Determinants influencing inbound arrivals to Africa

Abstract

The goal of this research was to investigate the determinants that influence foreign tourism arrivals to the African continent, firstly as a collective and secondly in different regions, with the aim to foster a greater understanding of how African countries and regions can grow their tourism economies. Using static and dynamic panel estimators, two key findings were identified from this research: (1) tourism to the continent is influenced by income in developed countries, prices, telecommunication infrastructure and geographical factors as well as conservation efforts and (2) the regions in Africa do not all react the same to changes in these factors.

Keywords

Africa bias-corrected least square dummy variable generalized method of moments panel data tourism demand

Introduction

In recent years, the African tourism economy has seen a steady growth in the number of foreign tourist arrivals; however, this growth is still only prominent in a few African countries. This is in contrast to the potential that tourism development offers the continent in terms of its impact on employment, exports, tax income generation and the stimulation of infrastructure provision. Fayissa et al. (2008), therefore, correctly point out that the potential for economic development through tourism growth has not been harnessed effectively in Africa. This is echoed by Sahli and Nowak (2007) who found that tourism-led growth is especially relevant for developing countries, while Christie et al. (2013) argue that many African economies still fail to fully understand tourism’s potential as a driver of economic development and sustained economic growth. The need for improvement and development is particularly evident when considering regional cooperation and integration of economic policy, infrastructure and management.

This is unfortunate, as developing African countries could procure the greatest benefits from tourism, since Africa offers something that few other destinations can in terms of scenery, natural and cultural resources and adventure. However, these particular characteristics are difficult to quantify and are often not included in traditional tourism demand models, contributing to the ignorance of the drivers of tourism in Africa. For example, during the late 1990s, attention was focused on economic drivers or price differentials as the main determinants of arrivals to a destination. Within the African context, Naudé and Saayman (2005) – as one of the earliest studies on tourism to Africa – noted that price competitiveness was not the primary inhibitor of African tourism growth. Analysing the factors influencing tourism demand in Africa as a whole, and also in various regions of the continent, would therefore foster a greater understanding of how African countries and regions can grow their tourism economies. This article aims to make a contribution within this context and distinguishes it from previous research on Africa using more recent data over a longer period of time and by recognizing that the African continent is diverse.

The goal of this research is therefore to investigate the determinants that influence foreign (non-African) tourism arrivals to the African continent, firstly as a collective and secondly in different regions. Using balanced panel data for 25 African countries over 10 years and for five regions in Africa, the research allows for regional analysis and comparison, which provide greater insight into tourism to Africa and, for the first time, the regional differences in African tourism are explored. This may contribute to better policy formulation as well as destination management and regional development.

Literature review

In determining the demand for a certain destination or country, one should consider how demand will be measured. According to Wu et al. (2017), the most prevalent measures are tourist arrivals, followed by tourist expenditure and number of nights stayed in a particular destination. Tourist arrivals are also used to measure tourism demand in this research. Concerning research on tourism demand, the following aspects are reviewed: (1) the standard variables and the evolution to ‘new’ variables; (2) common countries and regions investigated as well as the differences; and (3) the methods and data applied.

Standard and ‘new’ variables

According to Lim (1997), tourism demand can be determined by a few standard tourism flow factors, which, in general, include income from the country of origin, price differentials, travel costs, exchange rates, competing destinations and marketing expenditures, to name a few. Income of origin is the most popular measure used in tourism demand studies (Eilat and Einav, 2004; Lim, 1997; Naudé and Saayman, 2005; Witt and Witt, 2995), and gross domestic product (GDP) is most commonly used as a proxy for income in tourism demand studies (Dogru, Sirakaya-Turk & Crouch, 2017). During the early 2000s, most tourism demand studies relied on variables such as income, GDP, tourism prices, transportation costs, exchange rates, substitute prices, relative prices, hotel prices and various dummies controlling for adverse events such as the 1974 oil crises, 1991 Gulf war and 11 September 2001 or 9/11 (see Bonham et al., 2009; De Mello and Nell, 2005; Dritsakis, 2004; Garín Muñoz and Montero Martín, 2007; Li et al., 2006; Lim, 2004 as prime examples of tourism demand studies that included similar variables). The main differences were in the data ranges, the methods used and the origin and destination countries.

The influence of seasonality, weather patterns and religious holidays has also contributed to a wealth of information. Here, quarterly or monthly data are often used to illustrate the differences of seasonality or weather-permitting holiday periods. Studies related to holidays/Easter (González Gómez et al., 2011), seasonality (Deng and Athanasopoulos, 2011; Zhang and Kulendran, 2017) and weather patterns/climate change (Li et al., 2017; Priego et al., 2015; Ridderstaat et al., 2014; Saayman and Saayman, 2008) have contributed to an enriched understanding of cycles and trends in travel flows.

While standard determinates are still largely relied upon, more recently, studies have included variables such as user-generated data sources from search engines or Google Trends (Li et al., 2017; Rivera, 2016; Yang et al., 2015) and understanding the influence of acts of terrorism on tourism (Buigut and Amendah, 2016; Hamadeh and Bassil, 2017; Karl et al., 2017; Samitas et al., 2018). There is also a growing trend to investigate regional tourism economies (Bo and Ningqiao, 2018; Guizzardi and Stacchini, 2015; López et al., 2016; Luo, 2017).

For the purpose of this investigation, the inclusion of traditional determinants such as income from origin countries (GDP of Organisation for Economic Cooperation and Development (OECD)) and price are used. With reference to African arrivals from origin countries, the majority are comprised of the 36 OECD countries. The OECD countries are distributed within Western (17) and Eastern and Central Europe (9) also known as the European Union (EU)-28, North America (3), Asia (2), Middle East (2) and Australasia (2). Africa’s traditional source markets include Western Europe and North American countries, with the exclusion of Mexico, while the remainder of OECD countries are classified as emerging countries for African arrivals. Within the EU-28, it is estimated that around 57% of Europeans have the propensity to travel outbound (International tourism trends, 2016), furthermore, Europe (France and United Kingdom) and the United States are the largest inbound source markets to Africa. Therefore, the OECD GDP is an appropriate proxy for income of origin country, since traditional and emerging source markets are accounted for.

In addition to these, health, safety, telecommunication infrastructure, level of development and qualitative and geographic dummies are included. Research related to the African continent and determinants of tourist arrivals are scare with the exception of Naudé and Saayman (2005), Kareem (2008) and Fourie and Santana-Gallego (2013). Some variables used by Naudé and Saayman (2005) include political stability, communication infrastructure and marketing, level of development, health risks, tourism infrastructure and geographic dummies such as sharing a border with South Africa. Kareem (2008) made use of both income and consumer price index (CPI) as well as political stability, crime rate, exchange rate appreciation and infrastructure (number of telephone lines). Fourie and Santana-Gallego (2013) also made use of standard determinants such as income (both origin and destination countries), land size, regional trade partners, common borders, language, religion and former colonial ties. Therefore, from these articles, a list of variables were created based on available data, as subsequently discussed in the following sections.

Origins and destinations

The origin countries in many tourism demand studies are usually more sophisticated countries or regions, such as the OECD countries or the EU-28 countries. However, within these regional/economic compositions, some outliers are present. Origin countries usually investigated include the United Kingdom, United States, France and Germany (De Mello and Nell, 2005; Dritsakis, 2004; Li et al., 2006; Shahrabi et al., 2013); however, gradually, Asian-Pacific countries, such as China, Japan, Korea and Australia have attracted more attention (Bonham et al., 2009; Deng and Athanasopoulos, 2011; Lim, 2004; Zhou Grundy and Turner, 2014). Much of the country choice would be dictated by data availability, author preference or specific international events.

Destination countries are in many cases also part of the OECD or EU-28 countries, usually with a warmer climate such as Greece, Spain, Portugal and Italy, or island economies, including Hawaii (state) and Aruba (Bonham et al., 2009; De Mello and Nell, 2005; Dritsakis, 2004; Garín Muñoz and Montero Martín, 2007; Li et al., 2006; Nelson et al., 2011; Ridderstaat et al., 2014). Recent research has also focused on countries with other attractions, including countries such as Macau and Hong Kong for shopping and casino visits (Kim et al., 2011; Lee, 2011).

Concerning Africa, tourism demand studies have mainly focused on individual countries such as South Africa (Burger et al., 2001; Durbarry et al., 2009; Gil Alana, 2011; Moyo and Ziramba, 2013; Saayman and Saayman, 2008; Seetanah et al., 2010), Zimbabwe (Chigora and Vutete, 2015a; Chigora and Vutete, 2015b; Mpofu, 2009; Muchapondwa and Pimhidzai, 2011), Egypt (Faragalla, 2018; Ibrahim, 2013; Wahba and Elanani, 2017), Tunisia (Choyakh, 2008; Choyakh, 2009; Gasmi and Sassi, 2015; Hathroubi, 2011; Ouerfelli, 2008; Ouerfelli, 2010; Sassi, 2014), Mauritius (Gooroochurn and Sinclair, 2005; Gopaul and Cheeneebash, 2015; Seetanah et al., 2015) and Nigeria (Awaritefe, 2007; Bankole and Babatunde, 2010; Eja et al., 2012). Exploring more African countries or the continent as a whole or regionally has attracted less attention, with the exception of Naudé and Saayman (2005), Kareem (2008) and Adeola et al. (2018).

Methods and data

Data and the methods/procedures employed in tourism demand studies are vast. Goh and Law (2011) state that, in the 1990s, econometric approaches to tourism demand modelling focused on the use of single equation regression models. However, as noted by Crouch (1994) and Lim (1997), limitations in these early studies include the existence of multicollinearity and serial correlations as well as possible spurious results due to the lack of unit root tests. This resulted in more sophisticated econometric approaches, such as systems of demand equations, vector autoregression, cointegration, error correction model and panel data approaches (Goh and Law, 2011). The more sophisticated approaches incorporate dynamics of tourism demand such as time lags, information asymmetry, repeat visits, word-of-mouth recommendations, supply rigidities and long-term adjustments (Song et al., 2012). According to Lim and McAleer (2001), the performance of forecasting models can be affected by the type of data (monthly, quarterly and annually) used.

Methodology

The use of panel data is especially popular because it addresses the problem with omitted or unobserved variable bias more effectively than cross-sectional data regressions alone (Kirchgässner et al., 2013). The use of cross-sectional data limits the causal relationship due to (1) unobserved variable bias, (2) endogeneity bias and (3) indeterminacy of the causal mechanism (Berrington et al., 2006). For this reason, the use of panel data is more appropriate as it provides more reliable estimations and greater leverage on causality due to the time component and not the sequencing of the variables as is the case with cross-sectional data alone.The complete panel used in this article consists of 25 African countries (see Table 1) and 10 years of data (i.e. N = 25 and T = 10). It is therefore not a large panel in the time dimension, which causes some constraints in using time-series techniques in the analysis. The panel, therefore, calls for methods that can accommodate a relatively large cross section (when the total panel is analysed) and a relatively small-time dimension. Apart from the regular fixed-effects estimation – also referred to as the least square dummy variable (LSDV) method – these types of panels are also modelled using generalized method of moments (GMM) and bias-corrected LSDV to estimate reliable results (Williams, 2015).

Table 1.

Countries used in the analysis.

Region/country	Code	Region/country	Code
Northern Africa		Southern Africa
Algeria	DZA	Lesotho	LSO
Egypt	EGY	Madagascar	MGD
Morocco	MAR	Malawi	MWI
Tunisia	TUN	Mozambique	MOZ
Eastern Africa		Namibia	NAM
Burundi	BDI	South Africa	ZAF
Ethiopia	ETH	Central Africa
Kenya	KEN	Angola	AGO
Mauritius	MUS	Central African Republic	CAF
Seychelles	SYC	Zambia	ZMB
Tanzania	TZA
Uganda	UGA
Western Africa
Benin	BEN
The Gambia	GMB
Niger	NER
Nigeria	NGA
Sierra Leone	SLE

Generalized method of moments

According to Blackburne and Frank (2007), small T-panel estimations rely on random or fixed effects or a combination of fixed-effect estimators and instrumental-variable (IV) estimators such as the GMM estimator. These methods require pooling individual groups and allowing only the intercepts to differ across the groups. As a result of bias problems in dynamic panels, two solutions are proposed: (i) introduce exogenous variables (not viable with small T-samples due to bias) and (ii) use IV methods such as the GMM estimator (widely used in panels with small T) (Asteriou and Hall, 2015). According to Garín-Muñoz (2006), unless the time period in a panel is large (close to infinite), random-effect estimators are biased and inconsistent. A solution is to use fixed effects or the dynamic GMM procedure of Arellano and Bond.

Arellano and Bond (1991) state that the GMM estimator optimally exploits all linear moment restrictions that follow from the assumption of no serial correlation in the errors in an equation that contains individual effects, lagged dependent variables and no strictly exogenous variables. One of the main advantages of GMM is that it allows estimation in systems where the number of unknowns is smaller than the number of moment conditions and to test whether the moment conditions hold (Sheppard, 2005).

Two forms of the GMM are available: (1) difference GMM and (2) system GMM. According to Roodman (2009), the GMM estimation starts by changing all regressors, usually by differencing, therefore referring to the differenced GMM, where the first-differenced instrument variables are uncorrelated with the fixed effects. The system GMM derives from the differenced GMM by introducing more instruments for efficiency and building a system of two equations – the original and the augmented one (Roodman, 2009).

In determining GMM estimators, a one-step or two-step approach can be followed, with the latter being the most popular (Drukker, 2010). The main advantage of the two-step approach is that the number of equations and parameters in the nonlinear GMM step does not grow as the number of perfectly measured regressors grows. However, the one-step GMM does not offer the estimating simplicity of the two step but is asymptotically more efficient (Erickson and Whited, 2002). Bond and Windmeijer (2002) state that one-step GMM estimators are asymptotically equivalent to optimal two-step GMM estimators.

According to Mileva (2007), certain problems might arise from using panels with small T and moderate N (such as in this article T = 10; N = 25) including (1) causality might run in both directions, where regressors might correlate with the error term; (2) time-invariant country fixed effects might correlate with explanatory variables; and (3) the presence of a lagged dependent variable gives rise to autocorrelation, while the Arellano–Bond (AB) estimator was designed for small T and larger N panels (Mileva, 2007). This estimator is, therefore, applicable to the larger African panel, but not to the regional panels, where the cross-section dimension (N) is considered to be too small. For these panels, an alternative method should be considered.

Bias-corrected LSDV

Observed bias in panels characterized by small N, small T or both (such as with this panel) necessitates comprehending the properties of different estimators on the estimated fixed effects (Buddelmeyer et al., 2008). One solution to counter the bias within short panels is to make use of a bias-correction technique. In recent years, the bias-corrected LSDV method has become popular, especially the techniques based on Kiviet (1995) and Judson and Owen (1999), which have been shown to outperform GMM and IV estimators (Bun and Carree, 2005).

According to Bruno (2004), the bias-corrected LSDV is more accurate than traditional GMM estimators, especially when small panels are considered, or at least where N is only moderately large. According to Kiviet (1995), LSDV estimators are also biased, although the standard deviation is often much smaller compared to the first-differenced Anderson–Hsiao (AH) IV estimator models and various GMM estimator models. Moreover, the bias-corrected LSDV is not outclassed and is still reasonably easy to calculate, whereas the IV and GMM estimations deliver poor results, especially in restricted conditions (Kiviet, 1995). Panels with small time dimensions (T) deliver the most appropriate results by means of the bias-corrected LSDV.

It is due to these properties that the smaller panels in this research will be estimated using the bias-corrected LSDV. This estimator, together with the AB GMM, will also be applied to the total African panel, for robustness.

Case study

Within the context of this research, inbound tourism refers to tourism activity attracted from other continents to Africa and does not include inter- or intra-African travel. According to Rogerson (2007), the systematic colonization of Africa by Europeans contributed to the establishment of Africa’s major tourism source markets. The United Nations Economic Commission for Africa (TII, 2011) states that the traditional source markets (Britain, Europe and the United States) have been the backbone of the tourism industry to Africa, and the pace of African growth, especially tourism receipts, is heavily dependent on the welfare of these regions (presumably based on income of origin countries and consumer prices in Africa).

Traditional source markets have been sustained very successfully over the years, yet are dwindling and ageing rapidly (WEF, 2013a, 2013b). Emerging markets such as China, Latin America and Russia have started to travel more extensively, also to Africa (Christie et al., 2013; UNWTO, 2010). For example, China held a 9.5% global market share in 2012, which constitutes a 40% increase from 2011. This makes China the largest and most affluent travel market, spending over US$107 billion (UNWTO, 2013a). It may be argued that much of the expansion into Africa is due to trade between the BRICS (Brazil, Russia, India, China and South Africa) countries. This might mean that Africa has positioned itself to benefit from increasing tourism from emerging markets. International tourism arrival statistics show that the majority (52%) of international tourists still travel to Europe, followed by Asia and the Pacific (23%), the Americas (16%), Africa (6%) and lastly the Middle East (5%) (UNWTO, 2013b). However, the demand for destinations in Asia and the Pacific (+6%) and Africa (+6%) is growing rapidly. The growing importance of Africa as a tourism destination is not matched by an increase in research though, with Xiao and Smith (2006) describing the state of research on Africa as ‘low’ compared to tourism research on North America, Europe, Asia and Australasia.

However, Africa differs from other continents due to various difficulties such as a lack of infrastructure, lack of available data and in many cases political and social unrest. The fact that the majority of least developed countries worldwide (34 of 48) are African countries is very worrisome and could be a potential contributor to low inbound travel flows to the continent, since crime is the social cost of poverty and inequality (Bourguignon and Des Hautes, 2001). According to Christie et al. (2013), Africa’s tourism is threatened by land availability, investor access to finance, taxes on tourism investments, low levels of tourism skills, lack of security and safety, high crime, public health, visa requirements, red tape and bureaucracy. Fundamental developments by governments regarding air transport infrastructure, appropriate accommodation and access to tour operators are therefore needed.

Bloom et al. (1998) state that Africa’s geography makes cultural interaction between Africa and the rest of the world, as well as within Africa, very difficult. This is further complicated by the vast area (30,221,532 km²) to be travelled as well as the goods that need to be transported to destinations in and across Africa (Turvill, 2013). Moreover, the constraints that certain African countries have with respect to transport services and infrastructure make it increasingly difficult to accommodate economic activity such as travel and trade (Rodrigue et al., 2013). Seddighi et al. (2001) emphasize that political instability influences the tourism industry because of its sensitive nature, the increasing competition between destinations and the narrow profit margins.

Model specification, data and variables

This research followed the economic approach to demand, where the quantity demanded depends mainly on income, price and other factors that influence tastes and preferences. The following equation is a representation of the simplified model and forms the basis of the estimated models in this article

D_{i j} = f (Y_{j}, P_{i}, Z_{i})

where $D_{i j}$ = demand for destination i from origin j, Y_j = income in the origin, P_i = price in the destination and Z_i = vector of variables that influence tastes and preferences. The arrivals data were obtained from the United Nations World Tourism Organization, and the independent variables or determinants were acquired from the World Bank Development Indicators as well as the International Monetary Fund’s International Financial Statistics. The analysis includes 25 African countries (i.e. N = 25), listed in Table 1, and 10 years of data (T = 10), 2001–2010. The availability of data influenced the choice of countries, and only countries for which complete data are available are considered, since this ensures a balanced panel.

The models are estimated using the complete panel of data for all 25 African countries (i.e. balanced panel of N = 25 and T = 10) as well as for the five regional areas of the continent (i.e. Northern, Eastern, Southern, Western and Central Africa). For Northern Africa, the panel, therefore, consists of four countries (N = 4) and 10 years’ data (T = 10); for Eastern Africa N = 7 and T = 10, for Western Africa N = 5 and T = 10, for Southern Africa N = 6 and T = 10 and for Central Africa N = 3 and T = 10. In this research, demand is measured using foreign arrivals. Although demand can also be measured as real spending by tourists, this option is not available, since tourism receipt data for African countries is even scarcer than arrivals data. According to Sharma (2004), some problems are associated with arrivals data, especially concerning the accuracy.

Using the basic demand specification above (equation (1)), it is evident that the demand for a destination depends on income at the origin, prices at the destination and other tastes and preferences. To measure preferences and tastes for Africa as a destination, this research relies mainly on previous research done on the African continent, specifically that of Naudé and Saayman (2005). Factors to consider, according to these authors, are political stability and personal safety, health risks, the level of development, available infrastructure and tourism marketing. For all the panels, the basic model specification is

L A R R I V E_{i t} = α_{} + β_{1} L A G D P_{t} + β_{2} L R E L C P I_{i t} + β_{3} L T B_{i t} + β_{4} L S A F E T Y_{i t} + β_{5} L I N T E R N E T_{i t} + β_{6} L U R B A N_{i t} + β_{7} L D E A T H_{i t} + ε_{i t}

with L indicating that natural logs of all variables were taken to ensure standardization and ease of interpretation. Systematically defining the variables employed in equation (2) provides clarity. LARRIVE is the dependent variable, which includes all arrivals to Africa of non-African countries, ensuring that intra-African arrivals are not included, while the independent variables include the following:

LAGDP is the advanced economy GDP, which is expressed by the GDP of the OECD. GDP only has variation in the time dimension and not in the cross-sectional dimension, which implies that time-fixed effects cannot be used together with the income proxy. Since tourism is often classified as a luxury good, a positive sign is expected as well as a coefficient larger than 1.

LRELCPI is the relative CPI, which is expressed as the ratio of the African country CPI divided by advanced economy (OECD) CPI. Again, OECD is used due to the importance of OECD countries as African arrival source markets. According to this calculation, an increase in destination CPI causes the relative CPI to increase and due to the law of demand, it is expected that arrivals will decline. A negative sign is therefore expected.

LTB accounts for health risks, the incidence of tuberculosis per 100,000 inhabitants of an African country. It therefore measures tourists’ exposure to dangerous diseases. Higher incidence is expected to result in less tourism, and therefore, a negative sign is expected.

LSAFETY measures four dimensions of governance (personal safety, rule of law, accountability and corruption and national security) on a scale from 1 to 100. A lower score indicates that a country has more safety concerns and less rule of law, and therefore, a negative sign is expected.

LINTERNET is the measure of Internet penetration and subsequently refers to telecommunication infrastructure, the number of Internet users per 1000 people. A positive sign is expected with more Internet users indicating higher levels of communication infrastructure (i.e. accessibility and marketability the destination).

LURBAN and LDEATH are two proxies for the level of development of the African country as expressed by urbanization rate and the death rate. A more urbanized economy and an economy with a lower death rate are both indicators of a more developed economy. Therefore, a positive sign is expected for the urbanization rate and a negative sign for the death rate.

This basic specification is also expanded by including a number of variables accounting for specific qualitative characteristics (QDUMMY) and to control for the geographical attributes (GDUMMY) of African countries.

L A R R I V E_{i t} = α_{} + β_{1} L A G D P_{t} + β_{2} L R E L C P I_{i t} + β_{3} L T B_{i t} + β_{4} L S A F E T Y_{i t} + β_{5} L I N T E R N E T_{i t} + β_{6} L U R B A N_{i t} + β_{7} L D E A T H_{i t} + β_{8} Q D U M M Y + β_{9} G D U M M Y + ε_{i t}

In terms of the specific qualitative characteristics (QDUMMY), Africa is often visited for its natural beauty, wildlife and unspoiled marine life. This is added to the basic model by including two variables that were not used previously within the African context, namely:

LPROTECT measures the percentage of the total territory, that is, a terrestrial protected area.

LMARINE measures the percentage of the total territory, that is, a terrestrial or marine protected area – both obtained from the World Bank Development Indicators (WBDI) database.

Since the variable has little to no variation over time, it is constant in the time dimension, but varies in the cross-sectional dimension of the panel. Both, therefore, serve as fixed-effect dummies and are included in the regressions separately. All analyses were done using Stata 14.

Furthermore, as in the research by Naudé and Saayman (2005), geography is expected to influence tourism arrivals in Africa and three dichotomous (dummy) variables were coded to account for this (GDUMMY):

Landlocked countries (LLC) is a landlocked dummy, since being landlocked not only indicates lower accessibility but also excludes access to beaches. In the tourism environment, a negative sign is therefore expected.

NADUM is a Northern African dummy that takes the value of 1 for countries that are located in the northern part of the continent. Geographical proximity to Europe may create an advantage in terms of tourism flows to these countries and a positive sign is expected.

SABORDER is a dummy variable that measure closeness to South Africa – one of the main destinations in Africa. The variable is coded to take the value of 1 if the country borders South Africa; and 0 otherwise. It is expected that the relationship will be positive, since being close to a major destination may improve accessibility.

Results

The results are discussed for Africa as a whole and are then shown for each region, namely, Northern Africa, Eastern Africa, Southern Africa, Western Africa and Central Africa.

The models were estimated using the standard pooled estimator as well as fixed and random effects, and the two tests were conducted to determine the appropriate estimator: the Breusch–Pagan Lagrange Multiplier (LM) test for random effects and the Hausman test to determine whether random or fixed effects are more appropriate. The null hypothesis of the Breusch–Pagan LM test is that the pooled ordinary least square (OLS) is consistent, while the null hypothesis of the Hausman test is that the random-effects model is consistent.

The Breusch–Pagan LM test only rejects the pooled OLS estimator in favour of random effects in one instance, namely, for the total African panel. For the remainder of the panels, the pooled OLS estimator is consistent. The Hausman test subsequently indicated that for all the African panels, random effects could be rejected in favour of fixed effects (see Online Supplemental Appendix, Table A1). Therefore, the pooled OLS estimator together with the LSDV and bias-corrected LSDV estimators were used for all the smaller panels, while the larger African panel was estimated using fixed effects, the LSDV and bias-corrected LSDV estimators as well as the AB estimator.

Results for Africa as a whole

For the total African panel (i.e. 25 countries and 10 years), the basic regression (equation (2)) was first estimated using the LSDV to account for fixed effects as well as the robust option to account for heteroscedasticity in the error term. The results are shown in column (1) of Table 2. Since the fixed-effects estimator cannot accommodate any other dummy variable, the remainder of the static models was estimated using an LSDV approach, with a dummy variable for each destination country. This estimator is equivalent to fixed effects because it allows for different constants for each cross section, that is, α_i and not a constant intercept (α) as in equation (1) (Asteriou and Hall, 2015). Using the LSDV, it is also possible to include other dummy variables. The five dummy variables accounting for various geographical and conservation (or tourism-specific) effects were included separately to test their significance in explaining tourism flows to Africa. The results are reported in columns (2) to (6) of Table 2.

Table 2.

Static estimation results.

	(1)	(2)	(3)	(4)	(5)	(6)
Dependent variable: LARRIVE	Robust LSDV	Robust LSDV + NADUM	Robust LSDV + SAB	Robust LSDV + LLC	Robust LSDV + MARINE	robust LSDV + PROTECT
_cons	−50.32** (20.60)	−51.28** (20.55)	−50.32** (20.60)	−51.28** (20.55)	−55.87** (20.81)	−54.84** (20.66)
LAGDP	2.319*** (0.719)	2.319*** (0.719)	2.319*** (0.719)	2.319*** (0.719)	2.291*** (0.727)	2.306*** (0.722)
LRELCPI	0.0974 (0.249)	0.0974 (0.249)	0.0974 (0.249)	0.0974 (0.249)	0.0956 (0.248)	0.0966 (0.248)
LTB	0.280 (0.363)	0.280 (0.363)	0.280 (0.363)	0.280 (0.363)	0.303 (0.359)	0.303 (0.359)
LSAFETY	0.417 (0.433)	0.417 (0.433)	0.417 (0.433)	0.417 (0.433)	0.420 (0.435)	0.418 (0.435)
LINTERNET	0.195* (0.092)	0.195* (0.097)	0.195* (0.097)	0.195* (0.097)	0.190* (0.100)	0.190* (0.101)
LURBAN	−0.645 (1.896)	−0.645 (1.896)	−0.645 (1.896)	−0.645 (1.896)	−0.560 (1.941)	−0.556 (1.947)
LDEATH	−0.540 (0.739)	−0.540 (0.736)	−0.540 (0.736)	−0.540 (0.736)	−0.555 (0.726)	−0.552 (0.728)
NADUM		2.896*** (0.107)
SABORDER			−2.882** (1.190)
LLC				−3.539 (2.752)
LMARINE					0.552 (0.700)
LPROTECT						0.580 (0.767)
Number of cross sections	25	25	25	25	25	25
Number of observations	249	249	249	249	249	249
R ²	0.980	0.980	0.980	0.980	0.980	0.980

Source: Compiled by author from STATA output.

Note: LSDV: least square dummy variable; NADUM: Northern African dummy; SABORDER: South African border. Standard errors are given in parentheses.

*p < 0.01; **p < 0.05; ***p < 0.001.

The static regression results in Table 2 indicate that arrivals to Africa as a whole are sensitive to income of origin countries (LAGDP) and the sign is positive, as expected, and greater than 1. This indicates that tourism to Africa is a luxury good and that an increase in income in the developed world will lead to a relatively larger increase in tourism to Africa. This finding is contradicting to results by Khadaroo and Seetanah (2008) stating that Africa is considered a non-luxury destination.

Telecommunication infrastructure (LINTERNET) is also positively and moderately significantly (at a 10% level of significance in some regressions) related to tourist arrivals in Africa. Again, this is expected and indicates that African countries that are more easily accessible and marketable receive more tourists.

Concerning the dummy variables, it is noteworthy that the geography dummy, landlockedness (LLC), is not significant, but both the NADUM and South African border (SABORDER) are significant. As expected, the NADUM is positive, indicating that countries closer to Europe (one of the main markets for tourists to Africa) attract more tourists. Contrary to previous research by Naudé and Saayman (2005), the SABORDER dummy is found to be negative. This implies that closeness to the largest southern African tourism destination is not creating positive spillover effects to neighbouring countries.

Quite disappointingly, none of the conservation (or tourism-specific) dummy variables, namely, protected marine areas (LMARINE) and protected terrestrial areas (LPROTECT), were significant. This could indicate that overall (because of Africa’s size), the ratio of protected marine and terrestrial areas is insufficient, and therefore, more conservation attempts have to be made to either increase the size of protected areas or establish additional protected areas.

While Table 2 shows the results of the static regression models, the dynamic model results are presented in Table 3. Since the panel has a relatively large cross-sectional dimension (N = 25), but a small time dimension (T = 10), bias in a dynamic fixed-effects model, which results from the correlation of the lagged dependent variable with the individual effects, is addressed using two alternative estimators, namely, the AB estimator and the bias-corrected LSDV estimator.

Table 3.

Dynamic panel data regression results (GMM and bias-corrected LSDV).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Dependent variable: LARRIVE	One-step GMM	One-step GMM	One-step GMM	One-step GMM	One-step GMM	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator
_cons	−9.081 (28.60)	−19.34 (28.59)	−32.13 (37.37)	−13.91 (31.92)	−14.84 (32.58)
LARRIVE (lag)	0.576*** (0.114)	0.576*** (0.110)	0.694*** (0.168)	0.455*** (0.147)	0.474*** (0.139)	0.599*** (0.0691)	0.691*** (0.0613)	0.625*** (0.0999)
LAGDP	0.542 (1.023)	0.839 (1.034)	1.385 (1.421)	0.620 (1.180)	0.621 (1.201)	1.806* (0.720)	1.841* (0.763)	1.785* (0.772)
LRELCPI	−0.877** (0.405)	−0.695** (0.311)	−0.752** (0.343)	−0.862* (0.472)	−1.105*** (0.380)	−0.202 (0.184)	−0.243 (0.193)	−0.225 (0.198)
LTB	0.111 (0.147)	0.0821 (0.115)	−0.0889 (0.196)	0.197 (0.208)	0.0586 (0.165)	0.195 (0.263)	0.203 (0.286)	0.209 (0.284)
LSAFETY	−0.123 (0.371)	0.155 (0.323)	−0.485 (0.535)	0.312 (0.463)	0.577 (0.448)	0.357 (0.268)	0.317 (0.285)	0.377 (0.268)
LINTERNET	0.271** (0.100)	0.296*** (0.0959)	0.177 (0.152)	0.358*** (0.102)	0.391*** (0.0911)	0.131** (0.0411)	0.110* (0.0438)	0.126* (0.0516)
LURBAN	0.0636 (0.167)	−0.316 (0.289)	−0.144 (0.245)	−0.0940 (0.243)	−0.0548 (0.254)	−1.658 (0.933)	−1.687 (1.047)	−1.628 (1.016)
LDEATH	−0.881* (0.504)	0.0764 (0.546)	−0.278 (0.610)	−0.368 (0.698)	−0.292 (0.667)	−0.578 (0.508)	−0.644 (0.553)	−0.590 (0.520)
LLC		−0.483 (0.419)
NADUM		0.972* (0.549)	0.883** (0.403)	1.461** (0.702)	1.198* (0.667)
SABORDER			1.163 (0.681)
LPROTECT				0.352** (0.131)
LMARINE					0.335** (0.123)
Arellano–Bond test of the first-order autocorrelation	−2.50**	−2.77***	−2.22**	−2.30**	−2.49**
Arellano–Bond test of the second-order autocorrelation	−1.67*	−1.73*	−1.66*	−1.65*	−1.62
Sargan test of over-identifying restrictions	269.24***	251.09***	216.81***	243.61***	227.36***
Hansen’s test of over-identifying restrictions	18.02	18.08	19.60	17.93	15.78
Number of groups	25	25	25	25	25	25	25	25
Number of observations	224	224	224	224	224	224	224	224

Source: Compiled by author from STATA output.

Note: GMM: generalized method of moments; LSDV: least square dummy variable; AB: Arellano–Bond; BB: Blundell–Bond; AH: Anderson–Hsiao; NADUM: Northern African dummy; SABORDER: South African border.

* Significant at the 10% level; **Significant at the 5% level; ***Significant at the 1% level.

The reason for estimating dynamic econometric models is that it allows the time path of the dependent variable to be included in the model. According to Gujarati and Porter (2009), one of the main reasons for including lagged variable in econometric models is psychological, or in other words, habit formation (inertia). The inclusion of the lagged dependent variable due to aspects such as habit formation, explains a large portion of the variance in the dependent variable and therefore often leads to insignificance of other variables (that were significant in the static model) due to a reduction in their contribution to explaining the variance in the dependent variable. For this reason, there might be differences between the static and dynamic model estimates. The dynamic model estimates shown are then also the short-run coefficients, with the long-run coefficients that can be calculated as ${\hat{β}}_{i}^{*} = \frac{{\hat{β}}_{i}}{1 - {\hat{β}}_{Y t - 1}}$ , where $β_{i}^{*}$ is the long-run parameter, ${\hat{β}}_{i}$ is the estimated short-run coefficient and ${\hat{β}}_{Y t - 1}$ is the coefficient of the lagged dependent variable.

The system GMM was estimated because the models include dummy variables (which render them zero in the difference GMM). Furthermore, the robust one-step estimator was used, as in previous studies on tourism (e.g. see Naudé and Saayman, 2005; Seetaram, 2010). The results were obtained using the xtabond2 procedure in STATA, which allows more control over the instrument matrix. As in the static models, the dummy variables were not introduced all at once, and therefore, the first five columns, (1) to (5), report the system GMM results. The last three columns, (6) to (8), report the bias-corrected LSDV results. For the bias-corrected LSDV estimator, three underlying models were estimated, namely, the AB GMM, the Blundell–Bond (BB) GMM and the AH IV approach. The bias was corrected using the bootstrapped method with a variance–covariance of 100 (similar to that employed by Seetaram, 2010).

Concerning the dynamic regression results, it is evident that the lagged dependent variable is positive and significant in all specifications regardless of whether GMM or the bias-corrected LSDV estimator is used, thereby confirming results by Fourie and Santana-Gallego (2013). This indicates persistence in tourism flows to Africa, with the possibility of repeat visits and habit formation. This is contrary to the results found by Naudé and Saayman (2005) and Khandaroo and Seetanah (2008), which found a negative lagged dependent variable for Africa.

Relative prices become significant in the dynamic regression. The negative relationship is as expected and shows that tourists are becoming increasingly sensitive to prices in Africa, although the elasticity is still less than 1. Wakimina et al (2018) also found that the Association of Southeast Asian Nations (ASEAN)-5 countries are price sensitivity destinations, even though Khandaroo and Seetanah (2008) point out that travelling to low-income destinations (Asia and Africa) is less price-sensitive. This is also evident when the price elasticities in this study are compared to that of the United States (Yazdi and Khanalizadeh, 2017).

In the bias-corrected LSDV estimator, income is also positive and significant, similar to the results found in the static regression. The coefficient is larger than unity, confirming that tourism to Africa is viewed as a luxury product. Again, the income variable was also supported by Wakimina et al (2018) in similar research, however, contradicted by Yazdi and Khanalizadeh (2017) for research based on the United States. The insignificance of GDP in the Arrelano–Bond estimations may be due to the inclusion of the lagged dependent variable, as explained above.

While safety and health concerns for Africa as a whole seem to play a lesser role in influencing tourist arrivals, telecommunication infrastructure (as measured by Internet users) is a quite robust determinant of tourism. In almost all specifications, LINTERNET is positive and significant confirming results by Ghalia (2016) stating that Internet is essential to enhance the tourism industry.

The development proxies, that is, the urbanization and death rates, are not significant. However, geographical closeness to Europe (as measured by the NADUM) is clearly an advantage, confirming the results of the static regressions. In the dynamic GMM models, the tourism-specific dummy variables marine and terrestrial protected areas are positive and significant, indicating that African countries that increase their conservation efforts gain more from tourism.

In terms of the diagnostics, it is evident that the null hypothesis of the first-order autocorrelation cannot be rejected, although this is expected in GMM models (Roodman, 2009). The test for the second-order autocorrelation indicates that at a 5% level of significance, the null hypothesis of no autocorrelation cannot be rejected in all instances. The Sargan test of over-identifying restrictions indicates that the null hypothesis can be rejected in all instances. However, the Hansen test indicates that the null hypothesis cannot be rejected and that the instruments are valid. The difference-in-Hansen test results confirm the exogeneity of instruments. According to Baum et al. (2003), the Hansen J-statistic is consistent in the presence of intra-cluster correlation, which causes the standard over-identifying restriction test to over-reject the null.

Static results for African regions

Since the diagnostic tests rejected the random effects estimator for the various African regions, the pooled OLS results are shown together with the robust LSDV estimator in order to control for country-specific effects as well as differences between countries due to conservation efforts (i.e. LMARINE and LPROTECT variables). These static estimates are first discussed and summarized in Tables 4 and 5, before the dynamic estimator results follow.

Table 4.

Static panel data regression results.

	Northern Africa				Eastern Africa					Central Africa
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)
Dependent variable: LARRIVE	Pooled OLS	Robust fixed effects	Robust LSDV + MARINE	Robust LSDV + PROTECT	Pooled OLS	Robust fixed effects	Robust LSDV + LLC	Robust LSDV + MARINE	Robust LSDV + PROTECT	Pooled OLS	Robust fixed effects	Robust LSDV + MARINE	Robust LSDV + PROTECT
_cons	−17.808 (53.769)	−70.55** (19.29)	−65.76** (20.66)	−65.46** (20.56)	−14.55 (69.16)	−18.77 (47.37)	−17.57 (50.64)	−15.54 (48.25)	−19.18 (49.59)	−85.24 (105.19)	118.8 (70.81)	209.0 (80.97)	211.8 (81.22)
LAGDP	−0.456 (1.918)	2.518** (0.610)	2.508** (0.658)	2.501** (0.657)	0.152 (2.424)	1.608 (1.462)	1.608 (1.538)	1.417 (1.447)	1.418 (1.447)	2.625 (3.756)	1.679 (2.412)	1.677 (2.531)	1.677 (2.531)
LRELCPI	0.204 (0.513)	0.295** (0.0717)	0.297** (0.0802)	0.298** (0.0821)	0.542 (0.502)	0.240 (0.266)	0.240 (0.280)	−0.298 (0.274)	0.298 (0.274)	−0.601* (0.324)	<0.001 (0.106)	<0.001 (0.111)	<0.001 (0.111)
LTB	−0.601*** (0.112)	−0.102 (0.293)	−0.110 (0.306)	−0.114 (0.303)	1.868*** (0.188)	−0.658 (0.943)	−0.658 (0.992)	−0.427 (1.016)	−0.428 (1.016)	1.819** (0.873)	17.37** (3.184)	17.35** (3.340)	17.35** (3.340)
LSAFETY	7.260*** (0.733)	−0.397 (0.839)	−0.401 (0.895)	−0.402 (0.896)	4.792*** (1.120)	0.911 (0.823)	0.911 (0.866)	1.107 (0.825)	1.106 (0.825)	0.055 (1.180)	0.229 (0.463)	0.229 (0.486)	0.229 (0.486)
LINTERNET	0.285*** (0.093)	0.0870** (0.0211)	0.0887** (0.0273)	0.0896* (0.0283)	0.039 (0.102)	0.0829 (0.215)	0.0829 (0.226)	0.0865 (0.218)	0.0864 (0.218)	0.492** (0.212)	0.324 (0.116)	0.324 (0.122)	0.324 (0.122)
LURBAN	0.779 (0.857)	7.437*** (0.798)	7.415*** (0.905)	7.403*** (0.914)	0.657 (0.488)	−3.591 (2.732)	−3.591 (2.875)	−3.088 (2.868)	−3.090 (2.869)	2.667*** (0.656)	−51.45** (6.610)	−51.41** (6.939)	−51.41** (6.939)
LDEATH	7.424*** (1.722)	−7.324*** (0.819)	−7.335*** (0.870)	−7.339*** (0.878)	−3.497*** (0.802)	−1.382*** (0.300)	−1.382*** (0.316)	−1.156** (0.451)	−1.156** (0.451)	0.523 (1.501)	−21.88** (4.791)	−21.86** (5.027)	−21.86** (5.027)
LLC							−2.633 (2.596)
LMARINE			−0.0408 (0.0881)					1.982 (1.539)				−30.65* (4.794)
LPROTECT				−0.0645 (0.0966)					1.975 (1.534)				−31.45** (4.919)
Number of cross sections	4	4	4	4	7	7	7	7	7	3	3	3	3
Number of observations	40	40	40	40	70	70	70	70	70	30	30	30	30
R-sq	0.974	0.951	0.998	0.998	0.752	0.557	0.968	0.969	0.969	0.950	0.845	0.976	0.976

Source: Compiled by author from STATA output.

Note: OLS: ordinary least square; LSDV: least square dummy variable. Standard errors are given in parentheses.

*p < 0.1; **p < 0.05; ***p < 0.01.

Table 5.

Static panel data regression results.

	Western Africa					Southern Africa
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
Dependent variable: LARRIVE	Pooled OLS	Robust fixed effects	Robust LSDV + LLC	Robust LSDV + MARINE	Robust LSDV + PROTECT	Pooled OLS	Robust fixed effects	Robust LSDV + SABORDER	Robust LSDV + LLC	Robust LSDV + MARINE	Robust LSDV + PROTECT
_cons	33.24 (96.61)	−65.02* (27.02)	−66.93* (28.28)	−79.60** (26.68)	−64.71* (28.60)	−107.69 (135.96)	−86.15** (29.06)	−83.19** (30.56)	−85.79** (30.62)	−84.39** (30.15)	−88.44** (30.22)
LAGDP	−1.032 (3.445)	2.513** (0.868)	2.513* (0.913)	2.180* (0.945)	2.513* (0.913)	4.051 (4.956)	3.385** (1.014)	3.385** (1.066)	3.385** (1.066)	3.504** (1.060)	3.383** (1.070)
LRELCPI	−1.556 (1.399)	1.014 (0.575)	1.014 (0.605)	0.906 (0.642)	1.014 (0.605)	−1.778 (2.208)	−0.399 (0.676)	−0.399 (0.711)	−0.399 (0.711)	−0.411 (0.664)	−0.405 (0.690)
LTB	−0.745** (0.328)	−1.043** (0.325)	−1.043** (0.341)	−1.029* (0.381)	−1.043** (0.341)	−0.987** (0.461)	2.254* (0.896)	2.254* (0.943)	2.254* (0.943)	2.282* (0.939)	2.280* (0.964)
LSAFETY	−0.170 (1.528)	0.346 (0.365)	0.346 (0.383)	0.464 (0.393)	0.346 (0.383)	1.337 (2.255)	2.237*** (0.454)	2.237*** (0.477)	2.237*** (0.477)	2.259*** (0.455)	2.261*** (0.444)
LINTERNET	1.009*** (0.225)	−0.0575 (0.0855)	−0.0575 (0.0898)	−0.0384 (0.108)	−0.0575 (0.0898)	0.525 (0.395)	0.441** (0.166)	0.441* (0.175)	0.441* (0.175)	0.450* (0.178)	0.449* (0.185)
LURBAN	−0.070 (0.791)	2.229 (1.157)	2.229 (1.217)	1.922* (0.715)	2.229 (1.217)	2.108*** (0.609)	−2.720* (1.320)	−2.720 (1.388)	−2.720 (1.388)	−2.804 (1.424)	−2.790 (1.481)
LDEATH	3.290** (1.273)	1.073* (0.480)	1.073 (0.505)	0.854 (0.473)	1.073 (0.505)	−0.490 (0.538)	−4.553* (1.965)	−4.553* (2.067)	−4.553* (2.067)	−4.635* (2.006)	−4.630* (2.095)
SABORDER								−1.840** (0.529)
LLC			1.934 (1.095)						−2.017** (0.509)
LMARINE				103.7 (51.48)						−1.686*** (0.364)
LPROTECT					−0.703* (0.276)						−1.147 (1.217)
Number of cross sections	5	5	5	5	5	6	6	6	6	6	6
Number of observations	50	50	50	50	50	60	60	60	60	60	60
R ²	0.706	0.774	0.990	0.991	0.990	0.588	0.727	0.972	0.972	0.972	0.972

Source: Compiled by author from STATA output.

Note: OLS: ordinary least square; LSDV: least square dummy variable; SABORDER: South African border. Standard errors are given in parentheses.

*p < 0.1; **p < 0.05; ***p < 0.01.

The static regression results for Northern Africa (Table 4, columns (1) to (4)) indicate that the region is sensitive to income changes of origin countries, and the relatively large positive coefficient confirms that this is a luxury destination. Contrary to expectations, Northern Africa is not a price-sensitive destination, with the sign of the relative price variable indicating that an increase in prices in this region does not cause a decline in tourism. This is an indication that the region is still viewed as a value-for-money destination. Like the results for Africa as a whole, telecommunication infrastructure (proxied by LINTERNET) is positive and significant. It is also interesting that the development proxies are all significant and have the correct sign, indicating that when tourists choose between Northern African destinations, more developed countries are preferred. Finally, the two conservation proxies are not significant, which shows that tourists to Northern Africa might be less concerned about conservation and natural beauty.

Eastern Africa is the largest region in the analysis and the panel comprises seven countries: Burundi, Ethiopia, Kenya, Mauritius, Seychelles, Tanzania and Uganda. The static regression results in Table 4, columns (5) to (9), indicate that most of the variables are not significant predictors of tourism arrivals to the region. This indicates that tourism to this region is less concerned with price, health and safety concerns and telecommunication infrastructure. Two exceptions are the urbanization and death rates. The urbanization rate shows a fairly large and negative coefficient, indicating that tourists travelling here may be especially interested in rural tourism. The death rate is also significant and negative, which might indicate that tourists choose between countries in this region on the basis of levels of development. Interestingly, and contrary to the significant urbanization variable, the protected terrestrial and marine conservation measure is not significant, indicating that the natural environment might not be the main drawcard for tourists to this region.

The Central African panel consists of only three countries: Angola, Central African Republic and Zambia. The results are indicated in Table 4, columns (10) to (13). The regression results for Central Africa (column (10)) indicate that tourism depends positively on telecommunication infrastructure (LINTERNET), with countries that are more advanced benefiting more from tourism. This result is, however, only significant for the pooled regression. Both the development variables are significant in the fixed effects models, with an increase in the death rate associated with decreased arrivals, as expected. However, the urbanization rate also has a negative sign, indicating that more urbanized economies receive less tourists. The health risk (LTB) also show results contrary to expectations, which may indicate that health concerns are not hampering tourism in the region. The tourism and conservation proxies (LMARINE and LPROTECT) are both negative and significant, indicating that more protected areas are not necessarily associated with increased tourism flows in this region.

Western Africa consists of five countries: Benin, Gambia, Niger, Nigeria and Sierra Leone. The static results (Table 5, columns (1) to (5)) show that income (LAGDP) from origin countries is positive and significant, indicating that increased income in developed countries will lead to increased tourism arrivals in this region. The prevalence of the tuberculosis (LTB) variable indicates a significant negative coefficient, implying that arrivals to this region are deterred by concerns over health issues, such as tuberculosis. Furthermore, there is again a positive and significant relationship between telecommunication infrastructure (as measured through LINTERNET). The protected area dummy (LPROTECT) has a negative coefficient, indicating that arrivals again do not bear a positive relation to conservation efforts in the region.

Southern Africa, the second-largest region in the analysis, consists of six countries: Lesotho, Madagascar, Malawi, Mozambique, Namibia and South Africa. The static results (Table 5, columns (6) to (11)) show, by the significant positive sign of the coefficient, that the Southern African region is considered a luxury destination, similar to Northern Africa. The LSDV estimator shows that the prevalence of tuberculosis and the issue of safety and the rule of law in this region are not deterrents to tourism arrivals, although the pooled estimator does not confirm these results. This is surprising as Southern Africa has one of the highest concentrations of Tuberculosis (TB) in Africa. As in most other regions, the telecommunication (LINTERNET) variable is positive and significant. In the pooled estimation, the positive urbanization rate indicates that tourism is positively associated with higher levels of development. This is echoed by the negative and significant coefficient for death when the LSDV estimator is used.

The geographic dummies offer an interesting insight into the region. The significant and negative SABORDER proxy indicates that arrivals will not necessarily increase to the countries directly adjacent to South Africa. The negative and significant geographic proxy (LLC) indicates that tourism to this region is dependent on access to coastlines, although the protected marine areas (LMARINE) coefficient shows that this does not necessarily imply that conservation efforts will lead to increased tourism in this region.

Dynamic results per region

The dynamic regression results for the various African regions are indicated in Table 6. The dynamic models were estimated using the bias-corrected LSDV estimator and the results are summarized in the table for all the regions. The estimator used was developed by Kiviet (1995) to correct for bias in small samples, and Bun and Kiviet (2001) showed that this estimator is superior to any other estimator when the panel has both small N and T dimensions.

Table 6.

Dynamic panel data regression results (bias-corrected LSDV estimator).

	Northern Africa			Eastern Africa			Southern Africa			Western Africa			Central Africa
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)
Dependent variable: LARRIVE	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator	LSDV AB-estimator	LSDV BB-estimator	LSDV AH-estimator
LARRIVE (lag)	0.270** (0.131)	0.432** (0.162)	0.281 (0.164)	0.504*** (0.141)	0.659*** (0.121)	0.381** (0.168)	0.280** (0.141)	0.432*** (0.140)	0.255* (0.140)	0.0890 (0.137)	0.298** (0.138)	0.0742 (0.159)	−0.0807 (0.202)	−0.0127 (0.218)	−0.145 (0.210)
LAGDP	1.798*** (0.604)	1.616* (0.898)	1.779*** (0.835)	0.537 (1.352)	0.287 (1.455)	0.285 (11.67)	3.613 (2.228)	3.122 (2.805)	3.701 (2.310)	2.630** (1.211)	2.297 (2.039)	2.655 (1.734)	3.544 (2.775)	3.697 (4.007)	3.493 (4.823)
LRELCPI	0.0821 (0.157)	−0.0202 (0.229)	0.0768 (0.223)	0.127 (0.340)	0.0726 (0.371)	0.155 (2.948)	−0.857 (1.027)	−0.872 (1.275)	−0.891 (1.098)	0.454 (0.680)	−0.00791 (1.107)	0.480 (1.011)	−0.230 (0.660)	−0.241 (0.942)	−0.232 (1.020)
LTB	−0.505 (0.361)	−0.436 (0.565)	−0.529 (0.514)	−0.536 (0.892)	−0.321 (0.978)	−0.978 (7.936)	2.477** (1.057)	2.326* (1.345)	2.532** (1.151)	−0.973 (0.637)	−0.728 (1.045)	−1.002 (0.906)	21.87*** (5.556)	21.31*** (7.078)	22.98** (9.003)
LSAFETY	0.131 (0.622)	0.276 (0.952)	0.134 (0.834)	0.662 (1.118)	0.941 (1.205)	0.987 (9.160)	1.770** (0.774)	1.535 (1.004)	1.819** (0.856)	0.623 (0.513)	0.780 (0.816)	0.630 (0.755)	0.361 (0.854)	0.453 (1.161)	0.408 (1.310)
LINTERNET	0.101** (0.0405)	0.0912 (0.0608)	0.0994*** (0.0555)	0.0990 (0.129)	0.118 (0.144)	0.128 (1.096)	0.418** (0.189)	0.325 (0.242)	0.423* (0.223)	−0.0445 (0.138)	−0.0225 (0.237)	−0.0441 (0.195)	0.308 (0.207)	0.307 (0.282)	0.327 (0.336)
LURBAN	2.258 (1.841)	0.108 (2.655)	2.215 (2.453)	−3.033 (2.738)	−2.439 (3.055)	−4.092 (23.62)	−3.233 (2.563)	−2.796 (3.239)	−3.224 (2.722)	3.777 (3.465)	4.708 (5.698)	3.979 (4.830)	−62.54*** (17.13)	−61.96*** (22.44)	−65.64*** (25.46)
LDEATH	−2.122 (2.098)	−0.374 (3.064)	−2.023 (2.851)	−0.489 (1.091)	−0.187 (1.204)	−0.597 (9.028)	−4.990* (2.579)	−4.926 (3.342)	−5.201* (2.772)	1.070 (1.264)	1.104 (2.092)	1.166 (1.806)	−27.26*** (7.239)	−26.58*** (9.277)	−28.65** (13.25)
Number of cross sections	4	4	4	7	7	7	6	6	6	5	5	5	3	3	3
Number of observations	36	36	36	63	63	63	54	54	54	45	45	45	27	27	27

Source: Compiled by author from STATA output.

Note: LSDV: least square dummy variable; AB: Arellano–Bond; BB: Blundell–Bond; AH: Anderson–Hsiao.

* Significance at the 10% level; **Significance at the 5% level; ***Significant at the 1% level.

In the dynamic model, a lagged dependent variable is included, and it is evident that, as in the total African panel, there is habit formation and persistence in tourism to Northern Africa (Table 6, columns (1) to (3)) which confirms results by Fourie and Santana-Gallego (2013). The dynamic specification also confirms that Northern Africa is an income-sensitive destination and that changes in origin country incomes would have a large effect on tourism to this region. The influence of telecommunication infrastructure is also confirmed in the dynamic regression. However, the estimates do not confirm the idea that prices in Northern Africa can increase and still have a positive influence on tourism. Instead, it shows no effect of price changes on tourism. Furthermore, once the effect of repeat visitation (or persistence) is controlled for, the differences in the levels of development between Northern African countries become insignificant for tourism.

The dynamic regression results for Eastern Africa are indicated in Table 6, columns (4) to (6). As in the case of Africa as a whole and Northern Africa, the habit formation and persistence of tourism to Eastern Africa are evident through the significance of the lagged dependent variable. The region is, however, not income sensitive; this could be due to the large variety of expensive lodges and resorts in the region, indicating that very affluent people travel to this region. An attraction, which was not measured, could be cultural phenomena such as the Masai people of southern Kenya and northern Tanzania.

The dynamic regression results for Southern Africa are indicated in Table 6, columns (7) to (9). The lagged dependent variable again shows habit formation and persistence of tourism to the region. Southern Africa is not income or price sensitive. The positive effect of telecommunication infrastructure (LINTERNET) is confirmed in the dynamic regression. It is confirmed that higher levels of development (as measured by lower death rates) are associated with an increase in arrivals to Southern Africa. Like the static LSDV estimator, the dynamic regression shows that the prevalence of tuberculosis (LTB) does not have a negative influence on tourism arrivals.

The dynamic regression results for Western Africa are indicated in Table 6, columns (10) to (13). The results indicate that the dependent variable (LARRIVE) is positive and significant, but only when the BB estimator is the underlying technique, indicating limited evidence of habit formation and continuous travel to Western Africa. The income proxy (LAGDP) is the only other significant variable, which confirms that tourism to this region is dependent on income growth in developed countries.

The dynamic regression results (Table 6, columns (13) to (15)) confirm that the prevalence of tuberculosis does not deter arrivals to Central Africa. As in the static results, urbanization is negative and significant, implying that countries with lower levels of urban development attract more tourists to this region. However, tourism is adversely affected by increase in overall death rates.

Discussion

The first finding that follows from this research is that five determinants that influence travel to Africa as a whole were identified, namely, income, price, infrastructure, geographical and conservation factors. This research confirms that travel to Africa remains a luxury good, with an income elasticity greater than unity. A previous study on the determinants of tourism to Africa by Naudé and Saayman (2005) found price to be an insignificant predictor of arrivals. However, in this updated study with more African countries and a more recent time period, it has been found that tourism to Africa is becoming more price-sensitive. As in Naudé and Saayman’s (2005) findings, infrastructure remains a significant determinant of tourism to Africa. In terms of the significance of the different regions in Africa, the current research shows that the Northern African region and not the Southern African region, as found by Naudé and Saayman (2005), creates positive spillovers for tourism. In addition, conservation efforts have a positive effect on tourism to the continent.

Since this research shows that arrivals to Africa are becoming price sensitive, African countries have the opportunity to increase tourism by focusing on two distinct strategies. These strategies should be aimed at attracting: (1) tourists who are less price-sensitive and are willing to pay for exceptional five-star service at luxury lodges, and (2) attract tourists who are price sensitive by offering services at reduced rates, for example, backpacking or rustic resorts. A number of African countries follow mainly the first strategy and it can be worthwhile to develop the alternative lower cost strategies.

Telecommunication infrastructure (as proxied by Internet connectivity) remains a key determinant of African tourism. The results suggest quite robustly that the rapid expansion of communication infrastructure would benefit tourism in Africa. In future, it could be even more beneficial if communication operators in countries within the same region were connected to offer seamless integration to people travelling to multiple countries within the same region. This will also contribute to regional economic cooperation and integration.

Furthermore, it is confirmed that proximity to Europe creates benefits for tourism in Northern African countries. Since distance is reduced, travel costs are decreased, making Northern African countries ideal destinations for price-sensitive tourists. Moreover, Northern African countries have the ability to offer mass tourism products such as sun, sea and sand (beach) tourism. More affordable resort-style infrastructure could create even more benefits for the region.

Africa is home to many rare and indigenous species, which represents a key draw card for tourism. To ensure sustainable tourism, African countries are in an advantaged position where active conservation efforts may lead to increased tourism. However, it is imperative that developments and investments be made in a sustainable manner that takes into account both biodiversity and societal impact.

The second key finding is that the five regions in Africa are clearly very different and the variables that influence tourism to these regions vary significantly between them. This is largely explained by the fact that each region offers unique experiences and therefore faces unique challenges to attract more arrivals. An encouraging result is that lagged arrivals are positive and significant for almost all African regions, indicating that there is persistence in tourism to all regions of the continent.

A striking result in all the regions, except Western Africa, is that the death rate is almost always significant and have a negative influence on tourism arrivals. Since the death rate is a proxy for the level of development of the country, it clearly shows that more developed economies (i.e. lower death rate) attract more tourists. Not only does low death rates indicate more development, but it also shows better healthcare and less epidemics in a country, which is also important for tourists. Effective and efficient allocation of funds for health services in Africa would therefore also benefit the tourism sectors in these economies.

Northern African arrivals are susceptible to origin-country income and telecommunication infrastructure. Since these determinants are manageable, Northern African countries should look at more inclusive holiday destinations that can suit every budget. Merging with telecommunications operators in Europe might also be beneficial, because tourists will have access when visiting.

Different levels of development between Eastern African countries (as measured by the death rate) may account for differences in tourism between countries in this booming tourist region. The countries included in the region are quite diverse, ranging from islands to recently war-torn countries. This diversity might explain why the average coefficients estimated by panel models are not significant.

Infrastructure is a robust determinant of tourism to Southern African countries, which implies that the region can benefit from improving telecommunication infrastructure. Regional investment in the Southern African Development Community must therefore be promoted. Tourists to Southern Africa are informed and not deterred by the health, development and safety concerns that plague the region. Improvements in the region in terms of development, rule of law and safety can, however, be more beneficial to tourism.

As in the Northern African region, tourism to Western African countries is income elastic, indicating that this region is considered a luxury destination. With the exclusion of Niger, all the Western African countries have access to beaches and could offer cost-effective resort-style tourism. These countries are also much closer to Europe, resulting in reduced travel costs. However, the region still has a long way to go in establishing itself as a tourism destination.

As in Southern Africa, tourists to Central African countries seem to be informed travellers who visit the region despite its low levels of development and health concerns. In fact, tourists to Central Africa may prefer more rustic and natural settings. Greater government investment in public health and social welfare may, however, improve conditions in this region, where there is not yet persistence in tourism arrivals.

Conclusion

The goal of this research was to investigate the determinants that influence foreign tourism arrivals to the African continent, firstly as a collective and secondly in different regions, with the aim to foster a greater understanding of how African countries and regions can grow their tourism economies. The second aspect makes this research unique, since previous research has focused either on the entire African continent or on individual African countries. This research finds that inbound tourism to Africa is dependent on the income of source markets in developed countries, the price at the destination in Africa, challenges faced due to telecommunication infrastructure and geographic factors, as well as conservation efforts of African countries. Additionally, the regional analysis indicates that each region will react differently to the set of determinants that influence tourism arrivals to Africa as a whole.

The contribution that the research makes is firstly in identifying the determinants of tourism to Africa, using the most complete panel of African countries to date, and secondly in investigating and comparing tourism to different regions in Africa, which have not been done to date. Based on the results of the analysis, several recommendations have also been made to increase tourism arrivals to the African continent.

This research shows that although panel data are useful in analysing continents where there are limited data availability, one should also take note that the coefficients estimated represent the average effect and may not apply equally to all countries or regions. A closer analysis reveals that the African regions are diverse and, although some results are generalizable, each region and country should invest in assessing the opportunities (and threats) to benefiting more extensively from tourism. A simple example is conservation: in the complete African panel results, the conservation proxies are positive and significant, indicating that increased conservation efforts are associated with more tourist arrivals. However, conservation is not considered to influence the choice between destinations within a specific region. Therefore, African countries should not solely focus on conservation efforts to attract tourists, but rather expand current offerings to include various products and services, while still promoting conservation and increasing investment in conservation efforts. Another example is price competitiveness: while Africa as a whole is considered to have become more price sensitive, a closer inspection reveals that this result may be attributed to price increase in Northern African countries and not necessarily in all African countries.

Variables that consistently influence the total African panel, as well as most regions, include telecommunication infrastructure and, to a lesser extent, income in the source markets. The development of infrastructure, and specifically telecommunication infrastructure, could lead to increased tourism to the entire continent. Since most African destinations are considered luxury offerings, more could be done to market the continent as a value-for-money destination that offers more-than-luxury safaris. The research also highlights the influence that economic and social development in African countries has on tourism.

The limitations of the research are mostly attributed to the availability and quality of the data. Although data availability has certainly increased over the past decade, a lack of detailed records of tourism and economic data for African countries still hampers research. Future research could segment the continent into the different countries and employ other data analysis approaches.

Supplemental material

Supplemental Material, Table_A1 - Determinants influencing inbound arrivals to Africa

Supplemental Material, Table_A1 for Determinants influencing inbound arrivals to Africa by Armand Viljoen, Andrea Saayman and Melville Saayman in Tourism Economics

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Armand Viljoen

Andrea Saayman

Melville Saayman

Supplemental material

Supplemental material for this article is available online.

References

Adeola

Boso

Evans

(2018) Drivers of international tourism demand in Africa. Business Economics 53(1): 25–36. DOI: 10.1057/s11369-017-0051-3.

Arellano

Bond

(1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies 58(2): 277. DOI: 10.2307/2297968.

Asteriou

Hall

(2015) Applied Econometrics. New York: Palgrave Macmillan.

Awaritefe

(2007) Evaluating tourism resource areas in Nigeria for development. Tourism Review 62(2): 37–45. DOI: 10.1108/16605370780000313.

Bankole

Babatunde

(2010) Elasticities of demand for tourism in Nigeria: an Ardl approach. Tourism Analysis 15(4): 489–496. DOI: 10.3727/108354210x12864727453421.

Baum

Schaffer

Stillman

(2003) Instrumental variables and GMM: estimation and testing. Stata Journal 3(1): 1–31.

Berrington

Smith

Sturgis

(2006) An overview of methods for the analysis of panel data. Available at: http://eprints.ncrm.ac.uk/415/1/MethodsReviewPaperNCRM-007.pdf (accessed 10 June 2016).

Blackburne

Frank

(2007) Estimation of nonstationary heterogeneous panels. Stata Journal 7(2): 197–208.

Bloom

Sachs

Collier

, et al. (1998) Geography, demography, and economic growth in Africa. Available at: http://www.brookings.edu/∼/media/Files/Programs/ES/BPEA/1998_2_bpea_papers/1998b_bpea_bloom_sachs_collier_udry.pdf (accessed 10 June 2016).

10.

Ningqiao

(2018) The impact of high-speed trains on regional tourism economies: empirical evidence from China. Tourism Economics 24(2): 187–203. DOI: 10.1177/1354816617749346.

11.

Bond

Windmeijer

(2002) Finite sample inference for GMM estimators in linear panel data models. Available at: http://www.diw.de/documents/dokumentenarchiv/17/39273/59_1_windmeijer.pdf (accessed 14 June 2016).

12.

Bonham

Gangnes

Zhou

(2009) Modeling tourism: a fully identified VECM approach. International Journal of Forecasting 25(3): 531–549. DOI: 0.1016/j.ijforecast.2008.11.014.

13.

Bourguignon

Des Hautes

(2001) Crime as a social cost of poverty and inequality: a review focusing on developing countries. In: Yusuf

Evenett

(eds) Facets of Globalization: International and Local Dimensions of Development. Washington: World Bank, pp. 171–191.

14.

Bruno

(2004) Estimation inference and Monte Carlo analysis in dynamic panel data models with small number of individuals. Available at: http://wwwtstatit/novita/conferenze/brunopdf (accessed 14 June 2016).

15.

Buddelmeyer

Jensen

Oguzoglu

, et al. (2008) Fixed effects bias in panel data estimators. Bonn: Institute for the Study of Labor (IZA). Available at http://nbn-resolving.de/urn:nbn:de:101:1-2008052773

16.

Buigut

Amendah

(2016) Effect of terrorism on demand for tourism in Kenya. Tourism Economics 22(5): 928–938. DOI: 10.5367/te.2015.0467.

17.

Bun

Carree

(2005) Bias-corrected estimation in dynamic panel data models. Journal of Business and Economic Statistics 23(2): 200–210. DOI: 101198/073500104000000532.

18.

Bun

MJG

Kiviet

(2001) The accuracy of inference in small samples of dynamic panel data models. Tinbergen Institute discussion paper TI 2001-006/4. Available at http://www.tinbergen.nl (accessed 24 November 2017).

19.

Burger

CJSC

Dohnal

Kathrada

, et al. (2001) A practitioners guide to time-series methods for tourism demand forecasting—a case study of Durban, South Africa. Tourism Management 22(4): 403–409. DOI: 10.1016/S0261-5177(00)00068-6.

20.

Chigora

Vutete

(2015a) Comprehending determinants of demand: Zimbabwe tourism destination. Scenario Issues in Economics and Business 1(2): 17–23. DOI: 10.5296/ieb.v1i2.8689.

21.

Chigora

Vutete

(2015b) Elasticity of demand a critical form tourism market position: Understanding Zimbabwe tourism destination. Case Studies in Business and Management 2(2): 18–26. DOI: 10.5296/csbm.v2i2.8570.

22.

Choyakh

(2008) A model of tourism demand for Tunisia: inclusion of the tourism investment variable. Tourism Economics 14(4): 819–838. DOI: 10.5367/000000008786440238.

23.

Choyakh

(2009) Modelling tourism demand in Tunisia using cointegration and error correction models. In: Matias

Nijkamp

Sarmento

(eds) Advances in Tourism Economics. Heidelberg: Physica-Verlag, pp. 71–84.

24.

Christie

Fernandes

Messerli

, et al. (2013) Tourism in Africa: harnessing tourism for growth and improved livelihoods. Available at: http://wwwworldbankorg/content/dam/Worldbank/document/Africa/Report/africa-tourism-report-2013-overviewpdf (accessed 15 May 2016).

25.

Crouch

(1994) The study of international tourism demand: a review of findings. Journal of Travel Research 33(1): 12–23. DOI: 10.1177/004728759403300102.

26.

De Mello

Nell

(2005) The forecasting ability of a cointegrated VAR system of the UK tourism demand for France, Spain and Portugal. Empirical Economics 30(2): 277–308. DOI: 10.1007/s00181-005-0241-0.

27.

Deng

Athanasopoulos

(2011) Modelling Australian domestic and international inbound travel: a spatial–temporal approach. Tourism Management 32(5): 1075–1084. DOI: 10.1016/j.tourman.2010.09.006.

28.

Dogru

Sirakaya-Turk

Crouch

(2017) Remodeling international tourism demand: Old theory and new evidence. Tourism Management 60: 47–55. DOI: 10.1016/j.tourman.2016.11.010.

29.

Dritsakis

(2004) Tourism as a long-run economic growth factor: an empirical investigation for Greece using causality analysis. Tourism Economics 10(3): 305–316. DOI: 10.5367/0000000041895094.

30.

Drukker

(2010) Generalized method of moments (GMM) estimation in Stata 11. Available at: http://wwwstatacom/meeting/mexico10/mex10sug_drukkerpdf (accessed 14 June 2016).

31.

Durbarry

Nicolas

Seetanah

(2009) The determinants of tourism demand in South Africa using a dynamic panel data approach. Tourism Analysis 14(3): 375–385. DOI: 10.3727/108354209789704968.

32.

Eilat

Einav

(2004) Determinants of international tourism: a three-dimensional panel data analysis. Applied Economics 36(12): 1315–1327. DOI: 10.1080/000368404000180897.

33.

Eja

Joseph

Ojong

(2012) Success factors determining Nigeria as a tourist destination. Journal of Emerging Trends in Educational Research and Policy Studies 3(4): 426–432. DOI: 10520/EJC126559.

34.

Erickson

Whited

(2002) Two-step GMM estimation of the errors-in-variables model using high-order moments. Econometric Theory 18(03): 776–799. DOI:101017/s0266466602183101.

35.

Faragalla

WSS

(2018) Econometric analysis for tourism demand function in Egypt a dynamic panel data approach. Facta Universitatis Series: Economics and Organization 14: 321–332. DOI: 10.22190/fueo1704321f.

36.

Fayissa

Nsiah

Tadasse

(2008) Impact of tourism on economic growth and development in Africa. Tourism Economics 14(4): 807–818. DOI: 105367/000000008786440229.

37.

Fourie

Santana-Gallego

(2013) The determinants of African tourism. Development Southern Africa 30(3): 347–366. Available at: https://doi.org/10.1080/0376835X.2013.817302.

38.

Garín-Mun

(2006) Inbound international tourism to Canary Islands: a dynamic panel data model. Tourism Management 27(2): 281–291. DOI: 101016/jtourman200410002.

39.

Garín-Muñoz

Montero Martín

(2007) Tourism in the Balearic Islands: a dynamic model for international demand using panel data. Tourism Management 28(5): 1224–1235. DOI: 10.1016/j.tourman.2006.09.024.

40.

Gasmi

Sassi

(2015) International tourism demand in Tunisia: evidence from dynamic panel data model. Economics Bulletin 35(1): 507–518. Available at: www.accessecon.com/Pubs/EB/2015/Volume35/EB-15-V35-I1-P55.pdf (accessed 16 June 2017).

41.

Ghalia

(2016) Essays on tourism and its determinants (Doctoral dissertation, Brunel University London).

42.

Gil Alana

(2011) Tourism in South Africa-time series persistence and the nature of the shocks: are they transitory or permanent? African Journal of Business Management 5(12): 4686–4698. DOI: 10.5897/AJBM10.316.

43.

Goh

Law

(2011) The methodological progress of tourism demand forecasting: a review of related literature. Journal of Travel and Tourism Marketing 28(3): 296–317. DOI: 10.1080/10548408.2011.562856.

44.

González Gómez

Álvarez-Díaz

Otero-Giráldez

(2011) Modeling domestic tourism demand in Galicia using the ARDL approach. Tourism & Management Studies (7): 54–61. Available at: http://www.scielo.mec.pt/pdf/tms/n7/n7a06.pdf (accessed 12 June 2017).

45.

Gooroochurn

Sinclair

(2005) Economics of tourism taxation: evidence from Mauritius. Annals of Tourism Research 32(2): 478–498. DOI: 10.1016/j.annals.2004.10.003.

46.

Gopaul

Cheeneebash

(2015) A comparison of the ANFIS model with SARIMA for the forecasting of inbound tourism demand for Mauritius. University of Mauritius Research Journal 21: 1–16. Available at: https://www.ajol.info/index.php/umrj/article/viewFile/122067/111545 (accessed 25 July 2017).

47.

Guizzardi

Stacchini

(2015) Real-time forecasting regional tourism with business sentiment surveys. Tourism Management 47: 213–223. DOI: 10.1016/j.tourman.2014.09.022.

48.

Gujarati

Porter

(2009) Basic Econometrics (5th ed.). Boston: McGraw-Hill.

49.

Hamadeh

Bassil

(2017) Terrorism, war, and volatility in tourist arrivals: the case of Lebanon. Tourism Analysis 22(4): 537–550. DOI: 10.3727/108354217X15023805452086.

50.

Hathroubi

(2011) Modeling and forecasting international tourism demand for Tunisia: a time varying parameter approach interdisciplinary. Journal of Contemporary Research in Business 3(7): 261–279. Available at: https://www.researchgate.net/publication/285651247_Modelling_and_forecasting_international_tourism_demand_for_Tunisia_A_time_varying_parameter_approach (accessed 14 July 2017).

51.

Ibrahim

MAMA

(2013) The determinants of international tourism demand for Egypt: panel data evidence. European Journal of Economics, Finance and Administrative Sciences 30: 50–58. DOI: 10.2139/ssrn.2359121.

52.

International tourism trends in EU-28 member states. Current situation and forecasts for 2020-2025-2030 (2016). Available at: https://ec.europa.eu/growth/content/international-tourism-trends-eu-28-member-states-current-situation-and-forecast-2020-2025-0_en (published 27 May 2016).

53.

Judson

Owen

(1999) Estimating dynamic panel data models: a guide for macroeconomists. Economics Letters 65(1): 9–15. DOI: 101016/s0165-1765(99)00130-5.

54.

Kareem

(2008) A panel data analysis of demand for tourism in Africa, Ibadan. Journal of Social Sciences. Available at: https://www.researchgate.net/profile/Olayinka_Kareem/publication/228883530_A_panel_data_analysis_of_demand_for_tourism_in_Africa/links/0046353b4757722acd000000/A-panel-data-analysis-of-demand-for-tourism-in-Africa.pdf (accessed 17 May 2017).

55.

Karl

Winder

Bauer

(2017) Terrorism and tourism in Israel: analysis of the temporal scale. Tourism Economics 23(6): 1343–1352. DOI: 10.1177/1354816616686417.

56.

Khadaroo

Seetanah

(2008) Transport and economics: The case of mauritius. Journal of Transport Economics and Policy 42(2): 255–267. Available at: https://www-jstor-org-443.web.bisu.edu.cn/stable/20054047.

57.

Kim

Park

Gazzoli

, et al. (2011) Benefit segmentation of international travelers to Macau, China. Journal of Quality Assurance in Hospitality & Tourism 12(1): 28–57. DOI: 10.1080/1528008X.2011.541813.

58.

Kirchgässner

Wolters

Hassler

(2013) Nonstationary processes. In: Kirchgässner

Wolters

(eds) Introduction to Modern Time Series Analysis. Berlin: Springer, pp. 155–203.

59.

Kiviet

(1995) On bias inconsistency and efficiency of various estimators in dynamic panel data models. Journal of Econometrics 68(1): 53–78. DOI: 101016/0304-4076(94)01643-e.

60.

Lee

(2011) Forecasting long-haul tourism demand for Hong Kong using error correction models. Applied Economics 43(5): 527–549. DOI: 10.1080/00036840802599743.

61.

Wong

Song

, et al. (2006) Tourism demand forecasting: a time varying parameter error correction model. Journal of Travel Research 45(2): 175–185. DOI: 10.1177/0047287506291596.

62.

Song

(2017) A dynamic panel data analysis of climate and tourism demand: additional evidence. Journal of Travel Research 56(2): 158–171. DOI: 10.1177/0047287515626304.

63.

Pan

Law

, et al. (2017) Forecasting tourism demand with composite search index. Tourism Management 59: 57–66. DOI: 10.1016/j.tourman.2016.07.005.

64.

Lim

(1997) Review of international tourism demand models. Annals of Tourism Research 24(4): 835–849. DOI: 101016/s0160-7383(97)00049-2.

65.

Lim

(2004) The major determinants of Korean outbound travel to Australia. Mathematics and Computers in Simulation 64(3–4): 477–485. DOI: 10.1016/S0378-4754(03)00113-7.

66.

Lim

McAleer

(2001) Monthly seasonal variations: Asian tourism to Australia. Annals of Tourism Research 28(1): 68–82. DOI: 10.1016/S0160-7383(00)00002-5.

67.

López

Fernández

Incera

(2016) The economic impact of international students in a regional economy from a tourism perspective. Tourism Economics 22(1): 125–140. DOI: 10.5367/te.2014.0414.

68.

Luo

(2017) Special focus: identifying the influential elements in the regional tourism economy. Tourism Economics 24(2): 185–186. DOI: 10.1177/1354816617749342.

69.

Mileva

(2007) Using Arellano–Bond Dynamic Panel GMM Estimators in Stata. New York: Fordham University.

70.

Moyo

Ziramba

(2013) The impact of crime on inbound tourism to South Africa: an application of the bounds test. African Security Review 22(1): 4–18. DOI: 10.1080/10246029.2012.737815.

71.

Mpofu

TPZ

(2009) An analysis of factors that caused a downturn in tourism demand trends and patterns in Zimbabwe from 2000 to 2005. Journal of African Development Studies 2(1): 35–49.

72.

Muchapondwa

Pimhidzai

(2011) Modelling international tourism demand for Zimbabwe. International Journal of Business and Social Science 2(2): 71–81. Available at: http://ijbssnet.com/journals/Vol._2_No._2%3B_February_2011/8.pdf (accessed 15 July 2017).

73.

Naudé

Saayman

(2005) Determinants of tourist arrivals in Africa: a panel data regression analysis. Tourism Economics 11(3): 365–391. DOI: 105367/000000005774352962.

74.

Nelson

Dickey

Smith

(2011) Estimating time series and cross section tourism demand models: mainland United States to Hawaii data. Tourism Management 32(1): 28–38. DOI: 10.1016/j.tourman.2009.10.005.

75.

Ouerfelli

(2008) Co-integration analysis of quarterly European tourism demand in Tunisia. Tourism Management 29(1): 127–137. DOI: 10.1016/j.tourman.2007.03.022.

76.

Ouerfelli

(2010) Analysis of European tourism demand for Tunisia: a new approach. International Journal of Tourism Policy 3(3): 223–236. DOI: 10.1504/ijtp.2010.037994.

77.

Priego

Rosselló

Santana-Gallego

(2015) The impact of climate change on domestic tourism: a gravity model for Spain. Regional Environmental Change 15(2): 291–300. DOI: 10.1007/s10113-014-0645-5.

78.

Ridderstaat

Odube

Croes

, et al. (2014) Impacts of seasonal patterns of climate on recurrent fluctuations in tourism demand: evidence from Aruba. Tourism Management 41: 245–256. DOI: 10.1016/j.tourman.2013.09.005.

79.

Rivera

(2016) A dynamic linear model to forecast hotel registrations in Puerto Rico using Google Trends data. Tourism Management 57: 12–20. DOI: 10.1016/j.tourman.2016.04.008.

80.

Rodrigue

Comtois

Slack

(2013) The Geography of Transport Systems. London: Routledge.

81.

Rogerson

(2007) Reviewing Africa in the global tourism economy. Development Southern Africa 24(3): 361–379. DOI: 101080/03768350701445350.

82.

Roodman

(2009) How to do xtabond2: an introduction to difference and system GMM in Stata. Stata Journal 9(1): 86–136. DOI: 102139/ssrn982943.

83.

Saayman

(2008) Determinants of inbound tourism to South Africa. Tourism Economics 14(1): 81–96. DOI: 105367/000000008783554893.

84.

Sahli

Nowak

(2007) Does inbound tourism benefit developing countries? A trade theoretic approach. Journal of Travel Research 45(4): 426–434. Available at: https://doi.org/10.1177/0047287506295948.

85.

Samitas

Asteriou

Polyzos

, et al. (2018) Terrorist incidents and tourism demand: evidence from Greece. Tourism Management Perspectives 25: 23–28. DOI: 10.1016/j.tmp.2017.10.005.

86.

Sassi

(2014) Combination forecasts of international demand for tourism in Tunisia. Journal of Quantitative Economics 12(2): 94–109.

87.

Seddighi

Nuttall

Theocharous

(2001) Does cultural background of tourists influence the destination choice? An empirical study with special reference to political instability. Tourism Management 22(2): 181–191. DOI: 101016/s0261-5177(00)00046-7.

88.

Seetanah

Durbarry

Ragodoo

JFN

(2010) Using the panel cointegration approach to analyse the determinants of tourism demand in South Africa. Tourism Economics 16(3): 715–729. DOI: 105367/000000010792278437.

89.

Seetanah

Sannassee

Rojid

(2015) The impact of relative prices on tourism demand for Mauritius: an empirical analysis. Development Southern Africa 32(3): 363–376. DOI: 10.1080/0376835x.2015.1010717.

90.

Seetaram

(2010) The determinants of international tourism flows: empirical evidence from Australia. Available at: http://www.monash.edu/__data/assets/pdf_file/0005/925457/a_study_of_outbound_tourism_from_australia.pdf (accessed 15 June 2016).

91.

Shahrabi

Hadavandi

Asadi

(2013) Developing a hybrid intelligent model for forecasting problems: case study of tourism demand time series. Knowledge-Based Systems 43: 112–122. DOI: 10.1016/j.knosys.2013.01.014.

92.

Sharma

(2004) Tourism and Economic Development. New Delhi: Sarup & Sons.

93.

Sheppard

(2005) Generalized methods of moments (GMM). Available at: https://wwwkevinsheppardcom/images/5/55/Chapter6pdf (accessed 14 May 2016).

94.

Song

Dwyer

, et al. (2012) Tourism economics research: a review and assessment. Annals of Tourism Research 39(3): 1653–1682. DOI: 10.1016/j.annals.2012.05.023.

95.

TII (2011) Prospects for the emerging travel markets 2012. Tourism Intelligence International, see TII. Available at: http://wwwtourism-intelligencecom/pdf/NL%20Nov%202011%20%2811%29pdf (accessed 15 June 2016).

96.

Turvill

(2013) The true size of Africa. The Daily Mail, 5 October.

97.

UNWTO (2010) Study on Chinese outbound travel to Africa. Available at: http://pubunwtoorg/WebRoot/Store/Shops/Infoshop/4D13/2C30/152E/1FF1/08C9/C0A8/0164/8ED0/110204_chinese_outbound_to_africa_excerptpdf (accessed 10 May 2016).

98.

UNWTO (2013a) World’s top tourism spenders. Available at: http://dtxtq4w60xqpwcloudfrontnet/sites/all/files/pdf/tsen_0pdf (accessed 10 May 2016).

99.

UNWTO (2013b) Tourism highlights 2013 edition. United Nations World Tourism Organization, see UNWTO. Available at: http://wwwe-unwtoorg/doi/pdf/1018111/9789284415427 (accessed 14 June 2016).

100.

Wahba

Elanani

(2017) The exchange rate fluctuations and its effect on international tourism demand case study: Egypt the period (2000–2016). Journal of Tourism Research 17(1): 128–139. Available at: http://www.jotr.eu/index.php/volume17/169-the-exchange-rate-fluctuations-and-its-effect-on-international-tourism-demand-case-study-egypt-the-period-2000-2016 (accessed 12 November 2017).

101.

Wakimin

Azlinaa

Hazman

(2018) Tourism demand in Asean-5 countries: Evidence from panel data analysis. Management Science Letters 8(6): 677–690. DOI: 10.5267/j.esm.2018.9.002.

102.

WEF (2013a) The travel and tourism competitiveness report 2013. Available at: http://www3weforumorg/docs/WEF_TT_Competitiveness_Report_2013pdf (accessed 14 June 2016).

103.

WEF (2013b) The African competitiveness report 2013. Available at: http://www3weforumorg/docs/WEF_Africa_Competitiveness_Report_2013pdf (accessed 14 June 2016).

104.

Williams

(2015) Panel data: fixed effects vs random effects models University of Notre Dame. World Economic Forum, see WEF. Available at: https://www3ndedu/∼rwilliam/stats3/Panel04-FixedVsRandompdf (accessed 15 May 2016).

105.

Song

Shen

(2017) New developments in tourism and hotel demand modeling and forecasting. International Journal of Contemporary Hospitality Management 29(1): 507–529. DOI: 10.1108/IJCHM-05-2015-0249.

106.

Xiao

Smith

SLJ

(2006) The making of tourism research: insights from a social sciences journal. Annals of Tourism Research 33(2): 490–507. DOI: 10.1016/j.annals.2006.01.004.

107.

Yang

Pan

Evans

, et al. (2015) Forecasting Chinese tourist volume with search engine data. Tourism Management 46(2015): 386–397. DOI: 10.1016/j.tourman.2014.07.019.

108.

Yazdi

Khanalizadeh

(2017) Tourism demand: a panel data approach. Current Issues in Tourism 20(8): 787–800. DOI: 10.1080/13683500.2016.1170772.

109.

Zhang

Kulendran

(2017) The impact of climate variables on seasonal variation in Hong Kong inbound tourism demand. Journal of Travel Research 56(1): 94–107. DOI: 10.1177/0047287515619692.

110.

Zhou Grundy

Turner

(2014) The challenge of regional tourism demand forecasting: the case of China. Journal of Travel Research 53(6): 747–759. DOI: 10.1177/0047287513516197.

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

0.01 MB