Testing for horizontal and vertical foreign investment spillovers in China, 1998–2007

Journal of Asian Economics 23 (2012) 234–243

Contents lists available at ScienceDirect

Journal of Asian Economics

Testing for horizontal and vertical foreign investment spillovers in China, 1998–2007§ Luosha Du a,*, Ann Harrison b, Gary H. Jefferson c a

Department of Agricultural and Resource Economics, University of California, Berkeley, 310 Giannini Hall, #3310, Berkeley 94720-3310, United States The World Bank, 1818 H Street, N.W., Washington, DC 20433, United States c Brandeis University, P.O. Box 9110, Waltham, MA 02454, United States b

A R T I C L E I N F O

A B S T R A C T

Article history: Received 25 March 2010 Received in revised form 24 December 2010 Accepted 9 January 2011 Available online 14 January 2011

As with many developing countries, the Chinese government hopes that knowledge brought by multinationals will spill over to domestic industries and increase their productivity. In this paper, we show that foreign investment originating outside of Hong Kong, Macau, and Taiwan has positive effects on individual firm level productivity, while foreign investment from HKMT firms does not. We also test for both horizontal (within the same industry) and vertical (upstream or downstream) linkages from foreign investment. Using a manufacturing firm-level panel for 1998 through 2007, we find zero or weak positive horizontal externalities. However, our results show that foreign direct investment (FDI) has generated positive productivity spillovers to domestic firms via backward linkages (the contacts between foreign affiliates and their local suppliers in downstream sectors) as well as forward linkages (between foreign suppliers and their local buyers in the upstream sectors). ß 2011 Elsevier Inc. All rights reserved.

JEL classification: F2 L6 O1 O3 Keywords: Foreign direct investment Horizontal spillover effect Backward and forward linkages China’s manufacturing industries

1. Introduction Foreign direct investment (FDI) now accounts for an important source of capital inflows into developing countries, having increased from $22 billion in 1990 to about $200 billion annually in recent years. Developing countries are major recipients of global inward FDI, currently attracting about one third of total global inward FDI.1 FDI is an attractive source of global finance, because it is relatively stable compared to other capital flows and can also introduce advanced technologies. With rapid expansion in FDI throughout the world economy, the role of multinational enterprises in technology transfer and spillovers is receiving increasing attention. Technology spillovers in this paper are defined to take place when the entry or presence of multinationals increases the productivity of domestic firms, and the multinationals do not fully internalize the value of these benefits. Spillovers should be external to firms’ total factor productivity after controlling for inputs. We define intra-industry spillovers (also called

§ All the authors gratefully acknowledge the conference organizers at Fudan University, particularly to Peter Petri. We also would like to thank our discussant at the Fudan Conference, and an anonymous referee for excellent suggestions. This work is partially based on work supported by the National Science Foundation under grant no. 0519902. * Corresponding author. Tel.: +1 5105265612. E-mail addresses: [email protected] (L. Du), [email protected] (A. Harrison), [email protected] (G.H. Jefferson). 1 Obtained from ‘‘Trade, Foreign Investment and Industrial Policy’’, by Ann Harrison and Andres Rodriguez-Clare, forthcoming in the Handbook of Development Economics, September 2009.

1049-0078/$ – see front matter ß 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.asieco.2011.01.001

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

235

horizontal spillovers) to occur when domestic firms are affected by foreign corporations located in the same sector, while inter-industry spillovers (vertical spillovers) occur when foreign investors enhance the productivity of domestic firms through vertical linkages. Vertical linkages are possible through both backward linkages (between foreign affiliates and their local suppliers) and forward linkages (between foreign suppliers and their local buyers). There have been a number of recent papers that test for either horizontal or vertical linkages (see, for example Aitken and Harrison (1999), Hu and Jefferson (2002), Javorcik (2004), and Lin, Liu, and Zhang (2009)). Most of this literature finds mixed evidence on horizontal spillovers, with a number of authors even documenting negative spillovers. In this paper, we confirm that horizontal spillovers are not significant in Chinese manufacturing, suggesting that competition from foreign-owned firms in the same sector has neither strong positive nor negative effects on domestic plant productivity. In contrast, recent studies generally find support for positive vertical spillovers from downstream foreign firms to upstream suppliers (so-called backward linkages). In this paper, we find evidence of vertical linkages via both forward and backward channels. Since opening its economy to the outside world in late 1978, China has absorbed an increasing amount of FDI. It is now among the world’s largest hosts of FDI inflows. With the entry of China into the World Trade Organization, China’s economy is becoming increasingly open to foreign investors. Potential positive spillover effects are indeed the main rationale behind the Chinese government’s aggressive efforts over the past two decades to attract foreign investment to China (Hu & Jefferson, 2002). However, few studies estimate the impact of FDI spillovers on China’s economy. Hu and Jefferson (2002) are the first authors to use firm-level data to study horizontal FDI spillovers in China. They find that FDI reduces productivity and market share of domestic firms in the short run. One recent manuscript that investigates both horizontal and vertical FDI spillovers in China is Lin et al. (2009). In contrast to Javorcik (2004), they find bigger forward and smaller backward spillovers. Our results differ from theirs, in part because we focus on total factor productivity and they examine value-added productivity and also use a different estimation method. We use an approach to productivity measurement pioneered by Olley and Pakes (1996). This leads to bigger backward linkages and smaller forward linkages. Our empirical strategy follows Javorcik (2004) and Olley and Pakes (1996). First, we use Javorcik (2004)’s empirical strategies to calculate Backward and Forward linkages and follow her estimation models to test whether there are vertical FDI spillovers in the manufacturing sector in China, and if so, through what mechanism these spillovers occur. By using a rich set of data collected by the Chinese National Statistical Bureau from its annual survey of industrial firms with annual sales of more than five million yuan,2 we find that positive and significant vertical spillovers occurred through both backward and forward linkages. These results are consistent with the small set of case studies on vertical externalities from foreign investment on other countries: all previous research outside of China finds that vertical linkages operated through downstream linkages between domestic suppliers and foreign users of their inputs. We address the endogeneity of inputs by applying the strategy proposed by Olley and Pakes (1996). To address the potential problem of sector-level FDI variables and domestic plant productivity being jointly determined, we also explore the robustness of our results to using lagged measures of foreign investment at the sector level. We apply a variety of specifications to take into account firm-specific fixed effects, and find that our results are robust to these alternative approaches. The results indicate significant heterogeneity across firms in responding to foreign investment. Specifically, we find that the sources of foreign investment matter. Firms with equity participation from other foreign sources, principally the OECD economies, show positive effects on individual firm level productivity; while foreign investment from Hong Kong, Macao or Taiwan does not. The rest of paper is organized into five further sections. Section 2 summarizes previous empirical and theoretical work in this field. In Section 3, we describe the data used in this paper and review broad trends for the 1998 through 2007 period. Section 4 lays out the empirical strategy, discusses the econometric issues, and presents the empirical results. Section 5 concludes.

2. Literature review Research on the impact of FDI for domestic firms dates back to the study of Caves (1974), who examined FDI in manufacturing sectors within Canada and Australia. He found that productivity levels are higher for domestic Australian firms that compete in industries with a higher FDI presence. Following his pioneering work, Globerman (1979) also found some weak evidence of positive FDI spillovers. Summarizing these earlier studies,3 Blomstrom (1986) did not find evidence in support of the hypothesis that foreign investment speeds up the transfer of any specific technology to Mexico. The main disadvantage of these studies is that they all use industry-level data, which fails to disentangle the direction of causality between foreign presence and productivity improvement.

2

The data set includes data for all state-owned enterprises, regardless of whether they meet the 500 million yuan threshold or not. As Blomstrom summarizes, three available econometric studies deal with the influence of foreign investment on the technical efficiency of host country firms. One focuses on Australia (Caves, 1974), one on Canada (Globerman, 1979), one on Mexico (Blomstrom & Pearson, 1983). All three find some support for the spillover benefit hypothesis, but none of the studies analyzes the nature of spillover efficiency in depth. 3

236

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

As firm-level data became available in developing countries, a number of studies tested for FDI spillovers in the 1990s. Most of these studies, such as papers by Haddad and Harrison (1993) on Morocco, Aitken and Harrison (1999) on Venezuela, and Konings (2001) on Bulgaria, Romania and Poland, either failed to find evidence of horizontal spillovers or reported negative horizontal spillover effects. Aitken and Harrison (1999) showed that the earlier findings in support of positive spillover were likely to be driven by the endogeneity of FDI; FDI may choose to go to better performing industries. Once industry-specific factors are controlled for, there is no evidence of positive spillovers in their study on Venezuela. Another possibility, noted by Javorcik (2004), is that researchers have been looking for positive FDI spillovers in the wrong place. Since multinationals have an incentive to prevent information leakage that would enhance the performance of their local competitors, but at the same time may benefit from transferring knowledge to their local suppliers or clients, negative spillovers from FDI are more likely to be horizontal and positive spillovers are more likely to be vertical in nature. Javorcik uses firm-level data from Lithuania to show that positive FDI spillovers take place through backward linkages (between foreign affiliates and their local suppliers in upstream sectors); however, there is no robust evidence of positive spillovers occurring through either the horizontal or the forward linkage channel. There are also a set of theoretical studies about the impact of FDI on host-country industrial organization that demonstrate that positive FDI spillovers are more likely to operate at the inter-industry rather than the intra-industry level. For instance, there are studies on the choice by multinationals to use FDI as a model of market penetration. These studies emphasize apparent efforts to minimize the probability of imitation, especially under imperfect intellectual property rights in the host country. As Markusen and Venables (1998) point out, proximity to potential domestic competitors with absorptive capacity to reverse engineer proprietary technology would be detrimental to a multinational, causing subsidiaries to set up where potential rivals cannot erode its market share. By contrast, the multinational can benefit from knowledge diffusion when it reaches downstream clients and upstream suppliers, which will encourage vertical flows of generic knowledge that lead to inter-industry spillovers.

3. Data The data set employed in this paper was collected by Chinese National Bureau of Statistics.4 The Statistical Bureau conducts an annual survey of industrial plants, which includes manufacturing firms as well as firms that produce and supply electricity, gas, and water. It is firm-level based, including all state-owned enterprises (SOEs), regardless of size, and nonstate-owned firms (non-SOEs) with annual sales of more than 5 million yuan. We use a ten-year unbalanced panel dataset, from 1998 to 2007. The number of firms per year varies from a low of 162,033 in 1999 to a high of 336,768 in 2007. The sampling strategy is the same throughout the sample period (all firms that are state-owned or have sales more than 5 million RMB are selected into the sample); the variation of numbers of enterprises across years may be driven by changes in ownership classification or by increases (or reductions) in sales volume to in relation to the 5 million yuan threshold. However, the data show that 5 million yuan is not a strict rule. Among non-SOEs, about 6 percent of the firms report annual sales of less than 5 million yuan in 1998; this number rises to 8 percent by 1999 and falls after 2003. In 2007, only 1 percent of non-SOEs have annual sales below 5 million RMB. In terms of the full sample, the percent of firms with sales less than 5 million RMB stays at the same level for 1998 and 1999 and starts falling in 2000. In 2007, around 2 percent of the sample consists of firms with annual sales less than 5 million yuan. The original dataset includes 2,226,104 observations and contains identifiers that can be used to track firms over time. Since the study focuses on manufacturing firms, we eliminate non-manufacturing observations. The sample size is further reduced by deleting missing values, as well as observations with negative or zero values for output, number of employees, capital, and the inputs, leaving a sample size of 1,842,786. Due to incompleteness of information on official output price indices, three sectors are dropped from the sample.5 Thus, our final regression sample size is 1,552,557. The dataset contains information on output, fixed assets, total workforce, total wages, intermediate input costs, foreign investment, Hong Kong–Taiwan–Macau investment, sales revenue and export sales. These are the key variables from which we obtain measures of firm-level foreign asset shares and the FDI spillover variable, which are discussed in detail in the next section. In this paper, to test the impact of FDI spillovers on domestic firms’ productivity, we use the international criteria of 10% ownership to distinguish domestic firms and foreign owned firms, that is, domestic firms are those for which the share of subscribed capital owned by foreign investors is equal to or less than 10 percent.6 In the dataset, 1,210,804 observations (about 79% of the sample) fall below this threshold.7

4

The dataset we employed in this paper was purchased from Beijing Hongtao Times Information Consulting Co., Beijing, China. They are sectors of processing food from agricultural products; printing, reproduction of recording media; and general purpose machinery. 6 China’s National Bureau of Statistics employs 25% as the threshold for designating firms as ‘‘foreign-owned’’. However, to be consistent with Javorcik’s (2004) empirical strategy, we use 10% as the threshold. We have explored whether the results are sensitive to using 10% and 25% as the threshold and find that the magnitudes and the significance levels of the coefficients change little between these two thresholds. 7 Without dropping three sectors (that have missing output price indices), 1,461,475 observations meet this threshold. 5

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

237

4. Estimation and results To examine the correlation between a firm’s productivity and intra- and inter-industry FDI spillovers, we employ the following model inspired by Aitken and Harrison (1999) and Javorcik (2004) as the baseline specification in this study: lnY i jrt ¼ a þ b1 lnK i jrt þ b2 lnLi jrt þ b3 lnM i jrt þ b4 Foreign Sharei jrt þ b5 Horizontal jt þ b6 Backward jt þ b7 Forward jt þ ai þ at þ ei jrt Yijr stands for output value (quantities  prices) for firm i in sector j at region r and time t, which is deflated by the official sector-specific ex-factory industrial output price indices.8 Kijrt, capital, is defined as the value of fixed assets, which is deflated by the fixed assets investment index, and Lijrt is the total number of employees. Mijrt represents the intermediate inputs purchased by firms to use for production of final products, which is deflated by the intermediate input price index.9 Calculation of sector-specific price deflators is described in the data appendix. Firm-level foreign share, Foreign Shareijrt is defined as the share of the firm’s total equity owned by foreign and Hong Kong–Taiwan–Macau investors. By construction, ‘‘Foreign Share’’ is a continuous variable and has a range from 0 to 1.10 We separate the foreign share into two types: that which originates in Hong Kong, Taiwan, or Macao, and all other types, which is FDI that largely originates in the OECD countries. The motivation for this is two fold. First, we are curious to see whether some types of foreign investment are more likely to result in technology spillovers than others. Second, anecdotal evidence suggests large quantities of so-called foreign investors in China are actually domestic investors who channel investment through Hong Kong in order to take advantage of special treatment for foreign firms (so-called ‘‘round tripping’’). If this is the case, then we would expect that foreign investment of this type should have no special impact on domestic firms. Following Javorcik (2004), we define three sector-level FDI variables. First, Horizontaljt captures the extent of foreign presence in sector j at time t and is defined as foreign equity participation averaged over all firms in the sector, weighted by each firm’s share in sectoral output. In other words, "P

i for all i 2 j Foreign Shareit

P

Horizontal jt ¼

 Y it

#

i for all i 2 j Y it

Second, Backwardjt, captures the foreign presence in the sectors that are supplied by sector j.11 Therefore, Backwardjt is a measure for the foreign participation in the downstream industries of sector j. It is defined as Backward jt ¼

X

a jk Horizontalkt

ki fk 6¼ j

ajk is taken from the 2002 input–output table representing the proportion of sector j’s production supplied to sector k. Finally, Forwardjt is defined as the weighted share of output in upstream industries of sector j produced by firms with foreign capital participation. As Javorcik points out, since only intermediates sold in the domestic market are relevant to the study, goods produced by foreign affiliates for exports (Xit) should be excluded. Thus, the following formula is applied:

Forward jt ¼

X mi fm 6¼ j

d jm

P

  ðY it  X it Þ i for all i 2 m ðY it  X it Þ

i for all i 2P m Foreign Shareit

djm is also taken from the 2002 input–output table. Since Horizontaljt already captures linkages between firms within a sector, inputs purchased within sector j are excluded from both Backwardjt and Forwardjt. To begin the analysis, we estimate the model described above using ordinary least square (OLS) with firm fixed effects and time effects. The dependent variable is the log of the firm’s deflated output. According to Javorcik (2004), knowledge externalities from the presence of foreign firms may take time to manifest themselves. Therefore, two specifications are employed: one with contemporaneous and one with lagged spillover variables. The estimation is performed on the full 8 The official sector-specific price indices are taken from Urban Life and Price Yearbook 2008, Table 4-3-3 Ex-factory price indices of Industrial Products by Sector (1985–2007). According to the definition of the website of National Bureau of Statistics, ‘‘ex-factory’’ price index suggests the prices of first sale of industrial products. 9 The price indices for fixed assets and intermediate inputs are industry-wide, which are taken from China Statistical Yearbook 2008, Table 8-1 Price indices (1978–2007). The index is used to deflate fixed assets is called ‘‘investment in fixed assets’’ and the index used to deflate intermediate inputs is called the ‘‘purchasing price indices of raw material, fuel and power’’. 10 In some specification, we run regressions with domestic firms only. In these cases, we use the 10% threshold discussed in the last section to separate out domestic firms. Then we regress either the log of the firm’s output or productivity on sector-level FDI without the variable ‘‘Foreign Share’’. 11 For instance, both the furniture and apparel industries use leather to produce leather sofas and leather jackets. Suppose the leather processing industry sells 1/3 of its output to furniture producers and 2/3 of its output to jacket producers. If no multinationals produce furniture but half of all jacket production comes from foreign affiliates, the Backward variable will be calculated as follows: 1/3  0 + 2/3  1/2 = 1/3.

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

238 Table 1 Summary statistics for all years, 1998–2007.

Summary statistics for levels

Summary statistics for first differences

Number of

Mean

Standard deviation

Number of observations

Mean

Standard deviation

10.003 4.759 8.442 9.498 0.081 0.075 0.245 0.076 0.102

1.351 1.149 1.701 1.389 0.255 0.243 0.129 0.044 0.164

1,302,015 1,302,015 1,302,015 1,302,015 1,302,015 1,302,015 1,302,015 1,302,015 1,302,015

0.134 0.012 0.107 0.095 0.001 0.0003 0.004 0.002 0.004

0.564 0.504 0.756 0.666 0.148 0.141 0.045 0.015 0.064

observations lnY lnL lnK lnM Foreign share (contributed by HK–Taiwan–Macau) Foreign share (contributed by other countries) Horizontal Backward Forward

1,842,786 1,842,786 1,842,786 1,842,786 1,842,786 1,842,786 1,842,786 1,842,786 1,842,786

Notes: All the variables calculated in this table are based on annual survey of industrial plants (1998–2007), conducted by Chinese National Bureau of Statistics (refers to footnote 7 for the source of data). The data are at firm-level, including all state-owned enterprises (SOEs), regardless of size, and nonstate-owned firms (non-SOEs) with annual sales of more than 5 million yuan. Y, L, K, M represent output value, number of employees, capital, and intermediate input respectively. Y, K, and M are deflated by corresponding price indices and therefore all of them are in real terms. We define firm-level foreign share according to its different sources. Foreign share contributed by HK–Taiwan–Macau is defined as the share of firms’ total equity owned by investors from HK–Taiwan–Macau. Foreign share contributed by other countries is defined as the share of firms’ total equity owned by investors outside HK–Taiwan–Macau, principally from OECD countries. Horizontal captures the intra-industry FDI spillover while backward and forward catch the interindustry FDI spillover (see detailed discussion in Section 4).

sample and on the sample of domestic firms only. The specification becomes the following when the estimation is performed on the sample of domestic firms only: lnY i jrt ¼ a þ b1 ln K i jrt þ b2 ln Li jrt þ b3 ln M i jrt þ b4 Horizontal jt þ b5 Backward jt þ b6 Forward jt þ ai þ at þ ei jrt To study the impact of FDI spillovers on the performance of domestic firms, we are interested in how FDI invested in other firms affect the domestic firms located in the same sector. Therefore, the key parameters in the above specification are b4, b5 and b6. Table 1 describes the summary statistics for the main variables used in the regressions. We provide means on levels and first differences. The first difference results show that growth in output has average 13.4 percent per year. Most of that growth has been fueled by increases in the capital stock, which grew on average 10.8 percent per year, and material inputs, which grew 9.5 percent per year. Labor inputs actually grew very little, at only 1.2 percent per year, suggesting that growth of output was not skewed towards labor use. In Table 2, we provide summary statistics for the three spillover variables in each year. All measures increased significantly during the sample period. Foreign share increased from 20 to 27 percent between 1998 and 2007. Backward linkages also increased, as did forward linkages. These changes allow us to identify the effects of horizontal, backward, and forward linkages on domestic enterprises. 4.1. Baseline results Table 3 presents the results from OLS regressions. The first three columns present the results using contemporaneous spillover variables. To take into account the possible endogeneity of the foreign share, as well as the possibility that spillovers Table 2 Summary statistics for spillover variables. Backward

Horizontal

Forward

Year

Number of sectors

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

71 71 71 71 71 71 71 71 71 71

0.204 0.226 0.238 0.241 0.244 0.250 0.266 0.269 0.270 0.266

0.162 0.175 0.182 0.189 0.188 0.199 0.203 0.210 0.206 0.206

0.068 0.076 0.082 0.082 0.084 0.087 0.093 0.095 0.096 0.095

0.064 0.071 0.076 0.077 0.080 0.084 0.090 0.092 0.091 0.090

0.047 0.055 0.060 0.060 0.063 0.068 0.070 0.074 0.077 0.075

0.077 0.105 0.117 0.120 0.126 0.143 0.147 0.161 0.163 0.159

Notes: Horizontal, backward, and forward are calculated based on the same dataset as that in Table 1 (the annual survey of industrial plants, conducted by Chinese National Bureau of Statistics). Horizontal captures the intra-industry FDI spillover while backward and forward catch the inter-industry FDI spillover (see detailed discussion in Section 4). This table shows the basic summary statistics by year.

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

239

Table 3 OLS with contemporaneous and lagged spillover variables, 1998–2007 (using 29-industry specific official output deflators, dropping 3 sectors: processing food from agricultural products; printing, reproduction of recording media; general purpose machinery).

Log L Log K Log M Foreign share (by HK–Taiwan–Macau) Foreign share (by other countries) Horizontal

All firms

Domestic (10% foreign share)

Domestic (0 foreign share)

All firms

Domestic (10% foreign share)

Domestic (0 foreign share)

0.091*** (0.004) 0.028*** (0.002) 0.766*** (0.007) 0.002

0.082*** (0.003) 0.025*** (0.002) 0.776*** (0.007)

0.082*** (0.003) 0.025*** (0.002) 0.776*** (0.007)

0.093*** (0.004) 0.027*** (0.002) 0.761*** (0.007) 0.003

0.083*** (0.004) 0.025*** (0.002) 0.771*** (0.008)

0.083*** (0.004) 0.025*** (0.002) 0.771*** (0.008)

0.075 (0.095)

0.082 (0.098)

0.083 (0.098)

0.976*** (0.301)

0.931*** (0.315)

0.925*** (0.315)

0.251*** (0.085) 2.169*** (0.049) 1,455,873 0.828

0.196** (0.093) 2.122*** (0.054) 1,132,970 0.833

0.195** (0.092) 2.125*** (0.054) 1,127,471 0.833

(0.003) 0.008**

(0.003) 0.008*** (0.003) 0.127 (0.090)

(0.003) 0.132 (0.091)

0.133 (0.0907)

Horizontal_lagged Backward

0.776*** (0.273)

0.730** (0.277)

0.724** (0.277)

Backward_lagged Forward

0.206** (0.079)

0.152* (0.087)

0.151* (0.0864)

Forward_lagged Constant Observations R-squared

2.133*** (0.044) 1,552,557 0.831

1.699*** (0.045) 1,209,973 0.836

2.092*** (0.048) 1,203,687 0.835

Notes: Robust clustered standard errors are included in the parentheses. The dependent variable is the log of firm-level output, which is deflated by sector-specific ex-factory industrial output deflators (published by Urban Life and Price Yearbook 2008). Since the relevant data are incomplete for 3 sectors – processing food from agricultural products; printing, reproduction of recording media; and general purpose machinery – these sectors are excluded. In addition to the firm-level foreign share, three FDI spillover variables, firm-specific effect and time dummies are included in all regressions.

take time to be realized, we also present in the last three columns a specification using lagged spillover variables. Columns (1) and (4) present the results for all enterprises while the remaining columns restrict the analysis to firms with less than 10 percent foreign equity participation. In columns (2) and (5), domestic is defined as firms with less than 10 percent foreign equity, while in columns (3) and (6) domestic firms are defined as those with no foreign equity. All specifications allow for firm-specific effects and year effects. It is evident from columns (1) and (4) that foreign equity at the firm level is only associated with higher productivity for foreign participation originating outside of Hong Kong, Taiwan, and Macao (HTM). These results suggest that spillover benefits to individual firms are associated only with investments by foreign owners generally associated with the OECD economies. There may be several explanations for this. Foreign investment originating outside of Hong Kong, Taiwan, and Macao is generally more recent; perhaps the more recent vintage of foreign investors is more likely to convey new technology. Secondly, there is anecdotal evidence suggesting that some percentage of foreign investment from Hong Kong, Taiwan, and Macao actually originates in China and returns via these countries in order to exploit the benefits accorded to foreign investors; this is so-called ‘‘round-tripping’’ foreign investment. Consistent with the previous literature, our results show that there are no positive horizontal spillovers from foreign investment. However, by contrast, using either contemporaneous or lagged foreign shares, forward linkages do have a positive significant impact on productivity. This implies that enterprises benefit from externalities through foreign firms that are upstream to their operations, providing inputs of either higher quality or lower cost. Backward linkages also have a positive significant impact on productivity. These results are consistent with the literature on vertical linkages for other countries. The evidence suggests that externalities are generated when Chinese suppliers are linked with foreign buyers, leading to higher productivity for the Chinese suppliers. If we replace contemporaneous variables with lagged variables, the point estimates are higher for both vertical linkages, which indicates that lagged foreign investment has stronger effects on firms’ productivity. Our research suggests that foreign investors put significant downward pressure on the prices

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

240

of supplier industries, with somewhat less pressure on prices in the same industry. Failing to account for these price effects would lead to erroneous findings of productivity declines associated with horizontal linkages and with vertical (backward) linkages. 4.2. Correcting for endogeneity of input choice (Olley and Pakes) The earlier literature on production function estimation shows that the use of OLS is inappropriate when estimating productivity, since this method treats labor, capital and other input variables as exogenous. As Griliches and Mairesse (1995) argue, inputs should be considered endogenous since they are chosen by a firm based on its productivity. Firm-level fixed effects will not solve the problem, because time-varying productivity shocks can affect a firm’s input decisions. Using OLS will therefore bias the estimates of coefficients on the input variables. To solve the simultaneity problem in estimating a production function, we employ the procedure suggested by Olley and Pakes (1996) (henceforth OP), which uses investment as a proxy for unobserved productivity shocks. OP addresses the endogeneity problem as follows. Let’s consider the following Cobb–Douglas production function in logs: yit ¼ bk kit þ bl lit þ bm mit þ vit þ eit yit, kit, lit, and mit represent log of output, capital, labor, and materials, respectively. vit is the productivity and eit is the error term (or a shock to productivity). The key difference between vit and eit is that vit affects firm’s input demand while the latter does not. OP also makes timing assumptions regarding the input variables. Labor and materials are free variables but capital is assumed to be a fixed factor and subject to an investment process. Specifically, at the beginning of every period, the investment level a firm chooses together with the current capital value determine the capital stock at the beginning of the next period, i.e., kitþ1 ¼ ð1  s Þkit þ iit The key innovation of the OP estimation method is to use the firm’s observable characteristics to model a monotonic function of firm productivity. Since the investment decision depends on both productivity and capital, OP formulates investment as follows, iit ¼ iit ðvit ; kit Þ Given that this investment function is strictly monotonic in vit, it can be inverted to obtain

vit ¼ f t 1 ðiit ; kit Þ Substituting this into the production function, we get the following, yit ¼ bk kit þ bl lit þ bm mit þ f t

1

ðiit ; kit Þ þ eit ¼ bl lit þ bm mit þ ft ðiit ; kit Þ þ eit

The first stage of the OP estimation method yields consistent estimates of the coefficients on labor and materials as well as the estimate of a non-parametrical term (ft). The second step of OP provides an unbiased estimate of the coefficient on capital.12 The biggest disadvantage of applying the OP procedure is that 70% of the firms report zero or negative investment. To satisfy the monotonic relationship between investment and productivity, 70% of the sample has to be dropped. To address this problem, we construct our investment variable as the sum of the growth of capital stock and the current depreciation. With Olley and Pakes’s correction, we can get an unbiased estimate of the firm’s productivity. Therefore, the independent variable then becomes total factor productivity (TFP) instead of the log of output. Specifically, this is a two-stage estimation procedure when using TFP as the dependent variable. The first step is to use OP to obtain unbiased coefficients on input variables and then to calculate TFP (residual from the production function). The second step is to regress TFP on firm-level controls and FDI variables. Moulton (1990) showed that in the case of regressions performed on micro units that also include aggregated market (in this case industry) variables, the standard errors from OLS will be underestimated. As Moulton demonstrated, failing to take account of this serious downward bias in the estimated errors results in spurious findings of the statistical significance for the aggregate variable of interest. To address this issue, the standard errors in the paper are clustered for all observations in the same industry. Table A.2 compares the coefficients on the factor shares using both the OLS and the OP methods. The results indicate that the OP method successfully corrects the endogeneity problem of the input variables except for the coefficient on

12 The unbiasedness of the OP estimate on capital is based on two important assumptions. One is the first-order Markov assumption of productivity, vit and the timing assumption about kit. The first-order Markov assumption decomposes vit into its conditional expectation at time t  1, E[vit|vit1], and a deviation from that expectation, zit, which is often referred to the ‘‘innovation’’ component of the productivity. These two assumptions allow it to construct an orthogonal relationship between capital and the innovation component in productivity, which is used to identify the coefficient on capital.

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

241

Table 4 Results from OLS, Olley and Pakes regression (using 29-industry specific official output deflators, dropping 3 sectors: processing food from agricultural products; printing, reproduction of recording media; general purpose machinery). OLS

Log L Log K Log M Foreign share (by HK–Taiwan–Macau) Foreign share (by other countries) Horizontal Backward Forward Constant

Observations R-squared

Olley and Pakes

All firms

Domestic (10% foreign share)

Domestic (0 foreign share)

All firms

Domestic (10% foreign share)

Domestic (0 foreign share)

0.091*** (0.004) 0.028*** (0.002) 0.766*** (0.007) 0.001

0.082*** (0.003) 0.025*** (0.002) 0.776*** (0.007)

0.082*** (0.003) 0.025*** (0.002) 0.776*** (0.007)

0.088*** (0.002) 0.043*** (0.001) 0.771*** (0.004) 0.002

0.068*** (0.003) 0.043*** (0.002) 0.785*** (0.005)

0.068*** (0.002) 0.043*** (0.002) 0.785*** (0.005)

(0.003) 0.007**

(0.00) 0.008*** (0.003) 0.127 (0.090) 0.776*** (0.273) 0.206** (0.079) 2.133*** (0.044)

0.132 (0.091) 0.730** (0.277) 0.152* (0.087) 1.699*** (0.045)

0.133 (0.091) 0.724** (0.277) 0.151* (0.086) 2.092*** (0.048)

(0.003) 0.128 (0.090) 0.772*** (0.271) 0.204** (0.078) 1.703*** (0.018)

0.134 (0.090) 0.728** (0.275) 0.150* (0.087) 1.538*** (0.018)

0.135 (0.091) 0.720** (0.275) 0.149* (0.086) 1.908*** (0.022)

1,552,557 0.831

1,209,973 0.836

1,203,687 0.835

1,552,557 0.179

1,209,973 0.165

1,203,687 0.164

Notes: Robust clustered standard errors are included in the parentheses. The specification for the first three columns is the same as that in Table 3. When we apply Olley and Pakes regressions, the estimation procedure is actually two-stage. The first step is to use OP to obtain unbiased coefficients on input variables and then calculate TFP (residual from the production function). The second step is to regress TFP on firm-level controls and FDI variables. In addition to the firm-level foreign share, three FDI spillover variables, firm-specific effect and time dummies are included in all regressions.

material input. In theory, the OP procedure will increase the coefficient on capital and decrease the coefficients on labor and material compared to OLS results. In Table A.2, we summarize the estimated coefficients on the input variables by both OLS with fixed effects and the OP method. As shown, the changes in ln L and ln K follow the prediction, but the estimate of ln M changes in the opposite direction. What is unusual across both the OLS and OP specifications is that the labor share is very low in both regressions (compared to estimates for other countries) while the coefficient for input costs is extremely high. As a robustness check, we performed two tests. First, we calculated the share of labor expenditures in total output—the labor share in output according to the data. Under certain plausible restrictions (i.e., Cobb–Douglas production function, constant returns to scale, perfect competition) the coefficient on the factor inputs in our estimating equations should be equal to the factor shares. Labor’s share over the sample period is around 10 percent, which is very close to the coefficients we estimate on labor. Second, we compare the implied average wages from our sample (calculated by dividing total wages by the number of employees with average wages reported in the Chinese Statistical Yearbook (1998–2007)). The results are listed in Appendix Table A.1. From Table A.1, we can see that the average wages from the dataset are close to that from the statistical yearbook, although there are some discrepancies for 1998. We note that while the coefficient estimates for our OLS specification are plausibly close to the computed factor shares, this is only the case if one takes into account firm fixed effects in the estimation. In unreported results we model the fixed effect at the industry, and not the firm level, which yields OLS estimates with coefficient estimates which are even smaller for labor and implausibly high for material inputs. In Table 4, we reproduce the OLS results from the previous table in the first three columns and present the results with the OP correction in the last three columns. The results are consistent with Table 3. The point estimates are close to those reported in Table 3, particularly for backward and forward linkages which remain large and statistically significant, the approach also yield very precise estimates for the own firm productivity effect. In Table 5, we test the robustness of these estimates using derived input price deflators, which use output price indices and the input–output table for manufacturing industries. The results are consistent with Tables 3 and 4, although the magnitudes of the coefficients fall for all the FDI variables. However, the effects on backward and forward linkages still remain positively significant. The evidence in Table 5, combined with the previous results, suggests that externalities stemming from FDI yield observable productivity benefits through backward linkages (the contacts between foreign affiliates and their local suppliers in downstream sectors) as well as forward linkages (between foreign suppliers and their local buyers in the upstream sectors).

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

242

Table 5 OLS with contemporaneous and lagged spillover variables, 1998–2007 (using 29-industry specific official output deflators and derived input deflators, dropping 3 sectors: processing food from agricultural products; printing, reproduction of recording media; general purpose machinery).

Log L Log K Log M Foreignshare (by HK–Taiwan–Macau) Foreignshare (by other countries) Horizontal

All

Domestic (10% foreign share)

Domestic (0 foreign share)

All

Domestic (10% foreign share)

Domestic (0 foreign share)

0.088*** (0.004) 0.027*** (0.002) 0.771*** (0.007) 0.002

0.080*** (0.003) 0.024*** (0.002) 0.780*** (0.007)

0.080*** (0.004) 0.024*** (0.002) 0.780*** (0.007)

0.090*** (0.004) 0.027*** (0.002) 0.766*** (0.008) 0.004

0.082*** (0.004) 0.024*** (0.002) 0.776*** (0.008)

0.082*** (0.004) 0.024*** (0.002) 0.775*** (0.008)

0.016 (0.060)

0.010 (0.061)

0.011 (0.061)

0.379** (0.165)

0.402** (0.178)

0.399** (0.178)

0.102*** (0.037) 2.070*** (0.054) 1,456,660 0.832

0.085** (0.038) 2.020*** (0.062) 1,133,673 0.837

0.085** (0.038) 2.023*** (0.062) 1,128,171 0.837

(0.003) 0.006**

(0.003) 0.007** (0.003) 0.057 (0.057)

(0.003) 0.0489 (0.057)

0.049 (0.057)

Lagged horizontal Backward

0.273* (0.152)

0.295* (0.159)

0.291* (0.159)

Lagged backward Forward

0.078** (0.035)

0.062* (0.036)

0.062* (0.036)

Lagged forward Constant Observations R-squared

2.026*** (0.049) 1,553,344 0.835

1.980*** (0.056) 1,210,676 0.840

1.984*** (0.056) 1,204,387 0.840

Notes: Robust clustered standard errors are included in the parentheses. This table tests the robustness of previous results using derived input price deflators, which uses output price indices and the input–output table for manufacturing industries. In addition to the firm-level foreign share, three FDI spillover variables, firm-specific effect and time dummies are included in all regressions.

5. Concluding comments and suggestions for future work Given the intensive interest in FDI as a vehicle through which developing countries learn about new technology, many firm-level studies have investigated the presence of FDI spillovers. However, due to the lack of accessibility of firm-level data, few studies test for FDI spillovers in China. This paper, based on a rich firm-level dataset from China, tests particularly for vertical spillovers from FDI by applying the approaches of Javorcik (2004) and Olley and Pakes (1996). Taking into account the heterogeneity of firms’ absorptive capacities, we show that foreign investment originates principally from OECD sources (outside of Hong Kong, Macao, and Taiwan) has positive impacts on individual firm-level productivity. Across a variety of specifications, and controlling for firm and year effects, we find that positive productivity spillovers from FDI take place between foreign affiliates and their local clients in upstream sectors (backward linkages) and that there are also positive productivity spillovers from foreign suppliers to their domestic buyers (forward linkages). We also highlight the different effects played by the source of the foreign equity investment on domestic firm productivity. While foreign equity participation is generally associated with higher productivity at the firm level, this is not the case for foreign equity participation that originates in Hong Kong, Macao, or Taiwan. There are several possible explanations for this. One major reason could be that such investments actually originate in China, and are simply rechanneled through nearby locations to take advantage of special incentives offered to foreign investors. Another possible explanation is that nearby foreign investors are not sufficiently different technologically. Certainly more research is needed to fully understand the effect of FDI on host countries. In the case of China, the uniqueness of that country’s geography, institutions, and market conditions argue for differentiating our research strategy to address these special characteristics. Our research has done that in distinguishing between the firm-level productivity effects of investment originating from Hong Kong, Macao, and Taiwan versus that which originate primarily in OECD countries. In addition, more attention should be given to different spillover dynamics that operate in different regions of China in which local institutions and policies, the concentration and composition of FDI, and proximity to other contiguous or near-contiguous economies differ markedly. Thus, it will be more interesting if we can perform some disaggregated studies to calculate spillover effects in specific regions or sectors. Additional research exploring the role of trade policies and heterogeneity across ownership types would also be or particular interest.

L. Du et al. / Journal of Asian Economics 23 (2012) 234–243

243

Appendix A Tables A.1 and A.2.

Table A.1 Average wages comparison. Year

Mean of average wages from the data

Average wages from the national statistical yearbook (for the manufacturing industry)

Average wages from the national statistical yearbook (for SOEs manufacturing firms)

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

9795 8072 9038 10,329 10,586 11,002 13,588 14,087 16,925 19,957

7064 7794 8750 9774 11,001 12,496

6981 7611 8554 9590 10,876 12,601

15,757 17,966 20,884

16,963 20,317 23,913

Notes: Wages are measured in yuan/year for one person. To obtain means of average wages of the sample, we first calculate the average wage for each firm in each year by dividing total wages by the number of total employees then take the means of these averages. Since the official information from 2004 is missing, we leave them as blanks (for the second and third rows).

Table A.2 Summary of estimated elasticities of three input variables. OLS with firm fixed effects

ln L ln K ln M

Olley–Pakes

All firms

Domestic (10% foreign share)

Domestic (zero foreign share)

All firms

Domestic (10% foreign share)

Domestic (zero foreign share)

0.091 0.028 0.766

0.082 0.025 0.776

0.082 0.025 0.776

0.088 0.043 0.771

0.068 0.043 0.785

0.068 0.043 0.785

Notes: In this table, we compare two methods of calculating coefficients of inputs. With OLS specification, we regress ln Y on ln L, ln K, ln M and along with firm and year fixed effects. For each of two methods, we show input coefficients computed based on three samples: all firms; domestic firms with 10% or more foreign investment to their total equity; and domestic firms with zero foreign investment.

References Aitken, B. J., & Harrison, A. E. (1999). Do domestic firms benefit from direct foreign investment? Evidence from Venezuela. American Economic Review, 89(June (3)), 605–618. Blomstrom, M. (1986). Foreign investment and productive efficiency: The case of Mexico. Journal of Industrial Economics, 35, 97–110. Blomstrom, M., & Pearson, H. (1983). Foreign Investment and Spillover Efficiency in an Underdeveloped Economy: Evidence from the Mexican Manufacturing Industry. World Development, 11, 493–501. Caves, R. E. (1974). Multinational firms, competition and productivity in host-country markets. Economica, 41(May (162)), 176–193. Globerman, S. (1979). Foreign direct investment and ‘spillover’ efficiency benefits in Canadian manufacturing industries. Canadian Journal of Economics, 12(February), 42–56. Griliches, Z., & Mairesse, J. (1995). Production Functions: The Search for Identification. National Bureau of economic research (Cambridge, MA) working paper No. 5067, March. Haddad, M., & Harrison, A. E. (1993). Are there positive spillovers from direct foreign investment? Evidence from panel data for morocco. Journal of Development Economics, 42(October (1)), 51–74. Hu, A., & Jefferson, G. (2002). FDI impact and spillover: Evidence from China’s electronic and textile industries. World Economy, 25(8), 1063–1076. Javorcik, B. (2004). Does foreign direct investment increase the productivity of domestic firms? In search of spillovers through backward linkages. American Economic Review, 94(3), 605–627. Konings, J. (2001). Effects of foreign direct investment on domestic firms. Economics of Transition, 9(November (3)), 619–633. 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. Markusen, J., & Venables, A. (1998). Multinational firms and the new trade theory. Journal of International Economics, 46(2), 183–203. Moulton, B. R. (1990). An illustration of a pitfall in estimating the effects of aggregate variables on micro units. Review of Economics and Statistics, 72(May (2)), 334–338. Olley, G. S., & Pakes, A. (1996). The dynamics of productivity in the telecommunication equipment industry. Econometrica, 64(6), 1263–1297.