Hallyu tourism: The effects of broadcast and music

Introduction

The inbound tourists of South Korea (hereafter Korea) increased from 3.75 million in 1995 to 17.50 million in 2019. Such growth, according to researchers, such as Kim et al. (2007), was caused by the renowned Korean drama ‘Winter Sonata’ that was aired in Korea in 2002. Bae et al. (2017) state that a 2004 survey done by the Korea Tourism Organisation summarised 47% of the respondents visited Korea because of the influence of Korean dramas. Korean dramas are part of Hallyu, Korean popular culture consisting of dramas, popular music, movies, fashion, games and others. The influence of Hallyu begins with the spread of Korean dramas, but is reinforced subsequently by other Korean entertainment products, such as popular music and variety shows (Bae et al., 2017). Bang et al. (2021) state that Korean popular music, K-pop, has spread to even more countries than Korean dramas.

The potential positive impacts of Hallyu on Korean tourism industry, Hallyu tourism, have attracted the attention of more and more researchers, for example, the most recent work by Lee and How (2021). Lee and How (2021) explain Hallyu tourism as the interdependency of film tourism, celebrity-induced tourism, food tourism and culture tourism, caused by the spread of Hallyu. However, limited to our knowledge, only three studies, Bae et al. (2017), Lim and Giouvris (2020) and Lee and How (2021), formally test the effects of Hallyu on Korean tourism industry. But, these studies do not analyse the impacts of different Hallyu contents on inbound tourists.

This study aims to contribute to the literature, particularly to fill the gap identified above. It examines the effects of two main Hallyu contents: broadcast, inclusive of dramas and variety shows, and popular music on Korean tourism industry with disaggregated time series data. To identify the long-run causal effects of these two Hallyu contents on Korea’s inbound tourism, an appropriate cointegration analysis under small sample size constraint is conducted. Granger causality test is not performed in this study because this test does not show causal relationship and it only suggests a variable improves the forecast of another variable, if there is an existence of Granger causality (Seddighi, 2012: 339).

Data, methodology and results

Hallyu is intangible popular culture of Korea. The influence of Hallyu leads to the consumption of Hallyu-related consumer goods and contents. Therefore, consumption indicator can be used to measure the spread of Hallyu outside Korea. The exports of Korean content industry can be a proxy for Hallyu. However, we have to recognise that the exports of Korean content industry variable are not a perfect proxy for Hallyu because content piracy reduces the size of such exports and such exports cannot fully capture all intangible dimensions of Hallyu. Accepting these limitations, this study uses broadcast exports of the content industry to capture the Hallyu effect of Korean dramas and variety shows. The broadcast exports of Korean content industry comprise broadcasting channel and terrestrial operations of finished broadcast programmes, overseas Korean broadcasting support, video/DVD sales and others. The genres of the broadcast exports include drama, documentary, entertainment and others. The music exports of the content industry are used to represent the Hallyu effect of K-pop. The music exports of the content industry consist of popular music. The inbound tourism of Korea will be proxied by total visitor arrivals to Korea. These data are downloaded from CEIC. They are available in annual frequency from 2004 to 2018. Table 1 summarises these variables. Natural logarithm is applied on each of these variables to reduce its variability.

OPEN IN VIEWER

Table 1.

Summary of the variables.

Variable	Notation	Source
Broadcast exports of the content industry (in thousand USD)	BC	Korea Creative Content Agency
Music exports of the content industry (in thousand USD)	MC	Korea Creative Content Agency
Total visitor arrivals	VA	Korea Tourism Organisation

This study has a sample with only 15 observations. Therefore, only the bounds test developed by Pesaran et al. (2001) is suitable because this cointegration test has good small sample size properties (Pesaran and Shin, 1999). This cointegration test can be utilised for as long as the order of integration of each variable is at most equal to 1. Both the Dickey and Fuller (1979, 1981) test and Kwiatkowski et al. (1992) test indicate that the natural logarithm of each variable is integrated of order 1.

In this study, the impacts of the two identified aspects of Hallyu on the inbound tourism of Korea are expressed as

ln ( VA t ) = θ 0 + θ 1 ⁡ ln ( BC t ) + θ 2 ⁡ ln ( MC t ) + u t

(1)

ln ( VA t ) = β 0 + ∑ i = 1 1 β 1 i ⁡ ln ( VA t − i ) + ∑ i = 0 m β 2 i ⁡ ln ( BC t − i ) + ∑ i = 0 n β 3 i ⁡ ln ( MC t − i ) + ε t

(2)

Δ ⁡ ln ( VA t ) = α 0 + ∑ i = 1 l − 1 α 1 i Δ ⁡ ln ( VA t − i ) + ∑ i = 0 m − 1 α 2 i Δ ⁡ ln ( BC t − i ) + ∑ i = 0 n − 1 α 3 i Δ ⁡ ln ( MC t − i ) + δ 1 ln ( VA t − 1 ) + δ 2 ln ( BC t − 1 ) + δ 3 ln ( MC t − 1 ) + υ t

(3)

The null hypothesis of this test is δ 1 = δ 2 = δ 3 = 0 , indicating there is no cointegration. If the null is rejected, the derived ARDL will be reparameterised to obtain the corresponding cointegration regression as stated by equation (1).

This study builds models using the specific-to-general approach that focuses on a few key variables because a cointegrating relationship is invariant to different model extensions (Herzer et al., 2012; Lutkepohl, 2007). This means that if the included few variables are cointegrated, the found cointegrating relationship is robust against the inclusion of additional variables. Lutkepohl (2007) also emphasises that such approach is easier to identify a cointegrating relationship when the sample size is small. Therefore, using ln(VA) as the dependent variable, ln(BC) is included as the only independent variable (this model is reported as Model 1 in Table 2); then ln(MC) is added as the only independent variable (this model is reported as Model 2 in Table 2); finally, both ln(BC) and ln(MC) are included as the independent variables (this model is stated as Model 3 in Table 2). Model 3 is equation (1). Model 1 and Model 2 are nested in equation (1).

OPEN IN VIEWER

Table 2.

The results of the ARDL models, the bounds test and diagnostic tests.

	Model 1	Model 2	Model 3
l	1	2	1
m	0		0
n		0	0
F-statistic	9.6012**	9.8203**	13.9461***
JB	0.6664	0.5684	0.5304
BG	0.7357	1.0087	1.3018
ARCH	0.9702	3.4491	4.3345
RESET	1.5458	17.8229***	0.3293

The critical values provided by Narayan (2005), Pesaran et al. (2001) and Tang and Abosedra (2014) are used. The critical values provided by Pesaran et al. (2001), Narayan (2005), and Tang and Abosedra (2014) are based upon 30 observations, 1000 observations and 16 observations, respectively. At the 5% significance level, Model 1 and Model 2 reject the null of no cointegration for all sets of critical values. Model 1 passes all diagnostic tests. Model 2 fails Ramsey RESET test, suggesting that the derived model is mis-specified. The results of the corresponding cointegrating regressions for Model 1 and Model 2 are Model 1# and Model 2#, respectively, reported in Table 3. The only independent variable of Model 1# or Model 2# is statistically significant at the 1% level.

OPEN IN VIEWER

Table 3.

The results of the cointegrating regressions.

Variable	Model 1#	Model 2#	Model 3#
Intercept	6.0019*** (0.8094)	13.5095*** (0.3053)	8.8699*** (0.9164)
ln(BC)	0.8189*** (0.0655)		0.4855*** (0.0996)
ln(MC)		0.2329*** (0.0252)	0.1061*** (0.0299)

Model 3 also passes all diagnostic tests as shown by Table 2. Note that Tang and Abosedra (2014) generate critical values for only 1 independent variable. Because at a given significance level, when the number of independent variables increases, critical values will become smaller, the critical values of Tang and Abosedra (2014) are used as a guide for the case with 2 independent variables. At the 1% level, because of the large F-statistic, 13.9461, Model 3 rejects the null of no cointegration. The corresponding cointegrating regression is Model # reported in Table 3. Both independent variables are individually significant at the 1% level. Since the estimated models reported in Table 3 are log–log model and the estimated coefficient of each independent variable is positive and less than 1, the effect of BC or MC on VA is positive but becomes smaller when BC or MC increases. In Model 3#, given that the estimated coefficient of ln(BC) is larger than the estimated coefficient of ln(MC), and both BC and MC have the same unit of measurement, BC has a larger effect on VA than MC.

Conclusion

The main findings illustrate that broadcast has a stronger influence on the demand for inbound tourism of Korea than music. Korean drama and variety shows generally feature locations. Kim et al. (2007) emphasise that the featured locations experience an improvement in the destination image in comparison to those which are not featured. The findings of this study suggest that Korean government should encourage the producers of dramas and variety shows to feature different cities to improve their images as potential tourist attractions. This helps to strengthen the position of Korea as one of the main tourism destinations. This also implies that the negative exposures of a city, such as criminal activities, must be controlled to avoid jeopardising the image of that city as a tourist attraction. Other countries which intend to use their popular culture to expand their tourism industry can learn from the experience of Korea. Finally, this study is limited by the sample size. Future study can consider the effects of additional variables on inbound tourism when sample size is sufficiently large.