The role of endowments, technology and size in international trade: new evidence from product-level data

ABSTRACT

Using product-level trade data, we empirically investigate the export patterns of more than 150 countries in their exports to the USA, Brazil, India, and Japan. We document strong evidence that exporters specialize according to their relative factor endowments, technology, and economic size. More developed, capital abundant countries are found to export products of higher unit values and a wider range of products to developed, emerging and developing markets. More developed, economically larger, and technologically advanced countries are also the major exporters of new products, spanning a wide range of product categories with high unit values. Our findings provide important insights into the macro phenomenon that a large proportion of the global trade takes place among developed economies, and that the latter are also major exporters to developing markets.

KEYWORDS:

• New products
• product specialization
• unit value

JEL CODES:

• F20
• F14

Acknowledgements

Our special thanks go to Devashish Mitra and Mary E. Lovely for their extremely valuable comments on this paper. We would also like to thank Peter Schott for providing us with the US import data for the period 1972–1988. We would like to thank two anonymous referees, David Hummels, John Romalis, and seminar participants at Syracuse University, Deakin University, the University of Melbourne, Monash University and Australian National University for helpful comments on an earlier version of this paper. The usual disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. While Evenett and Venables (2002) and Kehoe and Ruhl (2013) overlap with some of our analysis, there are also major differences. For example, Evenett and Venables (2002) decompose trade growth by product line and destination using bilateral trade data of a sample of 23 developing economies. Yet, they use data at the three-digit level, which are much more aggregate than the data at 6-, 10-, or 12-digit levels used in this study. They also do not look into the extent to which the endowment and technology of exporters are related to the components of the intensive and extensive margins of trade. Kehoe and Ruhl (2013) use detailed trade data to investigate the determinants of the extensive margin of international trade, but their focus is on the effects of trade costs.

2. Schott (2004) focuses on multiple-sourced products, Hummels and Klenow (2005) make no distinction between products sourced from one or different cones of diversification.

3. While our static analysis of new products cannot provide a direct answer to the dynamic question of whether there exist product cycles in the US imports over time, our findings at the disaggregate level can provide certain insights into this phenomenon. For example, the evidence that low-wage exporters of the South and high-wage exporters of the North both export the ‘same’ new products to the US market seems to contradict the product cycles theory. Yet, we show that that while these new products may belong to the same product category in the data, they are of very different quality.

4. The gravity equation is also found to be robust and successful in explaining the volume of world trade both within and between the North and the South, but this model does not reveal what factors of standard trade theories (i.e., technology or factor endowments or both) explain the volume of bilateral trade. Deardorff (1998) showed that the gravity equation is not specific to any trade theory.

5. Costinot (2009) sets up another important model in which technology and factor endowments jointly determine the patterns of trade. See also Morrow (2010) for a related empirical study.

6. The reason for setting up a three-country Heckscher–Ohlin model is that it is straightforward to relate the three-country theoretical setting to the empirical analysis with three cones of diversification.

7. Letters (A, B, C...) can be considered as industry classifications (e.g., TV industry B) in the data, and numbers (z¹,z², z³...) can be considered to denote the products (e.g., black-and-white TVs, colored TVs and LCD TVs).

8. Another example is the difference in quality of cars made by local Chinese and Indian producers versus cars made by Japanese and German producers.

9. It is reasonable empirically to assume that products of a higher capital intensity and more advanced technology are of a better quality, and, consequently, have higher unit values.

10. Schott (2004) referred to the specialization in products D and C as ‘across-product’ specialization and the specialization in product B as ‘within-product’ specialization.

11. Schott (2003) show that empirical trade economists face a serious problem when using industry-level data to test the Heckscher–Ohlin model in that the three-digit ISIC manufacturing industries exhibit significant variation in terms of both input intensity and price across countries.

12. The prediction about the size effect is not limited to the DFS theoretical framework but it also applies to other trade models as well.

13. Specifically, Zhu and Trefler incorporate international technology differences in their model of trade between the North and the South, by assuming that, for each common set of factor prices the North has relatively lower marginal costs for relatively more skill-intensive goods.

14. In Figure 1 country Developing that is the least productive also has its unit cost lower than countries Emerging and Advanced on the range of products with the lower capital intensity. As it falls further behind other countries its unit cost will shift upward, which results in the country's having comparative advantage in a lower range of products.

15. It is worth noting that in order to meet the condition of full employment of the factors in the Heckscher–Ohlin model, the two arrows representing the vectors of capital-labor ratio of developing countries 5 and 6 must lie inside the vectors of the capital-labor ratio requirements of products D and A. In order to keep the diagram simple, only the two dotted rays connecting the origin to the tangencies between the isoquants and isocost lines of products 1 and 2 are drawn, to define the diversification cone of developing countries 5 and 6.

16. Large economies of the European Union such as France, Germany and United Kingdom are also good candidates to be included in our analysis. We decided not to include these countries for two reasons. First, like China these three largest economies of the EU have their data at six-digit HS classifications for the period starting 1992 only. Second, our sample already includes the U.S. and Japan, the two developed economies which are, we believe, very similar to France, Germany and the UK.

17. See Online Appendix 1 for the list of exporting countries used in our analysis.

18. An alternative way of identifying new products is to look for new product codes after taking into account changes in product codes over time and the fact that obsolete codes may be mapped to the new codes. We applied this method using the concordances constructed by Pierce and Schott (2012). This approach yielded essentially the same results as reported in the working version of this paper (link suppressed for reviewing purposes).

19. Penn World Table, Version 8, is available from the following link: http://www.rug.nl/research/ggdc/data/penn-world-table.

20. Hallak and Schott (2011), for example, can only estimate the cross-country differences in product quality for a sample of 50 countries.

21. Only 5% of the total number of observations correspond to products that are sourced exclusively from low-wage and/or middle-wage exporting countries. See Online Appendix 2 for the number of different types of products in the US import data from 1972 to 2004. Similar patterns are also observed in the import data of Brazil, India, and Japan.

22. The main disadvantage of GDP per capita to classify exporters into different cones is that it is not strictly based on the theoretical foundation of the Heckscher–Ohlin with multiple cones of diversification.

23. Specifically, Schott (2003) first constructs more theoretically appropriate industry aggregates by grouping together products that are manufactured with identical techniques and then identifies the cones by looking at how the output of those aggregates varies with endowment. The idea behind this method of identifying cones is that within the same diversification cone the Rybczynski relationship is expected to be linear.

24. While sample selection may be a problem in the regression Equations (5) and (6), it is impracticable to address this problem: for the USA there are 12,000 products over 20 years from 130 exporting countries, hence the sample size would make up more than 30 million observations. Note also that we do not include α_c, the exporter fixed effects, in Equations (1)(1) and (2(2) ). Since GDP per capita and capital stock per worker of exporters vary insignificantly from one year to another, if we include α_c (in addition to) there is not much within-exporter variation left in GDP per capita or relative factor endowment that can affect the unit value log(UV_pct). The regression results with α_c included showed that it was actually the case

25. Thus, the explanation for the trade among high-wage/developed economies based on differences in countries from the supply side differs from the explanation provided by Hallak (2006) on the demand side.

26. Note that this finding is in line with the recent study by Pierce and Schott (2014) who find that when the USA opened up its markets to Chinese exports, the US manufacturing employment declined by some 18%. The finding may have been the result of somewhat ‘sticky’ USA wages in the face of the downward pressure on the wages of low-skilled workers across all sectors which resulted from the increase in USA/China trade (Petit 2010).

27. These results are available in Online Appendix Table 3.

28. Khandelwal (2010) finds that unit values were more correlated with estimated qualities for products for which there is scope for quality differentiation, and that prices were appropriate proxies of quality for these products.

29. We used Rauch's (1999) conservative definition of differentiated products. The results remain essentially the same when Rauch's liberal definition of differentiated products is applied.

30. We control for a China dummy in all regressions in the rest of the paper.

31. For more details see the results on Table VI of Schott (2004).

32. Xiang (2013) identifies new products as those product classifications for which the assigned fraction of trade value is equal to or greater than 25%. That is, if a product category consists of, say, two products, with both of them identified as new, then the assigned fraction of trade value of new products in that product category is 100%. If one of them is identified as new and the other is not, then the fraction is equal to the value of the new product divided by the total value of trade of that category. When we restrict our analysis to product classifications only to those with assigned fraction of trade value being 100%, the sample size is reduced by around 20%, but the regression results remain essentially the same.

33. Note that if a new product was identified in a given year it remains new in subsequent years of our sample.

34. We also found that the numbers of single-sourced new products that are exported exclusively by low- or middle-wage exporters, and the numbers of double-sourced new products that are exported simultaneously by low- and middle-wage exporters, are insignificant.

35. According to Hall, Trajtenberg and Jaffe (2001), the patents included in their dataset are utility patents representing more than 90% of the total number of patents. An innovation is patentable only if it provides some identifiable benefit and is capable of use.

36. Note that the number of patents granted by the USA to its trading partners, as pointed out by Keller (2004), is not a perfect measure of technology of the latter. Yet, this is the measure that is available for a large number of countries for the period spanning the data of this study.

37. It is important to note a nuance between this paper and the study by Hausmann et al. (2007). While our study makes the case for specializing in and exporting products in accordance with a country's fundamentals, Hausmann et al. (2007) advocate a more flexible specialization policy that is based not only on a country's fundamentals but also on the number of entrepreneurs eager to engage in the process of discovering the underlying cost structure of the economy. More cost discovery will generate knowledge spillovers and will result in a better growth-enhancing mix of goods that a country produces with higher productivity and exports. Hausmann et al. (2007) call this mix of goods ‘rich-country products’ to distinguish them from the mix of ‘poor country products’ associated with low productivity. We thank an anonymous referee to bring this difference between our study and the study by Hausmann et al. (2007) to our attention.