Services Development and Comparative Advantage in Manufacturing

In Table 2, we estimate the specification in equation (1). The dependent variable is manufacturing export RCA calculated based on DVA in gross exports. The U.S. domestic services input intensity is averaged over 1995-2007 and treated as time-invariant. The financial (business) services development measure is defined as the ratio of U.S. financial (business) services valueadded to U.S. GDP. Because the embodied services measures are based on U.S. data, we drop the observations for the U.S. from the regressions to alleviate the potential endogeneity problem. In the first three columns, we consider financial services (f), business services (b), and the combined financial and business services (fb) respectively. Country fixed effects, year dummies, and manufacturing sector dummies are all included in the first three regressions. Standard errors are always robust to heteroscedasticity and are also clustered by country*sector to address the potential serial correlation in the error terms for a particular country-sector across years.

The coefficient of services development is negative and significant in the first regression for financial services, but not significant for business services in the second regression. The coefficient of the key interaction term is always positive and highly significant. The results imply that financial services development reduces manufacturing RCA when embodied financial services are sufficiently low. This is not surprising given the definition of RCA: services development tends to increase a country's services export RCA and in turn should lead to lower manufacturing export RCA when manufacturing sectors do not benefit much from services development due to low services input intensity. When embodied services are sufficiently high, services development can actually increase manufacturing RCA. These results provide strong support for our first hypothesis. We can calculate easily the cutoff value of$SII$. Taking regression (1) as an example, the cutoff $SII_{f}$ is about 0.046 as compared to its average value (0.035) reported in Table 1. The last three columns of Table 2 are analogous to the first three regressions except that we include time-varying country and time-varying sector fixed effects. As a result, services development measures and log(GDP/capita) are dropped from the regressions. The three interaction terms remain positive and highly significant, with similar magnitude as in the first three regressions. 

The control variables in Table 2 have the expected signs. Manufacturing productivity (TFP), the measure of scale economy (log(emp)), capital–labor ratio (K/L), and GVC participation increase manufacturing RCA. Other variables, including log(GDP/capita), relative wage, and the skill ratio, do not have significant effects.

In Table 3A, Table 3B, and Table 4, we perform various robustness checks. In Table 3A, we use an alternative measure of services development defined as average value added per worker in financial or business services. Our previous results continue to hold well. The interaction term D*SII is always positive and significant at the 1% or 5% level. Their magnitude is much smaller because the average values of the new services development measures are much bigger as shown in Table 1.

In Table 3B, we replace the services development measures in Table 2 by another two alternative measures for financial services as discussed in Section III.4. Because such a measure is not available for the corresponding WIOD business services sector, we perform this robustness check only for financial services. Our previous findings hold very well with or without timevarying fixed effects. The estimated cutoff$SII$ (about 0.04) is similar to what we got from Table 2.

In Table 4, we examine the sensitivity of our results to alternative measures of services input intensity. The time-varying country and sector fixed effects are used in all of the regressions, so both$SII$ and D variables are dropped. We consider here financial and business services together. In the first regression, we replace the average U.S.$SII$ with time-varying U.S.$SII$; our main findings remain unchanged, with a slightly smaller coefficient of the interaction term than the corresponding one reported in column (3) of Table 2. Although the services input intensity of the U.S. is arguably the best choice to capture the role of financial and business services in manufacturing sectors, it is still useful to check the robustness of the results when countries' own$SII$ measures are used. Regression (2) in Table 4 is analogous to those in the first column, except that we replace U.S.$SII$ with each country's own $SII$ (time-varying). We no longer drop the U.S. observations from this regression. The interaction term remains positive and significant at the 1% level, but the magnitude of the coefficient is much smaller than the one reported in the first column, probably because a country's own $SII$ may not capture well the potential role of services in manufacturing sectors if services sectors are under-developed as we would expect. In the last column, we use the average $SII$ of the U.K., another developed country with competitive services sectors. The results are similar to those when the U.S. data are used: the magnitude of the $D*SII$'s coefficient is similar to what is reported in Table 2 (30.3 vs. 27.72). 

Next, we test for the second hypothesis, which states that countries may bypass their own inefficient domestic services sectors by relying on imported foreign services. As defined in equation (11), the share of embodied foreign services in total embodied services (forsh) is used to measure the degree of a country's access to foreign services markets. Because our story is relevant only to the sectors that use a significant amount of services as inputs, we consider only the first seven manufacturing sectors with high services input intensity as listed in Appendix 4, and expect to see a stronger bypass effect than from sectors with lower services input intensity. We examine how the interaction between foreign services and domestic services development affects manufacturing export RCA based on specification (2) and report the results in Table 5A.  As in the previous table, in the first three columns, we include separate country, year, and sector fixed effects; the first two regressions consider financial and business services respectively; and the third regression combines the two types of services. The coefficients of D*forsh are always negative and significant at the 1% or 5% level. This shows that the benefit of foreign services inputs on manufacturing export RCA decreases with the level of domestic services development, suggesting that foreign and domestic services inputs are at least partially substitutable. Together with a positive coefficient of forsh, this also implies that the access to foreign services can help a country to bypass under-developed domestic services provision. In the last three columns of Table 5A, we include time-varying country and sector fixed effects. As a result, we cannot estimate the coefficient of D any longer. The absolute value of the estimated coefficients of the interaction term is even larger, although it is statistically less significant for financial services. 

If we include some additional sectors with medium levels of services input intensity, the above results are still robust, although a bit weaker as expected. For instance, including also sectors 9 and 7 does not lead to a dramatic change in the results, except that the interaction term turns slightly less significant in the regression for financial services. We also run similar regressions as in Table 5A for the other seven manufacturing sectors with low financial and business services input intensity as listed in Appendix 4. For these sectors, services development and access to foreign services markets should matter less. The results are reported in Table 5B. As expected, forsh and its interaction with D are mostly insignificant at the 10% level. Although the interaction term is significant at the 10% level for business services in column (2), the magnitude of the coefficient in absolute value is smaller than the corresponding coefficient in Table 5A. These results provide further support to the second hypothesis.