1 s2.0 S0969699723000078 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Journal of Air Transport Management 108 (2023) 102364

Contents lists available at ScienceDirect

Journal of Air Transport Management


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

The effect of airport efficiency on air traffic, using DEA and multilateral
resistance terms gravity models
Ah-Hyun Jo a, Young-Tae Chang b, *
a
Korea Maritime Institute, Busan, South Korea
b
Graduate School of Logistics, Inha University, Incheon, South Korea

A R T I C L E I N F O A B S T R A C T

Keywords: This study evaluates the bias-corrected airport efficiency by employing a slacks-based measurement-data
Airport efficiency envelopment analysis model with a bootstrapping technique. The impacts of airport efficiency on bilateral
Air traffic passenger traffic and freight volume are then assessed by developing a novel robust gravity model incorporating
Bootstrapping SBM-DEA model
multilateral resistance (MR) terms and the Poisson pseudo maximum likelihood (PPML) method to address
Gravity model
Multilateral resistance
heteroscedasticity. We collected air route data using Incheon airport with a sample of 50 airports in 28 countries
PPML estimator for 2006–2015. The results from all the applied models revealed a significant and positive effect of airport ef­
ficiency on passenger traffic and freight volume.

1. Introduction service qualities which customers experience. In these ways, the cost and
time reductions and enhanced customer satisfaction ultimately can
Today, as the aviation industry has developed rapidly, new demands make a bigger passenger traffic and freight volume. Also, cost reductions
for air traffic have emerged. Increasing international collaboration and in terms of airports caused from the increased efficiency allow the air­
growth of e-commerce and high value-added product have accelerated ports to establish a competitive price policy to attract bigger air traffics.
the increase of air passenger traffic and freight volume. The improving However, the relationship between airport’s efficiency and air traffics is
living standards worldwide and growing the middle class in developing not investigated yet. Therefore, the purpose of this paper is to empiri­
countries are also crucial factors boosting the growth of aviation in­ cally measure the impact of airport’s efficiency on both passenger traf­
dustry (Zhang and Zhang, 2016). The total number of air passengers fics and freight volume.
carried on scheduled services increased by 7.2% from 3.8 billion in 2016 Numerous studies have assessed the impact of transport infrastruc­
to 4.1 billion in 2017 (ICAO International Civil Aviation Organization, ture performance on bilateral passenger and cargo traffic (Sánchez et al.,
2018) and the global air freight measured in freight ton kilometers 2003; Clark et al., 2004; Wilson et al., 2005; Blonigen and Wilson, 2008;
increased by 9.0% from 2016 to 2017 year (IATA International Air Márquez-Ramos et al., 2011; Portugal-Perez and Wilson, 2012; Haus­
Transport Association, 2018). Therefore, many airports are now man et al., 2013; Bensassi et al., 2015; Bottasso et al., 2018). Efficiency
suffering from severe congestions such as delay at airport and these literature, however, points out that most efficiency studies are biased,
congestions have become a major problem worldwide (Rupp, 2009; mainly due to sampling problems, implying that those studies on airport
Santos and Robin, 2010). To solve these congestions and accommodate efficiency are not an exception (Simar and Wilson 1998, 2000). More­
increasing air traffic, many countries in the world are striving to develop over, the international trade literature also mentions other types of
their airport by expanding airport facilities and making the airport more serious biases in extant studies.
efficient. The typical methodologies used to measure the efficiency and
Since expanding airport facilities needs a huge amount of money and bilateral trade are data envelopment analysis (DEA) and the gravity
long time, governments and airport authorities have focused on model, respectively. DEA literature reveals an inherent bias of efficiency
enhancing the airport efficiency. The increase of airport efficiency will score caused by the data sampling process (Simar and Wilson, 1998,
bring cost and time reduction for both airport and its users. Or increased 2000). The data sampling bias lies in the presumption or research
airport efficiency can solve some congestions and, in turn, enhance the practice that the collected sample represents a population. In the air

* Corresponding author.
E-mail address: ytchang@inha.ac.kr (Y.-T. Chang).

https://doi.org/10.1016/j.jairtraman.2023.102364
Received 8 December 2021; Received in revised form 19 May 2022; Accepted 19 January 2023
Available online 11 February 2023
0969-6997/© 2023 Elsevier Ltd. All rights reserved.
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

transport management field, there have also been numerous DEA studies of studies have applied DEA models to various areas. However, these
that examine airport’s efficiency without consideration for the sampling DEA models have some drawbacks, including the radial assumption and
bias. Because the ignorance of this bias can expose stakeholders to orientation approach as they require all DMUs to change their inputs or
wrong information about benchmark points and cause wrong in­ outputs equi-proportionally to be efficient, which is therefore termed
vestments, correcting these biases in the methodology is important. the ‘radial assumption’. Moreover, the DMUs can improve their effi­
Further detail/corroboration on this shortcoming is provided below in ciencies either in the input dimension or the output dimension, which is
the literature and methodology sections. The international trade litera­ therefore called ‘orientation’. Tone (2001) proposed the SBM method­
ture shows that numerous studies using gravity models are biased due to ology to overcome the drawbacks of the DEA model’s radial assumption
the neglect of multilateral resistance (MR) or heteroscedasticity prob­ and orientation approach.
lems, or both (McCallum, 1995; Anderson and Van Wincoop, 2003; Silva A critical issue of DEA is that it cannot explain the statistical prop­
and Tenreyro, 2006). The application of a biased regression model can erties of the sample data. Unlike SFA, efficiency scores of DEA are totally
cause serious biases in the estimation and, in turn, offer wrong impli­ determined by the frontier line, which is composed of the sampled
cations to the industry and stakeholders such as policy makers. To the dataset, so the results can vary depending on which sample dataset is
best of our knowledge, no studies have been conducted yet to examine selected. Simar and Wilson (1998, 2000) introduced a bootstrapped DEA
the impact of airport efficiency on air freight volume and passenger method to overcome this limitation by re-calculating efficiency scores
traffic with a robust methodology. Even the numerous existing studies with a resampled dataset. The bootstrapped DEA method develops the
on airport efficiency or international trade are deemed to be not free data generating process (DGP) to resample data and draws the sampling
from the biases mentioned above. distribution of the measured efficiency score from the resampled data­
Our research makes the following contributions. First, this research set. By analyzing the approximated sampling distribution, the
is the first of its kind to investigate the impact of airport efficiency on bias-corrected efficiency score and confidence intervals for efficiency
both air passenger traffic and air freight volume using empirical data. scores can be obtained.
Since there only have been qualitative studies investigating the rela­ Following the pioneering study conducted by Gillen and Lall (1997)
tionship between airport efficiency and bilateral trade, it is important to in the airport efficiency literature, various studies have applied DEA
empirically measure the sign and magnitude of impact of environment models to investigate airport efficiency (Pels et al., 2003; Lin and Hong,
variable on air traffics. To address the impact of airport efficiency on 2006; Lozano and Gutiérrez, 2011; Wanke, 2012). More detailed re­
bilateral trade and passenger flow, this study develops testing models at views can be referred to Dmitry (2012), Merkert et al. (2012), and
two stages. In the first stage, airport efficiency from 2006 to 2015 is Fasone and Zapata-Aguirre (2016). However, among the numerous
analyzed using the bootstrapping SBM-DEA model. In the second stage, studies, only a few studies considered the biases of DEA models (Wanke,
the impact of airport efficiency on the air passenger traffic and freight 2012). As Simar and Wilson (1998, 2000) pointed out, without the
volume is examined by including the airport efficiency analyzed in the consideration for the sampling biases of DEA model, the result of esti­
first stage into the gravity model. mation can be biased and, in turn, offer wrong information about the
Second, this research employed novel and robust models to estimate DMU and industry. Since the result of DEA model offers benchmark
the impact of airport efficiency on air passengers and freight. Since the points to the DMUs that they can follow to be more efficient, correcting
application of biased methodology can offer wrong signals to the air for biases in the model is crucial.
market and stakeholders, it is crucial to adopt robust models. We correct
the sampling bias of DEA by combining SBM-DEA models with the 2.2. Gravity models to analyze bilateral trade
bootstrapping technique. We solve the infeasible problems arising when
combining the SBM-DEA models with subsample-based bootstrapping Moving to the trade literature, the gravity model has been used as
technique by applying the super-efficiency model approach. Also, this is one of the most successful measures to analyze international trade
the first study to apply a technical rule for selecting a sub-sample size (Anderson, 1979; Zhang et al., 2018). The idea of the original gravity
when conducting empirical bootstrapped DEA. Finally, to eliminate model stemmed from Newton’s law of gravity. His theory explains that
inherent biases in the gravity model, we adopt the MR term and Poisson the gravitation between two objects is proportional to their masses and
pseudo maximum likelihood (PPML) method while controlling for other inversely proportional to their inter-distance. Tinbergen (1962) and
variables, including macroeconomic factors and aviation demand- Pöyhönen (1963) are initial works that first introduced the law of
related factors found in the literature. By eliminating the biases in the gravity to the field of international trade. They replaced each
methodology, we expect that this model can estimate the impact of physics-related variable of the gravity model with trade-related vari­
airport efficiency on air passengers and freight precisely and in turn, ables. The bilateral trade volumes between the two countries correspond
offer correct field information to stakeholders such as airport, its users, to the gravitation between the two objects. The GDP, representing the
and policy makers. economic scale of the country, and the geographical distance between
The rest of this paper is organized as follows. The following section the two countries stand for the mass of the object and the distance be­
reviews the literature on airport efficiency and bilateral trade. Section 3 tween the two objects, respectively. From these studies, the literature
explains the methodology and section 4 presents the dataset and results. further includes diverse variables such as customs union, national ties,
Finally, section 5 concludes and states the limitations of this study with a trade agreements, and languages (Bougheas et al., 1999; Limao and
view to suggesting areas worthy of further research. Venables, 2001; Wilson et al., 2003; Clark et al., 2004; Nordas and
Piermartini, 2004; Portugal-Perez and Wilson, 2012; Bottasso et al.,
2. Literature review in efficiency and gravity models 2018).
The gravity literature raised a question about MR issues which means
2.1. DEA models to estimate efficiency score that the trade between the two countries is affected not only by bilateral
trade barriers but also by multilateral trade barriers of all countries
Two streams of research have investigated efficiency score: sto­ having a relationship with the two countries. One of the straightforward
chastic frontier analysis (SFA) and DEA. DEA has long been popular in ways to address MR is a fixed-effect model which considers the region or
the literature of efficiency measuring methodology because of its time or both of two effects (Feenstra, 2004). However, the fixed-effect
methodological advantages. It can handle multiple input and output model has some disadvantages in that it cannot explain the direct ef­
variables and does not require a definite production functional form. fect of region-specific variables on the trade since all the time-invariant
Charnes et al. (1978) first measured the relative efficiency score by effects are absorbed by the fixed effect terms (Agnosteva et al., 2014).
introducing a linear programming technique. Since then, a vast number Various types of panel analysis in many empirical studies have

2
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

continued to overlook the MR, while only a few studies have included
the MR terms in the gravity model (Baier and Bergstrand, 2009; Portu­
gal-Perez and Wilson, 2012; Zhang and Zhang, 2016; Zhang et al.,
2018).
Another critical issue in the gravity literature is heteroscedasticity.
The heteroscedasticity problem arises in the process of converting the
multiplicative error term into an additive error term, which makes the
use of the regression form a failure due to the violation of the funda­
mental assumptions of the model (Silva and Tenreyro 2006). In addition,
if some bilateral trade between the two countries has a zero value,
log-linearization turns this zero value into minus infinity, which renders
the regression model invalid. Contrary to existing nonlinear trans­
formation techniques such as non-linear least squares (NLS) and Tobit
regressions, PPML can simultaneously cope with heteroscedasticity and
the zero value problem.
Various studies investigated what factors determine the air passen­
ger traffic using the gravity models (Endo, 2007; Poole, 2010; Hwang Fig. 1. The development of methodology for DEA and the gravity model.
and Shiao, 2011; Grosso, 2012; Chang, 2014; Zhang and Zhang, 2016).
However, among the numerous studies, only a few studies addressed the the frontiers. We applied the idea of the super-efficiency method to solve
MR and heteroscedasticity issues of gravity models. Ignoring MR issues this problem (Tone, 2002). (5) The second problem is that the
can distort the effect of environmental variables on the bilateral trade sub-sampling technique requires an ex-ante selection of sampling size;
since international trade occurs simultaneously and also multilaterally. however, no empirical studies have yet examined how the ex-ante
According to the literature for the gravity model, the basic ordinary least sampling size has to be decided. The present study is the first to intro­
squared (OLS) method presents mis-specified results by ignoring MR duce a practical rule for selecting the subsample size when conducting
terms (Anderson and Van Wincoop, 2003). When examining bilateral an empirical study of bootstrapping based on the subsampling method.
trade between two countries, all trade-supporting factors or barriers The MR-related bias is a result of ignoring the necessity to evaluate
around them should be considered in the model because their effect is any trade barriers, such as free trade agreements (FTAs), according to
relative in relation to all other countries around the world. For example, their relativity rather than to their absolute magnitude (Anderson and
we can intuitively expect FTAs to increase the bilateral trade. However, Van Wincoop, 2003). (6) To deal with this, we introduce MR terms,
if all countries are tied in a same trade agreement with each other, the which are modified variables. (7) Moreover, the PPML method was
effect of an additional FTA between two trading countries becomes adopted which can simultaneously cope with heteroscedasticity and the
weakened or meaningless. zero value problem.
In summary, given that the airline industry is essential in both
regional and global economy and creates high added-value, it is 3.1. SBM-DEA model for measuring efficiency
important to examine what factors determine the growth of air pas­
senger traffic and air freight volume. Specifically, it is very timely to DEA is an efficiency measurement method based on linear pro­
discover the empirical evidence showing the effectiveness of logistics gramming. This study employed the SBM-DEA model combined with
infrastructure investment and efficiency. Nevertheless, in the absence of bootstrapping technology. Contrary to classic DEA, the SBM-DEA model
any empirical research examining these effects, this study intends to fill does not need to decide the specific orientation, in the direction of input
the gap in the literature by developing a novel and robust model to reduction (input orientation given output level) or output increase
measure airport efficiency and assess its impact on international trade. (output orientation given input level), to measure efficiencies, but rather
Applying a novel and robust two-stage model, a bias-corrected estima­ takes both input and output slacks into consideration simultaneously, i.
tion can be obtained and, in turn, offer useful information to stake­ e., with ‘no orientation’ (Tone, 2001). In addition, the SBM model can
holders such as airports, customers, and policy makers. In the next deal with changes of all input and output factors at differing ratios to
section, we developed the novel and robust two-stage model, the SBM- improve DMUs’ efficiencies, unlike the ‘equiproportional change’ of
DEA model with bootstrapping technique and gravity models with traditional DEA models, which are therefore ‘non-radial’. Based on these
multilateral resistance terms. features, the SBM-DEA model shows higher discriminating power and
estimates the efficiency scores more realistically. The production pos­
3. Methodology sibility set P containing p inputs, q outputs and n observations can be
defined as follows:
In this paper, we develop a novel and bias-correcting two-stage
model to measure the impact of airport efficiency on bilateral air traffic. P = {(x, y)| x ≥ Xλ. y ≤ Yλ, λ ≥ 0} (1)
Fig. 1 shows how this paper improves existing models to develop our
Where λ is an intensity vector with non-negative values representing
novel and robust model, which combines the two main methodologies of
benchmarking points, x is an input vector, y is an output vector, and X
DEA and gravity models. The numbers in the parentheses indicate each
and Y are the input and output matrix, respectively. The matrix form of X
rationale of methodology development. (1) First, a slacks-based mea­
and Y can be expressed as
surement (SBM)-DEA model was used to estimate the airport’s efficiency
⎛ ⎞ ⎛ ⎞
(Tone, 2001). (2) A bootstrapping technique is introduced to correct for x11 ⋯ x1n y11 ⋯ y1n
the inherent bias in the efficiency score caused by the sampling process. X= ⎝ ⋮ ⋱ ⋮ ⎠ andY = ⎝ ⋮ ⋱ ⋮ ⎠. (2)
The subsampling method was selected because it handles both input and xp1 ⋯ xpn yq1 ⋯ yqn
output variables simultaneously for bootstrapping, and is therefore
Using Tone (2001)’s SBM model, the framework to assess the effi­
suited for SBM-DEA models (Simar and Wilson, 2008).
ciency of certain DMU (x0 , y0 ) can be described as follows:
(3) The combination of SBM-DEA and the subsampling-based boot­
strapping technique leads to the following two problems. (4) The first is
an infeasible problem since newly generated frontiers formed by the
subsamples sometimes place some decision-making units (DMU) outside

3
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

∑p
1 s−i chosen. The stepwise algorithm for bootstrapping efficiency scores from
1− p i=1 xi0
min θ = ∑q s+
subject tox0 = Xλ + s− , y0 = Yλ − s+ , λ ≥ 0, s− ≥ 0, s+ the SBM with the subsampling technique is provided in the Appendix.
1
Fig. 2 shows the whole process of calculating the bias-corrected effi­
j
1+ q j=1 yj0

≥ 0. ciency score.
(3)
3.2. Gravity model
where s− and s+ represent input slacks (excess) and output slacks
(shortfall), respectively. Gravity model is the most popular methodology to investigate in­
Inherent biases in efficiency scores arising from an arbitrary sam­ ternational trade (Anderson, 1979; Zhang et al., 2018). There have been
pling, which can lead to a serious uncertainty in the measurement, can numerous studies analyzing the impact of environment variables on the
be corrected by incorporating the bootstrapping technique into a DEA bilateral passenger traffic and freight volume using Gravity models.
model (Simar and Wilson 1998, 2000). The basic concept of bootstrap in However, only a few studies addressed MR and heteroscedasticity issue
DEA is to imitate the distribution of an original estimator by replicating in their model. Ignoring these issues can cause biased estimation results
a proper DGP and repeatedly calculating the DEA scores. It is posited and, in turn, offer wrong signals to the market and stakeholders.
that the relationship between original and unknown efficiency score and Therefore, in this section, the MR terms and PPML method were intro­
estimated efficiency score from the initial sample corresponds to the duced to address above issues.
relationship between the estimated efficiency score from the initial The fundamental principle of the gravity model stemmed from the
sample and bootstrapped DEA score from the resampled dataset. In this physical law that the power of objects drawing each other is propor­
way, we can estimate and correct the inherent bias within efficiency tional to their mass and inversely proportional to the distance between
scores caused by data sampling. Following Simar and Wilson (2008), them. Likewise, bilateral trade increases with the economic scale of
two methods of DGPs can be used to apply the bootstrapping technique countries and decreases with the distance between the two trading
into the DEA score: the subsampling and smoothing technique. This countries. The fundamental form of the gravity model is expressed as
study employed the subsampling technique to generate resampled data follows:
since the smoothing technique only deals with a single factor, either
input or output, which is not suitable for the SBM-DEA model as it has to Tijt = αo GDPαit1 GDPαjt2 Distαij3 (6)
handle both input and output slacks simultaneously.
Implementing subsampling techniques with the SBM model, how­ where Tijt is bilateral trade volume between countries i and j at period t,
ever, we faced a limitation. Generally, linear programming posits all GDPit (GDPjt ) is the gross domestic product of country i (j), Distij is the
non-negative decision variables. However, because the subsampling distance between the two countries and α0 , α1 , α2 , and α3 are unknown
procedure often shrinks a frontier line (reference set), DMUs can be parameters. To apply the model (6) empirically, it is usually transformed
located outside of the frontier line, which consequently results in an into a log-linearized form. The uncertainty is a remarkable difference
infeasible solution in the SBM model. One possible solution is resorting between natural science theory and social science theory. ηijt in equation
to the super-efficiency of the SBM model, as it can handle DMUs located (7) represents the deviations from discrepancies between theory and
outside of the frontier line (Tone, 2002). Let us define a new production empirical data (Silva and Tenreyro 2006). By incorporating it into
possibility set as follows: equation (6) the stochastic version of the gravity equation takes the
} form:
∑ ∑
sub
P = {(x, y)| x ≥ λm xm , y ≤ λm ym , y ≥ 0, λ ≥ 0 , (4) Tijt = αo GDPαit1 GDPαjt2 Distαij3 ηijt (7)
m∈M m∈M
To estimate the parameters in equation (7), the trade literature has
which includes only sub-sampled m DMUs (xm , ym ). M is a newly used log-linearization, which enables the multiplicative equation to be
generated sample set from the initial dataset having size m out of n. solved by introducing regression methods such as least squares. This is
Considering sub-sampled m DMUs as a new reference set, the framework the transformed equation:
to assess the efficiency of certain DMU (x0 , y0 ) is described as follows:
ln Tijt = β0 + β1 ln GDPit + β2 ln GDPjt + β3 ln Distij + ln ηijt (8)
1
∑p x i
∑ ∑
An issue with log-linearization is that the error term ln ηijt does not
sub p i=1 x
min θ = ∑q yi0j s ubject to x ≥ λm xm , y ≤ λm ym , λ ≥ 0, x
1
q j=1 yj0 m∈M m∈M
satisfy the fundamental assumptions of regression analysis, which in
≥ x0 and y ≤ y0 , y ≥ 0 (5) turn causes serious problems such as heteroscedasticity. This problem
requires the adoption of a completely different estimation methodology,
Introducing this approach successfully solves the infeasible problem rather than least squares estimation, such as applying PPML estimation
in the bootstrapping SBM-DEA score caused by the reduced frontier line. into the model to solve these issues. Based on the constant-elasticity
To empirically measure the bias-corrected efficiency score of DMUs model, the measurement model for PPML is
with bootstrapped DEA, we first have to determine the subsampling size
m to simulate the DGP. However, the absence of any definite guideline yi = exp(xi βi ) + εi (9)
that can be applied to the general sample data has led some researchers
to display the coverage levels according to significance levels or the with yi ≥ 0 and E[εi |x] = 0. Silva and Tenreyro (2006) proposed a
characteristics of data such as the number of variables; nevertheless, no pseudo maximum likelihood estimator based on the assumption that the
golden rule has been determined for selecting subsample size m (Bickel conditional variance of εi is proportional to an exponential function of
and Sakov, 2008; Kneip et al., 2008). Alternatively, Politis et al. (2001) xi βi . Since this model deals with a pseudo-maximum likelihood esti­
introduced a data-driven method for choosing subsample size m, espe­ mator, it is not mandatory that the data should be from a Poisson dis­
cially when constructing confidence intervals of the estimated efficiency tribution. Parameters β of PPML are estimated by solving a set of
score. The principle of this method is to minimize the running standard first-order conditions:
deviation applied to the endpoints of confidence intervals, under the ∑
n

motivation of finding a certain point of m having the highest ‘stability’ [yi − exp(xi ̃
β)]xi = 0. (10)
for the fluctuation of confidence intervals. Among sample size M in the i=1

range of msmall to mbig , a specific value m having the smallest volatility in In addition, bilateral trade volume data often contain a zero value in
running standard deviation for the endpoints of confidence intervals is many cases, and log-linearization turns them into minus infinity, which

4
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

Fig. 2. Flowchart of efficiency estimation.

renders the regression model invalid. PPML can be an appropriate so­ interactions. A low-cost carrier (LCC) is also a dummy variable having a
lution for both obstacles. Silva and Tenreyro (2006) stated that other value of 1 if low-cost carriers operate on the air route, following Zhang
nonlinear transformation techniques, for example NLS and Tobit re­ and Zhang (2016). To investigate the differences between market types,
gressions, cannot cope with heteroscedasticity but PPML can. we introduced two dependent variables - passenger traffic and freight
We examined the international trade literature to select the explan­ volume. After taking log-linearization, the empirical gravity model
atory variables and decided our final variables in the following manner. equation estimating the impact of environmental factors on passenger
GDP and GDP per capita variables are traditional macroeconomic vari­ traffic and freight volume can be expressed as follows. An index i was
ables explaining the economic scale of countries. In particular, GDP per omitted because it solely represents Incheon.
capita generally means the income level of individuals. Distance vari­
ln Tjt = β0 + β1 ln GDPjt + β2 ln GDPPjt + β3 ln Distj + β4 FTAjt + β5 LCCjt
able means the length of the air route connecting trading countries
(regions). It is obvious that the more separated airports are, the less is + β6 ln Effijt + εjt
their trade and intercourse. Additionally, gravity models in the literature (11)
have employed extra explanatory variables such as common language,
land border, landlock, colonies, currency, and bilateral tariffs (Glick and where:
Rose, 2002; Nordas and Piermartini, 2004; Blonigen and Wilson, 2008).
As the main interest of this research is evaluating the impact of • Tjt denotes the bilateral air passenger traffic or freight volume be­
airport efficiency on the passenger traffic and freight volume, we chose tween Incheon airport and airport j at time t,
six independent variables, which are deemed to be the most important, • GDPjt denotes real GDP of the country of airport j at time t,
due to data and time constraints in the research. FTA is a dummy vari­ • GDPPjt denotes real GDP per capita of the country of airport j at time
able showing a value of 1 if trading countries have an FTA. An FTA t,
lowers various trade barriers and consequently increases international

5
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

• Distj is the length of the air route between Incheon airport and air traffic ranking third and fifth with 295 million tons of international
counterpart airport j, freight and 67 million international passengers, respectively, in 2018.
• FTAjt is a dummy variable which is unity if Korea and the country of Table 1 describes the input and output variables used for the first
airport j were in a FTA at time t, stage analysis. The data were obtained from the ATRS Global Airport
• LCCjt is a dummy variable representing the presence of LCC in the air Benchmarking Reports. Air Transport Research Society (ATRS) annually
route between Incheon airport and airport j in period t, reports the data on productivity, efficiency, and unit cost competitive­
• Effijt is an efficiency score of airport j at time t, ness of global airports to provide benchmarking points for the airports
• β is a vector of coefficient values of estimators, (ATRS, 2005–2018). The number of employees directly represents the
• εjt is the error term. level of labor power of airports. Runways, terminal space, and gates are
capital resources. Yoshida and Fujimoto (2004) found that the size of the
Bergstrand (2010) explained the MR issue by employing simple av­ terminal determines the airport’s capacity to operate passenger traffic
erages of the trade costs rather than GDP-shares weights. Following and cargo handling and hence plays an important role in the operating
Bergstrand (2010), MR for FTA impact can be defined as: activity of airports. These are standard input variables needed to mea­
( ) ( ) sure the efficiency of airport operation. Passengers, cargoes and aircraft
∑T ∑Nr T ∑
∑ Ns
MFTAijt =
1
FTArt +
1
FTAst movements were used as output variables. These output variables
T • Nr t=1 r=1 T • Ns t=1 s=1 represent the outcome of airport operation. The total number of DMUs
( )
∑T ∑Nr ∑Ns was 377. Fig. 3 depicts how the input and output variables were used in
1
− FTArst (12) the SBM model.
T • Nr • Ns t=1 r=1 s=1
Table 2 summarizes the independent and dependent variables of the
gravity models used in the second stage. At the initial data collecting
where r is an index for the agreement partner of the country (region) i,
stage, the air route data covered 102 airports in 40 countries worldwide.
and s is an index for the partner of country (region) j. Nr(s) represents the
After matching the data with ATRS Global Airport Benchmarking
number of agreement partners of country i (j) and T is the time period.
Report, the final dataset was reduced to 50 airports in 28 countries. The
The revised FTA variables for the MR term in the empirical gravity
total observations numbered 245 for passenger traffic and 345 for
model can be expressed as follows:
freight volume models. Meanwhile, our model has two ‘passenger’
FTA∗ij = FTAij − MFTAij (13) variables in the first stage and second stage. In the first stage, it means
the total number of air passenger for a single airport. On the other hand,
In summary, in the air transport management field, there have been in the second stage, ‘passenger’ indicates the number of passenger be­
numerous DEA studies examining airport’s efficiency without consid­ tween two specific airports. The correlation value between two ‘pas­
eration for the sampling bias. The ignorance of this bias can expose senger’ variables is 0.023 and the p-value is 0.7145, which indicated
stakeholders to wrong information about benchmark points and cause that the two ‘passenger’ variables are not statistically correlated.
wrong investments. The application of biased regression model also can Therefore, our two-stage model is free from the separability condition,
cause serious biases in the estimation and, in turn, send wrong signals to which can harm the robustness of the model as Simar and Wilson (2007)
the industry and stakeholders such as policy makers. Therefore, by pointed out.
developing and combining the two methodologies, SBM-DEA with
subsampling technique and Gravity model with MR terms and PPML, we
can measure the efficiency of airport and its impact on the bilateral 4.2. Results
passenger traffics and freight volume correcting for biases in estimation.
Fig. 4 shows how the volatility of confidence intervals is affected by
( 4 ) ( )
4. Empirical work the subsample sizes ranging from 30 = 50 ⋅377 to 354 = 47 50 ⋅377 .
Nine thin lines represent the running standard deviations according to
4.1. Data nine combinations of each k ∈ {1, 2, 3} and α ∈ {0.1, 0.05, 0.01}. The
thick line in the graph represents the average trend line of the rela­
Although testing the relationship between airport efficiency and air tionship between the two variables (sample size and volatility) in an
traffic ideally requires collecting the entire global data between world exponential function form. Measuring the volatility of the confidence
airports and their respective trading traffic volumes, such complete interval by subsample size necessitated making 37.7 million linear
population data were not available. Instead, we used sample data programming computations (377 DMUs and 2000 times bootstrapping
collected from the top 50 busiest air routes using Incheon airport for for each of 50 subsample sizes from m1 to mJ ).
2006–2015 in view of the fact that the traffic of Incheon airport suffi­ Table 3 displays subsampling size m∗ having the smallest SOV of
ciently reflects the general characteristics of air passenger and freight confidence interval for each k and α. As k increases, i.e., as the coverage
traffic along the major air routes in the world even if Incheon airport of running standard deviations becomes wider, the volatility of confi­
traffic has insignificant effect on trans-Atlantic air passenger and freight dence interval tends to increase. To conduct an empirical gravity anal­
)
traffic. Incheon International Airport plays a vital role as a global hub for ysis in the second stage, we chose the subsampling size 347 (= 46 50⋅377 as
it had the smallest volatility among the m∗ in Table 3. Though we

Table 1
Data description for the SBM-DEA model.
Category Variables Unit Mean Med Max Min Std.dev

Input variable Terminal Size m2 488,969 396,000 1,972,474 23,272 344,808


Runways Number 3 2 8 1 1
Gates Number 104 102 264 5 52
Employees Number 2296 1380 20,720 163 3149

Output variable Passengers Number (1000’s) 38,950 35,599 101,491 2714 21,034
Cargoes Metric tons 1,071,842 712,118 4,400,000 16,068 875,452
Aircraft Movements Number 341,097 305,808 981,402 28,991 195,973

Source: The ATRS Global Airport Benchmarking Report 2008–2017.

6
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

Fig. 3. The structure of the SBM-DEA model.

Table 2
Data description for the gravity model.
Variables Unit Mean Max Min Std.dev
a
Indepen-dent variables GDP 1000 USD 5,468,246 16,672,692 5000 6,052,705
(counterpart)
GDP per capita a USD 35,765 91,617 1880 20,780
(counterpart)
Airport efficiency – 0.519 0.999 0.067 0.247
Distanceb km 6543 12,466 821 3530
FTAc Binary – – – –
LCC Binary

Depen-dent variables Passengerd Number 788,477 3,084,931 167,641 594,328


Freightd Ton 64,971 280,595 8162 58,294

Source: a) World Bank, b) Korean Ministry of Land, Infrastructure, and Transport, Aviation Policy Division, c) World Trade Organization (WTO), d) Air portal
(http://www.airportal.go.kr).

estimation without affecting the result of the gravity model.


Table 4 displays the results of the first stage, which measures the
efficiency scores of world airports when α = 0.05 and m = 347. We only
present the result of 2015, which is the latest period since the original
result is so massive. After correcting the inherent bias of efficiency score
(Effi_bc), Hartsfield Jackson Atlanta International Airport (ATL), Dubai
y = 23.26 e(-0.065)x International Airport (DXB), Hong Kong International Airport (HKG),
R² = 0.93 Phuket International Airport (HKT), Honolulu International Airport
(HNL), Istanbul Atatürk Airport (IST), London Heathrow Airport (LHR),
Chicago O’Hare International Airport (ORD), Beijing Capital Interna­
tional Airport (PEK) and Taipei Taoyuan International Airport (TPE) had
efficiencies exceeding 0.9 in 2015. Throughout the whole period and for
all airports, the average bias was 0.014, accounting for about 3% of the
biased efficiency score.
Fig. 5 shows the changes of the average passenger traffic and average
efficiency score of the airports from 2006 to 2015 to grasp the rela­
tionship between them. A similar trend of two variables implies a pos­
itive correlation. The downward trend around 2008 presents the impact
Fig. 4. The change of the volatility of confidence intervals by subsampling size. of the economic crisis. After that, the average airport efficiency increases
gradually. Table 5 lists the results of the second stage gravity model
selected the subsampling size m∗ without a rigorous statistical verifica­ analysis. After the first independent variable names column, the next
tion in line with the practical method suggested by Politis et al. (2001), three columns (1), (2) and (3) display the results of the gravity models
this method had very little effect on the result of the second stage. As with the dependent variable being air passenger traffic. Overall, most of
evidence, all correlation coefficients between bias-corrected efficiencies the variables show statistical significance for the coefficient value. Based
from subsampling size m from 302 (= 40
) 49
) on the significance and positive sign of the GDP per capita variable, the
50⋅377 to 369 (= 50⋅ 377 were
results imply that an income level of the country plays an important role
more than 0.999. In other words, applying the practical rule to select the
in the passenger traffic. Columns (2) and (3) employed the PPML
subsample size successfully corrects the inherent bias of efficiency
method to make a meaningful change in the significance of the GDP

Table 3
Subsampling size m having the smallest volatility for each k and α.
α = 0.1 α = 0.05 α = 0.01
K=1 K=2 K=3 K=1 K=2 K=3 K=1 K=2 K=3

SOV m∗ 0.508 0.876 1.163 0.656 1.052 1.513 1.259 1.944 2.611
m∗ 347 332 324 354 354 302 369 354 354

7
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

Table 4 larger impact than the 25% value reported by Zhang and Zhang (2016).
The efficiency score of airports, 2015, α = 0.05, m = 347, B = 2000. This can be explained by the distinctive characteristics of the used
CODE Country City Effi_bc Bias lower upper dataset. Whereas Zhang and Zhang (2016) investigated China’s do­
bound bound mestic air passenger traffic, we utilized international passenger traffic
AMS Netherlands Amsterdam 0.445 0.006 0.423 0.451 and so the impact is more pronounced.
ATL USA Atlanta 0.996 0.004 0.984 1.000 Columns (4), (5) and (6) show the results when the dependent var­
BKK Thailand Bangkok 0.402 0.019 0.321 0.421 iable is freight volume. Compared to the passenger traffic model, the
CAN China Guangzhou 0.658 0.022 0.583 0.680 significance and magnitude of coefficients of the LCC variables were
CDG France Paris 0.286 0.003 0.265 0.289
CGK Indonesia Jakarta 0.523 0.058 0.164 0.582
decreased. This result can be ascribed to the fact that most LCCs
CVG USA Cincinnati 0.358 0.004 0.342 0.362 concentrate more on the passenger market than on the freight market.
DFW USA Dallas 0.408 0.006 0.389 0.414 Column (6) shows that the FTA variables became significant by
DXB UAE Dubai 0.989 0.011 0.950 1.000 employing the MR term, whereas FTA without the MR term in column
FRA Germany Frankfurt 0.284 0.005 0.257 0.289
(5) was insignificant. While we can expect that the FTA variable has a
GUM Guam Guam 0.077 0.001 0.073 0.078
HKG China Hong Kong 0.975 0.025 0.875 1.000 stronger and more direct impact on freight volumes than on passenger
HKT Thailand Phuket 0.947 0.053 0.759 1.000 traffic, reflecting MR term changed the impact of FTA into being sig­
HNL USA Honolulu 0.995 0.005 0.967 1.000 nificant. This result is different from that of Zhang and Zhang (2016),
IST Turkey Istanbul 0.975 0.025 0.885 1.000 who found that including the MR term did not lead to significant
JFK USA New York 0.805 0.027 0.728 0.832
KIX Japan Osaka 0.535 0.004 0.519 0.539
changes in the model. In our gravity model, the significant effect of
KUL Malaysia Kuala 0.220 0.003 0.211 0.223 introducing MR terms is identified. For the measurement of trade
Lumpur agreement’s effect on trade, there have been considerable debates. The
LAX USA Los Angeles 0.816 0.028 0.703 0.844 obstacle preventing the estimation of the exact effects of trade agree­
LHR UK London 0.995 0.005 0.977 1.000
ments is endogeneity, which is about the direction or causality of the
MIA USA Miami 0.506 0.005 0.489 0.511
MNL Philippines Manila 0.388 0.013 0.344 0.400 impact (Egger et al., 2011). More details about the endogeneity in
MXP Italy Milan 0.231 0.003 0.217 0.233 gravity models are discussed later in this session. The GDP variables
NGO Japan Nagoya 0.351 0.006 0.339 0.357 show fluctuations in the significance level and also changing signs
NRT Japan Tokyo 0.559 0.009 0.539 0.568 throughout models, possibly due to the tendency for some countries to
ORD USA Chicago 0.984 0.016 0.956 1.000
trade relatively larger amounts than the magnitude of their GDPs (Head
OSL Norway Oslo 0.444 0.059 0.005 0.503
PEK China Beijing 0.993 0.007 0.978 1.000 and Mayer, 2014). Also, Kupfer et al. (2017) explained that the corre­
PEN Malaysia Penang 0.392 0.009 0.367 0.401 lation between GDP and air freight is gradually weakening because the
PVG China Shanghai 0.522 0.012 0.462 0.534 portion of manufacture in GDP decreases, but the portion of service
SEA USA Seattle 0.500 0.017 0.461 0.517
industry increase. As recent studies are moving toward using other
SFO USA San 0.301 0.006 0.280 0.306
Francisco practices such as fixed effects model rather than GDP variables, further
SIN Singapore Singapore 0.507 0.009 0.481 0.516 study about the role of the GDP variable on passenger traffic needs to be
SZX China Shenzhen 0.438 0.009 0.392 0.447 conducted (Bensassi et al., 2015).
TPE Taiwan Taipei 0.969 0.031 0.896 1.000 Finally, even though some variables show fluctuations depending on
VIE Austria Wien 0.358 0.014 0.276 0.372
which model was applied, all coefficient values of efficiency variable
YVR Canada Vancouver 0.333 0.008 0.316 0.341
YYZ Canada Toronto 0.244 0.004 0.230 0.248 consistently represent the significant and positive effect on the depen­
dent variables throughout all models (1) ~ (6). On the basis of the
PPML-MR model, a 1% increase in efficiency score increases passenger
traffic and freight volume by 0.48% and 0.62%, respectively. The in­
crease of airport’s efficiency is expected to bring additional passenger
traffic and freight volume in two aspects. First, the increase of airport
efficiency will reduce the cost and time of its users (Henke et al., 2021).
Also, increased airport efficiency can solve some congestions and, in
turn, enhance the service qualities which customers experience (Oum
et al., 2003; Fodness and Murray, 2007). In these ways, the cost and time
reductions and enhanced customer satisfaction ultimately can make a
bigger passenger traffic and freight volume. Second, the airports also
benefit from the increase of efficiency and, in turn, can reduce their
operating cost. Then, by offering significant incentives to air cargo op­
erators and establishing a competitive pricing policy, the airport can
attract a bigger passenger traffics and freight volume (Katarelos and
Lagoudis, 2011).
On the other hand, the impact of efficiency increment can be more
emphasized in a hub airport (Gong et al., 2018). For example, if an
airport is operating tremendous cargoes and with full utilization of its
facilities, i.e., if an airport is crowdy, the increase of its efficiency would
directly attract a much bigger freight volume. On the other hand, Martín
et al. (2013) identified a clear positive relationship between pre-crisis
Fig. 5. Average air passenger and airport efficiency by year. efficiency and cost flexibility. Also, they found that there is a signifi­
cant positive relationship between the variation in passenger traffic and
variables. Column (3), which introduces the MR terms especially for the the efficiency. This result also supports the explanation that the airport’s
FTA(MR) variable, shows that the presence of the FTA increases air efficiency affects its cost flexibility and in turn, passenger traffics.
passenger traffic by 38.1% (i.e.,e0.323 − 1). The presence of an LCC in the Looking into the difference between the dependent variables, pas­
air route increases passenger traffic by 51.1% (i.e.,e0.413 − 1), which is a senger and freight, we can find more practical interpretations. The most
distinctive characteristic is that air passengers are more likely to prefer

8
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

Table 5
The result of the Gravity model.
Independent variable Dependent variable

Passenger Freight

OLS (1) PPML (2) PPML with MR term (3) OLS (4) PPML (5) PPML with MR term (6)

Constant 18.390*** (0.659) 20.172*** (0.672) 19.927*** (0.660) 11.622*** (0.810) 13.401*** (0.850) 13.246*** (0.831)
ln GDP (Counter) − 0.034• (0.020) − 0.081*** (0.019) − 0.067*** (0.019) 0.093*** (0.025) 0.028 (0.026) 0.049• (0.027)
ln GDP per capita (Counter) 0.147*** (0.033) 0.118*** (0.029) 0.091** (0.028) 0.073 (0.044) 0.112** (0.043) 0.111** (0.041)
ln Distance − 0.660*** (0.045) − 0.653*** (0.044) − 0.622*** (0.041) − 0.473*** (0.061) − 0.468*** (0.057) − 0.514*** (0.054)
ln Efficiency 0.361*** (0.0580) 0.479*** (0.059) 0.476*** (0.059) 0.564*** (0.083) 0.594*** (0.090) 0.624*** (0.087)
FTA(MR) 0.455*** (0.070) 0.449*** (0.079) 0.323∗∗∗(MR) (0.089) 0.323*** (0.089) 0.142 (0.102) 0.342∗∗(MR) (0.106)
LCC 0.395*** (0.103) 0.312*** (0.077) 0.413*** (0.154) 0.413** (0.154) 0.227• (0.126) 0.330** (0.127)
Observation 245 245 245 345 345 345
Adjusted R2 0.558 – – 0.293 – –

***: p < 0.01, **: p < 0.05, •: p < 0.1, Parenthesis: standard error.

convenient flight environments such as non-stop flight, attractive to get a neat result based on the obtained data.
airport environment, rich in diversions, or endowed with facilities Therefore, as an alternative approach to explain the cause-and-effect
(Zhang and Zhang, 2002a, 2002b). In contrast, air cargos more care relationship between airport’s efficiency and the two air traffics, we
about the condition of shipments such as a change of aircraft, whether it construct a definite and irreversible logic. In this study, we claim that
needs re-packaging or not, or trans-shipment costs. Also, the flow of air the increase of airport’s efficiency is expected to attract a bigger pas­
passenger is more balanced between the bilateral orientation and senger traffic and freight volume in two aspects. The increase of airport
destination since passengers tend to have round-trip flights. However, efficiency will reduce the cost and time of its users. Also, increased
cargo flows are more unbalanced because it is commonly unidirectional. airport efficiency can solve some congestions and, in turn, enhance the
Instead, they show ‘big circle’ networks connecting a number of related service qualities which customers experience. Then, these two factors
markets and production sites. This difference between two dependent ultimately can contribute to attracting bigger air traffics. Also, the air­
variables partially can explain the passenger model’s relatively bigger ports, which benefit from the efficiency increment and reduce their
significance and magnitude of coefficients of country-specific variables operating cost, can offer significant incentives to air cargo operators and
such as GDP, GDP per capita and Distance. Distance variable shows establish a competitive pricing policy to attract bigger air traffics.
negative and significant impacts throughout all models and dependent However, the reverse directional logic, i.e., ‘the increase of passenger
variables. However, the magnitude of the impact is bigger in the pas­ traffic and freight volume would cause user’s cost and time reduction
senger model than in the freight model, which means passenger traffic is and enhanced customer satisfaction, and in turn, enhance the airport
more sensitive to distance than freight volume is. Top 50 air route data efficiency’, does not make sense. Also, the increase of air traffics does
of Incheon airport used in this analysis show that during the recent five not logically cause the price to go down. Therefore, since we detected
years, 2012–2016, top 10 destinations are all Asia countries such as the positive relationship between the two value, airport’s efficiency and
Hong Kong, Tokyo, Bangkok and Pudong except Los Angeles once air traffics, and the reverse direction does not make sense, we can
ranked 10th in 2012. On the other hand, in the freight models, conclude that the increase of airport’s efficiency positively affects both
cost-related variables such as Efficiency and FTA show much more sig­ passenger traffic and freight volume.
nificant coefficients than passenger model does. This is a similar result to Several gravity model studies such as Magee (2003), Baier and
Murphy et al. (1989), who claimed that the introduction of FTA will Bergstrand (2007), and Egger et al. (2011) claim that FTA variables can
reduce the costs and time of cross-border shipments and increase the cause unstable results since the effect of the trade agreements on trade is
trade efficiency between the two countries. Also, freight-oriented air­ endogenous. Baier and Bergstrand (2007) explained that some existing
ports can consider actively investing in the airport’s infrastructure models in the literature ignored the endogeneity of FTAs so did not
because the effect is more emphasized in freight volume. However, this provide stable estimates. They also found that using pair-fixed-effects
result can be different according to the used air route data because the method with panel data can address the issue of endogeneity. Egger
pattern of air freight and passenger services is regionally diverse (Zhang et al. (2011) evaluated the general equilibrium-consistent effects of a
and Zhang, 2002b). For example, Incheon airport traffic used in this preferential trade agreement (PTA) on bilateral trade and demonstrated
analysis has a significant effect on Pan-Asia and trans-Pacific air pas­ that ignoring the endogenous issue in PTAs is critical. Although recent
senger and freight traffic. Also, in long-distance routes, Asian airlines studies recommend using fixed-effect models in order to capture the
often operate ‘combi’ aircraft which have a wide belly and thus can multilateral resistance, endogeneity issue, or even disaggregated data
accommodate both passenger and freight simultaneously. In contrast, in problem, we did not adopt the fixed-effect model. Instead, we employed
the U.S., there are several freight-oriented airlines such as FedEx and other alternative methods such as MR term and PPML to correct the
UPS using pure freight carriers. biases in the gravity model. As gravity literature stated, the critical
On the other hand, we should consider the simultaneity issue among drawback of the fixed-effect model is that it cannot explain other
several issues which can cause endogeneity problems. In other words, time-invariant effects since all time-invariant effects are absorbed by
we need to clarify a cause-and-effect relationship between airport’s ef­ fixed-effect terms (Yotov et al., 2016). Therefore, as the main interest of
ficiency and the two air traffics. In this paper, we concluded that the this research is to investigate the impact of airport efficiency on bilateral
increase of airport’s efficiency would cause an increase of bilateral air flows, we did not use fixed effect terms but used MR term and PPML to
traffics. To obtain the statistical evidence between the airport’s effi­ handle inherent biases of the gravity model. To examine the effect of
ciency and both air traffics, we conducted Granger causality test and airport efficiency on trade accurately considering the endogeneity
Transfer entropy method. However, because the time period of our data problem, future research needs to employ more advanced methodology
set is too short to conduct the two time-series analysis, we could not such as simultaneous equation models.
obtain a significant result. Another possible solution to identify the
endogeneity issue is to apply ‘instrument variable (IV)’ into the model.
However, the adoption of proper IVs is fairly complex and it is difficult

9
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

5. Conclusion its users, and policy makers.


The results demonstrated that airport efficiency has a significant and
The efficiencies of airports and analysis on global trade have been positive impact on bilateral passenger traffic and freight volume
major research themes in the aviation industry. Studies on the former throughout all models. In particular, the impact of efficiency was more
have focused on the methodology of measuring efficiencies and also how influential on freight volume than on passenger traffic. On the basis of
to enhance airport efficiency. On the other hand, studies on the latter the most robust model, PPML-MR, we estimated that the airport effi­
have centered on which factors are determinants of global passenger ciency elasticity of passenger traffic and freight volume was 0.48 and
traffic and trade volume. However, no linkage studies in the literature 0.62, respectively, indicating that a 1% increase in airport efficiency
have examined whether airport efficiency affects bilateral trade and increases passenger traffic and freight volume by 0.48% and 0.62%,
passenger traffic. To fill this research gap, this study investigated the respectively. In agreement with the literature, our results showed that
relationship between airport efficiency and bilateral flow, including LCCs played an important role in the aviation industry, but the effect
passenger traffic and freight volume. was stronger in the passenger market than in the freight market. This
In the air transport management field, there have been numerous result supports the global trend that as numerous LCCs are newly
DEA studies that examine airport’s efficiency without consideration for emerging, the air passenger market is dramatically growing. On the
the sampling bias. Because the ignorance of this bias can expose stake­ other hand, the impact of airport efficiency and trade agreement is
holders to wrong information about benchmark points and cause wrong bigger when it is estimated in the air freight model. Consequently, the
investments, correcting these biases in the methodology is important. To result of our research supports the tremendous efforts of government
eliminate the inherent bias caused by the deterministic nature of DEA, and authority to enhance the efficiency of their airport. Given that the
we introduced the SBM-DEA model combined with the bootstrapping air passenger traffic and air freight volume significantly affect both
technique. By applying the subsampling process, the bias-corrected ef­ regional and national economy, even though the magnitude of efficiency
ficiency scores of global airports were obtained. Then, to investigate the elasticity is not great, it is important to enhance the airport efficiency to
impact of airport efficiency on bilateral flows, the gravity model was accommodate growing air traffic. These findings may provide important
developed by incorporating the MR term and also employing the PPML managerial implications to stakeholders such as customers, airport au­
method to solve the heteroscedasticity problem. Since the application of thorities, and policy makers.
biased methodology can offer wrong signals to the air market and
stakeholders, it is crucial to adopt robust models. By eliminating the Acknowledgements
biases in the methodology, we expect that this model can estimate the
impact of airport efficiency on air passengers and freight precisely and, This work was supported by INHA UNIVERSITY Research Grant.
in turn, offer correct field information to stakeholders such as airports,

Appendix. The stepwise algorithm for bootstrapping efficiency scores from the SBM with the subsampling technique

In this appendix, we present the detailed algorithm for bootstrapping efficiency scores from the SBM with the subsampling technique and explain
how the bias-corrected efficiency score is obtained. The range of the M is not critical as long as it is not too narrow (Politis et al., 2001). Simar and
Wilson (2011) conducted Monte Carlo experiments and showed that the method of Politis et al. (2001) works well in finite samples using 49 subsample
{ n 2n 3n }
sizes m ∈ Mn = 50 , 50, 50,⋯, 49n
50 . Following Simar and Wilson (2011), we performed 2000 bootstrap replications for each of 49 subsample sizes mj ∈
Mn , where mj is jth subsample by size. The coverage of running standard deviation k ∈ {1, 2, 3} was set. k is used to determine how many subsample
sizes are included to calculate the standard deviation of the confidence interval’s end points. For each (1 − α) × 100-percent bootstrap confidence
interval, α ∈ {0.1, 0.05, 0.01} were used. Detailed explanations with examples are given after the generalized algorithm. Building on the work of
Politis et al. (2001) and Simar and Wilson (2011), we added the third line to obtain the value of sum of VIj (SOVj ) to clarify the procedure of selecting
m:

̂
i. For j ∈ {1, 2, ⋯, J}, create new samples mj with the subsampling method and compute the confidence intervals for the efficiency score ̂ θ
calculated from the resampled data at the (1 − α) × 100-percent confidence level, resulting in endpoints Ij,low and Ij,up .
ii. For a small integer k, let VIj be the sum of the standard deviation of the lower endpoints {Ij− k,low , …, Ij+k,low } and the standard deviation of the
upper endpoints {Ij− k,up , …, Ij+k,up }.

iii. For n DMUs, compute the sum of VIj (SOVj ), ni=1 (VIj )i .
iv. Pick m corresponding to the smallest volatility index SOVj and report [Ij∗ ,low , Ij∗ ,up ] as the final confidence interval.

For example, let’s assume that we have k = 1 and α = 0.05. For the specific sample size mj = 25n50 , n confidence intervals for n DMUs are calculated .
All 95% confidence intervals are calculated from B bootstrapped DEA scores by sorting them by size. Next, according to the level of k = 1, we get two
( ) ( 26n)
more confidence intervals for all DMUs with the adjacent sample sizes mj− 1 = 24n 50 and mj+1 = 50 repeating the bootstrapping calculation process.
Then, for all DMUs, standard deviations of three upper and lower confidence intervals are calculated. After that, summating the standard deviations of
upper and lower confidence intervals, we obtain VIj . Finally, summating n VIj of n DMUs, we get SOVj for the sample size m = 25n 50 . If k = 1 and the
number of M is 49, we can calculate total 47 SOVj for each of 47 sample sizes (see the second line in the calculation algorithm). For k = 2 and 3, 45 and
43 SOVj are obtained.
Bădin et al. (2014) demonstrated several alternatives to determine the sampling size m based on the above algorithm. One of the simplest ways is to
take a sample size m that appears most often for each k and α, another is to select it by calculating the average value of m. In this study, to estimate the
value of bias-corrected efficiencies as an empirical study, the subsample size m is chosen by comparing SOVj . The SOVj is an index representing the
volatility of confidence intervals. Based on the parameters, including k and α, we identified the subsample size m having the smallest SOVj . Again, this
is the process of finding the most ‘stable’ point having the smallest variance of confidence intervals (volatility), not the width.

10
A.-H. Jo and Y.-T. Chang Journal of Air Transport Management 108 (2023) 102364

References Limao, N., Venables, A.J., 2001. Infrastructure, geographical disadvantage, transport
costs, and trade. World Bank Econ. Rev. 15 (3), 451–479.
Lin, L.C., Hong, C.H., 2006. Operational performance evaluation of international major
Agnosteva, D.E., Anderson, J.E., Yotov, Y.V., 2014. Intra-national Trade Costs:
airports: an application of data envelopment analysis. J. Air Transport. Manag. 12
Measurement And Aggregation (No. W19872). National Bureau of Economic Research.
(6), 342–351.
Anderson, J.E., 1979. A theoretical foundation for the gravity equation. Am. Econ. Rev.
Lozano, S., Gutiérrez, E., 2011. Slacks-based measure of efficiency of airports with
69 (1), 106–116.
airplanes delays as undesirable outputs. Comput. Oper. Res. 38 (1), 131–139.
Anderson, J.E., Van Wincoop, E., 2003. Gravity with gravitas: a solution to the border
Magee, C.S., 2003. Endogenous preferential trade agreements: an empirical analysis.
puzzle. Am. Econ. Rev. 93 (1), 170–192.
Contrib. Econ. Anal. Pol. 2 (1).
ATRS, Airport Benchmarking Report, 2005-2018. Air Transport Research Society,
Márquez-Ramos, L., Martínez-Zarzoso, I., Pérez-García, E., Wilmsmeier, G., 2011.
Headquartered at the Robert H. Smith School of Business, University of Maryland.
“Special issue on Latin-american research” maritime networks, services structure and
Bădin, L., Daraio, C., Simar, L., 2014. Explaining inefficiency in nonparametric
maritime trade. Network. Spatial Econ. 11 (3), 555–576.
production models: the state of the art. Ann. Oper. Res. 214 (1), 5–30.
McCallum, J., 1995. National borders matter: Canada-US regional trade patterns. Am.
Baier, S.L., Bergstrand, J.H., 2007. Do free trade agreements actually increase members’
Econ. Rev. 85 (3), 615–623.
international trade. ? Journal of international Economics 71 (1), 72–95.
Merkert, R., Odeck, J., Brathen, S., Pagliari, R., 2012. A review of different
Baier, S.L., Bergstrand, J.H., 2009. Bonus vetus OLS: a simple method for approximating
benchmarking methods in the context of regional airports. Transport Rev. 32 (3),
international trade-cost effects using the gravity equation. J. Int. Econ. 77, 77–85.
379–395.
Bensassi, S., Márquez-Ramos, L., Martínez-Zarzoso, I., Suárez-Burguet, C., 2015.
Murphy, P., Dalenberg, D., Daley, J., 1989. Improving international trade efficiency:
Relationship between logistics infrastructure and trade: evidence from Spanish
airport and air cargo concerns. Transport. J. 27–35.
regional exports. Transport. Res. Pol. Pract. 72, 47–61.
Nordas, H.K., Piermartini, R., 2004. Infrastructure and Trade. World Trade Organization
Bergstrand, S.B.H., 2010. Approximating general equilibrium impacts of trade
Staff Working Paper. ERSD-2004-04.
liberalizations using the gravity equation. Grav. Model Int. Trade: Adv. Appl. 88.
Oum, T.H., Yu, C., Fu, X., 2003. A comparative analysis of productivity performance of
Bickel, P.J., Sakov, A., 2008. On the choice of m in the m out of n bootstrap and
the world’s major airports: summary report of the ATRS global airport
confidence bounds for extrema. Stat. Sin. 967–985.
benchmarking research report—2002. J. Air Transport. Manag. 9 (5), 285–297.
Blonigen, B.A., Wilson, W.W., 2008. Port efficiency and trade flows. Rev. Int. Econ. 16
Pels, E., Nijkamp, P., Rietveld, P., 2003. Inefficiencies and scale economies of European
(1), 21–36.
airport operations. Transport. Res. E Logist. Transport. Rev. 39 (5), 341–361.
Bottasso, A., Conti, M., de Sa Porto, P.C., Ferrari, C., Tei, A., 2018. Port infrastructures
Politis, D.N., Romano, J.P., Wolf, M., 2001. On the asymptotic theory of subsampling.
and trade: empirical evidence from Brazil. Transport. Res. Pol. Pract. 107, 126–139.
Stat. Sin. 1105–1124.
Bougheas, S., Demetriades, P.O., Morgenroth, E.L., 1999. Infrastructure, transport costs
Poole, J., 2010. Business Travel as an Input to International Trade. UC Santa Cruz.
and trade. J. Int. Econ. 47 (1), 169–189.
Portugal-Perez, A., Wilson, J.S., 2012. Export performance and trade facilitation reform:
Chang, L.Y., 2014. Analysis of bilateral air passenger flows: a non-parametric
hard and soft infrastructure. World Dev. 40 (7), 1295–1307.
multivariate adaptive regression spline approach. J. Air Transport. Manag. 34,
Pöyhönen, P., 1963. A tentative model for the volume of trade between countries.
123–130.
Weltwirtschaftliches Archiv 90, 93–99.
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision
Rupp, N.G., 2009. Do carriers internalize congestion costs? Empirical evidence on the
making units. Eur. J. Oper. Res. 2 (6), 429–444.
internalization question. J. Urban Econ. 65 (1), 24–37.
Clark, X., Dollar, D., Micco, A., 2004. Port efficiency, maritime transport costs, and
Sánchez, R.J., Hoffmann, J., Micco, A., Pizzolitto, G.V., Sgut, M., Wilmsmeier, G., 2003.
bilateral trade. J. Dev. Econ. 75 (2), 417–450.
Port efficiency and international trade: port efficiency as a determinant of maritime
Dmitry, P., 2012. Airport benchmarking and spatial competition: a critical review.
transport costs. Marit. Econ. Logist. 5 (2), 199–218.
Transport Telecommun. 13 (2), 123–137.
Santos, G., Robin, M., 2010. Determinants of delays at European airports. Transp. Res.
Egger, P., Larch, M., Staub, K.E., Winkelmann, R., 2011. The trade effects of endogenous
Part B Methodol. 44 (3), 392–403.
preferential trade agreements. Am. Econ. J. Econ. Pol. 3 (3), 113–143.
Silva, J.S., Tenreyro, S., 2006. The log of gravity. Rev. Econ. Stat. 88 (4), 641–658.
Endo, N., 2007. International trade in air transport services: penetration of foreign
Simar, L., Wilson, P.W., 1998. Sensitivity analysis of efficiency scores: how to bootstrap
airlines into Japan under the bilateral aviation policies of the US and Japan. J. Air
in nonparametric frontier models. Manag. Sci. 44 (1), 49–61.
Transport. Manag. 13 (5), 285–292.
Simar, L., Wilson, P.W., 2000. A general methodology for bootstrapping in non-
Fasone, V., Zapata-Aguirre, S., 2016. Measuring business performance in the airport
parametric frontier models. J. Appl. Stat. 27 (6), 779–802.
context: a critical review of literature. Int. J. Prod. Perform. Manag. 65 (8),
Simar, L., Wilson, P.W., 2007. Estimation and inference in two-stage, semi-parametric
1137–1158.
models of production processes. J. Econom. 136 (1), 31–64.
Feenstra, R.C., 2004. Advanced International Trade, Princeton (NJ) et al. Int. Econ.
Simar, L., Wilson, P.W., 2008. Statistical inference in nonparametric frontier models:
(Bachelor level) Forms Teach. 4.
recent developments and perspectives. In: The Measurement of Productive Efficiency
Fodness, D., Murray, B., 2007. Passengers’ expectations of airport service quality. J. Serv.
and Productivity Growth, pp. 421–521.
Market. 21 (7), 492–506.
Simar, L., Wilson, P.W., 2011. Inference by the m out of n bootstrap in nonparametric
Gillen, D., Lall, A., 1997. Developing measures of airport productivity and performance:
frontier models. J. Prod. Anal. 36 (1), 33–53.
an application of data envelopment analysis. Transport. Res. E Logist. Transport.
Tinbergen, J., 1962. Shaping the World Economy: Suggestions for an International
Rev. 33 (4), 261–273.
Economic Policy. Twentieth Century Fund, New York.
Glick, R., Rose, A.K., 2002. Does a currency union affect trade? The time-series evidence.
Tone, K., 2001. A slacks-based measure of efficiency in data envelopment analysis. Eur.
Eur. Econ. Rev. 46 (6), 1125–1151.
J. Oper. Res. 130 (3), 498–509.
Gong, Q., Wang, K., Fan, X., Fu, X., Xiao, Y.B., 2018. International trade drivers and
Tone, K., 2002. A slacks-based measure of super-efficiency in data envelopment analysis.
freight network analysis-The case of the Chinese air cargo sector. J. Transport Geogr.
Eur. J. Oper. Res. 143 (1), 32–41.
71, 253–262.
Wanke, P.F., 2012. Efficiency of Brazil’s airports: evidences from bootstrapped DEA and
Grosso, M.G., 2012. Air passenger transport in the APEC: regulatory impacts and
FDH estimates. J. Air Transport. Manag. 23, 47–53.
prospects for Asia Pacific integration. J. Econ. Integrat. 312–327.
Wilson, J.S., Mann, C.L., Otsuki, T., 2003. Trade facilitation and economic development:
Hausman, W.H., Lee, H.L., Subramanian, U., 2013. The impact of logistics performance
a new approach to quantifying the impact. World Bank Econ. Rev. 17 (3), 367–389.
on trade. Prod. Oper. Manag. 22 (2), 236–252.
Wilson, J.S., Mann, C.L., Otsuki, T., 2005. Assessing the benefits of trade facilitation: a
Head, K., Mayer, T., 2014. Gravity equations: workhorse, toolkit, and cookbook. Handb.
global perspective. World Econ. 28 (6), 841–871.
Int. Econ. 4, 131–195. Elsevier.
Yoshida, Y., Fujimoto, H., 2004. Japanese-airport benchmarking with the DEA and
Henke, I., Esposito, M., della Corte, V., del Gaudio, G., Pagliara, F., 2021. Airport
endogenous-weight TFP methods: testing the criticism of overinvestment in
efficiency analysis in europe including user satisfaction: a non-parametric analysis
Japanese regional airports. Transport. Res. E Logist. Transport. Rev. 40 (6),
with dea approach. Sustainability 14 (1), 283.
533–546.
Hwang, C.C., Shiao, G.C., 2011. Analyzing air cargo flows of international routes: an
Yotov, Y.V., Piermartini, R., Monteiro, J.A., Larch, M., 2016. An Advanced Guide to
empirical study of Taiwan Taoyuan International Airport. J. Transport Geogr. 19 (4),
Trade Policy Analysis: the Structural Gravity Model. World Trade Organization,
738–744.
Geneva.
IATA International Air Transport Association, 2018. Annual Review 2017, 73rd Annual
Zhang, A., Zhang, Y., 2002a. Issues on liberalization of air cargo services in international
General Meeting.
aviation. J. Air Transport. Manag. 8 (5), 275–287.
ICAO International Civil Aviation Organization, 2018. Annual Report 2017. ICAO
Zhang, A., Zhang, Y., 2002b. A model of air cargo liberalization: passenger vs. all-cargo
Annual Report of the Council.
carriers. Transport. Res. E Logist. Transport. Rev. 38 (3–4), 175–191.
Katarelos, E., Lagoudis, I.N., 2011. Greek airport capacity utilisation: what about
Zhang, Y., Zhang, A., 2016. Determinants of air passenger flows in China and gravity
airfreight? Int. J. Aviat. Manag. 1 (1–2), 124–139.
model: deregulation, LCCs, and high-speed rail. J. Transport Econ. Pol. 50 (3),
Kneip, A., Simar, L., Wilson, P.W., 2008. Asymptotics and consistent bootstraps for DEA
287–303.
estimators in nonparametric frontier models. Econom. Theor. 24 (6), 1663–1697.
Zhang, Y., Lin, F., Zhang, A., 2018. Gravity models in air transport research: A survey and
Kupfer, F., Meersman, H., Onghena, E., Van de Voorde, E., 2017. The underlying drivers
an application. Handb. Int. Trade Transport. 141.
and future development of air cargo. J. Air Transport. Manag. 61, 6–14.

11

You might also like