2.1 Theoretical Research

Economic studies on international trade in films and media programming have begun to explain why Hollywood dominates the international market. Focusing on product differentiation and the public good nature of films,Footnote ¹ Wildman and Siwek (Reference Wildman and Siwek1988) argue that the film industry is characterized by economies of scale and that market size is a key factor determining the competitiveness of a film. Filmmakers in a larger market produce a greater variety of high-budget films, which tend to have a higher quality.

Wildman and Siwek (Reference Wildman and Siwek1988) also consider the cultural discount and argue that the value of a film decreases when it is shown in a foreign country. Movie-goers prefer a film that is produced in their native language or reflects their own culture. A foreign film may be dubbed or subtitled, but subtle expressions of emotion cannot be easily delivered to a foreign movie-goer, and thus satisfaction with the film would be reduced. Therefore, all other things being equal, a country tends to import fewer films when the films are produced in a country that is more culturally distant from the importing country.

Waterman (Reference Waterman1988) also emphasizes market size and cultural discount as determinants of domestic market share. A large audience base enables producers to invest more in their programs, which leads to higher quality and thus to greater competitive advantage for their programs in the world market. Viewers in foreign countries prefer their local programs to ones imported from the US, but the competitive advantage of the US programs compensates for this effect.

Hoskins and Mirus (Reference Hoskin and Mirus1988) argue that the US programs are more competitive in the international market because they receive less cultural discount than those of other countries. Due to the diversity of the US population with immigrants from many countries, the US producers could make programs which are well received in foreign countries. They also note that the cultural discount explains why trade occurs predominantly in some genres, such as entertainment and drama that are less culture-specific than news and public affairs programming. Hoskins and McFadyen (Reference Hoskins and McFadyen1991) argue that the US competitive advantage in the global television market comes from economies of scale, first-mover advantage,Footnote ² and the characteristics of the market environment, such as private broadcasters seeking to maximize their audience, the heterogeneous nature of the US population, and the lower tolerance by US viewers to foreign programming.

2.2 Empirical Research

Based on the theoretical arguments above, empirical research has been conducted to test a hypothesis that a country with a larger market tends to have a greater market share of local films or media programming. This hypothesis was generally accepted by all research that we reviewed (Dupagne and Waterman, Reference Dupagne and Waterman1998; Jayakar and Waterman, Reference Jayakar and Waterman2000; Choi, Reference Choi2011; Oh, Reference Oh2001; Lee and Bae, Reference Lee and Bae2004; Lee and Waterman, Reference Lee and Waterman2007; Fu and Lee, Reference Fu and Lee2008). Market size is often measured by GDP or box office revenue.

The other key variable, cultural discount, represented by cultural distance (for example, between the US and a sample country) or English proficiency turned out to be insignificant or to have an inconsistent sign (Dupagne and Waterman, Reference Dupagne and Waterman1998; Jayakar and Waterman, Reference Jayakar and Waterman2000; Oh, Reference Oh2001;Footnote ³ Lee and Bae, Reference Lee and Bae2004; Chan-Olmsted et al., Reference Chan-Olmsted, Cha and Oba2008Footnote ⁴). Jayakar and Waterman, (Reference Jayakar and Waterman2000) interpret this result as implying that a foreign film can overcome differences in culture and/or language through dubbing.Footnote ⁵ Meanwhile, Hoskins et al. (Reference Hoskins, Mirus and Rozeboom1989) show that English-speaking countries tend to have greater willingness to pay for US television programming, which suggests linguistic differences as barriers to entry into foreign markets.

In contrast, some studies support the significance of the cultural discount in explaining international flow of films. For example, Fu and Lee (Reference Fu and Lee2008) analyze imported films in Singapore from various source countries during February 2002 to January 2004. They find that films from culturally more distant countries experience less box-office success in Singapore than films from countries with more similar cultures. Fu and Sim (Reference Fu and Sim2010) examine the international flow of films of nine major exporting countries and report that cultural distance reduces the flow of films between two countries while a common language increases such flow.

Dupagne and Waterman (Reference Dupagne and Waterman1998) and Lee and Bae (Reference Lee and Bae2004) have considered an import quota and a screen quota, respectively, as policy variables. Contrary to expectations, neither study lends support to the alleged protective effects of restrictive policies. Chan-Olmsted et al. (Reference Chan-Olmsted, Cha and Oba2008) consider the protection of intellectual property rights and show that it is significant in one model but not in another. Dupagne and Waterman (Reference Dupagne and Waterman1998) state that a quota as a dummy variable is a crude measure of regulations, and the results may have been different with a more refined measure. Lee and Bae (Reference Lee and Bae2004) point out that screen quota requirements in some countries are often ignored by local exhibitors who seek to maximize profits.

4.1 Two-Stage Least-Squares (LS) Estimation

This two-stage LS estimation method can isolate the effects of time-invariant variables from the country-specific effects if an orthogonality condition is satisfied (Breusch et al., Reference Breusch, Ward, Nguyen and Kompas2011; Greene, Reference Greene2011). The orthogonality condition is satisfied in our model, as explained below.

At the first-stage, we use dummy variables to represent the combined effects of the time-invariant variables and the country heterogeneity in equation (1); for country i (=1,···, 30) and year t (=2001,···, 2013)

(3)$$MS\_dom_{it}^{} = \theta _i + \alpha _3\log (market{\rm} \;size)_{it} + \delta _t + u_{it}^{} $$

where $\theta _i^{} = \alpha _1STRI_i + \alpha _2CD_i + f_i$, the sum of three terms. Since the country-specific effects (f _i) are captured by $\theta _i^{} $, we can obtain consistent LS estimates for α ₃ and δ _t in equation (3). These estimates are equivalent to the LS estimates obtained after the within-transformation approach is applied. Conditional on the consistent estimates of $\hat{\alpha} _3$ and $\hat{\delta} _t$, we calculate residuals which contain the combined effects of the time-invariant variables and the country-specific heterogeneity.

(4)$$\eqalign{ {\hat{\eta}} _{it} &= MS\_dom_{it}^{} -{\hat{\alpha}} _3\log (market{\rm} \; size)_{it}-{\hat{\delta}} _t \cr & {\rm} = \alpha _1STRI_i + \alpha _2CD_i + f_i + \psi _{it}^{}} $$

where $\psi _{it}^{} = u_{it} + e_{it}$ with e _it representing the sampling errors of $\alpha _3-\hat{\alpha} _3$ and $\delta _t-\hat{\delta} _t$. Since consistent estimates of $\hat{\alpha} _3$ and $\hat{\delta} _t$ are used, these sampling errors converge to zero as the number of cross-sectional units approaches to infinity.

At the second stage, we regress the residuals on the observed time-invariant variables, STRI _i and CD _i. Since the trade barriers and the cultural distance were determined before the sample period and remained the same afterwards, they are exogenous in the determination of $MS\_dom_{it}^{} $ and thus are uncorrelated with f _i.Footnote ¹³ Therefore, this second-stage LS estimation for equation (4) yields consistent estimates for α ₁ and α ₂ although f _i is not used as a regressor (Hausman and Taylor, Reference Hausman and Taylor1981; Breusch et al., Reference Breusch, Ward, Nguyen and Kompas2011; Greene, Reference Greene2011). To account for possible heteroscedasticity caused by the use of an estimated dependent variable and by the fact that the estimation error of ($\alpha _3-\hat{\alpha} _3$) is multiplied by log (market size)_it, we calculate White's heteroscedasticity-consistent standard errors (Lewis and Linzer, Reference Lewis and Linzer2005; Greene, Reference Greene2011).Footnote ¹⁴

According to the results in Table 3, log(BOR) and log(SCR) representing the realized market size are positively significant at the 1% and 5% levels, respectively. In contrast, the proxy for the potential market size, log(GDP), is not significant even at the 10% level. As expected, the realized market size is more relevant to explain the differences of the domestic market shares across countries. The cultural distance (CD) also turns out to be a positively significant factor for the domestic market share, indicating that cultural distance reduces the inflow of foreign films.

Table 3.

Estimates and test results by the two-stage LS estimation method

Notes: ^a This R ² is calculated without the variation explained by the country-specific effects.

^b This is to test whether there exist country-specific effects (i.e., differences between countries). Small p-values indicate that there exist significant country-specific effects in the dependent variable (MS_dom), given barriers, CD, and a market-size variable.

^c This is to test whether the country-specific effects are correlated with the market-size variable included in each regression. Small p-values indicate that the country-specific effects are significantly correlated with the market-size variable in each regression.

Year dummies are included in all regression models.

The standard errors are in the parentheses below their coefficient estimates.

***, **, and * denote a significant coefficient at the 1, 5, and 10% level, respectively.

To test whether the country-specific effects exist and are correlated with (market size)_it, we calculate the residuals using consistent estimates of $\hat{\alpha} _1$ and $\hat{\alpha} _2$ in equation (4).

(5)$$\hat{\hat{\eta}} _{it} = \hat{\eta} _{it}-\hat{\alpha} _1STRI_i-\hat{\alpha} _2CD_i = f_i + \hat{\psi} _{it}$$

where $\hat{\psi} _{it} = \psi _{it} + v_{it}$ with v _it representing the sampling errors of $\alpha _1-\hat{\alpha} _1$ and $\alpha _2-\hat{\alpha} _2$. Since consistent estimates of $\hat{\alpha} _1$ and $\hat{\alpha} _2$ are used, these sampling errors converge to zero as the number of cross-sectional units approaches to infinity. Thus, the country-specific effects (f _i) can be consistently estimated by the within-country averages of the residuals $\hat{\hat{\eta}} _{it}$.

We now test the null hypothesis of H ₀ : f _i = constant for all i, which means that there exist no differences across countries, i.e., no country-specific effects. The bottom panel in Table 3 supports the existence of significant differences across countries in the domestic market share, given the variables for trade barriers, cultural distance, and market size. The correlation coefficients indicate significant positive correlations between f _i and the variables for market size. Therefore, these test results suggest that country-specific effects have to be accounted for to avoid the endogeneity problem.

To evaluate the effectiveness of trade barriers (measured by STRI) in promoting the domestic market share, we focus on regression (i) which uses log(BOR) to proxy the market size. Its coefficient estimate of 5.456 indicates that the domestic market share ($MS\_dom$) increases by 0.05456 percentage point (%p) for a 1% increase in BOR. The impact of STRI is estimated as 0.340, indicating an increase in $MS\_dom$ by 0.340%p for an increase of 1 in STRI.Footnote ¹⁵ We now calculate the impact when each variable changes across its interquartile range in the data. To obtain this impact, we multiply the coefficient estimate by the interquartile range of each variable. A change by 569% in BOR across its interquartile range (from 111 to 743, in Table 1) is expected to promote $MS\_dom$ by 31.1%p (= 569 × 0.05456), while an interquartile change by 6 in STRI (from 13 to 19) is to promote $MS\_dom$ by 2.0%p (= 6 × 0.340). These estimates imply that the magnitude of the impact on the domestic market share is much larger when the film market grows than when the trade barriers become higher.

4.2 Instrumental-Variable (IV) Estimation

Hausman and Taylor (Reference Hausman and Taylor1981) suggest an IV estimation method for panel data models which include time-invariant variables and individual effects. We rewrite equation (1) as follows

(6)$$\eqalign{ MS\_dom_{it}^{} &= \alpha _1STRI_i + \alpha _2CD_i + \alpha _3\log (market{\rm} \; size)_{it} + \delta _t + \varepsilon _{it}^{} \cr {\rm} \varepsilon _{it}& = f_i + u_{it}} $$

where f _i is treated as a random variable and thus included in a new error term. To isolate the effects of the time-invariant variables from the unobserved individual effects (f _i), it is required that the model has to include a sufficient number of explanatory variables uncorrelated with f _i (and thus ɛ _it). As explained in subsection 4.1, the trade barriers (STRI) and the cultural distance (CD) are considered to be exogenous and uncorrelated with f _i, thus satisfying the requirement. However, the proxy variables for the market size are shown to be correlated with f _i in subsection 4.1. For such endogenous proxy variables, we use the deviations from their within-country means as instrumental variables, following the suggestions in Hausman and Taylor (Reference Hausman and Taylor1981) and Breusch et al. (Reference Breusch, Mizon and Schmidt1989).

Hausman and Taylor (Reference Hausman and Taylor1981) treat f _i as a random variable but allow f _i to be correlated with some explanatory variables. This is the key difference from the random-effects models which assume f _i to be uncorrelated with all explanatory variables. Treating f _i as a random variable, Hausman and Taylor (Reference Hausman and Taylor1981) apply the generalized method of moments (GMM) with the aforementioned instrumental variables. This GMM can account for disturbance covariances containing f _i and u _it, and it is thus expected to produce consistent and efficient IV estimators.Footnote ¹⁶

Table 4 reports the IV estimation results, which are consistent with the two-stage LS ones. The 95% confidence intervals of the coefficient estimates in Table 4 overlap with the two-stage LS ones in Table 3, indicating that they are not different at the 5% significance level. However, the IV estimation method produced larger standard errors and lower significance levels than the two-stage LS estimation. One reason for this is that the IV estimation method uses only the exogenous parts of the endogenous variables to identify its effects, thereby reducing the explanatory power. So, the R ²s in Table 4 are lower than the ones in Table 3. Another reason is that the two-stage LS method conditions the second-stage estimation on the first-stage estimates of $\hat{\alpha} _3$ and $\hat{\delta} _t$. Since the uncertainty associated with the estimates is ignored, the two-stage LS standard errors could be underestimated. In sum, we can say that the significance levels in Table 3 are the upper bounds, and the ones in Table 4 are the lower bounds.

Table 4.

Estimation results by the Hausman and Taylor (Reference Hausman and Taylor1981) instrumental-variable (IV) method

Notes: Year dummies are included in all of the regression models.

The standard errors are in parentheses below their coefficient estimates.

***, **, and * denote a significant coefficient at the 1, 5, and 10% level, respectively.

In the above estimation, we used the CD composite measure which is an average of the six indices weighted by their respective variances. We also include all of the six indices in the same regression model. Table 5 reports the estimation results when the market size is measured by log BOR _it and the model is estimated by the Hausman and Taylor (Reference Hausman and Taylor1981) instrumental-variable method.Footnote ¹⁷ The null hypothesis of the equality of their coefficients is rejected with a p-value of 0.001. In this regression shown in the second column (ALL), the coefficient for each index represents its net effect when the other five indices are held constant. Since the six indices are correlated with each other, their net effects could be positive or negative depending on their correlations. With the six indices included, the effect of log BOR _it is significant at the 10% level but the one of STRI is not.

Table 5.

Estimation results by the Hausman and Taylor (Reference Hausman and Taylor1981) IV method when all and each of the six indices in the CD composite measure are used

Notes: When all of the six indices are included (ALL), the null hypothesis of equal coefficients is rejected with a p-value of 0.001.

Year dummies are included in all of the regression models.

The standard errors are in the parentheses below their coefficient estimates.

***, **, and * denote a significant coefficient at the 1, 5, and 10% level, respectively.

We also include the six indices one by one. In these regressions (PDI ~ IVR), the coefficient for each index represents the total effect, i.e., the sum of its net effect and the indirect effect through the correlations with the other indices. Thus, the coefficient for each index captures the effect of cultural distance in all of the six aspects. Five indices of PDI, IDV, UAI, LTO, and IVR have significant total effects on the domestic market share at the 1% or 5% levels; only MAS is not significant. Their estimates are within the 95% confidence intervals. In the regressions (PDI ~ IVR), the effect of log BOR _it is significant at the 1% or 5% levels but the one of STRI is not.

4.3 Robustness Checks: Lagged Effects of the Market Size

It might take time for the market size to influence the domestic market share in each country. To consider such lagged effects of the market size, we use its lag-one variable.

(7)$$MS\_dom_{it}^{} = \alpha _1STRI_i + \alpha _2CD_i + \alpha _3\log (market{\rm} \; size)_{i,t-1} + f_i + \delta _t + u_{it}^{} $$

The estimation results for equation (7) are summarized in Table 6. These results are qualitatively the same as the ones with the contemporaneous market size in Tables 3 and 4. For regression (i) which uses log(BOR) to proxy market size, the two-stage LS estimates in Table 6 are 0.461 for STRI, 7.000 for CD and 1.971 for log (BOR) while the estimates in Table 3 are 0.340, 6.906, and 5.456, respectively. The corresponding estimates and their significance levels are qualitatively the same, although the lagged $\log (BOR_{i,t-1}^{} )$ has a weaker impact than the contemporaneous $\log (BOR_{it}^{} )$. When the Hausman and Taylor IV method is employed, the estimation results in Table 6 are also qualitatively the same as the ones in Table 4.

Table 6.

Estimation results when the lagged values of the market size are used

Notes: Year dummies are included in all of the regression models.

The standard errors are in the parentheses below their coefficient estimates.

***, **, and * denote a significant coefficient at the 1, 5, and 10% level, respectively.

In addition, use of the lagged values could reduce a possible simultaneity problem. For example, the box office revenue $(BOR_{it}^{} )$ as a measure of market size in equation (1) is assumed to have a causal effect on the domestic market share ($MS\_dom_{it}^{} $) during the same period. However, their relation could be simultaneous in that having a strong domestic film industry leads to higher box office revenue due to higher quality of domestic films. If there exists such two-way causation, the estimation results obtained with the contemporaneous $BOR_{it}^{} $ could be biased. Since the lagged values of box office revenue have been determined in the previous period t–1, the relation is one-way from $BOR_{i,t-1}^{} $ to $MS\_dom_{it}^{} $ in equation (7). As explained above, the results in Table 6 with the lagged market size are qualitative the same as the ones in Tables 3 and 4 with the contemporaneous market size. It appears that our main results about the determinants of the domestic market share and the effectiveness of trade barriers are not sensitive to a possible simultaneity.Footnote ¹⁸