Wage Differentials among Ownership Groups and Worker Quality in Vietnamese Manufacturing

Abstract Wholly foreign multinational enterprises (WFs), joint-venture multinationals (JVs), state-owned enterprises (SOEs) pay higher wages than domestic private firms in Vietnamese manufacturing. In large samples of medium–large (20+ employees) firms, conditional differentials accounting for worker education and occupation, as well as capital intensity, size, and shares of female workers, were substantially smaller, but positive and significant. Wage levels and differentials varied substantially among industries. Conditional differentials remained positive and significant for WFs and JVs in most of the 11 industries examined, but estimates of SOE-private differentials were insignificant in most industries. Robustness checks using 2007 data yielded similar results.


Introduction
There is a growing literature indicating that foreign-owned multinational enterprises (MNEs) normally pay higher wages than domestic firms in host, developing economies. The most sophisticated studies to date have analysed manufacturing plants in Indonesia and Malaysia, and accounted for the fact that multinational enterprises tend to hire relatively welleducated workers and be relatively large and capital or input intensive compared to local plants (Lipsey and Sjöholm 2004;Ramstetter 2014;Ramstetter and Narjoko 2013). These studies often found positive and significant wage differentials between foreign MNEs and local plants, even after controlling the influences of worker education and sex, as well as plant size and capital intensity. However, aside from limited evidence in Ramstetter and Phan (2007), Tran (2007), and Fukase (2014aFukase ( , 2014b, there is very little evidence regarding wage differentials among firm ownership groups in Vietnam, which accounts for the influence of worker quality. This paper partially fills the gap in the literature by using data on worker quality that were unavailable in previous years to analyse determinants of wages in substantially more open and competitive than even five years previous, and many firms were still in the process of adjusting to the large policy changes. During this period, firms were also affected by the world financial crisis, which was partially responsible for the decline of Vietnam's economic growth rate to 5.4-6.4 percent in 2008-13 from the 6.8-7.8 percent that were experienced in 2000-07 (Asian Development Bank 2014).
In this paper, we first review the literature on MNE-local wage differentials (Section 2) and describe the enterprise data that are used for the analysis, focusing on unconditional differentials in wages and worker skills between MNEs and private firms (Section 3). Then we test if wage differentials are statistically significant after accounting for firm size, capital intensity, worker sex, and worker education in both 2007and 2009 (Section 4). For 2009, it is also possible to control for the influence of worker occupation. The focus is on analysis of 2009 data because they allow better control for worker quality, but the estimates for 2007 provide an important robustness check. Finally, Section 5 offers some conclusions and suggestions for future research.

Literature review and Methodology
There is a compelling theoretical rationale suggesting that MNEs will often pay higher wages than corresponding domestic enterprises in host developing economies. On the demand side, MNEs are often argued to possess large amounts of knowledge-based, generally intangible assets such as production technology, marketing networks and management systems. Possession of these firm-specific assets suggests that MNEs will be likely to be more efficient than non-MNEs, which is reflected by larger firm size, higher factor productivity, and/or correspondingly higher factor rewards.
Many MNEs also require their employees, even relatively unskilled staff, to have engineering, marketing, and foreign language skills required to work with particular machinery and clients. In addition, many of these employees need to learn modern work disciplines, such as punctuality, tidiness and promptness, which may not be valued as highly in local firms, for example. Firms operating in developing economies like Vietnam often face shortages of skilled workers who have both engineering, foreign language, and modern management skills. Thus, MNEs relative unfamiliarity with local labor markets may make it more difficult for them to hire new skilled workers, or retain current skilled workers than domestic firms. This may motivate MNEs to pay relatively high wages as an incentive to increase the attractiveness of their firms to skilled workers or to reduce turnover.
On the supply side, workers may prefer to work for locally owned firms because they are more familiar with local management practices. In Vietnam, for example, it is clear that labor market practices often vary greatly between MNEs and local firms. Nonetheless, our impression is that most Vietnamese workers are not very opposed to working for MNEs and many might actually prefer MNE employment to the alternatives. This is supported by studies which suggest that internal migrants in Vietnam often prefer to work for MNEs over local firms (Fukase 2014b).
Some of the most comprehensive analyses of wage differentials to date have examined Indonesian manufacturing plants in 1996and 2006(Lipsey and Sjöholm 2004Ramstetter and Narjoko 2013). For 1996, estimates of Mincer-type wage equations at the plant level found strong evidence that MNEs paid higher wages than domestic firms after controlling for size, input intensity, the share of female workers, and worker education. For 2006, estimates in large samples of all manufacturing plants combined and a few individual industries also reveal positive and significant wage differentials, but many of the industry-level regressions indicate that conditional wage differentials were not significant in the latter year. 1 In addition, analyses of Malaysian plants in [2000][2001][2002][2003][2004] also suggest that conditional wage differentials accounting for both worker education and occupation were positive and significant in most of the individual industries examined and when all sample industries were combined (Ramstetter 2014). Although they do not control for the effects of worker education or occupation, other studies of Malaysia (Lim 1977) and Thailand (Matsuoka-Movshuk and Movshuk 2006; Ramstetter 2004) also found positive and significant wage differentials after controlling for plant-level differences in capital intensity and size, for example.
Similar studies of Vietnam are sparse. Most of previous studies of wage differentials primarily focused on gender wage gaps, finding that women tend to earn significantly less than men (Liu 2001(Liu , 2004McCarty 1999;Pham and Reilly 2007). Similar to this study, Ramstetter and Phan (2007) and Tran (2007) occupation. This study thus focuses on analyzing these years. A recent study by Fukase (2014a) used household data to compare the wages paid to workers in MNEs and domestic firms, also finding that MNEs and SOEs tended to pay higher wages than private firms and another study (Fukase 2014b) found that internal migrants were attracted by job opportunities in MNEs and SOEs.
In Vietnam, ownership-related wage differentials are also related to government regulations, which require MNEs to pay higher minimum wages than private companies (Nguyen 2014). For example, in 2006-2007 minimum wages in WFs and JVs were 58-93 percent higher than in domestic firms (private firms and SOEs combined), depending on the region. In 2009, these differentials declined to 38-50 percent. Foreign-domestic differentials in minimum wages were largest in Hanoi and Ho Chi Minh City and smallest in rural areas.
On the other hand, it is important to note that minimum wage requirements only affect base salaries, and domestic firms often pay much higher bonuses than multinationals. 2 As explained by Ramstetter and Phan (2007), SOEs were also required to pay relatively high minimum wages in previous years, though we have no new information on this point.

The data, wage differentials and worker quality
This study analyzes medium-large firms (20 or more employees) included in Vietnam's Annual Enterprise Surveys for 2007and 2009(General Statistical Office 2011, 2013. To date, only these two surveys have collected comprehensive information on employee education and wages. The 2009 data also have information on worker occupation but this indicator is not available for the 2007. All values are expressed in 2000 prices using appropriate deflators. 3 Wages are defined to include regular salaries and other compensation such as bonuses, subsidies, social security, health insurance, and pension insurance. Real wages are calculated using the consumer price index (CPI).
Most MNEs, including both WFs and JVs, and SOEs are medium-or large-sized firms, which differ in many respects from smaller firms, which are predominantly private. Therefore, it is more meaningful to compare wages among medium-large manufacturing firms with a workforce of at least 20 employees. In addition to making the comparison more consistent and meaningful, excluding small firms also allows us to remove most outliers and most firms reporting implausible data. 4 The analysis also excludes five industries with very few MNEs and/or SOEs (tobacco; publishing and printing; petroleum and gas; miscellaneous manufacturing; and recycling).
After eliminating firms that were small, had implausible data, or were in one of the five  (Table 1).
However, reflecting efforts to privatize and equitize many SOEs, SOE shares declined markedly after 2000, while MNE shares increased. Table 1 indicates these trends continued in 2007-09, with SOE shares of paid workers in the 11 sample industries declining from 13 to 10 percent while the total MNE (JV+WF) share rose from 43 to 47 percent. WFs accounted for the vast majority of MNE employment, their share rising from 37 to 42 percent while the JV share fell slightly from 5.3 to 4.6 percent. WFs are concentrated in labor-intensive industries such as wood and furniture, apparel, leather, and footwear; and electronics. WF employment shares also exceeded one quarter in three relatively capital-intensive industries: transportation machinery, textiles, and basic metals and metal products.
Of the 11 sample industries, paid employment was largest in apparel leather, and footwear, with 1.08 million paid workers in 2007 and 1.21 million in 2009, followed by food and beverages with 0.35 and 0.40 million, respectively, and wood and furniture, with 0.30 and 0.31 million, respectively. 54 and 60 percent, respectively of the paid workers in the apparel group worked in WFs WFs were also large in the smaller electronic machinery industry, accounting for 72 and 80 percent, respectively, and the paid workers in this industry. These two industries accounted for two-thirds of the paid workers in WFs. At the other end of the scale, WF shares were relatively small in food and beverages (13 percent) and non-metallic mineral products (6.1-6.4 percent). JV shares were almost one-fifth in transportation machinery, but much smaller (6.4 percent or less) in the 10 other sample industries. In 2007, SOE shares were one fifth or more in textiles, chemicals, rubber, and plastics, non-metallic mineral products, and transportation machinery, but in 2009 this was only true in transportation machinery.   Table 2) suggests a similar tendency for MNEs with large ownership shares (90 percent or larger) to have relatively small unconditional wage differentials compared to other MNEs. MNE-related wage differentials in Table 1 are also of similar as those for Indonesian production workers in 2006, though they are considerably smaller than differentials for 1996 and for non-production workers in 2006. This pattern makes sense because most of the paid workers in the Vietnamese samples are production workers or non-production workers in relatively lowwage occupations.
The size of MNE-local wage differentials may also be related to the size of the technology gap between MNEs and private plants, which is likely to be smaller at higher levels of wages and incomes. There is also a similar, though less consistent tendency for WFs or MNEs with relatively large ownership shares to have relatively small labor productivity differentials relative to local plants among ownership groups in Indonesia, Thailand, and Vietnam (Ramstetter 2004;Ramstetter and Phan 2013;Takii 2004;Takii and Ramstetter 2005).
Another factor leading to wage differentials is the previously noted tendency Vietnam's minimum, base wages to be highest in MNEs, though this difference is often offset by higher payments of other compensation in domestic firms and minimum wage requirements are probably not binding for many MNEs.
When the 11 sample industries are combined, shares of paid workers who completed tertiary education were also higher in SOEs and JVs than in WFs and private firms in both years (Table 3). In JVs this share increased from 16 percent in 2007 to 17 percent in 2009, in SOEs the share increased from 13 to 17 percent, respectively. Corresponding shares in WFs and private firms also increased but were much smaller (5.9-7.3 percent). Although it is reasonable to expect tertiary shares to rise during this period, the large increase for SOEs suggests substantial differences in the SOE sample between the two years, perhaps reflecting the influence of privatization.
There is also large variation in tertiary shares among industries (Table 3). For example, all ownership groups had relatively high tertiary shares in the chemicals group and electronic machinery, but relatively low shares in the apparel group. On the other hand, WFs and SOEs had relatively high tertiary shares in food and beverages, as did JVs in 2009, but tertiary shares were relatively low in private firms in both years. At the industry level, there are a number of other large changes in tertiary shares between 2007 and 2009 which suggest substantial differences in underlying sample firms in some industry-owner combinations. 5 Mean shares of moderately educated workers (those who completed secondary education (e.g., high school or vocational training college, but not tertiary education) in all sample firms were larger than corresponding tertiary shares all ownership groups in 2007 and for private firms and WFs, but not for SOEs or JVs in 2009 (Tables 3, 4). Moreover, differences between secondary shares and tertiary shares tended to be relatively small, six percentage points or less in absolute value. This pattern contrasts sharply with Indonesia in 2006, for example, where secondary shares tended to be substantially larger (e.g., 10-20 percentage points or even more) than tertiary shares. The contrast partially reflects the relatively heavy emphasis Vietnam has placed on higher education at relatively low levels of per capita income.
In addition to data on worker education, the 2009 survey also provides data on four types of worker occupations, two of which are highly paid, managerial employees, and professional, technical and supervisory employees. To further account for worker quality in this year, shares of these highly paid workers are also calculated (Table 5). In all 11 sample industries combined, SOEs and JVs also had the highest shares of high quality workers by this measure 24 and 22 percent, respectively, but in WFs and private firms, these shares were only 16 percent. Similar to tertiary shares, shares of highly paid workers were relatively large for all groups in the chemicals group and electronic machinery, in addition to the metals group, general machinery, and transportation machinery..

Conditional wage differentials from econometric approach
As emphasized in the literature, ownership-related wage differentials in the manufacturing sector are likely to be related to workforce characteristics such as education attainment and occupation. The literature also suggests that firm characteristics such as size, capital intensity, and the share of females in paid employees may also influence the extent of wage differentials.
Therefore, in this section we continue with an econometric analysis to examine the extent to which ownership-related wage differentials persist after controlling for the influences of worker education, occupation, and sex, as well as firm capital intensity and size. Similar to previous studies, we estimate the following model: where RW ij = Average real wage of firm i of industry j.
RO ij = Real output of firm i of industry j.
KI ij = Capital intensity of firm i of industry j, measured as the ratio of fixed capital stock over employment after deflating capital stock at a constant value.
SH ij = A share of highly educated employees in total employment of firm i of industry j (per cent).
SM ij = A share of moderately educated employees in total employment of firm i of industry j (per cent).
SP ij = A share of employees in highly paid occupation in total employment of firm i of industry j (per cent).
SF ij = A share of female employees in total employment of firm i of industry j (per cent).
DW ij = A dummy for wholly-owned, foreign-invested enterprises (wholly foreign firms -WF), taking a value of one if a firm is wholly owned FIE and zero otherwise.
DJ ij = A dummy for joint venture enterprises (JV), taking a value of one if a firm is FIE joint venture and zero otherwise.
DS ij = A dummy for state-owned enterprises (SOE), taking a value of one if a firm is state-owned and zero otherwise.
= A stochastic error term.
All estimates also include vectors of dummy variables identifying six regions and as many as 29 industries, usually defined at the two-or three-digit level of Vietnam's Standard Industrial Classification (VSIC) to account for region-specific and industry-specific influences on the constant which are not captured by the firm-level variables. 6 Industry-specific effects on constants and slopes are also accounted for in more detail by estimating equations for each of the 11 sample industries separately, as well as all 11 industries combined.
Coefficients on capital intensity (a 1 ) and real output (a 2 ) are expected to be positive because capital-intensive and large firms generally pay higher wages than labor-intensive or small firms. Coefficients on the shares of highly or moderately educated workers (a 3 , a 4 ) and shares of highly paid workers occupations (a 5 ) should also be positive because they suggest higher worker quality in firms with relatively high shares. In contrast, the coefficient on the share of female workers (a 6 ) is likely to be negative because firms with a higher proportion of female workers tend to have lower average wages. 7 Finally, if wage differentials between WFs JVs, and SOEs, on the one hand, and private firms, on the other, persist after controlling for worker education, occupation, and sex, as well as firm size and capital intensity, the signs of the coefficients on DW, DJ, and DS (a 7 , a 8 , a 9 ) will be positive.
Because data on worker occupation are only available for 2009, we focus on estimates for this year, but also provide estimates for 2007 without this variable as a robustness check.
Estimates are performed in cross sections, which mean that the coefficients on DW, DJ, and DS (a 7 , a 8 , a 9 ) can be interpreted as conditional wage differentials similar to the unconditional differentials in Table 2. However, it is also possible that wages could influence firm's capital intensity and size, creating potential simultaneity between the dependent and independent variables. To check for the robustness of the results to simultaneity concerns, estimates of both contemporaneous and lagged specifications, where capital intensity and output are lagged one year, are compared. All estimates use robust standard errors to account for heteroskedasiticity that can be expected when firm-level, scale variables (e.g., output, capital intensity) are used.
In large samples of firms in all 11 industries combined, estimated coefficients were always consistent with expectations for 2007 and generally consistent for 2009 (Table 6) suggesting that any simultaneity bias is likely to be small.
Most importantly, the estimates suggest that MNEs and SOEs paid significantly higher wages than local firms, even after controlling for the influences of capital intensity, firm size as well as worker education, sex, and occupation. However, conditional wage differentials were all substantially smaller than corresponding unconditional differentials in Table 2. respectively, and SOE-private differentials were 9-10 percent and 13-15 percent, respectively, and that all of these wage differentials were highly significant statistically. 8 These results are consistent with the patterns observed in Table 2 because they imply JVs pay the highest wages, followed by WFs, SOEs and lastly private firms. On the other hand, the conditional differentials were much closer in magnitude than unconditional differentials for WFs and JVs; in other words, the controls in equation (1)  Estimates of equation (1) (Table 7). However, in 2007 differentials were insignificant at standard levels in four industries: the apparel group, wood and furniture, paper, and general machinery. The JV-private differential was also rather small in the apparel group in 2009, though it was positive and highly significant. JV-private differentials were significant and tended to be largest in both years in the chemicals group, electronic machinery, and the metals group in both years. Estimated differentials were also relatively large in textiles in 2009, but smaller in 2007, while the reverse was true in transportation machinery. Wald tests again indicate that it is usually meaningful to distinguish JVs and WFs when estimates of equation (1) are performed at the industry level.
Although most WF-private and JV-private differentials were significant when estimated at the industry level, most SOE-private wage differentials were insignificant. There were three notable exceptions: food and beverages, the chemicals group, and electronic machinery. There was also some indication of positive and significant SOE-private differentials in transportation machinery in 2007 and in the apparel group in 2009 (lagged specification only).
In other words, most of the unconditional, SOE-private differentials are apparently explained by differences in worker education, occupation, and sex, as well as firm-level capital intensity and size.

Conclusions
This paper has examined the extent of wage differentials among medium-large MNEs,  Takii, S 2004, 'Productivity differentials between local and foreign plants in Indonesian manufacturing, 1995', World Development, vol. 32, no. 11, pp. 1957-69. Takii, S and Ramstetter, ED 2005     Note: See Table 1 for a precise definition of sample firms; highly paid occupations are defined as (1) managers and (2) Table 3 for other slope coefficients and indicators; full results including all coefficients and equation details are available from the authors.