1 Introduction

Tourism generated US$1482 billion in 2019 [1], representing 10.3% of global GDP, 6.8% of global exports, and 28.3% of global services exports [2]. Its direct impact also accounted for 330 million employments and US$948 billion capital investment (4.8% of total investment) [2]. During 2019, international tourist arrivals registered 1460 million worldwide, 14.7% of visits to the Americas [1]. In this context, Mexico counted 45 million international tourists and US$24.5 billion in international tourism arrivals, ranking 7th and 16th places respectively among countries in the world [3].

Tourism is an activity that is preferably measured by the number of national and international tourists who arrive at a certain destination. However, these arrivals are influenced by various aspects that limit or drive them. One of the most important driving factors in the tourism industry is the reputation of each tourist destination. According to Coelho and Gosling [4], the reputation of a touristic destination is influenced by four main factors: (1) communication (social media, internet, touristic guides), (2) individual consumers’ evaluations, (3) local specific experiences, and (4) time that creates reputation over a longer period. Tourist arrivals can also be influenced by dangerous situations that negatively affect the perception of tourist destinations. Sánchez Mendoza [5] points out that the proximity of these violent or dangerous events for tourists determines the perception of fear of a threat or danger.

The tourism industry includes different economic sub-sectors considered as characteristic activities like hospitality (accommodation services for visitors), transportation, food and beverage serving activities, travel agencies and other reservation services activities, cultural activities and sports and recreational activities [6]. Hospitality sector equals to 3.7% of the global GDP, which also represents 35.9% of the global touristic GDP [2]. In Mexico, the hospitality sector represents 2.5% of the total GDP and 29.2% of the total tourism GDP in the country [7]. In Mexico, during the last two decades, the arrival of national and international tourists has increased, as well as the offer of services. The GDP of hospitality went from US$ 16.084 constant (2013 = 100) billion in 1992 to US$ 20.284 constant (2013 = 100) billion in 2018, i.e., a growth of 28% during that period [8]. Employment in the hospitality sector grew by 33.8% from 148,713 people in 1993 to 198,977 in 2018 [8]. Similarly, the arrival of national tourists in hotel rooms grew from 7.4 to 27.1 million, while the arrival of international tourists grew from 34.7 to 100.4 million in the same period [9]. Hospitality infrastructure can be measured as the number of disposable rooms. Mexico has approximately 23,700 hotels, with an average of 646,304 available rooms daily in 2019, while in 1992, it was only 275,441. In 2019, 10 out of 32 Mexican states concentrated around 53% of all the hotels and 57% of hotel rooms in Mexico [9].

Although tourism generates significant revenues, a large percentage of these revenues is sent to the hotels’ international investors or gained by the local rich individuals, but only a few revenues belong to poor neighbours [10]. Therefore, it is important for the government to optimize resource allocation to tourism development, i.e., to foster tourism activities. Hospitality efficiency is an important aspect of tourism research. The level of contribution of the hospitality industry depends on the production factors acting efficiently to improve profitability and market position [11], and in the case of Mexico to increase the number of tourists in the states of the country.

The technical efficiency of hospitality is a comparative measure of how effectively it processes inputs to produce outputs in relation to its production possibility frontier, which represents its maximum potential for doing so. Technically, a hospitality unit can be ineffective if it operates below the boundary [12]. Efficiency in the hospitality industry must be based on the identification of new markets and products. Alberca and Parte [13] suggested that hotels should diversify their offerings according to the market, because tourist demand is based on more differentiated needs and expectations. Through this strategy, hotels can reach higher levels of efficiency since their demand will not be linked to specific seasons. Cázarez and Schütze [14] demonstrated that hotel organizations that use information obtained from online social networks to develop new products or services to reach new markets are increasing their efficiency.

As most of the countries all around the world are affected by the Covid-19 pandemic situation, it is crucial to search for policies of how to rescue the touristic industry. In this light, Mexico was identified among the 13 most vulnerable countries according to the tourism industry post-pandemic recovery, due to its orientation on international travellers and a high presence of tourism in the national economy [15]. The current situation is represented by the lack of economic resources due to the closure of economic activities in many countries. That is why governments must effectively use their resources and target their support effectively. Therefore, it is important to apply advanced tools to determine leverage points for their policies.

The article's objective consists of evaluating the efficiency of the hospitality industry in the 32 states of Mexico. To have a complex view, the analysis is based on the infrastructure (number of rooms available) each of the states has, the arrival of national and international tourists, and their length of stay for 1992–2018. This longitudinal analysis enables us to obtain a precise picture of tourism development in Mexico and its regions.

1.1 Efficiency and performance analysis in tourism

Efficiency in tourism is an important indicator for measuring the level and quality of tourism development. That is why such analyses have attracted considerable research attention. Many quantitative and statistical methods can be applied to evaluate the efficiency and performance in tourism. For the benchmarking techniques, the frontier analysis has become the most noteworthy approach in the tourism and hospitality literature. Data Envelopment Analysis (DEA) is a non-parametric modeling technique that belongs to the most often used methods for assessing the efficiency and performance of the set of decision making [16, 17] with successful applications in various industries.

De La Hoz et al. [18] evaluated academic efficiency of 256 engineering programs at Colombian universities. Flegl et al. [19] applied DEA to observe the production and investment efficiency of 1672 municipalities in the Mexican food industry. Chopra and Ramachandran [20] used DEA model to analyze water sector performance in 11 states in India. Linh Le et al. [21] investigated the total factor productivity growth and environmental efficiency of the agricultural sector in East Asian countries for the period from 2002 to 2010. Jablonsky [22] assessed countries’ performance at the 2016 Summer Olympic Games with respect to the resources each country can spent. Vikas and Bansal [23] evaluated efficiency of 22 Indian oil and gas sector during 2013–2017 period.

In tourism, the DEA has been applied to evaluate regional differences, the efficiency of the hotel industry, or to determine the influential factors in the tourism industry [24]. At the hospitality level, summarized in Table 1, Oliveira et al. [25] analyzed the impact of hotel quality (star rating) on the efficiency of 84 hotels in the Algarve, Portugal. Barros et al. [26] used the DEA to estimate the efficiency of operational activities of 15 Portuguese hotel groups. Hathroubi et al. [27] evaluated performance of 42 Tunisian hotels considering their environmental managerial practices to enhance their competitiveness. Higuerey et al. [11] measure efficiency and productivity of 147 hotels in Ecuador considering their quality and geographical location.

Table 1 A review of the literature on efficiency and performance assessment in tourism using DEA
Full size table

At the national and regional level, Corne [28] applied Data Envelopment Analysis to analyze efficiency in the French hospitality sector in 16 conurbations to identify possible improvement in the sector. Similarly, Liu et al. [29] evaluated the efficiency of 53 Chinese coastal cities from 2003 to 2013 to explore regional differences, whereas Chaabouni [30] investigated the tourism efficiency and its determinants in 31 provinces in China over the period 2008–2013. Song and Li [31] estimated the efficiency of the Chinese tourism industry from the sustainability point of view to increase tourist attraction, whereas Wang et al. [32] used a super-DEA model to evaluate the tourism efficiency of 30 provinces in China. Furthermore, Oukil et al. [33] applied DEA methodology to examine efficiency in the hotel industry in Oman to identify variables explaining inefficiency in the industry. Flegl et al. [34] analyzed the hospitality efficiency in 67 main touristic centers in Mexico regarding national and international tourism. Castilho et al. [35] observed the impact of tourism on the eco-efficiency of 22 Latin American and Caribbean countries, and Niavis and Tsiotas [36] used the DEA to evaluate the performance of 37 Mediterranean regions.

1.2 Efficiency and performance analysis in Mexican tourism

Nurmatov et al. [24] observed that the DEA applications in tourism are mainly concentrated in Europe and Asia regions, with negligible attention in Latin America. Up to our knowledge, the DEA analysis has been very little used to assess the efficiency and performance in Mexican tourism (Table 2). Camacho [37] used the DEA to determine the efficiency of 81 touristic destinations in attracting domestic and foreign tourists during the period 2000–2010. Cázares and Schütze [14] applied the DEA method to measure the online social network efficiency and their effect on new tourism products in Mazatlán, México. The analysis of the hospitality efficiency in 67 touristic centres made by Flegl et al. [34] found significant differences between foreign and national tourism efficiency. Finally, Kido-Cruz et al. [38] analyzed the technical efficiency of 59 Mexican municipalities and its impact on the poverty index. They concluded that these two variables are unrelated.

Table 2 A review of the literature on efficiency and performance assessment in tourism using DEA in Mexico
Full size table

2 Materials and methods

2.1 Data Envelopment Analysis

Data Envelopment Analysis (DEA) allows to evaluate the set of homogeneous decision-making units (DMU) regarding their capabilities to convert multiple inputs into multiple outputs [39]. Each DMU consumes \(m\) different inputs to produce s different outputs. The development of the DEA model theory started with the pioneering work [40]. The model formulated in their paper assumes constant returns to scale production technology. It is further referenced as CCR model. The linearized version of the CCR output-oriented model for the evaluation of the \({\mathrm{DMU}}_{0}\) is formulated as follows:

Minimize

$$q={\sum }_{i=1}^{m}{v}_{i}{x}_{i0}$$
(1)

subject to

$$\begin{gathered} \mathop \sum \limits_{i = 1}^{m} v_{i} x_{ij} - \mathop \sum \limits_{r = 1}^{s} \mu_{r} y_{rj} \ge 0,j = 1,2, \ldots ,n. \hfill \\ \mathop \sum \limits_{r = 1}^{s} \mu_{r} y_{r0} = 1, \hfill \\ \mu_{r} ,v_{i} \ge \varepsilon \hfill \\ \end{gathered}$$
(2)

where \({x}_{ij}\) is the quantity of the input \(i\) of the \(DM{U}_{j}\), \({y}_{rj}\) is the amount of the output \(r\) of the \(DM{U}_{j}\), and \({\mu }_{r}\) and \({v}_{i}\) are the weights (multipliers) of the inputs and outputs, and \(\varepsilon\) is a non-Archimedean constant necessary to eliminate zero weights of the inputs and outputs. \({\mathrm{DMU}}_{0}\) is efficient if \(q=1\), i.e., there is no other DMU that produces more outputs with the same combination of inputs. Whereas DMU is inefficient if \(q>1\).

2.2 Window analysis

To measure DMUs productivity over a longer period, the Windows Analysis (WA) approach can be used. This approach works on the principle of moving averages to detect DMUs performance trends over time [41]. The performance of a DMU in a particular period is compared to its performance in other periods, in addition to the performance of other DMUs. Therefore, there is \(n.k\) DMU in each window, where \(n\) is the number of DMUs in a given period (it must be the same in all periods) and \(k\) is the width of each window (same for all windows). This feature increases the discriminatory capacity of the DEA model, as the total number of \(T\) periods is divided into series of overlapped periods (windows), each with a width \(k\left(k<T\right)\) leading to \(n.k\) DMUs. The first window has \(n.k\) DMUs for periods \(\left\{1,\dots ,k\right\}\), the second period has \(n.k\) DMUs and periods \(\left\{2,\dots ,k+1\right\}\), and so on, until the last window has \(n.k\) DMUs and periods \(\left\{T-k+1,\dots ,T\right\}\). In total, there are \(T-k+1\) separate analyses where each analysis examines \(n.k\) DMUs.

An important factor is the determination of the size of the window. If the window is too narrow, there may not be enough DMUs which leads to a low power of model discrimination. Conversely, a too wide window can yield misleading results due to significant changes occurring during periods covered by each window [39]. Therefore, the size of the window should consider the structure of the DEA model and the characteristics of the analysed area.

The attractivity of a tourist destination can be significantly affected by negative reports by the media [4, 42]. The negative reputation reported by media can be linked to international conflicts, acts of terrorism, criminality, natural disasters or health concerns. There is no consensus about the length of the recovery time from each reported case. This recovery can range from several months to several years depending on the magnitude of each incident and the tourist’s personality type [43]. To minimize the effects of short-term negative events that would cause high volatility in the results obtained, the length of the window was selected as \(k=24\) (two-year window).

2.3 Data

For the analysis, we used data from the DATATUR database (Secretariat of Tourism, 2018). Monthly data related to hospitality activities of 32 Mexican states were collected for the period from 1992 to 2018 (i.e., 27 years or 324 months). According to Lee et al. [44] and Oliani et al. [45], the quality and capacity of hotels’ infrastructure (among others) play an important role in tourism. However, the quality of hospitality service and establishments are rarely used in the tourism analyses [24].

The star rating is commonly used to express the level of hotel quality [25, 28]. Therefore, to include the quality and capacity of the hospitality service in each state, we selected the following variables as the inputs of the DEA model: Number of one-star hotel rooms, number of two-star hotel rooms, number of three-star hotel rooms, number of four-star hotels room, and number of five-star hotel rooms. So, the increase of the inputs does not necessarily correspond with the investment as the construction of the one-star hotel room represents significantly lower costs in comparison with the higher quality rooms. Higher efficiency could represent the proper mix of rooms, which would fit the demand of the tourist for the specific region.

The objective of the hospitality sector is usually to maximize the occupancy rate and, consequently, their revenues. That is why the DEA analysis usually includes occupancy rate, tourists’ arrivals, and related revenues per available room as outputs [27,28,29,30]. However, the absolute number of tourists’ arrivals and the occupancy rate avoid reflecting the number of nights tourists stay in each destination. Instead, the output part of the constructed DEA model is represented by tourists’ nights (TN), which can be expressed as

$${\text{TN}} = {\text{tourists'}}\;{\text{arrivals*average}}\;{\text{number}}\;{\text{of}}\;{\text{nights}}.$$
(3)

Including the average number of nights each tourist stays in the model corresponds to the approach presented by Oukil et al. [33] or Hathroubi et al. [27]. The complete dataset is available in Flegl and Cerón-Monroy [46].

Tourism is an activity that is characterized by having months with a greater influx for reasons of already established holiday periods and climatic issues. Therefore, it is necessary to make a seasonal adjustment that eliminates the fluctuation that obscures the trend-cycle component of the series as much as possible [47]. The moving averages is the seasonal adjustment method used in this article. It consists in estimating coefficients that capture the difference between the real value of each month and the 12-month average value around the real value, i.e., 6 months ago and 6 months ahead. Each monthly observation has an adjustment coefficient which must be averaged with all the coefficients of all the years to have a single seasonal adjustment factor (\({F}_{i}\)) for each month throughout the study period.

$${F}_{i}=\frac{\sum_{j=1}^{n}{f}_{j} ({RV}_{1} / \left(\sum_{i=1}^{12} \left({RV}_{-6, -5,\dots -1.}{RV}_{+1,..+6}\right)/12\right)}{n}$$

where \({F}_{i}\) = Seasonality month factor, \({f}_{j}\) = seasonality month factor per year, \(i\) = month, \(j\) = year, \(RV\) = Real value.

2.4 Models

Three different models were constructed: (1) overall model, where the output side of the DEA model includes the total number of tourists; (2) international model, which only includes data for the international tourists’ arrivals to Mexico; and (3) national model, which includes data for the national tourists’ arrivals.

The advantage of the DEA methodology is the possibility of benchmarking DMUs of different sizes and locations if the homogeneity requirement is not violated [39, 48]. Although we evaluate Mexican states of different sizes and locations, the homogeneity is not violated as all operate on the same market (Mexico) and use the same type of inputs (technology). This consideration is alike to Chaabouni [30], who applied DEA to investigate the tourism efficiency of Chinese provinces, Corne [28] and the analysis of the French administrative departments or Liu et al. [29], who assessed the eco-efficiency of Chinese coastal cities.

Considering the operation of the Window Analysis method, 768 DMUs were available in each window, resulting in 20,736 analyses in total (\(n=32\) states, \(k=24\) width of the window, 27 years). This ensured sufficient discriminatory ability of the model [48]. Further, the output-oriented DEA model was used as the analysis aims to provide the optimal number of arrivals (TN) rather than to optimize the inputs structure of the model in each state. First, obtaining information about a target TN can be useful for decision-making strategies in attracting more tourists. Second, cutting the overall capacity (i.e., closing hotels) will likely lead to only short-term efficiency improvements, as the states may lack the capacity in the future. Finally, the CCR model was selected because we do not consider competition among the 32 Mexican states, as the Mexican Secretary of Tourism (SECTUR) develop economic activities in each state to encourage tourism to be competitive at a national and international level.

3 Results

The DEA model allows the maximum output to be set in a virtual way with the corresponding input. In this case, if the states of Mexico have an average number of rooms available per day and receive a greater flow of tourists, they will be efficient, otherwise, they will be inefficient. Table 6 presents the data for 1992 and 2018 of the inputs and outputs by state. We can observe that the available rooms grew by 130%, national tourists by 189% and international tourists by 265% between these two periods. There are significant differences in the growth rates among the states, as well as there are several states (Colima, Durango, Guerrero, Michoacán, and Oaxaca) with a negative rate in the case of international tourists.

With the data for the selected period, the efficiency levels are highlighted considering the model, that is, with the total number of tourists. Subsequently, the results are presented for the arrival of national tourists and the arrival of international tourists.

3.1 Overall model

The average overall efficiency of the hospitality sector in Mexico during the period 1992–2018 was 65.02% with a standard deviation (SD) of 12.98%, where 17 states were evaluated above the average (Table 3). The best evaluated state is Quintana Roo with an average efficiency of 88.25% (SD 8.19%), followed by Sonora (74.71%, SD 10.99%), Colima (74.41%, SD 12.52%), Puebla (74.09%, SD 12.19%) and Coahuila (73.54%, SD 17.72%). Contrary, the lowest efficiency of the hospitality sector is observed in Estado de México (51.29%, SD 10.32%), Baja California (52.61%, SD 11.25%), Aguascalientes (55.60%, SD 10.12%), Morelos (56.60%, SD 15.13%) and Veracruz (57.57%, SD 11.45%).

Table 3 Hospitality efficiency of Mexican states, overall model 1992–2018
Full size table

Figure 1 displays the evolution of the efficiency throughout the whole period. We can observe an initial drop in the overall efficiency during the year 1995, where the average efficiency from March to November in the entire Mexico was 61.20%, compared to the same period in 1994 with the efficiency of 71.53%. After the recovery of the hospitality sector in 1996 and 1997, the efficiency of the whole sector had a decreasing tendency during the following decade. At the end of this decade, the hospitality sector had an average efficiency of 54.41% from January 2005 to December 2006 (the lowest efficiency was recorded in April 2005 of 47.86%). After this period, the sector recovered during the following year, with the peak in March of 2008 (77.89%). The following decade presented a slight constant decrease, followed by a slight growth in the last year 2018. During this decade, the results indicate a significant drop in efficiency in May 2009 (50.11%, − 19.94% compared to April 2009) due to the peak of the H1N1 pandemic [49], which was recovered immediately in the next month, mainly thanks to the promptly implemented actions against the spread of the pandemic [50].

Fig. 1
figure 1

Average hospitality efficiency of Mexican states, comparison of overall, national and international models, 1992–2018

Full size image

Table 7 (in appendix) divides the average efficiency of all 32 Mexican states into nine 3-year-long periods to capture the variation in the hospitality efficiency on the period-to-period basis. The highest stability of the hospitality efficiency is observed in case of the best evaluated Quintana Roo, which average efficiency is 88.25% has remained mainly steady during the whole period with an average change from period-to-period of − 0.49% and SD of 3.66%. Sonora, as the second-best evaluated state with the efficiency of 74.71% reported the period-to-period volatility of 6.92% and, what is more, its average efficiency has grown by an average of 1.66%. The biggest positive period-to-period average change is linked to Nayarit (+ 5.09%) with the highest volatility of 22.23%. However, this is due to the huge drop of − 37.65% in its efficiency from 1995–1997 to 1998–2000, which was recovered in the following two periods. On the other hand, the biggest negative change can be observed in case of Tamaulipas (− 5.08%) with SD of 10.96%, Morelos (− 4.91%, SD 10.33%), Campeche (− 4.33%, SD 12.09%) and Baja California (− 4.31%, SD 7.03%). Finally, Guerrero represents the state with almost zero average period-to-period change of -0.01% (SD 10.68%).

3.2 International tourists

As the results of the analysis show differences across the analyzed period, we can also assume that similar differences can be observed considering the tourists’ origin. In the case of international tourists, the average efficiency of all 32 states for the entire period was 27.42% with SD of 3.25% (Table 4). The average efficiency is − 37.6% below the efficiency in the overall model, but, on the other hand, the sector is more stable (SD lower by − 9.73%). Several similarities can be observed in the evolution of efficiency. We can also observe the drop in the efficiency during the year 1995 (27.64% in 1995 compared to 30.88% in 1994), which represented a relative change of − 10.49% (slightly less compared to the overall model). Then, a similar decade of efficiency decrease can also be observed. However, the hospitality sector recovered from this decrease at the beginning of 2006, a year earlier than in the overall model (Fig. 1). Further, the results show the significant drop in the efficiency in May 2009 (14.60%, − 12.76% compared to April 2009), which was recovered in the following months, as in the overall model, but in a slower pace. International tourists returned slower after the H1N1 pandemic than the national tourists, which would be linked to students' return to class in May 2009 and the cancelation of the suspension of non-essential activities [49]. The following decade showed a stable level of efficiency around 26.07% with SD of 1.83%.

Table 4 Hospitality efficiency of Mexican states, international tourists 1992–2018
Full size table

The best evaluated state in the case of international tourists is Quintana Roo (86.86%), Baja California Sur (59.07%), Chiapas (55.05%), Campeche (49.23%) and Baja California (46.87%), as these states include the most important touristic destinations in Mexico. On the other hand, the lowest efficiency is reported in Querétaro (5.96%), Hidalgo (8.49%), Durango (8.60%), Veracruz (9.70%) and Morelos (9.71%), which represent mainly in-land states. This phenomenon was also observed by Flegl et al. [34] in their analysis of touristic centers in Mexico. Considering the period-to-period changes (Table 8), the lowest volatility in the hospitality efficiency is observed in the case of the worst evaluated state Querétaro with the average period-to-period change of + 0.20% and SD 1.29%, followed by Guanajuato with − 1.42% (SD 3.03%), Morelos (− 0.04%, SD 3.63%) and Nuevo León (− 0.13%, SD 3.63%). All these states belong within the states with the lowest efficiency (Table 4). This result indicates that regardless of changes in the tourism industry, the hospitality sector in these states has not shown any progress toward international tourists. On the other hand, Coahuila (+ 6.95%, SD 20.09%), Nayarit (+ 5.45%, SD 35.34%) and Tlaxcala (+ 3.89%, 14.54%) represent states with the highest period-to-period growth across the evaluated period. However, it is also important to mention that the growth is not linear, but a higher volatility is observed. Finally, Chiapas (− 5.51%, SD 12.86%), Yucatán (− 4.54%, SD 9.28%) and Durango (− 3.99%, SD 11.05%) reported the largest average period-to-period decrease in hospitality efficiency among all states. The efficiency of the entire hospitality sector in the international model decreased by − 0.54%.

3.3 National tourists

The hospitality efficiency in the case of the national tourists shows completely different level compared to the international tourists. The average level of efficiency in the whole Mexico is 65.22% with SD of 5.58% (Table 5), which is higher by 38.22% compared to the international model. The lower variation can be linked to a lower dependence of national tourism to a few specific touristic locations. The evolution of the efficiency throughout the evaluated period follows the overall efficiency in the sector (Fig. 1). The most significant difference between the overall and national models is the efficiency level before the drop in 2005–2006. The model indicates that the average efficiency in the national model before this drop in 2004 was 67.96%. However, from January 2005 to December 2006, the average efficiency in the sector dropped to 54.04% (− 13.92%), with the lowest efficiency of 48.49% recorded between January and March 2006. Two reasons can explain this: (1) the loss of competitiveness reflected in the decrease in tourist stay [51], and (2) the lack of convergence of public tourism policies that used to involve 14 public agencies [52]. After this period, during the sexennial period 2006–2012, the sector recovered with a peak in March 2008 (77.89%). The hospitality sector recovered in 2007, a year later compared to the international tourists. The rest of the analyzed period follows the trend of the overall model.

Table 5 Hospitality efficiency of Mexican states, national model 1992–2018
Full size table

The best evaluated states are Nuevo León (77.19%), Coahuila (77.09%), Colima (76.76%), Guerrero (75.11%) and Querétaro (73.84%). First, we do not observe big differences between the best evaluated states as in the international model. As a result, efficiency is not concentrated in a few main states with the main tourist attractions, but the hospitality sector shows a better distribution across the whole country. Second, the best evaluated states are the in-land states, which is opposite to the international model. This leads to the observation that the worst evaluated states are Baja California (41.44%), Estado de México (52.84%), Baja California Sur (53.95%), Yucatán (54.37%) and Nayarit (56.69%). Most of these states were within the best evaluated in the international model, which highlights the different priorities of international and intranational tourists (Table 4).

Further, the results do not indicate any state with very high stability over the period-to-period changes. The best evaluated state is Guerrero, with average volatility (SD) of 4.96% and an average growth of + 1.59%. In the previous model, Querétaro presented a volatility of only 1.29%. The highest growth of efficiency can be observed in the case of Nayarit (+ 4.54%, SD 10.90%), Sonora (2.07%, SD 11.12%) and Colima (1.76%, SD 7.35%). Only five more states have a positive average period-to-period change, but these averages are close to zero. Most of the states thus report negative average period-to-period changes. The biggest drop in the efficiency has Campeche (− 5.29%, SD 11.82%), Morelos (− 4.91%, SD 8.20%) and Tamaulipas (− 4.43%, SD 10.55%). As a result, the whole hospitality sector decreased by average − 1.38%. All period-to-period efficiencies are presented in Table 9.

4 Discussion

This research focused on the state level to provide quantitative elements for decision-making in the tourism sector. It should be noted that decisions in the tourism activity and the allocation of resources to increase the efficiency depend largely on the Federation and statal entities. The results obtained from the DEA models indicate significant differences between the regions and national and international tourism. More precisely, the average efficiency linked to international tourism is 27.42% (− 37.6% below the efficiency in the overall model), compared to the average efficiency 65.22% in the case of national tourism. What is more, the most efficient states for international tourism are mainly coastal states (Quintana Roo, Baja California Sur, Chiapas, Campeche and Baja California), whereas the lowest efficiency is mainly observed in the in-land states (Querétaro, Hidalgo, Durango, Veracruz and Morelos). Identifying such differences is crucial considering two factors.

First, as very little attention has historically been paid to the tourism performance and efficiency analysis in Latin America [24], the results may indicate a pattern for the rest of the region. Such a conclusion can be supported by previous research in the industry. For example, Higuerey et al. [11] observed that the most efficient hotels in Ecuador during 2013–2017 are located in zones with tourist attractions and activities. Flegl et al. [34] observed a similar pattern of the national and international tourism efficiency levels in the case of 67 main touristic centers in Mexico. In both cases, the conclusions correspond to the presented analysis in the article. However, more analysis in the hospitality sector must be done to confirm this pattern. In this case, the analysis can be replicated in other countries as the model does not include any country-specific variables.

Second, the analysis provides a valuable information about the development of the hospitality industry over a long period (27 years or 324 months). Thus, the calculated efficiency covers important national and international events (crises), such as the global economic crisis in 2008 and the H1N1 pandemic in 2009, which provides worthy information about the industry behavior. For example, in the case of the H1N1 pandemic, the analysis detected: (1) such an event has a negative effect on the whole tourism sector with a similar effect on both national and international tourism. However, (2) international tourism takes more time to return to its pre-crisis level compared to national tourism (Fig. 1). The recovery time depends on the magnitude of the event and can vary from one month to 93 months [53] and also depends on the taken strategies to reduce the negative consequences of such a crisis [54].

Such information is worthy, for example, for the current Covid-19 post-pandemic recovery of the tourism industry, as a similar pattern can be expected. Mexico is one of the most vulnerable countries as the industry depends on international arrivals [15]. Therefore, Quintana Roo, Baja California Sur, Chiapas, Campeche and Baja California can serve as a benchmark for the rest of the industry due to the high concentration of the tourists’ arrivals in these states. Such observation is not unusual as some destinations can be highly preferred by tourists. For example, Oukil et al. [33] found the high level of efficiency concentrated mainly in Muscat, the capital of Oman, Corne [28] showed that Paris is a benchmark for the whole French hospitality sector, while Liu et al. [29] observed that highly efficient Chinese coastal cities are concentrated in two main regions.

In this case, it is important to refer to the elements that have traditionally influenced the level of tourism in Mexico. The study carried out by Madrid and Cerón [55] observed six dimensions that influence achieving the maximum tourist performance of a destination: the tourist vocation of the destination, strategic planning and sustainability, infrastructure, governance, visitor satisfaction and security. The evidence of the analysis points out that there is an efficiency gap as the number of rooms available in the hospitality sector is not properly linked to the arrival of national and international tourists in some destinations. This observation results in lower efficiency and possible problems in the post-pandemic recovery.

Therefore, to minimize the efficiency gap between the high and low efficient states, it is necessary to prepare strategies to attract in-land touristic areas for international tourists. The government plays the crucial role in addressing these challenging issues [56], as the cooperation between national, regional and local governments is essential in this process [57, 58]. For Mexico, this goes along with the Tourism Sector Program 2019–2024 of the Mexican government [59]. One of the five main projects for this period aims at regionalization of tourism, to strengthen tourism within the whole country to make it more equal.Footnote 1 The calculated weights in the DEA models revealed that international tourists highly prefer five-star hotels (77.70%), whereas the national tourist have no clear preference. The development in lower efficient states should align their hospitality structure to this evidence. Each state must decide who their primary target is, whether national or international tourists, and find the optimal mix of hotels.

Further, to close the efficiency gap between the states and minimize the period-to-period volatility in the efficiency observed in some states, it is also necessary to continue with the planned regional infrastructure projects such as the Mayan Train in the Southern states of Mexico (Tabasco, Chiapas, Campeche, Yucatán and Quintana Roo), the Transítsmico Train in the states of Oaxaca and Veracruz, which will compete with the Panama Canal; the Toluca—Mexico City commuter rail in the State of Mexico and Mexico City, among others, in order to achieve a more inclusive and sustainable tourism [59]. For example, in the case of the Mayan Train, Tabasco (international efficiency 12.23%, Table 4), Campeche (49.23%) and Yucatán (37.21%) can benefit from the connection with Chiapas (60.55%) and Quintana Roo (86.86%) to increase the flow of tourists in the region.

Such projects can help with the development of the communities around the main tourist centres. As Holzner [60] and Ranjan et al. [61] pointed out, it is recommendable to invest apart from tourism specific also into traditional infrastructure, which can be used by both the tourism sector and the manufacturing sector. So, the Mayan Train connection can help Chiapas and Quintana Roo to overcome the post-pandemic recovery in a faster pace and maintain its efficiency level.

Similarly, the efficiency gap between the high and low efficient states can also be minimized by focusing on the local cultural specifics of each state. Oukil et al. [33] argue that the effect of cultural attractions appears among the most influential factors in tourists’ attractions. The cultural factor includes traditional villages, world heritage, museums, archaeological and religious sites, crafts, among others. Therefore, SECTUR should target its marketing operations in less efficient states (including small touristic centers) to promote their cultural history and traditions. However, such specific marketing operations must be cautiously planned and discussed with local authorities, as, for example, local problems in small municipalities can put barriers in promoting local touristic activities [62]. Similarly, it is important to consider the insecurity situation in several states (touristic centers), as the effect of the marketing operations can be easily neutralized by negative publicity due to local criminality [42]. Destination’s image is one of the crucial factors for tourists’ destination loyalty [63], which is the key element in marketing strategies [64].

5 Conclusions

The article presents the results of the efficiency analysis in the Mexican hospitality sector. The DEA Window Analysis model was constructed for 32 Mexican states covering a period of 324 months from 1992 to 2018. The analysis concludes that: Significant differences exist between the states with respect to their overall efficiency, as well as regarding the volatility across the analyzed period. Moreover, significant differences in the national and international tourism are observed. In this case, the international tourism is mainly concentrated in only a few coastal states, resulting in very low overall hospitality efficiency. On the other hand, national tourism is not characterized by such concentration in the coastal states, as it is rather in-land oriented. This different characteristic results in higher overall hospitality efficiency.

Finally, the analysis indicates that international tourism recovers slower from crisis events, which should be considered in creation post-crisis strategies in the industry as Mexico mainly depends on international arrivals. Therefore, in future research, it is required to extend the analysis to other activities that are considered touristic according to the Tourism Satellite Account [7], such as food and beverage activities, transport or travel agencies, in order to have a better overview of the efficiency of the Mexican tourism sector and aim correctly the decision-making strategies in attracting more tourists’ arrivals.