Intertemporal and cross-sectional contrasts in effects of trade: Significance of the technology content of exports

Abstract

This study identifies cross-country and intertemporal differences in the effects of trade exposure on per capita national income. We develop a small open economy endogenous growth model with high- and low-technology sectors and endogenous human capital accumulation. We then test the predictions of our model on a sample of 70 countries over the period 1980–2017. Our main assertion is that gains from trade are not only disproportionate across countries but also contrasting over time, depending on the technology intensity of exports. Countries with lower initial experience in the production of technology-intensive goods and services tend to specialize in sectors with lower demand for better-educated and high-skilled workforce, which lowers the return to and individual incentives for education. Consequently, trade-induced specialization patterns, due to their implications about technological progress, appear to be an important factor causing cross-national divergence in welfare. Our theoretical model implies that, in an unskilled labor-abundant country, higher exposure to international trade can decrease the long-run growth rate, even though it increases short-run per capita income. In a skilled-labor abundant country, both short-run and long-run effects are positive. Our empirical findings, which identify short-run and long-run effects separately, strongly support these predictions.

Keywords:

• Trade exposure
• technology intensity of exports
• human capital
• economic growth
• intertemporal effects

JEL Classifications:

• F16
• F43
• J24
• O31
• O33

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 One exception can be the unified growth theory (Galor 2005), which links human capital formation and technological progress in the very long-run, focusing on the demographic transition during the process of economic development.

2 Similar to the set-up in Grossman and Helpman (1991).

3 The model neglects the possible positive effects of technology transfers via channels, such as foreign direct investments (FDI) and multinational companies (MNCs) or disembodied technology in traded intermediates. These assumptions eliminate any direct effect that trade might have on the profitability of R&D. Hence, there are national as opposed to international returns to scale.

4 The technological innovation process will be of the ‘quality-ladder’ type, as in Aghion and Howitt (1992).

5 This implies that the productivity of skilled workers increases with the duration of education they get, but only when they are employed in the research labs.

6 It is important to note that, the assumption of unit input requirement depending on S which is determined endogenously creates 2 complementary channels through which opening up to trade changes steady-state growth rate: Without those, the only channel which would decrease (increase) the steady-state innovation rate (hence the economic growth rate) in an unskilled (skilled) labor abundant country after it opens up to trade would be the decrease (increase) in the resource base (h) as a result of the change in the relative wages of skilled vs unskilled labor. Whereas with these properties of the model, a second channel arises: after opening-up to trade, in unskilled-labor (skilled-labor) abundant countries, due to decrease (increase) in skill-premium, a new long-run equilibrium in labor market is possible where individuals are investing a lower (higher) amount of time for education, which causes the productivity of the skilled labor in research labs to go lower (higher) and consequently results in a decrease (increase) in the steady-state growth rate. In other words, the ‘effective value’ of skilled-labor changes due to economy-wide changes in supply and demand conditions both in the goods market and the labor market.

7 See Auer (2015) for a model based on heterogeneous workers who make educational decisions in the presence of complete markets. When such heterogeneity exists among workers, only ‘high-ability’ types earn a surplus from their investment in education.

8 This way of modeling individuals' decisions to invest in education builds on the framework of Findlay and Kierzkowski (1983).

9 Note that $\beta (1-\frac{1}{\lambda })$ is the profit rate in the intermediate sector.

10 See the functional form of the cost-minimizing unit input coefficient for H above (equation [16(16) $\begin{array}{rl}& a_{HY}=(1-\beta )\frac{P_Y}{w_H}{a}_{xY}=\beta \frac{P_Y}{P_X}\\ & a_{LZ}=(1-\beta )\frac{P_Z}{w_L}{a}_{xZ}=\beta \frac{P_Z}{P_X}\\ & a_{Hx}=\frac{\mathrm{\partial }{c}_x(w_{L,}{w}_H)}{\mathrm{\partial }{w}_H}{a}_{Lx}=\frac{\mathrm{\partial }{c}_x(w_{L,}{w}_H)}{\mathrm{\partial }{w}_L}\end{array}$(16) ]).

11 See equation (21(21) $y_{i,t}={\mathrm{\Sigma }}_i{\beta }_i{X}_{i,t}+{\alpha }_i+u_{i,t}$(21) ) and recall that $\frac{w_L}{w_H}<1$.

12 The human capital index, provided by PWT, is based on average years of schooling and on rate of return to education (estimated à la Mincer). See data appendix for further details. As another alternative, we used years of schooling (schooling); our findings persist.

13 There is also the issue of interdependence among countries, which arises possibly from spatial correlations, spillover effects, omitted global variables, and common unobserved shocks. This causes the standard panel estimators employed in the existing cross-country studies to be highly problematic (Pesaran 2006).

14 As shown by Pesaran and Shin (1998) and Pesaran, Shin, and Smith (1999).

15 Since these tests are performed post-estimation of auxiliary regressions with ‘dynamic common correlated effects’ estimator, we do not report the details of those results here; they are available upon request.

16 We used the panel unit root test developed by Im, Pesaran, and Shin (2003) because it relaxes the assumption of common auto-correlation parameter across cross-section units and does not require a balanced sample. Moreover, to mitigate the potential impact of cross-sectional dependence, we subtracted the cross-sectional averages from the series, as suggested by Levin, Lin, and Chu (2002).

17 Our original data set consisted of 92 countries. However, we had to drop several countries due to a large amount of missing observations (Bahrain, Barbados, Costa Rica, Dominican Republic, Ghana, Malta, Panama, Sri Lanka). Additionally, we dropped several other countries either because of very small economic/population size (El Salvador, Guatemala, Honduras, Jamaica, Mauritius, Malawi, Nicaragua, Togo, Trinidad and Tobago) or because they appeared to be outliers (Kuwait, Saudi Arabia, Venezuela – oil-dependent economies and Hong Kong, Singapore – city-states).

18 Penn World Tables – version 9.0; available at: www.ggdc.net/pwt

19 For technology intensiveness of products according to SITC Rev. 3, see https://unctadstat.unctad.org/EN/Classifications/DimSitcRev3Products_Ldc_Hierarchy.pdf.

20 For technology intensiveness of products according to SITC Rev. 2, see http://unstats.un.org/unsd/tradekb/Attachment398.aspx?AttachmentType=1.

21 We compared values estimated by linear interpolation with those estimated by cubic, spline, piece-wise cubic Hermite, and nearest neighbor interpolation; they were highly close, if not equal, for all cases.