An empirical analysis of the in fl uence of macroeconomic determinants on World tourism demand

This paper considers three econometric models to determine the relationship between macroeconomic variables and tourism demand. Tourism demand is measured by the inbound visitor's population and also by on-the-ground expenditures. The database is an unbalanced panel of 218 countries over the period 1995e2012. There is evidence that an increase in the World's GDP per capita, a depreciation of the national currency, and a decline of relative domestic prices do help boost tourism demand. The World's GDP per capita is more important when explaining arrivals, but relative prices become more important when we use expenditures as the proxy for tourism demand. We cannot reject the hypothesis of a relative prices unitary elasticity of expenditures. Additionally, we have partitioned our data by income level and by Continent. Results are robust in the first partition, but less robust in the second, although the main conclusions still hold. Finally, we draw policy implications from our findings. © 2017 Elsevier Ltd. All rights reserved.


Introduction
As one of the important industries of the tertiary sector, the tourism industry has been developing rapidly in the last decades and contributing signi…cantly for economic growth, especially in tourism-intensive countries. And the demand for tourism continues to rise, since the transports sector has also been signi…cantly developing. Consumers have more means of transportation at their disposal, which are faster and cheaper, allowing them to choose over more destinations.
With this growing trend in the travel and tourism industries, …rms can take the chance to increase their income, by attracting more customers, if they e¤ectively forecast their demand and allocate resources in a reasonable way. The macroeconomic determinants of tourism demand, at the world level, are the focus of our work. We will focus on three macroeconomic determinants of tourism demand -the nominal exchange rate, relative prices and world income per capita.
We based our choices on the results draw from previous literature, which we present in the next section.
The choice to include the nominal exchange rate as a determinant is obvious, since a depreciation of a given currency relative to others, can increase the demand for tourism, hence domestic prices become relatively cheaper than import prices. A substantial amount of previous research focused on analyzing the relationship between this variable and tourism demand and found a somewhat robust and positive relationship between the two. We have also chosen to include relative prices, i.e., the ratio of domestic prices over foreign prices (in our case, the consumer price index of the USA is the proxy chosen for foreign prices) as an important explanatory variable. This variable measures the cost of living in the country in comparison with the USA, so it measures the purchasing power in the visited country. The expected sign of this variable is negative, since the higher the purchasing power of the visited country vis-à-vis the USA, the lower the probability of having many tourists. The literature has focused its attention mostly on the consumer price level (CPI) of the country and not much on the comparison between the CPI of the country and that of the rest of the World. We think a comparison of purchasing power is more important for consumer (tourist) decision than a mere introduction of the price level of the country itself. Additionally, in the tourism literature, income or economic growth has been playing an important role, either as a source or as consequence of tourism demand. Since in our work we are dealing with a panel of 218 countries, we consider the World GDP per capita, i.e., the average of World income, as one of our determinants for tourism demand, because it re ‡ects the global economic environment and wealth. The expected sign for this variable is positive, since we expect that an increase in average World income increases tourism demand.
We model tourism demand, our variable of interest, from two perspectives -value and quantity. From the value perspective we use on-the-ground expenditures. From the quantity perspective we use arrivals to the destination country as our variable. Besides distinguishing the determinants for tourism value and quantity, we use a log-log model speci…cation so that the measurement units of the macroeconomic variables will not matter in ranking the importance of these in explaining tourism demand. Based on panel methods for count (Poisson regression) and real-valued data, we conclude that the World's GDP per capita is more signi…cant when explaining arrivals and relative prices is more relevant when we use expenditures as the proxy for tourism demand.
Additionally, many of the studies so far applied the data for a speci…c country, region, or small group of countries, which may ignore the heterogeneity among destinations and also World-wide e¤ects. Hence, these studies lack universality, making it di¢ cult to apply their results and conclusions to a larger extent. To increase the scope of the literature, our work is going to analyze a panel of 218 economies, spread throughout all continents, thus covering the entire world. Our micro panel covers the period between 1995 and 2012 and it is found that the number of arrivals grew 1.2% per year whereas relative expenditures declined at a rate of about 2% per year.
Finally, we have also partitioned our data by income level and by continent to check whether the relative importance of each macroeconomic variable is indi¤erent to these two world aspects.
Results are robust in the …rst partition, but less robust in the second, although the main conclusions still hold. Quite interesting, the results suggest that the world income is relevant to high income countries and the relative prices to low and middle income countries and that the relative prices have a much lower impact in Europe, when compared to other continents. This work is structured as follows. In Section 2 we perform a literature review of the works closely related to our topic of study and that motivates the choice we make for the macroeconomic variables. Section 3 describes the empirical approach, i.e., data and methodology, Section 4 discusses the results and Section 5 analyses two extensions: income levels and continents. Finally, Section 6 concludes.
In this section we analyze the most relevant literature related with the macroeconomic determinants, which we use in our study as explanatory variables for tourism demand.
One of the macroeconomic variables that we use as a possible determinant of tourism demand is (World) income per capita. Previous literature has mainly used economic growth, i.e., the growth rate of GDP, as the variable of interest, and not the level or average income. This previous literature was mainly concerned about the direction of causality between tourism and growth. Additionally, most of the literature explores the in ‡uence of tourism on economic growth and few explore the reverse causality. Some of the studies presented below also use the nominal exchange rate and some proxy for prices as explanatory variables, but not in the same context as we do. Sequeira and Campos (2007) investigated the causality between international travelling and economic development. The authors used variables such as the degree of openness, the investment-output ratio, tourist arrivals per head of population, tourism receipts in % of exports, the black market premium, real GDP, secondary male enrolment, and the government consumption-output ratio, from 1980 to 1999, obtained from the Penn World Tables and the World Bank. Using panel data regression (with …xed or random e¤ects), they reached the following conclusions: the chosen tourism variables are not closely correlated with the economic boom, regardless of tourism-specialized countries or a wider range of other countries. In latter research, Sequeira and Nunes (2008a) introduced three additional variables: secondary years of schooling above 25 years, life expectancy, and international country risk guide, using the corrected Least Square Dummy Variables (LSDVC) or the …xed E¤ects (FE) approach and the Generalized Method of Moments (GMM) estimator. Results show that poor countries can pro…t from specializing in tourism, not only in tourist receipts, but also in consumption, which contributes to the development of the economy. On the other hand, small countries are bene…ting less from the specialization in the tourism industry.
According to Odhiambo (2011), with data for 1980-2008, using the Autoregressive Distributed Lag (ARDL) bounds testing approach, unlike most of the previous research, in Tanzania, tourism development leads to more economic growth in the short term, however, in the long run, growth-led tourism plays the important role. Meanwhile, statistical analysis also indicates that in the short run, there are bidirectional relationships between exchange rate and tourism development, and between exchange rate and economic growth. Research for Mediterranean countries shows similar results. Dritsakis (2012), using the method of cointegration analysis and data for real GDP per capita, real tourism receipts per capita and real e¤ective exchange rate, in the period 1980-2007, reveals that tourism development is closely related to GDP in seven Mediterranean countries: Greece, Turkey, Cyprus, Spain, France, Italy, and Tunisia. Furthermore, the author suggests that governments should assist the tourism industry to grow as much as possible. study the relationship between tourism revenues (exports) and tourism spending (imports). This paper illustrates the exchange rate e¤ects on US tourism trade balance, using the SVAR model, using data from 1973 to 2007, for the exchange rate, tourism exports and imports. There is no evidence of a J-curve behavior (The J-curve behavior means that in the short run, currency depreciation leads to a trade balance de…cit, instead of a surplus, like it is expected) of the US tourism trade balance with the US dollar depreciation, and a unit elastic e¤ect hypothesis of US tourism trade balance was raised. Export revenue is …nitely sensitive to the exchange rate only. In these two works the nominal exchange rate was the only variable of interest that was related to tourism.
The following work only considers the impact of prices on tourism. The impact of prices on the number of tourists is di¤erent depending on the departing countries. The demand variation in tourism demand of New Zealand was estimated by Schi¤ and Becken (2011). The log-log speci…cation was chosen, which gives a direct elasticity estimate. Elasticities for not only international visitor arrivals but on-the-ground expenditure per arrival are estimated for each segment.
Analyzing the annual data for arrivals and the consumption from 16 countries (1997-2007), the authors concluded that the traditional segments, like the USA and Australia, were less price sensitive, while the Asian markets are relatively more sensitive to prices. Since the price is one of the critical components for tourists'decision, the inspection of price competitiveness relatively to the exchange rate and internal in ‡ation should be consider.
The following works relate both exchange rates and prices with tourism, and in some of these works income or growth is also used as an explanatory variable. Lee et al. (1996)  countries raised the price competitiveness compared with Australia. And the countries which kept relatively lower in ‡ation rates greater enlarged their competitive advantage. In the case of Taiwan, while the e¤ects of relative prices and exchange rate volatility tend to be di¤erent, the exchange rate typically has the expected negative impact on tourist arrivals to Taiwan.
Whereas exchange rate volatility can have positive or negative e¤ects on tourist arrivals to authors use daily data on exchange rates and its volatility; arrivals of tourism to Taiwan from Japan, the USA, and the Rest of the World from 1 January 1990 to 31 December 2008. To capture the approximate long-memory properties in the tourist arrivals series, the heterogeneous autoregressive model is applied.
Saayman and Saayman (2013) studied the impact of exchange rate volatility on tourism in South Africa. It is assumed that the volatility of the South African Rand, the local currency (the ZAR) has an important impact on both visitors' spending and arrivals only from 2000 onwards, when the South African currency was permitted to free ‡oat. Volatility is modeled using a GARCH model, while the in ‡uence thereof on tourism is modeled using an autoregressive distributed lag model (ADL) and a bounds test approach. Using quarterly data for the period between 2003 and 2010 for average spending, tourism arrivals, real GDP, CPI, nominal exchange rate, and the main sources (countries) of intercontinental arrivals, respectively Australia, Germany, the UK, the USA, France, Brazil, and China. The authors found that increased currency volatility is associated with an increase in on-the-ground expenditure in most of the countries, respectively China, Germany, the USA, and Brazil, while Australian tourists tend to take smaller risks, spending less when volatility increases. In terms of arrivals, most of the countries showed risk aversion behavior, at the exception of China. Due to increased currency volatility, arrivals declined. Last but not least, in the long term, spending would be in ‡uenced more than arrivals. This work will also focus on the relationship between exchange rates, prices, income, and the number and volume of expenditures of inbound tourists, but taking into account a panel of 218 countries between 1995 and 2012, allowing to reach robust conclusions, about the relevance of these macroeconomic variables as determinants for tourism demand. These three macroeconomic variables will be jointly combined in econometric speci…cations.

Empirical Approach
In this section we describe the data, as well as the econometric methodology that we use in our estimations.

Data
The sample of the variables used in the models was taken from several data sources.
To measure tourism demand we use the countries'number of arrivals (inbound visitors) and the on-the-ground expenditure level, between 1995 and 2012, collected from the World Tourism Organization. There are four approaches to compile the tourism arrivals, namely the arrivals of non-resident tourists at national borders (TF), arrivals of non-resident visitors at national borders (VF), arrivals of non-resident tourists in hotels and similar establishments (THS) and arrivals of non-resident tourists in all types of accommodation establishments (TCE). In this paper, we have used the criteria TF and VF, which give us a more precise …gure for the number of tourists'arrivals, so countries for which these criteria are not observed were eliminated from our database. For the volume of expenditures, the data compiled by the WTO comes from the International Monetary Fund (IMF). We end up with a database of 218 countries.  Table 1 below.

[TABLE 1 ABOUT HERE]
As a whole, the dataset we use is a signi…cant micro unbalanced panel. It covers essentially all countries in the world, although for some of them a few variables of the models are not observed for the entire time period. Nevertheless, the total number of observations used to estimate the tourism demand models is clearly meaningful: it ranges from a total of 1606 to 2190 data points.

Methodology
We specify three di¤erent econometric models for tourism demand as a function of the macroeconomic variables exchange rate (XR), relative prices (RP), and the World GDP (WGDP).
We measure tourism demand either in terms of quantity (tourism arrivals) or in terms of price (tourism expenditures). For the latter, tourism demand is proxied by Real Expenditures per Arrivals or Real Expenditures per Domestic GDP. 1 That is, instead of the total level of expenditures, we model how much a tourist spends on average in a journey or the weight tourism's expenditures have relatively to the GDP of each economy. 1 Nominal expenditures were de ‡ated by the CPI.
Besides modeling the number of arrivals, we also have estimations for the number of arrivals relatively to the domestic populations (Arrivals/Pop). In this paper, we do not present the results for Arrivals/Pop because the estimated coe¢ cients associated to the covariates XR and RP have the wrong expected signs (negative for XR and positive for RP). Nevertheless, it is worth mentioning that Arrivals/Pop has experienced an estimated increase of 3% a year, something that con…rms that tourism has grown as an industry in the past years.
These three models were estimated taking log-log speci…cations, so that the coe¢ cients are interpreted as elasticities and thus independent of the measurement units chosen for the variables. Additionally, we can also rank the three explanatory macroeconomic variables to see which has the highest importance for tourism demand, in terms of elasticity.
The models contain two more features in order to best capture the main drivers of tourism demand. To account for time-e¤ects, we add a deterministic linear time trend to the models. We have 18 years of observations in the sample and, therefore, it is plausible that tourism demand in the world has been following a deterministic path over time. Since we have a micro-type of panel (the number of years, T = 18, is "small" relatively to the number of countries, n = 218) with unbalanced data, we do not consider dynamic panel models where the lagged dependent variable is included in the list of regressors of the model.
The second feature of the models is the existence of a component that captures all unobservable country-speci…c characteristics that also helps determining the tourism demand and which is assumed to be time-invariant such as the risk of the country (see Sequeira and Nunes, 2008b, for example). Note that, by assuming that these individual e¤ects are unobserved, we do not have problems of mismeasurement of variables and the associated bias that it introduces in the estimation stage. Following the standard approach in panel data regression models, we test for the existence of country-speci…c e¤ects and, in the case of its presence, we further test for random e¤ects against …xed e¤ects. where E ( j ) is the conditional expectation, x it includes XR it ; RP it ; W GDP it (all in logs) and the time trend, t; (the intercept is merged into i ); is the vector of coe¢ cients, and i = log ( i ) controls for individual country-e¤ects. Taking logs, The values of j ; j = 1; 2; 3 shall be interpreted as point elasticities whereas 4 is the percentage change of E (Arrivals it jx it ; i ) per year.
In model 2, real Expenditures per Arrivals is the dependent variable, measuring average real spending that a tourist does in a country. Now, the standard methods for panel data with real-valued dependent data are applied, The model is given by: Likewise, in the third model, Expenditures per GDP is the dependent variable, measuring the relevance of the receipts of tourism for the wealth of an economy for each year: The choice for having the macroeconomic variables XR and RP is also motivated by the data itself. We tested for 2 = 1 which would mean that 1 log XR + 2 log RP equals RXR log(RXR); the log of the real exchange rate. This restriction is clearly rejected by the data in all models, i.e., the e¤ects of XR and RP on tourism demand are not symmetric, one dominating over the other (more details in the next section

Results
In the next subsections we analyze the results drawn from our three econometric models for tourism demand, using di¤erent proxies. Results are in Table 2 below. The last subsection will make a comparison between the three estimations for tourism demand. [

Arrivals
The second and third columns of

Real Expenditures per Arrivals
The fourth and …fth columns of Table 2

Real Expenditures per GDP
The last column of Table 2 Table 3 below. [

Comparing Results
In this subsection we discuss the results of the previous three subsections. There is clearly a The estimates of the country-speci…c …xed e¤ects in determining the level of expenditures per GDP highlight the fact that the wealth and the size of the economy might in ‡uence the tourism demand functions (c.f. Table 3 above). Thus, in this section we analyze the results of two extensions: we have partitioned the data by income level and also by Continent. These two extensions allow us to check the robustness of the macroeconomic determinants obtained before to explain the tourism demand.

Analysis by Income Level
We …rst break our data by income level using an adapted measure of the partition by income We extend the models (1), (5), and (6) by also including the covariates interacted with dl it and dh it so that the tourism demand functions for these two groups are compared to the reference group of middle income. More speci…cally, what was de…ned before as x 0 it is now 3 The World Bank revises the classi…cation of the world's economies yearly, based on estimates of the current gross national income (GNI) per capita for the previous year. As of 1 July 2012, the World Bank income classi…cations by GNI per capita are as follows: - The estimation results are in Table 4. In the second column (LOW) we have b + b ; then (MID-DLE) b ; and in the last (HIGH) b + b : We removed the parameters that were not statistically signi…cant and that is the reason why some elasticities are the same across two or more groups.
[  Table   5, distinguishing the estimated elasticities in Europe (EUR), America (AME), Asia (ASIA) and Africa and Oceania (AF&OC), after eliminating the parameters that were not statistically signi…cant in the models.
[ On average, the number of tourists (arrivals) grew 1.2% per year whereas relative expenditures declined about 2% per year. In a global World, it becomes more common to travel abroad, including more people of lower social economic status, and …erce competition between …rms involved in the tourism industry may be pushing prices down.
Additionally, we have also partitioned our data by income level and by continent. Results are robust in the …rst partition, but less robust in the second, although the main conclusions still hold. The results seem to emphasize the relevance of the world income to high income countries and of the relative prices to low and middle income countries, in determining tourism demand. The panel we use in this paper is very complete at the micro level since it covers essentially all countries in the world. Nevertheless, the time series information is somehow limited because the number of years it covers is not that signi…cant and some variables of the models are not observed for some countries over the entire period. In the future, it shall be important to study the macroeconomic determinants in tourism demand once more time observations become available to practitioners. Tourism real expenditure in the country-U.S. dollars in millions, 2005=100 Arrivals (TF) or (VF) Arrivals of non-resident tourists (visitors) at national borders-in thousands WGDP World Gross Domestic Product, Current prices-U.S. dollars in billions