The impact of tourism on carbon dioxide emissions: insights from 95 countries

ABSTRACT

Tourism development seems to have mixed effects on the level of CO2 emissions across the globe. This study thus provides international evidence on the impacts of tourism on carbon dioxide emissions in countries of arrival. We employ alarge panel of 95 countries, consisting of three subsamples of countries classified by income level over the period 1998–2014. The theoretical framework of this study is built based on an extended version of STIRPAT model combined with the EKC. The empirical results are following: (i) tourism (receipts and number of arrivals) reduces total CO2 emissions and CO2 emission from electricity and heat production in the countries of arrival; (ii) tourism increases CO2 emissions from transport, while the number of tourist arrivals increases CO2 emissions per capita; and (iii) The effects of tourism on emissions vary across different income levels. At the global level, tourism appears to increase CO2 emissions from transportation, suggesting that special attention should be paid to supporting green transportation infrastructure technologies and practices in the tourism industry. Overall, there is room for improvement in tourism management in countries of all income levels to promote the development of low-carbon tourism products and services.

KEYWORDS:

• Tourism
• CO2 emissions
• transportation
• electricity and heat production
• panel data analysis

JEL CLASSIFICATION:

• Q53
• Q56

Notes

¹ The World Bank’s new country classifications by income level: 2018–2019 is available at: https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups

² We conducted the variance inflation factor (VIF) test and confirmed that there is no evidence of multicollinearity. The results are not presented here to conserve space, but they are available upon request.

³ In the context of PCSE for panel data analysis, the stationarity of the data series is not required. Kao (1999) showed that estimates of the structural parameter binding two independent non-stationary variables converge to zero in probability in the case of panel data, instead of being arandom variable as in the case of time series. This means that although non-stationary panel data may lead to biased standard errors, the point estimations of the value of parameters are consistent. Instead, using residual diagnostic tests, CD tests (Pesaran 2007) and stationarity tests are important for analysing the goodness of fit of the model. We applied these diagnostic checks for all the panels and confirm that there is no problem with our PCSE estimation. The results are not presented here to conserve space, but they are available upon request.

⁴ Thanks to an anonymous reviewer’s suggestion, for robustness check, we estimated models with adummy that takes the value of 1 for the years 2008 to 2014 and 0 for the rest. The results are not reported here to conserve space, but they are available upon request.