Two-Stage Data Envelopment Analysis of Spanish Regions: Efficiency Determinants and Stability Analysis

The purpose of this paper is to examine the hypothesis that the efficiency of Spanish tourism regions for the period 2005-2013 is determined by a group of contextual variables. In contrast with monitoring reports based on descriptive methods, this paper uses the Data Envelopment Analysis (DEA) bootstrap semiparametric procedure to investigate efficiency determinants. An innovative analysis addresses the problem of the stability of efficiency estimates of random changes in the isolated exogenous variables. The statistical significance of the potential attractors can offer a tool for strategic decisions, and no previous work compares the stability analysis results to the estimates derived from the two-stage algorithm. The model appropriately fits the data, with all the coefficients being of the correct sign and statistically significant. Hence, the fact that the exogenous variables influence the hypothesis is confirmed by the results, and the stability analysis helps to verify the significance of each variable. We also extend the traditional DEA analysis by exploring efficiency and productivity changes using the slacks-based measure (SBM) model and the bootstrapped Malmquist index approach to obtain total productivity growth estimates.


Introduction
The Spanish-European tourism industry has recently experienced some of the most challenging times in its history. In 2010, recovery in Spain followed the global trend; in 2012, Spanish tourism revenues generated a surplus of 31,610 million euros, which was sufficient to cover the trade balance deficit of approximately In the current scenario of great pressure from the competition, performance is becoming a key issue.
In recent decades, the Destination Competitiveness Theory body of research has served as the basis for a number of studies, particularly in conceptual models such as Crouch and Ritchie (1999;2005), Ritchie and Crouch (2000b;, Mazanec, Wöber and Zins (2007), Crouch (2007;2011), Benito-López, Solana-Ibáñez andLópez-Pina (2014) and Assaf and Josiassen (2016). Emphasis has been placed on the clear need to direct research towards a better understanding of the attributes of competition. As a consequence, a growing number of initiatives has supported the need to measure and monitor tourist destinations. The 2015 Travel and Tourism (T&T) Competitiveness Index (TTCI) from the World Economic Forum (WEF) reveals that the world's leading country is Spain. The concern has sparked similar initiatives at the national level, such as the MONITUR report on Spanish Regions (also called Autonomous Communities -ACs).
Competitiveness refers to the ability to gain an advantage from available resources. However, the present study aims to extend the literature in Tourism Destination "Performance" by determining whether Spanish regions are using their resources optimally, or to what extent a destination is maximizing its outputs from its inputs. This paper therefore focuses on efficiency and productivity, as well as on testing the significance of the determinants potentially affecting performance. Malmquist index approach is also presented to obtain total productivity growth estimates.
Second, this work will analyze the hypothesis that the efficiency of Spanish tourism regions is deter-mined by a group of contextual or exogenous variables that can explain the level of efficiency. This analysis is conducted by applying the Simar and Wilson (2007; procedure to bootstrap the DEA scores with a truncated regression to estimate the effect of a selection of factors on robust DEA estimates. The identification of tourism performance determinants is not an aim of this study, however, as this has already been investigated in other works such as 2016).
Third, to determine the significance of each variable, an innovative analysis is included with the goal of studying the efficiency estimates and stability given small changes in the isolated variables of the problem. For this purpose we define the Stability Coefficient, whose magnitude reveals the effect of each exogenous variable in the efficiency estimates, thus complementing the estimates derived from the Simar and Wilson (2007; algorithm.
The study is important because the significance or non-significance of a certain factor can provide tourism policymakers with accurate information for future strategic decisions. Moreover, no previous work has compared stability analysis results to the estimates derived from the two-stage double bootstrap algorithm used. This paper is organized as follows. In the next section, the theoretical framework is presented. The third section explores the methodology. The fourth section is devoted to the sample and variables chosen for the first-stage and second-stage DEA analysis. In Section 5, we present the results of the DEA basic radial and SBM models, as well as the productivity growth estimates and the analysis of the efficiency determinants obtained from applying the two-stage procedure to Spanish regions for the period 2005-2013. Finally, we present our conclusions.

Theoretical framework
International tourism has the potential to be a driving force in the economies of industrializing countries during the 21st century, especially in Asia. Countries like Spain must develop strategies to make use of their comparative advantages to achieve competitive advantage, since, as Gooroochurn and Sugiyarto (2005, p. 25) predict: "the issue is especially important for countries that rely heavily on tourism". Strong competi-Two-Stage Data Envelopment Analysis of Spanish Regions: Efficiency Determinants and Stability Analysis tion remains a critical factor in Europe, where providers struggle to contain prices because tourists travel nearer to home and for shorter periods. In Spain, the government gives significant priority to the Travel and Tourism (T&T) sector; the government collects comprehensive data on it the and makes strong efforts to attract tourists through destination marketing campaigns. That said, during the last five years businesses have been forced to react with offers, discounts and deferred payment options.
It is therefore noteworthy that managing destination competitiveness has become a major topic of interest, and many researchers have studied its concepts, models and determinants: a good overview can be found in Mazanec et al. (2007), Tsai, Song andWong (2009), Crouch (2011), , Benito-López et al. (2014) or Marco-Lajara, Úbeda-García, Sabater-Sempere and García-Lillo (2014). The initial group of studies has sought to develop general models and theories of destination competitiveness. In the 1990s, Crouch and Ritchie established a comprehensive framework for tourism destination management - Crouch and Ritchie (1994;1995;2005), Ritchie and Crouch (1993;2000a;2000b; -with five main groups of destination competitiveness factors and 36 destination competitiveness attributes. In this regard, Heath (2003) developed a model based on Ritchie and Crouch (2000b), who established the initial framework of destination competitiveness. Furthermore, other models addressing this issue include those by Dwyer and Kim (2003), Dwyer, Mellor, Livaic, Edwards and Kim (2004), Enright and Newton (2004) and Crouch (2011).
In line with this rising interest, indices such as WEF TTCI, or the Spanish MONITUR, aim to measure the factors and policies that make it attractive to develop the T&T sector in different countries. The TTCI is based on three categories, each of which comprises a total of 14 pillars; within each pillar there are 75 final variables. The scores obtained by country are compared with those of the previous report; e.g., the final report of 2015 contains detailed information regarding each of the 141 countries covered by the study.
These TCCI type indices are descriptive, as noted by Assaf and Josiassen (2012, p. 394) "While the TTCI is probably the best known instrument used to rank nations according to their travel and tourism competitive-ness, it is important to note that it is not a performance index" … "it is not possible from this index to determine which inputs can be translated into industry performance most efficiently".
It must be determined whether a given tourism attractor is statistically significant. This could provide tourism policymakers with accurate information to use to make successful future strategic decisions. The TTCI calculates unweighted means, which implies that factors are equally important. Put differently, in Thailand, for example, the factor "hotel rooms" has the same importance as "primary education enrollment", which may be cause for suspicion. Furthermore, the impact of a competitiveness attribute on the destination relative to performance is a function of both the importance of the attribute and the degree to which destinations vary on the attribute. The same problem can be addressed in Spain, where the MONITUR report is of relevance and provides a comprehensive list of determinants that drive tourism performance, and a global index value of Spanish regions competitiveness is analyzed (ACs).
Beyond the previously mentioned link with Tourism Competitiveness, our goal is to study Tourism Performance with a special focus on the methodological aspects of the determinants of Tourism Destination Performance covered by statistical analysis. Recent publications have been devoted to investigating and testing which determinants concretely affect tourism performance, a main objective of this paper. The procedure consists of the development of a tourism performance index using the Data Envelopment Analysis (DEA) methodology, which involves the use of a linear programming formulation to construct a non-parametric frontier over the data. The statistical properties of DEA efficiency estimates can be explored via the use of the bootstrap approach, making it possible to obtain confidence intervals. The panel data structure is optimal for measuring whether the productivity of Spanish regions has progressed or regressed over time, and to this end, the Malmquist productivity index is used, which is a quantity index defined using the distance functions ratio and its classical decomposition. The bootstrapping procedure can thereby be extended to determine the statistical properties of the Malmquist index. In this line, several studies examine productivity using frontier models such as DEA, and a good overview in this regard can be found in Agbola (2011), Fuentes (2011), Barros et al. (2011) and Ribes, Rodriguez and Jiménez (2011).
The Simar and Wilson (2007) double DEA bootstrap procedure is used to evaluate how efficiency varies with the selection of determinants of tourism destination performance. This two-stage procedure used is relatively novelty because only a few very recent studies of this type can be found: Barros and Dieke (2008b) Farrell (1957) is the pioneering empirical work to estimate efficiency scores, which has been popular-ized by Charnes, Cooper and Rhodes (1978) and Banker, Charnes and Cooper (1984)

Data Envelopment Analysis
CRS measures the overall efficiency for each unit (pure technical efficiency and scale efficiency). The variable returns to scale (VRS) efficiency model, by Banker et al. (1984), is estimated by restricting Σλ j =1; it provides measures of pure technical efficiency. The scale efficiency score by Färe Grosskopf and Lovell (1985) is obtained by dividing the CRS score by the VRS score. The estimates of the efficiency scores, j δ (j=1,2,…n), are bounded between unity and infinity. A unitary value implies that the observed production coincides with the potential production and that the DMU is efficient. If it exceeds the unity, the DMU is not efficient.
As an alternative to conventional radial DEA models, we also consider the Slack-Based-Measured (SBM) model by Tone (2001). Its non-radial efficiency measure draws on all inefficiency sources, offering a more exhaustive explanation regarding why a destination may become relatively efficient or inefficient over time. The non-oriented CRS SBM The SBM efficiency score, θ, is between 0 and 1, considering that if θ=1, the region is efficient. When a region becomes SBM efficient, all slacks (regional input excesses and output shortfalls) are zero in any optimal condition, being the destination located on the efficiency frontier.

Malmquist Productivity Index
To measure whether the productivity of Spanish regions has progressed over time, we use the Malmquist productivity index, a quantity index defined using the ratio of distance functions that was originally introduced by Malmquist (1953). Following the decomposition by Färe, Grosskopf, Norris and Zhang (1994), the MI index between two periods, t and t + 1, is cal- The decomposition takes the Färe, Grosskopf, Lindgren and Roos (1994)

Efficiency Determinants and Stability Analysis
The causes of inefficiency are analyzed by considering a group of external factors, denoted by Z∈Ƶ⊂R r ; such variables, which are neither inputs nor outputs and are not under control of the DMU, may influence the production process.
The two-stage approach by Simar and Wilson (2007), which is complemented in Wilson (2011, 2015), has assumed the turning point in the treatment of exogenous factors. The model takes the following form: As true efficiency scores, δ i , are not observed in the first stage, technical efficiency is estimated by DEA ignoring Z. Estimates from the first stage, î δ , or biascorrected estimator, î δ , replace the unobserved δ i and, in the second stage, are regressed on environmental covariates, z i . In accordance with Simar and Wilson (2007), a truncated normal distribution is assumed.
The statistical significance of each exogenous variable under the two-stage procedure can be complemented through the Stability Analysis. Concretely, it is relevant to know how changes in the exogenous variables may affect efficiency. Suppose that x is denoted as a n-tuple of real numbers, representing one of the exogenous variables of our problem in a particular year. If we consider that the efficiency coefficients vector is an m-tuple of real numbers denoted by f, we introduce the stability coefficient, Ω, following Trefethen and Bau (1997):

Sample and variables
Our initial sample comprises data from 17 Spanish ACs between 2005 and 2013. We will consider that the regions' goal is to achieve maximum output once given inputs. In this sense, according to Botti, Peypoch and Solonandrasana (2008), Barros et al. (2011) or

Efficiency and Productivity
The first stage in the assessment, i.e., considering only the discretional input and output variables, provides the efficiency coefficients for the DEA ratio outputoriented models. As known from Simar and Wilson (1998) or Wheelock and Wilson (2008), these DEA estimators are biased downward, and this must consequently be considered.
The Farrell type DEA score is between 0 and 1, meaning the efficiency of a DEA score equal to 1. to achieve the optimal scale.

Two-Stage Data Envelopment Analysis of Spanish Regions: Efficiency Determinants and Stability Analysis
As an alternative to compare with the radial VRS model, the VRŜ δ in Table 1 and Under this assumption the average efficiency score is 0.656 (0.606 for the average period data in Table 2).  We also briefly analyze the productivity growth results from bootstrapped Malmquist Index decomposition. Table 3    The decrease in technical efficiency and more specifically, in PTE, is the most relevant factor in the example under analysis. Following Barros (2005, p. 181), this crucial member is a consequence of several factors; among them, in the Spanish case, structural rigidities associated with the labor market can be highlighted because they are potentiating the lack of a link between job tenure and performance. As an example, we can mention the Balearic Islands region.

Efficiency Determinants and Stability Analysis
Relative to the second-stage regression we have applied Simar and Wilson's (2007)

algorithm-II (their Monte
Carlo experiments confirm that it shows better functioning). In the two-stage Simar and Wilson (2007) procedure, the first stage estimated scores under the VRS assumption that are regressed in a truncated normal regression model on the group of environ-   The first columns include the Shephard (1970) output VRS distance function estimates (Eff), the es- As efficiency is measured in terms of Shephard's (1970) output distance function, which is the reciprocal of the Farrell (1957)

Discussion and conclusions
Tourism is an economic sector with a clear lack of research methodologies and applied studies, and the use of parametric and semiparametric techniques is   Table 4), suggesting that Spanish regions performed approximately 56% under their efficiency possibilities in these years. The productivity results follow the same trend.
At this stage in the literature development, there is a good basis of information of how to identify relevant attributes and, in particular, how to turn the focus of research toward assessing the relative importance of these attributes. Although the TTCI is the best known instrument used to rank nations according to their T&T competitiveness, it is important to note that it is not a performance index.
It is true that Tourism Attraction may increase the sources of revenue and subsequently improve destination performance, but we need to know whether the determinants are statistically significant and to rank them. To this end, we have addressed a promising path.
The significance of the factors under consideration, or lack thereof, can provide tourism policymakers with accurate information to use for future strategic decisions. Expert opinion is certainly a worthy mechanism, but mathematical programing techniques better allow us to draw the objective initial setting. Along these lines, the results in Table 4, and more concretely, the estimated coefficients in (7), have the correct sign and are statistically significant at 5% for COAST, MU-SEUM, MICE, NATUR, SKI, FOOD and SHOP in influencing Spanish regions' performance when these coefficients are considered as tourist attractors.
The Stability Analysis strengthens knowledge concerning the significance of the exogenous variables. Non-significant attractors are related to values of Ω less than unity, whereas significant ones have a stronger impact as Ω grows. The latter fact allows the attractors to be classified as strong and weak. That said, an additional question remains regarding the sustainability of the touristic model. Spain is currently receiving more tourists than in previous years, and these tourists are spending more than before.
At the same time, our mature model has moderated growth rates, so we are moving into an environment of increasingly strong competition where efficiency is the key subject. Following UNWTO recommendations, it is essential to foster responsible tourism in all aspects -economic, social and environmental -promoting sustainable growth as a consequence. Therefore, a future research line would be to perform a detailed analysis of the necessary link between performance and sustainability. A part of the Spanish success may Two-Stage Data Envelopment Analysis of Spanish Regions: Efficiency Determinants and Stability Analysis come through some new phenomena that is far from sustainability, with a special mention of pubcrawling, or the act of one or more people drinking in multiple pubs or bars in a single night. The inclusion of some variables acting as undesirable outputs could be a way to examine this imperative link.
Future research for the 2013-2015 period may confirm these results and include other potential performance determinants. Since 2010, the European Union (EU) has clearly been involved and is worried about the tourist industry, aiming to stimulate competitiveness and performance in the sector. That said, a new paradigm must be open to create a new link between tourism and the public and private sectors. New types of consumer behavior are inevitably linked to a generalized increase in levels of income, as well as to the demographic evolution of Spain's main original markets. Tourism must therefore respond, which entails a radical rethinking of the traditional ways of defining, structuring and distributing tourist products. Inefficient Spanish regions must respond to individuals who are seeking integrated experiences that surpass their expectations. This "cluster services approach" is now considered essential to ensuring tourists' full satisfaction. There is therefore a need for competitive interdependence on the part of all actors at destinations taking into account the management model used by each destination. The expectation is that regions representing a "cluster attraction" will have a better future projection and will gradually gain in competitiveness.
This study presents several limitations. The performance of each region depends to some extent on the type of tourist considered: resident or non-residents.
Furthermore, the exogenous variable selection is strategically crucial, and institutional involvement is required for deeper scrutiny.
In any case, the DEA technique offers new insights related to the topic of performance to be considered, i.e., the natural way to enhance competitiveness and performance.