High skills, high growth: Is tourism an exception?

Abstract

Despite the emphasis placed by growth models on technological progress, recent empirical evidence shows that tourism, a sector widely regarded as low-skill/low-tech and one of the fastest growing industries in the world, may offer a favorable strategy for growth. In addition, in this tourism-led growth literature it is not clear whether human capital plays a role. Using a panel of 72 countries (1980–2005) this study shed new light on the effect of tourism and human capital for economic growth. While our results confirm that the tourism sector indicator is always positive and significant in growth regressions they also show that increased education contributes to growth and that the role of the tourism sector is significantly larger in countries with higher aggregate levels of human capital. Our main results are robust to the inclusion of additional variables, the use of alternative estimators in the regression analysis and the use of different sub-samples. Overall, our results suggest that an increase in human capital endowments is always beneficial, even when the development strategy focuses on the expansion of a (successful) unskilled sector.

Keywords:

• economic development
• tourism
• human capital
• services sector
• human resources

JEL Classifications:

• I21
• O15
• O40
• O57

Notes

 1. Non positive results have been found by Di Pietro and Figini (2007), where specialization in tourism is never significant in growth regressions and Figini and Vici (2010) who found a positive relationship between tourism and growth during the 1980s but not afterwards.

 2. Sequeira and Nunes (2008) use the Blundell and Bond (1998) estimator, while Arezki, Cherif and Piotrowski (2009) suggest an instrument to control for the endogeneity of tourism specialization in growth regressions based on the UNESCO World Heritage List.

 3. For example, Blake et al. (2008) provide an economy-wide analysis of the distributional effects of tourism expansion focusing on weather and how this industry contributes to poverty relief.

 4. Among the most recent, see for example Figini and Vici (2010), Brau, Lanza, and Pigliaru (2007) and Sequeira and Nunes (2008).

 5. Two exceptions are present during the robustness analysis, when FDI and agriculture are included among controls. See Section 5.

 6. Problems arise since tourism specialized countries are often small or very small (islands) economies for which data are likely to be missing. Therefore, including/excluding some indicator or increasing/reducing years form the analysis implies including/excluding these specific countries from the sample and may possibly affect results on the tourism coefficient that represents the main focus of these studies.

 7. On this, see Jayawardena (2002) where there is evidence and abundance of data on the most tourism dependent area of the world, the Caribbean region, where, at the time the book was written, tourism earnings accounted for approximately 25% of the region's GDP and provided employment for 25% of the region's labor market.

 8. In particular, how the tourism product is marketed and the ways in which plants are operated. See for example Crick (2002).

 9. Two alternative indicators used to capture the specialization in tourism are the number of international tourist arrivals over population, and the numbers of establishment and bed places. However, these are not appropriate indicators for empirical macro-growth analysis.

10. On this, see Jayawardena (2002) for evidence on the Caribbean countries and Jonckers (2005) for Europe.

11. ILO Bureau of Statistics and the UNWTO Department of Statistics and Tourism Satellite Account (2008).

12. See for example Wood (1997) for UK and Unioncamere – Ministero del Lavoro (2008) for Italy.

13. See http://www.euklems.net/

14. More on endogeneity of the tourism variable in Sections 4 and 5.

15. Other studies on tourism that use this approach are Fayissa, Nsiah and Tadasse (2008), Cortés-Jiménez (2008), and Proença and Soukiazis (2008).

16. Within the empirical growth-convergence literature equation (1) represents a conditional convergence model. For more on this, see Durlauf, Johnson and Temple (2005).

17. We use Barro and Lee (2000) data on education attained by the total population aged 15 and over. See the Appendix for more details.

18. Quoting Solow (1999, 663) ‘Human capital is widely agreed to be an important factor in economic growth…. One difficulty is that the measurement of human capital is insecure’. For more on this see also Di Liberto (2007).

19. Unlike enrolment rates, for which we can also find annual data at cross country level, human capital stocks indicators are only available every five years (Barro and Lee 2000) or 10 years (see Cohen and Soto 2007; Morrisson and Murtin 2009).

20. On this, see for example Cellini (1997).

21. On this, see the extensive Durlauf, Johnson and Temple (2005) survey.

22. While micro panels are characterized by a large number of individuals and short T (assumed as fixed in asymptotics), macro panels have small N and a large T (not assumed fixed in asymptotics).

23. See Sequeira and Nunes (2008) for the tourism case.

24. Note that some feature of the Hauk and Wacziarg (2009) study make it open to criticisms and imply that their results are not fully comparable with that found in other similar studies. In particular, unlike all other studies (see footnote 25 below) in their Monte Carlo analysis they (i) do not consider the Kiviet estimator and (ii) do not compare results in terms of root mean square errors performances but only in terms of bias. The latter point is important since the between estimator is well known to be the less efficient estimator among the alternatives.

25. See Kiviet (1995); Judson and Owen (1999); Everaert and Pozzi (2007). In particular, these Monte Carlo studies find that the KIVIET and Anderson-Hsiao estimators consistently outperform alternative estimators in most cases but suggest using the KIVIET estimator for smaller panels, while Anderson-Hsiao should be preferred for large panels, as the efficiency of the latter improves with T.

26. The analysis is performed assuming a bias correction up to order O(1/T) and Arellano and Bond as consistent estimator in the first step. Results are not sensitive to the use of alternative options. Standard errors are calculated through bootstrapping (500 replications).

27. We have also replicated this analysis on different model specifications obtaining very similar results. These results are available upon request. However, it is well known that GMM-AB (and GMM-SYS) does not perform well when samples are characterized by small T (as in our case). See Bun and Windmeijer (2009).

28. We only briefly mention here that United Nation World Tourism Organization (UNWTO) updated the tourism data in 2004 asking countries to revise their figures in order to match them with the new standards only from 1995 onwards. For more details on this see Figini and Vici (2010).

29. Although we agree with authors concerns about the quality of this dataset. In general, data quality and measurement error represent a well-known problem in cross country growth analysis. See Durlauf, Johnson and Temple (2005).

30. In this study, the largest sample includes 166 countries, while the smallest one includes only 102 countries. Even if the numerosity of each sample depends on the set of variables included, there is a systematic difference of 25–30 observations between regressions performed over the 1990–2005 period and those that cover the longest period 1980–2005.

31. Even if its use is not appropriate in growth frameworks like ours, we have also estimated the model using a Random Effect estimator that take into account of both the within and the between variation in data and obtained a positive coefficient on tourism.

32. The Solow growth model implies that the speed of convergence, λ, can be calculated using . See Barro and Sala-i-Martin (1992); Islam (1995).

33. Such as in Eugenio-Martin, Morales and Scarpa (2004), Sequeira and Nunes (2008) and Figini and Vici (2010).

34. On this see Mulligan and Sala-i-Martin (2000).

35. See also Romer (1990) and Aghion and Howitt (1998).

36. See, for example, Wooldridge (2003) for who with interaction terms one should take care not to look separately at the estimates of single coefficients (see p. 195). In general, problems arise because of (among other factors) likely multicollinearity.

37. We thank an anonymous referee for this suggestion.

38. Durlauf, Johnson and Temple (2005) list 145 variables which have been found to be statistically significant in different studies.

39. These results can be found in the working paper version of the study. See Di Liberto (2010). In particular, our tourism indicator could capture the effect of exports that are recognized in both the theoretical and empirical literature on growth as one of its most important determinants since tourism is an export industry: foreign visitors who travel to a country purchase a service, the tourism experience, of that country.

40. Additional evidence not included here show that this result holds true even when we use lagged values of our investment indicator. However, in this case investments were never significant. See Di Liberto (2010).

41. See, for example, Figini and Vici (2010) and Sequeira and Nunes (2008).

42. There is an extensive literature that investigates the role of FDI on growth processes. See Alfaro et al. (2004).

43. All these regressions have been also performed including our investment share variable among regressors. Results do not change significantly. The same conclusions arise when we perform a regression including all the additional controls. These results are available upon request.

44. While missing observations in some t (mainly the earlier observations) have been found for El Salvador, Hong Kong, Singapore and Peru.

45. As expected, the correlation coefficient between these two variable is high, −0.74. In general, as noted by Durlauf, Johnson and Temple (2005) and Krueger and Lindhal (2001), the absence of a significant relationship between growth and other variables in many studies may be due to the model specification and the use of a parsimonious specification, as we have adopted in Table 6, is therefore preferable.

46. We thank a referee for suggesting this additional control.

47. Conversely, Sequeira and Nunes (2008) find a positive role for tourism both for the large heterogeneous sample and for a sub-sample of less developed economies.

48. Details on the World Bank country classification can be found in Table 2. When countries have changed their classification during the period, we have considered the prevalent (longest period) assignment/classification.

49. Additional evidence not provided here shows a non homogeneous role of education within the period covered by our analysis. In particular, our results seem to suggest that the secondary school indicator is highly significant only in the (1990–2005) more recent dataset while it is not during the 1980s. These results are available upon request.

50. See ILO, International Labour Office, (1997).

51. Note that models 2a/2b and 3a/3b show an overall decrease in coefficients significance levels. However, we claim that this is explained by the sample (and variance and so efficiency) decline caused by the smaller panel dimension involved in this part of the analysis (both in the T and N dimensions).