assessing the productivity of the italian hospitality sector: a post-wdea pooled-truncated and...
TRANSCRIPT
Assessing the productivity of the Italian hospitality sector:a post-WDEA pooled-truncated and spatial analysis
Claudio Detotto • Manuela Pulina • Juan Gabriel Brida
� Springer Science+Business Media New York 2013
Abstract This paper analyses the productivity of the
hospitality sector (hotel and restaurants) in Italy at a
regional level by using a mix of non-parametric and
parametric approaches. A novel pooled-truncated and
spatial analysis is employed, based upon a window data
envelopment analysis (WDEA), where pure technical
efficiency is computed. The WDEA results show that
Lombardy is the best relative performer. However, overall
Italian regions reveal important sources of inefficiency
mostly related to their inputs. As a post-WDEA, the
pooled-truncated estimation indicates that the rate of util-
isation and regional intrinsic features positively affect
hospitality efficiency. Nevertheless, the spatial analysis
does not support evidence of spill-over effects amongst
Italian regions.
Keywords Regional hospitality sector � Dynamic
window data envelopment analysis � Double
bootstrap � Pooled-truncated regression � Spatial
heterogeneity
JEL Classification C14 � C24 � L83 � R11
1 Introduction
During the second half of 2007, the sub-prime crisis
extended to many other sectors of the economy including
tourism. As the World Travel and Tourism Council
(WTTC 2010) reports, global travel and tourism (T&T)
economy GDP declined by 4.8 % in 2009, causing the loss
of almost 5 million jobs, while investment shrank by over
12 %. Nonetheless, this economic sector still employed
over 235 million people worldwide (8.2 % of all employ-
ment) and generated 9.4 % of the world’s GDP. The more
recent figures released by the World Tourism Organization
(UNWTO 2012) showed a slowdown in international
tourism arrivals with an overall growth of 4 % in 2011.
This setback was due to the on-going economic and
political instability. Nevertheless, in this economic turmoil,
Europe and Italy recorded a relatively positive perfor-
mance, with growth levels above the average figure reg-
istered worldwide.
The contribution of tourism demand as a driver of
economic growth is well established (Bimonte et al.
2012). As pointed by Federalberghi-Mercury (2010), the
Italian hospitality sector plays an important role as a
revenue generator. In terms of number of hotel rooms, it
ranks fourth after the United States, Japan and China
(WTTC 2010) and amongst the European countries, Italy
has the leadership in terms of hotel dimension and
quality (number of stars). Yet, Italian hospitality is
characterised by a strong seasonality, given that its rate
of utilisation (40 %) is much lower than the global
leading countries (e.g. Japan, 74 %; France, 61 %; U.S.,
60 %; Spain, 53 %). Such rate of utilization decreased
by 6.8 % during the period 2000–2004, although tourist
arrivals and nights of stay rose by 7.4 and 1.9 %,
respectively (ISTAT 2011).
C. Detotto (&) � M. Pulina
Researcher at Centro Ricerche Economiche Nord e Sud
(CRENoS) and Department of Economics (DiSEA), Universita
di Sassari, Via Torre Tonda, 34, 07100 Sassari, Italy
e-mail: [email protected]
M. Pulina
e-mail: [email protected]
J. G. Brida
Department of Economics at the School of Economics and
Management, Free University of Bolzano, Piazza
dell’Universita, 39100 Bolzano, Italy
e-mail: [email protected]
123
J Prod Anal
DOI 10.1007/s11123-013-0371-x
Overall, Italy has been defined as a pioneer in the hos-
pitality industry alongside other main Mediterranean des-
tinations (namely France and Spain) but while it is
characterised by a very long life cycle, it also displays quite
scattered renewal models within each of the regions (e.g.
Manera Erbina et al. 2010). As emphasised in Federal-
berghi-Mercury (2012), the structural characteristics of
Italian hotels at a regional level show that most of the
infrastructures are located in the North. However, over the
years, there has been a redistribution of the supply within
the country, characterised by an increase in hotel capacity
in the southern regions, where, between 2000 and 2010,
hotel accommodation proved to be particularly dynamic
(e.g. Basilicata, ?87.9 %; Sicily, ?58.5 %; Apulia,
?58.5 %; Calabria, ?57.3 %).
Within this setting, it seems interesting to investigate the
productivity of this rather heterogeneous Italian hospitality
sector on a regional basis. The question is particularly
important in the light of an increasing awareness of sus-
tainability issues that challenge the need for a further
expansion of tourism infrastructure (Bruni et al. 2011). The
question is particularly important as, over the time span
under investigation (2000 and 2004), overall Italian supply
capacity grew by 7.9 %, reaching more than two million
hotel beds in 2004 (ISTAT 2011).
One of the objectives of this paper is to investigate
whether the expansion of this capacity is supported by
productivity performance. As argued by Brown and Dev
(1999, 2000), there is no common agreement on the defi-
nition of productivity, which still remains a rather vague
concept. This is particularly true for the hospitality indus-
try, which is characterised by similarities with the manu-
facturing production process but also by distinctive
features that relate to its heterogeneous and intangible
nature. Within a production function setting, one needs to
take into account a key set of inputs and outputs employed
in the hospitality sector. In this respect, a further distinction
should also be made on the relevant methodological
frameworks, either microeconomic or macroeconomic, that
can help one to better identify the indicators of produc-
tivity. From an empirical point of view, data availability is
an important issue to be addressed.
At a microeconomic level, data are available either from
accounting books or through ad hoc questionnaires
designed to obtain very detailed information on both
demand and supply, as well as on those environmental and
intangible factors that may contribute to influence the
overall performance of the sector. Unfortunately, since
hotels and other types of accommodation produce imper-
fectly substitutable outputs, the firms operate on different
production frontiers, which makes it almost impossible to
compare them. This problem could be overcome by col-
lapsing firms’ heterogeneity into few aggregated indices,
mainly financial data such as value added, total revenues,
total costs and total investments. How to select these
variables is not trivial since it can drive the analysis and the
results could depend on the data. For an extensive study
about misspecification due to variable selection refer to
Melao (2005).
In present study, widely accepted indicators are used in
line with the existing empirical research. Once the objec-
tive measures of performance have been identified, it is
possible to conduct an empirical investigation assuming the
actual productivity of the sector under analysis. Several
approaches concerning data and methodological frame-
work can be used to measure productivity. The simplest
measures of productivity consist of indices and perfor-
mance indicators, such as the ratio of real outputs over real
inputs. However, more sophisticated tools, both parametric
(e.g. Deterministic Frontier Analysis or the Stochastic
Frontier Analysis) and non-parametric (Data Envelopment
Analysis) are also employed in order to calculate the effi-
ciency of a sector that actually incorporates the concept of
the production possibility frontier. The greater the output
for a given input, or the lower the input for a given output,
the more efficient the sector is.
To date, empirical research papers have mainly focused
on the productivity in the manufacturing sector, health
services, educational institutions, the services sector and
private organisations such as banks. On balance, less
attention has been given to analysing hospitality efficiency
and, more in general, tourism activity from a supply side
perspective (see Barros 2005; Wanhill 2011; Pulina et al.
2010; Brida et al. 2012). To this aim, the present paper
employs a novel combined approach involving three dis-
tinctive steps of research, underpinned to a macroeconomic
framework. First, a standard window data envelopment
analysis (WDEA) is run following the seminal papers by
Charnes et al. (1978), Banker et al. (1984), and Charnes
et al. (1985). WDEA is a non-parametric panel approach
that can be viewed as an appropriate tool to analyse the
productivity level of a group of Decision Making Units
(DMUs) with respect to its own performance over time, as
well as the performance of the relatively most productive
decision units within the sample set. Specifically, a macro
investigation is run by analysing Italian hospitality eco-
nomic efficiency at a regional level and its dynamic evo-
lution over time.1 Second, a double bootstrap procedure
and a pooled-truncated regression are run. This parametric
implementation adds new knowledge to policy makers on
the main factors that influence efficiency of the regional
1 As argued by Diewert and Mendoza (1995), DEA is highly
sensitive to data errors and outliers. Notably, since aggregation
alleviates measurement errors at the individual level, using regional
data allows one to reduce such biases.
J Prod Anal
123
hospitality sector. In this respect, the present paper expands
on other non-parametric studies on Italian hospitality (e.g.
Pulina et al. 2010; Brida et al. 2012). The last step of the
research involves a spatial heterogeneity investigation to
test the existence of possible spill-over effects amongst
Italian regions. This latter analysis helps public and private
agents to capture the actual degree of networking and joint
marketing strategies existing amongst the regions.
It is important to highlight that WDEA provides relative
efficiency scores only. This means that enlarging the
sample size can move the production frontier and hence
most efficient regions may be judged as inefficient. Para-
doxically, the WDEA procedure is ideal to detect ineffi-
cient DMUs since they still remain inefficient although
new DMUs are included. Besides, the present study does
not able one to analyse systemic factors that affect Italian
tourism supply and/or the efficiency of Italian sector with
respect to other European countries (that actually is out of
the scope of the present investigation). The proposed
framework allows one to gain insight into Italian regional
differences in tourism productivity proposing a novel post-
DEA technique, that can be considered as a benchmark for
further studies in the field.
The paper is organised as follows. In the next section, an
updated literature review is provided. In the third section,
the methodology adopted is highlighted. The fourth section
provides a DEA analysis for the hotel sector. The fifth and
sixth sections present post-DEA results. Finally, some
policy implications are drawn in the final conclusions.
2 An updated literature review
DEA and post-DEA have been widely used in the litera-
ture. These approaches have been adopted in the empirical
literature in recent years to analyse the efficiency in the
manufacturing sector (e.g. Kravtsova 2008), health services
(e.g. Olesen and Petersen 2002), education and institutions
(Robert et al. 2010), the services sector (Prado Lorenzo and
Garcıa Sanchez 2007) and private organisations such as
banks. Emrouznejad et al. (2008) provide an ample litera-
ture review on the most popular researched areas.
Less attention has been given to the hospitality sector,
which plays an important role in countries highly specia-
lised in tourism. Barros and Alves (2004), Barros (2005),
and more recently Pulina et al. (2010), and Assaf and
Agbola (2011), provide an extensive literature review on
efficiency in the hotel sector. One of the main features of
the reviewed empirical studies is the cross-section dimen-
sion and the relative low number of observations.
As a further update to this strand of the literature, similar
analytical features have been encountered for. For exam-
ple, Neves and Lourenco (2009) use a static input-oriented
DEA model to determine the efficiency of a sample of
worldwide hotel companies during the period 2000–2002.
The authors note that the majority of the hotel companies
are characterised by decreasing return-to-scale. Further-
more, specialised hotel companies perform better than
those characterised by a diversification strategy. Moriarty
(2010) applies the DEA method to establish relative tech-
nical and scale efficiency of hospitality divisions in New
Zealand during the period 1999–2003. He observes that the
majority of the hospitality divisions exhibit increasing
return-to-scale. Shuai (2009) analyses the impact of inter-
net marketing on hotel performance. It is found that
internet marketing (e-communication and e-transaction
orientations) is positively associated with hotel efficiency.
A possible rationale is that hotels’ efficiency is driven by
an increase in tourism demand due to a reduction in
‘‘search’’ price. Assaf and Cvelvar (2010) analyse the
performance of Slovenian hotels over the period
2005–2007. Bernini and Guizzardi (2010) employ a sto-
chastic translog production function to explore the effi-
ciency of a balanced panel of 414 Italian hotels over the
period 1998–2005. They find that efficiency is enhanced by
both firms’ location and the tourism vocation of a desti-
nation. A significant contribution to efficiency relates to the
firms’ location in seaside cities but especially in arts cities.
Italian business corporations are mainly characterised by a
relevant use of the labour factor of production and
decreasing return-to-scale. Sirirak et al. (2011) find an
empirical evidence of the positive influence of ICT adop-
tion on hotel performance in Taiwan. Shuai and Wu (2011)
make use of a DEA and a grey entropy approach to eval-
uate whether internet marketing affects the operating per-
formance of 48 international hotels in Taiwan for the years
2006 and 2007. Results suggest that hoteliers have to adopt
a more strategic approach to exploit the e-market oppor-
tunities by first preparing the ground for direct contact with
customers. Assaf and Agbola (2011) via a DEA double
bootstrap method estimate the performance of a total of 31
Australian hotels for the period 2004–2007. The findings
show that larger hotels and those located in cities are more
efficient than those in suburban and regional areas.
The majority of the reviewed studies relate to the per-
formance of a sample of hospitality firms. Less attention is
paid to the overall and comparative efficiency of a specific
geographical area. By applying DEA methods on aggre-
gated data, for example, Cracolici and Nijkamp (2006)
calculated a provincial efficiency score and a ranking of
efficient provinces in Italy while Cracolici (2008) explore
tourism competitiveness of Italian provinces for the years
1998 and 2001 and observe a weak decrease of efficiency
over the years. Suzuki et al. (2011) extend the work by
Cracolici (2008) and employ Euclidean Distance Minimi-
zation to investigate Italian provinces efficiency as tourist
J Prod Anal
123
destinations in 2001. The findings show that the perfor-
mance of many Italian provinces can be improved con-
siderably. Barros et al. (2011), analyse the determinants of
the technical efficiency of leading French tourism regions.
They find that the sea, sun and strategy based on the beach
endowment, along with the presence of museums and
monuments, play a relevant role in explaining efficiency in
French regions. Molina-Azorin et al. (2011) use a multi-
level approach and hierarchical linear models to show that
Spanish hotel performance is highly influenced by its
location.
To date, one of the main shortcomings of recently
published studies relates to the use of a relatively low
number of observations, although there are a few
exceptions (e.g. Cracolici 2008; Neves and Lourenco
2009, Bernini and Guizzardi 2010; Suzuki et al. 2011).
The present study stands as a novel example of a more
robust WDEA in analysing regional hospitality produc-
tivity and the main causes of economic inefficiency,
combined with a spatial analysis investigation to estab-
lish possible spill-over effects. To the authors’ knowl-
edge, so far, no study has been undertaken combining
WDEA with a spatial econometric investigation for the
hospitality sector.
3 Methodology: WDEA and post-WDEA
3.1 Window data envelopment analysis
In the literature two main DEA models are considered: the
DEA-CCR model developed by Charnes et al. (1978),
which assumes that all DMUs are operating at constant
returns to scale (CRS); the DEA-BCC model, developed by
Banker et al. (1984), which hypothesises variable returns to
scale (VRS).
The DEA approach estimates the (unknown) efficient
production frontier that allows one to calculate and com-
pare a firm’s efficiency to its own benchmark. Different
from a parametric approach, which requires an a priori
specification of the functional form of the production
function as well as an a priori hypothesis on the disturbance
term, DEA is a flexible linear technique that reduces a
multiple input–output productive structure into an easier
virtual uni-input–output analysis. As an additional advan-
tage, the economic variables of interest can be used jointly
despite their measure scale (Koksal and Aksu 2007).
Within a given sample of productive units, a subgroup will
achieve a relative efficiency equal to 1 (or 100 %) and the
residual DMU will be considered as inefficient if it has
reached a score of \1 (or less than 100 %). Mathemati-
cally, the linear programming solves the maximisation
problem in the following manner:
Max hi yi; xi; ui; við Þ ¼PN
n¼1 uniyniPK
k¼1 vkixki
ð1Þ
subject to:
XN
n¼1
uniyni ¼ 1 ð2Þ
uni � 0
vki � 0 n ¼ 1; 2; . . .;N outputs and k ¼ 1; 2; . . .;K inputs
ð3Þ
where yni is the quantity of output n produced by the DMU
i; uni is the weight of output n for the DMU i; xki is the
quantity of input k employed by the DMU i; vki is the
weight of input k for the DMU i.2 Furthermore, we employ
an input-oriented model as described in Eqs. (1–3) since it
is reasonable that in the short run tourist industry have
more control over its inputs than its outputs (see Wang
et al. 2006).
In this study, the Banker, Charnes and Cooper (BCC)
model is adopted, since a preliminary investigation depic-
ted VRS for the most of the regions. The WDEA analysis
has been further implemented by calculating the ratio
between CRS and VRS technical efficiency scores, which
gives scale efficiency scores that can be either CRS,
decreasing returns to scale (DRS) or increasing returns to
scale (IRS) (see Charnes et al. 1978; Banker et al. 1984;
Cullinane et al. 2004). Economic theory states that in the
long run the production level can change when inputs,
which are no longer fixed, vary in the same proportion. The
production function depicts IRS (DRS) when inputs are
increased by a factor a and the output increases by more
(less) than a, and CRS when inputs and output increase by
the same factor a.
3.2 Double bootstrap and pooled-truncated
specification
Though DEA has many advantages, it is not possible to
directly evaluate factors that influence DMUs’ efficiency.
2 The solution of Eq. (1) is given by either a maximisation or a
minimisation approach when either one input or one output is used.
However, in the presence of a multivariate input–output framework,
the problem can be solved with either an output-oriented method
(O-OM), by maximising the numerator while keeping the denomina-
tor constant, or an input-oriented (I-OM) method, by minimising the
denominator while keeping the numerator constant. Within the
O-OM, no DMU in the sample, with the same type of inputs, is able to
derive a higher quantity of output. In general, this setting is employed
for planning and strategic objectives. For example, it is used when a
DMU needs to understand whether an expansion of its capacity is
feasible, as long as the existing infrastructure has already been used at
its maximum capacity given the level of the inputs (Cullinane et al.
2004).
J Prod Anal
123
As an extension, a parametric specification can be used
such as a truncated regression since many of the DEA
efficiency scores typically equal to one. However, since
DEA efficiency scores are the result of a linear program-
ming, they are characterised by high correlation that leads
to bias parameter estimates. Recently, Simar and Wilson
(2007) proposed a double bootstrap method that overcomes
possible problems of serial correlation amongst the esti-
mated efficiency scores and approximates their asymptotic
distribution. They argue that conventional bootstrap tech-
niques used in the post-WDEA procedure do not allow for
valid inferences. Only very recently a few studies
employing the double-bootstrap procedure have appeared
in the hospitality literature (e.g. Barros and Dieke 2008;
Assaf et al. 2011; Assaf and Cvelvar 2010; Assaf and
Agbola 2011; Barros et al. 2011).
Following this methodological strand of the literature, in
the present paper, the following pooled-truncated specifi-
cation is used:
hit ¼ Zitbþ eit � 1 i ¼ 1; . . .; n t ¼ 1. . .T ð4Þ
where hit is the ith DMU’s efficiency score at time t (DMUs
are technically efficient or inefficient when hit = 1 or
hit [ 1, respectively3); Zit contains factors that are assumed
to affect the DMUs’ efficiency; b is the vector of param-
eters to be estimated; eit is the residual that is assumed to be
white noise. The b estimates and their confidence intervals
are calculated following a procedure that refers to the
Algorithm #2 proposed by Simar and Wilson (2007). First,
the DEA input-oriented is run for the DMUs under inves-
tigation using xit and yit as inputs and outputs, respectively.
Second, Eq. (4) is estimated by using the maximum like-
lihood method to obtain estimates of the parameters ðbbÞand the deviation standard of the residuals ðbreÞ. Third, for
each DMU the following loop is repeated L1 times (in this
case 10,000 times): (a) for each DMU, eitb is drawn from
the Nð0; breÞ distribution with left truncation at ð1� ZitbbÞ,
with b = 1,…,L1; (b) again, for each DMU, h�itb ¼ Zitbb þ
eitb is computed; (c) a new pseudo data set is defined where
x*it = xit and y�it ¼ yitðbhit þ bh�itbÞ; (d) using the constructed
pseudo data set ðy�it; x�itÞ, the input-oriented DEA is re-run
to compute efficiency estimates for all the DMUs ðbh�itbÞ.Fourth, the bias-corrected efficiency scores are computed
as follows:bbh it ¼ 2bhit �
PL1
b¼1bh�itb. Fifth, the maximum
likelihood method is used to estimate the truncated
regression of the bias-corrected efficiency scores, which
provides with marginal effects of the explanatory variables
ðbbÞ and estimated standard deviation of the residuals ðbreÞ.Sixth, again for each DMU the following bootstrapping loop
is repeated L2 times (again, 10,000 times): (1) for each DMU,
eits is drawn from the Nð0; bbreÞ distribution with left trunca-
tion at ð1� Zits^bÞ;with s = 1,…, L2; (2) for each DMU,
h��its ¼ Zit^bþ eits is computed; (3) the maximum likelihood
method is employed to estimate the truncated regression of
h��its on Zit that provides with a new set of estimated marginal
effects of the explanatory variables ðbbbb sÞ and standard
deviation of residuals (bbbr e;s). Hence, the L2 bootstrap esti-
matesbbbb s and
bbbr e;s are used to construct estimated confidence
intervals for each of the unknown element in (4) (see also
Balcombe et al. 2008; Assaf et al. 2011).
3.3 Spatial analysis
An important limitation when dealing with regional data is
the existence of spatial correlation between observations,
which produces unbiased but inefficient Ordinary Least
Squares (OLS) coefficients (Anselin 1988) due to the non
diagonal structure of the disturbance term. Spatial auto-
correlation means that similar values tend to be close to
each other. Unfortunately, model (4) does not take into
account spatial spill-over effects among Italian regions. In
this paper, we employ the Moran’s I test to the DEA
efficiency scores in order to check for the presence of
spatial autocorrelation among Italian regions. This way we
can identify the existence of clusters of regions character-
ized by similar level of technical efficiency.
The Moran’s I test is calculated as follows (Anselin
1988):
I ¼ n
S0
~h0W ~h~h0~h
ð5Þ
where n is the number of units, the vector ~hindicates the
deviation from the mean of the variable of DEA scores has
in Eqs. (1–3), W represents the spatial weight matrix and S0
is the sum of all components of the matrix W. Under the
null hypothesis, the expected value of Moran’s I test is:
EðIÞ ¼ �ðn� 1Þ�1 ð6Þ
In this study, the inverse of the distance between the cen-
troids is taken as a spatial weight, which signifies that the
higher the distance between two regions the lower is the
magnitude of the spatial effects between them.
By construction, the Moran I test ranges from -1 to ?1.
Specifically, positive values of the spatial autocorrelation
3 Standard DEA scores range between zero and one, as mentioned in
Sect. 3.1. Unfortunately, double bounds of the dependent variable
impose a more complex procedure in solving Eq. (4). In order to
simplify the econometric model, we follow the suggestion of Simar
and Wilson (2007), who propose to use the inverse of standard DEA
scores. This way, the new dependent variable has a single lower
bound equals to one.
J Prod Anal
123
indicate spatial similarity, while negative values are asso-
ciated with spatial dissimilarity (Griffith 2003; pp. 5). A
zero value indicates that the pattern is randomly spatially
distributed.
4 Italian regions economic efficiency
4.1 Hospitality productivity
As previously stated, the definition of productivity is
complex and includes several dimensions of performance
ranging from efficiency to effectiveness, from tangible
measures, such as outputs, rates of turnover and absen-
teeism, to intangible measures such as customer satisfac-
tion, loyalty and job satisfaction.
From an economic point of view, productivity is defined
as the amount of output of goods and services for each unit
of input used in a given time period. The concept of pro-
ductivity is closely linked to the notion of efficiency. If a
DMU is efficient it is said to be operating on the production
frontier. Hence, a rise in efficiency implies an increase in
productivity, while an outward shift in the production
frontier implies productivity growth. As Farrell (1957)
stated, the economic efficiency relates to how far a given
DMU can increase its output without using further
resources.
Mandl et al. (2008) emphasise that effectiveness relating
to the final outcome to be achieved is another important
concept. As far as the hospitality sector is concerned,
effectiveness can be influenced by several endogenous
determinants such as firms’ characteristics, pricing policy,
as well as exogenous factors such as the stage of the life
cycle of product in the destination and socio-economic
characteristics of the segments of demand.
To measure efficiency and effectiveness one needs to
employ observed data. To exemplify this semantic issue
and build a more systematic framework within an empirical
setting, two distinctive lines of research can be identified.
On the one hand, if one is analysing the productivity of a
DMU at a microeconomic level, it seems reasonable to use
a wider range of objective measures of efficiency of per-
formance (e.g. cost per unit; output per employee) and
outcomes of effectiveness (e.g. sales), possibly including
those intangible outcomes, such as customer experience
and satisfaction, that can be only quantified with ad hoc
survey data (Sigala 2004). In this respect, the implemen-
tation of categorical variables makes it possible to pick up
qualitative information, such as the quality of the services
offered and the quality of the working environment, which
can give a contribution in assessing DMUs performance.
On the other hand, if one is analysing the productivity at a
macroeconomic level, it seems reasonable to use standard
objective measures of efficiency and effectiveness such as
overall costs and overall revenues. These are generally
census data that tend to allocate firms to a specific industry
without providing much detailed information.
From a practitioner point of view, productivity, and
hence efficiency and effectiveness, is a strategic tool to
increase national wealth, particularly in mature economies
where insufficient productivity cannot be hidden by an
expansion in market size (Goh 2010). This is particularly
true for a labour intensive productive process, such as the
one which takes place in the hospitality sector. From
research related to the manufacturing sector it emerges
that, while less productive firms are not able to face
competition, an intermediate level of productivity allows
firms to only compete in internal markets and firms char-
acterised by high productivity are commonly more com-
petitive in external markets (e.g. Bernard et al. 2007).
Hospitality, and more in general tourism activity, is
regarded as a non-standard export sector, since consump-
tion happens in situ, and it is the consumer who moves
instead of the product (Cortes-Jimenez and Pulina 2010).
Hospitality can serve local, national and foreign consum-
ers. Hence, one can argue that the more productive a DMU,
the more it is able to equally compete within these three
segments of demand.
Increasingly, the hotel sector tends to operate in a
globalised and highly competitive marketplace, where
making profits becomes more and more challenging, given
the very volatile demand. Thus, monitoring hotel produc-
tivity is a key strategy to identify whether the sector is
profitable and is growing over time, and moreover to set ad
hoc marketing and pricing policies.
4.2 Inputs-outputs and data description
The DEA approach allows one to deal with multi-output
multi-input processes deriving a virtual input and output,
obtained as linear combination of the observed inputs and
outputs. The choice of inputs and outputs is a critical task
because a correct identification allows one to represent
properly the production function. Assaf et al. (2010)
present a comprehensive literature review that sheds light
on the indicators commonly used in the most relevant
studies on tourism hospitality to assess the DMUs perfor-
mance. The majority of these studies employ a microeco-
nomic framework, where efficiency is assessed at a firm
level. In general, inputs and outputs are obtained from
accounting books, which tend to gather data in a more
aggregated manner. Only a very few studies take into
account how intangible outputs, and specifically customer
satisfaction, contribute to the overall firms productivity.
Following previous studies, and based on the data
availability at a macroeconomic level, in the present study,
J Prod Anal
123
sales revenue (SR) and value added (VA) are employed as
outputs. Specifically, sales revenue is defined as the prod-
uct between the price at which goods and services are sold
and the number of units, or amount sold. This indicator has
been used in various studies and more recently in Reynolds
and Thompson (2007) to assess efficiency in brand res-
taurants. In the data provided by the Italian National
Institute of Statistics (ISTAT), value added is defined as the
market value of firms’ product, or service, minus the cost
of inputs purchased from other firms. It uses retail price
rather than market price (i.e. the price at which the man-
ufacturer recommends that the retailer sells the product).
Therefore, the value added in such a specification encom-
passes price-discrimination issues that are going to incor-
porate non-financial factors such as location and quality.
Nonetheless, these measures are recognised to be adequate
indicators of firms efficiency and have been used in other
empirical investigations (e.g. Barros and Santos 2006;
Wang et al. 2006; Min et al. 2008). Referring to the
aggregated business of the tourism sector, SR indicates the
market share in monetary terms of the DMUs. VA repre-
sents the ability to make profit of each region in the sector
under study.
Within the classical framework, two inputs are also
employed: labour costs (LC) and gross fix investments
(K)—used here as physical capital production factor.
Specifically, LC are defined as the total expenditure borne
by employers in order to employ workers; this indicator
includes direct remuneration, bonuses, payments for days
not worked, severance pay, benefits in kind. They also
include indirect costs linked to employees, such as con-
tractual and voluntary social security contributions, direct
social benefits, vocational training costs, other social
expenditure (e.g. medical services), and taxes relating to
employment regarded as labour costs, minus any subsidies
received. K are defined as the acquisition of fix capital,
which also comprises the value of capital goods produced
by the firm. In this context, each DMU combines labour
and capital to produce the outputs, namely sales revenue
and value added. The DEA approach allows us to detect
which DMUs use the inputs in the most efficient way.
The data used in this analysis are collected from ISTAT,
which assembles the variables at regional level. Table 1
shows descriptive statistics of the variables (in thousands
euros) under study in the DEA procedure. The analysis
concerns 19 Regions and the two autonomous Provinces of
Trento and Bozen for the time span 2000–2004.
5 Window-DEA results
At a macro level, a comparison of efficiency is provided
amongst all of the 20 Italian regions. One region, Trentino
Alto Adige, is reported as two provinces by the Italian
National Statistics Office (ISTAT) and this analysis follows
the national classification; thus, 19 regions and two
autonomous provinces (Trento and Bozen) are considered
resulting in 21 DMUs (see Fig. 1). Given the availability of
official statistic data, updated in December 2010, a time
span of 5 years (2000–2004) is considered. Over-fit prob-
lems are avoided as the minimum number of DMUs is
more than twice the total number of inputs and outputs in
the DEA (Min et al. 2008).
To run the analysis, the software package Frontier
Analyst 3.1.5 is used. The selection of the length of the
window is an important issue in window DEA because the
results may depend on the number of windows employed.
In the present analysis, the following formulas adapted
from Sun (1988) are applied (see also Cooper et al. 2000).
Given n DMUS and k periods, the length of window is
given by the following:
p ¼kþ1
2when k is odd
kþ12� 1
2when k is even
�
ð7Þ
In this case, as twenty-one regions and 5 years’ worth of
data are used, for a total of 105 observations, the chosen
window length is 3 years and three separate windows are
constructed as shown in Table 2; hence, the number of
Table 1 Descriptive statistics
Variables Code Model Obs. Mean St.Dev. Min Max
Capital costs K DEA (INPUT) 105 185,048.30 198,983.38 5,796 1,277,027
Labour costs LC DEA (INPUT) 105 482,125.35 482,844.53 14,581 2,157,532
Sales revenue SR DEA (OUTPUT) 105 2,380,467.55 2,367,530.27 100,890 10,770,620
Value added VA DEA (OUTPUT) 105 908,060.98 845,105.92 25,982 3,710,862
Bed places utilization NRU POST-DEA 105 37.93 8.53 21.8 58.8
Coeff. Variation of bed places utilization CV-NRU POST-DEA 105 14.91 4.91 3.65 23.32
Art cities DART POST-DEA 105 0.14 0.35 0 1
High quality hotels HQ-HOTEL POST-DEA 105 10.99 4.74 4.33 22.37
Calculation on data collected from ISTAT
J Prod Anal
123
observations reduces to 63 (21 regions over three separate
windows).
The first row (with values of 97.1, 98.0 and 88.3 %)
shows the relative technical efficiency of the Abruzzi
region in 2000, 2001 and 2002, respectively. The second
row (with values of 95.3, 88.3 and 93.2 %) shows the
relative technical efficiency of Abruzzi in 2001, 2002 and
2003, respectively, and so on. The same results, read by
column, represent the stability of efficiency scores both in
absolute terms as well as in terms of the relative perfor-
mance of that region with respect to the other regions in the
sample. It is worthwhile noting that two regions present a
few ‘‘extreme mismatches’’ that constitute a greater than
10 % points annual change, which might be due to mis-
reporting of data or computational errors. Examples of
misreporting are also found in Charnes et al. (1985) and
Cullinane et al. (2004). The last columns show the mean,
standard deviation (SD) and coefficient of variation (CV)
for each region. The latter is calculated as the ratio between
the standard deviation and the mean.
Overall, from Table 2 an increasing efficiency score
emerges for Emilia-Romagna, Lazio, Sardinia, Umbria and
Veneto. Reading the results by column, the best perfor-
mance was achieved consistently by Lombardy; however,
in 2003, the most efficient regions included Liguria, Molise
and Piedmont.
Following Pulina et al. (2010), a further step is to test
the relationship that exists between the mean efficiency and
its volatility measured in terms of the CV. A negative
correlation between the mean and the CV is expected as
high efficiency mean is most likely to be associated with a
low volatility. One reason for this negative correlation is
that the CV tends to zero as the mean approaches 100. High
CVs occur in regions such as Aosta Valley and Sicily,
which also have relatively low means. Spearman’s test is
then carried out in order to test whether the DMUs are
characterised by homoscedasticity (as a null hypothesis
q = 0) or heteroscedasticity (as an alternative hypothesis
q = 0). The calculated q for the Italian regions (Table 2)
equals -0.67. This value, in absolute terms, is higher than
the corresponding critical value (0.43), at the two-tailed
5 % level of significance. Hence, the null hypothesis can be
rejected and the Italian regions are characterised by a non-
constant variance at the cross-sectional level.
A synthesis of the main results achieved from the DEA
analysis is provided in Figs. 1 and 2. The former presents a
static analysis of the regional economic efficiency by
comparing the last year to the trend performance across the
time span under investigation. Clockwise, the top right
quadrant depicts the ‘‘moving ahead’’ regions, which
denote a score efficiency higher than both the trend average
score and the average score for 2004. Results show that
eight regions (namely Lombardy, Veneto, Umbria, Lazio,
Piedmont, Emilia Romagna, Tuscany and Liguria) belong
to this group, although only Lombardy is the peer that
shows the highest level of pure technical efficiency in each
year (Table 1). The bottom right quadrant includes the
‘‘catching up’’ regions, which reach a score efficiency
lower than the trend and higher than the average score for
2004. Sardinia and Marche are the sole regions that belong
to this group. The bottom left quadrant depicts the ‘‘falling
further behind’’ regions, which experience an average score
lower than both the trend and the average score in 2004.
Seven regions belong to this group namely: Aosta Valley,
0 0 0 050 55 60 65 70 75 80 85 90 95 100
70
75
80
85
90
95
100Moving aheadLosing momentum
Catching upFalling further behind
Tre
nd 2
000
-200
4
Score 2004
LO
VE
UMLA
ER
SA
LITO
MOPI
BZ
MA
AB
FVGCA
CAL
BA
TR
AP
AV
SI
Fig. 1 Italian regions: static
efficiency performance 2004
and trend (2000–2004). Notes:
‘‘Moving ahead’’: Molise (MO),
Lombardy (LO), Liguria (LI),
Veneto (VE), Lazio (LA),
Piedmont (PI), Umbria (UM),
Marche (MA), Emilia Romagna
(ER) and Toscany (TO);
‘‘Catching up’’ Sardinia (SA);
‘‘Falling further behind’’ Apulia
(AP), Sicily (SI), Trento (TR),
Calabria (CA), Aosta Valley
(AV), Bozen (BZ) and Campania
(CA); ‘‘Losing Momentum’’
Friuli Venezia Giulia (FVG),
Basilicata (BA) and Abruzzo
(AB)
J Prod Anal
123
Table 2 Regional efficiency: a window DEA approach (VRS)
Regions Years Statistics
2000 2001 2002 2003 2004 Mean SD CV
Abruzzo 97.1 98.0 88.3
95.3 88.3 93.2
97.2a 96.2 62.6 90.7 10.5 0.116
Aosta Valley 84.7 81.5 63.5
71.3a 63.5 52.1
62.0 54.7 69.0 66.9 10.4 0.155
Apulia 80.7 59.3 77.8
59.3 77.8 67.5
90.3a 70.7 68.2 72.4 9.6 0.133
Basilicata 86.3 – 78.6
– 78.6 80.3
78.6 86.3 60.4 78.5 8.0 0.102
Bozen 77.4 85.7 90.7
85.7 90.7 69.9
84.8 70.0 75.1 81.1 7.7 0.095
Calabria 74.6 79.6 96.7
79.6 96.7 68.1
99.8 72.4 58.6 80.6 13.5 0.167
Campania 89.6 94.6 100
94.6 100 75.9
100 72.3 59.6 87.4 13.8 0.158
Emilia Romagna 84.1 96.4 92.8
100 92.1 78.5
90.5 79.0 100 90.4 7.8 0.086
Friuli Venezia Giulia 89.5 82.6 93.7
82.5 93.7 90.3
94.1 100 65.2 88.0 9.6 0.110
Lazio 98.6 96.2 95.8
96.2 95.4 76.1
94.5 76.1 100 92.1 8.7 0.095
Liguria 100 91.8 69.0
89.4 68.9 100
65.6 100 85.3 85.5 13.5 0.157
Lombardy 100 100 100
100 100 100
100 100 100 100 0.0 0.000
Marche 90.1 79.9 69.9
73.8 67.3 88.4
68.6 87.9 82.0 78.6 8.5 0.108
Molise 87.7 100 100
100 100 100
100 100 76.7 96.0 7.8 0.082
Piedmont 87.1 100 100
100 100 100
100 100 73.5 95.6 8.8 0.092
Sardinia 66.5 74.5 70.7
74.5 70.7 62.6
J Prod Anal
123
Sicily, Apulia, Trento, Basilicata, Calabria, and Bozen.
Finally, the top left quadrant includes the ‘‘losing
momentum’’ regions, which denote a score higher than the
trend and lower that the average score in 2004. The regions
of Campania, Friuli Venezia Giulia, Abruzzi, Piedmont and
Molise fall into this category.
Figure 2 provides a dynamic picture of the performance
of each Italian region. The majority of the regions fall in
either the ‘‘moving ahead’’ or the ‘‘falling further behind’’
quadrant. There is a stable peer group that includes
Lombardy (as the peer) and Veneto, and a stable ‘‘falling
further behind’’ group that includes Aosta Valley, Apulia,
Basilicata, Trento and Sicily. All the other regions can be
considered in a transition phase. For example, Bozen and
Calabria, which were catching-up regions within the first
window, have become ‘‘falling further behind’’ DMUs
within the last window. Umbria, which in the first window
was in the ‘‘losing momentum’’ quadrant, has improved its
performance and in the second and third window has fallen
into the ‘‘moving ahead’’ quadrant. Marche and Sardinia,
which were ‘‘falling further behind’’ regions, have become
‘‘catching up’’ DMUs within the last window. Lazio and
Emilia Romagna regained a highest performance after
having lost momentum within the second window of
analysis. Abruzzi, Piedmont and Molise, which were
‘‘moving ahead’’ regions, have fallen into the ‘‘losing
momentum’’ quadrant within the last window of
investigation.
The window DEA analysis has been implemented by
calculating the scale efficiency scores, which can be either
CRS, IRS or DRS. Table 3 depicts the share of regions
denoting IRS and DRS, respectively. It emerges that the
share of regions characterised by IRS has increased over
time, while the share of regions that present DRS has
remarkably decreased. Hence, in the long run these firms
typically are small sized.
In order to compare regions characterised by IRS and
non-IRS, the average net rate of utilisation of bed places is
calculated for both groups (see Table 3). The rationale is
that higher ‘‘excess capacity’’, which corresponds to low
utilisation rate, should be associated with non-increasing
returns to scale, because it might indicate that hotels are
oversized. Although regions with increasing returns to
scale have higher utilisation rates, except for the last year
included in the sample, the differences in the average level
of utilisation rate between the two groups are not statisti-
cally significantly different from zero. As a result, there
does not seem to be any empirical evidence of the
Table 2 continued
Regions Years Statistics
2000 2001 2002 2003 2004 Mean SD CV
75.6 67.3 100 73.6 10.2 0.138
Sicily 87.2 77.8 69.8
79.8 69.7 63.4
65.9 63.0 49.6 69.6 10.4 0.150
Toscany 99.2 84.8 82.2
85.5 81.4 82.1
79.4 84.7 81.8 84.6 5.5 0.065
Trento 80.4 68.8 84.6
68.8 84.6 64.0
82.5 66.0 65.9 74.0 8.3 0.112
Umbria 77.7 100 84.7
100 84.7 98.8
89.1 98.8 100 92.6 8.2 0.088
Veneto 93.9 92.3 91.0
95.5 95.6 97.5
98.5 94.9 100 95.4 2.7 0.029
Italy 87.3 87.2 85.7
86.6 85.7 81.4 84.5 3.1 0.037
86.5 82.9 77.8
SD standard deviation, CV coefficient of variationa In 2001 the Basilicata region has not been included in the analysis since official data are missing, which does not affect the DEA overall
analysis. ‘‘Extreme mismatch’’ (over 10 % points change)
J Prod Anal
123
C
AV
SAAP
SIMA
BA
TR
LI
CALBZWUMTO FVG
ERVE
AB PICA
LO
LA MO
60.0
65.0
70.0
75.0
80.0
85.0
90.0
95.0
100.0
Score 2002
Tre
nd
200
0-20
02
A
B
BZCAL
TRSI
SA AP
AV
MABA
TOW
LACA ER FVG
LI
ABUM
VE
PI ; LO; MO
50.0
55.0
60.0
65.0
70.0
75.0
80.0
85.0
90.0
95.0
100.0
Score 2003
Tre
nd
200
1-20
03
CA
SIAV
TR
BACAL AP
ABFVG
PI MO
BZ
W
MATO
LI
SA
LAER
UMVELO
55.0
60.0
65.0
70.0
75.0
80.0
85.0
90.0
95.0
100.0
60.0 65.0 70.0 75.0 80.0 85.0 90.0 95.0 100.0
50.0 55.0 60.0 65.0 70.0 75.0 80.0 85.0 90.0 95.0 100.0
40.0 45.0 50.0 55.0 60.0 65.0 70.0 75.0 80.0 85.0 90.0 95.0 100.0
Score 2004
Tre
nd
200
2-20
04
Fig. 2 Italian regions: dynamic
efficiency performance (last
window year and window
trend). Notes: ‘‘Moving ahead’’:
top right quadrant; ‘‘Catching
up’’: bottom right quadrant;
‘‘Falling further behind’’:
bottom left quadrant; ‘‘Losing
Momentum’’: top left quadrant;
Italian average (W)
Table 3 Percentage of regions
showing IRS and DRS—
Window DEA (VRS)—
(2000–2004)
2000 2001 2002 2003 2004
Italy (IRS; DRS) 47.6; 42;9 57.1; 19.0 76.2; 4.8
66.7; 4.8 81.0; 0.0 52.4; 28.6
81.0; 0.0 66.7; 9.5 66.7; 4.8
Average net rate of utilization
IRS regions 40.04 41.19 37.35 38.63 35.76
No-IRS regions 37.84 38.00 35.83 34.71 39.63
J Prod Anal
123
existence of a relationship between capacity and returns to
scale in Italian tourist industry.
Although the DEA approach does not allow one to gain
a full understanding of the factors of inefficiency, varia-
tions in outputs and inputs can be calculated in order to
achieve the necessary improvement to obtain the score
within the benchmark efficiency frontier. Table 4 provides
useful information on possible ways to improve efficiency.
On the one hand, in a great part of the Italian regions,
potential improvements can be achieved via a consistent
decrease in terms of capital costs (K) and labour costs (LC).
On the other hand, important improvements have been
achieved in terms of sales revenue (SR) and gross value
added (GVA) in the last year under investigation. This is
especially true for the Bozen and Trento provinces.
Although the window-DEA is a powerful procedure,
particularly for entrepreneurs who rely on input-cost min-
imisation as a measure of organisational efficiency, it holds
important limitations since it does not allow for estimating
a random error in the estimation of efficiency (Reynolds
2003; Assaf and Agbola 2011). However, the statistical
properties of the window-DEA can be retained via a
bootstrap approach that can be further extended to a pooled
parametric estimation through which statistical hypotheses
can be directly tested.
6 Post-DEA results: a pooled-truncated regression
Following the econometric approach showed in Sect. 3.2, a
double bootstrapping method is followed and a pooled-
truncated regression is applied. To this aim the software
library FEAR 1.15, for the statistical package R, by Wilson
(2008) is used. The relevant regression equation is given by
the following:
hit ¼ aw þ axNRUit þ a2CV NRUit þ a3DARTit
þ a4HQ HOTELSit þ eit
i ¼ 1; . . .; n t ¼ 1; . . .; Tð8Þ
where hit is the DMU’s (i) (in)efficiency score at time t; a is
the constant term; a1, a2, a3are the parameters to be esti-
mated. NRU is the net rate of utilisation of bed places,
which proxies the capacity of fully utilised establishments
during the opening period. Usually highest levels coincide
with regions that are characterised by higher performances
in the tourism activity, and a negative sign is expected. In
other words, high levels of NRU are associated with (low)
high levels of (in)efficiency h. The net occupancy rate of
bed places is obtained by dividing total overnight stays in a
given period (in this case, during a year) by the product of
the bed places and the number of opening days. CV_NRU is
the annual coefficient of variation of the net rate of util-
isation to pick up the market volatility. It is calculated for
each region as the ratio of the standard deviation of the
monthly net rate of utilisation and the annual average of
NRU. High values of CV_NRU indicate the presence of
seasonality in the hotel demand. Hence, the relationship
between h and CV_NUR is expected to be positive. Usu-
ally, art cities reach high level of tourism industry spe-
cialisation; DART is a dummy variable that explicitly takes
into account the effects that well-known arts city (namely
Rome, Florence and Venice) have on regional hospitality
sector efficiency. Hence, a negative sign is expected.
HQ_HOTELS is defined as the share of high quality hotels
per region. It is calculated as the ratio between the number
of four and five stars hotels and the total number of hotels,
and is a proxy of the quality of the accommodation supply
in a given region. The expected sign is negative since high
quality hotels are more profitable than low quality hotels.
Finally, eit is the residual that is assumed to be white
noise. Some descriptive statistics are shown in Table 1.
The arts city regions (namely Lazio, Tuscany and Veneto)
account for 14 % of the sample. NRU, CV_NRU and
HQ_HOTELS have been multiplied by 100 in order to
express them in percentage terms.
Main results are reported in Table 5. Models (1) and (3)
contain the results of the pooled-truncated approach. In
Model (1), all the variables seem to explain Italian hospi-
tality sector’s economic efficiency. Specifically, an
increase in the net rate of utilisation causes an increase in
the pure technical efficiency (a1 ¼ �0:05), indicating that
the more the physical capital is used, the greater the effi-
ciency of the region.
Furthermore, an increase in the volatility of the net rate
of utilisation (CVNRU) leads to an increase in regional
inefficiency (a2 ¼ 0:05). Our findings are a further empir-
ical evidence of the negative impact of the seasonality on
tourism industry. Tourism seasonality imposes higher costs
to firms because they have to oversize their capability in
order to cope with the peak of the demand, thus reducing
their efficiency.
On average, regions specialised in cultural tourism, such
as Lazio, Veneto and Tuscany, tend to be more efficient
than the other Italian regions. Usually, cultural tourism is
concentrated in a few famous cities, namely Rome, Venice
and Florence, which allow one to realize scale agglomer-
ation effects. On the contrary, maritime and mountain
touristic regions are characterised by a wild spatial dis-
persion of attractiveness and destinations, which leads to
higher expenditures of transports and public services.
In Model (3), HQ_HOTELS is included into the
regression. Notably, all the estimates are congruent, with
the only exception of NRU which, although it still presents
a negative sign, is not statistically significant. HQ_HO-
TELS coefficient is negative (-0.07) and significant, indi-
cating that an increase in the share of high quality hotels
J Prod Anal
123
Ta
ble
4It
alia
nre
gio
ns:
tota
lp
ote
nti
alim
pro
vem
ents
(%)
inte
rms
of
sale
sre
ven
ue
(SR
),g
ross
val
ue
add
ed(G
VA
),la
bo
ur
cost
s(L
C)
and
fix
edin
ves
tmen
ts(K
);w
ind
ow
DE
A(V
RS
)—
(20
00
–2
00
4)
Reg
ions
Var
iab
les-
yea
rs
SR
00
GV
A0
0L
C0
0K
00
SR
01
GV
A0
1L
C0
1K
01
SR
02
GV
A0
2L
C0
2K
02
SR
03
GV
A0
3L
C0
3K
03
SR
04
GV
A0
4L
C0
4K
04
Ab
ruzz
o0
.02
6.6
-2
.9-
2.9
0.0
1.4
-2
.0-
2.0
0.0
0.0
-1
1.7
-5
3.9
0.0
0.0
-4
.7-
4.7
0.0
0.0
-1
1.7
-5
3.9
0.0
4.3
-6
.8-
6.8
0.0
0.0
-2
.8-
64
.30
.01
2.4
-3
.8-
3.8
0.0
0.0
-3
7.4
-4
0.5
Ao
sta
Val
ley
13
.50
.0-
15
.3-
15
.34
.80
.0-
32
.7-
18
.57
.10
.0-
36
.5-
38
.7
6.8
0.0
-2
8.7
-2
8.7
7.1
0.0
-3
6.5
-3
8.7
0.0
0.0
-4
7.9
-4
7.9
0.0
0.0
-3
8.0
-3
8.0
0.0
4.3
-4
5.3
-4
5.3
0.0
0.0
-3
1.1
-4
4.0
Ap
uli
a0
.00
.9-
19
.3-
19
.30
.00
.0-
40
.7-
58
.50
.00
.0-
22
.2-
70
.8
0.0
0.0
-4
0.7
-5
8.5
0.0
0.0
-2
2.2
-7
0.8
0.0
0.0
-3
2.5
-3
2.5
0.0
0.0
-9
.7-
66
.40
.00
.0-
29
.3-
40
.80
.00
.0-
31
.9-
57
.6
Bas
ilic
ata
0.2
23
.2-
13
.7-
13
.7-
--
–0
.80
.0-
21
.4-
65
.8
––
––
0.8
0.0
-2
1.4
-6
5.8
0.0
4.2
-1
9.7
-7
7.2
0.0
0.0
-2
1.4
-6
3.4
0.0
0.0
-1
3.7
-6
8.3
0.0
0.0
-3
9.6
-8
3.5
Bo
zen
42
.50
.0-
22
.6-
22
.66
4.0
0.0
-1
4.3
-2
3.1
54
.40
.0-
9.3
-2
1.3
64
.00
.0-
14
.3-
23
.15
4.4
0.0
-9
.3-
21
.32
6.2
0.0
-3
0.1
-9
0.9
0.0
0.0
-1
5.3
-6
2.7
0.0
0.0
-3
0.0
-9
4.6
0.0
0.0
-2
5.0
-5
2.6
Cal
abri
a0
.00
.0-
25
.4-
63
.60
.00
.0-
20
.4-
86
.37
.60
.0-
3.3
-6
1.2
0.0
0.0
-2
0.4
-8
6.3
7.6
0.0
-3
.3-
61
.20
.00
.0-
31
.9-
68
.6
0.0
0.0
-0
.3-
68
.70
.00
.0-
27
.6-
72
.50
.00
.0-
41
.4-
46
.6
Cam
pan
ia0.0
1.3
-1
0.4
-1
0.4
12
.10
.0-
5.4
-5
.40
.00
.00
.00
.0
12
.10
.0-
5.4
-5
.40
.00
.00
.00
.01
1.0
0.0
-2
4.2
-2
4.2
0.0
0.0
0.0
0.0
0.0
0.0
-2
7.7
-3
0.9
0.0
0.0
-4
0.4
-4
0.4
Em
ilia
Rom
agn
a2
.40
.0-
15
.9-
28
.43
.80
.0-
3.6
-3
.67
.80
.0-
7.2
-7
.2
0.0
0.0
0.0
0.0
11
.20
.0-
7.9
-7
.90
.00
.0-
21
.6-
21
.6
0.0
0.0
-9
.6-
9.6
0.0
0.0
-2
1.0
-2
4.9
0.0
0.0
0.0
0.0
Fri
uli
Ven
ezia
Giu
lia
0.0
10
.4-
10
.5-
10
.50
.00
.0-
17
.4-
17
.42
6.1
0.0
-6
.3-
57
.3
0.0
0.0
-1
7.5
-1
7.5
26
.10
.0-
6.3
-5
7.3
0.0
0.0
-2
1.6
-2
1.6
0.0
0.0
-5
.9-
74
.50
.00
.00
.00
.00
.00
.0-
34
.8-
34
.8
Laz
io0
.01
3.5
-1
.4-
42
.39
.30
.0-
3.8
-3
.81
.30
.0-
4.2
-4
.2
9.3
0.0
-3
.8-
3.8
3.2
0.0
-4
.6-
4.6
0.0
19
.6-
9.7
-5
3.2
0.0
0.0
-5
.5-
5.5
0.0
1.2
-2
3.9
-2
3.9
0.0
0.0
0.0
0.0
Lig
uri
a0
.00
.00
.00
.00
.02
.1-
8.2
-8
.27
.20
.0-
31
.0-
31
.0
0.0
0.0
-1
0.6
-1
0.6
7.6
0.0
-3
1.1
-3
1.1
0.0
0.0
0.0
0.0
0.0
0.0
-3
4.4
-3
4.4
0.0
0.0
0.0
0.0
0.0
0.0
-1
4.7
-4
0.7
Lo
mb
ard
y0
.00
.00
.00
.00
.00
.00
.00
.00
.00
.00
.00
.0
J Prod Anal
123
Ta
ble
4co
nti
nu
ed
Reg
ions
Var
iab
les-
yea
rs
SR
00
GV
A0
0L
C0
0K
00
SR
01
GV
A0
1L
C0
1K
01
SR
02
GV
A0
2L
C0
2K
02
SR
03
GV
A0
3L
C0
3K
03
SR
04
GV
A0
4L
C0
4K
04
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Mar
che
0.0
0.0
-9
.9-
9.9
0.0
11
.0-
20
.1-
20
.10
.03
.0-
30
.1-
30
.1
0.0
0.0
-2
6.2
-2
6.2
0.0
0.0
-3
2.7
-3
2.7
0.0
0.0
0.0
0.0
0.0
0.8
-3
1.4
-3
1.4
0.0
7.9
-1
2.1
-1
2.1
0.0
0.0
-1
8.1
-1
9.5
Mo
lise
1.3
0.0
-1
2.3
-1
2.3
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
10
.9-
11
.7-
11
.7
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2.3
0.0
-2
3.3
-5
1.2
Pie
dm
on
t9
.10
.0-
12
.9-
12
.90
.00
.00
.00
.00
.00
.00
.00
.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1.7
-2
6.5
-2
6.5
Sar
din
ia0
.01
.0-
33
.5-
34
.87
.30
.0-
25
.5-
43
.93
.30
.0-
29
.3-
51
.8
7.3
0.0
-2
5.5
-4
3.9
3.3
0.0
-2
9.3
-5
1.8
0.0
0.0
0.0
0.0
0.0
0.0
-2
4.4
-6
2.3
0.0
0.0
-3
2.7
-4
7.8
0.0
0.0
0.0
0.0
Sic
ily
0.0
5.4
-1
2.8
-1
2.8
16
.50
.0-
22
.2-
22
.27
.50
.0-
30
.2-
30
.2
8.6
0.1
-0
.2-
0.5
7.7
0.0
-3
0.3
-3
0.3
0.0
0.0
-3
7.4
-3
7.4
0.0
0.0
-3
4.1
-3
4.1
0.0
0.0
-3
7.0
-3
7.1
0.0
2.5
-5
0.4
-5
0.4
To
scan
y0
.00
.0-
0.8
-0
.81
1.2
0.0
-1
5.2
-1
5.2
14
.60
.0-
17
.8-
17
.8
12
.10
.0-
14
.5-
14
.51
9.6
0.0
-1
8.6
-1
8.6
1.8
0.0
-3
6.6
-3
6.6
0.0
0.0
-2
0.7
-2
0.7
4.9
0.0
-1
5.3
-3
8.3
0.0
0.0
-1
8.2
-1
8.2
Tre
nto
21
.90
.0-
19
.6-
19
.60
.30
.0-
31
.2-
31
.22
8.1
0.0
-1
5.4
-4
4.6
0.4
0.0
-3
1.2
-3
1.2
28
.10
.0-
15
.4-
44
.62
.00
.0-
17
.9-
17
.9
0.0
0.0
-1
7.5
-6
4.6
0.0
0.0
-3
4.0
-9
1.6
0.0
0.0
-3
4.1
-7
9.3
Um
bri
a0
.00
.0-
22
.3-
56
.50
.00
.00
.00
.00
.00
.0-
15
.3-
66
.6
0.0
0.0
0.0
0.0
0.0
0.0
-1
5.3
-6
6.6
2.5
0.0
-3
6.0
-9
0.9
0.0
0.0
-1
0.9
-6
9.6
0.0
0.9
-1
.2-
1.2
0.0
0.0
0.0
0.0
Ven
eto
9.5
0.0
-6
.1-
6.1
4.8
0.0
-7
.7-
7.7
0.0
0.0
-9
.0-
9.0
4.1
0.0
-4
.5-
4.5
0.0
0.0
-4
.4-
19
.20
.00
.6-
1.2
-1
.2
0.0
0.0
-1
.5-
45
.80
.90
.0-
5.1
-5
.10
.00
.00
.00
.0
Ital
y4
.83
.9-
12
.7-
18
.86
.70
.7-
13
.5-
18
.47
.90
.2-
13
.9-
29
.8
6.2
0.0
-1
2.4
-1
8.0
8.4
0.0
-1
3.9
-3
0.5
2.1
1.9
-1
8.4
-3
0.5
0.0
0.0
-1
3.5
-3
8.9
0.3
1.3
-1
7.1
-3
0.4
0.1
0.2
-2
2.2
-3
2.7
J Prod Anal
123
(represented by four and five-stars hotels) improves the
economic efficiency of the hospitality sector.
Finally, the Ordinary Least Squares (OLS) estimates of
Model (1) and (3) are presented in Table 5, Column (2) and
(4), respectively. None of the coefficients are statistically
significant except for CV_NRU (0.004). These empirical
results highlight that post-DEA analysis is very sensitive to
the approach used (see Simar and Wilson 2011, for an
extensive discussion). By ignoring the property of the multi-
step procedure by Simar and Wilson (2007), one could obtain
biased estimates that lead to wrong conclusions.
7 Spatial econometric analysis
Spill-over effect investigation of aggregated, or individual,
efficiency is a new branch of research in the empirical
analysis. In a post-DEA framework, Helfand and Levine
(2004) and Sampaio de Souza et al. (2005) find spatial
effects due to the existence of some functional relationship
between DMUs’ efficiency in two distinct points in space.
In a similar way, in the present study, the existence of
spatial autocorrelation between regional DEA scores is
tested by applying the Moran’s I test. Spill-over effects can
operate through different channels. For instance, tourism
hot spots can lead to positive effects for neighbouring
places, increasing their technical efficiency. Furthermore,
technical efficiency can be spread through ‘‘learning by
watching’’ or best practice imitation. In this sense, the
presence of a cluster of homogeneous regions, character-
ized by similar level of technical efficiency, is expected.
The analysis consists of two phases: first, a DEA
approach is run for each year considered; second, the
Moran’s I test is performed in each year.
Table 6 reports the test statistics and the respective
associated p values. Empirical evidence of the presence of
spatial effects is rather mixed. In four cases out of five, the
null hypothesis of absence of spatial autocorrelation cannot
be rejected; only in the last year of the sample, spatial
effects between DMUs’ scores efficiency are detected. The
Moran’s I test is also performed on the average DEA scores
of the period, and still the null hypothesis of absence of
spatial autocorrelation cannot be rejected (see last row of
Table 6).
8 Discussion and conclusions
This paper has added to the literature by examining the
productivity performance in the hospitality sector in a more
sophisticated manner through the use of a novel mix of non-
parametric and parametric approaches. Based on a macro-
economic framework, a dynamic WDEA has been employed
to evaluate and monitor the productivity of the hospitality
sector in Italy, over the period 2000–2004, based on the
official data availability. Assessing pure technical efficiency
in this economic sector has an important role in regional
planning and policy evaluation. Tourism activity and, hence,
hospitality has a key role in the Italian economy, though still
little is known on its performance.
Specifically, this paper has investigated the economic
efficiency of the Italian hospitality sector at a regional
level. Although Italy may be regarded as one single tour-
ism destination, given its geographical, cultural and his-
torical features, it also offers differentiated tourism
products and services that require a disaggregated
investigation.
Table 5 Post-DEA results (n = 105)
Variables Pooled-truncated estimation (1) OLS (2) Pooled-truncated estimation (3) OLS (4)
Coefficients Confidence intervals Coefficients Coefficients Confidence intervals Coefficients
Constant -28.73*** (99 %) = [-62.64, -27.69] 0.88*** -9.16*** (99 %) = [-23.96, -9.13] 0.86***
NRU -0.05** (95 %) = [-0.12, -0.03] 0.002 -0.01 (95 %) = [-0.06, 0.01] 0.002
CV_NRU 0.05*** (99 %) = [0.01, 0.10] 0.004*** 0.03** (99 %) = [0.02, 0.07] 0.004***
DART§ -1.56*** (99 %) = [-3.18, -1.56] -0.02 -1.26*** (99 %) = [-2.54, -0.93] -0.02
HQ_HOTELS -0.07** (95 %) = [-0.18, -0.05] 0.002
(1) Number of iteration = 10,000; (2) § reference group: no well-known art cities; (3) ** and *** indicate significance at the 5 and 1 %,
respectively
Table 6 Results of Moran’s I test
Year Moran I statistics p value
2000 0.078 0.130
2001 -0.031 0.633
2002 -0.083 0.427
2003 -0.051 0.505
2004 0.219 0.015
Average (00–04) 0.083 0.117
DEA scores are calculated with variable returns to scale and output
oriented
J Prod Anal
123
Empirically, the non-parametric WDEA has provided a
comparison of a DMU with respect to its own past per-
formance as well as the performance of the peer group.
Overall, the Italian regions show a relevant economic
inefficient performance throughout the period under
investigation. These findings are in line with other empir-
ical investigations showing that the hotel sector is charac-
terised by a lower productivity growth compared to other
businesses. This outcome is due, amongst other factors, to
the distinctive features of the hotel sector, which employs a
highly intensive labour productive process and is charac-
terised by excessive fixed costs, as well as a volatile
demand (Brown and Dev 2000; Kilic and Okumus 2005;
Goh 2010). On balance, the relatively most efficient region
is Lombardy, whose capital is Milan, which is one of the
most important centres of Italian finance and business
activity and is able to attract a relevant share of tourism
arrivals, particular foreigners with a high willingness to
pay, as reported in Federalberghi-Mercury (2005).
The key potential improvements that the hospitality sector
in Italian regions can achieve are based on the decrease of
labour and capital costs, though over time efficiency
improvements have been gained in terms of sales revenue
and added value. However, from the pooled-truncated ana-
lysis, it has emerged that various factors also influence the
level of Italian hospitality inefficiency. Poor utilisation of
infrastructure and high seasonal volatility have appeared to
be the main sources of economic inefficiency, while the share
of high quality hotels positively affects the sector efficiency.
This outcome has important implications. On the one hand,
the hotel quality variable can be thought as a proxy of the
overall environment, since it is plausible that 3–5 star hotel
accommodation are located in areas with a high level of
natural amenities; on the other hand, this finding links to
another empirical investigation run for the region of Sardinia
that shows how tourism demand is supply quality-driven in
the long run (Biagi and Pulina 2009). If one restricts the
attention to the Southern Italian regions, these results also
seem to support the work by Cracolici and Nijkamp (2009),
where Sardinia appears to be the region that has achieved the
best performance in comparison with its key competitors.
Environmental quality and a sustainable expansion of tour-
ism infrastructures turn out to be an important mix of
determinants which are not only able to drive tourism
demand, but are also likely to influence firms’ productivity.
The econometric investigation has further highlighted
that only well-known art cities, namely Florence, Rome
and Venice, drive the economic performance of the hos-
pitality sector in their regions, which are likely to be able to
compete more in local, internal and external markets.
Interestingly, the Moran’s I test has shown the lack of spill-
over effects amongst Italian regions. However, it is possi-
ble that spatial disaggregation matters. In fact, this outcome
is likely due to the loss of information during the process of
spatial data aggregation. Furthermore, the use of the
inverse of the distance between the centroids as a spatial
weight does not allow one to account for cross border
effects among tourist hot spots placed in different regions.
Future research may involve the use of provincial data,
characterised by a higher level of spatial disaggregation,
although it has to be said that such a data is rather difficult
to gather for a wider span of time.
Overall, the present findings have underlined important
features of the Italian hospitality sector. While the supply
of new infrastructure has continued to grow, Italian regions
have shown to be economically inefficient. Such an out-
come has also confirmed the empirical findings by Suzuki
et al. (2011) for the Italian provinces, although in that case
a statistic DEA is employed for the year 2001. The over-
investment, mainly designed to satisfy the high seasonal
demand, is one of the main sources of inefficiency. This
may turn into a large source of inefficiency and lack of
competitiveness. Additionally, the work force has less
incentive to perform efficiently since it is employed for
short periods of time throughout the year.
Although most of the Italian regions are characterised by
cities with an outstanding arts and historical heritage (e.g.
Bari, Bologna, Genoa, Palermo, Turin), this unique cultural
capital is not fully exploited as the driver of tourism supply
productivity. One of the reasons may be the low-profile
marketing campaigns, which are mostly run at a regional
level rather than by a central body able to activate a
comprehensive path of growth in decentralised areas and
less-known arts cities. Such a hypothesis has been further
confirmed by the spatial heterogeneity test, which has
shown no spill-over effects amongst the Italian regions,
confirming that Italian regions lack adequate networking
and joint marketing strategies. Ultimately, such a disen-
tangled development is likely to undermine the perfor-
mance of the less-known regions.
The present results are especially relevant for policy
makers and marketing managers, who should consider
synergic and complementary tourism policy for Italy as a
whole. This paper has drawn attention to the need to
reconsider an all-nation strategic approach to the tourism
activity. Although it is facing a long run life cycle, Italy is
endowed with an outstanding environmental and cultural
heritage that makes it inimitable, enabling it to operate in
the market place as one of the big players. Despite this
comparative advantage, Italy records a disharmonic growth
and a very heterogeneous performance at a regional level.
In a dynamic and globalised economy, characterised by
emerging destinations, there is the need to establish a
strong central body with a clear vision of the market
challenges, which may provide a substantive boost to
regain and maintain its actual position.
J Prod Anal
123
However, globalisation can not erase regional historical
and cultural roots. A comprehensive marketing strategy
should take into account the heterogeneous offer within the
Italian regions. One and a half century after the Italian
unification, the country is still characterised by ‘‘States’’ in
a ‘‘State’’ with their own arts, architecture, history, dialects
and enogastronomy and even secessionism ideologies. Yet,
this cultural diversity represents an asset for what is called
localnomics (or glocalisation). As argued by Salazar
(2010), localnomics is increasingly recognised to be a
winning strategy. In fact, the globalisation process of
tourism leads to an inevitably degree of worldwide inte-
gration and homogenisation in training, service and hos-
pitality benchmarks. However, tourism destinations seek to
diversify from the others in order to be more competitive
(Chang 1999). Such local differentiation, for instance
based on locally distinctive cultural identity and geo-
graphical peculiarity, can help to devise a solid strategy.
In this respect, Italian tourism, while developed and
distributed in a global market as a ‘‘unique product’’, still
needs to be tailored to accommodate heterogeneous con-
sumers’ preferences, allowing each region, and even pro-
vincial areas, to experience a renewal phase thanks to their
distinctive assets. Indeed, Italian tourism as an economic
activity, which has revealed to be more resilient to the
ongoing crisis than other economic sectors, offers an extra
advantage to competition if properly managed.
The dynamic non-parametric framework that has been
adopted in this study, combined with a parametric
approach, has provided insight into tourism industry dif-
ferences among the Italian regions. Unfortunately, this
powerful technique can not mitigate some relevant WDEA
limitations, such as variables selection and transferability
of the results, although it does provide an empirical
framework that can be used as a benchmark for further
investigations in the field.
Furthermore, since the efficiency of a given DMU is
measured with respect to other DMUs in the sample, the
obtained efficiency is always relative. The present paper
has two important implications. Firstly, it is not possible to
compare the efficiency of Italian regions sector with other
European competitor regions. Secondly, although the pro-
posed econometric approach captures the factors driving
the differences in efficiency among Italian destinations, it
does not allow one to detect the strengths and weakness of
Italian tourism industry in face of world competition.
From a methodological point of view, an interesting
result relates to the sensitivity of the post-DEA approach
used. Empirically, it has emerged that by ignoring the
property of the multi-step procedure by Simar and Wilson
(2007) one obtains biased results. Hence, the methodology
presented in this paper, although it refers to a single
country, can be applied to other destinations to find
common features and a more comprehensive assessment of
hospitality productivity. Future research can include the
collection and analysis of ad hoc questionnaires in each of
the Italian regions at a microeconomic level to obtain rel-
evant information on different segments of hospitality
supply. This information can be used to measure the spill-
over effects for the different categories of hotels and to
estimate a plant capacity or optimal size of hotels given the
state of travel and tourism in Italy.
Acknowledgments Manuela Pulina and Claudio Detotto acknowl-
edge the financial support provided by the Banco di Sardegna
Foundation (Prot. 1713/2010.0163). Juan Gabriel Brida acknowledges
the financial support provided by the Free University of Bolzano,
projects: ‘‘L’efficienza delle imprese turistiche in Italia’’ and ‘‘The
Contribution of Tourism to Economic Growth’’. Manuela Pulina
acknowledges the financial support provided by the Free University of
Bolzano (SECS-P/01—Economia Politica). The views expressed here
are those of the authors.
References
Anselin L (1988) Spatial econometrics: methods and models. Kluwer,
Dordrecht
Assaf AG, Agbola FW (2011) Modelling the performance of
Australian hotels: a DEA double bootstrap approach. Tour Econ
17(1):73–89
Assaf AG, Cvelvar LK (2010) The performance of the Slovenian
hotel industry: evaluation post-privatisation. Int J Tour Res
12(5):462–471
Assaf A, Barros CP, Josiassen A (2010) Hotel efficiency: a
bootstrapped metafrontier approach. Int J Hosp Man
29(3):468–475
Assaf AG, Deery M, Jago L (2011) Evaluating the performance and
scale characteristics of the Australian restaurant industry. J Hosp
Tour Res 35(4):419–436
Balcombe K, Fraser I, Latruffe L, Rahman M, Smith L (2008) An
application of the DEA double bootstrap to examine sources of
efficiency in Bangladesh rice farming. App Econ
40(15):1919–1925
Banker PC, Charnes A, Cooper WW (1984) Some models for
estimating technical and scale inefficiencies in data envelopment
analysis. Man Sci 30(9):1078–1092
Barros CP (2005) Measuring efficiency in the hotel sector. Ann Our
Res 32(2):456–477
Barros CP, Alves FP (2004) Productivity in the tourism industry.
IAER, August, 10(3). Technical University of Lisbon, Portugal
Barros CP, Dieke PUC (2008) Technical efficiency of African hotels.
Int J Hosp Man 27(3):438–447
Barros CP, Santos CA (2006) The measurement of efficiency in
Portuguese hotels using data envelopment analysis. J Hosp Tour
Res 30(3):378–400
Barros CP, Botti L, Peypoch N, Robinot E, Solonandrasana B, Assaf
G (2011) Performance of French destinations: tourism attraction
perspectives. Tour Man 32(1):141–146
Bernard AB, Jensen JB, Redding SJ, Schott PK (2007) Firms in
international trade. Working Paper 13054, http://www.nber.org/
papers/w13054. National Bureau of Economic Research,
Cambridge
Bernini C, Guizzardi A (2010) Internal and location factors affecting
hotel industry efficiency: evidence from Italian business corpo-
rations. Tour Econ 16(4):883–913
J Prod Anal
123
Biagi B, Pulina M (2009) Bivariate VAR models to test granger
causality between tourist demand and supply: implications for
regional sustainable growth. Pap Reg Sci 88(1):1–14
Bimonte S, Brida JG, Pulina M, Punzo L (2012) Tourism and growth:
stories of the two continents, Chapter 16. In: Punzo L, Feijo C,
Puchet Anyul M (eds) Beyond the global crisis: structural
adjustments and regional integration in Latin America and
Europe. Routledge Studies in the Modern World Economy,
USA, pp 252–268
Brida JG, Deidda M, Pulina M (2012) Investigating economic
efficiency in Italy: a regional comparison. Int J Rev Manag
6(3/4):175–198
Brown JR, Dev CS (1999) Looking beyond RevPAR; productivity
consequences of hotel strategies. Corn Hot Rest Admin Quart
40(2):23–33
Brown JR, Dev CS (2000) Improving productivity in a service
business: evidence from the hotel industry. J Serv Res
2(4):339–354
Bruni ME, Guerriero F, Patitucci V (2011) Benchmarking sustainable
development via data envelopment analysis: an Italian case
study. Int J Environ Res 5(1):47–56
Chang TC (1999) Local uniqueness in the global village: heritage
tourism in Singapore. Prof Geogr 51(1):91–103
Charnes A, Cooper WW, Rhodes E (1978) Measuring the efficiency
of decision making units. Eur J Oper Res 2(6):429–444
Charnes A, Clark CT, Cooper WW, Golany B (1985) A develop-
mental study of data envelopment analysis in measuring the
efficiency of maintenance units in the US air forces. Ann Oper
Res 2(1):95–112
Cooper WW, Seiford LM, Tone K (2000) Data envelopment analysis:
a comprehensive text with models, applications, references and
DEA-Solver software. Kluwer, Boston
Cortes-Jimenez I, Pulina M (2010) Inbound tourism and long-run
economic growth of Spain and Italy. Curr Issue Tour 13(1):61–74
Cracolici MF (2008) Assessment of tourism competitiveness by
analysing destination efficiency. Tour Econ 14(2):325–342
Cracolici MF, Nijkamp P (2006) Competition among tourist destina-
tion. An application of data envelopment analysis to Italian
provinces. In: Giaoutzi M, Nijkamp P (eds) Tourism and
regional development: new pathways. Ashgate, Aldershot, UK,
pp 133–152
Cracolici MF, Nijkamp P (2009) The attractiveness and competitive-
ness of tourist destinations: a study of Southern Italian regions.
Tour Man 30(3):336–344
Cullinane K, Song D-W, Wang T-F (2004) An application of DEA
windows analysis to container port production efficiency. Rev
Net Econ 32:184–206
Diewert WE, Mendoza MNF (1995) Data envelopment analysis: a
practical alternative. UBC Departmental Archives 95-30, UBC
Department of Economics
Emrouznejad A, Barnett RP, Tavares G (2008) Evaluation of research
in efficiency and productivity: a survey and analysis of the first
30 years of scholarly literature in DEA. Soc-Econ Plan Sci
42(3):151–157
Farrell J (1957) The Measurement of productive efficiency. J R Stat
Soc 120(3):253–290
Federalberghi-Mercury (2005) Rapporto 2005 sul sistema alberghiero
in Italia. www.federalberghi.it. Accessed on 31 Jan 2013
Federalberghi-Mercury (2010) Sesto rapporto sul sistema alberghiero
in Italia. www.federalberghiit. Accessed on 10 Jan. 2013
Federalberghi-Mercury (2012) DATATUR Trend e statistiche
sull’economia del turismo. Federalberghi & Format, Roma
Goh HN (2010) Recommending a productivity model for Singapore
hotels: A critical review of productivity models adopted by
researchers and hotel operators. UNLV Theses/Dissertations/
Professional Papers/Capstones. Paper 685
Griffith DA (2003) Spatial autocorrelation and spatial filtering.
Springer, Berlin
Helfand SM, Levine ES (2004) Farm size and the determinants of
productive efficiency in the Brazilian Center-West. Agric Econ
31(2/3):241–249
ISTAT (2011) Sistema di indicatori territoriali. http://sitis.istat.it/sitis/
html/index.htm. Accessed 12 Aug 2011
Kilic H, Okumus F (2005) Factors influencing productivity in small
island hotels: evidence from Northern Cyprus. Int J Cont Hosp
Man 17(4):315–331
Koksal CD and Aksu AA (2007) Efficiency evaluation of A-group
travel agencies with data envelopment analysis (DEA): a case
study in the Antalya region, Turkey. Tour Manag 28(3): 830–834
Kravtsova V (2008) Foreign presence and efficiency in transition
economies. J Prod Anal 29(2):91–102
Mandl U, Dierx A, Ilzkovitz F (2008) The effectiveness and
efficiency of public spending. Eur Comm Econ Pap 301:1–36
Manera Erbina C, Garau Taberner J, Molina de Dios R (2010) The
tourism revolution in the Mediterranean, 1950–2005. Documen-
tos de Trabajo, DT-AEHE 1014:1–17
Melao N (2005) Data envelopment analysis revisited: a neophyte’s
perspective. Int J Manag Dec Mak 6(2):158–179
Min H, Min H, Joo SJ (2008) A data envelopment analysis-based
balanced scorecard for measuring the comparative efficiency of
Korean luxury hotels. Int J Qual Rel Man 25(4):349–365
Molina-Azorin JF, Pereira-Moliner J, Claver-Cortes E (2011) The
importance of the firm and destination effects to explain firmperformance. Tour Man 32(1):22–28
Moriarty JP (2010) Have structural issues placed New Zealand’s
hospitality industry beyond price? Tour Econ 16(3):695–713
Neves JC, Lourenco S (2009) Using data envelopment analysis to
select strategies that improve the performance of hotel compa-
nies. Inter J Cont Hosp Man 21(6):698–712
Olesen OB, Petersen NC (2002) The use of data envelopment analysis
with probabilistic assurance regions for measuring hospital
efficiency. J Prod Anal 17(1/2):83–109
Prado Lorenzo JM, Garcıa Sanchez IM (2007) Efficiency evaluation
in municipal services: an application to the street lighting service
in Spain. J Prod Anal 27(3):149–162
Pulina M, Detotto C, Paba A (2010) An investigation into the
relationship between size and efficiency of the Italian hospitality
sector: a window DEA approach. Eur J Oper Res 20(4):613–620
Reynolds D (2003) Hospitality-productivity assessment using data-
envelopment analysis. Corn Hot Rest Admin Quart 44(2):429–449
Reynolds D, Thompson GM (2007) Multiunit restaurant productivity
assessment using three-phase data envelopment analysis. Int J
Hosp Man 26(1):20–32
Robert WJA, Haug AA, Jaforullah M (2010) A two-stage double-
bootstrap data envelopment analysis of efficiency differences of
New Zealand secondary schools. J Prod Anal 34(2):99–110
Salazar NB (2010) The glocalisation of heritage through tourism:
Balancing standardisation and differentiation. In: Labadi S, Long
C (eds) Heritage and globalisation. Routledge, London,
pp 130–147
Sampaio De Souza MC, Cribari-Neto F, Stosic BD (2005) Explaining
DEA technical efficiency scores in an outlier corrected environ-
ment: the case of public services in Brazilian municipalities.
Braz Rev Economet 25(2):287–313
Shuai JJ (2009) Web content and its influence on operational
performance-case of the hotel industry. Industrial engineering
and engineering management, IEEM, international conference,
8–11 December 2009, pp 885–889. doi:10.1109/
IEEM20095372884
Shuai JJ, Wu WW (2011) Evaluating the influence of E-marketing on
hotel performance by DEA and grey entropy. Exp Syst Appl
38(7):8763–8769
J Prod Anal
123
Sigala M (2004) Using data envelopment analysis for measuring and
benchmarking productivity in the hotel sector. J Trav Tour Mark
16(2):39–60
Simar L, Wilson PW (2007) Estimation and inference in two-stage,
semi-parametric models of production processes. J Econom
13(6):31–64
Simar L, Wilson PW (2011) Two-stage DEA: caveat emptor. J Prod
Anal 36(2):205–218
Sirirak S, Islam N, Khang DB (2011) Does ICT adoption enhance
hotel performance? J Hosp Tour Tech 2(1):34–49
Sun DB (1988) Evaluation of managerial performance in large
commercial banks by data envelopment analysis. IC2 Institute,
Austin
Suzuki S, Nijkamp P, Rietveld P (2011) Regional efficiency
improvement by means of data envelopment analysis through
Euclidean distance minimization including fixed input factors:
an application to tourist regions in Italy. Pap Reg Sci
90(1):67–89
UNWTO (2012) World tourism barometer, vol. 10, February 2012.
http://www.unwto.org/. Accessed on 28 May 2012
Wang FC, Hung WT, Shang JK (2006) Measuring the cost efficiency
of international tourist hotels in Taiwan. Tour Econ 12(1):65–85
Wanhill S (2011) What do economists do? Their contribution to
understanding tourism. Estudios de Economia Aplicada
29(3):679–692
Wilson PW (2008) FEAR: a software package for frontier efficiency
analysis with R. Soc-Econ Plan Sci 42(4):247–254
WTTC (2010) Progress and priorities 2009–10 http://www.
confederacaoturismoportugues.pt/downloads/get/id/192. Acces-
sed on 14 Oct 2013
J Prod Anal
123