of the Learning Organization. Doubleday/Currency, 1990. 7, 12, 35, 63, 70 [44] H. A. Simon. Models of Bounded Rationality. MIT Press, 1982. 10 [45] P. C. Smith and A. van Ackere. A note on the integration of system dynamics and economic models. Journal of Economic Dynamics and Control, 26(1):1–10, 2002. 7 [46] J. D. Sterman. Business Dynamics: Systems Thinking and Modeling for a Complex World. Irwin McGraw-Hill, 2000. viii, 7, 9, 10, 11, 13, 16, 32, 40, 63 [47] J. D. Sterman. An integrated theory of the economic long wave. Massachusetts Institute of Technology, 1984. 15 [48] A. J. Taylor. System Dynamics in Shipping. Operational Research Quarterly, 27 (1):41–56, 1976. 14 [49] D. R. Towill. Industrial dynamics modelling of supply chains. International Journal of Physical Distribution & Logistics Management, 26(2):23–42, 1996. 17 [50] D. R. Towill. Time compression and supply chain management - a guided tour. Supply Chain Management, 1(1):15–27, 1996. 17

BIBLIOGRAPHY

69

[51] K. Yamaguchi. Money Supply and Creation of Deposits - SD Macroeconomic Modeling (1). In Proceedings of the 22th International Conference of the System Dynamics Society, 2004. 15 [52] K. Yamaguchi. Aggregate Demand Equilibria and Price Flexibility - SD Macroeconomic Modeling (2). In Proceedings of the 23th International Conference of the System Dynamics Society, 2005. 15 [53] K. Yamaguchi. Integration of Real and Monetary Sectors with Labor Market SD Macroeconomic Modeling (3). In Proceedings of the 24th International Conference of the System Dynamics Society, 2006. 15 [54] K. Yamaguchi. Balance of Payments and Foreign Exchange Dynamics - SD Macroeconomic Modeling (4). In Proceedings of the 25th International Conference of the System Dynamics Society and 50th Anniversary Celebration, 2007. 15 [55] L. Zhou and B. Heimann U. Clausen. Forecasting a Logistic Service Demand based on Neural Network. In International Conference on Service Systems and Service Management, 2007. 6

Appendix A

Selected systems archetypes Selected systems archetypes from (43):

A.1

Success to the successful

+ Success of A

+ -

R

Resources to A +

Allocation to A instead of B -

Success of B +

R

Resources to B

Figure A.1: Success to the successful

70

APPENDIX A. SELECTED SYSTEMS ARCHETYPES

A.2

71

Growth and underinvestment

Growing action +

R

+ + Demand B Performance + Capacity +

B

+ Perceived need to invest

Investment in + capacity

Figure A.2: Growth and underinvestment

Performance standard