The determinants of African tourism

Abstract

Using a standard panel gravity equation of 175 origin/destination countries between 1995 and 2008, 43 of which are African, we identify the factors that drive African-inbound (arrivals to Africa from other continents) and within-African tourism (arrivals from and to an African country). We find that the determinants of African-inbound and within-African tourism are not all that different from global tourism flows; repeat tourism, income, distance, land area and the standard dummy variables not only drive global or OECD tourism, but also tourism within Africa, disproving the belief that African tourists ‘differ substantially’.

Keywords:

• sub-Saharan
• Africa
• gravity model
• cause
• trade
• VFR

1. Introduction

Tourism is a rapidly growing segment of African countries' export baskets. Between 1995 and 2008, tourism receipts increased by 13.70% on average for 28 African countries (for which sufficient data are available). This is higher than growth in the export of goods, for example, which increased by 11.97% over the period for the same sample of countries.

Tourism is often considered a catalyst for economic and social development; it tends to have a large trickle-down effect in terms of poverty alleviation, boosting employment creation and small business entrepreneurship. These theoretical benefits have recently found empirical support; Fayissa et al. (2008:807) show that ‘receipts from the tourism industry contribute significantly both to the level of gross domestic product and to the economic growth of sub-Saharan African countries’. The authors then conclude: ‘African economies could enhance their short-run economic growth by strengthening their tourism industries strategically’ (Fayissa et al., 2008:807).

The purpose of this paper is to identify the determinants of African-inbound and within-African tourism, and compare these determinants with those of OECD and global tourism flows. Building on a rich theoretical foundation, the paper empirically identifies the most critical sources that drive tourist arrivals. To do this, we define both a static and dynamic gravity equation for a panel comprised of 175 origin/destination countries, of which 43 are African. The dynamic panel data methodology adopted in this paper accounts for the possibility of endogeneity in tourism.

The paper is organised as follows. Section 2 presents an overview of African tourism trends. Section 3 presents the data and methodology used in the empirical analysis. Section 4 presents the main results where we not only measure the determinants of global tourist arrivals, but also differentiate between tourists from different regions. While Africa is still a fragmented continent characterised by low levels of inter-regional trade, we show that the inter-regional movement of people is on the rise, driven by a largely similar set of determinants. Finally, some conclusions are drawn in Section 5.

2. African tourism

Tourism, defined as Mode-2 travel service exports,3 is increasingly viewed as an export sector with high growth potential. A number of African countries, in particular, have begun encouraging the tourism industry as a means to earn foreign revenues, diversify their export baskets, create jobs and, ultimately, improve economic growth and development. Such support seems to be paying dividends. African countries have experienced strong growth in tourist arrivals during our sample period of 1995–2008. This is reflected in Figure 1, which shows the number of tourist arrivals by country for 1996 and 2006. We choose 1996 and 2006 because of data availability (in the absence of data for some countries, we use the most recent year available; see Appendix A) and to exclude the impact of the global economic recession beginning in 2008.

Figure 1: Number of tourist arrivals, 1996 and 2006

One striking feature of Figure 1 is the pervasive nature of tourism growth throughout the continent; the results are not dependent on the remarkable achievements by a select few countries. In fact, tourist arrivals increased in countries at both ends of the destination spectrum: South Africa – the largest tourism hub in sub-Saharan Africa – witnessed an average increase of 9.7% annually in tourism receipts between 1995 and 2008, as did many of the other large African countries, including Nigeria (21.42%), Ghana (21.03%), Angola (20.13%) and Ethiopia (15.74%). But growth was not restricted to the larger countries; where data are available, tourism seems to have performed particularly well in the tiny African countries of Rwanda (35.21%), The Gambia (8.75%) and Cape Verde (23.09%).

The rapid and widespread growth in African tourism has naturally provoked interest in its causes. Following the tourist demand literature, standard factors that explain tourism flows include the income of the country of origin, price differentials, travel costs (including flights, visas, insurance, etc.), exchange rate differences, competitor destinations, marketing expenditures, and various others (Lim, 1997). While such demand-type analyses date back to the 1960s, it is only recently that Africa has received more than mere footnote attention.

Principally, it seems that supply-side constraints inhibit the growth of the tourism industry: at the micro-level, the safety and security of tourists (Gauci et al., 2002), the quality–price offering, especially of standardised tour packages (Christie & Crompton, 2001), and the lack of tourism infrastructure, including availability of hotels and rental vehicles; while at the macro-level, the poor transport infrastructure, roads, railroads and airports (Kester, 2003; Estache, 2004), lack of development in the complimentary sectors of, for example, communications and finance (Cleverdon, 2002), high levels of political risk (Eilat & Einav, 2004) and a detrimental disease environment.

Of course, the direction of causality remains ambiguous: are these factors determinants of tourist arrivals or simply a consequence of tourism (or a lack thereof)? These questions can only be addressed through more precise empirical exercises, the most comprehensive of which – by Naudé & Saayman (2005) – use cross-section and panel methods to investigate the determinants of tourism to African countries. They find that tourism infrastructure, the level of a country's development and Internet usage are significant explanatory variables, while political and social instability also undermines tourism growth, confirming the earlier hypotheses. They find little impact of price differentials, suggesting that tourism to Africa is not determined by exchange rate movements.

Moreover, Naudé & Saayman (2005) account for dynamics in their panel data regression. They argue that there are ‘persistence/reputation effects’ that apply over time in the destination decision, for instance by tourists returning to a particular destination or recommending a country to friends and relatives – the word-of-mouth effect – after having a good experience. Their results show that the lagged tourist arrival variable is significantly negative, suggesting that African destinations do not generate repeat visits. Khandaroo & Seetanah (2008) also obtain an insignificant lagged tourist arrival variable.

More recently, and with the focus towards supply-side factors, geography has entered the fray. Saayman & Saayman (2008), looking only at South African tourist arrivals and in addition to the standard control variables, find that climate (measured as the number of sunny days in Cape Town) also contribute to tourist arrivals. Du Toit & Fourie (2012) also find proof that climate and environmental factors boost African countries' comparative advantage in travel service exports. But whereas environmental factors may of course explain the underlying reasons for tourist arrivals, being (relatively) constant, it cannot explain the rapid growth in tourist arrivals, except to the extent that other debilitating factors, acting as binding constraints, are now softened, enabling countries to realise their comparative advantage.

A trend that has escaped the discourse, however, is the stark growth of within-African tourism. While most marketing and promotion campaigns focus on the lucrative markets of Europe, North America and, increasingly, East Asia, within-African tourism has escaped attention of policy-makers, even though more than 20 million Africans travelled to other African countries in 2008, up from just over nine million in 1995. The literature also seems to eschew the significance of within-African trade.

Saayman & Saayman (2008), for example, differentiate between international travellers to South Africa and travellers from African countries and then continue to only estimate the determinants of international tourists, reasoning that ‘previous research … has shown that the spending of tourists from these markets is low compared to international markets and that the reasons for travelling to South Africa differ substantially from those of international travellers’ (2008:85). While it may be true that all movement across international borders in Africa is not strictly Mode-2 travel service exports – migrant labourers should be classified under Mode 4 – there is no denying that within-African tourism is both significant and increasing.

Figure 2 provides a snapshot of African and non-African tourist arrivals in African countries in 2005. Visually, the large percentage of African tourists in especially central and southern Africa is striking, compared with the very small percentage of African tourists in the North African countries. These characteristics will reappear in the regression analysis below.

Figure 2: Percentage of tourist arrivals by continent of origin, 2005

Table 1 provides a summary of changes in African tourist arrivals between 1995 and 2008. Although within-African tourism dropped off significantly as a share of total tourists between 1995 and 2000, it regained some of its lost ground leading up to 2008. More importantly, its growth was off a high base; even allowing for the strong growth of non-African tourists between 1995 and 2008, 36% of all tourists arriving in African countries in 2008 came from other African countries.

Table 1: Origin of tourists to African countries

	Number of tourist arrivals (in thousands)				Percentage of total
Region	1995	2000	2004	2008	1995	2000	2004	2008
Africa	9 039	11 239	14 428	20 410	40.01	34.75	35.53	36.06
Americas	844	1 377	1 275	1 921	3.73	4.26	3.14	3.39
East Asia	671	933	1 062	1 852	2.97	2.88	2.61	3.27
Europe	8 540	14 362	17 090	23 759	37.80	44.41	42.08	41.97
Middle East	1 554	1 805	3 026	3 857	6.88	5.58	7.45	6.81
South Asia	127	208	266	481	0.56	0.64	0.65	0.85
Not specified	1 821	2 417	3 456	4 324	8.06	7.47	8.51	7.64
Total	22 596	32 341	40 610	56 605
Source: Authors’ own elaboration of UN-WTO.

This must be seen as a positive sign. Africa remains a fragmented continent. Its low export diversity – which limits African countries' demand for their neighbours' produce – combined with poor transportation and communication infrastructure explain partly why African countries, relative to other regions, trade little with one another. In addition, historical remnants such as idiosyncratic national boundaries drawn up during colonial times or the practice of slavery that inhibited trade and the free movement of people create path-dependent distortions that still impact African countries today (Nunn, 2008). Export diversification into tourism services (and probably the de facto free labour market) may boost regional integration efforts, with spill-over effects into other service exports and goods.

The purpose of this paper is thus twofold: first, we aim to add to the literature on the determinants of inbound tourism by considering a static and dynamic version of the tourism gravity model. Second, by using a dataset that includes tourist movements for 175 countries worldwide from 1995 to 2008, we identify the factors driving inbound tourism to African countries. This will allow us to determine the extent to which tourism to Africa is ‘different’ from world tourism. Thirdly, we shed light on an enigma of African tourism: the determinants of within-African tourism.

3. Data and method of analysis

To analyse the determinants of African tourist arrivals, a gravity equation with tourism flow as the dependent variable is estimated. In this section, we discuss the features of the gravity equation, describe the dataset used and present the empirical strategy.

The gravity model is a workhorse in a number of empirical issues addressed within international economics. The origin of this model is Newton's Law of Universal Gravitation, and it was firstly proposed by Tinbergen (1962) to describe international bilateral trade. The main reason for its extensive use in empirical research is its goodness of fit, since international flows increase with the economic size of countries and decrease as the distance between them increases.

This type of specification has been used to estimate the effects of economic and non-economic events on international flows of goods (Armstrong, 2007; Fratianni, 2009), migrants (Karemera et al., 2000; Gil et al., 2006), foreign direct investment (Bergstrand & Egger, 2007; Eichengreen & Tong, 2007; Head & Ries, 2008) and tourism (Durbarry, 2000; Gil et al., 2007; Santana-Gallego et al., 2010a).

Indeed, this type of equation has been commonly used to investigate a number of empirical regularities, such as border effects (McCallum, 1995; Fitzsimons et al., 1999), regional trading blocs (Matyas et al., 2004; Cheng & Wall, 2005), currency unions (Rose, 2000; Rose & van Wincoop, 2001) and mega-events (Fourie & Santana-Gallego, 2011). Therefore, the following model is estimated:

where i indicates destination country, j origin country and t is time; is a constant; Ln denotes natural logarithms; , and are origin, destination and year fixed effects, respectively; and is a well-behaved disturbance term. In the analysis, as well as the standard gravity variables, the model is augmented using four different sets of factors: economic relationship variables, geographic variables, cultural affinity variables and development and stability variables. Table 2 presents a brief description of variables included in the analysis.

Table 2: Summary of variables used in the model

Variable	Definition
LnTou ijt	Log of tourist arrivals to destination country from the origin country
Standard gravity variables
LnGDPpc it	Log of gross domestic product per capita of the destination country
LnGDPpc jt	Log of gross domestic product per capita of the origin country
LnDist ij	Log of the distance between countries in the pair as a proxy of transport costs
Economic relationship
TouInv it	Capital investment in the tourism sector of the destination country as a percentage of total capital investment
LnTrade ijt	Log of bilateral trade as the sum of exports and imports between country pairs
CU ijt	Dummy variable: both countries in the pair share a common currency
RTA ijt	Dummy variable: both countries in the pair belong to the same regional trade agreement (different from the monetary agreement)
LnComp ijt	Relative purchasing power parity conversion factor (to market exchange rates) as a measure of the relative price competitiveness of the destination country to the origin one
Geography
LnArea i	Log of surface area of the destination country in square kilometres
Border ij	Dummy variable: both countries in the pair share a common land border
Coastline i	Length of the coastline of each destination country measured in kilometres
Temp i	Average annual temperature of each destination country in degrees Celsius
Cultural affinity
Lang ij	Dummy variable: both countries in the pair speak a common language
Colony ij	Dummy variable: both countries in the pair have ever had a colonial link
Relig ij	Dummy variable: More than 60% of the population of both countries are from the same religion
Development and stability
PS it	Political stability is measured in units ranging from about –2.5 to 2.5, with higher values corresponding to better governance outcomes
Lnlife it	Life expectancy index in the destination country (life expectancy at birth)

Our dependent variable is tourist arrivals by country of origin. The data are obtained from the United Nations World Tourism Organisation (UN-WTO). Sources of data are presented in Table A1 in Appendix A. The full sample includes 175 countries as origin/destination of tourists – of which 43 are African countries – over the period 1995–2008. The list of countries considered in the analysis is reported in Table A2 in Appendix A.

The WTO (2012) distinguishes between traveller, visitor and tourist. A ‘traveller’ is defined as a person who moves between different geographic locations, for any purpose and any duration. A ‘visitor’ is a traveller taking a trip to a destination outside his/her usual environment for less than a year and for any purpose other than to be employed by a resident entity in the country or place visited. Finally, a visitor (domestic, inbound or outbound) is classified as a tourist if his/her trip includes an overnight stay, or as a same-day visitor (or excursionist) otherwise. Hence, the visitor is a particular type of traveller and consequently tourism is a subset of travel.

The UN-WTO presents different measures of tourist arrivals because countries use varying methods of recording. These measures include the number of tourist arrivals at the border of a country, the number of visitor arrivals at the border of a country (tourists plus same-day visitors), the number of tourist arrivals at all accommodation types or the number of tourist arrivals at hotels and similar types of accommodation. The measure used in this paper is the number of tourist arrivals at the national border.

A limitation of the UN-WTO data, therefore, is that they do not include the objective of the trip (business, leisure or other personal purpose). From a tourism development perspective, it is assumed that choosing to travel and the choice of destination is somewhat discretionary for tourists; for economic migrants, people who travel to visit friends and relatives and overland traders, this is, of course, not the case. For this reason, earlier research excluded within-African tourists in an attempt to identify the determinants of tourist arrivals in Africa.

We take a different approach. Instead of investigating the reasons tourists travel to African countries, we use a gravity estimation strategy that helps to identify the determinants that affect both African-inbound tourists and within-African tourists. If the determinants differ between the two groups, a strong case can be made for treating African-inbound and within-African tourists as two distinct groups in empirical research, and also in policy-making and marketing campaigns. However, if the determinants are similar, tourism researchers and policy-makers should caution against such polarisation.

To investigate this, we use a static and dynamic gravity equation estimation strategy. Empirical research on the gravity equation commonly estimates by pooled ordinary least squares (OLS). However, if we assume that an unobserved heterogeneity exists, this technique can provide inconsistent and inefficient estimates. In this sense, a panel fixed-effects estimator (OLS-FE) offers a more suitable estimation technique to control for individual heterogeneity. Nevertheless, the fixed-effects approach does not allow for estimating coefficients of time-invariant variables such as the distance, the common border or language dummies.

In the recent econometric literature, a way to overcome this problem is to introduce individual country fixed-effects for the importers and the exporters in the gravity model (Matyas et al., 2004; Cheng & Wall, 2005; Kandogan, 2008). Moreover, the inclusion of country fixed effects is proposed by Rose & Van Wincoop (2001) as a way to approximate the multilateral resistances defined in the well-founded approach of Anderson & Van Wincoop (2004). In other words, the estimation of country-specific effects is suitable not only from an econometric point of view, but also adheres to the theoretical foundations of the gravity specification.

According to Martínez-Zarzoso et al. (2009), most of the existing trade gravity models based on panel data ignore dynamic effects; for example, only a few papers take into account persistence effects (de Nardis & Vicarelli, 2004; de Benedictis et al., 2005; Martínez-Zarzoso et al., 2009). Dynamics is introduced into the trade gravity model since exports tend to be highly persistent. Similarly, tourist arrivals can also present persistence or word-of-mouth effects. Moreover, tourist arrivals may change sluggishly due to supply constraints, such as shortages of accommodation, passenger transportation capacity or trained staff.

The introduction of dynamics into panel data models renders the OLS-FE estimator biased and inconsistent since the lagged endogenous variable correlates with the error term. The first differences-generalised methods of moments estimator by Arellano & Bond (1991) is commonly used in the literature to estimate dynamic panel data models. However, with a highly persistent dependent variable, it is more appropriate to use the system-generalised methods of moments (SYS-GMM) estimator proposed by Blundell & Bond (1998). Moreover, this method has the additional advantage that it allows us to obtain the estimates of time-invariant regressors included in the gravity model; that is, distance, common language, contiguity or colonial ties.

The SYS-GMM estimator is derived from the estimation of a system of two simultaneous equations, one in levels and the other in first differences. Where heteroscedasticity and serial correlation is a serious concern, the two-step SYS-GMM is asymptotically more efficient but standard errors tend to be severely downward biased. It is possible to solve this problem using the finite-sample correction to the two-step covariance matrix derived by Windmeijer.

The SYS-GMM allows endogeneity in some of the explanatory variables. In our case, apart from the lagged dependent variable, the gross domestic product (GDP) per capita of the destination country, the trade flows between countries and the investment in the tourist sector are considered endogenous. Lagged endogenous regressors are used as instruments in the estimation of the first-differenced equation while their lagged first-differences are instruments in the estimation of the level equation. Exogenous variables are used as standard instruments in both equations.

4. Results

The static version of the gravity model for tourist arrivals is estimated by OLS-FE where origin, destination and year fixed effects are included. The dynamic version is estimated by two-step SYS-GMM. We first estimate the determinants of tourist arrivals for the full sample of countries (175 × 175) to study the factors that drive global tourism. Then, we split the sample into OECD destinations (34 × 175) and African destinations (43 × 175) to analyse similarities and differences between tourist arrivals to developed countries and to the African continent. The results of the OLS-FE and SYS-GMM estimates are reported in Table 3. Results are discussed for the dynamic version of the gravity equation while the results for the static model are presented for comparison.

Table 3: Panel estimations of the gravity equation for tourist arrivals, full sample, period 1995–2008

	OLS-FE			SYS-GMM
Variable	World	OECD	Africa	World	OECD	Africa
LnTou ijt – 1				0.198*	0.270*	0.521*
				(8.44)	(8.07)	(16.20)
LnGDPpc it	0.258*	0.113	0.116***	0.275*	0.434*	0.273*
	(8.80)	(1.20)	(1.90)	(4.19)	(3.45)	(4.69)
LnGDPpc jt	0.095*	0.152*	0.152*	0.720*	0.366*	0.274*
	(3.33)	(3.28)	(2.62)	(16.49)	(6.14)	(6.09)
LnDist ij	–1.374*	–1.262*	–1.490*	–0.679*	–0.527*	–0.495*
	(–187.34)	(–84.71)	(–73.41)	(–17.69)	(–10.22)	(–7.48)
TouInv ijt	0.005*	0.006	0.007***	0.006*	0.067*	–0.005
	(3.67)	(1.36)	(1.87)	(2.90)	(7.04)	(–1.55)
LnTrade ijt	0.089*	0.126*	0.036*	0.089*	0.356*	0.068*
	(60.62)	(29.20)	(14.15)	(7.74)	(11.32)	(5.82)
LnComp ijt	–0.032	–0.173*	0.190*	–0.100**	–0.307*	–0.328*
	(–1.45)	(–4.25)	(4.30)	(–2.04)	(–3.94)	(–4.59)
CU ijt	0.216*	0.194*	0.388*	0.399*	0.002	–0.052
	(6.46)	(5.10)	(4.74)	(3.03)	(0.02)	(–0.22)
RTA ijt	0.339*	0.197*	0.411*	0.447*	0.070	0.387*
	(25.64)	(9.42)	(14.29)	(8.65)	(1.12)	(4.68)
LnArea i	0.279*	0.810*	0.083*	0.210*	0.131*	0.118*
	(6.89)	(79.02)	(2.58)	(12.73)	(6.53)	(4.89)
Border ij	1.112*	0.546*	1.346*	1.585*	0.546*	0.457**
	(37.37)	(10.19)	(21.88)	(11.91)	(2.36)	(2.13)
Coastline i	7.65×10^–6*	–1.29×10^–5*	3.62×10^–4*	6.31×10^–6*	–1.78×10^–6**	8.92×10^–5**
	(7.68)	(–17.10)	(10.11)	(6.99)	(–2.08)	(1.90)
Temp i	0.015**	–0.045*	–0.062*	0.026*	–0.008	0.025*
	(1.92)	(–5.34)	(–3.15)	(7.19)	(–1.48)	(2.11)
Lang ij	0.995*	0.644*	1.064*	0.412*	0.212**	0.353*
	(70.34)	(25.08)	(36.75)	(6.65)	(2.18)	(4.17)
Colony ij	0.835*	0.643*	0.790*	1.603*	0.772*	0.703*
	(26.31)	(15.40)	(11.57)	(11.55)	(3.80)	(3.85)
Relig ij	0.577*	0.335*	0.473*	–0.023	0.109	0.181**
	(42.63)	(13.98)	(13.48)	(–0.43)	(1.13)	(1.92)
PS it	0.209*	0.330*	0.148*	0.186*	–0.215*	0.088**
	(13.32)	(8.41)	(4.21)	(6.00)	(–4.53)	(1.90)
Life it	0.039*	0.064*	0.039*	0.015*	0.023	–0.008
	(7.46)	(3.70)	(5.97)	(2.47)	(1.34)	(–1.54)
cons	4.273*	–3.361**	7.777*	–2.855*	–7.160*	0.091
	(5.83)	(–1.97)	(8.63)	(–6.04)	(–6.20)	(0.10)
Number of observations	10 4271	27 493	18 701	93 245	25 074	16 375
R ^2	0.843	0.9039	0.8129
Year fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Origin/destination fixed effects	Yes	Yes	Yes	No	No	No
Number of instruments				201	128	176
Arellano–Bond test AR(1)				–18.384	–11.264	–14.010
				0.000	0.000	0.000
Arellano–Bond test AR(2)				–0.0974	0.945	0.337
				0.922	0.345	0.736
Notes: Origin, destination and year fixed effects are not reported. Huber–White estimator is computed to obtain robust standard errors for the OLS-FE model. Windmeijer bias-corrected estimator for the robust two-step GMM model is used. Significance at 1% (*), 5% (**) and at 10% (***) level, respectively.

The consistency of the SYS-GMM model requires autocorrelation of the first order and the lack of second-order autocorrelation. Arellano–Bond AR(1) and AR(2) report first-order and second-order autocorrelation tests, respectively. The null hypothesis is that there is no first-order/second-order autocorrelation. Results from Table 3 supports these diagnostic tests for the three samples and show the consistency of the GMM estimator.

The results presented in Table 3 support the notion that the determinants of tourism to Africa are not systematically different from factors that drive tourism to other regions. The importance of a lagged measure of tourism flows is reflected in the SYS-GMM specifications. The lagged dependent variable is positive and significant for the three sub-samples reflecting the importance of the repetition or the word-of-mouth effect. Moreover, the results show that the coefficient for African-inbound tourism is larger than for world and OECD tourism. In contrast to the earlier results of Naudé & Saayman (2005) and Khandaroo & Seetanah (2008) where the lagged variables were either negative or insignificant, we find that repeat tourism is actually of more importance for African-inbound tourism than it is for tourism flows in other areas.

As expected, GDP per capita of both the destination (LnGDPpc_it) and the origin (LnGDPpc_jt) are positive and significant across all three samples. The distance variable is also consistently negative and significant, with the African coefficient reflecting that of the other two specifications. Distance, ceteris paribus, does not have a different impact for African countries compared with other regions.

Capital investment in the tourism sector, measured here as the percentage of tourism investment over total investment, seems to have no impact for African countries compared with other regions, where it has a positive impact. This may reflect the low level of tourism infrastructure on the continent, or its inefficient use.4 An alternative explanation is that a large share of Africa's capital investment stems from the colonial era, which favoured investment in commodities that had little benefit for the tourism industry. A revealing result is the large coefficient of trade flows in explaining tourism flows for OECD countries, compared with the world and African coefficients. The positive coefficient suggests that people move to OECD countries along trade routes, perhaps as business tourism. Eilat & Einav (2004) suggest including trade, as the sum of exports and imports, in the gravity specification for tourism as a way to approximate for the intensity of the economic relationship between two countries. Moreover, tourism may either lead to an increase of domestic demand or an increase in the consumption of goods and services that are not produced in the tourist destination and as a consequence require being imported. The latter reason is a direct effect that can be illustrated by any international trade model in which consumers are allowed to consume abroad (see Santana-Gallego et al., 2010b).

The reported signs and size of coefficients for price competitiveness are robust across the different specifications, suggesting that price competitiveness is an important factor in explaining tourism flows into Africa. In contrast to the results of Naudé & Saayman (2005), of Eilat & Einav (2004) and of Crouch & Ritchie (2006), our results show that relative prices matter, especially in Africa, which supports the general belief that African tourism lags the rest of the world because of uncompetitive prices (Christie & Crompton, 2001). The positive and statistically significant coefficient of the OLS-FE shows that this result is heavily dependent on the type of estimation method used, which may explain the contradictions with the previous literature.

While the coefficients for the currency union dummy are significant in the fixed-effects specifications, the significance of both the OECD and African coefficients disappears in the SYS-GMM estimations. However, this may be because of high collinearity with the border regional trade agreement and border dummy, as is reflected when the CU dummy is not included.5 The high coefficient on the regional trade agreements dummy for African-inbound tourists suggests that tourists tend to visit countries with which their country of origin has signed a regional trade agreement. The large economic significance of the RTA dummy for Africa (in both the fixed-effects and SYS-GMM specifications) versus the small economic significance of the trade variable probably also reflects the well-known low inter-regional trade of African countries.

The variables related to geography also exhibit the same trends for African-inbound tourism as it does for tourism to the world. Land area is positive and significant while sharing a border is a strong predictor of tourism flows. While coastline, measured as the length of the coastline in kilometres, reveals conflicting results depending on the sample used, it has a significant and positive effect for the world sample and the Africa sample, but a significantly negative effect for the OECD sample. This may suggest that tourism to Europe is dominated by ‘cultural’ or ‘business’ tourists, rather than ‘sun, sand, and sea’ seekers. The positive and significant effects of the annual average temperature for the World and Africa samples also reveal the importance of the ‘sun and sand’ tourism since tourists prefer warmer destinations.

Regarding the cultural variables, results are similar across all three samples for two of the three variables. Sharing a common language and sharing a common colonial link all reveal large, positive coefficients for the world, OECD and Africa sample. The religion dummy, however, measured as one if more than 60% of the populations of the two countries share the same religion, reveals a large, positive and statistically significant coefficient for African tourism, while it is insignificant in the world and OECD sample. Tourists to Africa prefer visiting regions, ceteris paribus, which share the same religious affiliation. One mechanism why this effect may be larger for Africa is the large variation in twentieth-century European missionary activity on the continent (Gallego & Woodberry 2010). A more plausible explanation is that religion acts as a proxy for other historical or cultural connections (in addition to those historical and cultural variables already controlled for). Whatever the reason for the large coefficients, that historical and cultural linkages are strong determinants of African-inbound tourism may, potentially, have important policy implications; Fourie & Santana-Gallego (2013), for example, discuss how cultural affinity (the movement of tourists in the same direction as the flows of historical migration) and ethnic reunion (the movement of tourists in the opposite direction of historical migration flows) are significant components of tourism flows worldwide, but that these effects are particularly large for African countries. Such cultural determinants of African-inbound tourism are certainly fertile ground for future research.

Table 3 also reveals that political stability is an important determinant of tourism flows, notably in the world and Africa specifications. Life expectancy, as a measure of health and the standard of living, yields conflicting results, being positive in the world sample but insignificant in the OECD and Africa samples. The lack of consistency across samples raises doubts about the applicability of including this variable. While the Human Development Index by the United Nations is perhaps a better proxy for the development level of the destination country, given the many missing values for that variable, the sample size would be considerably reduced.

While these results reflect only small differences between African-inbound tourism and tourism to OECD countries and global tourism, our main concern is whether within-African tourism is significantly different from international tourism to Africa. Table 4 provides the results. We again use two estimation methods, fixed effects and SYS-GMM, and compare across two specifications, within-African tourism and African-inbound tourism. Note that even though the sample size falls to only 4894 in the SYS-GMM estimation of the Africa–Africa specification, the Arellano–Bond first-order and second-order autocorrelation tests again support the consistency of the SYS-GMM estimator.

Table 4: Panel estimations of the gravity equation for tourist arrivals, African sample; sample period 1995–2008

	OLS-FE		SYS-GMM
Variable	Africa–Africa	World–Africa	Africa–Africa	World–Africa
LnTou ijt – 1			0.584*	0.471*
			(15.35)	(11.13)
LnGDPpc it	0.133	0.108	0.262*	0.214*
	(1.24)	(1.54)	(3.91)	(2.74)
LnGDPpc jt	0.039	0.290*	0.096***	0.300*
	(0.49)	(3.67)	(1.80)	(3.84)
LnDist ij	–1.370*	–1.058*	–0.500*	–0.405*
	(–36.07)	(–38.16)	(–6.01)	(–4.53)
TouInv ijt	0.009	0.006	–0.017*	–0.009***
	(1.28)	(1.49)	(–3.16)	(–1.82)
LnTrade ijt	0.044*	0.032*	0.028*	0.088*
	(10.10)	(10.56)	(3.76)	(4.22)
LnComp ijt	0.226*	0.165*	–0.053	–0.403*
	(2.99)	(3.24)	(–0.66)	(–4.50)
CU ijt	–0.002	(omitted)	–0.233	(omitted)
	(–0.02)		(–1.27)
RTA ijt	0.487*	0.399*	0.381*	0.492*
	(11.02)	(10.81)	(4.01)	(3.90)
LnArea i	–0.038	0.312*	0.112*	0.121*
	(–0.46)	(6.90)	(3.22)	(3.92)
Border ij	0.887*	(omitted)	0.445**	(omitted)
	(12.44)		(2.29)
Coastline i	3.32×10^–4*	1.21×10^–4**	1.49×10^–4**	1.39×10^–4*
	(3.30)	(2.18)	(1.87)	(2.45)
Temp i	–0.218*	0.106*	0.024	0.030***
	(–4.65)	(3.91)	(1.53)	(1.87)
Lang ij	1.152*	1.014*	0.489*	0.355*
	(27.47)	(26.50)	(4.64)	(3.04)
Colony ij	0.760*	0.973*	–1.770	0.766*
	(4.27)	(14.25)	(–0.48)	(3.98)
Relig ij	0.586*	0.296*	0.322	0.185
	(9.03)	(7.17)	(2.57)	(1.49)
PS it	–0.043	0.245*	0.013	0.173*
	(–0.73)	(6.11)	(0.25)	(2.69)
Life it	0.039*	0.036*	–0.018*	0.002
	(3.06)	(5.02)	(–3.14)	(0.27)
cons	15.178*	–2.775**	2.255**	–1.129
	(8.31)	(–2.45)	(2.10)	(–0.79)
Number of observations	5 593	13 108	4 894	11 481
R ^2	0.8034	0.8504
Year fixed effects	Yes	Yes	Yes	Yes
Origin/destination fixed effects	Yes	Yes	No	No
Number of instruments			208	137
Arellano–Bond test AR(1)			–8.879	11.138
			0.000	0.000
Arellano–Bond test AR(2)			6.013	–0.9381
			0.548	0.384
Notes: Huber–White estimator is computed to obtain robust standard errors for the OLS-FE model. Windmeijer bias-corrected estimator for the robust two-step GMM model is used. Significance at 1% (*), 5% (**) and at 10% (***) level, respectively.

Considering only the SYS-GMM results, a large share of within-African tourism is repeat tourism. The notion, therefore, that within-African tourism is fundamentally different from African-inbound tourism in registering repeat visits is certainly unfounded. The GDP per capita coefficients are also positive and significant in both samples. While the GDP per capita coefficient of the origin country for within-African tourism is smaller than the coefficient for African-inbound tourism, there is no reason to suggest that poorer Africans travel, on average, to richer countries in search of job opportunities or for retail purposes. Distance, tourism infrastructure, trade flows, regional trade agreements, land area, coastline, the language dummy, and political stability for both within-African and Africa-bound tourism are similar in sign and significance. In other words, these determinants have no different impact for African tourists visiting other African countries than they have for foreign visitors to African countries.

Yet some minor differences do exist. The SYS-GMM estimation suggests that African-inbound tourists are to some extent price sensitive, while the same is not true for within-African tourism. As with the full sample, currency unions (in the presence of regional trade agreements and border effects) explain little within-African tourism while border effects explain a significant component of within-African tourism movements. The relatively poor infrastructure in Africa (with especially exorbitant air transport costs) may explain the reason why African tourists would rather choose to visit neighbouring countries (using mostly road infrastructure) than other African countries on the continent, without having to cross too many borders. Even in the absence of a land border with Europe, the large coefficient on distance and its proximity to Europe explain why three of the top four markets in Africa are located in North Africa.

Policy-makers should note the positive and statistically significant coefficient on our political stability variable for tourists travelling into Africa. For those countries targeting mostly non-African tourists, political instability may severely injure their tourism sector. In contrast, political stability is a less important indicator for within-African tourists. Countries that may have experienced recent instability, like Zimbabwe, may therefore do well to first focus on the African tourism market, instead of the lucrative but more risk-averse non-African markets, in securing new tourist arrivals.

Historical and cultural characteristics also matter for both within-African and African-inbound tourism. The large coefficient on the colonial dummy for African-inbound tourists suggests that colonial ties still explain a significant proportion of African-inbound tourism, even when controlling for language and trade flows This may suggest a greater propensity of Africans to visit friends and relatives across national borders (and across regional trade agreements), or because of similar tastes and preferences for consumer goods (Saayman & Saayman, 2012). While the coefficients for religion remain large (and statistically significant in the OLS-FE specification), they become insignificant in the SYS-GMM specification, suggesting that they should be interpreted with caution. Nevertheless, policy-makers would do well to be cognisant of these historical and cultural determinants.

5. Conclusions

Understanding the determinants of tourist arrivals to African countries is an important first step in alleviating the binding constraints that may inhibit further take-off of the fast-growing tourism industry in many African countries. This paper uses a dynamic gravity model specification to identify these determinants. The results suggest that most of the standard explanatory factors that explain global tourism are also significant in explaining African-inbound and within-African tourism. The estimation technique we use finds strong evidence that repeat tourism, in contrast to earlier evidence, is an important determinant of African tourism. Furthermore, the incomes of both the origin and destination countries, land size, partnering in a regional trade agreement and sharing a common border, language, religion or former colonial ties all increase tourist arrival to Africa, as it does for global tourism, while the greater the distance between two countries, the lower the tourism flows between them. African tourism does not ‘differ substantially’ from world or OECD tourism.

There are, however, some factors that explain tourism to Africa but do not also explain global tourism flows. Tourism infrastructure does not drive inbound-African nor within-African tourism, although it is positively correlated with OECD and global tourism flows, probably due to the low historical levels of investment on the continent. A shared religion as determinant of tourism flows is also unique to African countries, although this probably proxies for other historical and cultural linkages not already captured in the explanatory variables.

The results also show that the determinants of within-African tourism are in general not systematically different from African-inbound tourism. Political stability matter to African-inbound tourists; countries recovering from episodes of political instability should therefore focus on the African market. Historical and cultural characteristics matter both to inbound and within-African tourists: policy-makers should therefore pay more attention to the domestic tourism market, and specifically the growing visiting friends and relatives component of African tourism flows, an area that calls for greater research attention.

Acknowledgements

The authors acknowledge the financial support of Economic Research Southern Africa, who also published an earlier version of this paper as Working Paper 260.

Notes

³See Fourie (2011) for a discussion of tourism defined in the context of the service modes.

⁴We include a list of African countries in Appendix A with their share of capital investment in tourism as a proportion of total capital investment for 1995 and 2008 (Table A3).

⁵These results are available from the authors upon request.

Appendix A. Data descriptions

Table A1: Source of data

Variable	Source
LnTou ijt	The source of tourism data is the United Nations World Tourism Organisation (UN-WTO) and includes annual international arrivals by country of origin
LnGDPpc it	World Development Indicators from the World Bank. The variable used is GDP per capita (constant 2000 US$)
LnGDPpc jt
LnArea i	World Development Indicators from the World Bank. The variable used is land area (square kilometres)
LnDist ij	Centre d'Etudes Prospectives et d'Informations Internationales (CEPII) dataset
Border ij
Lang ij
Colony ij
Coastline i	Data obtained from CIA World Factbook
Temp i	Climatic Research Unit (University of East Anglia). http://www.cru.uea.ac.uk/~timm/cty/scen/TYN_CY_3_0_var-table.html. The variable used is tmp, average annual temperature in degrees Celsius
TouInv it	Calculated as the ratio of capital investment on the tourism sector of the destination country over total capital investment. Capital investment is obtained from the World Travel and Tourism Council (WTTC) database.
LnTrade ijt	Bilateral exports and imports are obtained from the International Monetary Fund Direction of Trade Statistics
CU ijt	Data obtained from Andrew K. Rose website. http://faculty.haas.berkeley.edu/arose/. Completed with data from CIA World Factbook
FTA ijt	Data obtained from the Regional Trade Agreement Database from the World Trade Organisation
LnComp ijt	World Development Indicators. The variable used is purchasing power parity conversion factor (to market exchange rate) ratio of the destination/origin country
PS	The Worldwide Governance Indicators (WGI) project produced by Kaufmann et al (2010) – political stability
Relig ij	CIA World Factbook –percentage of the population that declare a particular religion
Lnlife it	World Development Indicators. The variable used is life expectancy at birth

Table A2: Countries used in the analysis

Afghanistan	Czech Rep. (OECD)	Kyrgyzstan	Rwanda (African)
Albania	Denmark (OECD)	Laos	Samoa
Algeria (African)	Dominica	Latvia	Sao Tome & Principe (African)
Angola (African)	Dominican Rep.	Lebanon	Saudi Arabia
Argentina	Ecuador	Lesotho (African)	Senegal (African)
Armenia	Egypt (African)	Libya (African)	Seychelles (African)
Aruba	El Salvador	Lithuania	Singapore
Australia (OECD)	Eritrea (African)	Luxembourg (OECD)	Slovakia (OECD)
Austria (OECD)	Estonia (OECD)	Macao	Slovenia (OECD)
Azerbaijan	Ethiopia (African)	Macedonia	Solomon Islands
Bahamas	Fiji	Madagascar (African)	South Africa (African)
Bahrain	Finland (OECD)	Malawi (African)	Spain (OECD)
Bangladesh	France (OECD)	Malaysia	Sri Lanka
Barbados	French Polynesia	Maldives	St. Kitts & Nevis
Belarus	Gambia (African)	Mali (African)	St. Lucia
Belgium (OECD)	Georgia	Malta	St. Vincent & Grenadines
Belize	Germany (OECD)	Mauritius (African)	Sudan (African)
Benin (African)	Ghana (African)	Mexico (OECD)	Suriname
Bermuda	Greece (OECD)	Moldova	Swaziland
Bhutan	Grenada	Monaco	Sweden (OECD)
Bolivia	Guam	Morocco (African)	Switzerland (OECD)
Bosnia-Herzegovina	Guatemala	Mozambique (African)	Syria
Botswana (African)	Guiana, French	Myanmar	Tajikistan
Brazil	Guinea (African)	Namibia (African)	Tanzania (African)
Brunei	Guinea-Bissau (African)	Nepal	Thailand
Bulgaria	Haiti	Netherlands (OECD)	Togo (African)
Burkina Faso (African)	Honduras	New Caledonia	Tonga
Cambodia	Hong Kong	New Zealand (OECD)	Trinidad & Tobago
Cameroon (African)	Hungary (OECD)	Nicaragua	Tunisia (African)
Canada (OECD)	Iceland (OECD)	Niger (African)	Turkey (OECD)
Cape Verde (African)	India	Nigeria (African)	Turkmenistan
Central African Rep. (African)	Indonesia	Norway (OECD)	Uganda (African)
Chad (African)	Iran	Oman	Ukraine
Chile (OECD)	Iraq	Pakistan	United Arab Emirates
China	Ireland (OECD)	Palau	United Kingdom (OECD)
Colombia	Israel (OECD)	Panama	United States (OECD)
Comoros	Italy (OECD)	Papua New Guinea	Uruguay
Congo, Dem. Rep. (African)	Jamaica	Paraguay	Vanuatu
Congo, Rep. of (African)	Japan (OECD)	Peru	Venezuela
Costa Rica	Jordan	Philippines	Vietnam
Cote D'Ivoire (African)	Kazakhstan	Poland (OECD)	Yemen
Croatia	Kenya (African)	Portugal (OECD)	Zambia (African)
Cuba	Korea, Rep. of (OECD)	Romania	Zimbabwe (African)
Cyprus	Kuwait	Russia

Table A3: Capital investment in tourism sector as a percentage of total capital investment

Country	1995	2008
Algeria	1.7	4.1
Angola	2.7	2.5
Benin	2.0	4.4
Botswana	6.2	5.7
Burkina Faso	1.6	2.2
Cameroon	1.8	3.4
Cape Verde	3.2	26.1
Central African Rep.	1.0	4.1
Chad	1.7	5.3
Congo, Dem. Rep.	6.2	4.6
Congo, Rep. of	1.1	2.7
Cote d`Ivoire	10.0	5.9
Egypt	5.7	13.3
Eritrea	–	–
Ethiopia	5.9	10.1
Gambia, The	5.1	12.2
Ghana	3.6	4.9
Guinea	0.5	1.7
Guinea-Bissau	4.8	4.3
Kenya	11.6	9.7
Lesotho	4.1	8.0
Libya	1.7	2.3
Madagascar	4.7	8.5
Malawi	1.9	2.3
Mali	1.4	4.8
Mauritius	14.4	11.7
Morocco	6.2	12.6
Mozambique	2.0	5.3
Namibia	9.2	10.3
Niger	1.7	1.9
Nigeria	1.4	11.3
Rwanda	0.2	13.8
Sao Tome and Principe	0.8	4.1
Senegal	3.8	4.6
Seychelles	27.4	41.2
South Africa	4.5	9.3
Sudan	1.9	2.5
Tanzania	2.6	10.4
Togo	3.9	2.2
Tunisia	8.3	11.3
Uganda	2.6	5.1
Zambia	2.4	1.8
Zimbabwe	3.2	9.6