You are on page 1of 20

A stochastic frontier model of the efficiency of major container

terminals in Asia: assessing the influence of administrative
and ownership structures
Kevin Cullinane
a,
*
, Dong-Wook Song
a,1
, Richard Gray
b,2
a
Department of Shipping and Transport Logistics, The Hong Kong Polytechnic University, Hung Hom, Kowloon,
Hong Kong
b
Institute of Marine Studies, University of Plymouth, Drake Circus, Plymouth PL4 8AA, UK
Received 17 April 2001; received in revised form 6 September 2001; accepted 12 September 2001
Abstract
This paper applies a ‘port function matrix’ to analyse the administrative and ownership structures of
major container ports in Asia. The relative efficiency of these ports is then assessed using the cross-sectional
and panel data versions of the ‘stochastic frontier model’. The estimated efficiency measures are broadly
similar for the two versions of the model tested. From the results of the analysis, it is concluded that the size
of a port or terminal is closely correlated with its efficiency and that some support exists for the claim that
the transformation of ownership from public to private sector improves economic efficiency. While this
provides some justification for the many programmes in Asian ports which aim to attract private capital
into both existing and new facilities, it is also concluded that the level of market deregulation is an im-
portant intervening variable which may also exert a positive influence. Ó 2002 Elsevier Science Ltd. All
rights reserved.
1. Introduction
Asian container ports suffer from a number of problems including insufficient port and/or
terminal capacity, inefficient management and bureaucratic administration (Cullinane and Song,
*
Corresponding author. Tel.: +852-2766-7410; fax: +852-2334-1765.
E-mail addresses: stlkcull@polyu.edu.hk (K. Cullinane), stldsong@polyu.edu.hk (D.-W. Song), rgray@plymouth.ac.uk
(R. Gray).
1
Tel.: +852-2766-7397; fax: +852-2330-2704.
2
Tel.: +44-1752-232442; fax: +44-1752-232406.
0965-8564/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S0965- 8564( 01) 00035- 0
Transportation Research Part A 36 (2002) 743–762
www.elsevier.com/locate/tra
1998). This may be due in part to the fact that the majority of the region’s ports are controlled
and/or operated by public entities. To deal with these problems, the port authorities of a number
of countries in the region have launched programmes which aim to attract private capital into
both existing and new facilities. As a consequence, these schemes have kick-started a revo-
lutionary new operating milieu where inter- and even intra-port competition is rife. In many
instances, this has engendered the perception that organisational restructuring (including pri-
vatisation) is not only desirable, but necessary.
A key claim made with respect to organisational reforms is that the transformation of own-
ership from public to private sector will improve economic efficiency, as well as general welfare
(Yarrow, 1986; Vickers and Yarrow, 1991). Associated economic theories and existing empirical
studies, however, fail to establish clear-cut evidence supporting this claim (Vickers and Yarrow,
1988; Boardman and Vining, 1989). Indeed, econometric analysis of the relative productive effi-
ciency of the ports sector pre- and post-privatisation seems to suggest that ownership itself does
not seem to be categorically related to efficiency in port operations (Song and Cullinane, 2001). It
may well be the case, as proposed by the United Nations Conference on Trade and Development
(UNCTAD, 1995a), that the apparent absence of a clear-cut theoretical and empirical relationship
may reflect, to some extent, a unique socio-political situation in which these business entities
undertake their economic activities.
Against this conceptual background, this paper analyses the administrative and ownership
structures of the major container ports in Asia by relating them to the ‘port function matrix’
(Baird, 1995, 1997) and assesses their relative efficiency by applying an econometric technique
known as the ‘stochastic frontier model’.
2. Port administration and ownership
Stehli (1978) and Goss (1986) note that, although most of the physical methods used within
ports (e.g. loading and discharging) vary little between ports, there are a number of alternative
forms of port administration and ownership. Few ports can be described as either purely private
or public. Moreover, it is often even difficult to identify the extent of either public or private sector
involvement in a port. This situation does make it necessary, however, to distinguish between the
alternative approaches to port administration and ownership.
2.1. A conceptual framework for the classification of port ownership and administration
UNCTAD (1995b) classifies the list of facilities and services which should be provided by ports
for ships and cargoes: namely, infrastructure, superstructure, equipment, services to ships and
services to cargoes. According to which entity (i.e. private, public or joint) owns and provides
those facilities and services, ports can be divided into two distinct types (Goss, 1990; Heaver,
1995; De Monie, 1996). In the comprehensive port, the public port authority provides all facilities
and services within the port, thus having direct responsibility for the management and operation
of port services and facilities. Independent (private) operators are prohibited from undertaking
any port activity. This kind of port, therefore, can be said to be the ‘totally integrated port’. On
the other hand, in the landlord port, the activities of the port authority are limited simply to
744 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
providing and maintaining the basic infrastructure and essential services (e.g. fire services, security
etc.), while all the other facilities and services such as the superstructure and stevedoring labour
are provided by independent private (or public) companies. This port model can be referred to as
the ‘purely regulatory port’.
An alternative to this traditional framework for analysing port administration and ownership
is proposed by Baird (1995, 1997) who refers to a port function matrix. The starting point for
this conceptual framework is that, regardless of whether a port is in private or public hands,
within the port area there will generally be three essential functions the port must fulfil and
provide:
1. The regulatory function of a port can involve substantial powers being given to the port’s public
or private sector management, the majority of which will be of a statutory nature. In general,
this function may be regarded as the primary role of a port authority (Nagorski, 1972).
2. In expediting a landowner function, ports control significant land areas. Irrespective of whether
the land area of a port is large or small, however, the essential tasks involved would be to man-
age and develop the port estate, to implement policies and strategies for the port’s physical de-
velopment in terms of superstructure and (sometimes) infrastructure, to supervise major civil
engineering works, to co-ordinate port marketing and promotion activities, to provide and
maintain channels, fairways, breakwaters etc., to provide and maintain locks, turning basins,
berths, piers and wharves, and to provide or arrange road and rail access to the port facilities.
3. The operator function is concerned with the physical transfer of goods and passengers between
sea and land. In a comprehensive port, for example, the cargo-handling activity will be con-
trolled by state-owned organisations. Conversely, in a landlord port, private companies will un-
dertake this activity, while a mix of private and public companies may be involved as well.
According to which of these three functions are the responsibility of public or private organisa-
tions, the matrix presented in Table 1 makes it possible to ascertain the extent of the influence
exerted by public and private sectors within any given port. The matrix also suggests the four
main patterns (as defined by choice of port administration, ownership, management and opera-
tion) into which a government is able to organise its port industry.
According to Table 1, port administration and ownership models are divided into four types of
port administration: the PUBLIC port, the PUBLIC/private port with the public sector dominant,
the PRIVATE/public port with the private sector dominant, and the PRIVATE port.
Table 1
Port function matrix
Port models Port functions
Regulator Landowner Operator
PUBLIC Public Public Public
PUBLIC/private Public Public Private
PRIVATE/public Public Private Private
PRIVATE Private Private Private
Sources: Baird (1995, 1997).
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 745
APUBLICport can be regardedas synonymous with a comprehensive port. It is a port in which all
three functions are controlled by the government or public authority. In the PUBLIC/private port,
the operator function is controlled by the private sector, with both the regulatory and landowner
functions remaining in the hands of the government. This type of port, therefore, can be interpreted
to be a variant of the landlord port. In the PRIVATE/public port, both the landowner and operator
functions are in private hands, while the regulatory function remains within the public sector. Fi-
nally, in the PRIVATE port, all three essential functions are controlled by the private sector.
2.2. The economics of efficiency and port privatisation
Engendering greater competition in the market is the main mechanism by which enhanced
efficiency might be expected to be achieved under a policy of deregulation. Baumol et al. (1982)
provide a comprehensive explanation of the process and conditions under which this might prove
to be the case. Although deregulation policies have been commonly applied in many industries
and across many countries (especially to the land-based transportation sector), the perceived
wisdom is that privatisation is the most important policy for improving the efficiency of the ports
sector. This is the case even though it is by no means categorically proven that there exists a direct
causal link between the degree of private sector involvement and economic efficiency (see the
differing conclusions drawn from the analyses conducted by Millward, 1982; Millward and Par-
ker, 1983; Vickers and Yarrow, 1988; Boardman and Vining, 1989; Hutchinson, 1991; Parker,
1994). Perhaps the most obvious conclusion to draw from this morass of contradictory evidence
is, as Liu (1995) attests in relation to the ports sector, that the best way forward is to undertake a
specific empirical analysis to determine the relative efficiencies of alternative forms of ownership.
Hartley et al. (1991) and Parker (1994) attempt to set up a theoretical framework for imputing
and testing the importance of ownership. A conceptual schema due to the latter is illustrated in
Fig. 1, where Point A represents the position of a firm which is directly controlled by a gov-
ernment department. It is politically controlled and there are no tradable shares. Hence, from the
Fig. 1. Conceptual mapping for efficiency improvements.
746 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
theories of public choice and property rights, it is to be expected that efficiency will be low. Point
B represents an activity undertaken by a government agency which has some, if limited, autonomy
from the political process. Public corporations can be placed at point C since they have more
autonomy than quasi-governmental agencies. Points D, E and F correspond to forms of own-
ership in the private sector. Point D includes those private sector firms which are close to the
public sector because of government funding or a reliance on government contracts; character-
istics which might diminish incentives to be efficient. Point E is a joint stock company, while point
F represents private ownership where property rights are least attenuated, particularly the owner-
manager company.
An upward movement along the y-axis corresponds to a shift away from monopoly towards
competition and, thus, greater product market pressure to be efficient. Fig. 1, therefore, provides a
mapping of the expected relationship between ownership and performance. The schema implies
that changes in ownership involving movement away from political control and towards private
ownership, but with no change in competition, will be associated with improved efficiency due to a
change in the capital market (X to Y).
2.3. The Asian situation
As an obvious corollary of burgeoning international trade and the rapidly increasing seaborne
cargoes associated with this, Asian ports have played a pivotal role in national and regional
economic development (Haynes et al., 1997; Robinson, 1998). Table 2 shows that, in terms of
Table 2
Container port traffic league (Units: TEUs p.a.)
Rank
a
Port 1999 1998 Country
1 (2) Hong Kong 16,100,000 14,582,000 China
2 (1) Singapore 15,900,000 15,100,000 Singapore
3 (3) Kaohsiung 6,985,361 6,271,053 Taiwan
4 (5) Pusan 6,439,589 5,945,614 Korea
5 (4) Rotterdam 6,400,000 6,010,503 Netherlands
6 (6) Long Beach 4,408,480 4,097,689 USA
7 (10) Shanghai 4,210,000 3,066,000 China
8 (8) Los Angeles 3,828,851 3,378,219 USA
9 (7) Hamburg 3,750,000 3,550,000 Germany
10 (9) Antwerp 3,614,264 3,265,750 Belgium
11 (13) New York/New Jersey 2,863,342 2,465,993 USA
12 (11) Dubai 2,844,634 2,804,104 UAE
13 (12) Felixstowe 2,700,000 2,523,639 UK
13 (13) Tokyo 2,700,000 2,168,543 Japan
15 (21) Keelung 2,550,419 1,820,018 Taiwan
16 (19) Tanjung Priok 2,273,303 1,898,069 Indonesia
17 (16) Gioia Tauro 2,253,401 2,125,640 Italy
18 (17) Kobe 2,200,000 1,900,737 Japan
18 (15) Yokohama 2,200,000 2,091,420 Japan
20 (22) Bremen/Bremerhaven 2,180,955 1,812,441 Germany
Source: Derived from Boyes (2000).
a
Figures in brackets represent the rank position for 1998.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 747
annual container throughput measured in TEUs,
1
10 Asian ports are ranked among the top 20
container ports in the world and that the top four positions are held by ports in Asia. Fleming
(1997) and Robinson (1998) provide plausible explanations as to why these Asian container ports
have emerged as both regional and international load centres.
As far as the administrative and ownership structures of Asian container ports are concerned,
Cullinane and Song (2001) conclude that, in terms of the taxonomy developed by Baird (1995,
1997), ports and terminals in Asia have tended to move away from something akin to a pure
PUBLIC model and towards the PUBLIC/private model of port administrative and ownership
structures. While it has become almost commonplace for the terminal operator function to fall
within the purview of a private sector organisation, the current situation is that landowner and
regulatory functions are often maintained under public sector control, whether at national or at
regional level.
By borrowing the concepts of Table 1 and referring to the horizontal axis of Fig. 1, we now can
describe the selected major ports/terminals in the region as illustrated in Fig. 2. Hong Kong’s
main container operators – Hong Kong International Terminals (HIT) and Modern Container
Terminals (MTL) – are closest to the PRIVATE port, while Korea’s Pusan and Japan’s Kobe
ports are located somewhere between the PRIVATE and the PRIVATE/public model. Taiwan’s
Kaohsiung port can be allocated somewhere between the PRIVATE/public and the PUBLIC/
private port, while Singapore and Shanghai ports are closest to the PUBLIC port.
Based on the administrative and ownership characteristics of the major Asian ports presented
in Fig. 2, the following hypothesis is established for empirical investigation:
Fig. 2. The hypothesis for port administration, ownership and efficiency.
1
Twenty-foot equivalent unit; a standard size of container used, amongst other things, for denoting the container
carrying capacity (size) of ships and the handling capacity and throughput of ports.
748 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
The economic efficiency of container ports/terminals improves as ownership moves along the
continuum towards greater private sector participation (i.e., in line with a movement away from
the PUBLIC port to the PRIVATE port).
3. Methodology
Over the last decade a number of methods for measuring efficiency have been proposed, all of
which have in common the concept of the frontier; efficient units are those operating on the cost
or production frontier, while inefficient ones operate either below the frontier (in the case of the
production frontier) or above the frontier (in the case of the cost frontier).
The literature on frontier models begins with Farrell (1957) who suggested a useful, and sub-
sequently widely accepted, framework for analysing economic efficiency in terms of realised de-
viations from an idealised frontier isoquant.
A distinction exists between the methods employed to derive the specification of the frontier
model: either statistical or non-statistical methods may be used. The former technique makes
assumptions about the stochastic properties of the data, while the latter does not. Another dif-
ference concerns whether the chosen method is parametric or non-parametric. While the former
imposes a particular functional form, the latter approach does not. The non-parametric approach
revolves around mathematical programming techniques which are generically referred to as data
envelopment analysis (DEA). The parametric approach, on the other hand, employs econometric
techniques where efficiency is measured relative to a frontier production function which is sta-
tistically estimated.
Econometric approaches have a strong policy orientation, especially in terms of assessing al-
ternative industrial organisations and in evaluating the efficiency of government and other public
agencies. Mathematical programming approaches, on the other hand, have a much greater
managerial decision-making orientation (Aigner and Schmidt, 1980; Fare et al., 1994; Lovell,
1995). The policy orientation of econometric approaches more closely supports the purpose of this
paper, especially since they have a more solid grounding in economic theory (Forsund et al., 1980;
Pitt and Lee, 1981; Bauer, 1990). In addition, several studies (e.g. Gong and Sickles, 1992; Yu,
1995; Oum and Waters, 1996) have compared the performance of alternative methods for mea-
suring efficiency, focusing on the econometric method (in particular, the stochastic frontier model)
and the mathematical programming method. As measured by the correlation coefficients and rank
correlation coefficients between the true and estimated relative efficiencies, the results show that
when the functional form of the econometric model is well specified, the stochastic frontier ap-
proach generally produces better estimates of efficiency than the latter approach, especially when
measuring firm-specific efficiency where panel data are available.
The econometric approach involves the specification of a parametric representation of tech-
nology which itself can be divided into two different models; either deterministic or stochastic
frontiers may be specified according to whether or not certain assumptions are made concerning
the underlying data. The early parametric frontier models (e.g. Aigner and Chu, 1968; Afriat,
1972) are deterministic in the sense that all economic units share a common fixed class of fron-
tier. This is unreasonable and ignores the real possibility that the observed performance of the
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 749
economic unit may be affected by exogenous (i.e. random shock) as well as endogenous (i.e. in-
efficiency) factors. To allocate all these factors, whether favourable or unfavourable and whether
under or beyond the control of the economic unit, into a single disturbance term and to refer to
the mixture as inefficiency is clearly a dubious and imprecise generalisation.
As an alternative, the stochastic frontier model is motivated by the idea that deviations from the
production frontier might not be entirely under the control of the economic unit being studied
(Greene, 1993). Aigner et al. (1976, 1977) and Meeusen and van den Broeck (1977) independently
constructed a more reasonable error structure than a purely one-sided one. They considered a
linear model for the frontier production function as follows:
Y
it
¼ f ðX
it
; bÞ expðv
it
À u
it
Þ; i ¼ 1; 2; . . . ; N; t ¼ 1; 2; . . . ; T; ð1Þ
where Y
it
denotes the appropriate form of output for the ith firm at time t, X
it
is a vector of inputs
associated with the ith firm at time t and b is a vector of input coefficients for the associated
independent variables in the production function. In the disturbance terms, the component v
it
represents a symmetric disturbance term permitting random variation of the production function
across economic units due, not only to the effects of measurement and specification error but also,
to the effects of exogenous shock beyond the control of the economic unit (e.g. weather condi-
tions, geography or machine performance). The other component u
it
(P0) is a one-sided dis-
turbance term and represents ‘productive inefficiency’ relative to the stochastic production
function. The non-negative disturbance term u
it
reflects the fact that output must either lie on or
below its frontier. The deviation of an observation from the deterministic kernel of the stochastic
production function (Eq. (1)) arises from two sources: (i) symmetric random variation of the
deterministic kernel f ðX
it
; bÞ across observations that is captured by the component v
it
and (ii)
asymmetric variation (or productive inefficiency) captured by the component u
it
. The term u
it
measures productive inefficiency in the sense that it measures the shortfall of output Y
it
from that
implied by its maximum frontier given by f ðX
it
; bÞ expðv
it
Þ.
Nevertheless, any estimate of a firm’s efficiency level is not consistent, as it contains statistical
noise as well as productive inefficiency. In addition, stochastic frontier models suffer from two
other difficulties. One is the requirement of specific assumptions about the distributions under-
lying productive inefficiency (e.g. half-normal and exponential) and statistical noise (e.g. normal).
The other is the required assumption that regressors (the input variables contained in the vector
X) and productive inefficiency are independent. This may well be an unrealistic assumption since if
a firm knows its level of inefficiency, this should affect its input choices.
A further development in the modelling of frontiers lies with the use of estimation techniques
which involve panel data. Initially, the stochastic frontier model (1) was developed for cross-
sectional data. Hausman and Taylor (1981), Baltagi (1995) and Blundell (1996) list a number of
attractive features of using panel data, one of which is that panel data allows for the control of
individual effects which may be correlated with other variables included in the specification of an
economic relationship, thus making the analysis on the basis of single cross-sections extremely
difficult.
Initially, it was claimed that productive efficiencies for individual firms could not be estimated
and predicted. In an effort to explore this unsolved problem with the previous models, along with
attempting to reap the benefits of the aforementioned advantages of panel data, Pitt and Lee
750 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
(1981) were the first to develop techniques using panel data to estimate the frontier production
function. Jondrow et al. (1982) presented two estimators (i.e. for half-normal and exponential
cases) for the firm-specific effect for an individual firm under the assumption that the parameters
of the frontier production function were known and cross-sectional data were available for given
sample firms. Schmidt and Sickles (1984) suggested three different estimators for individual firm
effects and productive efficiencies for panel data.
A major breakthrough in the area of panel data models was achieved by Battese and Coelli
(1988), who presented a generalisation of the results of Jondrow et al. (1982) on the assumption of
a more general distribution for firm effects to be applied to the stochastic frontier model. Their
model is given by
Y
it
¼ f ðX
it
; bÞ expðv
it
À u
i
Þ; i ¼ 1; 2; . . . ; N; t ¼ 1; 2; . . . ; T: ð2Þ
It can be seen that the main difference between models (1) and (2) is the absence of the subscript t
associated with u in the latter. Thus, u captures firm-specific time-invariant variables omitted from
the previous function. The symmetric terms v
it
are assumed to be identically and independently
normally distributed with mean zero and variance r
2
v
, i.e., v
it
$ Nð0; r
2
v
Þ. The one-sided terms u
i
(P0) are assumed to be identically and independently distributed non-negative random variables
which capture a firm effect but no time effect (Schmidt and Sickles, 1984). In addition, the error
terms v
it
and u
i
are assumed to be independently distributed of the input variables as well as of one
another.
The most frequently defined distribution for the u
i
is the half-normal (i.e. u
i
$ jNð0; r
2
u
ÞjÞ
though other distributional assumptions for the u
i
terms have been proposed by several re-
searchers. For example, the exponential (Aigner et al., 1977), the truncated normal (Stevenson,
1980) and the gamma (Greene, 1980).
As far as the productive efficiency of a firm is concerned, Battese and Coelli (1988) define it as
the ratio of the firm’s mean production, given its realised firm-specific effect, to the corresponding
mean production with the firm effect being equivalent to zero. The productive efficiency of the ith
firm (PE
i
) is defined, therefore, as
PE
i
¼
EðY
it
à jui; X
it
Þ
EðY
it
à jui ¼ 0; X
it
Þ
; ð3Þ
where Y
Ã
it
represents the output of production for the ith firm at time t, and the value of the PE
i
lies
between zero and one (0 6PE
i
61). If a firm’s productive efficiency is calculated as 0.65, for
example, then this implies that, on average, the firm realises 65% of the production possible for a
fully efficient firm having comparable input values.
This technique, attributable to Battese and Coelli (1988), is relevant to a case where productive
efficiency is time-invariant. Schmidt (1985), however, states that unchanging inefficiency over time
is not a particularly attractive assumption; a criticism which is readily admitted by Battese and
Coelli (1988). With the assumption that productive efficiency does vary over time, an alternative
approach has been adopted by econometricians such as Cornwell et al. (1990) and Kumbhakar
(1990). None of these studies succeed, however, in completely separating inefficiency from indi-
vidual firm effects (Kumbhakar and Hjalmarsson, 1993) and, in any case, the methods proposed
thus far are too complicated for empirical application (Ferrantino and Ferrier, 1995).
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 751
4. An application to Asian container ports
4.1. Definition of variables
Dowd and Leschine (1990) argue that the productivity of a container port/terminal depends on
the efficient use of labour, land and equipment. Terminal productivity measurement, therefore, is
a means of quantifying the efficiency with which these three resources are utilised. As a measure of
the output of port production, Bernard (1991) questions whether total tonnage handled at a port,
should be applied to container terminals. Since the basic unit of output measurement is a con-
tainer and since, irrespective of its size and (especially) its weight, the facility inputs for the
movement of any container are more or less the same, his argument rests with the fact that the use
of total tonnage handled seems to be an illogical metric for the assessment of output at a container
terminal or port. One possible obvious solution to representing the output of a container terminal
may be provided by measuring the throughput in terms of the number of container movements
across the quayside or, alternatively, in terms of the monetary value of these movements as in-
dicated by the revenue associated with this operation.
In terms of a conventional categorisation of inputs, a typical expenditure structure of a port
over a given period of time is illustrated in Fig. 3.
As a proxy for the capital input variable, the combined values of buildings and equipment
(mainly cargo-handling equipment) accounts for 42% of total expenditure. Thus, the labour and
capital costs of a port or terminal together comprise 95% of the total cost structure of a port or
terminal operation. It seems reasonable enough to assume that this can be taken as sufficient to
describe the whole cost account.
It was originally intended that the basic economic inputs of capital and labour would fulfil the
data requirements of the study. This was the approach adopted in Song and Cullinane (2001)
where the focus was the productive efficiency of Korean and UK container terminals. Unfortu-
nately, across the Asian region in general, this sort of cost data proved impossible to collect from
secondary sources. In contrast with the situation in the UK, for example, accounting conventions
within the region have a general tendency not to require the publication of costs at a high enough
level of detail to allow their identification. Instead, an alternative approach was adopted which
utilises certain physical characteristics of the terminals as the required input data. This is related
to output data in order to assess relative efficiency. Thus, terminal quay length (X
1
), terminal area
in hectares (X
2
) and the number of pieces (X
3
) of cargo handling equipment (including gantry
Fig. 3. A port/terminal expenditure structure. (Source: derived from Sachish, 1996, p. 347.)
752 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
cranes, ship-shore gantries, yard cranes, and mobile cranes etc.) were employed. Such an ap-
proach has the advantage that the data on these measures of physical container terminal ca-
pacities are available within the public domain and precedents do exist where they have been used
in this way (e.g., Notteboom et al., 2000; Tongzon, 1995).
For the econometric estimation of efficiency using the stochastic frontier model, data on out-
puts from the container terminal sample are also required. In Song and Cullinane (2001), the
terminal output (Y) was defined as the turnover derived from the provision of container terminal
services but excluding property sales. Again, because of the range of accounting systems employed
by the sample of terminal operators, separating out the revenue attributable to different sources
proved to be an intractable problem. In parallel with the solution proposed for the data re-
quirements on production inputs, the readily accessible, physical measure of annual con-
tainer throughput in TEUs was adopted as the basis for measuring the productive output of
container terminals (Y). This approach also has its precedents (e.g., Bernard, 1991; Notteboom
et al., 2000) though it would be preferable to use (but impossible to collect information on) the
actual number of boxes handled. A summary of the major characteristics of the variables is
presented in Table 3.
4.2. Data sources
The sample comprised 15 container ports or terminals in Asia, namely: Singapore; HIT, MTL,
Sealand (all three in Hong Kong); Kaohsiung, Keelung (Taiwan); Pusan (Korea); Shanghai,
Dalian, Yantian (China); Tokyo, Yokohama, Kobe (Japan); Port Kelang (Malaysia); and Manila
(Philippines). In virtually all cases, annual data were collected for the 10-year period from 1989 to
1998. In the case of the recently developed Yantian container terminal, however, data are only
available for the 6 years from 1993 to 1998. The data were collected mainly from the Contai-
nerisation International Yearbook (various issues) but was validated and, in certain instances
supplemented, by approaching each of the terminals directly. This process yielded a total of 146
observations.
4.3. Model specification and assumptions
The estimation of relative terminal operator efficiency is conducted by assuming the ap-
propriateness of the log-linear Cobb–Douglas case. No other specifications were tested. The
Table 3
Statistical properties of the variables (1989–1998)
Variables
a
Mean Median Min Max S.D.
Y 2,476,032.00 1,812,710.00 105,736.00 15,100,000.00 2,524,487.00
X
1
3,483.00 3,192.00 305.00 8,754.00 2,397.00
X
2
111.89 94.43 12.50 275.50 73.45
X
3
73.23 52.50 4.00 433.00 86.37
a
Y is defined as the terminal output as measured by annual container throughput in TEUs. X
1
is defined as the
terminal quay length in metres. X
2
is defined as the terminal area in hectares. X
3
is defined as the number of pieces of
cargo handling equipment employed.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 753
logarithmic stochastic frontier model specified for the container terminal operating sector in the
cross-sectional case is defined by
ln Y
it
¼ ln f ðX
1it
; X
2it
; X
3it
; bÞ þ v
it
À u
it
; i ¼ 1; 2; . . . ; 15; t ¼ 1; 2; . . . ; T: ð4Þ
For application to the panel data, the model (2) is transformed into
ln Y
it
¼ ln f ðX
1it
; X
2it
; X
3it
; bÞ þ v
it
À u
i
; i ¼ 1; 2; . . . ; 15; t ¼ 1; 2; . . . ; T: ð5Þ
In both cases, Y
it
represents the output of the ith container terminal operator at time t; X
1it
; X
2it
and
X
3it
denote the respective input variables associated with the ith terminal operator at time t and b
is a vector of input coefficients associated with the independent variables in the model and is the
object of estimation. The disturbance term v
it
represents the symmetric (statistical noise) com-
ponent. In model (4), u
it
(P0) is the one-sided (inefficiency) component and, in model (5), the
one-sided disturbance term u
i
(P0) represents the ‘terminal operator-specific time invariant in-
efficiency’.
The terminal operators are assumed to be profit-maximisers and price takers in their input
markets. Hence, input prices may be treated as exogenous. Another assumption is that there is a
single-output production function. Thus, earnings from sources such as the sales of terminal
property are not classified as output and do not affect the production function frontier. Finally,
the concept of the average terminal frontier is applied as the definition of the frontier. Estimation
of terminal efficiency is conducted by the David–Fletcher–Powell algorithm (Fletcher, 1980;
Greene, 1997) using the LIMDEP 7.0 econometric software package (Greene, 1995).
Since the maximum likelihood method applied within this algorithm is a large-sample esti-
mation procedure (Maddala, 1992), it is required that an asymptotic test statistic be used. Thus,
since it is one of the general large-sample tests based upon MLE, the likelihood ratio test statistic
(LR) is applied to test whether or not model coefficients are significantly different from zero.
Under general conditions, the LR has a v
2
distribution with degrees of freedom equal to the
number of restrictions imposed and can be expressed as follows (Engle, 1984):
LR ¼ À2 ln
L
R
L
U
_ _
; ð6Þ
where L
R
denotes the ‘restricted’ likelihood function and L
U
the ‘unrestricted’ likelihood function.
5. Estimating the productive efficiency of Asian container ports
5.1. The cross-sectional model
The first step in the estimation procedure is to check the sign of the third moment and the
skewness of the OLS residuals associated with the sample data (Waldman, 1982). The third
moment for the terminal frontier model is )0.130; the negative sign implying that the residuals of
the sample data possess the correct pattern for the implementation of the MLE procedure. This is
reflected in the histogram of the residuals shown in Fig. 4 which is clearly negatively skewed.
Because estimation procedures yield merely the residuals e rather than the inefficiency term
u, this term in the model must be observed indirectly (Greene, 1993). In the case of the
754 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
cross-sectional model shown in Eq. (2), Jondrow et al. (1982) suggest the conditional expectation
of u
it
, conditioned on the realised value of the error term e
it
¼ ðv
it
À u
it
Þ, as an estimator of u
it
. In
other words, E½u
it
je
it
Š is the ‘mean productive inefficiency’ for the ith terminal operator in the
industry at any time t. Under each of the three assumed possible distributional forms for the
inefficiency term in the model, this means that:
• For the half-normal model:
E½u
it
je
it
Š ¼
rk
ð1 þ k
2
Þ
/
e
it
k
r
_ _
U À
e
it
k
r
_ _
_
À
e
it
k
r
_
: ð7Þ
• For the exponential model:
E½u
it
je
it
Š ¼ ðe
it
À hr
2
v
Þ þ
r
v
/ e
it
À hr
2
v
_ _
=r
v
_ ¸
U e
it
À hr
2
v
_ _
=r
v
_ ¸
: ð8Þ
• For the truncated normal model: the inefficiency term is obtained merely by replacing ðe
it
kÞ=r
in Eq. (7) with:
e
it
k
r
_
þ
l
rk
_
: ð9Þ
Based on the sample cross-sectional data, the OLS estimates and the MLEs for each of the three
assumed distributions of the inefficiency term in the frontier model (4) are shown in Table 4.
The estimated OLS coefficients are of limited value but do provide a starting point for the MLE
process. The goodness of fit of the estimated regression equation evaluated by R
2
for the least
squares method looks reasonably high at 0.649. This implies that the three inputs to the model do
satisfactorily explain the model output. In addition, the F-statistic of 87.86 shows that the rela-
tionship between exogenous and endogenous variables is significant at the 1% level.
The results also show that, except for the second input (X
2
) relating to the terminal area in
hectares, all variables are statistically significant at the 5% level and that the MLEs under the three
alternative inefficiency distributions yield parameters which are close to each other. In addition,
Fig. 4. Skewness of the OLS residuals.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 755
the signs of the parameters conform to a priori expectations. Another interesting point to address
here is that, except for the second input (X
2
) which becomes statistically significant at the 1% level,
the MLEs differ only marginally from the OLS estimates. This is to be expected since both
methods are consistent. The likelihood ratio test statistic of 153.24 reveals a high degree of sig-
nificance beyond the 1% level, thus leading to the rejection of the null hypothesis that the coef-
ficients are equal to zero.
2
As can be seen in Fig. 5, aside from the case of Kobe where a comparatively poor correlation of
approximately 0.67 was found when comparing the inefficiency estimates from each distributional
assumption, the worst correlation over time between different inefficiency estimates for any of the
other terminals or ports is impressively high at 0.995. The unusual result for Kobe is explained by
the influence exerted by the ‘strange’ data returned for the period immediately following its ca-
lamitous earthquake. Under the assumption of a half-normal distribution for the inefficiency
term, the overall inefficiency level of terminal operators over the sample period is illustrated in
Fig. 5.
The parameter k (as calculated by r
u
=r
v
) provides some insight into the relative variance of the
two composite errors that make up the total variation in the structural disturbance term. The two
variances of the two error components, in addition to k, indicate that the inefficiency component u
varies more widely than the uncontrollable random exogenous component v. This means that the
productive inefficiency u makes a more important contribution to the variability of the total error
in the cross-sectional frontier model. Prior to analysing the efficiency level of each terminal
Table 4
Frontier production function of Asian container terminals
Variables/parameters OLS MLE
Half-normal Exponential Truncated normal
Constant 9.875
Ã
10.234
Ã
10.187
Ã
10.265
(22.00) (21.245) (24.394) (10.324)
ln X
1
0.154 0.157 0.154 0.163
(1.58) (1.702) (1.736) (1.757)
ln X
2
0.345
ÃÃ
0.340 0.338 0.328
(3.20) (3.136) (3.240) (2.995)
ln X
3
0.440
Ã
0.432
Ã
0.431
Ã
0.432
(6.21) (5.806) (6.301) (5.378)
k – 0.824 0.495 0.963
r
2
v
– 0.249 0.247 0.237
r
2
u
– 0.168 0.060 0.220
h – – 4.067 –
l – – – 0.044
Log-likelihood – –121.635 )120.886 )121.630
Figures in parentheses indicate t-ratios.
*
Not significant at the 1% level.
**
Not significant at the 5% level.
2
The Likelihood Ratio test statistic is calculated as LR ¼ À2ðÀ198:45 þ 121:83Þ ¼ 153:24.
756 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
operator as it varies over time, it is useful to see each operator’s average efficiency level over the
whole sample time period as shown in Table 5.
Under all three assumed distributions for which the model parameters have been estimated, the
average efficiency level of Kaohsiung is consistently highest, followed by Pusan, Singapore,
Keelung and MTL’s Hong Kong terminal. The container ports of Kobe in Japan, Manila in the
Philippines and Yantian, Shanghai and Dalian in the Chinese mainland are consistently the most
inefficient ports in the sample, even though their precise rankings do not correlate perfectly across
all the distributional assumptions imposed.
One intriguing implication which can be drawn from this result is that the efficiency of a
container port or terminal appears to be very closely correlated to its size as measured in terms of
Fig. 5. Overall productive inefficiency levels (half-normal distribution).
Table 5
Average productive efficiency of Asian container ports/terminals (%)
Container port Half-normal Exponential Truncated normal
Singapore 79.43 (3) 85.25 (3) 79.83 (3)
HIT 75.55 (6) 82.12 (6) 75.57 (6)
MTL 75.75 (5) 82.36 (5) 75.85 (5)
Sealand 73.45 (7) 79.79 (7) 73.00 (7)
Kaohsiung 80.09 (1) 85.71 (1) 80.53 (1)
Keelung 79.39 (4) 85.17 (4) 79.61 (4)
Pusan 79.85 (2) 85.54 (2) 80.32 (2)
Shanghai 68.63 (11) 75.37 (11) 67.77 (12)
Dalian 65.43 (13) 71.45 (13) 63.90 (14)
Yantian 55.98 (15) 55.45 (15) 68.55 (11)
Tokyo 72.47 (8) 79.60 (8) 72.22 (8)
Yokohama 71.52 (9) 78.81 (9) 71.24 (9)
Kobe 67.28 (12) 74.54 (12) 66.36 (13)
Port Kelang 70.35 (10) 77.39 (10) 69.83 (10)
Manila 63.69 (14) 70.19 (14) 62.33(15)
(1) Figures are calculated by converting the inefficiency estimates using expðÀuÞ. (2) Efficiency rankings are shown in
parentheses.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 757
throughput; a result which is validated by previous empirical work (for a review, see Cullinane
and Khanna, 1999). Another possible interpretation might be that there is an inverse relationship
between the degree of centralised government control which is exerted over a port or terminal and
its level of efficiency. This second inference would appear to potentially validate the theories of
public choice and property rights.
5.2. The panel model
By generalising the cross-sectional results in Jondrow et al. (1982) to panel models, Battese and
Coelli (1988) propose the following estimation of the ‘time-invariant terminal operator-specific
inefficiency’ u
i
under the half-normal assumption:
E½u
i
je
i1
; . . . ; e
iT
Š ¼ l
Ã
i
þ r
Ãi
/
l
Ã
i
r
Ãi
_ _
U À
l
Ã
i
r
Ãi
_ _
_
_
_
_
; ð10Þ
where l
Ã
i
¼ c
i
l þ 1 À c
i
ð Þ Àe
i
ð Þ, c
i
¼ 1= 1 þ
k
T
i
_ _
and r
Ãi
¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r
2
u
1þkT
i
_ _
_
.
Based on the panel data, the MLEs of the terminal frontier model (5) are presented in Table 6.
As in the cross-sectional model, because the value of r
u
=r is high, the inefficiency disturbance u
makes a more important contribution to the total variation represented in the error component
than do the uncontrolled shocks denoted by v.
Given the parameters in Table 6, we are now able to compute the ‘time invariant terminal
operator-level efficiency’ by applying the formula given in (10). Fig. 6 shows the specific efficiency
for the seven major Asian container ports/terminals in the sample.
As distinct from the results achieved when considering the cross-sectional model, Singapore
is measured at the highest level of productive efficiency, with Pusan second, Kobe third and
Kaohsiung fourth out of the major ports in the region. The mainland Chinese port of Shanghai
again appears as the least efficient operator. The main difference in results compared to the cross-
sectional model is the dramatic improvement in the measured efficiency level of Kobe and indi-
Table 6
Maximum likelihood estimates of panel frontier model for Asian container ports/terminals
Coefficients Asymptotic t-ratios
Constant 9.648
Ã
11.427
lnX
1
0.363 1.973
lnX
2
0.427 1.991
lnX
3
0.145 1.741
k 1.437
r
2
v
0.119
r
2
u
0.171
Log-likelihood )81.913
k ¼ r
2
u
=r
2
v
.
*
Not significant at the 1% level.
758 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
cates very clearly the abnormal effect of its earthquake upon the data collected and the results
produced by the ensuing analysis.
6. Conclusions
Although not wholly conclusive, the results achieved from this application of the stochastic
frontier model provide evidence of the efficiency rankings of a selection of container ports and
terminals within the Asian region. There exist certain anomalies, however, which seem to suggest
that the method of analysis does not necessarily produce robust results. This might be inferred
from the divergent results obtained for Kobe from the cross-sectional, as opposed to the panel
data analysis. As such, the impact of ‘catastrophes’ may well be a fruitful avenue for further
investigation and provide much-needed justification for the omission of outliers from the sample.
One major intervening variable which might well have influenced the results achieved within
this study is the dichotomous positions of some of the sample ports/terminals with respect to the
level of market regulation of container terminal operations, particularly on the supply side. Whilst
recognising that there could well be other influential intervening variables, simultaneously con-
trolling for each of the individual effects of both private sector participation and market dere-
gulation will allow greater fine tuning in policy assessment and/or formulation. Within the context
of the container terminal industry, the greater cross-sectional variation in the combinations of
these two variables which comes from using a larger panel sample (particularly one with greater
diversity in geography and political economy) will permit the isolation of differences in their static
effects. Similarly, time-series variation in the sample will facilitate the assessment of the dynamic
impacts of changes in policy.
On the basis of purely a subjective appraisal of the results obtained, however, there does seem
to be some support for the notion that greater privatisation within, and/or deregulation of, the
market does seem to be closely associated with enhanced productive efficiency. This empirical
investigation, however, yields no definitive and irrefutable link between the degree of private
Fig. 6. Major Asian container port/terminal-specific time invariant efficiency.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 759
sector participation and the level of productive efficiency. Although there certainly appears to be
some justification for the belief that there exists a positive relationship between the two, the most
persuasive inference to be drawn from the analysis is the consistency with which large throughput
operations appear to outperform their smaller counterparts in terms of efficiency; a factor which is
likely to further reinforce the existing dominant market positions of certain ports and terminals in
the region.
Acknowledgements
The authors are grateful to the British Council, Hong Kong and the Hong Kong Polytechnic
University who provided the funding for this research under the UK/HK Joint Research Scheme
(Project Numbers JRS 94/38, G-T149 and G-S976). We would also take this opportunity to ex-
press our appreciation of the many container ports and terminals in Asia who provided us with
the supplementary data necessary to undertake this analysis. Finally, we would express our
heartfelt thanks to Professor Frank Haight and two anonymous referees for suggestions which
have undoubtedly improved the paper.
References
Afriat, S., 1972. Efficiency estimation of production functions. International Economic Review 13 (3), 568–598.
Aigner, D., Chu, S., 1968. On estimating the industry production function. American Economic Review 58 (4),
826–839.
Aigner, D., Amemiya, T., Poirier, D., 1976. On the estimation of production frontiers: maximum likelihood estimation
of the parameters of a discontinuous density function. International Economic Review 17 (2), 377–396.
Aigner, D., Lovell, C., Schmidt, P., 1977. Formulation and estimation of stochastic frontier production function
models. Journal of Econometrics 6 (1), 21–37.
Aigner, D., Schmidt, P., 1980. Specification and estimation of frontier production, profit and cost functions, editors’
introduction. Journal of Econometrics 13 (Suppl.), 1–3.
Baird, A., 1995. UK port privatisation: in context. In: Proceedings of UK Port Privatisation Conference. Scottish
Transport Studies Group, 21 September, Edinburgh.
Baird, A., 1997. Port privatisation: an analytical framework. In: Proceedings of International Association of Maritime
Economist Conference, City University, London, 22–24 September.
Baltagi, B., 1995. Econometric Analysis of Panel Data. Wiley, Chichester.
Battese, G., Coelli, T., 1988. Prediction of firm-level technical efficiencies with a generalised frontier production
function and panel data. Journal of Econometrics 38, 387–399.
Bauer, P., 1990. Recent developments in the econometric estimation of frontiers. Journal of Econometrics 46, 39–56.
Baumol, W., Panzar, J., Willig, R., 1982. Contestable Markets and the Theory of Industrial Structure. Harcourt Brace
Jovanovich, San Diego.
Bernard, J., 1991. European deep-sea container terminals: locational and operational perspectives. PhD Thesis,
University of Liverpool.
Blundell, R., 1996. Microeconometrics. In: Greenaway, D., Bleaney, M., Stewart, I. (Eds.), A Guide to Modern
Economics. Routledge, London, pp. 508–536.
Boardman, A., Vining, A., 1989. Ownership and performance in competitive environments: a comparison of the
performance of private, mixed and state-owned enterprises. Journal of Law and Economics 32, 1–33.
Boyes, J., 2000. Top 30 ports, Containerisation International, March 2000, p. 85.
760 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762
Cornwell, C., Schmidt, P., Sickles, R., 1990. Production frontiers with cross-sectional and time-series variation in
efficiency levels. Journal of Econometrics 46, 185–200.
Cullinane, K.P.B., Khanna, M., 1999. Economies of scale in large container ships. Journal of Transport Economics and
Policy 33 (2), 185–208.
Cullinane, K., Song, D.-W., 1998. Container terminals in South Korea: problems and panaceas. Maritime Policy and
Management 25 (1), 63–80.
Cullinane, K., Song, D.-W., 2001. Administrative and ownership structure of Asian container ports. International
Journal of Maritime Economics 3 (2), 175–197.
De Monie, G., 1996. Privatisation of port structures. In: Bekemans, L., Beckwith, S. (Eds.), Ports for Europe: Europe’s
Maritime Future in a Changing Environment. European Interuniversity Press, Brussels, pp. 267–298.
Dowd, T., Leschine, T., 1990. Container terminal productivity: a perspective. Maritime Policy and Management 17 (2),
107–112.
Engle, R., 1984. Wald, likelihood ratio and Lagrange multiplier tests in econometrics. In: Griliches, Z., Intriligator, M.
(Eds.), Handbook of Econometrics, vol. II. Elsevier, Amsterdam, pp. 775–826.
Fare, R., Grosskopf, S., Lovell, C., 1994. Production Frontiers. Cambridge University Press, Cambridge.
Farrell, M., 1957. The measurement of productive efficiency. Journal of the Royal Statistical Society 120 (3), 253–290.
Ferrantino, M., Ferrier, G., 1995. The technical efficiency of vacuum-pan sugar industry of India: an application of a
stochastic frontier production function using panel data. European Journal of Operational Research 80, 639–653.
Fleming, D., 1997. World container port ranking. Maritime Policy and Management 24 (2), 175–181.
Fletcher, R., 1980. Practical Methods of Optimisation. Wiley, New York.
Forsund, F., Lovell, C., Schmidt, P., 1980. A survey of frontier production functions and of their relationship to
efficiency measurement. Journal of Econometrics 13, 5–25.
Gong, B.H., Sickles, R., 1992. Finite sample evidence on the performance of stochastic frontiers and data envelopment
analysis using panel data. Journal of Econometrics 51, 259–284.
Goss, R., 1986. Seaports should not be subsidised. Maritime Policy and Management 13 (2), 83–104.
Goss, R., 1990. Economic policies and seaports: 2. the diversity of port policies. Maritime Policy and Management 17
(3), 221–234.
Greene, W., 1980. Maximum likelihood estimation of econometric frontier functions. Journal of Econometrics 13,
27–56.
Greene, W., 1993. The econometric approach to efficiency analysis. In: Fried, H., Lovell, C., Schmidt, S. (Eds.),
The Measurement of Productive Efficiency: Techniques and Applications. Oxford University Press, New York,
pp. 68–119.
Greene, W., 1995. LIMDEP (Version 7.0). User’s Manual. Econometric Software Inc, Castle Hill, Australia.
Greene, W., 1997. Econometric Analysis, third ed. Prentice-Hall International Inc, New Jersey.
Hartley, K., Parker, D., Martin, S., 1991. Organisational status, ownership and productivity. Fiscal Studies 12 (2),
46–60.
Hausman, J., Taylor, W., 1981. Panel data and unobserved individual effects. Econometrica 49, 1377–1398.
Haynes, K., Hsing, Y., Stough, R., 1997. Regional port dynamics in the global economy: the case of kaohsiung,
Taiwan. Maritime Policy and Management 24 (1), 93–113.
Heaver, T., 1995. The implications of increased competition among ports for port policy and management. Maritime
Policy and Management 22 (2), 125–133.
Hutchinson, G., 1991. Efficiency gains through privatisation of UK industries. In: Ott, A., Hartley, K. (Eds.),
Privatisation and Economic Efficiency: A Comparative Analysis of Developed and Developing Countries. Edward
Elgar, Aldershot, pp. 87–107.
Jondrow, J., Lovell, C., Materov, I., Schmidt, P., 1982. On the estimation of technical inefficiency in the stochastic
frontier production model. Journal of Econometrics 19, 233–238.
Kumbhakar, S., 1990. Production frontiers, panel data, and time-varying technical inefficiency. Journal of
Econometrics 46, 201–211.
Kumbhakar, S., Hjalmarsson, L., 1993. Technical efficiency and technical progress in Swedish dairy farms. In: Fried,
H., Lovell, C., Schmidt, S. (Eds.), The Measurement of Productive Efficiency: Techniques and Applications. Oxford
University Press, New York, pp. 256–270.
K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762 761
Liu, Z., 1995. The comparative performance of public and private enterprises: the case of British ports. Journal of
Transport Economics and Policy 29 (3), 263–274.
Lovell, C., 1995. Econometric efficiency analysis: a policy-oriented review. European Journal of Operational Research
80 (3), 452–462.
Maddala, G., 1992. Introduction to Econometrics, second ed. MacMillan, New York.
Meeusen, W., van den Broeck, J., 1977. Efficiency estimation from Cobb–Douglas production functions with composed
error. International Economic Review 18 (2), 435–444.
Millward, R., 1982. The comparative performance of public and private ownership. In: Roll, E. (Ed.), The Mixed
Economy. Macmillan, London, pp. 58–93.
Millward, R., Parker, D., 1983. Public and private enterprise: comparative behaviour and relative efficiency. In:
Millward, R. et al. (Eds.), Public Sector Economics. Longman, New York.
Nagorski, B., 1972. Port Problems in Developing Countries: Principles of Port Planning and Organisation.
International Association of Ports and Harbours, Tokyo.
Notteboom, T., Coeck, C., van den Broeck, J., 2000. Measuring and explaining the relative efficiency of container
terminals by means of Bayesian stochastic frontier models. International Journal of Maritime Economics 2 (2),
83–106.
Oum, T.H., Waters, W., 1996. A survey of recent development in transportation cost function research. Logistics and
Transportation Review 32 (4), 423–463.
Parker, D., 1994. Nationalisation, privatisation, and agency status within government: testing for the importance of
ownership. In: Jackson, P., Price, C. (Eds.), Privatisation and Regulation: A Review of the Issues. Longman, Essex,
pp. 149–169.
Pitt, M., Lee, L., 1981. The measurement and sources of technical efficiency in the indonesian weaving industry. Journal
of Development Economics 9, 43–64.
Robinson, R., 1998. Asian hub/feeder nets: the dynamics of restructuring. Maritime Policy and Management 25 (1),
21–40.
Sachish, A., 1996. Productivity functions as a managerial tool in Israeli ports. Maritime Policy and Management 23 (4),
341–369.
Schmidt, P., Sickles, R., 1984. Production frontiers and panel data. Journal of Business and Economic Statistics 2 (4),
367–374.
Schmidt, P., 1985. Frontier production functions. Econometric Reviews 4, 289–328.
Song, D.-W., Cullinane, K.P.B., 2001. Port Privatisation: A New Paradigm of Port Policy. In: Borgese, E.M., Chircop,
A., McConnell, M.L., Ocean Yearbook, Volume 16. University of Chicago Press, Chicago.
Stehli, H., 1978. Typology of ports. In: Beth, H. (Ed.), Port Management Textbook. Institute of Shipping Economics,
Bremen, pp. 25–35.
Stevenson, R., 1980. Likelihood functions for generalised stochastic frontier estimation. Journal of Econometrics 13,
57–66.
Tongzon, J., 1995. Determinants of port performance and efficiency. Transportation Research 29A (3), 245–252.
UNCTAD, 1995a. Comparative experiences with privatisation: policy insights and lessons learned. UNCTAD/DTCI/
23, New York..
UNCTAD, 1995b. Comparative analysis of deregulation, commercialisation and privatisation of ports. UNCTAD/
SDD/PORT/3, Geneva.
Vickers, J., Yarrow, G., 1988. Privatisation: An Economic Analysis. MIT Press, Cambridge, MA.
Vickers, J., Yarrow, G., 1991. Economic perspectives on privatisation. Journal of Economic Perspectives 5 (2), 111–132.
Waldman, D., 1982. A stationary point for the stochastic frontier likelihood. Journal of Econometrics 18, 275–279.
Yarrow, G., 1986. Privatisation in theory and practice. Economic Policy, 323–377.
Yu, C., 1995. A comparative study of alternative methods for efficiency measurement with applications to
transportation industry. Unpublished Ph.D. Dissertation, University of British Columbia, Vancouver, BC, Canada.
762 K. Cullinane et al. / Transportation Research Part A 36 (2002) 743–762