Foreign direct investment and relative wages

China Economic Review 12 (2001) 40 ± 57

Foreign direct investment and relative wages: The case of China Yaohui ZHAO China Center for Economic Research, Beijing University, Beijing 100871, China Accepted 17 February 2001

Abstract According to the conventional wisdom, foreign direct investment (FDI) can raise relative wages of skilled labor in a host country by bringing in skill-biased technology. This paper proposes an alternative hypothesis that in an economy characterized by labor market segmentation and high labor mobility costs, FDI could increase relative wages of skilled labor even without bringing in skill-biased technology. Chinese urban household survey data are used to test the hypothesis. We first estimate relative wages in Chinese state-owned enterprises (SOEs) and foreign-invested enterprises (FIEs) by correcting for possible sample selectivity caused by employment choice between SOEs and FIEs. Employment choice is then examined to provide evidence of the costs of labor mobility. The research implies that skill premium in a host country with labor market distortions may increase faster than when skill-biased technology is the only force for skill upgrading. D 2001 Elsevier Science Inc. All rights reserved.

1. Introduction Many studies using micro data from the 1980s have found that China's urban wage structure was extremely compressed (Byron & Manaloto, 1990; Knight & Song, 1991; Liu, 1998). There is indication that returns to education have gone up in the 1990s compared to the 1980s (Li, 2000; Zhao, Li, & Riskin, 1999). The rise of skill premium has paralleled the growth of foreign direct investment (FDI) in China. From 1979, the first year of legalizing FDI, to 1989, the annual flow of FDI averaged US$1.3 billion; from 1990 to 1998, the annual

E-mail address: [email protected] (Y. Zhao). 1043-951X/01/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved. PII: S 1 0 4 3 - 9 5 1 X ( 0 1 ) 0 0 0 4 2 - 6

Y. Zhao / China Economic Review 12 (2001) 40±57

41

flow increased to US$27.8 billion (calculated from State Statistical Bureau of China (SSB), 1999, p. 594). Could foreign investment influence wage structure in host countries? There is a small but growing literature on the relationship between FDI and host country wage structure. Aitken, Harisson, and Lipsey (1996) and Feliciano and Lipsey (1999), using data from the United States, Mexico, and Venezuela, find that foreign-owned establishments pay higher wages than domestic ones after controlling for other factors, which may potentially have the effect of bringing up the overall wage level in host countries. Using state-level data from Mexico, Feenstra and Hanson (1997) find empirical evidence that the growth of FDI is positively correlated with relative wages of skilled labor in Mexico. The authors attributed this effect to increased demand for skilled labor due to skill-biased technology brought in by foreign companies seeking outsourcing. Instead of directly testing whether FDI has contributed to rising relative wages of skilled labor in China, this paper intends to provide an alternative explanation to the correlation between FDI and relative wages. We argue that, in an economy with institutional segmentation in the labor market and high labor mobility costs, foreign investors have to pay much higher wages to hire skilled labor, but do not need to do so for unskilled labor. Consequently, relative wages are raised for skilled labor in a country even without bringing in skill-biased technology. In economies characterized by ``dual economy,'' the work force is segmented into a privileged sector and an unprivileged sector, with the privileged sector enjoying higher earnings. While more educated people born into the unprivileged sector can overcome barriers between the two sectors to join the privileged sector, less skilled workers are filtered behind, leaving the unprivileged sector with largely unskilled work force. As foreign investors come into a country, unskilled workers can be readily hired from the unprivileged sector at lower wages than in the privileged sector, but skilled workers have to come from the privileged sector. Furthermore, to entice skilled workers to leave the privileged sector, higher compensation must be offered. Thus, foreign firms pay higher relative wages to skilled workers even though foreign firms may face the same technological constraints as domestic firms in the privileged sector. In the empirical part of the paper, we use micro data from urban China to examine relative wages in the state-owned enterprises (SOEs) and foreign-invested enterprises (FIEs) and analyze the sector choice of employment by urban workers. We focus on urban China because this is where labor market segmentation is most pronounced. The wage estimation takes into consideration the employment choice between the two sectors by workers of different skill levels. Empirical results are consistent with the assumption that labor market is segmented in China and lend support to our hypothesis that FIEs pay relatively more to skilled labor because skilled workers are more expensive for FIEs. Results from employment choice models indicate that significant mobility costs exist and prevent skilled workers from taking full advantage of higher wage offerings in FIEs. The paper is structured as follows. Section 2 develops a simple-minded economic framework to illustrate the choice of employment of workers and the resulted static equilibrium of relative wages in SOEs and FIEs. Section 3 presents the empirical strategy and describes the data. Section 4 presents estimation results of wage equations and sectoral wage comparisons for workers of different skill levels. Section 5 presents structural and reduced form estimates of sectoral employment choice. Conclusions and discussions are presented in Section 6.

42

Y. Zhao / China Economic Review 12 (2001) 40±57

2. A simple analytical framework Prior to the economic reform, China displayed a sharp feature of a dual economy characterized by large urban±rural sectoral income differentials and barriers to entry into the urban sector (Putterman, 1992). The segmentation continued well into the 1990s. Workers in SOEs, the dominating sector of urban employment, enjoyed higher earnings over those in the nonstate sector due to government support and protection (Putterman, 1992). To simplify the discussion, let us suppose that there exist two types of workers, skilled and unskilled, and two sectors in the economy, state and nonstate. The labor market segmentation is characterized as wage advantage for the state sector for both skilled and unskilled workers: WHS > WHN

…1†

WLS > WLN

…2†

where superscripts S and N denote the state and the nonstate sector, respectively, and subscripts H and L denote skilled and unskilled workers, respectively. Under sectoral labor market segregation, education, especially university education, serves as a device in overcoming the entry barriers (Zhao, 1997). As a result of the education-selective labor mobility, coupled with underinvestment in rural education, a large imbalance in education levels emerged between rural and urban areas (Knight & Song, 1995). Almost all college graduates from rural families join the urban sector. Within the urban sector, workers with advanced degrees face much less entry barriers to SOE employment. Equilibrium wage rates paid by FIEs in relation to those in SOEs can be illustrated by the choice of employment by unskilled and skilled workers given market supply of workers in each category. 2.1. Unskilled workers The supply of unskilled workers is plentiful in China, not only in rural but also in urban areas. Although the majority of workers with urban resident status work for the state sector, rural migrants have increasingly become available to nonstate sector employment since the mid-1980s. Rural migrants are predominantly junior high school or primary school graduates (Zhao, 1999). Reservation wages are low for migrant workers because income levels in rural areas are much lower than in urban areas. Urban discrimination toward migrant workers blocks the access of higher-paying SOE employment to rural migrants. Existing SOE institutions also prevent SOEs from replacing high cost incumbent workers with cheaper migrant labor. SOE workers themselves, while receiving higher compensation in SOEs, certainly do not have an incentive to shift to lower-paying employment in FIEs. As a result, the wage gap between SOEs and FIEs prevails for unskilled workers, i.e., WLF < WLS

…3†

Y. Zhao / China Economic Review 12 (2001) 40±57

43

2.2. Skilled workers Because skilled workers are concentrated in SOEs, foreign investors must recruit skilled workers mainly from SOEs or compete for new workers with SOEs. A skilled worker chooses to work in a foreign firm only if the earnings premium in the foreign sector outweighs the costs of shifting employment (C), with earnings including both wages and nonwage benefits: WHF > WHS ‡ C

…4†

The mobility costs may include nonportable pension and housing benefits promised to SOE workers redeemable in a future date. The longer a worker stays with an SOE, the more he or she would lose when shifting employment. For a new labor market entrant, the mobility cost is zero because the person can choose FIE employment from the beginning. There may also be other costs of shifting employment imposed on skilled workers by SOEs. For example, SOEs often request university graduates to sign contracts agreeing to serve a certain number of years.1 Even after the expiration of the contracts, SOEs often use other administrative means to stop skilled workers from departing. The concentration of skilled workers in SOEs and the existence of mobility cost together make skilled workers more expensive to FIEs than SOEs: WHF > WHS

…5†

Inequalities (3) and (5) can be combined to become (Eq. (6)) WHF WHS > ; WLF WLS

…6†

or, relative wages for skilled workers are higher in foreign-invested firms than in SOEs. Notice that this inequality is derived without invoking assumptions on technological differences between SOEs and foreign firms. If skill-biased technological change is the sole explanation for rising relative wages in a host country, then we are expected to see an overall increase in relative wages with no significant sectoral discrepancy because the technological change should result in an economy-wide increase in demand for skilled labor. Our segmentation hypothesis predicts an overall rise in relative wage, too, but this is simply because higher wages for skilled workers in the foreign sector also raise the national average wages for skilled labor. A more important prediction is that the skill wage gap is wider in the foreign invested sector than in the privileged sector of the domestic economy. The empirical part of the paper will present empirical evidence from a household survey data to support this hypothesis. 1

The leverage often used is the promise of urban resident status. Generally speaking, only state units have access to urban resident quotas assigned by the government. This benefit is especially attractive to new college graduates whose precollege homes were in rural areas or in other cities.

44

Y. Zhao / China Economic Review 12 (2001) 40±57

3. Empirical model and data Assume that wages in the two sectors of interest are determined as follows: lnWsi ˆ qs0 ‡ qs1 Xsi ‡ qs2 Rsi ‡ qs3 Hsi ‡ usi

…7†

lnWf i ˆ qf 0 ‡ qf 1 Xf i ‡ qf 2 Rsi ‡ qs3 Hf i ‡ uf i

…8†

where subscripts s and f denote the state and foreign-invested sectors, respectively, Wi denotes total wages from working in the corresponding sector for individual i, including bonuses, subsidies and nonwage benefits, Xi is a vector of personal characteristics including gender, education, and work experiences, Ri is a vector of regional dummy variables representing regional differences in wage levels, Hi is a vector of industry dummy variables representing wage differences across industries, and us  N(0,ss2), uf  N(0,sf2) are random residuals. Eqs. (7) and (8) may be estimated independently using OLS regression method if the sector assignment of employment is random. However, as the analytical framework outlined in Section 2 shows, such may not be the case. In general, there is a two-way selection process involved in any employment outcome. To initiate the process, a worker makes a choice based on his or her preferences and personal constraints. In our case, we assume that a worker prefers to work in the foreign firms if the expected wage gain is higher than the cost of shifting employment. The transition costs vary by individuals. A younger worker may face less transition cost because the accumulated future benefits from SOEs are less Ð She has contributed fewer years into the pension program and work unit housing is generally allocated according to seniority. Given other things, more educated people may face higher transition cost because SOEs are less willing to let them go. If a worker does not expect to receive work unit housing, then he is less constrained by SOE employment. On the other hand, employment in SOEs may provide the benefit of a flexible work schedule, so given wage differences, parents with young children may value SOE employment more. The second part of the selection involves the selection by employers. When there is excess demand for employment in one sector, the sector may use nonprice rationing and rely on personal and family characteristics to determine admission into the sector. The availability of the preferred sector employment may also differ across geographical regions. The selection process described above can be summarized as follows. A worker will join a foreign firm if the expected percentage wage gain is higher than a function of individual, household and community characteristics (Eq. (9)):2 Wf i Wsi  lnWf i Wsi

lnWsi > g1 ‡ g2 Xi ‡ g3 Di ‡ g4 Ri ‡ ei

…9†

where Xi is a vector of individual characteristics (gender, education, and work experiences) influencing receptivity of the worker in the preferred sector, as well as costs of employment transition, Ri is a vector of regional dummy variables that control for the availability of FIE 2

We will construct the empirical model under the assumption that foreign firms are the preferred sector, but this assumption does not affect empirical results.

Y. Zhao / China Economic Review 12 (2001) 40±57

45

employment in different regions, Di is a vector of variables that do not affect wages but affect employment choice,3 and ei is a random residual assumed to be normally distributed with zero means. Following Lee (1978), define a latent index Ii* as I  ˆ d ‡ d …lnW lnW † ‡ d X ‡ d D ‡ d R e …10† i

0

1

fi

si

2 i

3

i

4 i

i

2

where ei  N(0,se ). Ii* is unobservable but has a dichotomous observable realization Ii, which is related to Ii* as follows: Ii ˆ 1 …employment in an FIE† if and only if Ii > 0 …11† Ii ˆ 0 …employment in an SOE† if and only if Ii  0

…12†

Conditional on selection rule in Eqs. (11) and (12), estimating wage Eqs. (7) and (8) by conventional OLS regression method may lead to estimation bias because E…uf j Ii ˆ 1† 6ˆ 0 and E…us j Ii ˆ 0† 6ˆ 0 The wages equations can be consistently estimated by a two-step procedure described in Heckman (1979) and Lee (1978). The first step is to estimate the following reduced form probit model obtained by substituting wage Eqs. (7) and (8) into Eq. (10): I ˆ a ‡a X ‡a D ‡a R ‡a H e …13† i

0

1 i

2

i

3 i

4

i

i

where ei* = ei d1(esi efi)  N(0,1). The second stage estimation of wage equations utilizes the fact that   f…Z† E…uf j Ii ˆ 1† ˆ sf e F…Z†  E…us j Ii ˆ 0† ˆ sse

f…Z† 1 F…Z†

…14†

 …15†

where Zi ˆ a0 ‡ a1 Xi ‡ a2 Di ‡ a3 Ri ‡ a4 Hi ; f…Z† and F(Z) are the standard normal density function and cumulative distribution function, respectively. Estimated parameters from the first stage probit model (10) are used to derive ZÃi = aà 0 ‡ aà 1 Xi ‡ aà 2 Di ‡ aà 3 Ri ‡ aà 4 Hi : The selectivity terms lf = (f(Zà ))/(F(Zà )) and ls = (f(ZÃ))/(1 F(ZÃ)) are then entered as regressors to obtain consistent estimations of wage equations: lnWf i ˆ qf 0 ‡ qf 1 Xf i ‡ qf 2 Rf i ‡ qf 3 Hf i ‡ qf 4 lf i ‡ hf i

…16†

lnWsi ˆ qs0 ‡ qs1 Xsi ‡ qs2 Rsi ‡ qs3 Hsi ‡ qs4 lsi ‡ hsi

…17†

where E(hfjIi = 1) = 0 and E(hsjIi = 0) = 0. 3

These include whether the worker is eligible for work unit housing and whether the worker needs flexible work schedule for the care of children.

46

Y. Zhao / China Economic Review 12 (2001) 40±57

Table 1 Descriptive statistics of the sample, 1996 SOEsa Sample distribution Age Male (%) Years of schooling (year) Illiterate (%) Primary school (%) Junior high school (%) Senior high school (%) Middle technical school (%) Advanced technical school (%) College or above Work experience (year) Tenure at current industry (year) Annual wages (yuan)

5345 39.8 53.3 11.4 0.1 3.9 29.6 28.6 13.4 15.6 8.7 22.4 17.9 8101.1

FIEsa (9.4) (2.3)

(9.9) (9.7) (4772.3)

188 34.8 47.9 11.5 1.1 3.7 20.7 34.6 13.3 20.2 6.4 17.3 10.9 11608.7

(10.6) (2.4)

(11.2) (10.3) (7929.4)

Source: sample survey. a In the parentheses are standard deviations.

An additional step is needed to derive asymptotic standard errors of estimated coefficients because the disturbance terms in Eqs. (16) and (17) are heteroscedastic. The asymptotic covariance matrix is derived according to Greene (1997, p. 981). Our data is from a 1996 household survey of 4798 urban registered households in six provinces.4 For our purpose, we select all workers in the state sector and foreign-invested enterprises (including Hong Kong-, Macao- and Taiwan-funded ones) but exclude those in the government. We call all state employers SOEs although some of them may not be in forprofit enterprises. The total number of valid observations is 5533, of which 3.5% are employed in FIEs.5 Table 1 presents basic statistics of the sample. As is shown, workers in FIEs are about 5 years younger than their counterparts in SOEs. Total years of work experience among FIE workers are also 5 years fewer, which means that the two groups of workers entered the work force at about the same age. However, FIE workers have about 7 fewer years of tenure in the current industry, implying that FIE workers have more recent job shifts than SOE workers. We also notice that there are more male workers in SOEs than in FIEs (53.3% vs. 47.9%). Looking at education, we see that years of schooling are nearly identical between the two sectors: 11.4 years for SOEs and 11.5 for FIEs. Some differences exist in levels of education. SOEs have more junior high school graduates (29.6% vs. 20.7% in FIEs); FIEs hire more senior high school graduates (34.6% vs. 28.6% in SOEs) and graduates of advanced technical 4

The six provinces are Liaoning, Zhejiang, Hubei, Guangdong, Sichuan, and Gansu. According to national statistics (SSB, 1999, p. 136), of all urban workers in the state sector (including government) and foreign-invested enterprises (including those from Hong Kong, Macao, and Taiwan), the share of foreign sector employment was 5.6% higher than the share in our sample. This is probably because coastal provinces are underrepresented in our sample. This problem does not hurt our analysis, however, as we do not claim that the results represent the whole country. 5

Y. Zhao / China Economic Review 12 (2001) 40±57

47

Table 2 Industry structure of workers (%) Industry

All

SOEs

FIEs

Agriculture and water conservancy Manufacture Geological survey Construction Transportation and communication Commerce Real estate Health, sport, and social welfare Education, arts, and news media Research Finance Other All

0.9 41.3 1.1 4.9 7.5 15.8 4.7 5.7 10.8 4.4 2.2 0.7 100

1.0 40.8 1.1 5.0 7.7 15.4 4.7 5.9 11.1 4.5 2.3 0.5 100

0.5 55.7 0.0 0.0 1.6 26.6 7.3 0.0 1.6 0.0 2.1 4.7 100

Source: sample survey

schools (20.2% vs. 15.6% in SOEs).6 The last row of the table presents crude averages of annual wages in the two sectors. It is shown that average wages in FIEs are higher than in SOEs.7 Table 2 gives an idea of the industry structure of workers in SOEs and FIEs. As is shown, manufacturing and commerce are the two largest industries in our sample. The concentration is more profound for FIE workers with the two industries totaling 82.3% of all workers, while the share is 56.2% for SOE workers. The uneven industry concentration is probably due to state monopoly over industries such as education, news media, transportation, communication, health care, etc., as well as the dominance of the state sector in nonprofit organizations such as social welfare, education, and research.

4. Wage equations estimates Estimation results of wage equations for workers in SOEs are presented in Table 3. Table 4 presents results for workers in FIEs. Two sets of equations are presented in each table Ð one with sample selectivity adjusted and one unadjusted. In each set, two models are estimated: one with schooling as a continuous variable and the other as discrete levels of education. We see that schooling is important in determining wages for both sectors, but returns to schooling in FIEs are roughly twice as much as in SOEs. Looking at continuous schooling models, each year of additional schooling raises earnings by 4.2% in SOEs; the return is 8.6% 6

In both sectors, there are very few observations with primary school education or lower and college education or higher, so these educational categories are merged with their nearest neighbors in regression models. 7 Wages include bonuses and cash subsidies but exclude nonwage benefits such as housing and pension due to data limitations. Independent estimates of nonwage benefits will be discussed following wage equation estimates. Another limitation of the data is that there is no information on working hours, thus, preventing the analysis of labor supply in both wage determination and sector choice models.

± ± ± Yes Yes ± 5345 0.382

7.727* 0.121* 0.049* 0.001* 0.006* 0.042*

±

0.097 0.014 0.003 0.000 0.001 0.003 ± ± ± 0.063* 0.167* 0.270* Yes Yes ± 5345 0.385

±

8.109* 0.123* 0.049* 0.001* 0.006*

±

0.090 0.014 0.003 0.000 0.001 ± 0.017 0.022 0.019 0.041 5345 0.382

± ± ± Yes Yes

7.713* 0.121* 0.049* 0.001 0.006** 0.042

0.142

0.108 0.014 0.003 0.001 0.003 0.026 ± ± ±

Standard errorb

b

Wages include bonuses and subsidies but do not include nonwage benefits such as pension, housing, and medical benefits. Standard errors adjusted for heteroscedasticity. * Coefficient different from zero at 1% significance levels. ** Coefficient different from zero at 5% significance levels.

a

Intercept Male Work experience Work experience squared Tenure at current industry Years of schooling Senior high school Middle technical school Advanced technical school or above Province dummies Industry dummies Selectivity Obs Adjusted R2

Estimated coefficient

Estimated coefficient

Estimated coefficient

Standard error

Model III: continuous schooling

Model II: discrete schooling

Model I: continuous schooling Standard error

Selectivity adjusted

Selectivity unadjusted

Table 3 Wage equations for workers in SOEs dependent variable: ln(annual wagesa)

0.062* 0.165* 0.268* 0.058 5345 0.385

Yes Yes

±

8.089* 0.124* 0.050* 0.001 0.006

Estimated coefficient

0.139

0.102 0.014 0.003 0.001 0.018 ± 0.022 0.020 0.026

Standard errorb

Model IV: discrete schooling

48 Y. Zhao / China Economic Review 12 (2001) 40±57

± ± ± Yes Yes ± 188 0.294

7.791* 0.152** 0.062* 0.001* 0.007 0.079*

±

0.289 0.074 0.012 0.000 0.005 0.016 ± ± ± 0.321* 0.167 0.527* Yes Yes ± 188 0.292

±

8.408* 0.169** 0.057* 0.001* 0.010***

±

0.226 0.073 0.012 0.000 0.006 ± 0.097 0.126 0.106 ± ± ± Yes Yes 0.230 188 0.290

7.485* 0.144** 0.059* 0.001 0.001 0.086

0.441

0.652 0.073 0.013 0.012 0.020 0.136 ± ± ±

Standard errorb

b

Wages include bonuses and subsidies but do not include nonwage benefits such as pension, housing and medical benefits. Standard errors adjusted for heteroscedasticity. * Coefficient different from zero at 1% significance levels. ** Coefficient different from zero 5% significance levels. *** Coefficient different from zero at 10% significance levels.

a

Intercept Male Work experience Work experience squared Tenure at current industry Years of schooling Senior high school Middle technical school Advanced technical school Province dummies Industry dummies Selectivity Obs Adjusted R2

Estimated coefficient

Estimated coefficient

Estimated coefficient

Standard error

Model III: continuous schooling

Model II: discrete schooling

Model I: continuous Schooling Standard error

Selectivity adjusted

Selectivity unadjusted

Table 4 Wage equations for workers in FIEs (dependent variable: ln(annual wagesa)

0.329** 0.193 0.551* Yes Yes 0.095 188 0.289

±

8.300* 0.166** 0.056* 0.001 0.007

Estimated coefficient

0.453

0.557 0.071 0.013 0.013 0.100 ± 0.170 0.152 0.134

Standard errorb

Model IV: discrete schooling

Y. Zhao / China Economic Review 12 (2001) 40±57 49

50

Y. Zhao / China Economic Review 12 (2001) 40±57

(Model III) and 7.9% (Model I) in FIEs. Discrete schooling models have the advantage of allowing for nonlinear returns to each level of education. As is shown, SOEs offer steadier increases of wages according to educational levels. Compared to people with junior high school education or less, other things being equal, senior high school graduates earn 6.2± 6.3% more, middle technical school graduates 16.5±16.7% more, and advanced technical school or higher 26.8±27.0% more. In contrast, FIEs value middle technical school graduates less than senior high school graduates, and value advanced degrees much higher. Graduates from senior high school, middle technical school, and advanced technical school or higher earn 32.1±32.9%, 16.7±19.3%, and 52.7±55.1% more than graduates of junior high school or less, respectively. Selectivity turns out statistically insignificant in wage equations for both sectors. This may be caused by the small size of the FIE sample. However, the signs of the selectivity effects (see definition in Eqs. (14) and (15)) are worth mentioning. The selectivity effects are positive in both sectors Ð indicating that if a person, due to some unobserved factors, earns more in a sector, then the person tends to work in that sector as well. The positive selection is more significant in FIEs than in SOEs. The estimation results of Tables 3 and 4 allow us to estimate predicted wages in the two sectors. Fitted wages in each of the two sectors are first calculated for each individual, then averaged for the whole sample to create mean wages in each sector. Thus, the calculated wage differences have controlled for all individual differences. When estimating wages by sex and educational levels, only the relevant group is used to derive the corresponding averages. The resulting wage differentials are presented in Table 5. Before examining the figures, it should be pointed out that due to the existence of nonwage benefits, wages do not equal total earnings and, thus, wage differentials do not equal earnings differentials. Using the same data set supplemented with aggregate statistics in 1996, Zhao (forthcoming) estimate that annualized difference in nonwage benefits including pension, housing and medical care was roughly 3430 yuan in favor of SOEs for an average unskilled

Table 5 Projected annual wages and wage differentials in the sample, 1996 (unit: yuan) SOE

FIE

FIE

SOE

All

7326.1

9064.3

1738.3

By sex Male Female

7884.4 6693.1

9871.2 8149.5

1986.8 1456.4

By education Junior high school or lower Senior high school Middle technical school Advanced technical school or above

6726.4 6542.5 7880.5 8765.7

7187.1 9328.5 8211.9 11784.3

460.7 2786.0 331.4 3018.7

Wages are projected using selectivity adjusted wage equations with discrete schooling variables in Tables 3 and 4.

Y. Zhao / China Economic Review 12 (2001) 40±57

51

worker in 1996. According to the data in Table 5, for workers of equal characteristics in both sectors, the average wage differential was 1738.3 yuan in favor of FIEs, which is less than the FIE shortfall in nonwage benefits, indicating that SOEs remained, on the average, the privileged sector in 1996. However, Table 5 also shows substantial differences among people of different characteristics. We first note that given other characteristics, the average wage for male workers is 1986.6 yuan higher in FIEs than in SOEs; the differential is only 1456.4 yuan for female. Looking at wage differential by education, we see that workers with junior high school education or less earn a wage premium of 460.7 yuan in favor of FIEs. The wage premium is also very small for middle technical school graduate (331.4 yuan). These FIE wage premia are definitely insufficient to overcome advantages in extra nonwage benefits enjoyed by SOE workers, thus, giving rise to overall earnings premium in SOEs for low skilled workers. The wage premium for workers with senior high school education is 2786.0 yuan; for graduates of advanced technical school or higher, it is 3018.7 yuan. Notice that the latter wage premium is very close to the aforementioned FIE shortfall in nonwage benefits for unskilled workers, 3430 yuan. We do not have an estimation of FIE shortfall in nonwage benefits for skilled workers, but anecdotal evidence suggests that it is much smaller than this figure. The reason is that while the distribution of benefits is more or less equal among SOE workers of different skill levels, skilled workers in FIEs are usually offered with much more generous benefits than unskilled ones. An extra annual benefit of 430 yuan in favor of skilled workers over unskilled ones in FIEs would be enough to generate an overall earnings premium for skilled workers in FIEs over their counterparts in SOEs, and this is easily conceivable. For example, while unskilled workers in FIEs are at most provided with dormitories, important technical personnel are often offered with free or subsidized apartments. Furthermore, we have not been able to derive an independent set of estimates for college graduates because the sample size is too small for this group. There is reason to believe that college graduates definitely earn more in FIEs than in SOEs. To summarize, our empirical results are consistent with the hypothesis that total earnings for skilled workers are more in FIEs than in SOEs, while the reverse if true for unskilled workers. 5. Employment choice models Given projected earning differentials, we can now estimate the structural form sector selection model specified in Eq. (10). Such an estimation, by controlling for wage differentials, would be able to separate the effects of various factors on wages and on the likelihood of obtaining a FIE job. To prepare for interpretation of estimation results of sector selection models, we first present means of variables used in the models (Table 6). Maximum likelihood estimates of the structural form probit model of employment status in foreign-invested firms are presented in Table 7. Two models are presented: Model I uses discrete schooling variables, whereas Model II uses a continuous schooling variable. Wage à s), an approximation of percentage wage advantage à f ln W differential is represented as (ln W

52

Y. Zhao / China Economic Review 12 (2001) 40±57

Table 6 Variables used in probit models Definition FIE Male Work experience Tenure at current industry Senior high school Middle technical school Advanced technical school or above Spouse housing Number of children Ãs à f ln W ln W

Dependent variable: 1 if FIE employment; 0 if SOE employment Dummy variable. Reference group: female workers Total years of work experience Total years working at the current industry Dummy variable. Reference: junior high school or lower Dummy variable. Reference: junior high school or lower Dummy variable. Reference: junior high school or lower Dummy variable: 1 if spouse has public housing. Reference: spouse does not have work unit housing Number of children under 18 years Difference of logarithm of fitted wages projected from regression equations with selectivity adjusted

Mean 0.034 0.531 20.396 17.626 0.288 0.134 0.244 0.299 0.622 0.230

Ãf W Ã s)/W Ã s.8 Estimation results show that the wage differential of FIEs over SOEs, (W between FIEs and SOEs is an important consideration in sector choice. Increasing the percentage wage advantage by 10 percentage point increases the probability of employment in FIEs by 0.4 (Model I) to 0.5 (Model II) percentage points. Since average probability of employment in FIEs is only 3.5% in the sample, this effect represents an over 10 percentage points increase in the probability. Given expected earnings differential, gender does not seem to be important in determining the sector of work. Work experience matters: More experienced workers are less likely to work in FIEs. The negative effect of work experience is consistent with our hypothesis that newer job market entrants face less mobility costs because they have relatively less to lose facing nonportable social security benefits. Table 7 also tells us that given earnings differentials, more educated workers are less likely to be selected into FIEs. Evaluating the effect at mean values, 1 additional year of schooling reduces the probability of FIE employment by 0.15 percentage points, or 4.3% of the probability of FIE employment. The negative effect of schooling is consistent with our assumption that more educated workers may face higher mobility cost than less educated ones because SOEs are less willing to let more educated workers leave. Looking at the discrete schooling model, we see that although graduates of senior high school and people with

8

Because fitted values are used in place of their true values in the structural probit model, the usual estimates of standard errors are biased. While recognizing this problem, this paper does not attempt to correct for the bias.

0.359* 4.608*

Number of children à f ln W Ãs ln W

535.676 5533 188

0.384* 3.636*

0.229*

4.789* 0.117 0.009** 0.129*

571.888 5533 188

No

No

±

± ±

Yes

0.0038 0.0004

0.0061

0.0123

0.0126 0.0098

0.0015 0.0010

Estimated coefficient

0.080 0.233

0.090

±

0.644 0.083 0.004 0.021 ± ±

Standard error

Model II: continuous schooling

Yes

±

±

Marginal effect on probability

* Coefficient different from zero at 1% significant levels. ** Coefficient different from zero at 5% significant levels. *** Coefficient different from zero at 10% significant levels.

Provincial dummy variables Industry dummy variables Log likelihood Number of observation Number in FIEs

0.093

0.256*

0.082 0.266

0.134

1.116* 0.228***

0.967*

±

0.565 0.086 0.005 ± 0.132 0.138

3.369* 0.163*** 0.014*

Standard error

Intercept Male Work experience Years of schooling Senior high school Middle technical school Advanced technical school or above Spouse Housing

Estimated coefficient

Model I: discrete schooling

±

± ±

±

0.0061 0.0005

0.0036

0.0016 0.0011 0.0015

Marginal effect on probability

Reference: no spouse or spouse has no work unit housing One child vs. no child One percentage point above mean vs. mean

Reference: junior high or lower

Reference: female Ten years above mean vs. mean One year above mean vs. mean Reference: junior high or lower Reference: junior high or lower

Definition of marginal effect

Table 7 The structural form probit estimates of foreign firm employment status equation (dependent variable: employment in foreign firm = 1; else = 0)

Y. Zhao / China Economic Review 12 (2001) 40±57 53

0.156** Yes Yes 650.129 5533 188

Number of children Provincial dummy variables Industry dummy Log likelihood variables Number of observation Number of FIEs

0.074

33276.28 0.078 0.005 0.006 ± 0.099 0.126 0.108 0.085

Standard error

* Coefficient different from zero at 1% significance levels. ** Coefficient different from zero at 5% significance levels. *** Coefficient different from zero at 10% significance levels.

0.121** 0.299* 0.297* 0.092

±

38.909 0.017 0.005 0.030*

Intercept Male Work experience Tenure at current industry Years of schooling Senior high school Middle technical school Advanced technical school or above Spouse Housing

Estimated Coefficient

Model I: discrete schooling

0.0001

± 0.0000 0.0000 0.0001 ± 0.0001 0.0001 0.0000 0.0001

Marginal Effect on probability

0.082

0.169** Yes Yes 653.470 5533 188

± ± ±

39.544 0.014 0.005 0.030* 0.032***

Estimated Coefficient

0.074

33364.14 0.078 0.005 0.006 0.018 ± ± ± 0.084

Standard error

0.0001

± 0.0000 0.0000 0.0001 0.0000 ± ± ± 0.0001

Marginal Effect on probability

Model II: continuous schooling

± Reference: female Ten years above mean vs.mean Ten years above mean vs. mean One year above mean vs. mean Reference: junior high or lower Reference: junior high or lower Reference: junior high or lower Reference: no spouse or spouse has no work unit housing One child vs. no child

Definition of marginal effect

Table 8 The reduced form probit estimates of foreign firm employment status equation (dependent variable: employment in foreign firm = 1; else = 0)

54 Y. Zhao / China Economic Review 12 (2001) 40±57

Y. Zhao / China Economic Review 12 (2001) 40±57

55

advanced degrees are less likely to be selected into FIEs than people with junior high school education or less, the probability is the highest for middle technical school graduates. One possible explanation is that FIEs might have paid some junior high school graduates to undergo specific training in middle technical schools and promised subsequent employment. A supporting evidence is that middle technical school graduates earn less than their senior high school counterparts in FIEs, as shown in wage equations (Table 4). This is consistent with predictions of the theory of specific human capital, but more evidence is needed to substantiate the conjecture. Another set of interesting findings from the structural form probit model concerns with the impact of housing status and family composition on employment choice. The existence of work unit housing assigned to one's spouse has a statistically significant effect on which sector the worker chooses to work. This explanation is obvious: If one's spouse has been assigned work unit housing, then according to commonly observed housing allocation rule among SOEs, one would not expect work unit housing from his or her work unit. Since housing is a very important component of nonwage benefits, not expecting it would certainly reduce one's attachment to SOEs. The number of children also has statistically significant effect on employment choice. Going from zero to one child, the probability of FIE employment increases by 0.24 percentage points, representing a 6.9% increase in the probability of FIE employment. A possible explanation is that SOEs usually provide more flexible work schedule than FIEs. In Table 8, the net effects of various factors on employment status are presented as reduced form probit models. The results summarize cost±benefit considerations of individual workers and selectiveness of firms. According to estimation results from wage equations (Tables 3 and 4), returns to education are higher in FIEs than in SOEs. Furthermore, Table 5 and discussions of nonwage benefits tell us that skilled workers almost certainly receive higher compensation than unskilled workers. The higher earnings serve as a magnet attracting skilled workers to FIEs. However, selectivity results (Table 7) show that controlling for earnings difference, SOEs select more educated workers. The net effect as presented in Table 8 is that controlling for work experience and other factors, more educated workers tend to work in FIEs. 6. Conclusions and discussions In this paper, we investigate a mechanism through which FDI may affect relative wages in host countries. Feenstra and Hanson (1997) suggest that FDI can raise relative wages in host countries by bringing in skill-biased technology. Here, we propose an alternative hypothesis that applies to countries with segmented labor market and high labor mobility costs. Since labor markets in many developing countries receiving FDI display features of segmentation, the explanation may be relevant outside of China as well. Under labor market segmentation in China, education is used to access the privileged sector Ð the state sector Ð by people born into the unprivileged sector. As a result, unskilled workers are abundant in unprivileged or nonstate sector, but skilled workers are concentrated in the state sector. Therefore, as a third sector in the economy, foreign firms can readily hire unskilled labor from the unprivileged sector at lower costs, but have to offer skilled labor

56

Y. Zhao / China Economic Review 12 (2001) 40±57

much better compensation than that in the privileged sector. The existence of mobility costs by state-sector workers further raises wage premium for skilled labor in the foreign sector. As a result, relative wages of skilled labor are higher in the foreign sector than in the state sector. Using micro data from an urban household survey in 1996, we estimate relative wages of skilled workers in both foreign sector and the privileged sector Ð SOEs in China. We do so by controlling for sample selection bias using a standard two-stage process described in Heckman (1979) and Lee (1978). Empirical results are consistent with our hypothesis. Less educated workers earn significantly less in FIEs than in SOEs, but more educated workers earn more in FIEs than in SOEs. These results indicate that significant labor market segmentation exists in China, at least in 1996, our year of data, and the mere entry of foreign firms can automatically raise returns to schooling regardless of whether technology is skill-biased or not. If the technology in FIEs is indeed skill-biased, then the labor segmentation effect discussed in this paper would reinforce the skill premium. Our results also indicate that skilled workers are very expensive for nonstate firms in China. Facing high cost of skilled labor, domestic private enterprises of China often choose a technology to minimize the use of advanced degree holders, especially college graduates.9 Thus, in the absence of competition for skilled labor from foreign firms, the state sector is monopsony of skilled labor and is able to practice wage compression that gives rise to low returns to education. In the absence of competition for skilled workers, as long as the absolute earnings are higher in SOEs, skilled workers have no choice but to stay in the sector. The market entry of the foreign sector brings competition for skilled labor. The competition may have been limited in our data year, 1996, because significant mobility costs existed, as is shown by our employment choice model estimates, but as labor market reform gradually reduces mobility costs, competition will intensify. This implies that SOEs may eventually be forced to raise returns to schooling to retain skilled work force. This may have already occurred, but to confirm it would require panel data and is beyond the scope of this paper. Finally, our study implies that FDI may have a bigger impact on skill upgrading in China than in other host countries where less labor market distortions exist and skill-biased technology is the only force raising skill premium. A larger skill premium is likely to induce faster increase in private investment in education in China. Although the foreign sector may still be too small in the Chinese economy to have significant impact on education attainment, the effect is expected to be much more significant after China joins WTO. Acknowledgments The research is supported by the Teaching and Research Award Fund for Outstanding Young Teachers in Higher Education of the Ministry of Education, PRC. Part of the paper is completed when the author visited National Bureau of Economic Research and Harvard University under a visiting scholarship provided by NBER. Helpful comments from D. Gale Johnson, Robert Lipsey, and an anonymous referee are gratefully acknowledged. 9

In our data, only 2% of workers in domestic private enterprises have college education.

Y. Zhao / China Economic Review 12 (2001) 40±57

57

References Aitken, B., Harrison, A., & Lipsey, R. (1996). Wage and foreign ownership: a comparative study of Mexico, Venezuela and the United States. Journal of International Economics, 40, (3/4), 345 ± 371 (May). Byron, R.P., & Manaloto, E.Q. (1990). Returns to education in China. Economic Development and Cultural Change, 38, (4), 783 ± 796 (July). Feenstra, R., & Hanson, G. (1997). Foreign direct investment and relative wages: evidence from Mexico's Maquiladoras. Journal of International Economics, 42 (3/4), 371 ± 393. Feliciano, Z., & Lipsey, R. (1999). Foreign Ownership and Wages in the United States, 1987 ± 1992. NBER Working Paper, No. W6923, February. Greene, W.H. (1997). Econometric analysis (3rd ed.). Upper Saddle River: Prentice-Hall. Heckman, J. (1979). Sample selection bias as a specification error. Econometrica, 47, (1), 153 ± 162 (January). Knight, J., & Song, L. (1991). The determinants of urban income inequality in China. Oxford Bulletin of Economics and Statistics, 53, (2), 123 ± 154. Knight, J., & Song, L. (1995). Towards a labor market in China. Oxford Review of Economic Policy, 11 (4), 97 ± 117. Lee, L.-F. (1978). Unionism and wage rates: a simultaneous equations model with qualitative and limited dependent variables. International Economic Review, 19, (2), 415 ± 433 (June). Li, H. (2000). Economic Transition and Returns to Education in China. Unpublished manuscript, Georgia Institute of Technology. Liu, Z. (1998). Earnings, education and economic reform in urban China. Economic Development and Cultural Change, 46, (4), 697 ± 725 (July). Putterman, L. (1992). Dualism and reform in China. Economic Development and Cultural Change, 40, (3), 467 ± 493 (April). State Statistical Bureau of China (SSB). (1999). China statistical yearbook, 1999. Beijing: China Statistical Press. Zhao, R., Li, S., & Riskin, C. (Eds.). Zhongguo jumin shouru fenpei zai yanjiu (Re-studying income distribution of Chinese households). Beijing: Publishing House of Chinese Finance and Economy. Zhao, Y. (1997). Labor migration and returns to rural education in China. American Journal of Agricultural Economics, 79, (4), 1278 ± 1287 (November). Zhao, Y. (1999). Labor migration and earnings differences: the case of rural China. Economic Development and Cultural Change, 47, (4), 767 ± 782 (July). Zhao, Y. (forthcoming). Earnings Differentials between State and Non-State Enterprises in Urban China, Pacific Economic Review.