WIOD Conference on Industry‐Level Analyses of Globalization and its Consequences

1. Output Quality, Skill Intensity, and Factor Contents of Trade:An Empirical Analysis Based on Micro-Data of the Census of Manufactures WIOD Conference on Industry‐Level Analyses of Globalization and its Consequences Technische Universitaet Wien, Vienna May 26-28, 2010 Kyoji Fukao (Hitotsubashi University and RIETI)  Keiko Ito (Senshu University)

2. 1. Introduction • Recent studies on intra-industry trade (IIT) have brought to light rapid increases in vertical IIT (VIIT) (for example, see Fukao, Ishido and Ito (2003) and Schott (2004)). • Many theoretical models assume that developed economies export physical and human capital-intensive products of high quality and import unskilled labor-intensive products of low quality from developing economies. • Through this mechanism, an increase in vertical IIT may have a large impact on factor demand and factor prices.

3. 1. Introduction (Contd.) • Empirical studies use information on the unit value of commodities as a proxy for product quality. • Most of empirical studies based on the unit value data take the positive relationships between commodity prices and factor intensities as given. • Yet, to the best of our knowledge, few studies have empirically examined the relationship between unit values of commodities and their factor contents at the factory level. (Notes: Some recent studies try to take quality difference among firms (plants) into account. <Quality Heterogenous-Firms model> Baldwin & Harrigan 2007; Kugler & Verhoogen2008; Hallak & Sivadasan 2009, etc.)

4. 1. Introduction (Contd.) • The Purposes of this study is: - To develop a theoretical framework to estimate the relationship between the unit values of gross output and factor intensities - To test whether factories that produce goods with a higher unit value tend to input more skilled labor and capital stock services, using micro data of the Census of Manufactures for Japan -To calculate the factor contents of trade for Japan, based on the estimated relationship between the unit values and factor intensity.

5. 2. Theoretical Analysis of Factor Contents in VIIT • We assume that factories, in order to produce commodities of a high quality, engage in production processes that are intensive in both skilled labor and capital. • Suppose that N commodities are produced in an industry. For each commodity, there is a continuum of different qualities. • Each “commodity” in our model corresponds to one product item in the most detailed commodity classification of production and trade statistics and that products that differ only in quality are not recorded as different products in the statistics.

6. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.) • Each commodity, (n, q), is produced by a Leontief-type constant-returns-to-scale production function. where LU, q, i, t, LS, q, i, t, Kq, i, t and Mq, i, t denote unskilled (blue-collar) labor, skilled (white-collar) labor, capital, and intermediate input. Yq, i, t denotes the gross output of factory i. a i, t denotes factory i’s total factor productivity (TFP) level in comparison with the industry average TFP level in year t. We express elasticity values by ηY=(qi, tde(qi, t))/(e(qi, t) dqi, t), ηS=(qi, tdf(qi, t))/(f(qi, t) dqi, t), ηK=(qi, tdg(qi, t))/(g(qi, t) dqi, t), ηM=(qi, tdh(qi, t))/(h(qi, t) dqi, t), respectively.

7. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.) From our production function, we have the following factor demand relationships: We assume that the price elasticity of demand for each factory’s output in this industry is constant and takes the same value for all factories producing commodity n.

8. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.) From an unit production cost function and constant mark-up ratio, we have the following relationship between p and q. By making a linear approximation of equation (3.3) and subtracting average values across all factories from both sides of the equation, we have

9. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.)

10. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.) where

11. 2. Theoretical Analysis of Factor Contents in VIIT (Contd.) Since we assume constant returns to scale and a constant mark-up ratio, we have the following identity among the coefficients of the above four equations. This constraint means that a one percent increase in the unit price of output corresponds to a one percent increase in the unit production cost. We estimate the four equations under the constraint (3.19) using SUR techniques.

12. 3. Factor Contents in Japan’s VIIT • Using equations (3.15) and (3.18), we can express the ratio of the white-collar labor input to the output quantity for a factory which produces commodity (n, q) as follows: where c’n, t denotes a commodity- and year-specific constant term. Let φD, n, t(pt)denote the distribution function of output quantity by all the factories producing commodity n in Japan over unit value p. Then, we can derive the following equation from the above equation:

13. 3. Factor Contents in Japan’s VIIT (contd.) • In a similar way, we can derive

14. 4. Data • Census of Manufactures for Japan - Larger Establishment Sample: all mfg. plants with 30 or more employees - 6-digit commodity level information on shipment and quantity  Unit values can be calculated for 800 commodities out of 2,000 commodities - Number of blue-collar and white-collar workers available for years 1981, 1984, 1987, and 1990. • Trade Statistics for Japan - Values and quantities of exports and imports at the HS 9-digit commodity level (7,000 commodities for exports and 9,000 commodities for imports)

15. 4. Data (contd.) • To estimate the relationship between the unit value of gross output and factor intensities, we select only single-product establishments, which we define as establishments where one commodity accounts for more than 60% of total shipments. • By using the unit value and factor intensity information taken form the CM and the TS, we calculate the factor contents of Japan’s trade, taking account of quality (price) difference between exported and imported goods.

17. 5. Empirical Results on the Relationship between Output Unit Values and Factor Intensities • Estimate equations (3.15)-(3.18) by SUR estimations subject to the constraint expressed by equation (3.19)----using factory-level data from the Census

18. 5. Empirical Results on the Relationship between Output Unit Values and Factor Intensities: Basic Result

19. 5. Empirical Results on the Relationship between Output Unit Values and Factor Intensities: TFP controlled

20. 5. Empirical Results on the Relationship between Output Unit Values and Factor Intensities: Estimation without the Constraint

21. 5. Empirical Results on the Relationship between Output Unit Values and Factor Intensities: based on data of factories belonging to firms with no additional factory and whose headquarters are located in the same place

22. 5. Robustness checks: Relationship between Unit Production Costs and Factor Intensities

23. 5. Robustness checks: Relationship between Output Unit Values and Factor Intensities including Multi-Product Establishments

24. 6. Calculated Factor Contents in VIIT (contd.) • Using unit value information taken from the CM and the trade statistics as well as data on factor intensities, we estimate the factor contents of Japan’s VIIT. • For number of white-collar workers embodied in Japan’s exports for commodity n are calculated as: Unknown Estimated

25. 6. Calculated Factor Contents in VIIT (contd.) • Assuming that φE, n, t(pn, t) and φI, n, t(pt) follow a log normal distribution, μE : log of unit value of Japan’s exports μD : average of the factory-level unit values in logarithm σE: standard deviation of the distribution functions of exports σD : standard deviation of the distribution functions of all shipments by single-product factories

26. 6. Calculated Factor Contents in VIIT (contd.) • Due to data constraints, we use average unit value differences (μE – μD, μI – μD) at the broad industry level. • Average difference between ln(unit value for exporting factories) and ln(unit value for non-exporting factories), calculated using the CM for 2001-2004. • Average difference between ln(export unit value) and ln(import unit value), calculated using the TS • We assume that σE= σI= σD

27. Table 7. Difference in average unit values Estimate relative unit values for domestic shipments, exports, and imports

28. Tables 8 & 9. Estimated factor contents of net exports LS +7,000 persons (1.5%) LU +5,000 persons (2%) K +41 bil. yen (0.4%) LS +16,000 persons (10%) LU -21,000 persons (-18%) K +310 bil. yen (2%)

29. 7. Conclusion • As for the relationship b/w the unit value of a product and its white-collar labor intensity, the significant and positive relationship we find is important empirical evidence which supports the assumption widely employed in theoretical models. • On the other hand, we find that the widely employed assumption that commodities with higher prices are more physical capital-intensive does not always hold. • The results of the factor contents of trade estimation suggests that the implication of international trade on the domestic factor markets should be very different if we take account of VIIT.

30. 8. Extensions • Incorporate headquarter-level data in our analysis in order to take account of white-collar tasks provided by headquarters.  Using information from “Kigyo Katsudo Kihon Chosa (BSBSA)? • What is the definition of “skill”?  We may use the wage information as a proxy of skill. • Distinguish between intra-firm trade and inter-firm trade?  If we can match the trade statistics with the firm- or plant-level data, it would be possible. (c.f. France, U.S.)