A Multivariate Prediction Copula Model to Characterize the Expenditure Categories in Tourism

Abstract

The aim of this article is to introduce a multivariate statistical model that represents the expenditure of tourists disaggregated by categories. The model is applied to study the distribution of the expenditure of cruise passengers in Uruguay, using data of the 2016-2017 cruise season survey (collected by the Ministry of Tourism). Given the mixed distribution in each component of the main variable, the model is implemented in two stages and using copulas to obtain a conditional distribution of the different items of expenditure, characterizing the dependence between them. The empirical results show that the key variables that determine the average spending of cruise tourists are their residence and the port of arrival of the cruise. The parameters representing dependence of the copula show moderate association between the different categories of expenditure, in particular for cruisers disembarking in Montevideo, the capital of Uruguay. In addition, it can be noted that the expenditure pattern in each item shows time dependence. In general, the empirical results show that a cruiser that spends more on one item is likely to spend more (less) on a complementary (noncomplementary) items of expense.

Keywords

copula cruises tourist expenditure Uruguay

Introduction

The accelerated expansion of cruise ship tourism has led the literature on tourism economics becoming more and more interested in its analysis. The so-called “cruise industry” has consistently grown, and in 2017 it broke a new record with 25.8 million global ocean cruise passengers according to the Florida–Caribbean Cruise Association (2018). Since 2011, the increment was more than 25%.

In Uruguay, a small South American country, cruise industry has become a significant component of the tourism sector (Brida et al., 2015). Last season (2017-2018) almost 250,000 tourists arrived, which represents about 7% of the foreign tourists in Uruguay. A key factor on the dynamism of cruising is the strategic location of Montevideo and Punta del Este ports, between Buenos Aires and Río de Janeiro. It is important to have in mind that in the Uruguayan economy, tourism sector accounts for more than 7% of GDP (gross domestic product), according to the Tourism Satellite Account (Ministerio de Turismo, 2017). As has been stated in previous researches, tourism has turned out to be an engine of Uruguayan economic growth (Brida & Scuderi, 2013). To determine the economic impacts of cruise activity on a destination, it is important to understand the different items of cruise-related expenditure. These include vessel-related expenditure and supporting expenditure, which includes direct payments by ship owners to the destination authorities (see Dwyer & Forsyth, 1998; Seidl et al., 2007).

The present article considers cruise passenger expenditure in Uruguay as a key variable in the economic analysis of the costs and benefits of the cruise industry. In this article, the analysis focuses on passenger- and crew-related expenditures and assumes that cruise ship tourists make two types of decisions about their expenditures: extensive and intensive decisions. The extensive decision is the decision to spend money on one or more of the different items. The intensive decision has to do with how much of the budget is spent on each item. Among the different approaches to study the tourist expenditure (and in particular, cruise passenger expenditure), the classical regression techniques—including OLS, quantile, Tobit and two-step, logistic—are the most frequently applied because of their simplicity and generality (Brida & Scuderi, 2013). However, all these methodologies model the expenditures in one item as a function of different explanatory variables but not the interdependence between the different types of expenditure.

Recently, copula methods have been applied to model the temporal dependence of some variables referring to tourism through copulas (Tang et al., 2017; Zhu et al., 2017). But there are few studies that treat dependence simultaneously of different tourism variables by copulas; one example is Cai et al. (2019), which studies the impact of certain climatic variables on the tourist destination.

The present article contributes to the empirical literature that analyzes cruise passengers’ expenditure by applying an innovative methodology, a prediction copula model, to characterize the dependence between the expenditure categories and identify their determinants. This method, applied to the analysis of all categories of expenditure, introduces some challenges as will be explained later. The analysis uses data from the cruise survey conducted by Ministry of Tourism during the 2016-2017 cruise season (Ministerio de Turismo, 2017).

Modeling the Determinants of Tourist Expenditure

The impulse of tourism on global economic development has been largely examined in the empirical literature on the so-called Tourism-Led Growth Hypothesis (Brida et al., 2016; Castro-Nuño et al., 2013; Pablo-Romero & Molina, 2013). Many of the empirical studies and review papers have the objective of modeling and forecasting the aggregate tourism demand (Divisekera, 2013; Dogru et al., 2017; Dwyer et al., 2012; Lim et al., 2006; Lin & Song, 2015; Peng et al., 2014; Peng et al., 2015; Song & Li, 2008).

Nevertheless, microeconomic analysis on the determinants of tourist spending is less numerous. According to Mak et al. (1977) and Wang and Davidson (2010), except for two case studies, research on this topic starts in the 1990s, and most of the papers are case studies for specific tourism destinations (Brida & Scuderi, 2013; Mayer & Vogt, 2016; Xiao & Smith, 2006).

Previous studies on the determinants of individual tourist expenditure consider different spending measurements, type of models, and covariables (Brida & Scuderi, 2013; Mudarra-Fernández et al., 2019). In the models, many features of cruise passengers have been tested in the demand models: economic attributes (income, preferences, occupation), sociodemographic characteristics (age, gender, place of residence, education, marital status), trip-related features (accommodation type, transport, destination, travel information source), and psychographic variables (opinions, attitudes, motivations). Besides, there is considerable heterogeneity in the use of statistical and econometric methodologies.

A set of recent studies propose innovative and original methodologies to study the determinants of tourist spending. In Marrocu et al. (2015), it is analyzed that the determinants of tourist expenditure in Sardinia introduce the quantile regression approach; Olya and Mehran (2017) use complexity theory to explore a variety of configurations for the simulation of outbound tourism expenditures. Moreover, Melstrom (2017), as an alternative to standard OLS and Tobit estimators, presents an exponential model of tourist expenditures estimated, and Abbruzzo et al. (2014) present a scad-elastic net in the assessment of the determinants of individual tourist spending. Recently, Gómez-Déniz and Pérez-Rodríguez (2019) analyzed the distributional characteristics of aggregate tourist spending and length of stay in the Canary Islands. The method proposed enables to model both variables simultaneously by estimating two distribution models: for the length of stay of tourists and for the tourist expenses.

In this article, we propose a prediction copula model to analyze and represent the dependence between the different items of expenditure of cruise tourists in Uruguay and their determinants.

Empirical Literature on Cruise Tourism Economic Affects

The research on cruise tourism does not have a long tradition, and as pointed by Hung et al. (2019), it has been focused on studies for the Western hemisphere (Papathanassis & Beckmann, 2011; Sun et al., 2011). Moreover, Hung et al. (2019) and Sun et al. (2014) analyze cruise tourism industry in non-Western countries (e.g., China).

The empirical analysis of the dynamics of cruise passengers expenditure and their determinants has been approached from different points of view. From a macroeconomic point of view, the analysis focuses on the impact that this activity has on the country of destination (Brida & Zapata, 2009; Dwyer & Forsyth, 1998; MacNeill & Wozniak, 2018; Seidl et al., 2007). From a microeconomic point of view, the determinants of tourism and cruise spending have been explored from diverse empirical and econometric schemes (Bellani et al., 2017; Brida et al., 2010; Brida, Fasone, et al., 2014; Brida, Garrido, & Such Devesa, 2012; Brida, Pulina, et al., 2012; Brida & Scuderi, 2013; Brida, Scuderi, & Seijas, 2014; Henthorne, 2000; Morrison, 1996; Risso, 2012; Seidl et al., 2007), among others.

Cruise tourist expenditure in Uruguay was analyzed by Bellani et al. (2017) and Brida et al. (2010; Brida, Garrido, & Such Devesa, 2012) from a microeconomic point of view and by applying an econometric approach (Heckman selection models and Tobit and probit models). In general, these investigations show that nationality, the size of the group of travelling, the mobility, and their degree of satisfaction are the main variables to explain their spending pattern.

With an alternative methodological approach, Brida et al. (2017) analyzes the determinants of spending and behavior of tourists disembarking in Uruguay in the ports of Montevideo and Punta del Este. The authors introduce graph representation to synthesize and visualize the relationships between the set of variables that characterize tourists or groups of tourists and the determinants of their expenditure. They show that variables that best explain the behavior of visitors are those linked to the port of disembarkation and spending. Nevertheless, none of these works considers the possibility of dependence between the different items of expenditures.

Brida et al. (2018) applies nonlinear regression techniques with LASSO (Least Absolute Shrinkage and Selection Operator) penalty and nonparametric techniques to analyze the conditional distribution of global expenditure to a set of sociodemographic, contextual, and behavioral variables. Their findings reveal that the residence and the port of arrival of tourists are the main variables that explain the amount of spending.

The present article applies a similar methodology to analyze how the expenditure in one item could affect the expenditure in an alternative item (Bellani et al., 2017; Brida et al., 2010; Brida, Garrido, & Such Devesa, 2012). This model is implemented in two stages and using copulas to obtain a conditional distribution of the different items of expenditure, characterizing the dependence between them.

Methodology

This section introduces the dataset, the research questions, and the methodology used to perform the statistical analysis.

Study Sample

This study is based on a survey conducted by the Uruguayan Ministry of Tourism during the 2016-2017 cruise season (from November 2016 to April 2017). The sample design was carried out based on a two-stage stratified probabilistic sample. In the first stage, the sampling units are cruises, stratified according to the destination port (Montevideo or Punta del Este). The probability of selecting a cruise is proportional to the capacity of each ship. In the second stage, a fixed number of cruise groups are sampled, the number of people in each group being variable. A total of 1,644 interviews were conducted in Montevideo and 1,129 in Punta del Este. The total number of tourists who disembarked from the 150 cruises that arrived during the season was 260,704.

Table 1 shows a brief summary of the profile of the respondents. This table shows that Argentines and Brazilians have a greater presence within all the surveyed groups. More than 60% of the total respondents belonged to the age group of 25 to 55 years, and the majority were Brazilian. With regard to respondents of the age group between 55 and 70 years, it can be noted that they were mostly cruise passengers from North America and other countries. Note that around 43% of the respondents of Type 1 are from North America and Europe. Other categories show a high presence of Brazilians and Argentines. The table shows that the number of respondents in Type 2 is marginal. In addition, cruise passengers aged between 40 and 55 years were mostly in Type 4 category.

Table 1

Profile of Respondents According to Occupation, Age, and Nationality (Absolute Totals in the Sample).

Average age	Nationality					Occupation
Average age	Latin American (Rest)	Argentina	Brazil	North American	Others	Type 1	Type 2	Type 3	Type 4	Type 5
0-25	6	59	60	6	4	29	1	29	50	26
25-40	36	165	254	14	21	50	4	110	219	107
40-55	80	238	303	117	109	182	10	221	317	117
55-70	52	144	186	204	173	609	6	42	82	20

Note: Type 1 occupations are noted for retirees, students, housewives; Type 2 to professionals, technicians, teachers, journalists, athletes, and military; Type 3 to entrepreneurs, managers, merchants, and industrialists; Type 4 to skilled workers; and Type 5 to the administrative, cashiers, and sellers.

Source: Authors’ calculations.

The general profile of the respondent is tourist from Argentina and Brazil, of Type 1 and between 55 and 70 years old. The model introduces a set of nonnegative dependent variables concerning the expenditure on different items per capita. The databases used have the advantage of including a sample of a large cruise ship and report information on a large number of individuals and for various cruise seasons. However, the survey itself has limitations in terms of information that would be useful but not included, in particular, income data of respondents, education, tastes, of time outside the cruise, behaviors in the port, and other variables referring to the characteristic of the trip.

Questionnaire

The questionnaire was administered in three languages: Spanish, English, and Portuguese. The survey was administered on passengers older than 15 years of age when they return to the ship.

The data collection consisted of a series of questions about the sociodemographic characteristics of the cruise ship such such as age, gender, country of residence, and occupation. It also considered the number of times the respondent has visited the country. Another set of questions refers to the trip they take and includes variables such as the size and composition of the group in which they travel, the port where they disembark, individual spending (total and then broken down into transportation, food and beverages, purchases, and others). Finally, the questionnaire introduces questions about the pleasure the tourist felt about some aspects of the visit, such as landscape buildings, hygiene, cordiality, tranquility, beaches, and others.

They are mixed random variables, where distribution is partially discrete and partially continuous, with a positive probability at zero. Covariates can be grouped into four groups: travel variables (descending port, number of previous visits, sites visited, month), context variables (group type—mixed, only women, or only men—and number of group members—one, two, and more than two), satisfaction variables (variables of displeasure and pleasure), and socioeconomic variables (occupation, age, residence).

The questions about pleasant or unpleasant experiences are open. The possible answers are “yes” or “no,” that is, they are dichotomous variables. Interviewees can declare up to 3 aspects of pleasant or unpleasant experiences.

The satisfaction variables are dichotomous. Each respondent responds on the factors of main satisfaction or dissatisfaction. The categories (items) of expenses are the following: shopping, tours, food, transportation, and other expenses (including expenses in casinos). The total expense is satisfied with the sum of these expenses.

Procedure

As a way to capture the dependency relationships between the different expenses categories, the statistical procedure of this article is based on copula’s methodology. The definition of “Copula” or “Copula function” was introduced by Sklar (1959) originally in the context of probabilistic metric spaces. This concept is based on the idea that multivariate distributions can be separated as the univariate marginal distributions and their dependence structure (that can be represented by the copula). In statistics, the “copula” describes the function that “unites” one-dimensional distributions to shape a multivariate distribution and can be used to even several types of nonlinear dependence. They allow to represent all the possible cases of dependence, both for perfect dependence or independence.

Conceptually, Sklar’s theorem holds that for any multivariate distribution function, there is a copula that links the univariate distributions and that it contains all the information about the nature of the dependence between two random variables regardless of their marginal distributions.

From a practical perspective, one can decompose any estimation problem into two steps: the first modeling the marginal distributions and the second to model the parameters of the copula function. The classical approach of measuring dependency, through the function of linear correlation, has its disadvantages since it is a valid measure of dependence only within a restrictive class of distributions, for example, ellipticals, whereas the functions of copulas do not have this limitation. The bivariate case is developed for a better understanding.

The copula is a bivariate distribution function for which the marginal probability distribution of each variable is uniform (see Nelsen, 2007, for a formal definition). According to the Sklar theorem (see Sklar, 1959), let $H$ be the joint distribution function of a vector random variable $(X, Y)$ with $F$ and $G$ the marginal distributions, then there is a copula $C$ such that $\forall (x, y) \in {\bar{ℝ}}^{2}$ is satisfied that

H (x, y) = C (F (x), G (y)) .

If $F$ and $G$ are continuous, the copula $C$ is unique. In another case, the solution copulas coincide in $I m (F) \times I m (G)$ , where $I m$ denotes the image of the function. Conversely if $C$ is a copula with $F$ and $G$ two distribution functions then $H (x, y) = C (F (x), G (y))$ is a joint distribution function with marginals $F$ and $G$ . In general, the density of copula c describes in a more explicit way the dependence relations that it determines,

c (u_{1}, u_{2}) : = \frac{\partial^{2} C (u_{1}, u_{2})}{\partial u_{1} \partial u_{2}} .

In the present work, two parametric families of copulas are used, the Gaussian copula and the Student copula (see Embrechts et al., 2002; Lee, 1983, respectively). The Gaussian copula allows modeling with great flexibility for equal degree both positive and negative dependencies. The Student copula add a parameter, the degrees of freedom $ν$ . This dependency parameter controls the “heavy” of the queues.

An advantage that the Gaussian and Student copulas have with respect to the distribution functions from which they are derived is that from the copulas it is possible to use random variables that follow marginal distributions that are not of the same type.

As a disadvantage, as indicated by Sklar’s theorem, if the marginal variables are discrete, the uniqueness of the copula function could not be assured. Due to this factor, the use of copulas with discrete margins has a recent development (see Disegna et al., 2016; Genest et al., 2013; Zilko & Kurowicka, 2016).

As in Brida et al. (2018), the aim of this study is to model the cruise passengers’ expenditure in Uruguay. The present article aims to model not only the total expenditure as a function of a set of covariables but also the expenditure by item and their dependence. The total expenditure $Y$ can be decomposed into $d$ types of expenditures (in this article, travel, shopping, and food expenses are taken into account),

Y = Σ_{i = 0}^{d} Y_{i},

where $Y_{i}$ is a mixed random variable (nonnegative and with positive probability in $0$ ) that represents the expense of the ith item of the tourist passengers. The vector $(Y_{1}, \dots, Y_{d})$ has a dependency structure that can be modeled by copulas. Due to the mixed structure of the variables, it is necessary to perform a stepwise estimation separating the discrete part of the continuous (see Humphreys, 2013), but at the multivariate level. For this purpose the auxiliary variables $Z_{i} \sim B e r (p_{i})$ are defined, with $p_{i} = P (Y_{i} > 0)$ . Let $X = (X_{1}, \dots, X_{m}) \in ℝ^{m}$ be a random vector that represents the set of explanatory variables. The joint distribution of $Y$ , through the Total Probability Theorem, can be rewritten. For instance, when $d = 2$ the following representations are obtained for $y_{1}, y_{2} \geq 0$ (see Figure 1):

\begin{array}{l} P (Y_{1} \leq y_{1}, Y_{2} \leq y_{2} | X) \\ = P (Y_{2} \leq y_{2} | X, Z_{1} = 0, Z_{2} = 1) P (Z_{1} = 0, Z_{2} = 1 | X) \\ + P (Y_{1} \leq y_{1} | X, Z_{1} = 1, Z_{2} = 0) P (Z_{1} = 1, Z_{2} = 0 | X) \\ + P (Z_{1} = 0, Z_{2} = 0 | X) . \end{array}

The first factor of each term is modeled marginally when a single variable intervenes or by a copula with continuous marginals when there is a multivariate structure. The marginal distributions of the random variable $Y_{i}$ are estimated, with

Y_{i} | X, Z_{1} = 0, . . ., Z_{i} = 1, \dots, Z_{d} = 0,

$Y_{i}$ being the item in the ith expenditure through a model of the type $\log Y_{i} = α + β X^{t} + ε_{i}$ with $β = (β_{1}, \dots, β_{d})$ , $X^{t}$ being the transposed matrix of explanatory variables. It assumes log-normality ( $\log (Y_{i}) \sim N (μ (x), σ^{2} (x))$ ), independence and homoskedasticity of the residuals. To obtain a parsimonious model with high predictive power when many explanatory are available, a careful process of selection is required. For the selection of the variables in the regression model the LASSO penalty is used.

On the other hand, for the second factor of each term, it is necessary to model the joint distribution of $Z = (Z_{1}, \dots, Z_{d})$ conditioned to the vector of covariables $X$ . For this, the methodology proposed by Genest et al. (2013) is used and subsequently applied to tourism data by Disegna et al. (2016). It is assumed that the distribution of $Z$ is completely determined by the marginal conditional distributions $F_{1} (\cdot | X), \dots, F_{d} (\cdot | X)$ of $Z_{1}, \dots, Z_{d}$ , respectively, and a copula $C$ (independent of $X$ ). These marginal distributions are completely determined by $π_{j} (x) = P (Y_{j} = 1 | X = x)$ and modeled through a logistic model of regression,

π_{j} (x) = \frac{\exp {x^{t} β_{j}}}{1 + \exp {x^{t} β_{j}}},

where $β_{j}$ is a dimension vector $p \times 1$ to be estimated by maximum likelihood. A parametric copula family is subsequently adjusted (e.g., Gaussian copula or Student copula). This copula is uniquely determined in the image of the marginal distributions $[0, 1 - π_{1} (x)] \times \dots \times [0, 1 - π_{d} (x)]$ . Given the parametric copula $C_{θ}$ is estimated $θ$ maximizing the log-likelihood function.

Figure 1

Scheme of the Modeling Stages.

That is, in the first stage the distribution of the $Z$ vector conditioned to the covariables of the matrix $X$ is estimated. Since the components of $Z$ are not independent, a modeling through copula with discreet marginals is used (see Disegna et al., 2016). In the second stage, the distribution of the vector $Y$ also conditioned to $X$ and $Z$ is estimated. If only one component of $Y$ is considered positive ( $Z_{i} = 1$ and $Z_{j} = 0$ when j ≠ i), the $Y$ is modeled according with (Brida et al., 2018). If more than one component of $Y$ is positive (let $β = {i / Z_{i} > 0}$ then $# β > 0$ , where $#$ is the cardinal of the set), the vector ${Y_{j}}_{j \in β}$ is modeled through a standard approach of copula with continuous marginal distributions. For a better understanding, this procedure is depicted in the Figure 1.

Note that, the component covariables of matrix $X$ were not collapsed through a method of dimensionality reduction. That is, a new scale was not developed (for instance through factorial analysis or principal component); instead, a method of variable selection (LASSO method) was applied, as in Brida et al. (2018).

Results

This section presents the empirical results and is organized as follows. First, a descriptive statistics of the data is performed, and second, the estimation of the marginal distributions of each item of expenditure when positive is shown. Then, the joint distributions showing the association of the different expenditure items for this case is presented. Finally, the joint distribution of the Bernoulli variables which indicates positive values of expenditure for each item is modeled. Given that the variable “Transportation Expenditure” has a low impact on total expenditure, all the analysis is concentrated using the other three items of expenditure, namely, tours, shopping, and food, is studied.

Figure 2 shows the sample correlation between the items by items. The item “other expenses” is not included in Figure 2 given that it presents very small correlations with respect to the other items (all these correlations are below 0.01), but it is part of the total expenditure. The high correlation between total expenditure and shopping expenditure is appreciated.

Figure 2

Matrix of Correlations for Expenditures by Items.

Note that for each category the highest correlation is with total expenditure, reflecting the fact that each category is part of total expenditure.

It can be noted that the most significant correlations take place between the total expenditure and the shopping, and between the first and the power. The remaining bivariate correlations do not reach high levels, implying that expenditures in the different categories must be analyzed jointly.

In addition, the expenditure in shopping being the most correlated to the total expenditure, this item could be a good proxy of total expenditure. This fact can be a characteristic of the cruise business case in Uruguay. In this country, the supply for cruisers expenditure on tours and transportation is marginal. Probably the case of expenditure in food and beverage is a category that in general is not correlated with total expenditure.

Table 2 shows the different percentages of expenditure for items, conditioned to the variables Zone, Port, and Type of group (by gender of its members).

Table 2

Percentages of Expenditure for Items Conditioned to the Variables Zone, Port, and Type of Group.

Expenditures	Zone					Port		Type			Mean
Expenditures	Latin America (Rest)	Argentina	Brazil	North American	Others	Montevideo	Punta del Este	Mixed	Only women	Only men	Mean
Tour	16.3%	5.0%	7.3%	20.9%	13.5%	10.4%	13.6%	12.9%	7.8%	6.1%	5
Food	24.0%	10.8%	13.0%	25.5%	27.6%	16.6%	22.3%	19.7%	11.5%	21.0%	7
Transport	1.8%	2.8%	1.9%	1.4%	0.8%	1.8%	1.8%	1.9%	1.9%	1.9%	1
Shopping	56.0%	79.6%	76.5%	50.2%	54.0%	69.9%	58.5%	64.2%	76.7%	62.4%	25
Other	1.9%	1.8%	1.3%	2.0%	4.0%	1.3%	3.8%	1.3%	2.1%	9.3%	1
Percentage	100%	100%	100%	100%	100%	100%	100%	100%	100%	100%
Mean	57	26	39	45	37	46	26	36	40	58

Source: Authors’ calculations.

Different behaviors can be noted in certain categories of expenditure. For example, Argentinians and Brazilians have a higher percentage of shopping expenditure compared with other nationalities. With respect to the port, those tourists disembarking at Montevideo show a higher percentage of shopping than tourists who disembark at Punta del Este. In addition, mixed groups present a higher percentage of expenditures in tours than groups integrated only by women or men.

Groups of only women show a higher expenditure in shopping than other type of groups (mixed or men only). Finally, groups of only men present a higher expenditure in “other expenses.” The last column of Table 1 shows that the average per capita expenditure (in dollars) for each of the categories. Note that shopping is the higher item.

Figure 3 relates the satisfaction with the visit with the expenditure in the different items. In particular, the figure shows some percentages of likes with different attractions of the destination in each expenditure item. Note that those tourists who spent in transport dislike most of the attractions. This could indicate an unsatisfactory transportation service, indicating the necessity of improving transport services at both destinations. In general, the percentage of likes on local people amiability is almost the same in all items. This fact confirms one of the strengths of tourism in Uruguay: the kindness of its people.

Figure 3

Proportions of Tastes for People Who Have Positive Spending in Each Item.

Note also that the tourists who show positive expenditure in tours are those with higher likes in buildings and architecture; what is not surprising given that are those visitors that can have a general view of the destination. Given that the supply of tours (in particular in Punta del Este) is low, improving the number of tours within the city could be a good strategy for increasing the satisfaction of cruisers with the visit. Analogously, the small proportion of visitors spending in gourmet food are highly satisfied with the quality of Uruguayan cuisine. This is a good opportunity for local gourmet restaurants.

Supplemental Table 3 shows that the expenditure in tours has a positive association with U.S. and Canada tourists, not like hygiene and not like lack of shows, and a negative association with the January month. In particular, this implies that a cruiser from the United States or Canada has a higher proportion to spend in tours than cruisers of other nationalities. The rest of the coefficients must be interpreted in a similar way. Note that “Like food” is associated with a higher propensity to consume “food,” which is an intuitive result. Similarly, the satisfaction with the supply of stores is positively associated with a higher expenditure in shopping.

In Supplemental Table 4, note that the correlation parameters of the copula of tour–food and food–shopping are positive, while the copula tour–shopping is negative. This indicates that the expense in case tour–food and food–shopping are positively associated, that is, who else spends on tour has a positive probability of spending more on food (and vice versa). The same for food shopping, while who else spends on tour has a positive probability of spend less in shopping. These three results seem intuitive, because who take the tour hardly have time to spend shopping (and vice versa), while whoever spends more on shopping and tour is likely to spend more on food (and vice versa), which indicates a complementarity between these items of expenses.

Analogously it is possible to estimate a copula for the three expenses.

As a special case, the behavior of tourists who travel for the first time is analyzed, who have a high liking for the country (they like everything), middle aged, traveling as a couple to Montevideo and Punta del Este. The temporal effect is also analyzed, that is, the difference of the monthly predictions is observed. In the following, ^πk is written down to the estimated marginal probability of spending on the item $k$ ( $k = 1$ represents the food expense, $k = 2$ represents the expenditure on shopping, and $k = 3$ the expenditure on tour). The estimation of the matrix of parameters of the Student copula with 4 degrees of freedom is estimated by maximum likelihood. These results are compared with those obtained using a Gaussian copula. According to the AIC criterion, the Student copula had a better performance.

The higher shopping expenditures of Brazilians compared with Argentines observed can be explained by the higher income of Brazilian passengers compared with Argentines (see Supplemental Tables 5 and 6).

Regarding spending on shopping, it is observed that it increases in the month of March and is significant in tourists who disembark in Montevideo. These could be a consequence of the greater supply in Montevideo than Punta del Este, and that March is the month when low season starts in Punta del Este and higher prices in Punta del Este.

Note that the proportion of food expenditures are notoriously lower during November/December (see Supplemental Table 5). This could be a consequence of being in the low season of tourism, when most restaurants are closed (particularly in Punta del Este). In addition, there are only marginal differences between Argentines and Brazilians with respect to the consumption of food and beverage.

The estimates of the variances and covariances matrices are shown in Supplemental Table 7. There are slight variations in the correlations of the item expenses with reference to the Argentines and Brazilians. For example, spending on shopping has greater variability in Montevideo than in Punta del Este for Argentines, but is only slightly higher for Brazilians.

In this case, significant differences are observed between correlations estimated in each port, being of greater magnitude in Punta del Este. For example, for Argentines, between the Shopping and Food items, the correlation is 7%, while in Punta del Este it is 90%. In Brazilians, a similar behavior is observed.

Discussion, Conclusions, and Policy Implications

Several articles focus on modeling the total expenditure of tourists, but without taking into account the dependence of the different items that make up this expenditure (see Brida & Scuderi, 2013). This article focuses on studying and modeling not only the different items of expenditure but also the dependence between them. To this aim, a novel methodology that allows modeling the dependency structure through the use of copulas is introduced. To model nonindependent Bernoulli variables, a two-step methodology is applied to estimate the joint distribution of expenses.

The study shows that behavioral differences can be observed regarding the expenditure according to the nationality and the port of disembarkation. In particular, it shows that Brazilians tend to spend more in shopping than Argentinians and that cruisers disembarking in Montevideo tend to spend more in shopping and tours than those disembarking in Punta del Este. The results are consistent with those found in Bellani et al. (2017); nevertheless, they offer a different way to interpret the dependency. According to Brida et al. (2018), the study shows that variables of tourist residence and port of arrival are the main determinants of aggregate expenditure.

The parameters of dependence of copula show little association between the items of expenditure, but this dependence increases for the case of Punta del Este. This can indicate that cruise tourists do not consider the fact of consuming one item of expenditure in the decision of expending in another item. In other words, that the items of expenditure are independent in the decision of consuming. Additionally, a variable pattern is observed in the expenses in each item according to the month of the year. In particular, the study shows that there is an increase of expenditures in shopping during March, particularly in Montevideo, and the proportion of food expenditures are notoriously lower during November/December.

Some extensions to this work can include the analysis of cruise passengers that did not spend, to create packages of activities and programs that are appropriate to their profile. Another extension includes the analysis of cruise passengers who did not get off the cruise and/or compare the expenditure tourists of stay against cruises. Another line of analysis would be to compare the expenditure carried out by a cruise passenger in Uruguay with the expenditure in other destinations of cruisers traveling at the same cruise. The goodness of the model can be analyzed.

As a final remark, given that the dataset is available for a time period of several years, future research can include considering if the dependence is time-varying by applying dynamic copulas. The flexibility of the copula method allows to use various functions to describe different dependence structures, and this can be used to apply the methodology to other databases that can bring more interesting results. In particular, databases given information of expenditure by items for cruisers or other type of tourists for different countries could be a good start point. In general, copulas give an interesting method for measuring the dependence between different tourism variables including tourism demand and its determinants that do not need any assumption for the distribution of tourism demand variable (in other models tourism demand is assumed to follow certain a normal distribution). Finally, a potential application of the methodology is in forecasting in tourism.

Summary Conclusions

To summarize, the results presented in this study, if compared with those of previous studies analyzing cruisers expenditure, are contribution to this literature on giving some information about the different items of expenditure. For this specific case study, the results are not particularly surprising, but in our opinion this is because of data characteristics.

Since cruise tourism in Uruguay is a very dynamic sector, and due to its growing importance, the focus should be on more studies to improve managerial and policy decisions. The empirical results of this work can suggest some policies that allow improving the benefits of cruise tourism in Uruguay. Since the nationality and size of the group with which the cruiser is traveling explains the behavior of spending individual, entrepreneurs and public policymakers should promote Uruguay as a destination with appropriate facilities for group activities, such as entertainment and cultural events. In addition, the cruise tourism sector should promote Uruguay as a destination instead of a particular port in Uruguay. Since there are important groups of visitors that repeat the visit to the cities, the tour operators should improve on the development of new products that can induce passengers to spend more time at the destination.

Supplemental Material

sj-pdf-1-jht-10.1177_1096348020973266 – Supplemental material for A Multivariate Prediction Copula Model to Characterize the Expenditure Categories in Tourism

Supplemental material, sj-pdf-1-jht-10.1177_1096348020973266 for A Multivariate Prediction Copula Model to Characterize the Expenditure Categories in Tourism by Juan Gabriel Brida, Bibiana Lanzilotta, Leonardo Moreno and Florencia Santiñaque in Journal of Hospitality & Tourism Research

Supplemental Material

sj-tex-1-jht-10.1177_1096348020973266 – Supplemental material for A Multivariate Prediction Copula Model to Characterize the Expenditure Categories in Tourism

Supplemental material, sj-tex-1-jht-10.1177_1096348020973266 for A Multivariate Prediction Copula Model to Characterize the Expenditure Categories in Tourism by Juan Gabriel Brida, Bibiana Lanzilotta, Leonardo Moreno and Florencia Santiñaque in Journal of Hospitality & Tourism Research

Footnotes

ORCID iD

Juan Gabriel Brida

Supplemental Material

Supplemental material for this article is available online.

Juan Gabriel Brida (corresponding author; email: elbrida@gmail.com) is full professor of Economic Dynamics at the Faculty of Economics - Universidad de la República, Uruguay. Bibiana Lanzilotta (email: bibiana.lanzilotta@gmail.com) is professor in Economics at the Institute of Economics and in Economic Dynamics at the Department of Quantitative Methods at the Universidad de la República, Uruguay. Leonardo Moreno, PhD (email: mrleo@iesta.edu.uy), is associate professor of the Faculty of Economics and Administration (FCEyA), Universidad de la República, Uruguay. Florencia Santiñaque (email: florsanmes@gmail.com) is assistant professor of the Instituto de Estadística (IESTA), Department of Mathematical-Quantitative Methods of the Faculty of Economics and Administration, Universidad de la Rep�blica, Uruguay.

References

Abbruzzo

Brida

J. G.

Scuderi

(2014, October). Scad-elastic net and the estimation of individual tourism expenditure determinants. Decision Support Systems, 66, 52-60. https://doi.org/10.1016/j.dss.2014.06.003

Bellani

Brida

J. G.

Lanzilotta

(2017). El turismo de cruceros en uruguay: determinantes socioeconómicos y comportamentales del gasto en los puertos de desembarco [Cruise tourism in Uruguay: Socioeconomic and behavioral determinants of spending at landing ports.]. Revista de Economía del Rosario, 20(1), 26. https://doi.org/10.12804/revistas.urosario.edu.co/economia/a.6149

Brida

J. G.

Bukstein

Tealde

(2015). Exploring cruise ship passenger spending patterns in two Uruguayan ports of call. Current Issues in Tourism, 18(7), 684-700. https://doi.org/10.1080/13683500.2013.861391

Brida

J. G.

Cortes-Jimenez

Pulina

(2016). Has the tourism-led growth hypothesis been validated? A literature review. Current Issues in Tourism, 19(5), 394-430. https://doi.org/10.1080/13683500.2013.868414

Brida

J. G.

Fasone

Scuderi

Zapata-Aguirre

(2014, November). ClustOfVar and the segmentation of cruise passengers from mixed data: Some managerial implications. Knowledge-Based Systems, 70, 128-136. https://doi.org/10.1016/j.knosys.2014.06.016

Brida

J. G.

Garrido

Such Devesa

M. J.

(2012). Cruise passengers’ satisfaction: Cartagena de Indias. Benchmarking: An International Journal, 19(1), 52-69. https://doi.org/10.1108/14635771211218353

Brida

J. G.

Lanzilotta

Moreno

Santiñaque

(2018, December). A non-linear approximation to the distribution of total expenditure distribution of cruise tourists in Uruguay. Tourism Management, 69, 62-68. https://doi.org/10.1016/j.tourman.2018.05.006

Brida

J. G.

Osti

Barquet

(2010). Segmenting resident perceptions towards tourism—A cluster analysis with a multinomial logit model of a mountain community. International Journal of Tourism Research, 12(5), 591-602. https://doi.org/10.1002/jtr.778

Brida

J. G.

Pulina

Riaño

Zapata-Aguirre

(2012, January). Cruise passengers’ experience embarking in a Caribbean home port: The case study of Cartagena de Indias. Ocean & Coastal Management, 55, 135-145. https://doi.org/10.1016/j.ocecoaman.2011.10.003

10.

Brida

J. G.

Santiñaque

Lanzilotta

(2017). Modelos basados en grafos: una aplicación al estudio del gasto de cruceristas en Uruguay [Graph-based models: An application to the study of the expense of cruise passengers in Uruguay]. Revista de Métodos Cuantitativos para la Economía y la Empresa, 24(3), 270–291.

11.

Brida

J. G.

Scuderi

(2013, April). Determinants of tourist expenditure: A review of microeconometric models. Tourism Management Perspectives, 6, 28-40. https://doi.org/10.1016/j.tmp.2012.10.006

12.

Brida

J. G.

Scuderi

Seijas

M. N.

(2014). Segmenting cruise passengers visiting Uruguay: A factor–cluster analysis. International Journal of Tourism Research, 16(3), 209-222. https://doi.org/10.1002/jtr.1916

13.

Brida

J. G.

Zapata

(2009). Cruise tourism: Economic, socio-cultural and environmental impacts. International Journal of Leisure and Tourism Marketing, 1(3), 205-226. https://doi.org/10.1504/IJLTM.2010.029585

14.

Cai

Liu

(2019). Estimation of the spatial suitability of winter tourism destinations based on copula functions. International Journal of Environmental Research and Public Health, 16(2), 186. https://doi.org/10.3390/ijerph16020186

15.

Castro-Nuño

Molina-Toucedo

J. A.

Pablo-Romero

M. P.

(2013). Tourism and GDP: A meta-analysis of panel data studies. Journal of Travel Research, 52(6), 745-758. https://doi.org/10.1177/0047287513478500

16.

Disegna

Durante

Foscolo

(2016). A multivariate analysis of tourists’ spending behaviour. In Ferraro

et al. (Eds.), Soft Methods for Data Science. SMPS 2016. Advances in Intelligent Systems and Computing (Vol. 456). Springer. https://doi.org/10.1007/978-3-319-42972-4_24

17.

Divisekera

(2013). Empirical estimation of tourism demand models: A review. In Tisdell

C. A.

(Ed.), Handbook of tourism economics: Analysis, new applications and case studies (pp. 67-85). World Scientific. https://doi.org/10.1142/9789814327084_0003

18.

Dogru

Sirakaya-Turk

Crouch

G. I.

(2017, June). Remodeling international tourism demand: Old theory and new evidence. Tourism Management, 60, 47-55. https://doi.org/10.1016/j.tourman.2016.11.010

19.

Dwyer

Forsyth

(1998). Economic signiﬁcance of cruise tourism. Annals of Tourism Research, 25(2), 393-415. https://doi.org/10.1016/S0160-7383(97)00098-4

20.

Dwyer

Gill

Seetaram

(Eds.) (2012). Handbook of research methods in tourism: Quantitative and qualitative approaches. Edward Elgar. https://doi.org/10.4337/9781781001295

21.

Embrechts

McNeil

Straumann

(2002). Correlation and dependence in risk management: Properties and pitfalls. In Dempster

M. A. H.

(Ed.), Risk management: Value at risk and beyond (pp. 176-223). https://doi.org/10.1017/CBO9780511615337.008

22.

Florida–Caribbean Cruise Association. (2018). Cruise industry overview.

23.

Genest

Nikoloulopoulos

Rivest

Fortin

(2013). Predicting dependent binary outcomes through logistic regressions and meta-elliptical copulas. Brazilian Journal of Probability and Statistics, 27(3), 265–284. https://doi.org/10.1214/11-BJPS165

24.

Gómez-Déniz

Pérez-Rodríguez

J. V.

(2019, May). Modelling distribution of aggregate expenditure on tourism. Economic Modelling, 78, 293-308. https://doi.org/10.1016/j.econmod.2018.09.027

25.

Henthorne

T. L.

(2000). An analysis of expenditures by cruise ship passengers in Jamaica. Journal of Travel Research, 38(3), 246–250. https://doi.org/10.1177/004728750003800306

26.

Humphreys

B. R.

(2013). Dealing with zeros in economic data.

27.

Hung

Wang

Guillet

B. D.

Liu

(2019, January). An overview of cruise tourism research through comparison of cruise studies published in English and Chinese. International Journal of Hospitality Management, 77, 207-216. https://doi.org/10.1016/j.ijhm.2018.06.031

28.

Lee

L.-F.

(1983). Generalized econometric models with selectivity. Econometrica, 51(2), 507-512. https://doi.org/10.2307/1912003

29.

Lim

(2006). A survey of tourism demand modeling practice: Issues and implications. Dwyer

Forsyth

(Eds.), International handbook on the economics of tourism (pp. 45-72). Edward Elgar. https://doi.org/10.4337/9781847201638.00008

30.

Lin

V. S.

Song

(2015). A review of Delphi forecasting research in tourism. Current Issues in Tourism, 18(12), 1099-1131. https://doi.org/10.1080/13683500.2014.967187

31.

MacNeill

Wozniak

(2018, June). The economic, social, and environmental impacts of cruise tourism. Tourism Management, 66, 387-404. https://doi.org/10.1016/j.tourman.2017.11.002

32.

Mak

Moncur

Yonamine

(1977). Determinants of visitor expenditures and visitor lengths of stay: A cross-section analysis of US visitors to Hawaii. Journal of Travel Research, 15(3), 5-8. https://doi.org/10.1177/004728757701500302

33.

Marrocu

Paci

Zara

(2015, October). Micro-economic determinants of tourist expenditure: A quantile regression approach. Tourism Management, 50, 13-30. https://doi.org/10.1016/j.tourman.2015.01.006

34.

Mayer

Vogt

(2016). Economic effects of tourism and its inﬂuencing factors. Zeitschrift für Tourismuswissenschaft, 8(2), 169-198. https://doi.org/10.1515/tw-2016-0017

35.

Melstrom

R. T.

(2017). Estimating a model of sportﬁshing trip expenditures using a quasi-maximum likelihood approach. Tourism Economics, 23(2), 448-459. https://doi.org/10.5367/te.2015.0511

36.

Ministerio de Turismo. (2017). Anuario, estadísticas de turismo [Yearbook, tourism statistics]. Uruguay Natural, Ministerio de Turismo.

37.

Morrison

A. M.

(1996). Hospitality and travel marketing (Food and Hospitality Series). Delmar.

38.

Mudarra-Fernández

A. B.

Carrillo-Hidalgo

Pulido-Fernández

J. I.

(2019). Factors inﬂuencing tourist expenditure by tourism typologies: A systematic review. Anatolia, 30(1), 18-34. https://doi.org/10.1080/13032917.2018.1495086

39.

Nelsen

R. B.

(2007). An introduction to copulas. Springer Science & Business Media.

40.

Olya

H. G.

Mehran

(2017, June). Modelling tourism expenditure using complexity theory. Journal of Business Research, 75, 147-158. https://doi.org/10.1016/j.jbusres.2017.02.015

41.

Pablo-Romero

M. d. P.

Molina

J. A.

(2013, October). Tourism and economic growth: A review of empirical literature. Tourism Management Perspectives, 8, 28-41. https://doi.org/10.1016/j.tmp.2013.05.006

42.

Papathanassis

Beckmann

(2011). Assessing the “poverty of cruise theory” hypothesis. Annals of Tourism Research, 38(1), 153-174. https://doi.org/10.1016/j.annals.2010.07.015

43.

Peng

Song

Crouch

G. I.

(2014, December). A meta-analysis of international tourism demand forecasting and implications for practice. Tourism Management, 45, 181-193. https://doi.org/10.1016/j.tourman.2014.04.005

44.

Peng

Song

Crouch

G. I.

Witt

S. F.

(2015). A meta-analysis of international tourism demand elasticities. Journal of Travel Research, 54(5), 611-633. https://doi.org/10.1177/0047287514528283

45.

Risso

(2012). El gasto de los cruceristas en Uruguay 2008-2010 [The spending of cruise passengers in Uruguay 2008-2010]. Revista de Turismo y Patrimonio Cultural, 10(3), 393-406. https://doi.org/10.25145/j.pasos.2012.10.044

46.

Seidl

Guiliano

Pratt

(2007). Cruising for colones: Cruise tourism economics in Costa Rica. Tourism Economics, 13(1), 67-85. https://doi.org/10.5367/000000007779784533

47.

Sklar

(1959). Fonctions de répartition àn dimensions et leurs marges [N-dimensional distribution functions and their margins]. Publications de l’Institut Statistique de l’Université de Paris, 8, 229-231.

48.

Song

(2008). Tourism demand modelling and forecasting—A review of recent research. Tourism Management, 29(2), 203-220. https://doi.org/10.1016/j.tourman.2007.07.016

49.

Sun

Feng

Gauri

D. K.

(2014, September). The cruise industry in China: Efforts, progress and challenges. International Journal of Hospitality Management, 42, 71-84. https://doi.org/10.1016/j.ijhm.2014.05.009

50.

Sun

Jiao

Tian

(2011). Marketing research and revenue optimization for the cruise industry: A concise review. International Journal of Hospitality Management, 30(3), 746-755. https://doi.org/10.1016/j.ijhm.2010.11.007

51.

Tang

Ramos

Cang

Sriboonchitta

(2017). An empirical study of inbound tourism demand in China: A Copula-Garch approach. Journal of Travel & Tourism Marketing, 34(9), 1235-1246. https://doi.org/10.1080/10548408.2017.1330726

52.

Wang

Davidson

M. C.

(2010). A review of micro-analyses of tourist expenditure. Current Issues in Tourism, 13(6), 507-524. https://doi.org/10.1080/13683500903406359

53.

Xiao

Smith

S. L.

(2006). Case studies in tourism research: A state-of-the-art analysis. Tourism Management, 27(5), 738-749. https://doi.org/10.1016/j.tourman.2005.11.002

54.

Zhu

Lim

Xie

(2017). Analysis of tourism demand serial dependence structure for forecasting. Tourism Economics, 23(7), 1419-1436. https://doi.org/10.1177/1354816617693964

55.

Zilko

A. A.

Kurowicka

(2016, November). Copula in a multivariate mixed discrete–continuous model. Computational Statistics & Data Analysis, 103, 28-55. https://doi.org/10.1016/j.csda.2016.02.017

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.12 MB

0.01 MB