Вы находитесь на странице: 1из 23

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/265849794

A MODEL OF TRANSHIPMENT PORT


COMPETITION: A TEST WITH CROSS-SECTION
AND TIME-SERIES DATA...

Conference Paper January 2008


DOI: 10.13140/2.1.4304.5128

CITATIONS READS

3 558

4 authors, including:

Simme Veldman
ECORYS Netherlands
13 PUBLICATIONS 137 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Port choice of West African container ports View project

All content following this page was uploaded by Simme Veldman on 21 September 2014.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
A MODEL OF TRANSHIPMENT PORT COMPETITION: A TEST WITH
CROSS-SECTION AND TIME-SERIES DATA FOR THE MEDITERRANEAN

Dr. Simme J.H. VELDMAN


Senior Consultant
ECORYS Transport
P.O. Box 4175, 3006 AD Rotterdam
The Netherlands
E-mail: simme.veldman@ecorys.nl
Tel: +31-10-4538855
Fax: +31-10-4523680

&

Dr. Adnan Rachman


Managing Partner
ECORYS Transport
P.O. Box 4175, 3006 AD Rotterdam
The Netherlands
E-mail: adnan.rachman@ecorys.nl
Tel: +31-10-4538855
Fax: +31-10-4523680

January 2008

Abstract
In the growing market of container transhipment ports worldwide and in the faster growing
market of the Mediterranean it is important for proper planning to understand the competitive
position of these ports and port choice elasticties in particular.

Statistical tests are applied using a 10-year time series of aggregate transhipment flows between
15 transhipment ports and 9 feeder regions. Tests of Logit Models with regression analysis show
that variables such as feeder costs, mainline port access costs and Mohring effects are statistically
significant.

Comparison of the results with those of earlier research concerning the Northwest European
transhipment market shows that the outcomes correspond rather well in terms of the resulting
choice or demand elasticity. It also shows that the value of this elasticity in some cases compares
rather well with those estimated in port choice studies for the non-transhipment market
concerning Northwest European, US and Chinese container trades.

This paper shows that the use of Logit Models with respect to transhipment port choice leads to
useful findings for port planning. This research in combination with earlier research by one of the
authors for transhipment port choice in Northwest Europe is a step forward in the field of
transhipment port choice.

Keywords: Port choice, port competition, container transhipment, aggregate Logit Model,
Mediterranean

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


1 Veldman and Rachman.doc
A MODEL OF CONTAINER PORT COMPETITION: AN APPLICATION FOR
THE TRANSHIPMENT MARKET OF THE MEDITERRANEAN

1. INTRODUCTION

1.1 Problem setting and relation to earlier research

Port stakeholders and financiers of port investment projects require financial and
economic assessments of investments in infrastructure and other types of intervention. If
investments do not take place, this might lead to a deterioration of the quality of service
being offered, congestion and loss of market share. For financiers it is important to know
the relation that exists between the price of port services and the related level of demand.
For public authorities it is also important to know to what extent the investment
contributes to public welfare. In this case the cost of the use of the port also includes not
traded aspects such as transit times, inter-arrival times and quality of service all referred
to as generalised costs. As container ports generally work in a competitive situation it is
important to know the relation between demand of port services and the price of using the
port expressed in generalised costs.

The objective of this paper is to establish the relation between demand for container
transhipment services and the generalised costs of these services. The demand of these
services concerns the transhipment market of the Mediterranean.

The subject of this paper is analysed earlier with respect to container port competition in
Northwest Europe. Veldman and Bckmann (2003) estimated demand functions for both
the continental and the overseas hinterland of the West European major container ports
and assessed the demand function for a port expansion project for the port of Rotterdam.
Veldman et al. (2005) estimated demand functions for a project to improve the
accessibility of the Port of Antwerp by deepening the Scheldt River and thereby reducing
waiting times for the tide and the ability to accommodate bigger ships. In both
publications the parameters of a Logit Model were estimated with regression analysis and
the demand function could be derived by systematically changing cost and assessing the
resulting market shares. Veldman and Vroomen (2007) presented a version based on
cross-section data only.

1.2 Literature review

1.2.1 Models on port choice

Choice between transport options receives a great amount of attention with respect to
surface transport and subsequent research has lead to extensive model development and
related practical applications and commercial software. Research in the maritime field
started much later.

The early work on port choice started with modal split studies such as by Winston (1981),
where seaports were part of a logistic chain. Malchow and Kanafani (2001) tested the
factors determining port choice in the US by applying a multi-nomial logit model for US
exports using discrete disaggregate data. The availability of combined trade and vessel

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


2 Veldman and Rachman.doc
data in the US offers a rich source of information making it possible to combine detail on
cargo, port of origin and destination, ship type and size and shipping route.

Malchow and Kanafani (2004) worked it out further and show how the market share
predicted for a port varies with commodity-type and carrier.

Blonigen and Wilson (2006a) used the same rich source of information to test port choice
for US imports using aggregate data. The level of aggregation was very low resulting in a
multitude of information on cargo type, trade partner and port in the US and abroad.
Given these enormous of amount of data, nearly 100,000 observations, the models tested
could include port efficiency data based on Blonigen and Wilson (2006b), which
measured port efficiency for a great number of US and foreign ports.

Tiwari e.a. (2003) used a discrete choice model where shippers choose among
combinations of shipping line and port and make decisions based on shipper and port
characteristics. The situation concerns the Chinese foreign trade.

Veldman and Bckmann (2003) tested port choice models for the continental and
overseas hinterland of West European container ports using aggregate container flows,
where type of cargo, type of container (loaded or empty) and direction of trade were
disregarded. Port access costs, hinterland transport costs by mode and proxy variables for
quality of service proved to be significant. Veldman et al. (2005) tested similar models for
a larger continental and overseas hinterland and included a variable expressing container
hub port draft restrictions explicitly, which proved to be significant.

Other studies with respect to port choice, which are not dealt with further, concern Chien-
Chang Chou, (2006), who studied port choice applying the Stackerlberg optimization
method. Meifeng Luo and Grigalunas (2005) applied a simulation model for competition
between US coastal container ports. Guy and Urli (2006) studied port selection on the
Montreal-New York port range and applied multi-criteria analysis.

With respect to modal split studies involving maritime transport and using Logit Models a
study by Garcia-Menendez et. al. (2004) may be mentioned, who tested logit models for
exports from Valencia. Ortuzar and Gonzalez (2002) studied inter-island passenger
transport options for passenger transport comparing the market shares of air transport,
high speed and normal ferry boats.

1.2.2 Models on transhipment port choice

The situation with respect to the choice of transhipment port, the subject of this paper,
needs some further explanation. Liner shipping operators design their shipping networks
with main lines, feeder lines and transhipment ports in order to meet their future demand.
These networks include transhipment services for both hub-spoke systems and for relay-
services. The users of the competing services punish bad alternatives by not choosing
them. This occurs according to an iterative process as schematised in Figure 1. It can be
stated therefore, that finally the users of the services decide on network design, given the
generalised costs derived from the geographical setting and shipping technology.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


3 Veldman and Rachman.doc
Supply of shipping services

Producer choice User choice

Demand for shipping services

Figure 1: Interactions between users and producers of container liner services

At a high abstract level Jansson and Shneerson (1987) developed a mathematical model
of the economics of hub-spoke systems versus multi-porting systems. They demonstrated
the trade-offs between both. At a detailed level Aversa e.a.. (2005) used mixed integer
programming model for the selection of a transhipment port in the East Coast of South
America among a set of 11 ports servicing the regional demand. Baird (2005) analysed
the detailed costs and operational advantages of a new transhipment port on Orkney vis a
vis the existing transhipment ports in the North Sea area. A potential market share was
assessed on the basis of comparative costs. None of these studies, however, appears to be
directly suitable for deriving a demand function, so that other solutions have to be found.

Lirn e.a. (2004) analysed transhipment port selection by ship and terminal operators
applying the Analytic Hierarchy Process and did in-depth interviews with the major
global ocean container operators and transhipment service providers. They analysed and
ranked the importance of 47 relevant service attributes as recorded from a literature
review and measured the performance criteria ranking and revealed some marked
differences between the perceptions of ship and terminal operators.

Veldman and Bckmann (2003) tested port choice models also for the overseas hinterland
of West European container ports using aggregate data on container flows in a similar
way as for the continental hinterland. Port access costs, feeder line costs and a port market
share variable as an indicator of quality of service aspects appeared to be statistically
significant. Veldman et al. (2005) tested the same models including a maritime access
resistance variable that proved to be statistically significant. Veldman and Vroomen
(2007) tested a transhipment model for the Mediterranean. The model findings are
elaborated here by using a combination of cross-section and time-series data.

1.2 Container shipping lines in the Mediterranean

1.2.1 The importance of transhipment

Container throughput of the ports in the Mediterranean nearly tripled from 13.5 to 35.7
million TEU for the period 1995 2004. In the same period the throughput of the ports in
Northwest Europe doubled from 20.6 million to 41.5 million. In terms of overall port
throughput the Mediterranean ports appear nearly to have reached the same level as the
Northwest European ports. In the same period the share of transhipped containers has

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


4 Veldman and Rachman.doc
increased strongly, the strongest for the Mediterranean from 23.6% to 36.6%. See table
below.

Table 1: Container throughput volumes Northwest Europe and Mediterranean (million TEU)
Year 1995 2004
Total container throughput
- Northwest Europe 20.6 41.5
- Mediterranean 13.5 35.7
Share transhipped containers
- Northwest Europe 18.7% 22.8%
- Mediterranean 23.6% 36.6%
Source: Derived from Ocean Shipping Consultants Ltd, (2006)

The ongoing increase in ship size is fuelling the importance of transhipment. Around
1990 the 10 largest containerships in operation had a capacity of 4,500 TEU and it took
about 15 years to double the size to 9000 TEU. It may take another 15 years to double the
size of the largest ships to 18,000 TEU, the so-called Malacca-Max (Wijnolst e.a.1999) in
about 2020. This increase in ship size demonstrates a long-term need for transhipment
port planning.

In 2004 the Mediterranean shows 14 ports with transhipment volumes in excess of 50,000
TEU and 9 ports with volumes in excess of 500,000 TEU. It can be stated that in the
Mediterranean transhipment is spread over more ports than in Northwest Europe. Market
studies such as OSC (2006) expect the importance of transhipment to increase further.
This growing importance of transhipment within a trade which is also growing strongly
will make the transhipment port capacity planning an issue of constant attention.

1.2.2 Linkage of Mediterranean port to world trade routes


The major container trade routes of the world are the routes connecting the industrialized
regions of Europe, North America and the Far East and concern the Europe Far East,
the Transatlantic and the Transpacific trade routes. These routes are also referred to as the
world major East - West trade routes. Other secondary trade routes along the East West
axis concern the trades generated by the Middle East and by South Asia. These secondary
routes are competing strongly with the shipping capacity offered on the major East West
trades routes, as the ships on the latter routes pass these areas picking up and dropping
containers on their way.

For the trade between the Mediterranean and the Far East shipping capacity is offered by
routes connecting both areas directly and also indirectly by ships sailing between
Northwest Europe and the Far East. Capacity is offered also by Pendulum services and
Round-The-World (RTW) services, which are passing the Mediterranean, thereby
connecting it with the Far East and North America.

Important for the assessment of the transhipment potential of ports in the Mediterranean is
therefore the structure of:
The shipping routes having the Mediterranean ports as beginning or end, i.e. end-to-
end routes from the point of view of the Mediterranean; and
The shipping routes passing the Mediterranean with ships on their way between for
instance Northwest Europe and the Far East and said Pendulum and RTW services.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


5 Veldman and Rachman.doc
An analysis of shipping capacities employed in 2005 shows that 76 shipping routes
connect the Mediterranean with North America, the Far East and Northwest Europe with
a total annualised capacity of 17.3 million TEU1. The two most important categories of
routes concern the 20 services connecting the Mediterranean directly with the Far East
and the 29 services connecting Northwest Europe with the Far East passing the ports of
the Mediterranean. The latter services offer a connection with both Northwest Europe and
the Far East. Summary data on these routes are given in Table 2.

In 2005 the largest container ships operational on the Northwest Europe - Far East trade
route at were about 8000 to 9000 TEU. Larger ones are in the process of being employed.
The largest ship of the world, commissioned recently, is the Emma Maersk of 11,000
TEU. As this trade route has the longest distance, it is the most attractive to introduce the
largest ships first. The average size of ships is 5,655 TEU and shows a large variation in
size from 2,600 TEU (deployed by the consortium PIL/Wan Hai) to about 8,200 TEU (by
the D-Loop of the Grand Alliance).

Table 2: Major East West shipping routes passing the Mediterranean mid 2005
No of Route Average
b a
Trade route Type routes capacity ship size
000 TEU TEU

1 Mediterranean Far East ETE 20 3,911 3,761


2 NW Europe - Far East ETE/passing 29 8,528 5,655
3 Mediterranean -
North Am. East Coast ETE 15 1,850 2,371
4 Mediterranean -
North Am. West Coast PDM/passing 2 582 5,596
5 NW Europe -
North Am. West Coast PDM/passing 6 1,659 5,318
6 North America EC -
Far East PDM/passing 1 233 4,481
7 RTW Services RTW/passing 3 487 4,329
All trade routes 76 17,250
Source: Derived from Drewry Shipping Consultant (2006)
a)
annualised capacity one way
b)
ETE: End-to-End service, PDM: Pendulum service, RTW: Round-The-World service

The average size of containerships employed on the Mediterranean Far East trade route
is with 3,911 TEU, smaller than with the Northwest Europe Far East trade route. This
reflects the smaller scale of operations, where container demand for Northwest Europe is
more than three times the volume of the Mediterranean, whilst these lower volumes are
divided over a greater number of ports.

The North American East Coast is connected with the Mediterranean by 15 services with
an annualised capacity of 1.85 million TEU. The average size of ships on this route is
2,371 TEU, less than the one connecting with the Far East. The smaller size reflects the
lower volume of trade.

1
The annualised capacity of a shipping line is calculated as the product of the average ship (in TEU) and the number of
trips per year of 52

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


6 Veldman and Rachman.doc
In half of the cases the liner services passing the Mediterranean do not call at a port in the
Mediterranean. If so, they generally call at one port only. On some 14 routes, out of a
total of 30 routes, container ships call at ports in the Mediterranean either on their way
eastbound, westbound or on both. Generally one call and sometimes two calls are made at
ports located near to the main navigation course, so that the mainline ships have to make
small deviations. Of these ports Algeciras, Malta, Gioia Tauro and Port Said are the
prominent ones.

The 9 Pendulum and 3 RTW services offer a total capacity of 2.96 million TEU and the
size of ships offered is comparable with those of the liner services of the Northwest
Europe Far East trade route and bigger than those with the Mediterranean Far east
routes.

The main differences between the linkage of the Mediterranean and Northwest Europe to
the world container network and impacting the model specifications tested hereafter, are
the greater importance of transhipment in the Mediterranean, the greater importance of
passing shipping lines and the greater number of transhipment ports.

2. THE TRANSHIPMENT PORT CHOICE MODEL

2.1 Model specification


The probability that shippers or receivers in region (r) select transhipment port (p) can be
expressed as:
U rp

Ppr = e
p= P
, (p=1, , P) (1)
e
U rp

p=1

where U pr is the utility attached to transhipment port (p) by shippers and receivers in
region (r) and p the index of the transhipment port in a total of P ports.

For region (r) the utility function is:


U pr = p D p + 1 CF pr + 2 CM p + 3 M p + 4CAPp , (2)
where Dp is a dummy variable indicating a preference, if any, for transhipment port (p);
CFpr is the feeder line transport cost between transhipment port and feeder region (p,r);
CMp is the mainline access cost; Mp represents quality of service aspects of port p related
to the frequency of services offered and CAPp represents the attraction of a port given its
volume of captive cargoes. The latter are part of Mohring-effects 2 and expressed as a
function of the level of port throughput. The Greek symbols m,p, 1, 2, 3 and 4 are
the coefficients of the utility function.

2
Users of transport facilities impact the situation of other users. In the negative case, e.g. when the activity of one user
causes extra costs for others, we talk of congestion costs. In the positive case, when users activities improve the
welfare situation of other users we talk about the Mohring effect. The project UNIfication of accounts and marginal
costs for Transport Efficiency (UNITE), is a project under the Fifth Framework package by the European Commission,
where a great amount of attention is paid to the quantification of Mohring effects through case studies concerning
passenger and freight transport. Logit models were used in particular with respect to freight transport. See UNITE
(2003)

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


7 Veldman and Rachman.doc
By taking for each region (r) the ratio of the market share of transhipment port (p) and of
an arbitrarily chosen basic port (p*), it follows from (1):
Ppr U rp
= e = U p U p*
r r

P*
r r
U * e (3)
p e p

Combination of equations (2) and (3) and taking of logarithms leads to:
Ppr
Ln r = p D p P D p * + 1 (CF pr CF p * r )+ 2 (CM p CM p * )
P* (4)
p
+ 3 (M p M p * ) + 4 (CAP p CAP p * )
Equation (4) is the basis for the empirical analysis hereafter. The equation will be tested
with annual data for the period 1995 2004. For simplicity reasons the index for time is
omitted

2.2 Variables used


The dependent variable is the observed share of containers generated by a hinterland
region, which are routed via a transhipment port. This concerns an aggregation of
containers by cargo type, direction (incoming and outgoing) and by type (full and empty).

The AHP survey by Lirn e.a. (2004) found that handling costs of containers, proximity
to main navigation routes, proximity to import/export areas, basic infrastructure
condition (water access, e.g. depth) and existing feeder network were the service
attributes with the highest importance for transhipment port selection.

Veldman and Bckmann (2003) and Veldman et al. (2005) adopted a maritime access
cost variable expressing port access cost differences based on the sum of port charges on
the ship, terminal handling charges per container, main line deviation costs of the ship
and interest cost of cargo aboard. This covers the first two attributes mentioned in the
previous section. The variable did not prove to be significant for the West European
transhipment ports. The variable is significant for the situation in the Mediterranean, as
will appear later on. The variable cost is comparable with the attribute proximity to
import/export areas and appeared to be significant. The variable maritime access
resistance expressed the expected average waiting times for large container ships and
appeared to be significant. This variable corresponds with basic infrastructure condition
and than in particular with respect to draft. Proxy variables for quality of service aspects
were included intended to capture Mohring effects. For the latter it was expressed as 1
minus the inverse of port throughput, i.e. the sum of captive and transhipment containers.
This variable appeared to be significant. The variable comes close to existing feeder
network. Another way to capture Mohring effects was to take the share of total container
throughput of a port as the share of the total of all transhipment ports. Also this variable
proved to be statistically significant.

In this study the following variables are used:

Mainline deviation costs


For this study we dispose of annual data from 1995-2004, but not of detailed data on port
charges and terminal handling charges such as those included in the maritime access
cost mentioned in the previous section. At this stage we neither dispose of port

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


8 Veldman and Rachman.doc
limitations related to ship dimensions such as vessel draft. The geography of the
Mediterranean, however, is quite different from Northwest Europe leading to a greater
variation of mainline deviation distance and we therefore adopt a main deviation cost
variable.

The mainline deviation distance is measured as the extra distance needed to call at a
transhipment port compared to the distance of the shortest navigation course between the
entrance of the Suez Canal and Gibraltar. This distance precisely corresponds with the
distance sailed by the passing shipping lines, but not with the end-to-end lines turning
their itinerary in the Mediterranean.

Feeder costs
The feeder costs concern the shipping costs between a transhipment port and the gravity
point of a feeder area.

Proxies for quality of service and related aspects


Quality of service aspects are manifold and are often listed in user surveys. Gardner A
number of these can be included explicitly such as transit time or included in other
variables such as port efficiency, which is part of port access costs. Some quality of
service aspects are more difficult to capture such as those related to the commercial
performance of ports and shipping lines. These aspects are expected to work neutral for
all transhipment ports. Moreover, by adopting time-series data their effects are likely to
be levelled out, as an outlying performance by player A will be copied after some time by
players B and C.

The attraction of a port for a transhipment function, because of its size, however, needs
attention. The greater a port is, the more it has shipping lines calling, a higher frequency
of service of the main lines and feeder lines calling. With surface and air transport these
aspects are generally referred to as Mohring effects (UNITE, 2003). Parallel to this
greater ports also have a greater supply and range of supporting services.

After reaching some level Mohring effects become exhausted. To capture this they are
often expressed by functions such as the inverse value of the number of ship calls or, if
these are not available, by the inverse of port throughput. Taking one minus the inverse
value, results in a variable developing as throughput levels increase, asymptotically to 1.
This variable proved to be significant in Veldman et al. (2005). Another way to express
Mohring effects is to take the share of the transhipment port in the total of all
transhipment ports. This variable also appeared significant in the previous studies.

An aspect running parallel to Mohring effects concerns the level of captive cargoes
generated by the transhipment port itself. If an operator has to choose between two
transhipment ports which only differ in the amount of captive cargo, it seems logic that he
selects the one with most captive cargo as this would result in a lower volume of
transhipment needed. Veldman and Vroomen (2007) show that a variable expressing the
amount of captive cargo appears to be significant, performing best when using the
logarithmic form.

Finally, it should be noted that two issues arise when trying to capture these aspects by
more than one proxy variable:

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


9 Veldman and Rachman.doc
The proxies are most likely correlated
The use of time-series information makes that the absolute levels of variables vary
over the years, so that relative expressions have to be used.

For the statistical tests three variables related to quality of service aspects are used:
Throughput share (p, t) or in short THS: this variable is intended to capture quality
of service aspects having to do with the level of port activities thereby capturing
Mohring effects.
Captive share (p, t) or in short CPS: this variable is intended to capture cost
efficiency considerations related to avoided transhipment.
Transhipment ratio (p, t) TRATIO: this variable is intended to distinguish between
pure transhipment ports such as Algeciras, Gioia Tauro and Marsaxlokk at one hand
and hub-ports attracting substantial transhipment volumes such as Barcelona, Genova
and La Spezia.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


10 Veldman and Rachman.doc
Table 3 Description of variables used
Name of variable Unit Description
Share routing P(p, r, t) Ratio Dependant variable: ratio of the number of
containers measured in TEU carried via
transhipment port (p) and the number feeder
containers generated by feeder region (r) in year
t
Deviation cost (p) per Mainline deviation costs based on deviation
(DEVC) TEU from main navigation course for transhipment
port (p)
Feeder cost (p, r) per Costs of feeder transport stack-to-stack between
(FEEDC) TEU feeder region (r ) and transhipment port (p)
Throughput share (p, t) Ratio The share of throughput of transshipment port p
(THS) in the total throughput of all transshipment ports
in year t
Captive share (p, t) Ratio The share of captive throughput of
(CAPS) transshipment port p in the total captive
throughput of all transshipment ports in year t
Transhipment ratio (p, t) Ratio The share of transshipped containers on the
(TRATIO) total throughput of transshipment port p in year
t
Port preferences (p) Binary 1: if a port preference exists for port p; 0
otherwise

3. MODEL TESTS

Testing the basic variables


As a first step we test the impact of all five non-binary variables on port choice. FEEDC,
THS and TRATIO appear to be significant, while DEVC and CAPS are not. FEEDC
shows little correlation with the other independent variables, but the other variables to
some extent do. CAPS shows the highest correlation with the other variables and is
therefore skipped. As a result the t-values of all variables improve, except for DEVC.
DEVC is strongly correlated with TRATIO, which is not surprising. The typical
transhipment ports all have a high TRATIO and are also closely located along the main
navigation course leading to low values of DEVC. From the point of view of theory there
is a strong preference to use DEVC instead of TRATIO. Skipping TRATIO leads to an
improvement of DEVC in terms of its t-ratio. This shows that mainline deviation costs,
feeder line costs and the ports share in the total regional port throughput, as a proxy
variable of quality of service aspects, are the three key variables. The results are given in
table 4.

Adoption of port dummies


In a second step dummy variables are adopted to measure preferences for certain ports.
Dummy variables cannot be used for all ports, as this would lead to the inclusion of
variables being linear combinations of other ones. By including all dummy variables
some are skipped automatically while three appear not to be significant. After that all
dummy variables are adopted again exclusive of the three insignificant ones. As a result
only four of the maximum fourteen dummy variables appear to be significant, i.e. with t-
values in excess of 2. See table 4.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


11 Veldman and Rachman.doc
A negative value of the coefficient of a dummy indicates that the ports market share is
over-estimated by the three explanatory variables and has corrected for that. In case of a
positive value the market share is under-estimated needing an upward adaptation. If the
coefficient is not significant no adaptation is needed.

The ports can be grouped in typical transhipment ports, where transhipment is the main
element, and hub-ports that have considerable volumes of transhipment. It appears that
the predicted market shares of the ports of Taranto, Limassol and Livorno are too high
and that of Gioia Tauro too low. No specific reasons can be given why there are so.

Table 4: Models tested for routing choice with respect to overseas hinterland
Variable Coefficient t-statistic P value
Basic model
DEVC -0.077 -4.86 0.000
FEEDC -0.030 -9.55 0.000
THS 53.230 15.25 0.000
R-squared 0.254 Observations 995
Adjusted R-squared 0.251

Model with combined feeder and deviation costs


COST -0.032 -10.01 0.000
THS 51.438 14.92 0.000
R-squared 0.247 Observations 995
Adjusted R-squared 0.246

Model with port dummies


DEVC 0.051 1.11 0.266
FEEDC -0.027 -8.43 0.000
THS 29.187 5.107 0.000
Algeciras 0.681 0.089 0.371
Gioia Tauro 3.344 4.716 0.000
Piraeus Excluded
Port Said -0.044 -0.053 0.958
Damietta -0.423 -0.527 0.598
Genoa -1.007 -1.385 0.166
La Spezia Excluded
Livorno -4.598 -6.176 0.000
Barcelona Excluded
Valencia Excluded
Taranto -2.679 -2.837 0.005
Cagliari Excluded
Constanza -5.506 -1.800 0.072
Limassol -3.022 -4.456 0.000
R-squared 0.305 Observations 995
Adjusted R-squared 0.297

Coefficient values over the years


Statistical test for each year separately show that the values of the coefficients of THS and
FEEDC are rather stable over the years, while being significantly different from zero. The

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


12 Veldman and Rachman.doc
coefficient values for THS vary around the value of 0.030 for the test for all years and
those for THS close to the corresponding value of 55.2 for all years. The coefficient for
DEVC shows for each year the right sign, but low t-values indicating a weak statistical
relationship. Compare results in table 4 and 5.

Table 5: Models tested per year

Year DEVC FEEDC THS R-square Observ.


coefficient t-value coefficient t-value coefficient t-value
1995 -0.052 -0.55 -0.025 -2.34 39.64 2.59 0.15 85
1996 -0.130 -1.27 -0.034 -3.16 51.81 2.25 0.15 85
1997 -0.140 -2.16 -0.031 -3.14 62.18 4.40 0.25 93
1998 -0.115 -1.88 -0.028 -2.84 67.59 5.05 0.29 93
1999 -0.107 -1.76 -0.028 -2.84 66.91 5.33 0.31 93
2000 -0.128 -2.00 -0.027 -2.64 60.91 5.15 0.29 93
2001 -0.120 -1.98 -0.031 -2.30 57.65 5.54 0.29 102
2002 -0.120 -2.04 -0.031 -3.10 55.64 5.78 0.29 111
2004 -0.113 -1.85 -0.033 -3.18 57.84 5.37 0.26 111
2004 -0.028 -1.03 -0.035 -3.63 55.97 5.19 0.29 120

4. Comparison with other studies

The test results on transhipment with the Northsea ports show that feeder costs are also
statistically significant with somewhat smaller coefficient value ranging from -0.014 to -
0.020. See Veldman e.a. (2005). The variable comparable with deviation costs is
performing poorly and skipped for final runs.

The competitive position of a transhipment port changes as a result of changes in the


relative costs of using the port. Such a change may result from changes in port
productivity, port dues, terminal charges, economies of scale effects with respect to
mainline and feeder lines and so on. The result of model simulations shows that, given the
situation as measured with transhipment levels of 2004, an increase in the costs of
transhipment with 10 per TEU for Marsaxlokk will lead to a decrease in transhipment
demand of 24%. This is based on the coefficient estimated for the model with the
combined feeder and deviation costs as presented in table 4.

Table 6 The impact of costs on transhipment demand for Marsaxlokk

Additional cost in Additional demand


in %
-50 217%
-40 161%
-30 111%
-20 67%
-10 30%
0 0%
10 -24%
20 -43%
30 -57%

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


13 Veldman and Rachman.doc
40 -68%
50 -77%

The elasticity of transhipment demand with respect to transhipment costs can be derived
and shows a value of -2.2 compared. For the Northsea ports the corresponding value for
the Port of Rotterdam3 is practically the same with -2.0.

As mentioned above Lirn e.a. (2004) found 5 service attributes to be considered as of the
highest importance for port selection. The attribute proximity to main navigation routes
corresponds with DEVC, which is found to be of little impact. Proximity to
import/export areas corresponds with FEEDC and is clearly significant. Existing feeder
network is part of a proxy for Mohring aspects as included in THS, which is clearly
significant. Handling costs of containers is not included here. Veldman e.a. 2005 also
tested a variable with maritime access costs for port choice in the Northsea and found that
the impact was not significant4. The variable basic infrastructure condition (water access,
e.g. depth) is not included. In the North Sea port choice study this aspect is dealt with
partly under the maritime access costs variable, which did not appear to be significant
and partly under the maritime access resistance 5 variable, which appeared to be
significant.

In the North Sea studies port choice was subject of analysis not only for the overseas
hinterland, but also for the continental hinterland and more or less similar types of models
were tested. The feeder cost variable included here is comparable with the variable of
hinterland transport costs between hub-port and hinterland regions. The resulting
elasticity of port demand with respect to hinterland transport costs was -1.3, which is
lower than the -2.0 for the overseas hinterland. Hereafter some comparisons are made
with the outcomes of port choice studies with respect for the continental hinterland.

Malchow and Kalafani (2004) studied port choice for US containerised exports. They
used amongst other the overseas distance and the inland distance as explanatory variables,
which appeared to be highly significant. For the inland distance 6 the choice elasticity
ranged from -1.03 to -2.17 for the eight individual US ports studied. These values seem to
correspond rather well with those of demand for North Sea ports with respect to the
continental hinterland. Blonigen and Wilson (2006a) analysed port selection for US
exports and also used inland transport costs as explanatory variable. They found a value
for the elasticity of -0.5, which is lower than the results of our analysis.

Tiwari e.a (2003) analysed port choice for Chinese importers and exporters. The distance
to the port appears to be an important explanatory variable for port choice. The elasticity
of demand with respect to distance appears to be considerably higher than for the
previous studies mentioned and ranges -2.9 to -14.0 for the combinations of ports and
shipping companies concerned.

3
Derived from demand function as presented in Figure 5
4
A variable named maritime access cost including terminal handling charges, port charges and deviation cost, which are small in case of
the North Sea hub-ports, was tested. The variable appeared not to be significant.
5
The main objective of the study was to estimate the impact of the depth of the hub-ports access channel on market share
6
Inland transport distance is closely related to inland transport costs so that resulting values of the elasticity can be compared with the
results of this paper.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


14 Veldman and Rachman.doc
Some final remarks needs to be made with respect to the variables included. Quality of
service aspects such as Mohring aspects are difficult to capture. In this paper the market
share of a container ports throughput within the total feeder region is used as proxy and
includes the sum of captive and transhipment cargoes. Such a proxy was not adopted in
the Logit Models tested for the US and Chinese trade studies. Blonigen and Wilson
(2006a) adopted a port efficiency variable reflecting costs spend in port and derived from
the same data set by Blonigen and Wilson (2006b). Malchow and Kalafani (2001) used a
variable named sailing frequency which appeared to be statistically significant, but had
the wrong sign. Malchow and Kalafani (2004) did not adopt quality of service proxies.

Tiwari e.a (2003) used port related variables such as total TEU throughput, number of
berths, number of ship calls and number of cranes. These variables are positively related
to a ports size thereby reflecting Mohring aspects. The test shows a positive coefficient
value for the number of berths and negative ones for TEU throughput and number of
cranes. The correlation existing between the variables most probably leads to the
unexpected results. A port advantageous because of its location attracts traffic, resulting
in more throughput and ship calls, and thereby invests in berths and cranes. Including
such variables as explanatory variables disturbs the causal relationships. In this paper we
analysed the transhipment market and adopted the sum of captive and transhipment
cargoes, thereby minimising causal disturbances, while still including Mohring type
effects.

5. CONCLUSIONS

The market share of the transhipment ports serving the overseas hinterland of the
Mediterranean can be explained by variables such as feeder costs, to a lesser extent by
mainline deviation costs and further by quality of service effects as measured by the
transhipment ports share in the total sum of transhipment and captive cargo of these ports.

The value of the demand or choice elasticity that can be derived from the estimated
coefficients, is -2.2 and compares rather well with the value of -2.0 estimated by Veldman
and Buckmann, 2003 for the ports in the North Sea transhipment market.

Port choice for the overseas hinterland has some marked differences compared to port
choice for the continental hinterland. These differences have to do with the roles of
decision-makers in the choice process. In case of merchant haulage shippers and receivers
are directly involved in the choice of seaport. In the other cases port choice is part of the
liner route design process that is conducted by the carriers. Shippers and receivers are not
directly involved in this, but have a crucial influence as they punish bad network designs,
whether it concerns the choice of the final port of call or the choice of transhipment port
en route.

Feeder transport costs in case of the transhipment port choice and inland transport costs in
case of hub-port choice correspond with each other. Veldman and Buckmann, 2003
measured that the choice elasticity for the port of Rotterdam was -2.0 and -1.3
respectively. Malchow and Kalafani (2004) used the distance between inland point of
origin and port of transfer as a proxy for inland transport costs. The resulting choice

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


15 Veldman and Rachman.doc
elasticity for each of the eight individual US ports studied ranges from -1.03 to -2.17.
Blonigen and Wilson (2006a) analysed port selection for US imports and also used inland
transport costs as explanatory variable. They found a value for the elasticity of -0.5,
which appears to be low.

This paper shows that the use of Logit Models with respect to transhipment port choice
leads to useful findings for port planning. This research in combination with earlier
research by one of the authors for transhipment port choice in Northwest Europe is a step
forward in the field of transhipment port choice.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


16 Veldman and Rachman.doc
REFERENCES
Aversa, A., Botter, R.C., Haralambides, H.E., Yoshizaki, H.T.Y.: (2005), Mixed Integer
Programming Model on the location of a Hub Port in the East Coast of South America, Maritime
Economics & Logistics, March 2005, 1-18

Baird, A., (2005): Optimising the container transhipment hub location in northern Europe,
Journal of Transport Geography, Article in Press, www.sciencedirect.com

Blonigen, B. and Wilson, B.(2006a): International Trade, Transportation Networks and Port
Choice, Internet, May, 2006, 1-22

Blonigen, B. and Wilson, B.(2006b): New measures on port efficiency using international trade
data, NBER Working Paper Series Number 12052, February, 2006, 1-30

Bruno and Guy MEL 2006

Chien-Chang Chou, (2006): A comparative study of models for port choice, Proceedings of the
Eastern Asia Society of Transportation Studies, 2005, 608-616

Drewry (2006): Annual Container Market Review and Forecast 2005/06, Drewry Shipping
Consultants Ltd, 2006

Garcia-Menendez, L., et. al. (2004): Determinants of Mode Choice between Road and Shipping
for Freight Transport, Evidence for Four Spanish Exporting Sectors, Journal of Transport
Economics and Policy, September 2004, 447-466

Guy, E. and Urli, B. (2006): Port Selection and Multi-criteria Analysis: An Application to the
Montreal-New York Alternative , Maritime Economics & Logistics, June 2006, pp. 169-186

Lirn T.C., Thanopoulou H.A., Beynon, Beresford A.K.C., (2004): An application of AHP on
Transhipment Port Selection: A Global Perspective, Maritime Economics & Logistics, 2004, 6,
(70-91)

Jansson, J.O. and Shneerson, D, (1987): Liner shipping economics, Section 5.5, Chapman and
hall Ltd, London, 1987

Magala, M., and Sammons, A. (2007): A new Approach to Port Choice Modelling, Conference
Proceedings Challenges and Trends in Shipping: Market Investments and Policies, IAME
Conference 2007, July 4-6, 2007 Athens Greece

Malchow, M. B., and Kanafani, A., (2001): A disaggregate analysis of factors influencing port
selection, Maritime Policy & Management 28:3, 265-277

Malchow, M. B., and Kanafani, A., (2004): A disaggregate analysis of port selection,
Transportation Research Part E 40 (2004) 317-337

Meifeng Luo and Grigalunas, (2002): T. A spatial-economic multimodal transportation


simulation model for US coastal container ports, IAME Paper, 2002

Ortuzar J. and Gonzalez R, (2002): Inter-Island Travel Demand Response with Discrete Choice
Models, Functional Form, Forecasts and Elasticities, Journal of Transport Economics and Policy,
January 2002, 115-138

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


17 Veldman and Rachman.doc
OSC (2006): The European & Mediterranean container markets to 2015, Ocean Shipping
Consultants Ltd, 2006

Panayides P. M. (2006): Maritime logistics and global supply chains: towards a research agenda,
Maritime Economics & Logistics, 2006, 8(1) 3-18

Tiwari, P. Itoh, H. and Doi, M. (2003), Shippers port and carriers selection behaviour in China: a
discrete choice analysis, Maritime Economics and Logistics, 5: 23-39

UNITE (2003), UNIfication of accounts and marginal costs for Transport Efficiency, Project
funded by the EC under the Fifth Framework Transport RTD

Veldman S. and Bckmann E. (2003): A Model on Container Port Competition, An application


for the West European Container Hub-ports, Maritime Economics & Logistics, Vol 5, No 2,
March 2003

Veldman S., Bckmann E. and Saitua R. (2005): River Depth and Container Port Market
Shares: The Impact of Deepening the Scheldt River on the West European Container Hub-Port
Market Shares, Maritime Economics & Logistics, Vol 7, No 4, December 2005

Veldman, S., and Vroomen B., (2007): A model of container port competition: an
application for the transhipment market of the Mediterranean, IAME Conference Athens
2007

Winston, C. 1981, A multi-nomial probit prediction of the demand for domestic ocean
container service, Journal of Transport Economy and Policy, 15(3), 243-252

Wijnolst, N., Scholtens, M.and Waals F. (1999) , Malacca-Max, The ultimate container
carrier, Delft University Press

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


18 Veldman and Rachman.doc
APPENDIX: ON CONTAINER FLOWS AND PORT CHOICE ATTRIBUTES
The basis of the statistical analysis of the transhipment market shares is the present
pattern of transhipment flows in the Mediterranean, the container flows between the 15
transhipment ports and the 9 feeder areas in the Mediterranean for the period 1995-2004.
The incoming and outgoing container flows for the hub-ports in the Mediterranean are
derived from publications by OSC (2006), Containerisation International and some port
specific sources.

Data on the origin and destination of feeder flows are not published and need to therefore
constructed on the basis of the sailing patterns of feeder lines. The starting point of such a
pattern is Drewry (2000) 7 , which is updated to 2004 and serves as a basis for the
assessment of the feeder flows in that year. The feeder flows generated by areas outside
the Mediterranean are excluded. This means that for some transhipment ports the volumes
have to be reduced8.

Table A.1 Estimated container feeder flows in the Mediterranean in 2004 (000 TEU)

Feeder area
1 2 3 4 5 6 7 8 9
NW NE Black Near
Port Spain Africa Africa France Italy Adriatic Aegean Sea East Total
1 Marsaxlokk 154 98 22 8 131 34 141 0 70 658
2 Algeciras 463 0 61 74 225 0 0 0 0 823
3 Gioia Tauro 229 55 61 12 254 120 475 93 258 1558
4 Piraeus 155 0 15 8 112 0 0 36 68 395
5 Port Said 0 0 0 0 70 0 89 0 195 354
6 Damietta 0 0 0 0 58 0 73 0 160 290
7 Genoa 22 0 11 13 0 0 30 3 21 101
8 La Spezia 14 0 7 8 0 0 18 2 13 62
9 Livorno 0 0 0 30 0 0 0 0 0 30
10 Barcelona 0 0 29 41 67 0 83 18 49 287
11 Valencia 0 0 36 46 0 0 131 18 73 305
12 Taranto 0 0 0 0 187 145 0 0 0 332
13 Cagliari 0 114 0 26 59 46 0 0 0 244
14 Constanza 0 0 0 0 0 0 0 61 0 61
15 Limassol 0 0 0 0 0 0 0 0 21 21

Total 1036 267 243 267 1163 344 1041 231 928 5521
For comparison with port throughput figures, note that transhipment flows are handled twice in a port, i.e. by mainline
carrier and feeder ship. Due to rounding totals may differ from additions

The totals of incoming and outgoing container flows of the transhipment ports in 2004 are
distributed over the 9 feeder areas. Subsequently, per feeder area feeder flows are area
summed over all transhipment ports. The resulting totals are compared with the captive,

7
Mediterranean Container Ports and Shipping, January 2000, Drewry Shipping Consultants
8
This applies for instance for Algeciras, which transhipment volumes are only partly related to feedering of ports in the
Mediterranean.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


19 Veldman and Rachman.doc
i.e. non-transhipment totals of the 9 feeder areas, which consist of containers carried
directly through multi-porting practices and indirectly through mainline-feeder line
systems. Per feeder region the composition of both is not known to us. It is assumed
therefore that the share of feeder shipping in the total is equal for all feeder regions. As a
final step the feeder flows between the transhipment ports and feeder ports are assessed
with the RAS method, using the initial structure as the starting situation. This is applied to
each year separately. The results for 2004 are given in Table A.1.

The competitive position of a transhipment port is compared to 14 competing ports varies


per hinterland area. The port of Marsaxlokk is taken as a basis. Per feeder area Maltas
market share is set against the share of the other transhipment-ports and a mathematical
relationship according to equation (4) is tested.

The mainline deviation costs per TEU can be assessed on the basis of deviation distance,
mainline sailing speed, shipping costs per hour sailing and interest costs of cargo aboard.
The resulting costs per TEU are given in Table A.2.

Table A.2 Mainline deviation costs and captive containers per transhipment port
N Transhipment port Distance in Deviation costs in Captive cargoes
Naut. Miles Per trip Per TEU 000 TEU

1 Marsaxlokk 15 1,420 1 146


2 Algeciras 10 947 1 146
3 Gioia Tauro 43 4,072 3 144
4 Piraeus 161 15,246 10 751
5 Port Said 0 0 0 161
6 Damietta 0 0 0 76
7 Genoa 353 33,428 22 1,418
8 La Spezia 339 32,102 21 916
9 Livorno 296 28,030 19 579
10 Barcelona 190 17,992 12 1,343
11 Valencia 143 13,542 9 1,536
12 Taranto 260 24,621 16 99
13 Cagliari 56 5,303 4 6
14 Constanza 1.065 100,852 67 265
15 Limassol 183 17,330 12 256

In a similar way feeder line costs are assessed on the basis of roundtrip trip calculations
for a standard feeder ship operating between transhipment ports and feeder area. The level
of resulting feeder costs is made comparable with a sample set of feeder line freight rates
provided by ship operators.

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


20 Veldman and Rachman.doc
Table A.3: Distances between transhipment ports and feeder areas in nautical miles
Feeder area
1 2 3 4 5 6 7 8 9
N Africa N Africa Black Near
Port Spain West East France Italy Adriatic Aegean Sea East

1 Marsaxlokk 750 583 516 645 408 737 517 1.026 1.037
2 Algeciras 389 414 1.506 692 897 1.650 1.483 1.998 2.006
3 Gioia Tauro 768 620 555 584 278 620 481 990 1.009
4 Piraeus 1.237 1.071 362 1.065 759 839 0 548 643
5 Port Said 1.669 1.503 438 1.512 1.206 1.294 593 982 228
6 Damietta 1.629 1.463 398 1.472 1.166 1.254 553 942 228
7 Genoa 510 528 1.047 201 217 1.111 972 1.481 1.500
8 La Spezia 537 533 1.011 228 180 1.075 936 1.445 1.464
9 Livorno 534 523 974 232 144 1.038 900 1.408 1.427
10 Barcelona 164 279 1.188 185 452 1.300 1.156 1.665 1.673
11 Valencia 0 226 1.260 345 585 1.388 1.237 1.746 1.760
12 Taranto 973 825 606 789 483 507 485 999 1.013
13 Cagliari 455 326 850 352 223 947 808 1.317 1.336
14 Constanza 1.746 1.580 883 1.574 1.268 1.348 548 0 1.006
15 Limassol 1.643 1.477 485 1.475 1.169 1.250 525 888 132

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


21 Veldman and Rachman.doc
Figure A.1: Map Feeder areas and hub-ports in the Mediterranean

C:\Documents and Settings\dorsmanj\Desktop\IAME 2008 paper


22 Veldman and Rachman.doc

View publication stats

Вам также может понравиться