Evaluating the dynamics and impact of terrorist attacks on tourism and economic growth for Turkey

ABSTRACT

Turkey is one of the most advanced economies in the MENA region and one of the most important tourist destinations in the world, but with a long lasting history of terrorism that has significantly increased during the last years after the Syrian Civil War. This paper uses an updated sample of quarterly data on GDP, inbound tourism and terrorism to explore the relationship between tourism and economic growth adding information on terrorist activity. Different specifications of cointegrating regressions are used to quantify the impact of terrorism on the relationship between tourism demand and economic growth, including the novelty approach of a threshold cointegrating regression, where the stationary transition variable is given by a standard measure of terrorist activity. Results show that, with a delay between three to six months, and even for a relatively small number of estimated terrorist attacks, there is a negative impact of terrorism on real GDP of around 10% through a reduction of the contribution of tourism demand on economic growth.

RESUMEN

Turquía es una de las economías más avanzadas en la región MENA (Medio Oriente y Norte de África) y uno de los destinos turísticos más importantes del mundo, pero con una duradera historia de terrorismo que se ha incrementado significativamente en los últimos años tras la Guerra Civil Siria. Este trabajo utiliza una muestra actualizada de datos trimestrales del PIB, la llegada de turistas y terrorismo para explorar la relación entre turismo y crecimiento económico añadiendo información sobre la actividad terrorista. Se utilizan diferentes especificaciones de regresiones cointegradas para cuantificar el impacto del terrorismo en la relación entre la demanda turística y el crecimiento económico, incluyendo la novedosa perspectiva de un umbral de regresión cointegrada, donde la variable de transición permanente es dada por una medida estandarizada de la actividad terrorista. Los resultados muestran que, con un retraso de entre tres y seis meses, e incluso para un número relativamente pequeño de ataques terroristas estimados, hay un impacto negativo del terrorismo en el PIB real de alrededor del 10% a través de una reducción de la contribución de la demanda turística al crecimiento económico.

PALABRAS CLAVES terrorismo; crecimiento económico; llegadas de turistas; análisis de cointegración

RÉSUMÉ

La Turquie est l'un des pays les plus développés sur le plan économique dans la région du Moyen-Orient et d'Afrique du Nord (MOAN) et l'une des destinations touristiques les plus importantes du monde, mais dont la longue histoire du terrorisme a été exacerbée par la guerre civile syrienne au cours des dernières années . Cet article utilise un échantillon mis à jour des données trimestrielles sur le PIB, des arrivées des touristes et du terrorisme en vue d'explorer la relation entre le tourisme et la croissance économique vis-à-vis des informations sur les activités terroristes. Différentes spécifications des régressions de cointégration ont été utilisées pour quantifier l'impact du terrorisme sur la relation entre la demande touristique et la croissance économique, y compris la nouveauté de l’approche du seuil de la régression de cointégration, où la tension stabilisée variable s’obtient par une mesure standard de l'activité terroriste. Avec un retard de trois à six mois, et même pour un nombre relativement faible d'attaques terroristes estimés, les résultats montrent qu’il y a un impact négatif du terrorisme sur le PIB réel d’environ de 10% à cause de la réduction de la contribution de la demande touristique sur la croissance économique.

MOTS-CLÉS terrorisme; croissance économique; les arrivées de touristes; analyse de cointégration

摘要

土耳其是中东及北非地区经济最发达的国家之一，也是世界上最重要的旅游目的地之一，但是在叙利亚内战结束之后，该国恐怖主义活动加剧。本文通过恐怖主义与更新的入境旅游GDP季度数据，研究在恐怖活动影响下旅游与经济增长的关系。通过不同的协整回归规范，量化恐怖主义对旅游需求与经济增长间关系的影响，包括门限协整回归的新途径，给定一个标准的平稳转移变量以测量恐怖活动。研究结果显示，即便是规模相对较小的恐怖袭击，在其发生后的三到六个月之间，由于旅游需求下滑，恐怖活动对实际GDP的负面影响大约在10%左右。

关键词 恐怖主义； 经济增长；旅游到访人数；协整分析

KEYWORDS:

• Terrorism
• economic growth
• tourist arrivals
• outliers
• persistence
• cointegration analysis

Acknowledgements

We wish to thank two anonymous referees for helpful suggestions and comments that led to a substantial revision of the first draft version.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes on contributor

Julio A. Afonso-Rodríguez is Senior Lecturer at the Department of Applied Economics and Quantitative Methods belonging to the University of La Laguna, in Tenerife (Canary Islands, Spain) since the end of 1996. The main research interests are time series analysis (univariate and multivariate), applied macroeconometrics, and applied financial econometrics.

Notes

1. For a review of many of these studies and procedures, see the surveys by Durlauf and Quah (1999) and Durlauf, Johnson, and Temple (2005), as well as the survey by Ehrhart (2009) for the interaction between growth and inequality.

2. For more details on the literature for this phenomenon see, for example, Marin (1992), Giles and Williams (2000, 2001) and Dreger and Herzer (2013) and the references therein.

3. For a review of the literature on this question see for instance the recent surveys by Cortés-Jiménez and Pulina (2006), Brida and Pulina (2010) and Pablo-Romero and Molina (2013).

4. See Table 1 in Sandler and Enders (2008) for a list of many of these macroeconomic empirical studies of the impact of terrorism, with a description of the methodology and the main findings.

5. Differences in the empirical results on the impact of terrorism on macroeconomic variables can be explained by differences in sample of countries and time periods under analysis as well as differences in the indicators or measures of terrorist used. Frey and Luechinger (2003) provide additional possible effects of terrorism activity.

6. For the ELGH, Taban and Aktar (2012) found evidence to support the existence of both a short and long-run bidirectional causality relationship between export growth and real GDP growth for this country and for the period 1980:Q1-2007:Q2.

7. For a complete and updated summary of empirical results on causality analysis between tourism and economic growth for Turkey, see Table 1 (p. 168) in Terzi (2015).

8. The source of these data are Quarterly National Accounts from OECD statistics (https://stats.oecd.org) for real GDP and Turkish Statistical Institute (Turkstat) (http://www.turkstat.gov.tr/Start.do) for tourist arrivals.

9. In this database, maintained by the National Consortium for the Study of Terrorism and Responses to Terrorism (START) which is based at the University of Maryland, terrorism is defined as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation”. (https://www.start.umd.edu/gtd)

10. In the basic case, that is, in absence of deterministic component d_t= 0, the IV-type estimator is based on the raw observations x_t= η_t, while that OLS detrended observations are used if d_t≠ 0. Originally, this estimator, and the corresponding statistics for testing the null hypothesis of a unit root, is proposed to deal with serially correlated error terms ε_t being generated by a moving average (MA) process of order q₀ ≥ 1 as , with e_{t ∼} iid(0, ). For any chosen lag q ≥ q₀, these test statistics are pivotal with limiting null distributions free of nuisance parameters.

11. Also, following Hansen and Lunde (2014), it can be shown that the unit root tests based on these IV estimators of the first-order serial correlation coefficient α are also robust to measurement errors or aberrant observations in the series.

12. As has been recommended by a referee, the results for the DF tests should not be reported given the need to account for serial correlation in the error terms evidenced by the MAIC criteria in computing the ADF test. This is due to the fact that the true limiting null distribution of the DF statistics has changed and the outcomes of these tests are not longer valid using the usual critical values. However, given that the estimated values of these statistics clearly point in the direction of stationarity, this evidence (being strictly invalid) could still be useful.

13. Although Vogelsang’s procedure has the right size under the null hypothesis of no single outlier, Perron and Rodríguez (2003) have shown that when applied in an iterative fashion to select multiple outliers, this test exhibits serious size distortion as an excessive number of outliers will be detected. To avoid this possibility, we use the appropriate limiting distribution of the test at each step of the iterative search obtained by Perron and Rodríguez (2003) and also the modified version proposed by these authors (Panel B of Table 3). It should also be noted that even when the Vogelsang’s test is used to detect a single outlier, Haldrup and Sansó (2008) have shown that this procedure is generally inconsistent, with very low power unless the outliers are huge. However, despite the possible unreliable outcomes resulting from this procedure, the results shown in Table 3 indicate close agreement between the two procedures.

14. See footnote 12 for the justification of the inclusion of the results based on the DF test statistics, despite the apparent requirement of serial correlation adjustment.

15. For testing the null hypothesis of stationarity against the alternative of a unit root, in addition to the KPSS test, we also use the MLH₁ test proposed by McCabe et al. (2006), which have the advantage that its limiting null distribution is standard normal and that it is invariant to the particular specification of the deterministic component.

16. Particularly, under exogeneity of a subset of stationary regressors, it can be shown that the no stationary parameters can be consistently estimated by OLS under cointegration with the same limiting distribution as if the stationary regressors are omitted from the estimated regression. As is well known, this distribution is only useful for constructing pivotal test statistics under the additional condition of strict exogeneity of the integrated regressors. In such a case, hypothesis testing for these parameters is standard.