MA Thesis M. Jansen 386267
MA Thesis M. Jansen 386267
MA Thesis M. Jansen 386267
Preface
This report is my final effort in order to graduate the Master Urban, Port & Transport Economics at
the Erasmus University Rotterdam. It is written commissioned by Witteveen+Bos to give guidelines
for their container cargo forecasting practices.
I would like to thank my supervisor at the EUR, Michiel Nijdam, for his guidance and feedback. Also, I
would like to thank my supervisors at Witteveen+Bos: Bert Burgers for his detailed feedback and
Ben-Jaap Pielage for making sure I would not lose sight of the bigger picture. Acknowledgements go
out to Lrnt Tavasszy (TNO and TU Delft) and Aernoud Willeumier (Port of Rotterdam) who
managed to make time to provide me with useful information.
Mathilde Jansen
Rotterdam, September 2014
Abstract
At the time of writing this thesis, no universal method for making forecasts on container cargo
throughput in seaports is conducted by consultancy firms and other interested parties. This report
gives an overview of the different models used for freight modelling. These models are based on the
four step approach that is used in modelling demand for passenger transport: production and
attraction, distribution, modal split and assignment to the transport network. Based on this
overview, essential elements of a good forecasting method are identified. These are a model to
determine the generation and distribution of freight flows and a model to explain port choice and
competition. The input and output of these models should be evaluated using expert and
commodity specific knowledge, resulting in possible modification of the results. Case studies on
different port categories - i.e. transhipment, import-export, Greenfield, Brownfield, well-developed
and less-developed economies - are used to illustrate the factors that explain the fluctuations of
container cargo throughput volumes in practice. These findings have possible implications for the
optimal set of variables and/or port choice model used. The purpose of this thesis is to provide a
solid basis to develop an integrated and universal forecasting method.
Table of contents
1
Introduction
2.1
Models for production and attraction
2.1.1 Time series models
2.1.2 System dynamics models
2.1.3 Regression analysis on cross-sectional data
2.1.4 I/O models
7
8
10
10
14
2.2
Models for distribution
2.2.1 Gravity model
2.2.2 Multiregional I/O model
15
15
16
2.3
17
2.4
20
2.5
22
2.6
Qualitative assessment
22
26
3.1
26
3.2
26
3.3
Assignment of container flows to the port network
3.3.1 The World Container Model
3.3.2 The Port Competition Model
28
29
31
3.4
33
35
4.1
Transhipment vs. import-export
4.1.1 The port of Salalah, Oman
4.1.2 The port of Auckland, New Zealand
4.1.3 Comparison Salalah with Auckland
35
37
39
41
4.2
Greenfield vs. Brownfield
4.2.1 The port of Filyos, Turkey
4.2.2 Port of Rotterdam
4.2.3 Comparison Filyos with Rotterdam
42
43
45
47
4.3
Developing vs. developed economy
4.3.1 The port of Dar es Salaam, Tanzania
4.3.2 Copenhagen Malm Port
4.3.3 Comparison Dar es Salaam with Copenhagen-Malm
48
50
52
53
4.4
Findings and implications for each port type
4.4.1 Transhipment port
4.4.2 Import-export port
55
55
55
Greenfield port
Brownfield port
Port in a less developed country
Port in a well-developed country
56
56
56
57
58
5.1
58
5.2
60
5.3
61
References
62
1 Introduction
In the past decades, globalization caused a huge increase in trade volumes. Because of this increase,
new port areas had to be developed. This process is still going on, as can be seen from the
development of Maasvlakte II at the Port of Rotterdam for example. Investment decisions on these
kind of expansion projects are usually based on port traffic and cargo throughput forecasts. These
forecasts are made for at least 25 years ahead, since ports generally have a long technical and
economic lifetime, and investments made in port infrastructure have a long pay-back period (De
Langen, Van Meijeren & Tavasszy, 2012a). Until now, no universal method is developed to do these
forecasts. The main approach used for existing ports is based on trend extrapolation techniques that
rely on historic data. However, such models give little insight in the variables that drive cargo
throughput in ports other than economic growth. For Greenfield ports it is noted that trend
extrapolation is not possible, thus requiring an alternative approach.
The purpose of this thesis is understanding the variables that affect container cargo throughput. The
analysis is made for this specific type of cargo, because in contrast to bulk cargo containers can be
used for a wide range of consumer, construction and industry related commodities and clients. This
makes its development subject to trends and general economic development. Mass containerization
in the past decades, for example, caused a major growth in container throughput in ports.
Therefore, many port expansion or development projects are about developing (additional)
container terminals (De Langen, Nijdam & Van der Lugt, 2012b). It is assumed that in the future,
more and more commodities are being containerized (Havenga & Van Eeden, 2011) causing the
container shipping industry to expand even more. Havenga & Van Eeden (2011) also predict a
maturing in the containerization trend, simply because on a given point in time every commodity
that can be shipped in containers shall be shipped in containers.
The focus is not solely on the variables. Different modelling techniques are evaluated to prepare a
general pragmatic approach for container cargo throughput forecasting in ports. These methods are
generally based on the four step approach that is also used in passenger transport modelling:
production & attraction, distribution, modal split and route choice. The evaluation of these
techniques serve as the basis for identifying the essential elements of a good container cargo
forecasting approach.
To verify whether the drivers mentioned in the literature are indeed the factors that determine
container throughput volumes, the developments of throughput volumes in several ports are
2.1
Models for production and attraction illustrate the quantities of goods that are transported between
two regions. These goods start at the production zone and are transported to the attraction zone.
Zones can be countries, a group of countries or a region within a country. This is up to the researcher
to decide. Attraction exists when there is demand for a certain product that is produced in another
region. This causes a flow of a goods from zone to zone.
7
10
Import prices (Coto-Milln, Baos-Pino & Castro, 2005), the higher the import prices, the
lower the transport volume.
Trade value (Seabrooke et al., 2003), because demand for port services is an outcome of
demand for imports and exports.
Population (Seabrooke at al., 2003; Gosasang et al., 2010), the larger the population, the
higher demand, which causes increasing trade flows.
Exchange rates (Gosasang et al., 2010), unfavourable exchange rates discourage trade which
will make the trade flows decline.
Interest rates (Gosasang et al., 2010), interest rates partly determine the value of GDP and
GDP is a good indicator of port throughput because this is determined by imports and
exports, which are a function of GDP.
Inflation rates (Gosasang et al., 2010), inflation rates also partly determine the value of GDP.
Trade between biggest trade partner of the country and the country itself (Hui, Seabrooke &
Wong, 2004), this determinant is derived from the variable trade value, that is mentioned
11
Research on NNs shows that these models perform better than regression analyses. NNs seem to be
able to follow volatilities better than regression analyses. The latter can only follow the middle of the
trend. For forecasting cargo throughput it is better to use NNs because sometimes the volumes
depend on explanatory variables that are very volatile over time. In short, neural network models
give more trustworthy results when forecasting cargo throughput, than regression analyses (Lam et
al., 2004). However, although in terms of results a neural network model outperforms other causeand-effect models, it is difficult to use in practice, because it is a very complicated mathematical
model for which specified knowledge is required. Also, a lot of data are required which are generally
hardly available for ports.
13
Advantages
Easy to use in practice
Differenced regression
Neural networks
Disadvantages
Does not capture the real
causalities because variables are
nonstationary
Only considers short term
adjustments
Co-integration between variables
is required, otherwise the model is
useless
A lot of data is needed as input
Difficult to use in practice
Table 1: Advantages and disadvantages of the different cause and effect models. Based on Hui et al. (2004), Gosasang et
al. (2010), Van Dorsser et al. (2011) & Lam et al. (2004).
Advantages
Limited data requirements
(but for many years)
System dynamics
Disadvantages
Little insight into causality
and limited scope for policy
effects
Not perfectly suitable for
long term forecasts
No statistical tests on
parameter values
14
Variables
None
Mostly GDP
Regression
analysis
Input-output
Designed to illustrate
causal effects
Link to the economy
Can give land-use
interactions
Policy effects in case of
elastic coefficients
Table 2: Freight transport models for production and attraction. Based on De Jong et al. (2004).
2.2
The distribution of freight transport is based on the output of the production and attraction step
(the quantities of trade) and a measure of transport resistance, expressed as generalized transport
costs.
15
The gravity model is a clear model that is easy to understand and use. Data requirements are
limited, which is positive. However, policy effects and other explanatory variables than trade
resistance and trade arrangements cannot easily be included in the model. These variables can play
a role in the next step that covers port choice. Overall, the model is considered useful in forecasting
freight transport.
Advantages
Limited data requirements
Some policy effects
through transport cost
function
Input-output
(I/O)
Disadvantages
Limited scope for including
explanatory variables and
policy effects
Limited number of calibration
parameters
I/O table required, preferably
multiregional
Restrictive assumptions if fixed
coefficients
Need conversion from values to
tonnes
Table 3: Freight transport models for distribution. Based on De Jong et al. (2004).
16
Variables
Potential supply
Potential demand
Trade resistance
Trade arrangements
The amount (in money
units) that a sector delivers
to other sectors
Final demand
Imports
Exports
2.3
In the third step, the flows of goods from the two previous steps are distributed to different modes.
Modal split concerns the distribution of cargo between transport modes, i.e. marine, road, rail and
inland waterway (IWT). There are several kinds of models that deal with modal split. De Jong et al.
(2004) mention seven models in their paper, those are all briefly described below.
The first are elasticity models. These models use elasticities that are adopted from other models or
expert judgement. The elasticities show the effects of changing a single variable and are mainly used
when very little data are available, or for a quick first analysis.
Aggregate modal split models use data on shares of a mode in a specific zone and put these in a
binomial or multinomial logit model. So the output is a share instead of an absolute number and
thus the elasticities in the model are conditional on the quantity demanded. Although little data is
required as input for this model, it has a weak theoretical basis and there is little insight into
causality.
Next, neoclassical models are based on microeconomic theory. The share of a particular mode in the
total cost is usually the explanatory variable in these kind of models. So the share in total costs
instead of the share in transport volume, which is the relevant variable in the greater four step
system, is important in neoclassical models. This difference makes it hard to incorporate neoclassical
models output in the greater transport model system.
Models that are also hard to incorporate in the four step system are direct demand models. This is
because transport routes for a specific mode are predicted directly by these models, while in the
greater system the total demand over all modes is used.
Disaggregate modal split models use data from surveys conducted under shippers or commodity
surveys. These data are used in a multinomial or nested logit model. The approach is based on
passenger transport models: the utility functions of such models are adjusted to a profit function
that is suitable for freight transport. Disaggregate modal split models have more advantages than
drawbacks. It is possible to include many causal variables in the model and there is a solid
theoretical basis: microeconomic theory. However, disaggregated data is required as input, which
means surveys have to be conducted. This is a time consuming way of data collection.
A microsimulation model concerns trips for freight transported by trucks. It is actually a part of a two
stage model of which the first part determines flows between regions and the second part, a
17
18
Advantages
Very limited data
requirements
Fast in application
Neoclassical
Direct demand
Disaggregate modal
split
Microsimulation
approach
Multimodal network
(WCM & PCM)
Theoretical basis
Opportunity to include
many causal variables and
policy measures
Many behavioural choices
Links to theory included
Disadvantages
Elasticities may not be
transferable
Only impact of single
variables, no interactions
between variables
Weak theoretical basis
Little insight into causality
Limited scope for policy
effects
Hard to combine in larger
transport model system
Hard to combine in larger
transport model system
Table 4: Freight transport models for modal split. Based on De Jong et al. (2004).
19
Variables
Freight transport flows,
aggregate transport
networks, vehicle
usage/specifications
Market share of a mode
Observed data on
distributions
Transport time
Terminal cost
Transport cost
Other port specific
variables
2.4
In the previous step, freight in tonnes is converted into vehicle-units. In the last step these vehicleunits are assigned to networks. Theoretically, this means the truck, rail or inland waterway trips are
assigned to routes that consist of links between these different modes. In practice, many models do
not even incorporate the separate assignment step and most models only consider assignment to
trucks. In this case, this is often done together with the assignment of passenger road traffic. This is
because trucks usually make up a small proportion of total road traffic. It is also possible to replace
the assignment step by a multimodal network that is described in the previous section (De Jong et
al., 2004). The fact that a separate assignment model focuses merely on road traffic, makes it less
relevant for this research. Therefore it is recommended to use the multimodal network approach so
that the last two steps of the system can be covered jointly. However, it is important to keep in mind
that there is one major drawback on the use of this model in step 4. Because an optimization
mechanism (in the shape of a cost minimization algorithm) is used in this step, sometimes the
output contains illogical mode-route solutions. Indeed travel choices, are in addition to cost also
based on transport time.
Type of model
Separate
assignment
Multimodal
network
Advantages
Mode choice model can be
disaggregate
Allows interaction with
passenger trips if freight
and passenger trips are
assigned together
Substitution takes place
between mode-route
combinations
Chains with different modes
on a route can be handled
Disadvantages
No interaction between
demand and assignment can
be unrealistic
Transport chains are difficult
to incorporate
Variables
?
Transport time
Terminal cost
Table 5: Models for the assignment step in freight transport models. Based on De Jong et al. (2004).
21
2.5
Tavasszy (2006) argues that a fifth logistics step is needed to complete the freight transport
modelling process, because logistics are changing and have impact on the distribution of freight. In
the past decades it can be seen that transport costs have declined substantially. This is caused by the
fact that supply chains are more efficient, which is made possible by globalization and the
improvement and developments of information technology. Because of these cost and efficiency
improvements in terms of logistics services, we are now heading to a time wherein masscustomized logistics services are the standard (Tavasszy, Ruijgrok & Davydenko, 2012). In this step
the locations of distribution centres should be determined. This can be done by incorporating an
inventory location model that helps to give a more realistic estimate of interaction costs, and so
decreases the probability of bias in volume forecasts. Also, it helps to forecast the effect of changes
in logistics services.
An approach to account for logistics cost can be to extend the gravity model, which is done by
Hausman, Lee and Subramanian (2005) and Hummels & Schaur (2012). Not only transport costs are
of influence of the probability of trade, but logistics costs (including transport costs) act as a
resistance to trade. In both researches it is found that logistics cost have a significant effect on trade.
According to Tavasszy et al. (2009), to incorporate logistics choices in freight models it is necessary
to use the Generalized Cost concept. This concept includes all costs that are involved in overcoming
time and space (Tavasszy et al., 2009). It is assumed that companies have the objective to minimize
these generalized cost. Tavasszy et al. (2009) propose this concept has to include shipment size,
speed, value density, demand uncertainty, scale economies and network synchronization to cover
logistics choices. A disadvantage of using the Generalized Cost concept is that it does not take into
account trade-offs between variables. Models like the World Container Model and the Port
Competition Model do take into account these trade-offs. They do, however, do not take into
account the locations of distribution centres. Hence, the models should be adjusted to suit this layer
or new models have to be developed.
2.6
Qualitative assessment
In the previous sections it became clear that some variables are hard to include in the model. This is
predominantly caused by the fact that a lot of data is needed to make a variable usable. In this
event, qualitative methods can also be used to forecast freight flows. This is especially done when
historical data is not available. If these data are missing, a qualitative analysis and prediction of the
variables can give insight in the movement and the influence of the variable in the future. The
22
23
24
Expert judgement
Advantages
Gives more insight and awareness of
disruptions that cannot be included in the
model
Disadvantages
Data is not measurable and cannot be
verified
Table 6: Advantages and disadvantages of the expert judgement method. Based on De Langen et al. (2012a).
25
3.1
From the previous chapter it becomes clear that a good cargo forecasting method has several
characteristics. First, in the maritime industry data is often not available in large quantities and
detail. This makes it better to use a model that does not require a lot of data. Also because using a
lot of different data makes the forecasts less reliable, since it is more prone to error (Bankes, 1993).
Second, the model should make clear how trade flows are generated. This is based on information
about supply and demand in different regions. Based on this information, supply and demand (or
production and consumption) regions are linked to each other to show how these trade flows are
distributed over the world. Here, macroeconomic variables are used to forecast freight flows. Third,
factors that determine the competitiveness of a port have to be taken into account. Because not
only the existence of trade explains the cargo throughput in ports but also the attractiveness of the
port. This is because many ports in the world serve the same hinterland and port specific
characteristics determine which port gets the biggest share of the trade flow that is destined for this
hinterland. Finally, preferably the model accounts for shocks in the economy and trends and
developments in the industry. It has shown that this is very difficult to include in a mathematical
model. Therefore, it is also possible to analyse and if it is desired - adjust the results of quantitative
models by use of expert judgement on these trends and shocks.
3.2
Flows of trade form the basis of a freight transport model. Therefore, as a first step, generation and
distribution of trade flows should be determined. As mentioned before, GDP is an important
indicator of trade. This is supported by the papers of Stack (2009), Hausman et al. (2005) and De
Groot, Linders, Rietveld & Subramanian (2003). The results of these researches state that 70 percent
of bilateral trade is explained by GDP, GDP per capita and distance between countries. This arises
26
Yi,j
yi,j
Dij
Aij
As mentioned above and can be seen in the model, 70 percent of bilateral trade is determined by
GDP, GDP per capita and the distance between the countries. A trade arrangement dummy is
included to account for factors that reduce the resistance of trade, measured by distance. Note that
27
3.3
In the second step, the container flows that are determined in the previous step should be assigned
to the maritime transport network. To do this, first a maritime networks should be mapped,
containing nodes and links between different ports and (main) sea routes. For example, Tavasszy et
al. (2011) build their transport network based on weekly container liner routes that are provided by
container shipping lines. Information on liner schedules and transport fees that are publically
available were used to construct a database containing 800 maritime services and 437 ports.
Eventually, the network was developed by using one land node per country (origin and destination),
port nodes and maritime nodes. As a result the transport network was illustrated as shown below.
Figure 3: Maritime network with country access links o and from ports. Source: Tavasszy et al. (2011).
Next, port competition should be included in the approach. In the previous chapter, two port choice
models are discussed briefly: the World Container Model and the Port Competition Model. Here, a
more extensive (mathematical) description is given. The objective of incorporating a port
competition model is to account for characteristics of a port that determine its competitive position
with respect to other ports in the region.
28
This was done because in the simple logit model, the routes where bundled and therefore the routes that
overlap (are in the same bundle) would be overestimated. So, a path size overlap variable is added to the
model. This is a variable that depicts the degree of overlap based on the sum of the relative lengths of the
routes, divided by the number of times a link is found in alternative routes (Tavasszy et al., 2011).
29
za
length of link a
zr
length of route r
Nah
Ap
cl
Tp
tl
Modal split is not explicitly expressed in the model. Instead the developers chose to use a modeabstract formulation, which enables the researcher to use a more detailed underlying multimodal
network. The parameter Ap is made of several characteristics that measure the service level at the
port, like fuel costs, handling costs, congestion costs, etc. Another important parameter in the model
is the value of time. This variable indicates the preference of shippers for fast and more expensive or
slow and cheaper transport. This parameter contains all the characteristics of the good, that are also
mentioned above. Next a set of route choices is determined using a shortest path algorithm to
decide what is the preferred route between the origin country and the destination country, via the
ports in the country. This can be pictured as follows.
30
Where O and D stand for origin and destination, respectively, and S and E are the ports of these
countries. The shortest route is the thick scattered line (Tavasszy et al., 2011).
In their paper, Tavasszy et al. (2011) use the WCM to estimate the effect of different scenarios that
illustrate a change in the transport system, like opening a shipping route over the North instead of
via the Suez canal. Using the WCM, the authors show the changes in the proportion of total
container flows that comes into a port, per scenario. It is an interesting feature that the model can
account for possible changes and show the effect of these changes. The forecasting of effects of
changes is very hard to do with a model and therefore, qualitative assessments are used. The WCM
gives the opportunity to analyse the effects of changes in the transport system quantitatively.
31
Where,
Pm
Um
routing index
Where,
Dm
Cm
shipping cost of routing m (incl. freight rate, handling charges, hinterland transport)
Tm
Fm
32
So, the market share ratio is a function of the differences in utilities, determined by characteristics of
the port in terms of shipping cost, transit time and frequency of service. The dummy variable Dm
stands for all other factors that affect port choice, like the unquantifiable variables mentioned
above. The ratio of two different coefficients is calculated to find the trade-off between two
variables. For example 1/2 illustrates the trade-off between shipping costs and transit time. The
values of these coefficients depend on several factors like geographical setting, model specification,
choice situation and the level of aggregation (Veldman & Bckmann, 2003).
The variable transport costs stands for the costs of transporting a container from the port of choice
to the destination in the hinterland. The costs for sea transport are deliberately omitted, because
carriers charge the same tariff to each port. Transport time is the time measured between the
departure of the container from the port and the arrival at the final destination in the hinterland. For
the frequency of service variable the reciprocal of the frequency is used. This is expressed by the
average inter-arrival time (IAT) of two sequential calls of a liner vessel or another hinterland
transport mode. This displays the fact that a port that offers many services is more attractive. This
concept can also be applied to the different hinterland modes. The busier the mode, the more
attractive it is assumed to be. It is also noted by Veldman and Bckmann (2003) that the model is
improved by adding dummy variables for the three different inland transport modes: road, rail and
inland waterway.
3.4
Many forecasting approaches use scenarios as input for their quantitative models. These scenarios
are often developed by public agencies and include assumptions on future developments. Although
scenarios account for uncertainty and give some direction for the future, they do not account for
external shocks in the economy, because they are mainly based on historical data.
For quantitative models in freight forecasting, usually publicly available data is used. These data are
globally focused and thus not very detailed. No wonder these models cannot account for strategic
developments in production networks or technological innovation, for example. Therefore, it can be
33
Figure 5: Combination of model results with expert judgment. Source: De Langen et al. (2012a).
34
4.1
Ports can offer import, export and transhipment services. Import and export services are the obvious
ones and have a clear relation with economic growth as they directly explain trade and thus port
throughput. Transhipment is a more complicated issue in forecasting container throughput as it has
less to do with the economic situation of the hinterland of the transhipment port and thus the
destination of the transported product. As a matter of fact, port throughput can be explained based
on macroeconomic factors or regional port competition variables. Transhipment has something to
do with the latter category. The success of a port that focuses on transhipment services mainly,
heavily depends on the main routes of the biggest liner vessels. This is because the costs of diverting
from these routes have to be low in order to make transhipment profitable, so the port should not
be situated too far from the main routes. Also, the distance from the port to the market served and
the handling costs in the transhipment port are of importance for the competitive position of the
port as an important transhipment hub in the maritime network (Zohil & Prijon, 1999). To include
this view into a model, Zohil and Prijon (1999) proposed a simple regression model with the
estimated number of TEU transhipped in a port as dependent variable. This number is explained by
35
Million TEU
500
400
Import-export
300
Transhipment
200
100
0
1990
1995
2000
2005
2009
2012
Year
Figure 6: World container throughput volumes and the share of transhipment (million TEU). Based on: Tavasszy et al.
(2014).
This increase is also due to the fact that ships will get bigger, so that shipping lines can profit from
economies of scale (Baird, 2006; De Langen et al., 2012a; De Langen et al., 2012b). By making use of
hub-and-spoke systems, the main liner vessels only have to enter big transhipment ports that can
handle big ships with a deep draft. From there, feeder ships can take the freight to smaller regional
ports (Lam & Iskounen, 2010). In general, the main focus is on the cost savings shipping lines achieve
by making use of hub-and-spoke systems. However, Lam and Iskounen (2010) and Baird (2006)
mention the downside of the emergence of transhipment ports. Baird (2006) states that the benefits
of transhipment are a trade-off between extra feeder and handling costs against the extra costs of
calling at an additional port. He argues that for none of the transhipment ports in Nothern Europe
the feeder and handling costs of transhipment services are lower than the costs of calling an extra
port. Thus, a dedicated transhipment port should be established to serve the whole Northern
36
www.vacationstogo.com.
container throughput volumes in the port. The drop in volumes in 2006 is caused by the shipping line
Maersk, that decided to realign its vessel schedule. This illustrates that seaports are very dependent
on the strategic decisions of other players in the market, like shipping lines. Unfortunately, strategic
decisions of shipping lines cannot be predicted, which makes it hard to account for its effects.
The fast growth between 2006 and 2008 is caused by the high oil price in that period. This increased
investments in infrastructure in the Middle East, causing container throughput to grow by 12%. The
position of the port of Salalah at this point is stable, having the certainty of long term contracts with
the three major shipping lines Maersk, Mediterranean Shipping Company and American President
Line. However, in 2008 the credit crisis started, causing freight rates to fall by 15%. Together with
the high oil prices, shipping lines went to a hard time. On top of that, shipping lines started to invest
in Ultra Large Container Ships (ULCS) to profit from economies of scale, triggered by the high levels
of uncertainty in the market. This resulted in over capacity. This all led do a decrease in container
throughput between 2010 and 2011. However, economies of scale is one of the biggest drivers of
transhipment, causing the port of Salalah to recover quickly.
As can be seen in the last few years, the container transhipment market is volatile. More than
import-export markets, because shipping lines have more options to tranship their cargo somewhere
on the route. The final destination of the freight does not matter in that context. Hence,
transhipment ports highly dependent on shipping lines. These shipping lines have more and more
power as many of the pure transhipment terminals are (partly) owned by carriers and international
terminal operators. Therefore, fluctuations and transhipment volumes are hard to predict
(Notteboom et al., 2014).
38
3800
3600
3400
3200
3000
Container throughput volume in
TEU (x1000)
2800
2600
2400
2200
2000
Figure 8: Development of container throughput in the port of Salalah between 2004 and 2013.
Based on this analysis it can be concluded that it is very hard to model what drives the container
throughput in transhipment ports, because this heavily depends on the strategic decisions of
shipping lines and these cannot be predicted. It is true that transhipment ports compete in service
levels and location.
found for this port. However, the review on the development of the throughput volumes starts from
2008 onwards.
39
1000
950
900
850
800
750
700
650
600
Figure 10: Development of container throughput in the port of Auckland between 2004 and 2013.
In the years 2008-2010 container volumes were small because of volatility in the shipping industry.
Shipping lines were rationalizing their schedules to save costs and achieve greater efficiency. This
effect is also visible in the port of Salalah. Transhipment volumes all over the world increase caused
by liner shippers search for cost savings and greater efficiency, investing in bigger vessels to benefit
from economies of scale. This also effects the port of Auckland, where transhipment volumes
increase by 55% in 2008 (Port of Auckland, 2009). Because the port of Auckland is located far away
from the main transhipment routes, it is assumed these transhipment volumes are destined for
smaller ports in New Zealand. This means the same country is served and thus the same
macroeconomic variables are of importance to predict the freight flows. This is exactly the difference
between transhipment in the port of Salalah and in Auckland, as for the first macroeconomic
variables do not predict transhipment volumes. In 2009, transhipment volumes increased even more
by 10%. Although this is a big increase, import and export seem the core services of the port. This
indicates that transhipment rates before 2008 were probably extremely low, causing an increase in
percentages to have little impact on entire throughput volumes.
The abrupt fall in throughput volumes in 2011 is a result of an industrial dispute, causing strikes and
actions on the terminal. Container volumes fell down by 9.6% because of loss of significant services
due to the strikes on the terminal (Port of Auckland, 2011). No agreement was reached for the
beginning of 2012, which explains the small growth in container volumes. Meanwhile, the port
started with re-organizing its business into a more customer focussed organisation. On top of that,
operations became more efficient and flexible. This made it possible to attract Maersk back, one of
40
4500000
4400000
40%
4300000
4200000
20%
4100000
4000000
0%
2008
2009
2010
Import
2011
2012
2013
Export
Population
4.2
Port expansion projects for container terminals are based on a container cargo throughput forecast.
This is very important because the required capacity of the new terminals is determined based on
this forecast. Port expansion projects can either concern new ports (Greenfield) or expansion of
existing ports (Brownfield). Unsurprisingly, historical data on throughput volumes is not available in
case of Greenfield projects. This rises the assumption that when it comes to forecasts, Greenfield
and Brownfield projects require different approaches based on the amount and nature of the data
available. In case of a Greenfield port it can be that another port is already situated in the country.
So there is already a flow of goods, and the question is how this new port will attract cargo and so
competes with the existing port. This relies heavily on competitive characteristics like if the port is
suitable for transhipment (see 4.1), or the hinterland connections of the port.
Most of the above described models raise the assumption that they focus on existing ports. For
these ports, historical data on throughput volumes is available and can be extrapolated into the
future. However, extrapolation of trends is not the desired approach for forecasting future
throughput. So the question is, is there a substantial difference between Greenfield and Brownfield
projects in choosing the best models? The difference in this situation is that for Brownfield ports,
this competitive position can be expressed quantitatively by making use of, for example, the LSCI
(see 2.1.2.1) (Zohil & Prijon, 1999). In case of Greenfield ports, no numbers are available and so the
analysis of its competitiveness should be a qualitative comparison with the biggest competitors or
based on goals. This makes the forecast less reliable, because it is based on predictions on the
competitiveness of the port rather than on current situations.
42
Figure 13: An overview of the transport routes in Turkey. Based on Donders (2010).
43
44
Reference scenario
Sensitivity low scenario
No Filyos port
Certain investments,
Middle parameters
Do nothing scenario
-
To not fully ignore competition between ports, the writers of the report made an analysis of the cost
structures. All possible routes per OD pair are compared in terms of generalized costs and the trade
flow is assigned to the cheapest route. This is approach is similar to the multiregional I/O model
described in chapter 2. It is especially important for container traffic to take into account generalized
costs, because these ships are more flexible compared to bulk carriers when it comes to route choice
(CFCU, 2009). A disadvantage to this approach is that it does not include trade-offs between costs
and quality of service (Veldman & Bckmann, 2003). The forecast for Filyos would therefore benefit
from including a port competition/choice model. However, many port characteristics are not yet
known for sure and thus such a model would depend too much on speculation. That makes its
reliability questionable.
Overall, the forecasting method used for the port of Filyos seems appropriate. Use is made of a
regional model to determine trade flows and assign these flows to the possible routes according to
generalized costs. Hence, macro-economic factors (GDP and population) are taken into account for
determining trade flows and competition factors are included by assigning to routes based on
generalized costs. Also, the essential qualitative analysis is used to fill in the gaps of the quantitative
variables. Unfortunately, because the port of Filyos is not yet constructed, no analysis can be made
on the fit of the forecasts with the realized throughput volumes.
Low Growth
Low economic growth and a low fuel price. Fossil fuels stay dominant and environmental
policy is moderate.
European Trend
Continuing existing policy and a moderate growth of the economy.
Global Economy
Further globalisation combined with low fuel price causing high economic growth and
moderate environmental policy.
3,5
3
2,5
2
1,5
1
0,5
0
1990-2000
Low Growth
2000-2010
2010-2020
Global Economy
2020-2030
European Trend
Figure 14: GDP growth per 10 year period, per scenario. Based on HBR (2011).
46
2030-2040
High Oil Price
47
4.3
A country in the third world is not as well developed economically as for example North-Western
countries. This makes that trade flows to and from these countries, and thus their transport facilities
like seaports, are also underdeveloped. This can have the effect that administration in such countries
is less up to date. This makes it more difficult to prepare an O/D-table, because proper data is
48
49
Source: dlca.logcluster.org.
import rates are partly due to the fact that Tanzania serves as a gateway for several land-locked
countries, i.e. Zambia, Democratic Republic of Congo, Malawi, Rwanda and Burundi. Transits of
freight to these countries account for approximately 35% of total traffic. This share should be
greater, given the strategic location of the port to serve as a transport hub. However, hinterland
connections are poor and most of the countries in the hinterland are politically unstable. This is
considered typical for a developing country and although freight for the hinterland does not make
50
300
250
200
Import container throughput in
TEU(x1000)
150
100
50
0
2007
2008
2009
2010
2011
2012
Figure 16: Development of container throughput in the port of Dar es Salaam. Source: Tanzania Port Authority.
Over the years, the proportion of imports and exports is more or less stable. Only in 2008 the
exports exceed the imports. In this year, also a small decrease in throughput volumes is observed.
This is due to the fact that less dry cargo was containerized in 2008 compared to the previous year.
Which is remarkable as the main trend is that more and more cargo is containerized.
Throughput volumes are incessantly increasing in the years under consideration, indicating the port
is able to attract more container traffic. Relative to 2010, there was an increase of 21% in container
traffic calling the port in 2011. Also, the market trend of bigger vessels is affecting the port of Dar es
Salaam in a positive way. At first, the port had to adjust to this movement, as stated in the annual
51
This is one of the reasons why throughput volumes Based on: www.greenroofs.com.
fell heavily with 24% after 2008, the start of the credit crisis. Consumer goods are less cyclical than
industrial products and consumer confidence declined, causing them to purchase less products. CMP
responded to the decline by changing its opening hours to become more flexible and lowering
operating costs in order to become more efficient and attract new shipping lines. These new
customers, CMA CMG and Teamlines, provided the port a marginal increase in volumes in 2010.
However, the volumes were disappointing for CMP because the economy was on its return.
Unfortunately, growing demand for consumer goods did not yet pick up sufficient momentum at
that time, forcing the port to cut its staff.
In 2011, throughput volumes stayed more or less the same. Although there was an increase of
activity in Copenhagen by 6%, caused by higher productivity during vessel operations and the return
of some of the greatest shipping lines in the world, the volumes for Malm are unsatisfactory. This is
probably due to the relocation of the container terminal to the northern harbour. This new
container terminal resulted in overcapacity, responsible for the small drop in container volumes in
2012 as developments in Copenhagen remained stable (CMP, 2012).
52
250
200
150
Container throughput in
TEU(x1000)
100
50
0
2007
2008
2009
2010
2011
2012
Figure 18: Container throughput volumes in Copenhagen Malm Port. Source: CMP.
53
10
8
6
4
Sweden
Denmark
0
-2
2007
2008
2009
2010
2011
2012
Tanzania
-4
-6
-8
Figure 19: Percentage GDP growth for Sweden, Denmark and Tanzania in 2007-2012. Based on: IMF (2012).
Another explanation is that the content of the containers differ between the ports. It is not clear
what is the content of the containers handled in Dar es Salaam, but the containers in CMP contain
consumer products destined for the regional market. The containers handled in Tanzania can also be
destined for their expansive hinterland. Also, it can be reasoned that the economies of African
countries mainly depend on natural resources while the economies of European countries depend
more on the service industry and consumer goods. The latter markets are more volatile and have an
effect on the economy quicker. For example, one of Zambias core businesses is the exportation of
copper. Demand for copper increased in the past decade, causing the price to rise. This is a boost for
the Zambian economy and transport over Dar es Salaam is intensified. Accompanied with the
development of the Zambian economy, demand for commodities transported in containers can go
up, causing throughput volumes in the port of Dar es Salaam to rise.
When it comes to import and port of call, CMP probably faces heavier competition than the port of
Dar es Salaam. Containers destined for Denmark or Sweden can also be transported via another port
in the region, because hinterland connections in that part of Europe are well developed in general.
Hence, costs and time will not necessarily increase by choosing a port that is further away from the
region of destination. For the containers that have to be transported to Tanzania there are not as
much options as for Denmark and Sweden. First of all, the country is a lot bigger and thus the ports
are located further away from each other. Hence, differences in speed and costs will be bigger for
Tanzania than for Northern Europe. Second, the port of Dar es Salaam has the biggest market share
in Tanzania (approximately 90%) and it is therefore assumed that it has the best hinterland
connections, because hinterland connections in developing countries are very poor in general. This
means that Dar es Salaam is the optimal port of choice in terms of costs and time. Therefore,
54
4.4
In this concluding paragraph of the chapter, the findings of the case studies per port category are
discussed and the implications for the form of the models are mentioned. An abstract and holistic
approach is used to describe these implications, leaving the exact formulation of the models to
further research.
56
57
5.1
Forecasting methods nowadays build on the four step approach derived from the passenger
transport literature. These are production & attraction, distribution, modal split and assignment to
the network. For the first step, I/O models or O/D tables are the most suitable solution. These
matrices show per region (or sector in case of I/O) which products are produced and which products
are attracted or demanded. Strongly related to this is the distribution step in which regions of supply
and demand are linked. In this way, the distribution of trade flows arises. The model that is usually
used for this is the gravity model, ascended from physics. Big economies attract many trade flows,
hence there is many trading going on between the richest countries in the world. There is however
also the resistance of trade, for example physical distance or transport cost, causing a reduction in
potential trade flows. Often, assignment to a mode and a certain route are done simultaneously. The
best way to do this is by use of a multimodal network model. These are mainly multinomial logit
models that deal with discrete choice models.
According to this research the essential elements of a good container cargo forecast are threefold.
First, a good forecast has to be made on the level of GDP in the future. This can be done based on
the forecasts of the IMF or other public organizations that are committed to making good GDP
forecasts. Often, scenarios are used to give an idea of the development of GDP based on factors like
oil price, environmental policy, economic growth etc. It is useful to use scenarios when making long
term forecasts, because in that way different trends are considered and assumed possible. This
lowers uncertainty, because when the actual volumes deviate from the forecasted ones there is
always another forecasted trend that is close to the real values. These GDP forecasts are used as
input for the gravity model that estimates the distribution of trade flows. So in this first step towards
a good cargo forecast, macroeconomic variables are taken into account. These together form the
first essential element of container cargo forecasting.
58
59
5.2
Whilst the models recommended in this thesis seem to work in theory, it is necessary to come up
with a more practical advice for Witteveen+Bos. It is possible for the consultancy firm to use the
gravity model, the WCM and the PCM. However, datasets are not always available and the gathering
of data can be labour-intensive. If the formulation of the gravity model is simplified, it should be
feasible to use it for any organisation or person. That is, when trade flows depend on the GDP in the
origin and destination and trade resistance is formulated as the distance between countries. A
disadvantage of this is that the explanatory power of the model decreases. One of the developers of
the WCM, L. Tavasszy, claims that the WCM is useable for everyone and that datasets are publically
available (L. Tavasszy, personal communication, July 11, 2014). The issue here is whether the
datasets are up to date and ready to use, which should be verified before use. In this thesis, an old
version of the PCM is presented that is publically available. The model is developed by the company
Ecorys and is adjusted several times after 2003. The improved version of the model is not available
for free and should be bought at Ecorys. A method that requires less data, but does not account for
the trade-off between costs and quality is to assign flows to the network directly according to
generalized cost. This is also done for the port of Filyos and requires less data.
60
GDP growth is a very solid indicator for cargo growth and they move in the same direction.
Population and income growth are important indicators for the development of import
volumes. They also move in the same direction.
Level of service, costs and connectivity are important variables for transhipment ports. The
higher the level of service, the lower the costs and the better the connectivity (or the shorter
the time of deviation from the main routes), the higher transhipment volumes.
Hinterland connections are important for every port except the transhipment port.
Generalized costs for the route from port to final destination can be used to qualify the
connections.
It is also possible to come up with an advice that is based on the developments in these drivers of
changes in trade flows and throughput volumes in ports. The results are however not verifiable,
because they are not tested statistically.
5.3
There are several shortcomings of this research. One is that case studies of only two opposing ports
are considered for each port category. Therefore, findings cannot be generalized as there is always
the probability that they correspond with port specific element instead of the whole category.
Further and more extensive research on several kind of ports and their implications on the
forecasting method would contribute to the freight modelling literature.
Second, because developing a model that is ready to use does not fit into the scope of this thesis, it
cannot be tested if the recommendations made on the elements of the model are grounded. Further
research is necessary to test whether these elements are indeed essential.
61
References
Baird, A. J. (2006). Optimising the container transhipment hub location in northern Europe. Journal
of transport geography, 14(3), 195-214.
Bankes, S. (1993). Exploratory modeling for policy analysis. Operations Research, 41(3), 435-449.
Central Finance and Contracts Unit (CFCU) (2009). Technical assistance for construction of new port
in Filyos.
Charlier (1983). Ports et regions franaises: line analyse macro-gographique. Acta Geographica
Lovaniensia. 24, 1-198.
Chou, C. C., Chu, C. W., & Liang, G. S. (2008). A modified regression model for forecasting the
volumes of Taiwans import containers. Mathematical and Computer Modelling, 47(9), 797807.
Clarkson Research Services (2013). Container Intelligence Monthly. 15(6).
Copenhagen Malm Port (CMP) (2007,2008,2009,2010,2011,2012). Annual report.
Coto-Milln, P., Baos-Pino, J., & Castro, J. V. (2005). Determinants of the demand for maritime
imports and exports. Transportation Research Part E: Logistics and Transportation
Review, 41(4), 357-372.
Cullinane, K., Teng, Y., & Wang, T. F. (2005). Port competition between Shanghai and Ningbo.
Maritime Policy & Management, 32(4), 331-346.
De Groot, H. L., Linders, G. J., Rietveld, P., & Subramanian, U. (2003). The institutional determinants
of bilateral trade patterns.
De Jong, G., Gunn, H., & Walker, W. (2004). National and international freight transport models: An
overview and ideas for future development. Transport Reviews, 24(1), 103-124.
De Langen, P.W., Nijdam, M.H. & Van der Lugt, L.M. (2012a). Port Economics, Policy and
Management [Syllabus]. Erasmus Universiteit Rotterdam, Erasmus School of Economics.
De Langen, P. W., Van Meijeren, J., & Tavasszy, L. A. (2012b). Combining Models and Commodity
Chain Research for Making Long-Term Projections of Port Throughput: an Application
to the Hamburg-Le Havre Range. European Journal of Transport & Infrastructure
Research, 12(3).
Dikos, G., Marcus, H. S., Papadatos, M. P., & Papakonstantinou, V. (2006). Niver lines: a systemdynamics approach to tanker freight modeling. Interfaces, 36(4), 326-341.
Donders, L. B. (2010). Layout design for greenfield port Filyos (Masters thesis, Delft University of
Technology, The Netherlands).
Europe, R., & Brinckerhoff, P. (2002). Review of Freight Modelling. Report B2Review of Models in
Continental Europe and Elsewhere, DfT.
Garatt, M. (2006). Forecasting for long term investment in the container shipping industry
[Powerpoint-presentation]. Retrieved from www.mdst.co.uk
62
63
64