Cultural Preferences in International Trade: Evidence from the Globalization of Korean Pop Culture ∗

Korean pop culture (soap operas and K-pop music) has grown immensely popular across the globe over the past 15 years. This paper analyzes its impacts on international trade. We compile cross-country panel data on South Korea’s TV show exports to 136 countries for the period 2001– 2014. These data, when combined with Korea’s exports of goods (HS 4-digit level), exhibit large variations across years, countries, and products. We demonstrate that more exposure to Korean culture increases foreign consumers’ demand for Korean goods, and the eﬀect diﬀers across types of products. First, given the fact that women spend more time consuming Korean soap operas and music, we show that higher Korean TV show exports signiﬁcantly increase female visitors to Korea and its exports of women’s clothing and cosmetics, while the eﬀects are smaller on male visitors and men’s clothing. This provides strong evidence for the demand-side preference mechanism, because supply-side factors can hardly generate such gender bias within the same product category. Furthermore, we ﬁnd that Korean TV show exports have signiﬁcantly positive eﬀects on consumer goods, but no eﬀects on capital or intermediate goods. Together, these ﬁndings highlight the role of cultural ﬂows on consumer preferences in international trade.


Introduction
"The booming South Korean presence on television and in the movies has spurred Asians to buy up South Korean goods and to travel to South Korea, traditionally not a popular tourist destination. The images that Asians traditionally have associated with the country -violent student marches, the demilitarized zone, division -have given way to trendy entertainers... " -The New York Times, 2005 (by Norimitsu Onishi) Are cultural preferences important in international trade flows? It is difficult to systematically identify such effects because most cultural variables -language, ethnicity, and religion, for example -are strongly correlated with geographical factors, as well as communication and information costs. This paper overcomes this difficulty by using the phenomenon that the Korean pop culture has rapidly spread to many other countries in the last two decades, and demonstrates that this leads to increased exports of South Korea by changing foreign consumers' preferences for Korean products. Because the measure of cultural flows in our study varies substantially over time and across countries, and because its effects on consumers preferences are not uniform across products, these rich variations allow us to identify the cultural effects on trade. To our knowledge, this is the first paper in the trade literature that demonstrates the causality in which cultural shocks of a clear source affect international trade through the demand channel.
Over the last fifteen years, the Korean popular culture, especially soap operas (television dramas) and K-pop music, has become immensely popular across the globe. This phenomenon is called "Korean Wave" (or "Han-Ryu" in Chinese), a term that was coined by the Chinese media around Le Figaro reported with the headlines "Korean Wave Reaches Europe" and "Korean Wave Hits Zénith." Although K-pop is better known in the western world, it is the Korean soap operas that initially lead the wave in Asia and in many countries in the Middle East and South America. The Korean wave first began in China in 1997 with the drama series What is Love All About?. The soap opera recorded a 15% audience share, meaning that over 150 million Chinese watched it. 1 Meanwhile, the wave arrived in Japan in 2003, with another Korean drama series Winter Sonata, which recorded a sensational audience rating of 22.5%. Koreans were pleasantly surprised by this, because previously in Japan, Korean cultural contents had hardly received any attention despite the geographical proximity. The popularity has since grown dramatically in both countries, and nowadays, the Korean culture has pervaded the everyday life of the people in these countries. 2 Interestingly, the Korean culture has become even more popular in farther Asian/Central Asian 1 The audience ratings of Korean dramas are documented by numerous news articles and various reports by South Korea's Ministry of Foreign Affairs and the Korea Foundation for International Culture Exchange.
2 For example, another Korean drama series, Descendants of the Sun, has become extremely popular in China, hitting more than 2.6 billion viewership within the two months starting February 2016 (the sum of 16 episodes' viewership, reported by the Chinese online video platfrom, Iqiyi). Similarly, it is no longer surprising to see a K-pop song ranked at the top of the most popular music charts in Japan.
The Korean wave has also reached the Americas as well as the Middle East and Europe in various years during 2007-2012. 3 In Ecuador, for example, the wave started in 2009 with a phenomenal 55% audience rating. More surprisingly, in Cuba, yet another two drama series recorded more than 80% audience ratings in 2012-2013. 4 As another example, in the Middle East, Islamic Republic of Iran Braodcasting (IRIB) reports that Jewel in the Palace was ranked as the most popular drama from March to April in 2007, with a 57% audience rating and 97% satisfaction.
The spread of the Korean wave, as suggested above, surpasses physical distance, language barriers, ethnicity and religion differences (all of them the conventional measures of cultural affinity) and thus offers a unique source of cultural shocks that we can exploit to study its effects on consumer preferences and demand behaviors. In this study, we construct two unique measures of exposures to the Korean popular culture in a destination and year. First, we compile South Korea's TV show exports to over 150 countries for the period of 1998-2014. This measure exhibits significant variations across countries and over time. Second, we construct a popularity index, where countries are classified into five levels of affinity for the Korean culture as of 2015. This is based on our own reading of various reports prepared by the Korean government agencies on the Korean wave situation in each country. For example, one of the criteria for a country to be classified as Level 5 (where the Korean culture is very popular) is: 'The Korean dramas and K-pop have become so popular that the local government publicly expresses concerns and tries to restrict the inflow of the Korean cultural contents.' We find that the countries assigned with a Level-5 index constitute more than 30% of the world population.
To identify the effects of the Korean wave on foreign consumer preferences, we adopt two strategies. First, we use industry-specific trends in Korean exports to control for any industryspecific supply shocks (at the HS 4-digit level). This reduces the concern of omitted variable bias if the Korean wave is correlated with the supply-side shocks such as improvement in the Korean product quality. It also helps control for general shifts in the Korean trade structure. For example, South Korea's industrial specialization has moved upstream, with downstream tasks being relocated overseas in the past two decades. As a result, the Korean exports of durable and semidurable consumer goods have decreased, while its exports of intermediate and capital goods have increased dramatically. In contrast, if the Korean wave is to affect the Korean exports via changes in foreign consumer preferences, we would expect such effects to be stronger on consumer products than capital/intermediate goods. Thus, it is important to control for these confounding supply-side factors. Second, we address any remaining concerns that the Korean wave and the Korean exports may be determined by some unobserved common factors by showing that the trade effects of the Korean wave are systematically heterogeneous across types of goods. Such heterogeneity cannot be explained by plausible alternative hypotheses (common factors), but is perfectly consistent with the demand-driven hypothesis.
First, we find that more TV show exports significantly increase foreign female visitors to Korea and Korea's exports of women's clothing, while the effects are much smaller for men. It is well known that women are stronger fans for the Korean soap operas and K-pop music. 5 This result strongly supports the demand-side preference mechanism, because supply-side factors can hardly generate such gender bias within finely disaggregated product/service categories. The most striking result appears in cosmetics, an iconic women's product. We find that doubling the Korean TV show exports increases Korean exports of cosmetic products by 40% (the total TV show exports have grown about tenfold during [2002][2003][2004][2005][2006][2007][2008][2009][2010][2011][2012][2013][2014]. Further, we find that doubling the TV show exports increases exports of consumer goods by 16%, while it has no effects on capital or intermediate goods. When we refine the set of consumer goods to those that are identified by surveys to have been heavily influenced by the Korean wave, we find that the effect on these "Korean wave goods" is even larger at 26%. That the result with the more specific indicator is stronger is consistent with the proposed preference mechanism and increases our confidence in the estimation results. Lastly, we note that there exist significantly positive Korean-wave effects even for foods and clothes, which are not often advertised through the mass media in foreign countries, suggesting the power of preference diffusion. Relatedly, the Korean TV show exports are found to have promoted Korean outward FDI, and such effects are significantly larger in sectors that embody stronger country-oforigin symbolic values (such as restaurants, entertainment, grocery stores, hair salons and aesthetic clinics) than the other sectors. Collectively, the results suggest a strong causal effect of the Korean wave on foreign consumer preferences and demand behaviors.
The rest of the paper is organized as follows. In Section 2, we discuss related literatures.
Section 3 documents measurement of the Korean wave. Section 4 proposes plausible mechanisms through which the global spread of Korean pop culture affects Korea's exports of goods and services.
In Section 5, we build a simple model and derive the estimation frameworks. Section 6 describes the estimation strategies and results. Section 7 concludes.

Related Literature
The current studies on the Korean wave are mostly conducted by the Korean government agencies or foundations (MOFA, 2015;KOTRA, 2011;KOTRA and KOFICE, 2015;KOFICE, 2015a,b) in the form of summary reports. There also exist economic analysis on the subject by Korean researchers, but these studies typically cannot identify robust evidence on the causal effects of the Korean wave on aggregate bilateral exports. See for example, Park andChoe (2008, 2009) and Choi (2012). Indeed, as we will demonstrate with our data below, one cannot find robust effect estimates of the Korean wave by exploiting only country and time variations. The product variation (in addition to the country-time variation) is the key to the identification.
In using the Korean TV show exports as a measure of cultural influence, our study is related to the literature that studies the economic and social impacts of the media. As the survey by DellaVigna and Ferrara (2015) indicates, exposure to the media contents could lead to imitative behaviors. For example, Disdier et al. (2010a) show that exposure to foreign media contents affect French parents' naming pattern through selective imitation. Jensen and Oster (2009) demonstrate that introduction of cable television is associated with increases in women's autonomy, and decreases in fertility and son preferences in rural India. Similarly, Chong and Ferrara (2009) find that the availability of the TV signals significantly raised the divorce rate in Brazil. 6 On the other hand, Bursztyn and Cantoni (2016) find that exposure to Western TV signals in the former East Germany affects the composition of consumption biased in favor of categories of goods with a high intensity of pre-reunification advertisement. In our study, we show that both mechanisms (preference diffusion and advertisement) play important roles in the Korean wave's influence on consumption behaviors.
The paper is closely related to the literature that studies the effect of cultural affinity on trade.
For example, Guiso et al. (2009) show that lower trust reduces bilateral trade across countries.
Similarly, Felbermayr and Toubal (2010), Disdier and Mayer (2007), and Disdier et al. (2010b) construct alternative proxies for cultural proximity and document their positive effects on trade volumes. In general, this literature does not differentiate whether the cultural proximity variable affects trade through lower trade costs (on the supply-side) or stronger preferences (on the demandside). On the other hand, there exist several papers that focus on the supply-side. For example, Melitz (2008), Melitz and Toubal (2014), and Egger and Lassmann (2015) study the role of common language in lowering communication costs. In this paper, we focus on the demand-side changes in consumer preferences due to the Korean wave. As indicated by the discussions in the introduction, the reach of the Korean wave is not systematically related to any conventional cultural proximity measures (such as language, religion, ethnicity, or physical distance). 7 Thus, it offers a distinct source of cultural shocks that vary across a long period of time and a large sample of countries.
The heterogeneous pattern of the trade effects found across products is consistent with systematic heterogeneous responses in foreign consumer preferences to the Korean wave, but difficult to reconcile with supply-side shocks. This paper is also related to the growing literature on the role of information in trade. To illustrate, Rauch and Trindade (2002) and Wagner et al. (2002) find that ethnic networks facilitate information flow and help match foreign buyers and sellers, which promotes international trade by 6 See also Kearney and Levine (2015), Olken (2009), andFerrara et al. (2012) for the effect of media on teen pregnancy, social capital, and fertility rates, respectively. 7 In fact, the Korean language is considered as a "language isolate," sharing no genealogical relationship with other languages (see, e.g., Ethonologue). lowering search costs. Cristea (2011) shows that high-quality information flow facilitated by inperson business meetings increases international trade. Allen (2014) demonstrates that information frictions are quantitatively important in regional agricultural trade in the Philippines. Like the literature on communication costs discussed above, this literature has predominantly focused on the supply-side. Our work contributes to the literature by showing that cultural information flows on the demand-side can have substantial impacts on international trade as well.

Korean Wave and Data
In this section, we document the measurement of the Korean wave. Details on the trade data and the other control variables used in the analysis can be found in the Appendix.

South Korea's TV program exports
We compile a cross-country panel dataset of South Korea's TV program exports for the period of 1998-2014. The data are sourced from the "Annual Report on the Actual Condition of Korean Broadcasting Industry" by Korea Communications Commissions, a government agency. The report is published annually since year 2001, but with data dating back to year 1998. This publication reports South Korea's TV program exports (excluding exports for overseas Koreans) to over 150 countries for each genre (drama, reality shows, music, documentary, sports, movies, etc.). The publication also separately reports the aggregate TV program exports for the overseas Korean across the world. We allocate the sum across countries by the proportion of overseas Koreans residing in a country, and incorporate it in the total TV show export to each country. Note that the lump-sum export value for the overseas Korean is trivial, at less than 0.5% compared to the total TV show exports (as of 2014). Figure 1 illustrates the trend of the aggregate Korean TV program exports. It has grown substantially from US$8 to US$336 million during the period 1997-2014. 8 On the other hand, the corresponding imports only increased slightly from US$57 to US$64 million. Figure 2 shows that the genre composition is highly biased toward dramas, which account for 90% of the total TV show exports on average during 2010-2013, followed by reality shows at 5%. Figure 3 illustrates the TV program exports to a subset of destinations for each year in 2001, 2004, and 2014. We see exponential increases of Korean TV show exports to Japan, China, Singapore, Taiwan, Vietnam, Thailand, etc. In contrast, the TV show export values stay close to zero in countries such as India, Russia, France, and the UK. Note that the official export values likely have under-represented the actual extent of culture exposure, because many people watch the Korean TV shows using the internet. 9 Nonetheless, the TV show export data provide an excellent measure of the Korean wave, as it exhibits substantial variations across years and countries.
The global spread of Korean pop music is another key feature of the Korean wave. As shown in Figure 4, the export value of K-pop has grown dramatically from US$22 million to US$381 million during 2005-2015, while music imports by South Korea increased only from US$8 million to US$14 million. Many countries' enthusiasm for the Korean culture started with its soap operas, before spilling over to the K-pop music. Accordingly, Figure 4 shows that the exports of the K-pop music started increasing dramatically only around 2008, several years after the success of Korean TV dramas overseas. Unfortunately, data on bilateral exports of K-pop music does not exist, so we cannot incorporate it systematically in our analysis. However, it is likely that Korean TV shows have caused greater impacts on Korea's merchandise exports than K-pop music, because viewers of Korean TV shows spend an extended period of time experiencing indirectly a large array of the Korean culture (e.g., the fashion, foods, and electronic products used by the characters in the TV shows). The exposure to the Korean culture is thus likely richer through the contents of soap operas than music clips, leading to potentially stronger affinity for Korean goods and services.

Popularity Index
In addition to the overseas sales of the Korean TV programs, we construct an alternative measure that focuses on how intensely and extensively people in a country follow the Korean pop culture.
To do this, we identify the criteria and classify countries into five levels of popularity with which the Korean pop culture is received in the country. We use the following Korean government agency reports that document the situation of the Korean wave in more than 100 countries. The documents  shows or K-pop music, and not aware of the Korean wave phenomenon. A total of 47 countries belong to this category: including India, Pakistan, Switzerland, Austria, Sweden, the UK, Germany, and Spain.
Level 2 (Recognized): The K-pop music (or Korean soap operas) are very popular among a small fraction of the population, to the degree that the national media give major coverages of the phenomenon. A total of 25 countries belong to this category: including Russia, Zimbabwe, Canada, the US, Israel, France, Chile, Mexico, Argentina, UAE, and Morocco.
Level 3 (Somewhat Popular): The majority of the population have experienced the Korean soap operas, and they are popular at a moderate degree. The K-pop music is highly popular among the majority of teens and early twenties. A total of 14 countries belong to this category: including Iran, Hungary, Romania, Paraguay, Kyrgyzstan, Tajikistan, Bulgaria, Bolivia, and Ukraine.
Level 4 (Popular): Almost everyone in the country is likely to admit that the Korean soap operas and K-pop music have been very popular. Due to its popularity, the major channels of the country have been airing a number of Korean TV shows during prime time slots for many years.
The 7 countries that belong to this category are: Uzbekistan, Peru, Panama, Ecuador, Laos, Cuba, and El Salvador.  The countries classified as Level 5 (very popular) are typically East Asian, South East Asian, and Central and West Asian countries. The total population of these countries exceeds 30% of the world population. Remarkably, the Korean wave extends beyond Korea's immediate neighbors and reaches countries as far as Ecuador, and as unexpected as Cuba, Iran, Morocco and Zimbabwe.

Mechanism
Based on the Korean government agency reports and the literature on the impact of the media, we propose that the Korean wave could have affected the foreign consumer demand for Korean products through the following mechanisms: (i) the diffusion of preferences and (ii) advertisement. While the second mechanism is limited to the products that are advertised (through product placement or celebrity promotion), the first mechanism is potentially far-reaching if consumers' affinity for Korean products improves in general as a result of exposure to the Korean pop culture.
First, as the literature on the impact of the media suggests, exposure to the media content could induce powerful imitation effect on its viewers. Similar imitative behaviors could develop when foreign consumers watch the Korean dramas. To provide some evidence on this mechanism, Figure 6 presents the survey results of 6,500 foreign participants across the world conducted by KOFICE (2015b) for people who have experienced the Korean pop culture. Interestingly, over 50% of the people surveyed responded that they become interested in Korean foods and visiting Korea after watching Korean TV shows. Because one's demand for foreign foods and travel is highly dependent on tastes and cultural preferences (rather than the objective quality of the product), these results provide strong support for the mechanism of preference diffusion. In fact, 45% of the sample replied that they become interested in purchasing 'Korean products in general' after experiencing the Korean pop culture.
It is plausible that the power of imitation is so strong that viewers not only develop desire to experience Korean culture, products, and services, but also fundamentally change their perception about South Korea. For example, in 1990, 50.8% of the Japanese respondents had negative feelings toward Korea according to the surveys conducted by the Cabinet Office, Government of Japan. 12 However, the Japanese perception of Korea had dramatically turned around, after the Korean drama Winter Sonata was aired in Japan with multiple encore runs in 2003-2004. In 2004 of the respondents had positive views toward Korea. According to another survey conducted by KOTRA (2005), the positive rating was 66.6% in 2004 and increased to 77.8% in 2005. Such positive images of Korea could in turn affect the consumer behavior through the country-of-origin effect (Obermiller and Spangenberg, 1989;Ozsomer and Cavusgil, 1991;Elliott and Cameron, 1994). 13 Second, recognizing the strong imitation effects induced by the Korean pop culture, Korean firms have often exploited the Korean wave phenomenon in their marketing strategies. To illustrate, Korean firms often sponsor the production of soap operas on the condition that their products be placed naturally in the soap opera scenes. These include, e.g., cosmetics, home appliances, or electronics. The promotion effects can be substantial. As an example, when Descendants of the Sun was released in China on the online video website Iqiyi in 2016, the sales of the compact powder used in the drama by the main actress increased by ten-times compared to the same period in the previous year. 14 Celebrity branding is also heavily used by many Korean multinationals (e.g., Samsung, LG, Hyundai and cosmetic companines). They strategically use Korean-wave celebrities who have strong persuasive power on foreign viewers to promote the sales abroad. To illustrate, when the Korean drama Descendants of the Sun had a sensational hit in China and the region, J.Estina and Laneige (two Korean brands that sell jewelry and skin-care products) saw a sudden increase in their sales (and their stock prices) during the period. In fact, the two brands were using the lead actress of the drama in their advertisements. 12 https://survey.gov-online.go.jp/h26/h26-gaiko/zh/z11.html. 13 In general, the country-of-origin effects could be driven by demand-side preference changes or supply-side quality changes. In the case of the Korean wave, the former plays a dominant role, because the drastic changes in the sales of several Korean products can hardly be explained by their quality changes.
14 Similarly, the lipstick that she used in the drama was immediately sold out within three days, based on one of the largest online shopping websites (SK Planet) selling Korean products.

Framework
We apply the conceptual framework of Anderson and van Wincoop (2003) [AvW] and allow the two channels discussed above to work through the preference parameter. Let there be C countries and I industries, with an upper-tier Cobb-Douglas preference (and expenditure share α i ) over the industries and a lower-tier CES preference over goods imported from different sources of origin within each industry. Specifically, in each industry i, goods are differentiated by the country of origin, and buyers in each country c choose imports q oci from origin o to maximize the lower-tier utility, where b oci is the taste parameter for goods produced in o perceived by buyers in country c, which can vary across industries i. The parameter σ > 1 indicates the elasticity of substitution across sources of imports; Y c is country c's nominal income; and p oci ≡ p oi τ oci is the destination price equal to the exporter's supply price p oi scaled up by the iceberg trade cost factor τ oci . Solving the utility maximization problem of (1) gives us the export values from country o to country c in industry i: where P ci is the consumer price index of industry i in country c given by P ci = o b oci (p oi τ oci ) 1−σ 1/(1−σ) .
Since the sample we will study is fixed in terms of Korea as an origin, we will omit the origin subscript below. We hypothesize that the Korean wave affects the foreign consumer preferences towards Korean products over time b ci,t through the two mechanisms (preference diffusion and marketing) as discussed in Section 4. Specifically, we assume that the taste parameter is a loglinear function of the lagged Korean TV show exports to the country (T V showExp c,t−1 ), and a destination-industry parameter shifter ( where ε ci,t absorbs any other idiosyncratic shocks to the taste parameter over time not accounted for by the TV show exports. Note that we allow the elasticity θ i (of the taste parameter with respect to the TV show exports) to vary across industries i.
Combine (2) and (3), and assume that the unobserved trade cost τ ci,t is log-linear in observable trade cost proxies X c,t . We obtain an estimable equation: where the error term u ci,t absorbs the destination-industry specific effectb ci , the Korea-industry-time specific effect p i,t , the industry specific effect α i , the shocks to the taste parameter ε ci,t , and any other unobserved variables. We include in the list of trade cost proxies X c,t : the physical distance between Korea and the destination (ln Dist c ), the regional trade agreement dummy (RT A c,t ), the Korean embassy dummy (Embassy c,t ), the number of Koreans residing in the destination country (ln Koreans c,t ), and the cross exchange rate (ln ExRate c,t , Korean Won / destination currency).
Framework 1: In the basic exercise, we focus on specific industries and estimate (4) for each of the industries. This leaves a panel data structure with variations across destination and time.
We allow for two alternative specifications of the error term such that: ln In (5), we include the year fixed-effects (FEs) and exploit the variations across countries, while (6) controls for country FEs and relies on with-in country variations across time. 15 Framework 2: In the second framework, we pool observations across industries. To control for the complicated error structure in u ci,t , we do the following. First, we take the difference of (4) across time within a period. for heterogeneous growth rates of Korean exports across industries. Third, to control for potential heterogeneous growth rates of Korean exports across destinations, we construct destination-specific growth rates of Korean exports during the period ∆GoodsExp c ≡ i ∆ ln Exp ci /I, or simply include a destination-specific constant in the difference equation. Finally, we assume that the preference elasticity varies across industries in a systematic manner. For example, one of the specifications we will use is θ i = β 1 · ConsumerGoods i + β 2 , where ConsumerGoods i is a dummy variable that indicates whether an industry belongs to the set of consumer goods, while β 2 represents the base effect of the Korean TV show export on non-consumer goods, and β 1 the additional effect on the consumer goods. In sum, the estimation equations are: Note that we have relaxed the unit income elasticity implied by theory and allow δ to differ from one.
The coefficients γ on the trade-cost proxies in principle can vary across industries as Framework 1 explicitly allows. In Framework 2, however, we do not further model how the elasticity of the trade-cost proxies differ across industries. One can take their estimates as the mean across sectors.

Identification Strategies
The estimation equations (7)-(9) are robust to a large set of unobservables as suggested above.
First, the Korea's export structure has exhibited systematic differential trend across sectors. For example, by the UN Broad Economic Categories (see Table 1 (7)-(9) controls for such heterogeneous export growth rates across industries.
Second, it is plausible that the increased popularity of Korean culture in a country has occurred simultaneously with some unobserved factors that improved Korea's trade relationship with the destination country. Such destination-specific shocks to the trend in the Korean TV program exports and in its merchandise exports, however, would be controlled for by g c in (9) or by ∆GoodsExp c in (8).
Third, it is also possible that technology changes such as the availability of internet have enabled the Korean wave to develop worldwide, and at the same time lowered the communication and information costs of international trade. Again, such unobserved shocks would be controlled for by the g i and g c terms for the industry-or country-specific trend, and by the differencing if they are country-industry time invariant.
Finally, we demonstrate that the Korean-wave effects are systematically different across types of goods and that such heterogeneity is difficult to reconcile with any supply-side factors but highlight the importance of the demand-driven mechanisms. For example, as introduced in (7), we examine whether the Korean-wave effects are stronger on consumer goods than on capital or intermediate goods (since consumer goods embody stronger cultural contents). If this is the case despite Korea's strengthening comparative advantage in non-consumer goods as suggested by Figure 7, this will imply that the preference mechanism plays a dominant role. We will also exploit the fact that women have shown stronger affinity for the Korean pop culture and test whether exports of goods/services for women are more responsive to the Korean wave. More identification strategies based on the variations across types of goods will be elaborated in the next section. Last but not least, we conduct falsification tests by showing that future changes in TV show exports have no effects on past changes in goods exports. These tests ensure that our results are not driven by some common long-run factors that determine both Korean merchandise and cultural exports.
Extensive robustness checks are presented along with the main analysis.
We close this section by countering the potential concern of reverse causality (that the Korea's exports of goods/services have caused the overseas popularity of the Korean pop culture). First, in the regression, we use lagged TV show exports in predicting the future merchandise trade flows.
Second, the falsification tests confirm that these two series are not co-determined in the long run by some unobserved factors. Third, the systematic differences in the economic impact of TV program exports across types of goods/services are difficult to reconcile with shocks to goods/services trade flows, which then affect the demand for Korean TV shows. For example, it is hard to argue that 6 Evidence on the Korean-wave effects and the importance of the

Preference Mechanism
This section provides econometric evidence that more exports of the Korean TV shows to a country, by changing its consumers' preferences, induce more demand for and hence imports of Korean goods and services. We employ three identification strategies. First, given the observation that the Korean pop culture has attracted more fans from women than men, we test whether the TV show effect is stronger on goods/services designed for women. Second, we examine whether the effect is stronger for final consumer goods than intermediate or capital goods, given that consumer goods embody stronger cultural contents. Third, to identify the effects of preference diffusion (beyond direct marketing effects), we study the effect on goods/services that are rarely advertised. We explain the estimation approaches, results, and their implications below.

Gender-Biased Preferences
It is a well-documented fact that women have shown stronger affinity for the Korean soap operas and K-pop music in the Korean-wave phenomenon. For example, the largest online video website Iqiyi in China reports the gender composition of the viewers for each video. More than 70% of the viewers are female for most Korean TV dramas. Given that women consume Korean pop culture more than men, our proposed mechanisms in Section 4 imply that products and services designed for women will be more responsive to the Korean wave. We examine this hypothesis based on three industries whose sales are observed to be highly sensitive to the Korean-wave phenomenon: tourism, beauty products (e.g., skin-care and cosmetics), and fashion goods. A positive finding will provide strong evidence for the preference mechanism, because it is difficulty to argue that the production technology or the qualities of the Korean products/services are systematically biased across genders.

Female versus Male Tourists
As suggested by Onishi (2005)  Organization. This is an excellent test of the demand-side mechanism, because tourism is strongly driven by consumers' tastes and preferences.
Columns 1-3 of Table 2 provide the estimation results of equations (5) and (6) Taking this into consideration, we create a dummy P opular c , which indicates whether a country c is highly influenced by the Korean wave with a popularity index of 4 (popular) or 5 (very popular).
This indicator variable is included as an interaction term with the Korean TV show exports.
The result in Column 3 indicates that the effect is statistically significant at 1.89 for these (and only these) countries. The findings on the percentage of female visitors are thus very similar in magnitudes based on either cross-country or within-country variations.
As a robustness check, we also conduct the analysis for the number of tourists (instead of visitors culture is highly influential.

Beauty Product Exports
According to the survey conducted by KOFICE (2015b), more than 60% and 50% of the 6500 foreign respondents consider 'attractive appearances of the actors and actresses' and 'leading the fashion and beauty trend' as the main pull factor of Korean soap operas, respectively. As a matter of fact, Korean exports of beauty products (including, e.g., skin-care products, cosmetics, and hair preparations) have increased dramatically by more than tenfold during 2003-2015. As shown in Figure 9, the growth in the Korean exports of beauty products is exponential in destinations where the Korean pop culture is very popular (e.g., China, Japan, Thailand, Singapore, and Vietnam). In contrast, the Korean exports stay relatively flat to the UK, Germany, and France, where the Korean pop culture is not popular. These export patterns of Korean beauty products suggest that changes in foreign consumer preferences induced by the Korean wave are the most plausible mechanism, because supply-side shocks such as product quality improvement cannot explain such systematic heterogeneities across destinations and the dramatic growth in exports to specific markets. Table 4 reports the estimation results based on equations (5) and (6) Table 3 for the detailed description). Based on the cross-country variation, Column 1 shows that doubling the Korean TV show exports leads to a 40% increase in beauty product exports.
On the other hand, the effect estimate based on within-country variations in Column 2 is smaller at 25%. Still, Column 3 shows that for countries with a K-wave popularity index of 4 or 5 (very popular), the within-country variation in trade is highly responsive to the Korean wave with the corresponding effect at 37% (in comparison with 5% for the other countries). Given that beauty products are mainly consumed by women, the results in this section offer another layer of evidence on the importance of the preference mechanism in the Korean wave phenomenon.

Women's versus Men's Clothing Exports
Following the same line of arguments, we test whether the TV show effect is stronger on Korea's exports of women's clothing than men's. As discussed in Section 5, Korea's exports of clothing have been in the decline in the past two decades due to production offshoring (see Figure 7 on semi-durables). To control for such confounding effects, we use the difference estimation equations (7)-(9) in Framework 2 that pools observations across industries and allows for industry-specific export trend. We include in the analysis all the HS 4-digit industries classified under 'Section XI -Textiles and Textile Articles' (see Panel B of Table 3).
To test the preference hypothesis, we construct two dummies: 'W omensClothing i ' and 'M ensClothing i ' that indicate whether a HS 4-digit industry i is one of the eight women's clothing industries and men's, respectively. The base category thus includes all industries in textiles and textile articles that are not clothing. We replace the ConsumerGoods i dummy with the W omensClothing i and M ensClothing i dummies in equations (7)-(9). Thus, the estimation equation will have two interaction terms: ∆ ln T V showExp c interacted with W omensClothing i and M ensClothing i , respectively. We expect that: (i) the two coefficients on the interaction terms be both positive and significant, if the Korean wave has stronger effects on consumer goods than intermediate goods; (ii) the coefficient on the interaction term with W omensClothing i be larger than for M ensClothing i , if women have responded more strongly to the Korean soap operas than men. Next, to address potential unobserved confounding factors as discussed in Section 5.2, we construct destination-specific export growth rates ∆GoodsExp c , defined as the mean export growth rates across all HS 4-digit industries under study. In other words, ∆GoodsExp c represents Korea's overall export trend (of the industries under study) to a specific destination c. Columns 2 & 5 of Ta Table 5 show that there is no correlation between changes in Korea's clothing exports in the 1990s and changes in Korea's future TV show exports in the 2000s, and there is no systematic difference in the effects on women's versus men's clothing exports as well. This is true in either long-period growth regressions (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001) or stacked-period regressions (1991)(1992)(1993)(1994)(1995)(1996)(1996)(1997)(1998)(1999)(2000)(2001). These falsification test results thus alleviate the concern that our findings of positive Korean-wave effects might be driven by some long-run coincidental factors.
To summarize, the results in this section support the demand-side driven mechanism that the Korean wave has changed consumers' preferences and their consumption behaviors. That the effects are stronger for products/services used by women is especially revealing of this preference mechanism, because such gender-biased trade effects within fine product categories is difficult to reconcile with supply-side explanations.

Consumer Goods Effects
In this section, we extend the analysis to all HS 4-digit industries and examine whether there are systematic differences in the Korean-wave effects across types of goods in ways that are consistent with the preference mechanism. In particular, we test two hypotheses: of the reliability of our analysis. If the preference mechanism is the key in the Korean-wave phenomenon, we expect the effect estimate associated with KoreanW aveGoods i to be larger and more significant than the estimate associated with ConsumerGoods i .

Consumer-Goods dummy Specification
We use the UN Broad Economic Categories (BEC) as shown in Table 1  are substantial. Next, by controlling for Korea's destination-specific export trend, ∆GoodsExp c , across all HS 4-digit industries, the effect on the consumer goods remains statistically significant at 12%. In fact, there appear to be negative effects on non-consumer goods, but the magnitude is relatively small in comparison with the positive effect on consumer goods. As seen in Table 5, including the destination-specific export trend, ∆GoodsExp c , absorbs the explanatory power of most other gravity variables, but leaves the differential positive effect on consumer goods intact.
Finally, by controlling for country FEs (in the difference equation), hence eliminating all countryspecific variables but the interaction term, the effect remains statistically significant at 11% on the consumer goods. As we break the period into two sub-periods (2002-2007 & 2007-2015) and stack the observation, the hypothesis continues to hold as indicated in Columns 4-6. 21 We use the concordance table between the HS 6-digit sectors and the BEC provided by the UN. About 10% of the HS 4-digit industries are mapped to both consumer and non-consumer goods, and we drop such industries. The results in Table 6 remain robust, if we force the problematic industries to be classified as either consumer goods or non-consumer goods. Tables A.1-A.3 report the results when these double-matching HS 4-digit industries are classified as consumer goods. The magnitudes of the estimates if we classify them as non-consumer goods generally fall in between these two sets of estimates (where we simply drop them and where we classify them as consumer goods).
In Panel B of Table 6, we conduct falsification tests, by regressing past (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001) changes in exports of HS 4-digit industries on future (2003-2013) changes in TV show exports. The effect estimates on the interaction term and the TV show exports are all arbitrarily close to zeros and statistically insignificant. This increases our confidence that the positive finding above is not due to spurious correlations. Overall, our findings in Table 6 suggest that the Korean pop culture has economically and statistically significant impacts on the exports of consumer goods, but not on capital or intermediate goods.

Korean-Wave Goods dummy Specification
The Korean wave goods as introduced above are often identified by the mass media and government agency reports/surveys as products whose sales have directly benefited from the Korean wave.
These Korean wave goods are typically food/beverages, cosmetics, clothing/accessories, certain home appliances, electronic products, and cars. We collect such information and construct the indicator KoreanW aveGoods i -which takes on value 1 if a HS 4-digit industry i belongs to the set of Korean wave goods and 0 otherwise. The industries classified as Korean wave goods are listed in Panel A of Table 3.
As a preliminary analysis and illustration, we estimate (5) for each of the HS industry clusters identified as Korean wave goods. The results are summarized in Table 7. We see that exports of most Korean wave goods are highly responsive to the TV show exports. Most pronounced impacts are found in beauty products and jewelry, reiterating our earlier conclusion of genderbiased Korean-wave effects. Economically large and statistically significant effects are also seen in foods and clothing, and home appliances such as washing machines and refrigerators. However, the effects are not significant for aircons, TV and cars.
We next use the robust specifications (7)-(9) with the KoreanW aveGoods i dummy in place of ConsumerGoods i . These specifications, by pooling over all HS 4-digit industries, allow us to control for many potential confounding factors as emphasized in Section 5.2. As shown in Table 8, doubling the Korean TV show exports raises the exports of Korean wave goods by 26% in the baseline specification using the long period regression of 2002-2015 (Column 1). The magnitude of this effect is significantly larger than the estimated effect on general consumer goods (16% cf. Column 1 of Table 6). This finding, where the effect on the Korean wave goods is larger than on the consumer goods, holds true throughout all the other specifications in Columns 1-6 of Tables 6 & 8. These results thus support the second hypothesis that the Korean wave goods are more responsive than the whole set of consumer goods.
Given the important role of the Korean wave goods in leading the diffusion of preferences for Korean products, we look further into the impacts in each sub-period of 2002-2007 and 2007-2015 in the next six columns of Table 8. In spite of the shorter time frames and potentially more noises, the effects on Korean wave goods remain statistically significant and robust as in the benchmark.
The corresponding falsification tests in Table 9 assure that there are no systematic spurious effects of the future TV show export growth on the past changes in Korea's exports, across all the twelve specifications.
Overall, the results in this section demonstrate that there exist significantly positive effects of the Korean wave on exports of consumer goods but not on capital or intermediate goods, and that the effects for the Korean wave goods are further larger than for the consumer goods. These findings provide further evidence for the demand-side mechanism. It is also worthwhile noting that during the decade of 2002-2015, Korea's exports of consumer goods have in general gone downtrend relative to capital or intermediate goods. This makes the finding of positive cultural effects on the exports of Korean consumer goods appear even more riveting. In a way, the Korean wave has helped buck the downward trend of Korean consumer-goods exports, with the Korean wave goods leading the way.

Diffusion of Preferences
In this section, we argue and show that of the two mechanisms: (i) diffusion of preferences and (ii) advertisement, the diffusion of preferences is the dominant driver of the Korean-wave effects documented above.
First, we argue that without the diffusion of preferences, the advertisement efforts cannot lead to such phenomenal impacts. For example, despite substantial efforts by the Korean cosmetics firms to penetrate lucrative markets such as France, UK, US, and Canada, where the Korean pop culture is not popular, the market shares of Korean beauty products remain negligible. Advertisement using Korean actresses simply does not work in these countries where the preferences for Korean products are weak. In contrast, in countries where the Korean wave popularity index is high, Korean cosmetics firms find it extremely lucrative to ride on the Korean wave by involving Korean stars in marketing activities. Thus, conceptually, the diffusion of preferences for Korean culture develop first before the advertisement channel becomes effective.
Next, we present two sets of evidence to support this argument, where the products/services are typically not advertised via product placement in the Korean TV shows or promoted by celebrity.
First, Korean clothing and foods are rarely advertised overseas. 22 However, as we have demonstrated in Table 5, Korean TV show exports have significant and positive effects on Korea's exports of clothing. Similarly, Table 7 suggests that Korea's exports in many of the food and beverage industries are highly responsive to the Korean wave. These findings strongly support our hypothesis above that diffusion of preferences is the key: the label 'Made in Korea' or 'Korean style' that emphasizes Korea as the country of origin evokes in foreign consumers the cognitive and affective effects in their decisions to purchase a product (Obermiller and Spangenberg, 1989;Verlegh and Steenkamp, 1999). To illustrate, as shown by the survey results in Figure 6, 45% of the 65,000 foreign respondents indicated that they would like to purchase products made in Korea after experiencing the Korean pop culture (and 32% even without knowing the brands, reported in the same 22 Though there are some famous Korean clothing brands, those are typically made in and exported from China or Vietnam. Clothing that are made in and exported from South Korea are mostly individual designer's clothes, not marketed on a large scale.

survey).
Next, we examine the Korean wave's impacts on South Korea's outward FDI across countries and sectors, noting that the flow of Korea's services via FDI are typically not advertised via the mass media in foreign countries. The list of sectors are given in Table 10. As in the case of Korean wave goods, we can identify certain service sectors whose sales likely have benefited from the Korean wave. For example, Figure 6 shows that more than 25% of the respondents would like to have Korean medical services, learn the Korean language, and more than 50% would like to eat and purchase Korean foods, after experiencing the Korean pop culture. Using the same sources of information (government reports) that helped us define the Korean wave goods, we classify the FDI sectors in Table 10 into Korean wave service sectors and non-KoreanWaveService sectors. The Korean wave service sectors include, e.g., Korean restaurants, grocery stores, K-pop concerts and performances/events of Korean celebrities, aesthetic and medical clinics, language institutes, and hair salons, which are set up overseas by Koreans.
If the Korean wave has induced the diffusion of foreign consumers' preferences, we expect Korea's FDI in the Korean wave service sectors to be positively affected by the Korean TV show exports. Table 11 provides a preliminary analysis for each of the seven Korean wave service sectors based on equation (5), using the FDI flows as the dependent variable. Consistent with our hypothesis, we find that the Korean wave effects are positive and significant for all the Korean wave service sectors, except 'medical services' whose estimate is positive but not precisely estimated. For example, doubling the TV show exports increases FDI in restaurants and retail/wholesale services by 23% and 41%, respectively. Strong impacts are also found for education and personal services (33% and 30%, respectively). 23 In contrast, as shown in Table 12, the impacts of Korean TV show exports are in general absent in the other sectors, including, agriculture, manufacturing, mining, construction, science and technology, public administration, real estate, and business/finance services. The exceptions are FDI in shipping, electricity, and waste management, for which we provide some plausible explanations in the footnote. 24 Following the above motivating evidence, we pool observations across sectors and allow the Korean wave effect to differ across the set of Korean wave service sectors and the set of non-KoreanWaveService sectors. Unlike the previous analysis for merchandise trade, We do not use the time-difference equations (7)-(9). This is because Korea's FDI data at the sectoral level across 23 Understandably, FDI in broadcasting and entertainment is strongly associated with Korean TV show exports. 24 The positive effect on Korean FDI in shipping is likely an induced demand for logistics services due to increased sales of Korean wave goods or increased demands for materials and equipments required in the Korean wave service sectors in the destination. For FDI in electricity and waste management, we suspect that this could be due to the state-controlled nature of Korea Gas Corporation (KOGAS) and Korea Electric Power Corporation (KEPCO), which are respectively the top 4th and 8th Korean multinationals in 2013. South Korea strongly promoted investment in energy during the five years of the Lee Myung Bak administration (2008)(2009)(2010)(2011)(2012), and the geographical distribution of these foreign investment tends to coincide with that of the Korean wave. For example, the amount of Korean outward FDI in energy is topped by Asia and reared by Africa and Middle East (Moon and Yin, 2015). Electricity generation (by nuclear or fossil fuels) produces waste at each step of the fuel cycles (Tsyplenkov, 1993). Thus, waste management by environmental regulations is a production process that accompanies energy production. This may help explain the rise of outward Korean FDI in electricity and in waste management that coincides with the countries affected by the Korean wave.
destinations are quite sparse with many zero entries; taking log difference across two years leads to a large drop in the sample size. Instead, we run the FDI regression in levels, but control for sector-year and destination FEs. This allows us to control for sector-year and destination-specific unobserved confounding factors. Table 13  In sum, this section demonstrates that even for Korean goods and services that are rarely advertised by the mass media, the Korean wave has a significant impact on their overseas sales.
The finding that the effects on women's clothing (rarely advertised in the mass media) and on beauty products (intensively advertised by famous Korean actresses) are similar in magnitudes and both economically large is especially revealing of the preference-diffusion mechanism. Finally, our finding of a significant and positive effect of the TV show exports on the Korean wave service FDI sectors reinforce the evidence for this channel.

Conclusion
Identifying a causal link from culture to international trade flows is challenging. This is because the notion of culture is so broad, and culture tends to develop slowly over time and thus highly correlated with geographic factors. There are also potentially many confounding factors that could affect both culture and trade flows. This paper exploits the phenomenon of the Korean wave, where the Korean pop culture dramatically spreads across many countries (regardless of physical distance, religion, language, and ethnicity differences) in a short time frame, to identify the effect of culture on consumer preferences. In particular, we assemble a unique dataset on Korean TV program exports across more than 150 countries for the period of 1998-2014. This measure of culture shocks is unique because the Korean TV program exports vary substantially across countries and time, and when combined with disaggregated trade flows across industries, allows us to adopt very robust estimation specifications that control for several potential unobserved confounding factors (such as destination-industry time-invariant fixed effects, Korea-industry-specific time-varying fixed effects, and destination-specific time-varying fixed effects). We hypothesize that the exposures to the Korean culture affect foreign consumer preferences for the Korean products/services, and test the implications of this demand-driven mechanism that on the other hand cannot reconcile with supply-side explanations.
Our estimation results provide strong evidence that increased exposure to Korean pop culture has changed foreigner consumer preferences for Korean goods/services, and increased Korea's ex- 25 The data on the presence of Korean embassy and the size of overseas Koreans are available only since 1995 and 1997, respectively. Thus, for the falsification tests, we use the future values for these two variables. We also verify the robustness of the results in Table 13 by dropping these two variables from the regressions. ports and FDI in (and only in) relevant industries. First, we find that the impact is stronger for products/services designed for women, consistent with the fact that proportionally more females are documented to spend more hours watching Korean TV programs and hence more intensively exposed to the Korean cultural contents. Second, the effects are found to be significantly positive for consumer goods while insignificant for capital or intermediate goods, consistent with the fact that consumer goods embody stronger cultural contents. Third, we find that the effects are significant even for the goods and services which are rarely advertised by the Korean firms overseas, which further strengthens the preference-diffusion argument. These systematic differences in the impact across products support the demand-driven hypothesis, because it is difficult to reconcile these patterns with supply-side shifts or reverse causality.
By the estimate in Column 1 of Table 6 The results in this paper suggest an interesting policy implication: promoting a country's culture and its country image such that they affect foreign consumer preferences for its products/services might just prove to be an effective export-promotion policy. This is especially relevant in a world when technology has lowered communication cost in trade to unprecedentedly low levels and policy barriers are overall liberalized by the multilateral trade negotiations via GATT/WTO. The demandside mechanism via preference diffusion, in comparison, leaves a large room for policy manipulation and could play a complementary role to efforts on the supply-side to reduce market penetration cost (Arkolakis, 2010).

A.3 Gravity variables
Data on GDP (current US$), GDP per capita (current US$), and populations are downloaded from the World Development Indicator of the World Bank. Distance between two countries is measured by distw from CEPII; in particular, the measure calculates the weighted average distance between the biggest cities of two countries, using population shares of the cities as weights. The information on RTA is based on the RTA dataset maintained by José de Sousa, and supplemented by the CEPII data on RTA.

Embassy (source, year available)
The size of overseas Koreans is retrieved from the "Report on the present state of overseas Koreans," published by South Korea's Ministry of Foreign Affairs. This report is published every two years (1997, 1999, . . . , 2015). We intrapolate the series linearly when the data for two adjacent odd years are available; i.e., Korean t = (Korean t−1 + Korean t+1 )/2. As indicated, this series is available only from year 1997 onward.
Exchange rates are sourced from the Penn World Table, and supplemented by the World Bank data when the whole exchange rate series for a country is missing from the Penn World Table. The cross exchange rate (Korean won / national currency) is calculated from the Korean Won/USD and the national currency/USD exchange rates.

A.4 Concordance of HS and BEC
The UN provides concordance between the HS 6-digit and the BEC codes. Given this concordance, we classify each HS 6-digit as consumer goods or non-consumer goods (where the cluster of consumer goods are as indicated in Table 1). We then further classify each HS 4-digit as consumer goods if all of the HS 6-digit sub-industries under a HS 4-digit industry are consumer goods, as non-consumer goods if all of the HS 6-digit sub-industries under a HS 4-digit industry are non-consumer goods, and double-matching otherwise. About 10% of the HS 4-digit industries are in the category of double-matching.
Notes: OLS estimation of equations (5) and (6). Each observation is an origin country (of the tourists) and a year. P opularc = 1 if the country c's K-wave popularity index is 4 (popular) or 5 (very popular). Standard errors are clustered by the origin country (for the year-FE regressions) and by year (for the country-FE regressions). Sample is restricted to origin countries whose GDP per capita in 2015 is more than US$ 4000. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. Men's or boys' shirts Notes: (i) Panel A: Korean Wave Goods indicate the subset of consumer goods whose sales overseas are documented by surveys and reports to have increased largely due to the Korean wave. (ii) Panel B: the HS industries included in Section XI (textiles and textile articles). (iii) Panel C: the HS industries that correspond to women's and men's clothing.  (5) and (6). Each observation is an export destination and a year. Beauty products include cosmetics, skin-care, and hair products. P opularc = 1 if the country c's K-wave popularity index is 4 (popular) or 5 (very popular). Standard errors are clustered by export destination (Column 1) or by year (Columns 2 & 3). Missing trade values are not replaced by zeros. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively.  2002-2015 stacked (2002-2007, 2007-2015) 1991-2001 stacked (1991-1996, 1996-2001) Notes: OLS estimation of equations (7)-(9) using the Women's and Men's clothing dummies. Observations include all HS 4-digit industries under Section XI (textiles and textile articles). Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in merchandise exports. Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.  2002-2015 stacked (2002-2007, 2007-2015) 1991-2001 stacked (1991-1996, 1996-2001) Notes: OLS estimation of equations (7)-(9). Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in merchandise exports. Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.
Notes: PPML estimation of equation (5). Standard errors are clustered by the export destination. Missing trade values are not replaced by zeros. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively.  (5). Standard errors are clustered by the export destination. Missing trade values are not replaced by zeros. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively.  2002-2015 stacked (2002-2007, 2007-2015) 2002-2007 2007-2015 (1) (2) Notes: OLS estimation of equations (7)-(9) using the Korean Wave Goods dummy. Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.  1991-2001 stacked (1991-1996, 1996-2001) 1991-1996 1996-2001 (1) (2) Notes: Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in merchandise exports. OLS estimation of equations (7)-(9) using the Korean Wave Goods dummy and the future changes in Korea's TV show exports. Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.
Notes: PPML estimation of equation (5). Missing FDI values are replaced by zeros. Standard errors are clustered by FDI destination. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The sector 'International' is dropped because there are insufficient observations.  Table 10. The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in FDI. The data for the two control variables, Embassyc,t and ln Koreansc,t, are available only since 1995 and 1997, respectively. Thus, we use their future values in the falsification tests. The alternative results are provided as robustness checks if we simply drop them from the regression. stacked (2002-2007, 2007-2015) 1991-2001 stacked (1991-1996, 1996- Notes: OLS estimation of equations (7)-(9). Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in merchandise exports. Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.  2002-2015 stacked (2002-2007, 2007-2015) 2002-2007 2007-2015 (1) (2) Notes: OLS estimation of equations (7)-(9) using the Korean Wave Goods dummy. Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.  1991-2001 stacked (1991-1996, 1996-2001) 1991-1996 1996-2001 (1) (2) Notes: Falsification tests explore the relationship between the future changes in Korea's TV show exports and the past changes in merchandise exports. OLS estimation of equations (7)-(9) using the Korean Wave Goods dummy and the future changes in Korea's TV show exports. Observations include all HS 4-digit industries. Each observation is a destination country and a HS 4-digit industry (and a period for the two-period stacked regressions). Standard errors are clustered by export destination in specifications (7)-(8) and robust in specification (9). The symbols * , * * , and * * * indicate statistical significance at the 10%, 5%, and 1% level, respectively. The two control variables, ∆RT Ac and ∆Embassyc, are dropped, because the RTA and Embassy status of Korea with most trading partners have not changed during this period.