Productivity Spillovers from Foreign Direct Investment: Firm-Level Evidence from China

World Development Vol. 40, No. 1, pp. 62–74, 2012 Ó 2011 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2011.05.006

Productivity Spillovers from Foreign Direct Investment: Firm-Level Evidence from China XINPENG XU Hong Kong Polytechnic University, Kowloon, Hong Kong

and YU SHENG * Australian National University, Canberra, Australia Summary. — Using firm-level census data, this paper examines the spillover effects of foreign direct investment (FDI) on domestic firms in the Chinese manufacturing industry between 2000 and 2003. Our analysis takes into account the endogeniety of input choices, simultaneity bias, and clustering errors that are known to cause biased and inefficient estimations. Our results suggest that positive spillovers from FDI arise from forward linkages where domestic firms purchase high-quality intermediate goods or equipment from foreign firms in the upstream sectors. Our results also show that domestic firms differ significantly in the extent to which they benefit from FDI. Ó 2011 Elsevier Ltd. All rights reserved. Keywords — foreign direct investment, spillover effects, Asia, China

1. INTRODUCTION

negative. For example, being a supplier to a multinational firm does not necessarily lead to positive benefits. If the market for supplying inputs to foreign firms is competitive, it is possible that foreign firms in downstream sector will undercut prices to take advantage of the competitive market in the upstream sector. Empirical evidence of the benefits of FDI spillovers is limited (Rodrik, 1999). Due to a lack of detailed firm-level data, researchers have focused mainly on developed countries such as the United Kingdom (Haskel, Pereira, & Slaughter, 2007), where firms, as technological leaders, may have little to gain from FDI spillovers. Other studies focus on small developing countries where the amount of FDI is relatively small and domestic industries are not sufficiently diversified to reap significant benefits from FDI. For example, Aitken and Harrison (1999) estimate the productivity effects of FDI to a sample of Venezuelan manufacturing plants during 1976–89, and find that plants in industries with a higher foreign presence actually had lower productivity than those in other industries. Javorcik (2004) finds that domestic firms in Lithuania only benefit from FDI when they are the suppliers to foreign firms. Blalock and Gertler (2007) find positive vertical spillover effects from FDI in Indonesian manufacturing firms. Lopez (2008) examines the effect of foreign technology licensing in Chile and finds that licensing in upstream sectors increases the productivity of plants that purchase intermediate inputs from them while

Over the past two decades, cross-border flows of foreign direct investment (FDI) have taken center stage in the globalization process, with increasing numbers of firms (usually based in developed countries) investing in foreign countries (either developed or developing countries). According to UNCTAD (2008), the global flows of FDI increased from US$324 billion in 1995 to US$1.3 trillion in 2006. In 2006 inflows of FDI to developed countries amounted to US$857 billion, while inflows to developing countries rose to a record US$379 billion. The global stock of FDI has thus more than quadrupled from US$2.76 trillion in 1995 to $12 trillion in 2006. A commonly-held belief among policy makers is that FDI benefits recipient countries through knowledge transfer from multinational firms, which helps improve the productivity of domestic firms. As such, governments around the world provide policy incentives to attract multinational firms. There are several channels through which FDI may affect domestic productivity. First, domestic firms may benefit by observing and imitating multinational firms in the same industry; however, it is also possible that the presence of foreign firms increases competition and reduces the market share of domestic firms, which may lower domestic firms’ productivity (horizontal spillovers). Second, productivity spillovers may occur through labor turnover, as former employees of multinationals who have acquired managerial expertise, production or marketing skills, resurface in domestic firms or set up their own firms to which they can transfer that knowledge (horizontal spillovers). Third, domestic firms may also benefit through backward linkages, by being a supplier to multinational firms and thereby obtaining some free technology transfer, or through forward linkages, by having a foreign supplier and gaining access to better machinery equipment or intermediate inputs, which may lower costs and increase productivity (vertical spillovers). However, it is important to note that horizontal and vertical spillovers from FDI may be positive or

* We are grateful to the Editor of this journal, four anonymous reviewers, Ann Harrison, Gary Jefferson, Hongbin Li, Kuijs Louis, Dong He, participants at the conference “Investments, Technology Spillovers, and East Asian FTA” organized by Fudan and Brandeis Universities (Shanghai, October 2008) as well as at “First Annual International Conference on the Chinese Economy” organized by the Hong Kong Institute for Monetary Research in 2010 for comments. Financial support from the Hong Kong Polytechnic University (G-U685) is gratefully acknowledged. All remaining errors are our own. Final revision accepted: April 8, 2011. 62

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

licensing in downstream sectors has a negative effect on the productivity of suppliers of intermediate inputs. Recently, Suyanto, Salim, and Bloch (2009) have shown that there are positive productivity spillovers to FDI in the Indonesian chemical and pharmaceutical industry. However, there is a lack of firm-level study on a large FDI recipient country in the developing world, such as China, where any spillover effects may be most important. 1 The case of China is of interest for several reasons. First, China is the largest recipient of FDI in the developing world, recording US$ 106 billion of inflows in 2010 and a total FDI stock of US$ 384 billion at the end of 2010, and accounting for about 7% of fixed asset investment in China each year in the past decade. 2 This level of FDI appears sufficiently large for China to reap horizontal benefits. Second, China’s history under centralized planning led to unique industry development. As the economy has opened to foreign direct investment, the existence of a wide spectrum of industries provides domestic firms with opportunities to benefit through backward and forward linkages with foreign firms. Third, as a developing economy, China’s distance from the technology-and-management frontier may place it in an ideal position to exploit the potential benefits of FDI, relative to more advanced industrialized countries (Findlay, 1978). Finally, over the years Chinese governments at various levels have provided substantial amounts of subsidies to foreign firms, ranging from land at price that is much lower than the market price to tax exemptions on corporate income for the first few years of their investment in China. An important question is whether these policy incentives to foreign multinationals are justified, which depends on whether there are spillovers from FDI in China. Although there have been many studies of FDI spillovers in China, most studies use industry-level data (see the reference to Hale and Long (2007)). For example, Wang and Zhao (2008) use a panel dataset for Chinese industry over the period from 2000 to 2002 and find both positive horizontal and vertical spillovers. However, industry-level studies suffer from problems such as aggregation bias and endogeneity, as discussed in Hale and Long (2007) and Haskel et al. (2007, footnote 2). There are also a few firm-level studies using small sample dataset with mixed results. For example, Hu and Jefferson (2002) study FDI spillovers in China’s electronic and textile industries while Hale and Long (2007) use a sample of 1500 firms in five Chinese cities in 2000. Fleisher, Li, and Zhao (2010) examine province-level data and find that FDI had positive productivity spillovers before 1994 but not after. Sun (2009) analyzes how FDI affects domestic firms’ exports using firm-level data in one industry. Wei and Liu (2006) use a panel of more than 10,000 domestic and foreign-invested firms for the period from 1998 to 2001 in China and find spillovers occur within regions. More recently, Lin, Liu, and Zhang (2009) examine productivity spillovers using valueadded production function (instead of gross output value production function as in our paper and many others) and find positive vertical linkage effects but negative horizontal spillovers. This paper uses a comprehensive micro dataset—the Chinese manufacturing census data of firms (including all stateowned enterprises and non-state-owned firms with annual sales of more than 5 million renminbi (about US$600,000)) for the years 2000 to 2003 (with about US$500,000 firms each year)—to study the effects of FDI on domestic-firm productivity. We contribute to the literature in several ways. First, by using census data, instead of a sample survey for an industry or a region, we are able to undertake a full-scale examination of firm-level FDI spillovers in China. Second, our empirical

63

analysis overcomes a variety of problems typically associated with this type of analysis, including endogeneity of input choices, simultaneity bias, and clustering effects in standard errors. In particular, we differ from the literature by dealing with clustering effects through a new approach recently proposed by Woodridge (2006), and using first-differencing and the instrumental variables approaches to deal with simultaneity bias. Third, we find negative backward spillovers in the case of China, which is in contrast to Javorcik (2004), who finds positive backward spillovers in the case of Lithuania. Our further analysis suggests that the presence of negative backward spillovers may be due to the fact that many FDI firms in China are export-oriented. We believe we are the first to explore the issue as to why in the case of China there exist negative backward spillovers. Finally, we explore the role of heterogeneity in firms to see whether certain firm characteristics (such as ownership structure and export orientation) have implications for FDI benefits. Our results indicate that positive spillovers from FDI operate through forward linkages where domestic firms purchase high-quality intermediate goods with lower input prices, or equipment from FDI firms in the upstream industry. With high FDI presence in their upstream industry, Chinese domestic firms in an industry can produce a greater output (for a given level of inputs) than otherwise similar firms in industries with lower upstream FDI. Furthermore, the expected positive knowledge spillovers of FDI firms in the same industry as domestic firms are counterbalanced by competition effects arising from the entry of FDI firms, resulting in negative horizontal spillovers after controlling for a firm’s market power. The finding of negative backward spillovers may be a bit puzzling. Our further investigation suggests that this may be the result of a set of unique Chinese FDI policies that encourage foreign firms to import raw materials and equipment from the international market. We do find support that more export-oriented firms have weaker vertical linkages. Finally, we also find that domestic firms differ significantly in the extent to which they benefit from FDI, with large and medium-sized, non-state-owned enterprises, and exporting firms accruing the greatest benefits from foreign firms in China. The rest of the paper is organized as follows. The next section briefly provides a background to FDI in China. Section 3 discusses the construction of our dataset and provides basic statistics, as well as the parameter-identification strategy implemented. Section 4 discusses the results. Section 5 concludes. 2. OVERVIEW OF FOREIGN DIRECT INVESTMENT IN CHINA Although China’s first experience with FDI came after the reforms of 1978, it was not until 1992 that high levels of FDI started to flow into the country. Between 1992 and 2006, FDI inflows increased from US$1.1 billion to $73 billion. In particular, after its entry into the WTO in 2001, China’s commitment to broader and deeper liberalization in trade and investment further accelerated FDI inflows and increased the share of foreign ownership of Chinese assets. In 2006, the share of FDI inflow in total fixed-asset investment reached 5.28%, with the manufacturing sector having the largest recipient of FDI in China, accounting for 63.6% of the total FDI. 2. China’s policy objectives in attracting FDI are to advance China’s technology and to promote exports, as articulated in Article 3 of the Law of the People’s Republic of China on

64

WORLD DEVELOPMENT

Foreign-owned Enterprises, by: “. . . [encouraging] the establishment of foreign-owned enterprises that are export-oriented or technologically advanced.” To promote exports by foreign firms, China offers import tariff and value-added tax (VAT) exemption for imported raw materials and parts used in export processing. This tax incentive encourages foreign firms to purchase inputs from, and to export their output to, the international market. In fact, imports by foreign firms accounted for almost 59% of China’s total imports while exports by foreign firms accounted for 57% of China’s total exports in 2007. 2 Consequently, most foreign firms in China are exported-oriented. An unintended consequence of the tax incentive has been a weakening vertical linkage between foreign firms and local Chinese firms, in particular, a lack of backward linkage with those Chinese firms in the upstream industry. 3 China also offers various preferential treatments to foreign firms if their investment falls into the so-called “high-tech” sector. Within the manufacturing sector, FDI has started to move away from labor-intensive industries, where FDI was initially concentrated, to capital-intensive and technologyintensive industries. Between 2001 and 2005, the growth in total assets of foreign firms was greatest in the most technologyintensive industries—increasing by 137%—followed closely by capital-intensive industries, where total assets increased by 125%. In contrast, over the same period foreign firms’ total assets in labor-intensive industries increased by 81%. 4 FDI inflows into China contribute significantly to the process of marketization in the manufacturing sector. In 2006, the total output value of FDI firms was 6.09 trillion renminbi, accounting for 47.5% of the total output value of private enterprises in the Chinese manufacturing sector. With more foreign firms entering into the Chinese manufacturing sector, state owned enterprises (SOEs) are less dominant. Due to the intensified market competition, more productive firms enter while less efficient firms exit. Between 2000 and 2003, the average eight-firm concentration ratio (CR8), 5 defined as the sum of the market share of the eight largest firms across 21 two-digit level manufacturing industries decreased from 8.7% to 8.5% as the average FDI output share increased from 29.0% to 30.5%.

3. DATA AND ESTIMATION STRATEGY (a) Data collection and variable definition The data used in this study is derived from the Annual Industrial Enterprise Census conducted by the National Bureau of Statistics (NBS) of China. The census covers all state-owned firms and non-state-owned enterprises with annual sales above 5 million renminbi in the mining, manufacturing and public-utility sectors, across all provinces. It is estimated that the firms covered in the dataset account for about 90% of the total industrial output. The data has been checked by NBS for consistency and an independent examination by Holz (2005) suggests that it is likely to be of high quality. The dataset used in this study is an unbalanced dataset at the firm level for the manufacturing sector (China Industry Classification Code: 13–42), which spans the four year period from 2000 to 2003. The number of firms sampled varies from 134,130 in 2000 to 169,810 in 2003. 6 Table 1 provides summary statistics for the number of firms, the value of output, the amount of labor, capital, and intermediate inputs for domestic firms and FDI firms. 7 The real output value of firms, Y, is defined as the firm’s output value, deflated by the producer price index at the firm level, with

1990 as the base year. 8 Labor input, L, is measured by total number of employees which includes all the full-time production and nonproduction workers recorded for a firm, excluding part-time workers and casual workers. As China implemented reform on SOEs in the sample period, there are workers who are classified as redundant workers or have been laid off and thus are not involved in the production activity but are still kept in the firm’s employee name list. A better measure on labor usage would be a measure of labor who are actively involved in the production activity of the firm (or so-called “zai gang” workers), or more accurately, hours worked. However, these data are not consistently available to us in the period under study. Since the total number of employees does not include part-time workers and casual workers, it may underestimate firms’ labor usage. On the other hand, as the total number of employees also includes redundant workers or laid-off workers who do not contribute to productive activity, it may tend to overestimate firms’ labor usage. Taken together, it is not clear whether the variable over- or underestimate labor usage of a firm. Capital, K, is defined as the value of fixed assets at the end of the year, deflated by the sectoral price index for investment goods, with 1990 as the base year. As defined by the NBS, intermediate goods, M, is the value of total output less value added, plus the net value-added tax, deflated by the intermediate-input deflator. Price deflators for investment goods and for intermediates are taken from China Fixed Asset Investment Year Book (various years) published by the NBS. Following Javorcik (2004), we measure the presence of FDI in an industry by calculating the weighted sum of foreign capital, with the weight being each firm’s share of industry output (Horizontaljt): !, ! X X ð1Þ ForeignShareit  Y it Y it ; Horizontaljt ¼ i2j

i2j

where i denotes firm, j denotes industry and t year, ForeignShareit measures the share of firm i’s total equity owned by foreign investors. The index is calculated at the two-digit level of industry classification code based on a firm’s registered capital. 9 We follow the concordance table published by the NBS to convert China Industrial Classification Code to International Standard Industrial Classification (see Zhao 2004). The average share of foreign equity in an industry during the period 2000–2003, Horizontaljt, as shown in Table 1, registered a significant increase over time. The backward and forward linkages of FDI are captured by Backwardjt and Forwardjt, which are defined, following Javorcik (2004), as follows: X Backward jt ¼ ajk Horizontalkt ð2Þ k–j

and Forwardjt ¼ " X

X

"" ujm

m–j

ðY it  X it Þ

X

#, ForeignShareit  ðY it  X it Þ

i2m

## ;

ð2AÞ

i2m

where ajk is the proportion of industry j’s output supplied to industry k, derived from the 1997 input–output table at the two-digit International Standard Industrial Classification (ISIC) level, 10 and /jm is the share of inputs purchased by industry j from industry m in total inputs sourced by industry j. Yit is the total output and Xit is the export of firm i at time t.

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

65

Table 1. Summary statistics for domestic and FDI firms in China: 2000–2003 2000 Total Output (million US$)

106.2 (150.4) Employment (person) 315 (1182) Capital (million US$) 46.2 (73.4) Intermediate inputs (million US$) 82.5 (121.5) Horizontal 21.7 (10.7) Forward 8.5 (6.2) Backward 6.7 (4.8) Number of firms 134130

2001

Domestic

FDI

86.2 (120.9) 313 (1248) 40.4 (70.5) 66.4 (93.7) –

191.3 (237.1) 324 (859) 70.4 (82.1) 150.6 (199.6) –

– – 108714

Total

112.5 (187.5) 336 (1249) 46.7 (89.9) 87.6 (149.7) 22.8 (11.2) – 8.8 (6.3) – 7.0 (4.9) 25416 147690

2002

Domestic

FDI

89.4 (128.47) 332 (1335) 40.3 (83.1) 69.2 (102.2) –

208.5 (334.8) 354 (886) 72.7 (109.5) 163.7 (267.2) –

– – 119113

Total

126.6 (214.4) 365 (2461) 47.9 (94.6) 97.9 (172.9) 23.2 (11.2) – 9.0 (6.5) – 7.1 (4.9) 28577 154317

Domestic 101.0 (148.5) 362 (2766) 41.4 (91.4) 77.7 (117.5) – – – 123816

2003 FDI

Total

231.8 148.5 (379.3) (288.3) 382 276 (962) (921) 74.4 47.3 (105.9) (84.8) 180.8 114.5 (309.3) (238.2) – 24.1 (11.7) – 9.5 (6.8) – 7.5 (5.2) 30501 169810

Domestic

FDI

115.3 (189.4) 259 (957) 40.5 (78.8) 88.1 (148.6) –

279.1 (519.9) 345 (756) 73.9 (102.3) 218.4 (440.6) –

–

–

–

–

135355

34455

Note: Output is defined as the total output value, employment is defined as the total number of employees, and capital is defined as the net fixed asset value. “Horizontal”, “Forward”, and “Backward” are defined in the text. All financial variables are measured with US dollars at the 1990 constant price, and the exchange rate used for the conversion is 4.7832 (China Statistical Yearbook, 2000, 2001, 2002, 2003). Numbers in parenthesis are standard errors.

Table 2. Correlation matrix for FDI and industry concentration variables

Horizontal Forward Backward CR8

Horizontal

Forward

Backward

CR8

1.000 0.378 0.369 0.192

1.000 0.292 0.061

1.000 0.023

1.000

Sources: Authors’ calculation.

Correlations among the three FDI variables are low. As shown in Table 2, the correlation between Horizontaljt and Backwardjt (Forwardjt) is 0.369 (0.378) while the correlation between Forwardjt and Backwardjt is 0.292. The market concentration ratio CR8 in an industry has a low correlation with Horizontaljt (0.192) and a much lower correlation with Forwardjt (0.061) and Backwardjt (0.063). Table 3 shows the distribution of FDI firms and their shares (output-weighted) across industries at the two-digit level within the manufacturing sector during the sample period. The industry that had the largest share of foreign investment is “Instruments, Meters, Cultural, and Clerical Machinery” (57.9%), followed by “Communication Equipment, Computers, and Other Electronic Equipment” (55.2%), “Cultural, Educational, and Sports Goods” (49.4%) and “Leather, Fur, Feather, and Related Products” (40.6%). (b) Specification and identification To examine whether FDI generates intra-industry and/or inter-industry productivity spillovers to domestic firms, we start with a specification that has been used extensively in the literature (see Aitken and Harrison (1999) and Javorcik (2004)): ln Y ijrt ¼ a0 þ a1 ln Lijrt þ a2 ln K ijrt þ a3 ln M ijrt þ a4 Horizontaljt þ a5 Backward jt þ a6 Forward jt þaX þ aj þ ar þ at þ eijrt ;

ð3Þ

where Yijrt denotes the real output of domestic firm i operating in industry j and region r at time t, Lijrt, Kijrt, and Mijrt are labor, capital, and intermediate production inputs, respectively. X is a vector of three control variables. The first control

variable, fdi, is defined as the capital share of foreign investment in a domestic firm, is included to isolate the impact of foreign capital’s participation in a firm on its own productivity from spillovers from FDI presence in the industry. The second variable is the eight-firm concentration ratio (CR8), defined as the sum of the market share (in revenue) of the largest eight firms in an industry. The third variable is the market share of a firm in the industry, MS. The last two variables are included to further isolate the spillover effects from those due to the market Ppower of thePfirm. The three sets Pof dummy variables, aj = j-jdj, ar = rvrd,r and at = tdtdt, are used to control for the industry-, region-, and time-specific effects, respectively. Several econometric issues, including the endogeneity of input choices, cluster effects, and simultaneity bias, are central to correct estimation of the effects of FDI on domestic productivity. We address them as follows. (i) Endogeneity of input choices Ordinary least squares (OLS) is inappropriate for estimating the impacts of labor and capital on productivity, since the factors of production should be treated as endogenous. Olley and Pakes (1996) (OP method), followed by Levinsohn and Petrin (2003) (LP method), point out that inputs like capital should be considered endogenous since producers choose the level or usage rate based on cost and productivity considerations. These considerations are observed by the producer but not by the econometrician. Thus, productivity estimates may be biased if the endogeneity of input choice is not taken into account. To address this concern, we employ a semi-parametric estimation procedure suggested by Levinsohn and Petrin (2003) (We also employ the OP method and GMM method for robustness check). Compared with the approach of Olley and Pakes (1996), the LP method allows for firm-specific productivity differences that exhibit idiosyncratic changes over time, and use intermediate inputs rather than long-term capital investment as a proxy for unobserved productivity. A detailed exposition of the estimation procedure using the LP method is available in the working paper (online). The production function estimation by LP (OP and GMM) provides consistent estimates of input coefficients that can be

66

WORLD DEVELOPMENT Table 3. The share of FDI by industry in China: 2000–2003 (Unit:percent) Sector ID Sector name

2000

2001

2002

2003

Total

FDI Number of FDI Number of FDI Number of FDI Number of FDI Number of share FDI Firms share FDI firms share FDI firms share FDI firms share FDI firms 13 14 15 16 17 18 19 20

21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 39 40

41 42 Total

Food processing Food production Beverage production Tobacco processing Textile industry Clothing and footware Leather, fur, feather, and related products Timber processing, wood, bamboo, rattan, palm, and straw products Furniture manufacturing Paper and paper products Printing and medium reproduction Cultural, educational, and sports goods Petroleum refining, coking, and gas production and supply Raw chemical materials and chemical products Medical and pharmaceutical products Chemical fibers Rubber products Plastics products Nonferrous mineral products Smelting and pressing of ferrous metals Smelting and pressing of nonferrous metals Metal products General purpose machinery Special purpose machinery Transportation equipment Electrical machinery and equipment Communication equipment, computers, and other electronic equipment Instruments, meters, cultural, and clerical machinery Other manufacturing All manufacture

16.8 29.3 24.2 0.1 14.1 33.7 43

[1017] [765] [380] [4] [2003] [2705] [1153]

17.9 33.2 26 0.3 15.1 33.7 40.6

[1154] [823] [404] [4] [2274] [2984] [1305]

17.5 30.5 27.9 0.2 15.5 33.2 39.5

[1225] [876] [406] [4] [2450] [3203] [1423]

17.3 30.7 30.2 0.2 17 34.7 39.9

[1377] [951] [460] [6] [2812] [3670] [1658]

17.4 30.9 27.1 0.2 15.5 33.8 40.6

[1193] [854] [413] [5] [2385] [3141] [1385]

18.5

[453]

20.1

[531]

19.1

[529]

20.7

[610]

19.7

[531]

34.5 23.6 20.5

[399] [631] [442]

35.4 23.8 23.2

[436] [684] [488]

36.5 25 22.8

[479] [688] [504]

40.9 23.7 23.8

[548] [741] [556]

37.1 24 22.6

[466] [686] [498]

48.3

[770]

48.3

[820]

49.3

[917]

51

[1131]

49.4

[910]

4.8

[112]

4.5

[89]

7.6

[121]

7.1

[92]

6.2

[104]

15.4

[1274]

16.5

[1502]

17.4

[1596]

19.6

[1850]

17.4

[1556]

13.5 21.5 25.6 33.2 12.3 5.1

[464] [164] [319] [1730] [1245] [198]

12.3 16 25.8 32.3 13.1 5.6

[554] [154] [348] [1914] [1402] [223]

13.3 17 28.4 31.2 12.9 6.3

[545] [162] [342] [2010] [1445] [223]

13 13.8 27.2 33 11.9 7

[597] [181] [393] [2194] [1553] [245]

13 16.8 26.8 32.4 12.6 6

[540] [165] [351] [1962] [1411] [222]

7.7

[213]

8.3

[188]

8.2

[222]

9.5

[309]

8.5

[233]

27.8 15.9 10.6 20 25.1 47.7

[1336] [961] [651] [865] [1578] [1927]

26.2 15.2 13.7 20.9 26.4 55.1

[1525] [1093] [810] [998] [1819] [2167]

26.3 16.3 12.8 20.4 27.1 56.3

[1616] [1235] [823] [1103] [1963] [2352]

26 18.8 14.9 22.5 28 59.9

[1756] [1592] [929] [1224] [2163] [2712]

26.5 16.7 13.1 21 26.7 55.2

[1558] [1220] [803] [1048] [1881] [2290]

53.6

[542]

58.9

[608]

57.9

[628]

60.1

[781]

57.9

[640]

32.6 21.7

[1115] [25416]

32.1 22.8

[1276] [28577]

35.1 23.2

[1411] [30501]

31.1 24.1

[1364] [34455]

32.8 23

[1292] [29737]

Note: Numbers in the rectangle parenthesis are number of foreign firms. The share of FDI in an industry is Output-weighted. Sources: Authors’ calculation.

used to derive total factor productivity (TFP) at the firm level, LP LP that is ln TFP ijrt ¼ ln Y ijrt  aLP L ln Lijrt  aK K ijrt  aM ln M ijrt LP LP LP where aL ; aK , and aM are the LP estimates of the production function coefficients for labor, capital, and materials. Using the derived productivity as the dependent variable, 11 we estimate the impact of spillovers from FDI in an industry on the productivity of domestic firms as follows. ln TFP ijrt ¼ a0 þ a4 Horizontaljt þ a5 Backward jt þ a6 Forward jt þ aX þ aj þ ar þ at þ eijrt :

ð4Þ

(ii) Cluster effect The OLS estimates may overestimate the spillover effects of FDI on domestic firm productivity unless corrected for clustering. Moulton (1990; p.334), followed by Bertrand, Duflo,

and Mullainathan (2004), argues that “when one tends to use the aggregate market or public policy variables to explain the economic behavior of micro units, it is possible that the standard errors of estimated coefficients of those aggregate variables from OLS might be underestimated, which would lead to the overstated significance of coefficients.” The presence of group-level variables in such a “structural” model can be viewed as putting additional restrictions on the intercepts in separate-group models, which can cause the residual to deviate from the i.i.d. assumption. Failure to address this type of cluster error problem may cause a serious downward bias in the estimated errors, resulting in spurious findings of statistical significance for the aggregate variable of interest (industry-level FDI variables in this case). Javorcik (2004) uses a simple cluster-robust option to correct for any intra-group correlation in standard errors between

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

observations belonging to the same industry in a given year. Although this represents an improvement over previous studies that do not correct for cluster effects, the method of allowing for differences in the variance/standard errors due to arbitrary intra-group correlation has limitations (Woodridge, 2006). To illustrate the potential risk that the simple clusterrobust correction can bring about, we suppose there is a cluster effect in Eqn. (4). Then, the residual part can be decomposed into two components: eijrt ¼ ugjt þ vir . Thus, the variance of the residual in the regression could be written as: eijrt : re ¼ r2u þ r2v =M g ;

ð5Þ

where eijrt is the residual of Eqn. (4), re is the variance of eijrt, r2u is the variance of the inter-group residual (ugjt ), r2v is the variance of the intra-group residual (vir), and Mg is the number of observations in each group. In such a situation, the cluster-robust option will work only when ugjt is normally distributed with constant variance and when it dominates eijrt so that either r2v is small relative to r2u ; Mg is large, or both. However, in many FDI studies the number of groups (say, two-digit industries in a single time period) is small (M<<50) (that is r2v is small relative to r2u ) and there are very unbalanced cluster sizes in the sample (some Mg may be small) so that re may not be constant and dominated by r2u . 12 Therefore, the cure provided by the cluster-robust correction can be even worse than the disease, since using the wrong weights may bias the standard errors of the estimated coefficients in an unclear direction. 13 To properly correct for cluster effects in standard errors of the estimated coefficients, we follow a two-stage estimation procedure recently proposed by Woodridge (2006). In the first stage, we treat each industry-year as a group and regress firm productivity on firm-level variables within each group, separately controlling for regional effect. 14 The equation used for the first-stage estimation can be written as: ð6Þ ln TFP ir ¼ djt þ c fdi þ c MS ir þ ar þ vir ; 1

ir

2

where ln TFPir is firm i ’s total factor productivity in region r (given industry j at time t), the control variables, fdiir and MSir, are firm level variables as defined earlier. The ar are regional dummies capturing the regional disparity of domestic djt Þ) firms. The constant term ( djt ) and its standard error (seð are then extracted from each of these regressions, capturing firm characteristics at the industry-year group level, or firm’s industry characteristics. In the second-stage, we use weighted lest squares to estimate regressions of firm i’s industry characteristics on FDI variables, controlling for other factors, where group g is weighted by 1=½seð djt Þ2 . 15 Hence, groups for which there are more data and a smaller variance receive a greater weight, which is similar to M g =r2v (See Woodridge (2006, p.21)). In doing so, our estimation equation for the second stage becomes (controlling for concentration ratio of an industry (CR8):  djt ¼ b þ b Horizontaljt þ b Backward jt þ b Forward jt 0

4

5

þ b7 CR8jt þ aj þ at þ ujt :

67

studies) through the two-stage estimation. Third, it is compatible with all other methods (such as the instrumental variable approach and first differencing methods) which deal with the problem of simultaneity bias discussed below. (iii) Simultaneity bias Another threat to model identification is that there may be certain unobserved factors at the industry level, such as changes in business-cycle conditions or industry-wide implementation of new technologies that may affect domestic firms, but may be closely correlated with FDI in the industry. For example, FDI may flow into industries that are a priori more productive for reasons that are unclear. 16 An increase in the productivity of domestic firms in an industry could coincide with an increase in the presence of FDI in the industry, even though the increase may merely reflect an improvement in business-cycle conditions in that industry. Failure to account for omitted variables would lead to biased regression results, or simultaneity bias. Few papers on productivity spillovers have tackled this issue satisfactorily. First differencing and incorporating the fixed effect is the most commonly used method to deal with the simultaneity bias problem. Taking the first difference and including time dummy variables may remove any unobserved firm-specific, industry-specific, and region-specific effects. 17 By firstdifferencing, Eqn. (8) becomes: D djt ¼ b0 þ b4 DHorizontaljt þ b5 DBackward jt þ b6 DForward jt þ b7 DCR8jt þ at þ ujt :

ð8Þ

However, taking the first-difference and including time dummy variables may not remove the simultaneity bias problem arising from important unobservables that vary both across industries/regions and over time. We address the issue by using the instrumental variable (IV) approach. It is well-known that finding a strong instrument is difficult. For our purpose, we need to find a set of instrumental variables that are correlated with the FDI variables (namely, Horizontaljt, Backwardjt, and Forwardjt), but not the productivity of domestic firms. Our selection of instruments is guided by the empirical literature on FDI, which suggests that the level of China’s foreign direct investment is positively related to the levels of inward direct investments to economies in East and Southeast Asia and that the industrial structure between China and East and Southeast Asian countries are similar (Chantasasawat, Fung, Iizaka, & Siu, 2010). We thus choose the FDI variables from Southeast Asian countries as instruments for the corresponding FDI variables in China. 18 In sum, our final preferred approach to estimating spillovers of FDI in China takes the form of (8) but using the instrumental variable approach since it addresses the econometric problems we have discussed above. 4. ESTIMATION RESULTS

6

ð7Þ

Compared with the simple cluster–robust correction, Woodridge (2006) two-stage method has three advantages. First, it has some explicit assumptions for the intra-group and intergroup components in the random-error term, so that the cluster effect can be better controlled. Second, it helps to avoid the potential multi-collinearity and identification problems between regional dummies and industry dummies (where interaction terms between regional dummies and industry dummies should have been, but are not, incorporated in previous

(a) Basic results We now turn to our estimation results. To conserve space, we do not report results using output as the dependent variable. 19 Although Aitken and Harrison (1999) argue that a regression of output on FDI that controls for inputs allows for an estimate of productivity, there are concerns that the endogeneity of input choices may be a threat to identification. We thus report results using TFP as the dependent variable where TFP is derived using the LP method described in Section 3.

68

WORLD DEVELOPMENT Table 4. Productivity spillovers from FDI (Dependent variable: lnTFP)

Dependent variable:lnTFP Horizontal Forward Backward CR8 Region dummy Industry dummy Year dummy Constant Number of observations R2 F-statistics (weak instrument identification test)

(1) Simple OLS

(2) Simple OLS with simple cluster correction

(3) FDIV (no control of firm’s market power)

(4) FDIV (control of firm’s market power)

0.992*** (0.009) 0.929*** (0.157) 3.658*** (0.208) 0.848*** (0.015) Yes Yes Yes 0.935*** (0.007) 3,63,704 0.096 –

0.992** (0.448) 0.929 (9.643) 3.658 (12.655) 0.848 (0.717) Yes Yes Yes 0.935*** (0.205) 3,63,704 0.096 –

0.117*** (0.009) 1.967*** (0.072) 4.135*** (0.110) 0.022*** (0.001) Yes Yes Yes 0.143*** (0.000) 1,99,845 – 275.34

0.011*** (0.001) 0.157*** (0.010) 0.275*** (0.014) 0.201*** (0.007) Yes Yes Yes 0.046*** (0.002) 1,99,845  338.18

Note: Numbers in parenthesis are standard errors. Cluster errors in column (2) are estimated using general cluster-error correction option available in most econometrics packages while column (3) and (4) are estimated using Woodridge (2006)’s two-stage procedure in correcting cluster errors. Regression results in column (3) do not control for a firm’s market power in an industry at the first-stage estimation while those in column (4) do. FDIV refers to estimations using first differencing and instrumental variable methods. “CR8” is defined as the sum of the market share of the largest eight firms in an industry. The F-statistics reported in the above table for the FDIV regressions are the Kleiberge–Papp rk Wald F-statistics, which is used to test whether instrumental variables are strong for identifying the regression. The null hypothesis is that the function is weakly identified with the current instrumental variables. Usually, the F-statistics should be higher than 10 to reject the null hypothesis. ** Significance at 5%. *** Significance at 1%.

To compare with the literature, we report three sets of regression results in Table 4. Columns (1) and (2) provide two sets of regression results obtained using OLS: with and without corrections for cluster errors using the general cluster-robustness approach. These two specifications are most common in the literature and are close to those of Aitken and Harrison (1999), Javorcik (2004) and Haskel et al. (2007). Without correcting for cluster effects in the standard error, the presence of FDI has positive and statistically significant spillover effects on domestic firms in the same industry, as well as in the downstream industry. The spillover effect on domestic firms in the upstream industry is negative and statistically significant (column (1)). However, after correcting for cluster effects in the standard error using the general clusterrobust option available in most statistical packages, the spillover effect on domestic firms in the same industry remains positive and statistically significant while the spillover effect on domestic firms in the downstream industry becomes insignificant (column (2)), suggesting that the estimation results are sensitive to cluster effects, as pointed out by Moulton (1990; p.334). However, OLS is not able to deal with the simultaneity-bias problem wherein unobserved factors at the industry level that affect domestic firms’ productivity may also be closely correlated with FDI in the industry. Moreover, as discussed in Section 3, the cure provided by the cluster-robust correction can be even worse than the disease, since using the wrong weights may bias the standard errors of the estimated coefficients in an unclear direction. We thus use Woodridge (2006) approach to carefully deal with cluster effects and then employ first differencing with the instrumental variable method to correct for the simultaneity-bias problem. These results are reported in columns (3) and (4) of Table 4. By using FDI variables from Southeast Asian countries as

instruments for the corresponding FDI variables in China deals with time-variant unobserved factors that are correlated with the extent of FDI presence in an industry and the productivity of domestic firms in the same industry. 20 To isolate the effects of market competition (firm’s market power and market concentration in an industry) from productivity spillovers from FDI, we report these first differences with IV results with and without controlling for a domestic firm’s market share (MS) in an industry at the first stage regression (Woodridge, 2006; also see discussion in previous section) in columns (3) and (4), respectively. Without controlling for a firm’s market share, results from first difference with IV are similar to the OLS results with same signs for the three FDI variables but small coefficients (column (3)). Note that the coefficient of market concentration ratio (CR8), the control variable in our second stage regression, is negative after controlling for a firm’s market power in the first stage regression, as shown in column (4) of Table 4. This negative coefficient suggests that higher market concentration in an industry is associated with lower firm-level productivity, which is consistent with theory. (b) Horizontal spillovers However, controlling for a firm’s market share reverses the sign for Horizontal, which becomes negative. This suggests that the seemingly positive productivity spillovers to domestic firms from FDI presence in the same industry (Horizontal) that emerge from previous estimations in columns (1) to (3) is in fact to do with a firm’s market power. Controlling for the effects of a domestic firm’s market power, the presence of FDI in an industry has a negative and statistically significant effect on the productivity of domestic firms in the same industry. A one percentage point increase in the share of foreign firms in an industry leads to, on average, a 0.01%

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

productivity decrease in the productivity of domestic firms in the same industry. It is worth pointing out that these negative horizontal spillovers do not imply negative knowledge spillovers. It is well known that the coefficient for Horizontal, the spillover effects of FDI presence in the same industry on the productivity of domestic firms, captures both knowledge spillovers and competition effects from FDI presence (Nickell 1996; Aitken & Harrison 1999; Haskel et al. 2007). This is because the FDI variable (Horizontal) is commonly defined as the output weighted average of foreign equity participation in a firm. The value of this variable increases with the output of the FDI firms and the share of foreign equity in these firms. Therefore, the measure of FDI presence (Horizontal) has already reflected the market power of all FDI firms in the industry. Given the nature of this variable, it is difficult, if not impossible, to disentangle effects arising from knowledge spillovers from those due to competition. Controlling for a domestic firm’s market share in an industry removes the “market reallocation” effects among domestic firms (from less efficient domestic firms to more efficient ones) arising from intensified competition due to FDI’s presence, which may generate positive effects on domestic firms’ overall productivity. As a consequence, after controlling for a domestic firm’s market share, the competition effect only captures “market stealing” effects of FDI firms (but not “market reallocation” effects). Our finding of a negative coefficient for Horizontal thus suggests that knowledge spillovers within an industry are counterbalanced by the “market stealing” effects as in Aitken and Harrison (1999). 21 (c) Vertical spillovers There are clear positive spillovers from the presence of FDI firms in the upstream industry who supply high-quality, and perhaps lower cost intermediate goods or equipment to domestic firms (forward linkages, Forward). The signs of Forward are consistently positive across all specifications (Table 4), and as shown in column (4), a one percentage point increase in the share of foreign firms in an industry leads to, on average, a 0.16% productivity gain for domestic firms in the downstream industry. The spillover effects on domestic firms in the upstream industry (the coefficient of Backward) are negative and statistically significant. Again, this result is robust and consistent across all specifications (Table 4). As shown in column (4), a one percentage point increase in the share of foreign firms in an industry leads to, on average, a 0.28% decrease in the productivity of domestic firms in the upstream industry. The finding of negative backward spillovers to domestic firms is puzzling and worth further investigation. First, it is possible that the entry of FDI firms weakens the vertical linkages between FDI firms and domestic firms. After being taken over by foreign multinationals, domestic firms that were originally sourcing from firms in the upstream industry became foreign firms and the foreign firms might replace their purchase of intermediate inputs from the domestic market by sourcing from the international market. As discussed in Section 2, to promote exports of foreign firms, China offers import tariff and value-added tax (VAT) exemption for imported raw materials and parts used in the export processing. This tax incentive encourages foreign firms to purchase inputs from, and to export their output to, the international market. An unintended consequence of the tax incentive has been a weakening of vertical linkage between foreign firms and local Chinese firms, in particular, a lack of backward

69

linkages with those Chinese firms in the upstream industry. This weakening of domestic vertical linkages reduces the output of domestic firms in the upstream industry and pushes up average costs, leading to negative productivity effects. It is also possible that there exists a “market competition” effect. An increase in the share of foreign firms may result in monopsony which in turn introduces tougher competition among domestic firms in the upstream industry, thus lowering the productivity of domestic firms in the upstream industry. 22 To investigate why backward linkages between FDI firms and local firms in China are weak, we ask whether the negative backward spillovers arise from the nature of FDI firms. We pursue this issue in the following two dimensions. First, since firms produce different types of products and some products in nature have greater industry linkages than others, it is possible that FDI firms producing more sophisticated products generate spillovers to domestic firms in the upstream industry by direct knowledge transfer, through higher requirements for product quality (Javorcik 2004), while FDI firms producing simple homogenous products may have less strong backward linkages. If the types of products FDI firms produce matter, we would find negative backward effects only for FDI firms producing simple homogenous products but not others. To this end, we split our sample of FDI firms into three categories based on the type of goods they produce by following the methodology suggested by Rauch (1999). The three types of FDI firms are “organized-exchanges” goods producers, “reference prices” goods producers and “differentiated” goods producers. The “organized-exchanges” goods refer mainly to agricultural and resource products (mostly from Standard International Trade Classification Code (SITC) 0–4). The “reference prices” goods refer to the limited differentiated goods (mostly staple industrial materials such as chemical fiber, lead alloy, SITC 5–6) and the “differentiated” goods refer to other manufactured goods (clothing, electronics and machinery, mainly SITC 7–9). In doing so, we match Chinese Industrial Classification code (CIC) (for FDI firms) with the Standard International Trade Classification code (SITC revision 2) by using Rauch’s conservative classification in the concordance table. 23 Though this classification is made according to each product’s trading manner in the market, it reflects different degrees of sophistication in the production process and, therefore, the different degrees of industrial linkages for firms producing those goods. We expect that firms producing “differentiated” goods have more vertical linkages with other firms than firms producing “organized-exchanges” goods. Based on this classification, we first re-calculate the three sets of FDI variables, namely, Horizontal, Backward, and Forward for each of the three categories of FDI firms, differentiated by their product features. We then run regressions using each of the three sets of FDI variables. Our objective is to determine whether FDI firms producing more sophisticated products generate positive spillovers to domestic firms in the upstream industry. We report the results from the three sets of regressions in Table 5. Consistent with the main results in Table 4, the coefficients for Horizontal are all negative and mostly significant. Surprisingly, the backward spillovers of FDI firms producing more sophisticated products are negative. This suggests that it may not be the type of goods FDI firms produce that matters for vertical linkages. Another possibility is that the export-orientation of FDI firms matters for spillovers. An important feature of FDI firms in China is that exporting FDI firms tend to source intermediate inputs from the international market, as encouraged by

70

WORLD DEVELOPMENT Table 5. Productivity spillovers by FDI firms classified by type of goods produced

Dependent variable:lnTFP Horizontal Forward Backward CR8 Constant Number of observations F-statistics (weak instrument identification test)

(1) Homogenous goods

(2) Referenced goods

(3) Differentiated goods

0.099*** (0.007) 0.124*** (0.021) 1.537*** (0.077) 1.841*** (0.061) 0.026*** (0.001) 1,99,845 197.11

0.020 (0.065) 0.283 (0.307) 0.857 (0.652) 2.651** (1.099) 0.009 (0.005) 1,99,845 11.68

0.003*** (0.000) 0.001 (0.001) 0.017*** (0.000) 1.040*** (0.035) 0.020*** (0.001) 1,99,845 1782.91

Note: Numbers in parenthesis are standard errors. Cluster errors in column (2) are estimated using general cluster-error correction option available in most econometrics packages while column (3) and (4) are estimated using Woodridge (2006)’s two-stage procedure in correcting cluster errors. Regression results in column (3) do not control for a firm’s market power in an industry at the first-stage estimation while those in column (4) do. FDIV refers to estimations using first differencing and instrumental variable methods. “CR8” is defined as the sum of the market share of the largest eight firms in an industry. The F-statistics reported in the above table for the FDIV regressions are the Kleibergen–Papp rk Wald F-statistics, which is used to test whether instrumental variables are strong for identifying the regression. The null hypothesis is that the function is weakly identified with the current instrumental variables. Usually, the F-statistics should be higher than 10 to reject the null hypothesis. ** Significance at 5%. *** Significance at 1%.

China’s exemption of import tax policy. Imports by foreign firms accounted for almost 59% of China’s total imports while exports by foreign firms accounted for 57% of China’s total exports in 2007, as discussed in Section 2. We would, therefore, expect that FDI firms that mainly export to international markets would have weaker domestic vertical linkages than those FDI firms serving domestic market. To test whether exporting FDI firms have weaker domestic vertical linkages, we re-calculate the three sets of FDI variables, namely, Horizontal, Backward, and Forward for each of the three categories of FDI firms differentiated by their export orientation: (1) FDI firms that serve domestic market; (2) FDI firms that report positive export values; (3) FDI firms that have export values higher than 50% of their output. We seek to identify whether there is a weaker domestic vertical linkage for firms that are very export-oriented than those FDI firms that serve domestic markets only. Table 6 reports regression results for spillovers from FDI firms by their export orientation. There is a clear pattern of weaker linkages between high-exporting FDI firms and domestic firms. FDI firms that serve domestic market generate positive forward spillovers while high-exporting FDI firms exert negative horizontal as well as vertical spillovers to domestic firms. Therefore, it is quite likely that the negative backward spillovers that we find in our main results arise from the unique feature of export-oriented FDI in China. (d) Firm size, ownership, export-orientation, and FDI spillovers The above results suggest that on average domestic firms do benefit from FDI firms in the upstream industry but are negatively affected by FDI firms in the same industry as well as those in the downstream industry. It is possible that there are significant heterogeneities across different types of firms affected by the presence of FDI firms. Domestic firms may have varying degrees of absorptive capacities as well as responses to heightened competition from the presence of FDI firms. Thus, the spillovers from FDI presence may vary across types of domestic firms. In this section we consider the relationship between three attributes of domestic firms and their benefits from the presence of FDI firms: firm size (large and

medium-sized firms vs. small firms), export orientation (export-oriented vs. domestic-market-oriented firms) and firm ownership (state-owned enterprises (SOEs) vs. non-SOE firms). 24 First, consider firm size. It is possible that large domestic firms have better capacities to absorb knowledge spillovers from FDI firms as they have more resources (for example, formal R&D capacity or more skilled workers) than small domestic firms. However, whether large firms are better equipped with the ability to compete with FDI firms than small firms is less obvious. To explore whether domestic firm size matters for spillovers from the presence of FDI, we split the sample into two subsamples: large and medium-sized firms vs. small firms. According to the National Bureau of Statistics, a firm is classified as large and medium-sized firm if it satisfies the following three criteria: (1) annual sales value of 30 million renminbi or above, and (2) 300 employees or above; and (3) asset values of 40 million renminbi or above. A firm that does not satisfy any one of the three criterias is classified as a small firm. We report our regression results in columns (1) and (2) in Table 7. These regression results are obtained by controlling for a domestic firm’s market share (MS) in an industry at the first stage regression and taking first difference with IV methods at the second stage regression (Woodridge, 2006), as in previous subsection. Our results suggest that there is a significant difference between large and medium-sized firms and small firms in absorbing the spillovers from the presence of FDI firms. Large and medium-sized firms benefit positively and strongly from spillovers of FDI firms located in all industry. A one percentage point increase in the share of foreign firms in an industry leads to, on average, a 0.08%, 0.74%, and 1.51% increase in the productivity of domestic large and medium-sized firms in the same industry, downstream industry, and upstream industry, respectively. In contrast, small firms only benefit positively from spillovers from FDI firms located in their upstream industry but are affected negatively by FDI firms located in the same industry or their downstream industry. Second, consider domestic firm ownership. As shown columns (3) and (4) of Table 7, the presence of FDI firms benefits

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

71

Table 6. Productivity spillovers by export orientation of FDI firms (1) Nonexporting FDI

(2) Exporting FDI

(3) High Exporting FDI

0.463*** (0.062) 0.560*** (0.097) 0.747*** (0.101) 7.604*** (1.121) 0.086*** (0.012) 1,99,845 208.50

0.002* (0.001) 0.123*** (0.013) 0.220*** (0.009) 1.183*** (0.058) 0.106*** (0.005) 1,99,845 18.76

0.102*** (0.006) 0.328*** (0.019) 0.394*** (0.014) 2.405*** (0.097) 0.066*** (0.003) 1,99,845 165.21

Dependent variable: lnTFP Horizontal Forward Backward CR8 Constant Number of observations F-statistics (weak instrument identification test)

Note: Numbers in parenthesis are standard errors. Cluster errors in column (2) are estimated using general cluster-error correction option available in most econometrics packages while column (3) and (4) are estimated using Woodridge (2006)’s two-stage procedure in correcting cluster errors. Regression results in column (3) do not control for a firm’s market power in an industry at the first-stage estimation while those in column (4) do. FDIV refers to estimations using first differencing and instrumental variable methods. “CR8” is defined as the sum of the market share of the largest eight firms in an industry. The F-statistics reported in the above table for the FDIV regressions are the Kleibergen–Papp rk Wald F-statistics, which are used to test whether instrumental variables are strong for identifying the regression. The null hypothesis is that the function is weakly identified with the current instrumental variables. Usually, the F-statistics should be higher than 10 to reject the null hypothesis. * Significance at 10%. *** Significance at 1%.

Table 7. FDI productivity spillovers and domestic firm heterogeneity

Dependent variable: lnTFP Horizontal Forward Backward CR8 Constant No. of obs. F-statistics (weak instrument identification test)

(1) Large and medium-sized firm

(2) Small firm

(3) SOEs

(4) Non-SOEs

(5) Exporting firms

(6) Nonexporting firms

0.080*** (0.002) 0.737*** (0.148) 1.513*** (0.232) 0.567*** (0.172) 0.249*** (0.022) 82,106 50.11

0.038*** (0.001) 0.219*** (0.022) 0.444*** (0.025) 0.519*** (0.020) 0.035*** (0.002) 1,93,651 117.38

0.031*** (0.001) 1.084*** (0.059) 1.477*** (0.076) 0.390*** (0.031) 0.082*** (0.005) 49,185 27.25

0.126*** (0.048) 1.725** (0.830) 1.923** (0.881) 1.938*** (0.139) 0.052 (0.035) 2,38,716 217.41

0.813*** (0.006) 0.937*** (0.012) 0.755*** (0.042) 30.917*** (0.261) 0.227*** (0.003) 24,095 673.24

0.025*** (0.002) 0.346*** (0.032) 0.359*** (0.040) 1.688*** (0.020) 0.030*** (0.004) 1,75,750 67.15

Note: Numbers in parenthesis are standard errors. Cluster errors in column (2) are estimated using general cluster-error correction option available in most econometrics packages while column (3) and (4) are estimated using Woodridge (2006)’s two-stage procedure in correcting cluster errors. Regression results in column (3) do not control for a firm’s market power in an industry at the first-stage estimation while those in column (4) do. FDIV refers to estimations using first differencing and instrumental variable methods. “CR8” is defined as the sum of the market share of the largest eight firms in an industry. The F-statistics reported in the above table for the FDIV regressions are the Kleibergen–Papp rk Wald F-statistics, which are used to test whether instrumental variables are strong for identifying the regression. The null hypothesis is that the function is weakly identified with the current instrumental variables. Usually, the F-statistics should be higher than 10 to reject the null hypothesis. ** Significance at 5%. *** Significance at 1%.

non-SOEs in the same industry, and those that are in the downstream industry of foreign firms. Consistent with the regression results from the whole sample; foreign firms have a negative impact on domestic firms in the upstream industry. In contrast, SOEs do not benefit from the presence of FDI firms in the same industry or in the upstream industry. In fact the spillover effects are negative and statistically significant. These results suggest that the overall spillovers mask significant differential impacts of FDI on domestic firms, with non-SOEs benefiting most from the presence of foreign multinationals. Finally, there are differences in the spillovers to domestic firms differentiated by export orientation. As can be seen in

the last two columns of Table 7, domestic exporting firms benefit from the presence of FDI firms located in the same industry as well as in their downstream industry but are affected negatively by FDI firms located in their upstream industry. On the other hand, domestic nonexporting firms only benefit from FDI firms located in their upstream industry and are affected negatively by FDI firms located in other industries. The results indicate that whether a firm is exporting or not has significant implications for their benefits from FDI. Taken together, domestic firms differ significantly in the extent to which they benefit from FDI, with large and medium-sized firms, non-SOE and exporting firms accruing the greatest benefits from foreign firms in China.

72

WORLD DEVELOPMENT

(e) Robustness check

5. CONCLUSION

In this section we check whether our results are robust to different methods for deriving total factor productivity. We use the Levinsohn and Petrin (2003) (LP) method in this study, and in this section determine whether our results are sensitive to other popular methods, specifically, the Olley and Pakes (1996) (OP) and GMM methods for deriving total factor productivity. To address this concern, we re-estimate firm productivity (TFP) using the OP and GMM methods. In OP, “investment” (instead of “intermediate inputs”) is used as a proxy for unobserved productivity. To conserve space, we refer readers to Olley and Pakes (1996) and others for detailed techniques in deriving TFP using OP. We report regression results using firm level TFP as a dependent variable, derived from OP, GMM, and LP (value-added production function) in Table 8. Again, the results are obtained after correcting for cluster effects using Woodridge (2006) two-stage estimation method and dealing with simultaneous bias issues using first difference and IV methods. As shown in Table 8, both coefficients of Horizontal and Backward are negative and significant and we find positive and statistically significant coefficient for Forward variable. These results are obtained for all three alternative productivity estimations (OP, GMM, and LP value added methods) and are consistent with the specification using LP gross output value method in column (4) of Table 4, thus confirming our main results. We also experiment with an alternative measure of productivity by using domestic firms’ output as the dependent variable while controlling for labor, capital and intermediate inputs. For FDI variables, we have re-run regressions using their first-period lag and second-period lag variables to examine whether the results are sensitive to time-lag effects. Also, we considered additional scenarios by controlling for additional firm-level variables such as the number of products, firms’ age, and share of new products in total revenue. Our results are robust to all different specifications.

China has emerged as the largest recipient of foreign direct investment (FDI) in the developing world, yet little is known about the benefits from FDI for domestic firms. Using firm-level census data from China for the period of 2000–2003, this paper examines the various channels of FDI spillovers. We find that FDI has had a significant positive impact on productivity of domestic firms that purchase high-quality intermediate goods with lower input prices, or equipment from FDI firms in the upstream industry. However, we find negative horizontal effects in the case of China after controlling for firm-level market share. The expected positive knowledge spillovers of FDI firms in the same industry as domestic firms are counterbalanced by competition effects arising from the entry of FDI firms. The finding of negative backward spillovers is puzzling. Our further investigation suggests that this may be the result of a set of unique Chinese FDI policies that encourage foreign firms to import raw materials and equipment from the international market. Further, we find that the more export-oriented firms have weaker vertical linkages. Moreover, firms in China do not benefit uniformly from foreign investment, with large and medium-sized firms, non-SOE and exporting firms accruing the greatest benefits from the foreign presence in China. The positive forward spillovers to domestic firms from FDI suggest that Chinese governments’ preferential treatment of foreign firms in past decades may be justifiable. However, the negative backward linkage effect we identify in this paper requires further study when longer time series data are available. For a long time the Chinese government has provided exemption of import duty to those firms that import equipment and inputs, provided that those firms exported their goods to the international market, for fear of a shortage of foreign exchange. This export incentive, although successful in encouraging exports, comes with an implicit cost by reducing the incentive for foreign firms to source domestically (i.e.,

Table 8. Alternative estimations on productivity spillovers from FDI

Dependent variable:lnTFP Horizontal Forward Backward CR8 Constant Number of observations F-statistics (weak instrument identification test)

OP

GMM

LP (value added)

0.024*** (0.001) 0.168*** (0.028) 0.283*** (0.031) 0.808*** (0.040) 0.047*** (0.002) 1,50,788 126.52

0.018*** (0.001) 0.159*** (0.027) 0.266*** (0.030) 0.932*** (0.037) 0.045*** (0.002) 1,50,860 93.23

0.022*** (0.004) 0.861*** (0.098) 0.752*** (0.117) 0.872*** (0.071) 0.065*** (0.014) 2,31,178 225.30

Note: OP refers to using Olley and Pakes (1996) method in deriving total factor productivity at the firm level while GMM uses generalized methods of moments. LP(value added) uses LP method but applies it to the value added function (excluding intermediate inputs) instead of gross output value function. Numbers in parenthesis are standard errors. Cluster errors in column (2) are estimated using general cluster-error correction option available in most econometrics packages while column (3) and (4) are estimated using Woodridge (2006)’s two-stage procedure in correcting cluster errors. Regression results in column (3) do not control for a firm’s market power in an industry at the first-stage estimation while those in column (4) do. FDIV refers to estimations using first differencing and instrumental variable methods. “CR8” is defined as the sum of the market share of the largest eight firms in an industry. The F-statistics reported in the above table for the FDIV regressions are the Kleibergen–Papp rk Wald F-statistics, which is used to test whether instrumental variables are strong for identifying the regression. The null hypothesis is that the function is weakly identified with the current instrumental variables. Usually, the F-statistics should be higher than 10 to reject the null hypothesis. *** Significance at 1%.

PRODUCTIVITY SPILLOVERS FROM FOREIGN DIRECT INVESTMENT

in China) parts, intermediate inputs, and equipment. Now that China’s foreign exchange reserves have reached about US$2 trillion, concerns about a shortage of foreign exchange are remote and a rethinking of the tax incentive may be warranted. However, notwithstanding the government policy, the benefits from FDI are not automatic, but also require some support

73

for domestic firms. Perhaps more R&D is needed to increase the quality of domestic products to induce foreign firms to source products from the domestic market. Furthermore, as non-SOE firms accrue greater benefits from foreign firms in China than SOE firms, there is an incentive to continue the reform of SOEs.

NOTES 1. Du et al. (2010) is one of the first studies of FDI spillovers in China using census firm-level data. 2. Source: http://www.fdi.gov.cn. 3. China also allows imported inputs sold to downstream firms to be exempt from import tariff and VAT tax as long as they are processed for export. 4. The nine sectors with the most significant expansion of foreign firms are furniture (183%), chemical materials and products (128%), ferrous metal smelting (297%), non-ferrous metal smelting (193%), general machinery (145%), special machinery (206%), transport equipment (134%), electronics and telecommunications equipment (146%), and instruments (169%), most of which are capital-intensive and technologyintensive industries (China Statistical Yearbook, various years). 5. The Concentration Ratio and the Herfindahl index are two standard indicators to measure market concentration (whether an industry is comprised of a few large firms or many small firms). The former is defined as the sum of the market share of the largest firms (usually take 4 or eight firms) while the latter is defined as the sum of squared market shares of all firms. A small Concentration Ratio or Herfindahl index indicates that the industry is competitive. In this paper, we use the Eight-firm Concentration Ratio (CR 8) as a measure of market concentration. 6. Firms with missing value of major variables such as fixed assets, labor, and output are dropped from our dataset.

12. See Blalock and Gertler (2007) for the argument on the overcorrection of the cluster effect for FDI studies. 13. Moreover, in Javorcik (2004), the introduction of industry dummies into the regression between firms’ productivity and the FDI variables at the industry level tends to reduce the freedom of estimation leading to over-identification in the regression. 14. We treat each industry in each year as a group rather than each industry over time as a group because our observations on FDI at the industry level are changing over time for each sector. 15. If we assume that Zg is the group-specific effect and ug is the residual ^FE Þ ¼ EðP Z0g Z g Þ1 ^ from the second-stage estimation, we have: Avarrð b P P ^FE Þ ¼ EðP Z0g Z g Þ1 ðP Z 0 Xg Z g Þ ^ ðP Z 0g ^ug ^u0g Z g Þð Z 0g Z g Þ1 o r Avarrð b g ð Z 0g Z g Þ1 when G is large. If a cluster effect arises due to correlation ^ among intra-group firms, we have Avarð^ ug Þ ¼ Eð^ug ^u0g Þ ¼ EðXg Þ ¼ r2g þ ¼ r2u =M g . Given that the first-stage estimation yields r2u =M g , the OLS with the analytical weight correction adjusts Xg by dividing it by r2u =M g . The ^ adjusted Avarð^ ug Þ ¼ r2g =r2u  M g þ 1, which may be biased when M g =r2u is small. Thus, we use frequency weights in applying weighted least squares. The correlation between the weights and group size is high (0.74). 16. Sometimes this problem is also referred to as the endogeneity of FDI. See, e.g., Hale and Long (2007).

7. FDI firms refer to those firms that are wholly owned by foreign investor or in the case of Chinese-foreign equity joint venture, the foreign investor has no less than 25% share in the joint venture. The 25% threshold is set by The law of P.R.C. on Chinese-Foreign Equity Joint Ventures and its implementation regulations. All other firms are referred to as domestic firm.

17. Although time differencing removes unobservable factors that are not changing over time while inclusion of the industry and time dummy variables in the first-difference specification controls for unobserved factors that may be driving changes in the attractiveness of a given industry or year, it may not remove those factors at the firm level that may change over time. Our LP method deals with unobservable factors changing over time at the firm level, such as quality of management, which may not be fixed over time within firms.

8. Some studies have used the industry-specific price index to deflate firm output, which may not be appropriate as it imposes a strong assumption that all firms faced the same prices (see Klette and Griliches (1996) for a discussion).

18. Note that the Backwardjt and Forwardjt for Southeast countries are calculated based on the Chinese input–output table.

9. This follows Aitken and Harrison (1999) and Javorcik (2004). To test for robustness, we also provide an alternative measure of FDI in a sector by calculating the weighted sum of foreign capital, with the weight being each firm’s share of capital in the sector. 10. There is a significant revision in industry classification in 2003 initiated by the NBS. The 2003 input–output table uses the new industry classification and is not consistent with industry classification for firms before 2003. Since our data covers the period 2000 to 2003, it is important to use the 1997 input–output table which is consistent with a firm’s classification of industry. 11. Both the OLS and LP estimates of the production function are shown in Table 3.

19. Results using output as the dependent variable are available from the authors on request. 20. Note that our instrumental variable has the desirable property of being strong predictor of the endogenous variable. The first stage F-test for weak identification is well above 10, a value commonly suggested as a sign of variables to be strong instruments. 21. Without controlling for a firm’s market share in an industry, the competition effect captures both “market stealing” effects by FDI firms as well as market reallocation effects among domestic firms. As in the working paper version of this paper, the net effect on Horizontal from FDI knowledge spillovers, FDI “market stealing” effects as well as market reallocation effects among domestic firms is positive.

74

WORLD DEVELOPMENT

22. It is sometimes argued that the increase in measured productivity may reflect the degree of market concentration in the industry. To control for the degree of market competition, we include the Eight-firm Concentration Ratio (CR8) in all regressions. But as argued in previous paragraphs, the nature of the FDI variables makes it difficult, if not impossible, to distinguish the effects arising from knowledge spillovers from those due to competition effects, even after controlling for market competition at the industry level.

23. Available from Jon Haveman’s International trade database (http:// www.macalester.edu/research/economics/page/haveman/trade.resources/ tradedata.html#Rauch).

24. The state-owned enterprises (SOE) are defined as domestic firms that the state or collective owner has at least 50% share in the registered capital. Other domestic firms are classified as non-SOE firms.

REFERENCES Aitken, B., & Harrison, A. (1999). Do domestic firms benefit from foreign direct investment? Evidence from panel data. American Economic Review, 89(3), 605–618. Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates?. Quarterly Journal of Economics, 119(1), 249–276. Blalock, G., & Gertler, P. J. (2007). Welfare gains from foreign direct investment through technology transfer to local suppliers. Journal of International Economics, 74(2), 402–421. Chantasasawat, B., Fung, K. C., Iizaka, H., & Siu, A. (2010). FDI flows to Latin America, East and Southeast Asia and China: Substitutes or complements?. Review of Development Economics, 14(3), 533– 546. Du, L., Harrison, A., and Jefferson, G. (2010). Do institutions matter for FDI spillovers? The implications of China’s “special characteristics”, NBER working paper No. 16767. Findlay, R. (1978). Relative backwardness, direct foreign investment, and the transfer of technology: A simple dynamic model. Quarterly Journal of Economics, 92(1), 1–16. Fleisher, B., Li, H., & Zhao, M. (2010). Human capital, economic growth, and regional inequality in China. Journal of Development Economics, 92(2), 215–231. Hale, G. and Long C. (2007). Are there productivity spillovers from foreign direct investment in China? Federal Reserve Bank of San Francisco Working Paper series. Haskel, J. E., Pereira, S., & Slaughter, M. (2007). Does inward foreign direct investment boost the productivity of domestic firms?. Review of Economics and Statistics, 89(3), 482–496. Holz, C. (2005). China Governance Project: The institutional arrangements for the production of statistics, OECD Statistics Working Paper no. JT00177141. Hu, A. G., & Jefferson, G. H. (2002). FDI impact and spillover: Evidence from China’s electronic and textile industries. The World Economy, 25(8), 1063–1076. Javorcik, B. S. (2004). Does foreign direct investment increase the productivity of domestic firms? In search of spillovers through backward linkages. The American Economic Review, 94(3), 605– 627. Klette, T., & Griliches, Z. (1996). The inconsistency of common scale estimators when output prices are unobserved and endogenous. Journal of Applied Econometrics, 11, 343–361.

Lin, P., Liu, Z., & Zhang, Y. (2009). Do Chinese domestic firms benefit from FDI inflow? Evidence of horizontal and vertical spillovers. China Economic Review, 20, 677–691. Levinsohn, J., & Petrin, A. (2003). Estimating production functions using inputs to control for unobservables. Reviews of Economic Studies, 70(2), 317–342. Lopez, R. (2008). Foreign technology licensing, productivity, and spillovers. World Development, 3(4), 560–574. Moulton, B. R. (1990). An illustration of a pitfall in estimating the effects of aggregate variables on micro units. The Review of Economics and Statistics, 72(2), 334–338. Nickell, S. (1996). Competition and corporate performance. Journal of Political Economy, 104, 724–746. Olley, S. G., & Pakes, A. (1996). The dynamics of productivity in the telecommunications equipment industry. Econometrica, 64(4), 1263–1297. Rodrik, D. (1999). The new global economy and developing countries: Making openness work. Overseas Development Council (Baltimore, ND) Policy Essay No. 24. Rauch, J. E. (1999). ‘Networks versus markets in international trade’. Journal of International Economics, 48(1), 7–35. Sun, S. (2009). How does FDI affect domestic firms’ exports? Industrial evidence. The World Economy, 32(8), 1203–1222. Suyanto, S., Salim, R., & Bloch, H. (2009). Does foreign direct investment lead to productivity spillovers? Firm level evidence from Indonesia. World Development, 37, 1861–1876. UNCTAD. (2008). World Investment Reports 2008. UN Conference for Trade and Development. Wang, C. and Zhao, Z. (2008). Horizontal and vertical spillover effects of foreign direct investment in Chinese manufacturing. Journal of Chinese Economic and Foreign Trade Studies (1), 8–20. Wei, Y., & Liu, X. (2006). Productivity spillovers from R&D, exports and FDI in China’s manufacturing sector. Journal of International Business Studies, 37, 544–557. Woodridge, J. (2006). Cluster-sample methods in applied econometrics: An extended analysis. Mimeo. Zhao, T. (2004). Comparison between Industrial Classification in Chinese National Accounts and International Standard Industrial Classification for All Economic Activities. Paper presented to the 8th OECD–NBS Workshop on National Accounts, 6–10 December 2004. Paris: OECD Headquarters.