Connecting Logistics Performance To Export: A Scenario-Based Approach

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

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

net/publication/325899732

Connecting logistics performance to export: A scenario-based approach

Article  in  Research in Transportation Economics · June 2018


DOI: 10.1016/j.retrec.2018.05.007

CITATIONS READS

11 375

3 authors:

Ozgur Kabak Füsun Ulengin


Istanbul Technical University Sabanci University
86 PUBLICATIONS   982 CITATIONS    51 PUBLICATIONS   789 CITATIONS   

SEE PROFILE SEE PROFILE

Şule Önsel
Dogus Universitesi
59 PUBLICATIONS   926 CITATIONS   

SEE PROFILE

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

BSc Final Year Project View project

Climate Change View project

All content following this page was uploaded by Şule Önsel on 06 July 2018.

The user has requested enhancement of the downloaded file.


Research in Transportation Economics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Research in Transportation Economics


journal homepage: www.elsevier.com/locate/retrec

Connecting logistics performance to export: A scenario-based approach


Özgür Kabaka, Füsun Ülenginb,∗, Şule Önsel Ekicic
a
Istanbul Technical University, Department of Industrial Engineering, Maçka 34356, Istanbul, Turkey
b
Sabanci University, Sabanci School of Management, Istanbul, Turkey
c
Doğus University, Department of Industrial Engineering, Istanbul, Turkey

A R T I C LE I N FO A B S T R A C T

Keywords: We propose a novel scenario analysis-based approach to investigate the relationship between logistics perfor-
Logistics performance mance and exports at a country level. Different from the existing scenario analysis methods that contain sub-
Scenario analysis jectivity in measuring consistency, the proposed approach uses objective information rather than expert opi-
ExportJEL classification: nions. We propose a binary integer program to select significantly different, consistent and small number of
R42
efficient scenarios. On the basis of the selected scenarios, we develop a new approach for evaluating the per-
C61
formance of strategies across a series of scenarios.
C65
F10 Furthermore, the proposed approach is used to analyse the relationships between exports and the six in-
dicators of the logistics performance index (LPI): customs, infrastructure, international shipments, logistics
quality and competence, tracking and tracing, and timeliness. With the help of these case studies, conducted for
Turkey and four other selected countries, namely Burundi, Zimbabwe, Brazil and Portugal, the proposed method
is proved to be generally applicable.

1. Introduction Gunasekera, Anderson, & Lakshmanan, 2008; Lean, Huang, & Hong,
2014).
Development of public policies for logistics that covers freight Infrastructure has a very close relationship with economic devel-
transportation, border clearance, warehousing, payment systems and opment and international trade. The improvement of the logistics in-
many other functions is an important area for national governments to frastructure will reduce the travel time, and the resulting time savings
support regional and international organisations (Arvis, Mustra, Ojala, will facilitate producers’ access to distant markets and the drawing of
Stepherd, & Saslavsky, 2012). For a country, it is necessary to under- inputs from distant markets. Fedderke and Garlic (2008) show a posi-
stand its logistics performance to better evaluate and target trade and tive relationship between infrastructure and economic development in
transport facilitation policy efforts over time and across countries. A South Africa. Bensassi, Márquez-Ramos, Martínez-Zarzoso, and Suárez-
competitive advantage in terms of logistics performance can lead a Burguet (2015) analysed the relationship between logistics infra-
country to increase its international trade potential and help to reach structure and trade. According to the evidence from Spanish regional
new markets and encourage business. The policymakers of a country exports, they find the importance of logistics for the analysis of trade
can also benefit from a country-based logistics performance evaluation flows. Liu, Li, and Huang (2006) find a unidirectional causality from
by understanding the relative position of their country. A guide can be logistics to economic growth. Carruthers, Bajpai, and Hummels (2003)
developed to decide which specific logistics function to focus on. highlight that the economic growth of many countries, such as Singa-
Although the logistics functions are mostly performed by private pore and Hong Kong, has been partly due to their transport invest-
service providers for private traders and owners of goods, if country- ments. Chen and Novy (2011) underline that the modern logistics in-
based improvements can be realized, companies with access to high- dustry plays an important role in national economic development.
quality and cost-efficient logistical capabilities can perform better. For These researchers use econometric methods to establish the relation-
instance, high-quality roads and certain customs clearance will de- ship between logistics and gross domestic product (GDP). Demurger
crease delivery time and its variance. The development of the logistics (2001) provides empirical evidence of the links between infrastructure
sector is expected to have a positive impact on increasing production, investment and economic growth in China. Esfahani and Ramirez
consumption and trade, thus stimulating the economic growth. (2003) develop a structural growth model of infrastructure and output
(Barnister & Berechman, 2001; Berechman, Ozmen, & Ozbay, 2006; growth. Their model indicates that infrastructure services greatly


Corresponding author.
E-mail addresses: kabak@itu.edu.tr (Ö. Kabak), fulengin@sabanciuniv.edu (F. Ülengin), sonsel@dogus.edu.tr (Ş. Önsel Ekici).

https://doi.org/10.1016/j.retrec.2018.05.007
Received 5 January 2018; Received in revised form 22 May 2018; Accepted 22 May 2018
0739-8859/ © 2018 Elsevier Ltd. All rights reserved.

Please cite this article as: Kabak, Ö., Research in Transportation Economics (2018), https://doi.org/10.1016/j.retrec.2018.05.007
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

contribute to GDP and that this contribution more than offsets its cost. The proposed methodology is applied to analyse the interaction
Fan and Chan-Kang (2008) use an econometric model to estimate the between logistics and export at a country level. LPI indicators and
impact of road investments on overall economic growth. export levels of countries are used to select and evaluate scenarios.
As the literature shows, the impact of logistics on growth and in- Furthermore, Turkey is selected as an example to analyse the impact
ternational trade is generally analysed from the transport infrastructure of the logistics indicators on international trade by using the proposed
perspective. However, logistics should be investigated from a wider scenario analysis approach. One of the basic reasons for selecting
spectrum. Indeed, the logistics performance index (LPI) is a very com- Turkey specifically is that Turkey, an important logistics centre in
monly used index for evaluating the logistics performance of 155 Europe that exhibits high trade values with its regional partners, has a
countries based on six areas: customs, infrastructure, international large population, diversified economy and strategic geographical lo-
shipments, logistics quality and competence, tracking and tracing, and cation. It is considered as a critical actor in the trade between Europe,
timeliness (Arvis et al., 2012; Rantasila & Ojala, 2012). The World Bank the Commonwealth of Independent States (CIS) and the Middle East
and the Turku School of Economics have administered the index every (ITF, 2015). Subsequently, we have also applied the proposed scenario
two years since 2007, rating countries based on a scale of 1 (worst) to 5 approach to selected four countries, namely Burundi, Zimbabwe,
(best) (Arvis et al., 2012). Therefore, one of the significant contribu- Brazil and Portugal, each having a different export level, to analyse
tions of this study is to analyse logistics performance based on this their logistics performance in relation to international trade. Ac-
wider perspective. cording to the classification of countries that is explained in Section
Although deterministic forecasting models are widely used to ob- 3.1. (See Table 2), Burundi is an example of countries that have very
serve the impact of logistics on the improvement of trade or economic low export level. Similarly, Zimbabwe, Brazil and Portugal are ex-
development, the basic drawbacks of such models are that they typi- amples of countries that have low, medium and high export levels,
cally provide a single prediction and that they do not include qualita- respectively. Policy suggestions are developed for these countries to
tive systemic change (Tietje, 2005). Indeed, particularly for long-term underline which logistics indicators should be focused on to improve
planning, backcasting, scenario analysis and foresight techniques are their export level.
accepted as suitable for providing information to logistics and trans- The main contributions of this study are twofold. Initially, we pro-
portation decision makers in highly uncertain future conditions pose a new scenario analysis methodology that uses objective in-
(Schuckmann, Gnatzay, Darkow, & Heiko, 2012). formation rather than expert opinions. Subsequently, we apply this
Backcasting is an approach for exploring long-term scenarios to methodology to investigate the interaction between logistics perfor-
reach a desirable future. It first creates a desirable future vision and mance and export as well as to develop policy suggestions for the se-
then looks back at how this desirable future can be achieved (Quist & lected countries.
Vergragt, 2006). It provides an approach that is the opposite of fore- The relative advantages of our proposed method can be listed as
casting, which aims to predict the future based on current conditions. below:
However, because of the high variability in the set of identified back-
casts, no specific policy recommendation can be made, and the inter- • Different from the current literature, our method analyses the in-
relationships among the variables cannot be easily observed. Moreover, teraction between logistics performance and economic development
as underlined by Wangel (2011), few backcasting studies include the from a wider spectrum. It does not focus on just one perspective at a
actors or governance. time (like transport infrastructure) but investigates the logistics
Foresight, on the other hand, is an important part of participatory performance depending on six different components of LPI.
strategy process helping organisations to make effective long-term • In the literature, the interaction between logistics performance and
planning under uncertainty. However, the basic drawback of foresight economic development is generally investigated with deterministic
processes is that they do not take into account certain complication models that use past data. However, future can be totally different
factors, such as non-linearities and thresholds, in their attempts to from the past, making all the models depending on past data useless
project the future. Scenario analysis, on the other hand, is a well-re- in the effort to create road maps for the future. Hence, our proposed
cognized method in the planning process (Önsel Sahin, Ülengin, & method uses scenario analysis to deal with uncertainty about the
Ülengin, 2004). Scenarios encourage participants to think through the future.
consequences of decisions and actions especially when dealing with • Scenario analysis methods mostly depend on subjective expert jud-
the macro environment. Scenario analysis focuses the attention of the gement that makes them hard to create and analyse. Our method
decision makers on a set of different descriptions of the future, which proposes to use objective information (LPI data and export levels) to
are explicitly designed to be feasible but are not necessarily the most constitute a sufficient number of plausible and consistent but still
likely. different scenarios.
In this study, a new scenario analysis approach is proposed. An • Although there are a number of quantitative scenario analysis
efficient scenario analysis should be composed of a small number of methods such as Cross-Impact Analysis or Trend Impact Analysis,
significantly different, reliable and efficient scenarios (Tietje, 2005). they combine historical trends with expert judgments by using some
However, in general, scenario analysis is highly subjective in nature. analytical methods. However, using historical data makes these
Because scenarios are generated based on expert opinions, the basic methods similar to deterministic models. Our model also uses past
motivation of this study is to use objective information rather than data about the countries. But, the rationale of using these data is not
subjective expert judgments during the scenario generation process. to make projections about future but only to generate a set of sce-
Additionally, a mathematical program is proposed to select sig- narios.
nificantly different, consistent and small number of efficient scenarios.
Finally, on the basis of the selected scenarios, this research proposes a Section 2 underlines the framework of the proposed methodology,
new approach that is developed to evaluate the performance of stra- and section 3 presents the results of the application of the proposed
tegies across a series of scenarios. Two measures are suggested to methodology and provides examples from five selected countries with
identify the most important variables to focus on for a given state (i.e. each having a different level of export, namely Turkey, Burundi, Zim-
scenario). As opposed to a correlation analysis where only the sum of babwe, Brazil and Portugal. The interpretations of the results are given
the independent variables is regarded, the proposed approach con- to highlight the key areas of logistics that should be improved in the
siders different combination effects of the variables and hence uses first run in each selected country to obtain the highest increase in their
information that is disregarded in traditional linear correlation export level. Finally, conclusions and suggestions are presented in
approaches. section 4.

2
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

2. Framework of the proposed methodology socio-economic development and technological change (Pittock, 2009).
Cuenot, Fulton, and Staub (2012) have used the International En-
The proposed methodology aims to analyse the interactions be- ergy Agency (IEA)'s set of travel projections to 2050 (IEA, 2008), in-
tween the variables in a system for strategic planning by using a sce- cluding a Baseline and ‘Modal Shift’ scenario, in order to estimate the
nario analysis-based approach. In this section, we first state the reason potential impact of reducing car and air travel by 2050, relative to the
behind the use of scenario analysis and then present our new approach Baseline in that year. The BLUE scenarios of the IEA aim to reduce
for scenario analysis. greenhouse gas emissions in the energy sector by reducing individual
motorized modes of transportation and aircraft use. Cuenot et al.
2.1. Why use scenario analysis? (2012) added the capability to estimate energy use and greenhouse gas
emissions from modal shifting at the national/regional and global level.
Scenario analysis provides an important tool in the strategic plan- However, as they also mention, their study requires a more detailed
ning process. Scenarios are not forecasts, but they give an internally policy-oriented analysis at the national/global level. Their study also
consistent view of what the future may be. Scenario analysis has fails to provide a broader review of the problem taking into account
emerged as a useful tool for long-term planning when the future is socio-economic, demographic and technological variables with a
perceived as being subject to a high degree of uncertainty and com- system approach.
plexity (Nguyen & Dunn, 2009; Tietje, 2005). Ülengin, Kabak, Önsel, Ülengin, and Aktaş (2010) discuss a decision
Scenario analysis technique is a well-known tool for strategic support framework that guides policymakers for analysing the effects of
planning where uncertain situations for complex systems have to be transportation policies on the environment, society, economy and en-
analysed. Both qualitative and quantitative approaches are used in ergy. A three-stage problem structuring model is developed for this
different areas analysed in the literature ranging from simplistic to purpose. Initially, experts' opinions are structured by using a cognitive
complex (Amer, Daim, & Jetter, 2013). The most popular three quan- map to determine the relationships between transportation and en-
titative scenario development methodologies used in the literature are vironmental concepts. Then a structural equation model (SEM) is con-
Interactive Cross-Impact Simulation, Interactive Future Simulations and structed, based on the cognitive map, to quantify the relations among
Trend Impact Analysis methodologies. Interactive Cross-Impact Simu- external transportation and environmental factors. Finally, the results
lation methodology was developed by Enzer at the Centre for Futures of the SEM model are used to evaluate the consequences of possible
Research, University of Southern California, and explores alternative policies by using scenario analysis.
futures as part of an iterative procedure (Bradfield, Wright, Burt, In fact, logistics and export is a very complex issue having different
Cairns, & Van Der Heijden, 2005). Interactive Future Simulations stakeholders, interacting with government objectives, institutions and
technique was developed by the Battelle Memorial Institute in the socio-economic objectives as well as technological objectives.
1970s. Different from Interactive Cross-Impact Simulation, it does not Therefore, it is not possible to obtain reliable forecasting for such a
use Monte Carlo simulation and does not require an independent complex system. Hence, scenario analysis is selected as a suitable
forecast of the key indicators or variables (Huss & Honton, 1987). As method of projection for this study.
the third well-known quantitative scenario analysis method, the Trend There are several critics that question the predictive power (Wright,
Impact Analysis has been used since the 1970s (Chermack, Lynham, & Cairns, & Goodwin, 2009), credibility, consistency and legitimacy of
Ruona, 2001). It is a combination of statistical extrapolations with scenario planning (Duckett et al., 2017). In fact, the internal con-
probabilities and combines historical trends with judgment about the sistency of the scenarios, which refers to the way in which levels of
probabilities and impacts of selected future events. However, all of factors in the scenario hold together depending on their plausibility, is
them combine historical trends with expert judgments by using some of primary importance. Indeed, long-term planning necessitates an
analytical methods. Fuzzy cognitive maps, which is another popular understanding of multidisciplinary connections among several factors,
scenario development approach combines qualitative storylines with such as demographic, technological, economic and political develop-
quantitative models. It can be concluded that, although there are many ments. However, the human mind is limited in mentally processing
quantitative scenario analysis methodologies in the literature, most of these interdependencies (Weimer-Jehle, 2006). Therefore, different
them are based on the combination of qualitative and quantitative scenario generation techniques are developed that consist of phases of
techniques in which subjectivity (expert judgement) has a great role. problem analysis, system analysis and synthesis. Problem analysis helps
For example, Walz et al. (2007) used a participatory scenario analysis the related experts and stakeholders to have a common understanding
technique for regional planning including landscape and economic of the problem at hand. System analysis defines the problem as a set of
changes, climate change and changes in the region's popularity where interrelated subsystems, identifying the relevant external influences on
scenarios were defined based on qualitative local knowledge. the investigated problem. Brainstorming, brain-writing, the Delphi
Bryan, Crossman, King, and Meyer (2011) have introduced a technique etc. can be used for these two phases. Finally, the process of
‘landscape futures analysis’ method which combines linear program- synthesis is used to examine the interdependencies among the influ-
ming optimisation with scenario analysis in order to quantify the en- encing factors and to develop scenarios. For this last stage, two different
vironmental, economic and social impacts to identify specific geo- groups of methods are generally proposed: Non-Bayesian methods (e.g.
graphic locations in the landscape for six natural resource management Morphological Analysis, Battelle Approach and Field Anomaly Relaxa-
actions. They developed scenarios with the participation of the stake- tion) and Bayesian Methods (e.g. Cross-Impact Analysis and Goal Pro-
holders and used a linear programming-based approach to develop gramming) (for a detailed overview of these techniques, see Nguyen
landscape futures integrating a variety of detailed spatial information and Dunn (2009), Bishop, Hines, and Collins (2007), and Amer et al.
and models to see the results of several policies under the developed (2013)).
scenarios. However, the scenario development process is subjective, Developing a small number of scenarios that represent possible
that is, based on expert opinions. states of a system is another appealing alternative. Scenario analysis
There is also a wide use of scenario analysis in transportation and provides an important tool in the strategic planning process. A scenario
logistics. In order to provide policy-relevant advice on the consequences is a plausible description of some future state with no statement of
of human-induced climate change in the twenty-first century, the probability. Scenarios are alternative pictures of how the future may
Intergovernmental Panel on Climate Change (IPCC) developed a series develop. They are used to highlight the consequences and thus provide
of scenarios of greenhouse gases and sulphate aerosol emissions up to a basis for policies that may influence future developments or help
the year 2100. The IPCC report accepts that future emissions are the governments to address future issues. Projections are sets of future
product of complex interacting systems driven by population change, conditions based on different scenarios.

3
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

A scenario selection method that yields consistent scenarios and that number of scenarios is generally desirable (Tietje, 2005).
supports the quality of scenario analysis is desirable. Indeed, incon- In the literature, the consistency is based on a specific scale, namely
sistent scenarios do not draw a realistic image of the future. Therefore, the consistency indicator. However, the consistency rating of all pairs of
scenario selection should take into consideration the following rules impact variables is performed by the experts. Subsequently, these rat-
(Tietje, 2005): ings, which are summarised in the consistency matrix, are formed, and
Significantly different scenarios: The decision maker is interested in a different indicators, such as overall consistency, multiplicative con-
set of principally possible cases, and small differences between sce- sistency, the number of inconsistencies and the minimum consistency
narios are not very relevant. level of a scenario, can be calculated (for the calculation details and
A small number of scenarios: The main reason for having a small characteristics of these indicators, see Tietje, 2005).
number of scenarios is that decision makers can hardly compare nu- In this research, instead of relying on experts, the consistency of a
merous qualitatively different scenarios and a large number of scenarios scenario as a whole is estimated by assessing the consistency of the
may indicate a large redundancy. levels of all pairs of variables. The frequency of the levels of variables
A reliable set of scenarios: Different scenario analysts should arrive at generated from past data is used to determine the consistency. The
the same results when they use different scenarios based on the same consistency of the levels of a pair of variables is calculated by the fol-
scenario selection procedure. lowing formula:
Efficient scenarios: These are the most consistent scenarios within a
group of similar scenarios because they are the most relevant re-
mj
⎛ f (yimi , yjmj ) f (yimi , yjmj ) ⎞
presentatives (note that a consistent scenario is not necessarily effi- c (yimi , yj ) = max ⎜⎜ ; ⎟
f (yimi )
cient). ⎝ ( )
f yjmi ⎟
⎠ (1)
In this research, taking into consideration the aforementioned rules,
a new approach is proposed for scenario selection. where i = 1, …,n are the impact variables, and m is an index for lin-
guistic terms to state the levels of the variables;
m
2.2. Proposed scenario selection procedure f (yimi , yj j ) is the number of scenarios where the level of variable i is
mi and the level of variable j is mj ; and f (yimi ) is the number of scenarios
A scenario describes a possible future state of a system through where the level of variable i is mi .
impact variables. In order to generate a number of scenarios, for each of We have provided a hypothetical example in the Appendix to il-
the n impact variables yi, i = 1, …,n, possible different levels are de- lustrate the proposed scenario selection procedure. Please see that ex-
fined, and the combinatorial number of scenarios is constructed by ample for calculation of consistency level.
taking the product of all of the different levels. Linguistic terms can be After calculating the consistencies in a pairwise manner, the con-
used to define the levels of variables. sistency of a scenario Sk = (y1m1 , y2m2 , …, ynmn ) is specified based on the
As proposed by Tietje (2005), the most important characteristic of consistency of the levels of a pair of variables, as given in Eq. (2):
an efficient scenario analysis is to work with a small set of consistent mj
C (Sk ) = min c (yimi , yj )
and different scenarios that are reliable and efficient. For this reason, it j = 1, … n,
is necessary to define the consistency and the neighbourhood of sce- i = 2, … n,
i>j (2)
narios prior to proposing a mathematical program to select the sce-
narios. The selected scenarios are evaluated to identify the important In the literature, additive and multiplicative operators have been
variables to focus on. Fig. 1 shows the main steps of the proposed used for similar purposes (please see examples in Tietje (2005)). The
methodology. rationale underlying our use of the minimum operator is to eliminate
those scenarios with the weakest possibility of occurrence in any pair of
2.2.1. Consistency variables within the scenario. Please see example of calculation in
The key issue in projecting the future on the basis of a scenario is the Appendix.
consistency of the scenario (Pittock, 2009). The consistency of a sce-
nario refers to the compatibility of combined variations of various
impact factors and is estimated by assessing the consistency of the le- 2.2.2. Neighbourhood
vels of all pairs of impact factors (Tietje, 2005). The consistency ana- To have significantly different and efficient scenarios, the closeness
lysis is the core part of a formative scenario analysis because incon- of the scenarios must be considered. We used the neighbourhood de-
sistent scenarios do not draw any realistic image of the future. A finition of scenarios based on their closeness to each other to ensure
scenario is taken as a set of system variables, each of which is allowed that neighbour scenarios are not selected in the final set of scenarios.
to take only a small number of different levels. However, the resulting To define the neighbourhood relation, the following general rules
combinatorial set of scenarios may be very large. Therefore, a scenario are proposed:
selection method that yields a consistent, reliable, different and small Neighbourhood rules:

Fig. 1. Main steps of the proposed methodology.

4
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

1) Two scenarios differ at most in the Δ variables; Max ∑k C (Sk ) xk


2) The total linguistic variable difference in all variables is at most δ; subject to
3) The linguistic variable difference in variable i is at most δi; xk + ∑l Nkl xl ≤ 1 ∀k
xk ∈ {0, 1} ∀k (4)
where Δ, δ, and δi (i = 1, …,n) are parameters determined according
to the problem under consideration. The first rule defines the neigh-
where C (Sk ) is the consistency of scenario Sk;
bourhood based on the number of different variable levels. In fact, a
Nkl is a binary parameter to indicate whether scenario k and l are
similar rule is defined in Tietje (2005) — two scenarios are a neighbour,
neighbours; and xk is a binary variable to identify whether scenario k is
if and only if the two scenarios differ only in one variable. We have
selected.
generalized this rule by defining Δ to be used for the cases of a high
The binary program in Eq. (4) is designed to select scenarios that
number of variables. For instance, if there are 10 variables, the differ-
have the basic properties of a well-organized set of scenarios, i.e. being
ence in two or three variables can be acceptable to declare two sce-
consistent, being significantly different, being efficient, having a small
narios as neighbour.
number of scenarios and having a reliable set of scenarios. The objec-
The second and third rules are based on the distance between the
tive function maximizes the total consistency of the selected scenarios
variable levels. The second rule considers the total difference in the
to ensure consistency. Significantly different and efficient scenarios are
variable levels of all variables in consideration. We decided to use this
gathered by the first constraint, in which if scenario k is selected, none
rule due to the fact that although there might be an acceptable differ-
of its neighbours can be selected. The reliability of the scenarios can be
ence in the variables according to the first rule, if the range of the
ensured by a robustness analysis after solving the model.
difference is high (for instance, very high versus very low), then it will
The proposed binary program is a kind of a set packing problem and
not be realistic to define the related scenarios as ‘neighbour’.
thus an NP-complete problem. Therefore, if the number of scenarios
The third rule considers the difference between particular variables
increases due to an increase in the number of variables or levels of
of interest. This rule can be used to give more importance to specific
variables, the solution time can increase considerably. To tackle this
variables in the system. For instance, in our application (Section 3), the
issue, it is suggested to include the scenarios which have consistency
‘export’ is the variable of interest; therefore, a special rule is dedicated
level higher than a threshold. It is noticed that even upon determining
to this variable in order to make a better discrimination among the
the threshold level as 0 (i.e. only the scenarios with non-zero con-
scenarios.
sistency are considered), the number of candidate scenarios decreases
According to the above given rules, the distance between the sce-
dramatically.
narios is calculated as follows:

⎛ n ⎧ 1 if yi (Sk ) ≠ yi (Sl ) 2.2.4. Scenario evaluation


∑ n
⎜ i=1 ⎨
⎩0 otherwise ∑ yi (Sk ) − yi (Sl )
d (Sk , Sl ) = Max ⎜ , i=1 , Despite its advantages, one element of scenario planning that has
Δ δ been underdeveloped is the evaluation of the performance of strategies

⎜ across a range of scenarios (Goodwin & Wright, 2001). As Chermack

(2004) has stated, although scenario planning has been applied in or-
⎞ ganisations for years, little or no guiding theory for implementation or

⎛ yi (Sk ) − yi (Sl ) , i = 1, …n ⎞ means for sound evaluation has been given. The absenteeism of eva-

δ

⎟ luation component from the literature of scenario planning is one of the
⎝ i ⎠⎟
⎟ most important drawbacks of scenario analysis. Once the scenarios have
⎠ (3)
been generated, they can be used in the evaluation of proposed stra-
tegies and in the selection of the most appropriate strategy. One pos-
A scenario Sk is a neighbour of scenario Sl if d (Sk , Sl ) ≤ 1. Please see
sible method is to put ticks and crosses in the matrix, where the ticks
examples in the Appendix.
represent the positive performance of a strategy under a given scenario
and the crosses represent negative performance. The strategy that has
the least negative performance may be judged as the most robust op-
2.2.3. Scenario selection
tion. However, this evaluation will need to be performed through
After the consistency assessment, the set of all plausible scenarios
multiple objectives. The integration of scenario planning and multi-
that match specific criteria related to the maximum number of incon-
criteria decision analysis (MCDA) offers the possibility of specifying the
sistencies or a minimum consistency level can be obtained. However, it
strategic options that are necessary to focus on by informing within and
is necessary to underline that, even after this filtering, the number of
across scenario comparisons of options. Ram, Montibeller, and Morton
scenarios left may be in the hundreds or, in some cases, even in the
(2011) measure how each strategic option performs under a given
thousands. Therefore, it is necessary to use an additional procedure to
scenario based on a weighted sum of the preference of each option
find representative scenarios that are distinct from each other. One
against each criterion. Goodwin and Wright (2001, 2014) use Swing
possible method is the use of cluster analysis. However, this approach is
Weight to calculate the weights of the criteria. Srdjevic, Medeiros, and
criticized because it does not provide a good proposal for which sce-
Faria (2004) rank the scenarios with TOPSIS and Compromise Pro-
nario of a cluster to take as a representative. Tietje (2005) proposes
gramming after criteria weights have been derived by the entropy
three different procedures to obtain a set of final scenarios: local effi-
method. Diakoulaki and Karangelis (2007) use multi-criteria and cost-
ciency, distance-to-selected and max-min selection (for details, see
benefit analysis to calculate the economic, technical and environmental
Tietje, 2005).
performances of the examined scenarios.
The objective of using a mathematical model in the scenario selec-
In this study, in order to evaluate the performance of strategies
tion phase of our proposed method is to eliminate the heuristic per-
across a series of scenarios, in the last phase of the proposed scenario
spective that is used in Tietje (2005) and thus provide an optimum set
analysis, we develop measures to identify the most important variables
of scenarios.
to focus on for a given state (i.e. scenario). For this purpose, initially, a
In this study, we propose to use the following binary program to
similarity measure is defined to determine which selected scenario is
select a small set of efficient and consistent scenarios:
more appropriate for a given state.

5
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

n 2) Customs: the efficiency of customs and border management clear-


∑i = 1 yi (Sg ) − yi (Sk ) 2
σ (Sg , Sk ) = 1 − ance. Simplifying documentation for imports and exports,
n( m − 1)2 (5) prompting initiatives to bring border agencies together and to create
where m is the total number of linguistic terms, n is the total number of a single window for trade have always been on the trade facilitation
variables and σ (Sg , Sk ) is a similarity measure of a given state g and agenda. Efficient clearance procedures at the border are critical to
selected scenario k. σ (Sg , Sk ) takes values between 0 and 1 and the eliminate avoidable delays and improve supply chain predictability.
greater the σ (Sg , Sk ) value, the greater is the similarity between g and k. 3) International shipments: the ease of arranging competitively priced
yi (Sg ) − yi (Sk ) is the distance between linguistic variables of state g and shipments. Competitive pricing with the sub-dimensions of port
scenario k for variable i. By this measure, we can determine the most fees, airport fees, railroad fees, storage/loading fees and agent's fees
similar selected scenario for a given state and interpret some sugges- will have an important potential in decreasing trade costs.
tions accordingly. Please see the Appendix for example of similarity 4) Logistics quality and competence: the competence and quality of lo-
measure calculation and developing suggestions. gistics services. Service quality differs substantially at similar levels
The similarity measure finds the nearest scenario for a given state of perceived infrastructure quality. This indicates that even high-
and, thus, only some general suggestions can be derived by comparing quality hard infrastructure cannot substitute or replace operational
the current state with the nearest scenario. However, this information is excellence, which is based on the professional skills of service pro-
insufficient to understand how and by which means to improve this viders, well-functioning soft infrastructure and smooth business and
third variable. administrative processes. Higher service quality is driving logistics
In order to provide more detailed policy suggestions which will performance in emerging and richer economies. Yet, the develop-
especially take into account the variable of interest in the system, we ment of services, as in third- or fourth-party logistics, is a complex
propose to combine the selected scenarios for the levels of the variable policy agenda because the provision of these more advanced ser-
of interest and compare the given state and combined scenarios. In this vices cannot be created from scratch or developed purely domes-
respect, for each level of the variable of interest, the weighted average tically.
of the other variables is calculated. To that end, the consistencies of 5) Tracking and tracing: the ability to track and trace consignments
related scenarios are considered as their weights to give more emphasis concerns a process of determining the current and past locations
to the most consistent scenarios: (and other information) of a unique item or property. Smart trans-
port systems play an important role in monitoring the flow of ma-
∑k y (Sk ) = mj yi (Sk ) C (Sk )
μ (mj , yi ) =
j terials and, hence increase the competitive advantage by instantly
∑k y (Sk ) = mj C (Sk )
j (6) monitoring the speed and routes of vehicles, effective fleet man-
agement, increasing customer satisfaction by providing customers
where j is the variable of interest, mj is the level of variable j and yj is with information, ensuring flexibility through the monitoring of
another variable. μ (mj , yi ) is defined as the weighted average level of cargo on the vehicles and directing new cargo to the closest vehicles
variable j when the level of variable j is mj . Then, values of given state at the time of order.
and μ (mj , yi ) values are compared for the target level of the variable of 6) Timeliness: the frequency with which shipments reach consignees
interest to develop policy suggestions. Section 3.3 gives the details of within scheduled or expected delivery times. A useful outcome
using the weighted average method for scenario evaluation. measure of logistics performance is the time taken to complete trade
transactions. Logistics performance depends on the availability to
3. Application of the proposed methodology to analyse the traders of reliable supply chains and predictable service delivery.
interaction between logistics and international trade Global supply chains are becoming more complex, and the safety,
social, environmental and other regulations affecting traders and
In this research, the proposed scenario-based methodology is ap- operators are becoming more demanding.
plied to the analysis of logistics performance indicators to improve the
export level of a country. In fact, the six LPI indicators explained above can also be grouped
As is underlined in the introduction section, most of the studies that into two main categories: the topics indicating the main inputs to the
analyse the impact of logistics on international trade especially focus on supply chain, namely customs, infrastructure and services, necessitate
transport infrastructure. However, logistics should be investigated from policy regulation. On the other hand, time and reliability: timeliness,
a wider perspective. For this reason, in this study, the LPI is used to international shipments and tracking and tracing correspond to the
represent countries’ logistics performance. The variables considered in supply chain performance outcomes (Arvis et al., 2016).
the scenario analysis are presented in Fig. 2.
Logistics performance is central to the economic growth and com- 3.1. Data collection
petitiveness of countries, and the logistics sector is now recognized as
one of the core pillars of economic development. Efficient logistics As stated before, one of the important contributions of the proposed
connects firms to domestic and international markets through reliable scenario analysis procedure is the use of objective data instead of expert
supply chain networks (Arvis et al., 2016). The World Bank's LPI report opinions. For the application of the procedure to the interaction be-
analyses countries based on six components: tween logistics and exports, the logistics performance indicators and
export levels of 154 countries in 2007, 2010, 2012 and 2014 are used.
1) Infrastructure: the quality of trade and transport infrastructure. The related LPI indicator data are provided from World Bank's webpage
Logistics is not limited to transportation or trade facilitation. It is (http://lpi.worldbank.org/).
part of a broader agenda that includes services, development of To specify the level of the variables, a 5-term linguistic set, defined
facilities, infrastructure and spatial planning. Infrastructure devel- as very high (VH), high (H), medium (M), low (L) and very low (VL), is
opment continues to play an important role in assuring basic con- used. The determination of the number of linguistic terms in our ap-
nectivity and access to gateways for most developing countries. plication is rather intuitive. In fact, according to Chen and Hwang
However, hard infrastructure is not enough. There also needs to be a (1992), very few terms will not provide detailed information, while
soft component, involving regulatory reform in service markets such excess of terms may make the system very complex to be practical.
as transport, logistics and telecommunications. Efficient manage- Additionally, LPI itself is measured in 1–5 scale. Hence, in this study, a
ment and information technology (IT) solutions in both the private 5-term linguistic set is preferred.
and public sectors are vital to provide high-quality logistics. The LPI indicators are measured on a scale of 1 (worst) to 5 (best),

6
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Fig. 2. Variables in the logistics and exports interaction.

Table 1 are kept for further analysis.


Intervals specified for levelling LPI values. Then, the neighbourhood relations are determined considering
Level Linguistic term LPI values Midpoint values Δ = 2, δ = 3 and δ7 = 1. By this way, if two scenarios differ in more
than 2 variables (i.e. Δ = 2), the total variable difference is more than 3
VL Very Low x<2 1.50 (i.e. δ = 3) and difference in the export variable is more than 1
L Low 2≤x < 2.67 2.33
(δ7 = 1), then we conclude that these scenarios are not neighbour to
M Medium 2.67 ≤ x < 3.33 3.00
H High 3.33 ≤ x < 4.00 3.67
each other. We have identified δ7 = 1 because the export is the variable
VH Very High 4.00 ≤ x < 5.00 4.50 of interest in our application. By using Eq. (3), 67,608 pairs are found to
be neighbours.
Finally, the consistencies of the selected 1395 scenarios and their
where values lower than 2 and higher than 4 are considered very low neighbourhood relation are used in the proposed binary program given
and very high, respectively. Therefore, the values corresponding to the in Eq. (4). As a result, 16 scenarios are selected, as shown in Table 3.
LPI indicators are divided into five categories, as shown in Table 1, These 16 scenarios are the optimum set of scenarios that are consistent,
where values lower than 2 are classified as VL, higher than or equal to 4 reliable, efficient and different from each other in our problem.
as VH and the values in between are divided into three equal intervals.
The export variable is measured by exports per capita to avoid any 3.3. Scenario evaluation and policy suggestions
bias that may occur due to the population size of the country. To take
the effect of inflation into account, the current series of World bank is In the final step of the methodology, initially, the final set of sce-
used. The data (exports of goods and services (BoP, current US$) and narios is evaluated and policy suggestions are developed for the se-
Population, total) provided by the World Bank database (http://data. lected countries, based on the scenario evaluations.
worldbank.org/) are used to construct the export data. When the data
are analysed, it is observed that the data are highly skewed (skew- 3.3.1. Scenario evaluations
ness = 5.74). For this reason, a logarithmic transformation is applied to Scenarios are evaluated to show the interaction between logistics
the data to obtain a more symmetric distribution (skewness drops to performance and export. For this, the average level of the logistics
−0.036). Finally, the categorisation is made using the final transformed variables for each particular export level is calculated, and the weighted
data by dividing the entire range into five equal intervals (see Table 2.). sum of the logistics variables is specified according to Eq. (6). The
midpoint value of each linguistic term (see Table 1) is used to represent
3.2. Scenario selection it as a numerical value. The consistency of each scenario is considered
to be its importance weight of the selected scenarios for each export
Initially, all possible scenarios are generated, considering the com- level. The results are presented in Table 4, which shows that, for ex-
binations of levels of variables. As a result, 57 = 78,125 (i.e. 7 variables ample, due to the fact that scenarios 105, 138 and 938 have a VL export
and 5 linguistic terms) scenarios are identified, and their consistencies level, the weighted average of these three scenarios are calculated for
are calculated according to Eq. (1) and Eq. (2) given in Section 2.2.1. ‘international shipment’ as follows (note that in scenarios 105, 138 and
Scenarios that have a non-zero consistency (a total of 1395 scenarios) 938, the level of international shipment are L, VL and M, respectively):
μ (Export = VL, International Shipment )
Table 2 MP (‘L’) C (S105) + MP (‘VL’) C (S138) + MP (‘M ’) C (S938)
Intervals specified for levelling export data. = C (S105) + C (S138) + C (S938)
2.33 ⋅ 0.273 + 1.50 ⋅ 250 + 3 ⋅ 0.021
Level Linguistic term Export (current US$ per capita) = 0.273 + 0.250 + 0.021
= 1.98
VL Very Low < 77
L Low 77 ≤ x < 564 where MP(mi ) is the midpoint of linguistic term mi (such as L, VL or M)
M Medium 564 ≤ x < 4149
given in Table 1.
H High 4149 ≤ x < 30478
VH Very High ≤30478 According to values given in Table 4, there is a direct relationship
between the logistics indicators and exports. For example, when the

7
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table 3
Selected scenarios in the application.
Scenario ID Customs Infrastructure International Shipment Logistics quality and Competence Tracking and Tracing Timeliness Export Consistency

2 H H H H H VH H 0.533
7 VH VH H VH VH VH VH 0.462
8 L VL VL L L M L 0.450
27 L L M M M M M 0.365
45 VL L L VL L L M 0.343
105 L L L L VL L VL 0.273
138 VL VL VL VL VL VL VL 0.250
172 M M M M H H H 0.230
603 M L L M L H L 0.054
725 L L L L VL H H 0.042
858 M M M L M L L 0.029
882 M H H M M M VH 0.027
893 L M M L L M VH 0.022
938 VL VL M L M H VL 0.021
1161 L M L L H VH M 0.010
1382 H M H L L H M 0.009

export level is VH, then the customs, infrastructure, logistics quality Table 5
and competence, tracking and tracing and timeliness are VH (i.e. > 4.0) Similarity measures of selected countries’ current states and scenarios.
and level of international shipments is H (i.e. > 3.33) and when export Scenario ID Turkey Burundi Zimbabwe Brazil Portugal
level is VL, then level of all variables is very low (i.e. ≤2).
2 0.811* 0.457 0.457 0.701 0.866*
7 0.659 0.268 0.268 0.509 0.717
3.3.2. Policy suggestions for Turkey 8 0.537 0.836 0.836 0.673 0.465
After the evaluation of the scenarios for specific export levels, in- 27 0.733 0.750 0.789 0.866* 0.701
itially, Turkey is selected as an example to show how these values are 45 0.518 0.750 0.836* 0.673 0.482
used for policy development and then the methodology is applied to 105 0.509 0.866* 0.811 0.634 0.441
138 0.345 0.717 0.717 0.482 0.280
some other selected countries.
172 0.836* 0.599 0.622 0.836 0.836
Turkey is selected in this study as it is an important logistics centre 603 0.687 0.811 0.769 0.789 0.610
in Europe. Further, in 2009, the Turkish Exports Strategy for 2023 was 725 0.577 0.687 0.750 0.733 0.557
initiated by the Ministry of Economy and Turkish Exporters Assembly 858 0.687 0.769 0.733 0.750 0.634
(TSV, 2015). The main purpose of this strategy is to reach an export 882 0.733 0.509 0.547 0.733 0.811
893 0.634 0.599 0.673 0.750 0.659
volume of 500 billion dollars by 2023, the centenary anniversary of the 938 0.557 0.789 0.789 0.701 0.500
Turkish Republic, with an average increase in exports of 12% annually. 1161 0.733 0.659 0.687 0.811 0.673
In this respect, the exports strategy of Turkey establishes a production 1382 0.687 0.646 0.622 0.750 0.687
plan oriented towards export performance. Basic elements of this
strategy are shifting production from low-technology sectors to high *Indicates the most similar scenario(s) for the related country.
value-added areas and achieving new investments in high-tech sectors.
To reach these goals of the strategy, some crucial actions must be im- better than or equal to the corresponding variable levels. When com-
plemented, including the improvement of logistics facilities. pared to scenario 2, on the other hand, Turkey is underachieving in
Turkey is ranked 30th out of 160 countries in the LPI 2014. It is variables Customs, International Shipment and Timeliness.
above the average of the upper-middle-income economies to which it Second, the indicators that Turkey must focus on are investigated by
belongs. Although Turkey's LPI score showed a significant increase of comparing Turkey's current level of exports (M) with the H and VH
approximately 9% from 2010 to 2012, it showed almost no change levels of exports using the weighted averaging approach. Turkey's LPI
between 2012 and 2014, resulting in a slight decrease in its ranking 2014 values are presented in Table 7, and they are compared with the
during this period (Arvis et al., 2014). average scenario scores for the export level H and VH (see the last two
In order to analyse Turkey's position, initially, similarity measures rows of Table 7). The results show that Turkey should improve all lo-
of Turkey's current state (LPI 2014 scores given in Arvis et al. (2014), gistics indicators to achieve a VH level of exports but should particu-
see Table 6) and the scenarios are calculated based on Eq. (5). Ac- larly focus on customs, international shipment and timeliness for an H
cording to the similarity measures presented in Table 5, Turkey is most level of exports.
similar to scenarios 172 and 2. In both scenarios, export level is high Indeed, in the LPI report, the international shipment indicator is
(H), whereas Turkey's current export level is medium (M) (see Table 6). defined as ‘The ease of arranging competitively priced shipments’ (Arvis
This shows that Turkey has to focus on strategies to improve its un- et al., 2014). Turkey has a score of 3.18, and it ranks 43th out of 160
derachieved export situation. When compared to Scenario 172, which is countries. It is Turkey's lowest score among the 6 LPI indicators. One
the most similar one, Turkey's levels in the LPI related variables are basic method of achieving an improvement in international shipment

Table 4
Weighted average level of variables for the levels of export.
Export Level Related Scenarios Customs Infrastructure International Shipment Logistics quality and Competence Tracking and Tracing Timeliness

VL – Very Low 105, 138, 938 1.92 1.92 1.98 1.95 1.56 2.00
L – Low 8, 603, 858 2.44 1.67 1.67 2.40 2.37 3.03
M – Medium 27, 45, 1161, 1382 1.96 2.35 2.68 2.27 2.69 2.71
H – High 2, 172, 725 3.41 3.41 3.41 3.41 3.55 4.22
VH – Very High 7, 882, 893 4.33 4.39 3.64 4.33 4.33 4.36

8
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table 6
Turkey's most similar scenarios and current level of Turkey.
Scenario ID Customs Infrastructure International Shipment Logistics quality and Competence Tracking and Tracing Timeliness Export

2 H (4) H (4) H (4) H (4) H (4) VH (5) H (4)


172 M (3) M (3) M (3) M (3) H (4) H (4) H (4)
Turkey M (3) H (4) M (3) H (4) H (4) H (4) M (3)

can be realized through border procedures and time. Indeed, the LPI 105. Therefore, it is not possible to generate policy suggestions ac-
measures this factor based on several indicators. The import and export cording to similarity analysis.
time is a useful outcome measure of logistics performance that analyses On the other hand, for Zimbabwe that is selected as one of the ‘low’
the time taken to complete trade transactions. The lead time for port or export level countries, the most similar scenario is scenario 45.
airport supply chains, which is nearly twice as long in low-performance Zimbabwe is ranked 137th out of 160 countries. Its worst performance is
countries as it is in high-performance countries, should be improved. for Customs (ranked as 154th) and best is for Timeliness (ranked as
Geographical hurdles, and possibly internal transport markets, continue 104th) (Arvis et al., 2014). It was one of the top 10 low-income per-
to pose substantial difficulties in the country. It is necessary to reduce formers on the 2012 LPI (Arvis et al., 2012). Zimbabwe is under-
time across all dimensions of the border process and enact reforms to achieving in export compared to scenario 45, while it performs better or
focus on the prevalence of physical inspection, the proliferation of the same for the LPI-related variables. For Brazil, as an example of
procedures and red tape. Although the time taken to clear goods medium export level countries, the most similar scenario is scenario 27.
through customs is a relatively small fraction of the total import time, Brazil's is ranked 65th in 2014 LPI report. As a large country, Brazil has
this time increases when the goods are physically inspected. created high level interagency bodies to manage logistics system ef-
Export supply chains typically face fewer procedural burdens than fectively (Arvis et al., 2014). It was one of the top 10 upper middle-
imports, as evidenced by the shorter lead time for exports than for income performers on the 2012 LPI (Arvis et al., 2012). Brazil's scores
imports. Customs is not the only agency involved in border manage- are higher than or equal to the scores of Scenario 27. Finally, Portugal,
ment; collaboration among all border management agencies and the as an example of high-level export countries, has scores most similar to
introduction of modern approaches to regulatory compliance are Scenario 2. Portugal is 27th in the 2014 LPI report (Arvis et al., 2014).
especially important. Indicators of red tape also illustrate a lack of Its worst performance is for Timeliness (ranked as 35th) and best is for
coordination at the border and the burden it imposes on private logis- ‘Logistics Quality and Competence’ and ‘Tracking and Tracing’ (ranked
tics operators. Turkey must reduce the number of government agencies as 20th for both) (Arvis et al., 2014). Portugal is underachieving in
involved and the documentary requirements. Turkey must also improve Customs and Timeliness compared to Scenario 2. According to simi-
its connectivity to the market through the hierarchical hub-and-spoke larity analysis, it was difficult to derive policy suggestions by directly
network of international trade. Another problem related to the inter- looking at the most similar scenarios. It is required to have a detailed
national transportation of Turkey is the quotas and transit procedures analysis to specify the policy measures to focus on in order to reach the
that make the movement of goods possible with the payment of duties target level for the variable of interest (e.g. Export in our analysis).
and excessive control. Because of the insufficiency of the similarity measure to specify such
policies, in this study, the weighted averaging approach is developed.
The results of applying the weighted averaging approach to the
3.3.3. Policy suggestions for selected countries selected countries are given in Table 8. It can be seen for Burundi that it
Subsequently, using the proposed methodology, similar analyses has to focus on improving ‘Timeliness’ because the only improvement
can be conducted for other countries based on the scenario evaluation area with respect to the next export level belongs to this indicator.
results given in Table 4. For instance, we have considered four coun- Zimbabwe, on the other hand, has to focus especially on two LPI in-
tries, namely Burundi, Zimbabwe, Brazil and Portugal, each having a dicators, namely International Shipment and Tracking and Tracing in
different export level (i.e. VL, L, M and H, respectively) (see Tables 5 order to improve its export level to Medium. Brazil has to improve all of
and 8). its LPI values in order to upgrade to the next high export level but
First, we demonstrate the use of similarity measures for the selected should especially concentrate on Customs and Timeliness where a sig-
countries. As an example of countries that have very low export level, nificant difference occurs. Finally, Portugal has to improve all of its LPI
initially, Burundi is investigated. Burundi is ranked 107th out of 160 values but especially focus on Customs and Infrastructure.
countries. It is also in the top 10 low-income performers on the 2014 Note that this analysis recommends evolutionary policies to im-
LPI and is among the economies that have a statistically significant prove its current export level to next higher level. For instance, from
positive change in LPI score during 2012–2014 (Arvis et al., 2014). Low (L) to Medium (M) for Zimbabwe. Because it is not realistic to
Burundi's current state is found to be most similar to Scenario 105. expect a sudden jump from low to very high. Rather, the evaluation is
When Burundi's current state and scenario 105 are compared, all scores expected to be from low to medium at the first glance.
of Burundi are higher and/or equal to corresponding scores of Scenario

Table 7
Strategy proposal for Turkey.
Export Level Customs Infrastructure International Shipment Logistics quality and Competence Tracking and Tracing Timeliness

TURKEYc Export level – M 3.23 3.53 3.18 3.64 3.77 3.68


H – Highb 3.41 3.41 3.41 3.41 3.55 4.22
Difference to Export H level 0.17a −0,13 0.22a −0,23 −0,22 0.53a
VH – Very Highb 4.33 4.39 3.64 4.33 4.33 4.36
Difference to Export VH level 1.09a 0.86a 0.46a 0.69a 0.55a 0.67a

a
The variables in which Turkey needs improvement.
b
Weighted average levels presented in Table 4.
c
Source of LPI data: Arvis et al. (2014).

9
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table 8
Strategy proposal for the selected countries.
Export Level Customs Infrastructure International Shipment Logistics quality and Competence Tracking and Tracing Timeliness

b
Burundi VL 2.60 2.40 2.60 2.51 2.51 2.76
Difference to Export L level −0.16 −0.73 −0.93 −0.11 −0.14 0.27a
Zimbabweb L 1.89 2.25 2.25 2.50 2.22 2.93
Difference to Export M level 0.07a 0.10a 0.44a −0.22 0.47a −0.22
Brazilb M 2.48 2.93 2.80 3.05 3.03 3.39
Difference to Export H level 0.93a 0.48a 0.60a 0.36a 0.53a 0.83a
Portugalb H 3.26 3.37 3.43 3.71 3.71 3.87
Difference to Export VH level 1.06a 1.02a 0.21a 0.61a 0.61a 0.49a

a
The variables in which the country needs improvement.
b
Source of LPI data: Arvis et al. (2014).

4. Conclusions and further suggestions as the one proposed by Goodwin and Wright (2001, 2014) and Stewart,
French, and Rios (2013), can be added to the proposed scenario
This study makes two significant contributions. First, a novel sce- methodology in order to provide a formal strategy evaluation process.
nario-based methodology depending on objective information is pro- By this way, the synergies between scenario-planning and quantitative
posed, and second, this methodology is applied to investigate the in- decision modelling can also be exploited.
teraction between logistics performance and export in a wider The second contribution of this research is the application of the
perspective. proposed methodology to analyse the link between a country's logistics
The most important part of scenario analysis is generating a con- performance and its export performance. For this purpose, six LPI in-
sistent, reliable, different and small set of scenarios. In the literature, dicators and the export level of countries are used to find the interac-
consistency checks have been performed using expert judgments, which tions between them. In fact, the LPI Index provides a general view of the
are subjective in nature. In the proposed approach, the biases that may current state of the countries from the logistics perspective. It does not
occur due to the subjectivity of the scenario analyst are eliminated by provide a roadmap for the countries on their way to improve their lo-
using the past data. For generating a small number of scenarios, the gistics performance. Additionally, it does not give any information
literature suggests using a neighbourhood distance. However, calcu- about the interaction between logistic and export activities. This study
lating the size of differences between two levels is impossible with a highlights how to increase the level of export in a specific country by
number of variables that is on a nominal scale (which is the general improving particular logistics activities. The proposed methodology
case in the literature). To obtain a final set of scenarios, Tietje (2005) allows to develop different strategies for different countries depending
uses three different procedures: the local efficiency, distance-to-selected on their LPI and export characteristics.
and max-min selection procedures. Nevertheless, all these procedures Subsequently, 16 different and efficient scenarios are generated
depend on consistency scores, which, as stated above, are subjective in using the proposed methodology. After the identification of the scenario
nature. The present paper elaborates further Tietje (2005)'s procedures set, initially, Turkey was selected as a case study. The detailed analysis
to remedy the subjectivity. For this purpose, we propose to use a binary of the scenarios shows that Turkey should improve all logistics in-
program to select a small set of efficient and consistent scenarios. In- dicators to achieve a very high (VH) level of exports but should parti-
itially, different from the studies found in the literature, we calculate cularly focus on customs, international shipment and timeliness to
consistency with a completely objective method based on observed reach a high (H) level. Subsequently, in order to show the general ap-
frequencies of levels of variables. In addition, because we use an ordinal plicability of this methodology, four other countries, each having a
scale to generate scenarios, simply taking the difference between two different export level, are also investigated and policy measures to
levels of a variable is an easy and efficient method that we use in the improve the export level of each to the next higher level are specified.
neighbourhood calculation step. Subsequently, the research proposes a According to the weighted average level of variables for the levels of
method to determine the most important variables to focus on to im- export provided in Table 4, there is an obvious relationship between
prove the variable of interest in relation to its target value. This task is logistics variables and export. When the export level is high or very
performed by finding the scenarios that provide the target level for this high, the logistics performance of countries is more than 3.4. This
variable and obtaining the weighted average of the levels of the other shows that in order to increase export level of a country to high or very
variables. In this manner, it becomes possible to find the variables that high level, the performance of a country in logistics indicators must also
are important for improving a specific variable. be in high or very high level. In order to increase the export level from
The basic disadvantage of traditional quantitative scenario analysis very low to low, a country has to focus on customs (1.92 → 2.44), lo-
methodologies is their dependence on historical data extrapolation gistics quality and competence (1.95 → 2.40), tracking and tracing
without considering the effects of unprecedented future events. (1.56 → 2.37) and timeliness (2.00 → 3.03). While moving from low
However, our proposed method uses past data not to make projections export level to medium, infrastructure (1.67 → 2.35), international
about future but only to generate significantly different, reliable and shipment (1.67 → 2.68), and tracking and tracing (2.37 → 2.69) are the
efficient set of scenarios. three most important variables to focus on.
As stated by Amer et al. (2013) ‘both qualitative and quantitative One of the important properties of our proposed methodology is
approaches are complementary and strengthen each other when used that it enables to consider the relative priorities according to possible
together’. Therefore, our proposed quantitative scenario generation LPI and export configurations (i.e. scenarios). For instance, see our
technique that depends heavily on large-scale data can easily be com- analysis for Turkey in Table 7. If only, the LPI scores are considered,
bined with some qualitative scenario analysis method in order to con- Turkey should focus on international shipment, because it has the
duct an efficient analysis. However, when the past data are missing and lowest level for this indicator (highest distance to 4.0). But in fact, 4.0 is
solely expert judgments are employed for scenario development, it an unrealistic target for international shipment due to the fact that even
would not be effective to apply the proposed methodology. the countries having high and very high export level have an average of
As a further suggestion, if some strategies are generated based on 3.41 and 3.64, respectively. On the other hand, according to our
the most important variables, then a multi-attribute value model, such methodology, timeliness is suggested as the indicator that we have to

10
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

concentrate on at the first glance because the countries having high and it would be worthwhile to analyse and ascertain the extent to which this
very high export level have high performances in terms of timeliness interaction is bidirectional.
(4.22 and 4.36, respectively). As an additional further study, we aim to combine a previous study
In the literature, there is a debate regarding whether logistics and (Onsel Ekici, KabakÖ, & Ülengin, 2016) that analyses the relationship
economic growth have a bidirectional interaction (Nguyen & Tongzon, between global competitiveness and the logistics efficiency of a country
2010). This study shows that improvement in some of the logistics with the results of this paper. Thus, we will be able to analyse the effect
performance indicators has an important positive impact on the export of a country's global competitiveness on its export level through logis-
level of a country. Indeed, improving exports, in turn, is expected to tics performance. To achieve this goal, a Bayesian Network will be used
enhance logistics performance because the increased demand for goods to combine and analyse the results of the two different studies, both of
and services will necessitate investments and improvements in the lo- which analyse a country's logistics performance from different per-
gistics indicators (Lean et al., 2014). Therefore, as a further suggestion, spectives.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.retrec.2018.05.007.

Appendix B. A hypothetical example to illustrate the proposed scenario selection procedure

This Appendix presents examples of calculations presented for scenario selection procedure in Section 2. Suppose a hypothetical example where
three variables (y1, y2, y3) are measured in three levels (L - Low, M - Medium, H - High). The observed frequencies of the levels of variables are shown
in Table B1.

Table B1
Frequencies of the scenarios for variables.

y1 y2 y1 y3 y2 y3

L M H Total L M H Total L M H Total

L 3 7 4 14 L 13 1 0 14 L 5 4 3 12
M 6 9 1 16 M 3 12 1 16 M 7 12 3 22
H 3 6 11 20 H 0 5 15 20 H 4 2 10 16
Total 12 22 16 50 Total 16 18 16 50 Total 16 18 16

Consistency

In order to calculate the consistency of scenarios, initially, the consistency of the levels of a pair of variables are calculated using E. (1). For
example, according to the given data in Table B1, the consistency of ‘variable 1 is H, and variable 2 is M’, c (y1H , y2M ) is equal to 0.3, which is
calculated based on Eq. (1) as follows:

f (y1H , y2M ) f (y1H , y2M ) ⎞ 6 6


c (y1H , y2M ) = max ⎛⎜ ; ⎟ = max ⎛ ; ⎞ = 0.3
⎝ f (y1H ) f (y2M ) ⎠ ⎝ 20 22 ⎠

The consistencies calculated for the hypothetical example are shown in Table B2.

Table B2
Consistency of the levels of a pair of variables.

y1 y2 y1 y3 y2 y3

L M H L M H L M H

L 0.250 0.500 0.286 L 0.929 0.071 0.000 L 0.417 0.333 0.250


M 0.500 0.563 0.063 M 0.188 0.750 0.063 M 0.438 0.667 0.188
H 0.250 0.300 0.688 H 0.000 0.278 0.938 H 0.250 0.125 0.625

Consistencies of scenarios are calculated using Eq. (2). For instance; the consistency of scenario S20 , where (y1, y2, y3) = (H, L, M), is calculated as
follows:

C (S20) = C (y1H , y2L , y3M )


= Min (c (y1H , y2L ), c (y1H , y3M ), c (y2L , y3M ))
= Min (0.250, 0.278, 0.333)
= 0.250
The consistencies of all possible scenarios in the hypothetical example are given in Table B3.

11
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table B3
Consistencies of the scenarios.

Scenario ID (k) y1 y2 y3 c (y1m1 , y2m2 ) c (y1m1 , y3m3 ) c (y2m2 , y3m3 ) C (Sk )

1 L L L 0.250 0.929 0.417 0.250


2 L L M 0.250 0.071 0.333 0.071
3 L L H 0.250 0.000 0.250 0.000
4 L M L 0.500 0.929 0.438 0.438
5 L M M 0.500 0.071 0.667 0.071
6 L M H 0.500 0.000 0.188 0.000
7 L H L 0.286 0.929 0.250 0.250
8 L H M 0.286 0.071 0.125 0.071
9 L H H 0.286 0.000 0.625 0.000
10 M L L 0.500 0.188 0.417 0.188
11 M L M 0.500 0.750 0.333 0.333
12 M L H 0.500 0.063 0.250 0.063
13 M M L 0.563 0.188 0.438 0.188
14 M M M 0.563 0.750 0.667 0.563
15 M M H 0.563 0.063 0.188 0.063
16 M H L 0.063 0.188 0.250 0.063
17 M H M 0.063 0.750 0.125 0.063
18 M H H 0.063 0.063 0.625 0.063
19 H L L 0.250 0.000 0.417 0.000
20 H L M 0.250 0.278 0.333 0.250
21 H L H 0.250 0.938 0.250 0.250
22 H M L 0.300 0.000 0.438 0.000
23 H M M 0.300 0.278 0.667 0.278
24 H M H 0.300 0.938 0.188 0.188
25 H H L 0.688 0.000 0.250 0.000
26 H H M 0.688 0.278 0.125 0.125
27 H H H 0.688 0.938 0.625 0.625

Neighbourhood

Suppose that if two scenarios differ in more than 2 variables (i.e. Δ = 2), the total variable difference is more than 3 (i.e. δ = 3) or difference in
variable 3 is more than 1 (δ3 = 1), then we conclude that these scenarios are not neighbour to each other.
For instance, Scenario 1 and 2 are neighbours because they are different in one variable (i.e. their y1 and y2 levels are the same, y3 level is
different), total linguistic variable difference is 1 (i.e. difference of L and M in y3 is 1) and their difference in y3 is 1 (i.e. difference between M and L).
However, scenario 1 and 12 are not neighbours because although they are 2 different variables (i.e. their y2 level is the same, y1 and y3 levels are
different), and total linguistic variable difference is 3 (i.e. difference of L and M in y1 is 1 and difference of L and H in y3 is 2), their difference in y3
which is equal to 2 (i.e. difference of L and H in y3 is 2) is more than the threshold. Neighbourhood relations between selected scenarios are given in
Table B4.

Table B4
Neighbourhood relation between selected scenarios.

Scenario ID (k) y1 y2 y3 Scenario ID (l) y1 y2 y3 Rule 1 Rule 2 Rule 3 Neighbour?


Variable difference Total difference Difference in variable 3

1 L L L 2 L L M 1 0 + 0+1 = 1 1 Yes
1 L L L 12 M L H 2 1 + 0+2 = 3 2* No
3 L L H 23 H M M 3* 2 + 1+1 = 4* 1 No
3 L L H 5 L M M 2 0 + 1+1 = 2 1 Yes
19 H L L 27 H H H 2 0 + 2+2 = 4* 2* No
19 H L L 20 H L M 1 0 + 0+1 = 1 1 Yes
*shows a violation of a rule for Δ = 2, δ = 3, and δ3 = 1.

Scenario selection

In order to select scenarios, the mathematical program defined in Eq. (4) is used. To apply the proposed program to the hypothetical example, 21
scenarios which have consistency higher than 0 (scenarios 3, 6, 9, 19, 22 and 25 were eliminated) and for the neighbourhood-related parameters
Δ = 2, δ = 2, and δ3 = 0 are considered. When the proposed program is applied, scenarios 4, 14 and 27 are selected (see Table B5).

12
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table B5
Selected scenarios in the hypothetical example.

Scenario ID y1 y2 y3 C (Sk )

4 L M L 0.438
14 M M M 0.563
27 H H H 0.625

Scenario evaluation

In order to evaluate scenarios, the first approach is the similarity measure given in Eq. (5). We calculate the similarity measure between a given
state and selected scenarios to develop suggestions. Suppose a given state is Sg = (y1H , y2H , y2M ) , then the similarity measure to the selected scenario 4
(see Table B5) is calculated as follows:
3
σ (Sg , S4 ) = 1 − ∑i = 1 yi (Sg ) − yi (S4 ) 2 / 3(3 − 1)2
2 2 2
= 1− y1H − y1L + y2H − y2M + y3M − y3L / 12
= 0.29
for scenarios 14 and 27, similarity measures are calculated as σ (Sg , S14 ) = 0.59 and σ (Sg , S27) = 0.71, respectively. According to the calculated
similarity measures, the given state is most similar to scenario 27. When we compare Sg and S27 , although the first two variables are the same,
performance of Sg in the third variable is lower than S27 ’s. Therefore, we can recommend focussing on the third variable for the given state.

References Fan, S., & Chan-Kang, C. (2008). Regional road development, rural and urban poverty
evidence from China. Transport Policy, 15, 305–314.
Fedderke, J., & Garlic, R. (2008). Infrastructure development and economic growth in South
Amer, M., Daim, T. U., & Jetter, A. (2013). A review of scenario planning. Futures, 46, Africa: A review of the accumulated evidence. Economic research Southern Africa, policy
23–40. paper 12.
Arvis, J.-F., Mustra, M. A., Ojala, L., Stepherd, B., & Saslavsky, D. (2012). Connecting to Goodwin, P., & Wright, G. (2001). Enhancing strategy evaluation in scenario planning: A
compete 2012: Trade logistics in the global economy. World Bank. role for decision analysis. Journal of Management Studies, 38(1), 1–16.
Arvis, J.-F., Saslavsky, D., Ojala, L., Shepherd, B., Busch, C., Raj, A., et al. (2016). Goodwin, P., & Wright, G. (2014). Decision analysis for management judgment (5th ed.). UK:
Connecting to compete 2016 trade logistics in the global Economy: The logistics perfor- John Wiley and Sons.
mance index and its indicators. The World Bank. Gunasekera, K., Anderson, W., & Lakshmanan, T. R. (2008). Highway-induced develop-
Arvis, J.-F., Saslavsky, D., Ojala, L., Stepherd, B., Busch, C., & Raj, A. (2014). Connecting to ment: Evidence from Sri Lanka. World Development, 36(11), 2371–2389.
compete 2014: Trade logistics in the global economy. World Bank. Huss, W. R., & Honton, E. J. (1987). Alternative methods for developing business sce-
Barnister, D., & Berechman, Y. (2001). Transport investment and the promotion of eco- narios. Technological Forecasting and Social Change, 31, 219–238.
nomic growth. Journal of Transport Geography, 9(3), 209–218. IEA (International Energy Agency) (2008). World energy outlook. Paris, France: OECD
Bensassi, S., Márquez-Ramos, L., Martínez-Zarzoso, I., & Suárez-Burguet, C. (2015). Publishing.
Relationship between logistics infrastructure and trade: Evidence from Spanish re- ITF (International Transport Forum) (2015). Drivers of logistics Performance: Case study of
gional exports. Transportation Research Part A: Policy and Practice, 72, 47–61. Turkey. Corporate Partnership Board Report http://www.
Berechman, J., Ozmen, D., & Ozbay, K. (2006). Empirical analysis of transportation, in- internationaltransportforum.org/Pub/pdf/15CPB_Logistics-Turkey.pdf.
vestment and economic development at state, country and municipality levels. Lean, H. H., Huang, W., & Hong, J. (2014). Logistics and economic development:
Transportation, 33(6), 537–551. Experience from China. Transport Policy, 32, 96–104.
Bishop, P., Hines, A., & Collins, T. (2007). The current state of scenario development: An Liu, W., Li, W., & Huang, W. (2006). Analysis of the dynamic relation between logistics
overview of techniques. Foresight, 9(1), 5–25. development and GDP Growth in China. IEEE international conference on service op-
Bradfield, R., Wright, G., Burt, G., Cairns, G., & Van Der Heijden, K. (2005). The origins erations and logistics, and informatics, 21-23 June 2006, Shanghai (pp. 153–157). .
and evolution of scenario techniques in long range business planning. Futures, 37, Nguyen, M. T., & Dunn, M. (2009). Some methods for scenario analysis in defence strategic
795–812. planning. Australian Government Joint Operations Division Defense Science and
Bryan, B. A., Crossman, N. D., King, D., & Meyer, W. S. (2011). Landscape futures ana- Technology Organisation DSTO-TR-2242.
lysis: Assessing the impacts of environmental targets under alternative spatial policy Nguyen, H.-O., & Tongzon, J. (2010). Causal nexus between the transport and logistics
options and future scenarios. Environmental Modelling & Software, 26(1), 83–91. sector and trade: The case of Australia. Transport Policy, 17(3), 135–146.
Carruthers, R., Bajpai, J. N., & Hummels, D. (2003). Trade and logistics in East Asia: A Onsel Ekici, S., Kabak Ö, & Ülengin, F. (2016). Linking to complete: Logistics and global
development agenda, EASTR working paper, No. 3. Transport Sector Unit, Infrastructure competitiveness interaction. Transport Policy, 48, 117–128.
Department, East Asia and Pacific Region, The World Bank. Önsel Sahin, S., Ülengin, F., & Ülengin, B. (2004). Using neural networks and cognitive
Chen, S. J., & Hwang, C. L. (1992). Fuzzy multiple attribute decision making methods: mapping in scenario analysis: The case of Turkey's inflation dynamics. European
Methods and applications. Lecture Notes in Economics and Mathematical Systems. Journal of Operational Research, 158(1), 124–145.
Springer, Berlin, Heidelberg. Pittock, A. B. (2009). Climate change: The science, impacts and solution (2nd ed.). Australia:
Chen, N., & Novy, D. (2011). Gravity, trade integration, and heterogeneity across in- CSIRO Publishing.
dustries. Journal of International Economics, 85, 206–221. Quist, J., & Vergragt, P. (2006). Past and future of backcasting: The shift to stakeholder
Chermack, T. J. (2004). A theoretical model of scenario planning. Human Resource participation and a proposal for a methodological framework. Futures, 38(9),
Development Review, 3(4), 301–325. 1027–1045.
Chermack, T. J., Lynham, S. A., & Ruona, W. E. A. (2001). A review of scenario planning Ram, C., Montibeller, G., & Morton, A. (2011). Extending the use of scenario planning and
literature. Futures Research Quarterly, 17, 7–31. MCDA for the evaluation of strategic options. Journal of the Operational Research
Cuenot, F., Fulton, L., & Staub, J. (2012). The prospect for modal shifts in passenger Society, 62, 817–829.
transport worldwide and impacts on energy use and CO2. Energy Policy, 41, 98–106. Rantasila, K., & Ojala, L. (2012). Measurement of national level of logistics costs and per-
Demurger, S. (2001). Infrastructure development and economic growth: An explanation formance, international transport forum discussion paper, No 2012–4.
for regional disparities in China. Journal of Comparative Economics, 29, 95–117. Schuckmann, S. W., Gnatzay, T., Darkow, I.-L., & Heiko, A. (2012). Analysis of factors
Diakoulaki, D., & Karangelis, F. (2007). Multi-criteria decision analysis and cost-benefit influencing the development of transport infrastructure until the year 2030-A Delphi
analysis of alternative scenarios for the power generation sector in Greece. Renewable based scenario analysis. Technological Forecasting and Social Change, 79, 1373–1387.
and Sustainable Energy Reviews, 11, 716–727. Srdjevic, B., Medeiros, Y. D. P., & Faria, A. S. (2004). An objective multi-criteria eva-
Duckett, D. G., McKee, A. J., Sutherland, L.-A., Kyle, C., Boden, L. A., Auty, H., et al. luation of water management scenarios. Water Resources Management, 18, 35–54.
(2017). Scenario planning as communicative action: Lessons from participatory ex- Stewart, T. J., French, S., & Rios, J. (2013). Integrating multicriteria decision analysis and
ercises conducted for the Scottish livestock industry. Technological Forecasting and scenario planning—review and extension. Omega, 41, 679–688.
Social Change, 114, 138–151. Tietje, O. (2005). Identification of a small reliable and efficient set of consistent scenarios.
Esfahani, H. S., & Ramirez, M. T. (2003). Institutions, infrastructure and economic European Journal of Operational Research, 162, 418–432.
growth. Journal of Development Economics, 70, 443–477. TSV (Turkey’s Strategic Vision 2023 Project 2008-2023) 2015, http://www.tsv2023.org/.

13
Ö. Kabak et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Ülengin, F., Kabak, Ö., Önsel, Ş., Ülengin, B., & Aktaş, E. (2010). A problem-structuring backcasting studies. Futures, 43, 880–889.
model for analyzing transportation-environment relationships. European Journal of Weimer-Jehle, W. (2006). Cross-impact balances: A system –theoretical approach to
Operational Research, 200, 844–859. cross-impact analysis. Technological Forecasting and Social Change, 73, 334–361.
Walz, A., Lardelli, C., Behrendt, H., Gret-Regamey, A., Lundstrom, C., Kytzia, S., et al. Wright, G., Cairns, G., & Goodwin, P. (2009). Teaching scenario planning: Lessons from
(2007). Participatory scenario analysis for integrated regional modelling. Landscape practice in academe and business. European Journal of Operational Research, 194,
and Urban Planning, 81, 114–131. 323–335.
Wangel, J. (2011). Change by whom? Four ways of adding actors and governance in

14

View publication stats

You might also like