A REPORT ON VIETNAM’S LABOR MARKET

CIEM-DANIDA PROJECT: POVERTY REDUCTION GRANT (PRG)

CORRESPONDING AUTHOR: DIEP N PHAN

LAST UPDATED: JULY 23RD 2009

CONTENTS

1. Introduction...... 2 2. Diagnosis of Labor Market Data in Vietnam...... 2 2.1. Population and Housing Census (1989, 1999, 2009, GSO)...... 2 2.2. Labor Force Survey (1996-2007 MOLISA)...... 3 2.3. Labor Force Survey (2007 GSO)...... 6 2.4. Non-farm Individual Business Establishment Census/Survey (1995, 2002-2006, GSO)...... 6 2.5. Enterprise Survey (2000-2007, GSO)...... 7 2.6. Vietnam (Household) Living Standard Survey (1993, 1998, 2002, 2004, 2006, 2008, GSO)...... 9 2.7. Agricultural Census (1996, 2001, 2006, MARD)...... 9 2.8. DANIDA’s SME Surveys...... 10 2.9. Summary...... 11 3. Literature Review...... 11 3.1. The Demand Side of the Labor Market...... 11 3.2. The Supply Side of the Labor Market...... 15 3.3. The Dis-equilibrium: Wage Determinants & Wage Inequality...... 16 4. Current Situation of the Vietnamese Labor Market...... 19 4.1. Labor Supply...... 20 4.2. Labor Demand and Employment Structure...... 23 4.3. Labor Productivity...... 25 4.4. Wage...... 28 5. Concluding Remarks...... 31 Introduction

Labor is the primary asset of the poor. Thus, understanding how the labor market operates and its role in determining poverty and inequality outcomes is the key to formulating any development policy. This is particularly true in the context of Vietnam, where the importance of pro-poor gains from agricultural growth in the 1990s are gradually diminishing relative to wage employment in manufacturing and service sectors. This report attempts to summarize the current findings and identify the gaps in the literature on the labor market in Vietnam.

This report has three primary objectives:

1. To present a diagnosis of labor market data in Vietnam

2. To review empirical studies on the labor market in Vietnam, in particular on

a. the demand side

b. the supply side

c. the (dis)equilibrium: wage determinants and wage differential/inequality

3. Compile the most up-todate data set on the Vietnamese labor market, using the various sources discussed in (1), then descriptively study the current situation of the Vietnamese labor market using this data set

The ultimate goal of this report is to answer the questions of what do we know and what don’t we know about the Vietnamese labor market. Given the gaps in the literature and the available data, the report provides some suggestions for future research and for future data collection.

1. DIAGNOSIS OF LABOR MARKET DATA IN VIETNAM

Currently, labor market data in Vietnam are insufficient in scope, low in quality, and piecemeal in sources, but improvements are being made in various ways. Followings are the main sources:

1. Population and Housing Census

2. Labor Force Survey (LFS) by MOLISA

3. Labor Force Survey (LFS) by GSO

4. Establishment census/survey

5. Enterprise survey

6. Vietnam Living Standard Survey 7. Agriculture Census (for rural labor market)

8. DANIDA’s SME surveys

1.1. POPULATION AND HOUSING CENSUS (1989, 1999, 2009, GSO) The first data source to study the labor market in Vietnam would be the Population and Housing Census, collected every 10 years (the 2009 census was just finished in April 2009). Given its large coverage, the census questionnaire is kept to a minimum. It is among the best sources to give a snap-shot of: the population (by ethnicity, religion, gender, education, etc.), housing status, birth and death rates, basic employment statistics (labor supply by skills, employment by industry and occupation, unemployment and underemployment rates) and migration statistics. For the 1999 census, a 3% sample is available. However, data on income or wages are not collected in the census, so these data are only good for “counting” purposes, not for deeper analyses, unless one combines this census with some other data sets (for example, poverty mapping is typically done by combining census data with living standard survey data).

This census is currently the only source that provides migration data that is representative of all migration flows in Vietnam. There are various ad-hoc migration surveys whose questionnaires are much more extensive but whose scopes are limited to a few rural villages or urban areas. Despite its representativeness, migration data from this census have a major disadvantage. A migrant is defined as someone who changed place of residence in the past 5 years. But this definition largely ignores all seasonal and temporary migrants who move around much more frequently than the 5-year interval. Seasonal and temporary migrants are ignored also because many of them are unregistered and thus not interviewed during time of data collection. The census thus severely underestimates the actual number of in-migrants at the destination while overestimates the actual number of out-migrants at the origin.

Note that the annual statistical year book series, published by GSO each year since 1990, has data on employment by 20 activities. It is unknown which survey or census these data come from. They may be “estimates” from the base-year data in the 1989 and 1999 Population and Housing Census, adjusted for by the LFS survey results collected each year.

1.2. LABOR FORCE SURVEY (1996-2007 MOLISA) The LFS is a household survey, collected annually from 1996 through 2007 by MOLISA. In 2007 two different versions of the LFS were collected separately by MOLISA and GSO. The one by GSO was evaluated to be better (larger sample size, more than twice longer questionnaire content, and also higher quality), although the two surveys provided consistent broad results. For a detailed discussion, please see Roubaud (2008). In 2008, the LFS was not collected by either MOLISA or GSO, but GSO collected the Movement of Population, Employment, and Family Planning Survey (MPEFPS), which partially covered labor market indicators. Starting in 2009, the LFS is expected to be collected permanently by GSO, possibly turned into a quarterly survey, and its content “may” change to accommodate the collection of information on the informal sector.

From 1996-2007, the LFS has a sample size of around 100,000 households each year, and is representative at the provincial level. Sample design is a classic two-stage stratification, with primary sample units being census- enumerated villages, and secondary sample units being households. It is designed to study the supply side of the labor market. The survey’s content is light and declined over the years, but core questions normally include households’ socio-demographic characteristics and employment status of each household member aged 15 years and older. Since 1999, data on earned income of working members were also collected. Table 2.2.1, taken from Roubaud (2008), summarizes this survey.

The quality of this data set from 1996-2007 is highly questionable. Raw data are not available, only aggregate indicators and tabulations printed in a series of annual publications. The content of these publications vary from one year to the next. In some years, employment data by 20 activities are available. In other years, employment data by only 3 sectors (agriculture, industry, and service) are available. This makes it difficult to create a consistent time-series of employment by activity. Furthermore, even though the data are supposedly representative at the provincial level, it has been advised that province-level data would reveal inconsistencies over time. Data on earned income may not be reliable either, given that the questionnaire contains a simple question of the type “How much did you earn last month?” (Razafindrakoto 2008)

In short, the use of this 1996-2007 data set is generally not recommended, mostly because of low quality and inaccessibility to raw data. At best, it can be used for simple “counting” purposes, similar to the Population and Housing Census. That is, it can provide indicators on labor supply (and its qualification, i.e., by age, gender, ethnicity, education), labor force participation, unemployment and underemployment rates. Its main advantage over the Population and Housing Census is that it is collected annually. Table 2.2.1: Main characteristics of GSO LFS 2007 & MPEFPS 2008 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Sample design Total # of provinces 53 53 53 61 61 61 61 61 64 64 64 64 2,05 2,85 2,85 2,85 2,85 3,23 3,23 3,23 3,23 3,23 3,35 3,35 # of selected PSUs 8 6 6 6 6 8 8 8 0 0 6 6 144,00 125,09 125,09 109,54 109,54 109,54 109,54 109,54 97,14 97,14 100,68 100,69 # of households (theoretical) 0 0 0 0 0 0 0 0 0 0 0 0 125,09 126,59 105,83 83,20 105,86 # of households (real) 0 5 0 0 0 N/A N/A N/A N/A N/A N/A N/A Panel component No No No No No No No No No No No No Content # of questions 60 49 49 67 59 62 59 53 34 33 24 22 Unemployment Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Underemployment Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Employment Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Institutional sector Part Part Part Part Part Part Part Part Part Part Part Part Business size No Part Part Part Part Part No No No No No No Monthly income No No No Part Part Part Part Part Part Part No No Second job No No No No No No No No No No No No Source: Roubaud (2008) 1.3. LABOR FORCE SURVEY (2007 GSO) As mentioned earlier, in 2007, the LFS was also collected by GSO and its quality seemed improved. The sample size was 173,000 households, randomly selected from 5768 enumeration areas in a 2-stage stratification sampling design. The table below summarizes the 2007 LFS and the 2008 MPEFPS by GSO.

Table 2.3.1: Main characteristics of GSO LFS 2007 & MPEFPS 2008 2007 2008 Sample design 64 64 Total # of provinces 5768 3840 # of selected PSUs 173040 380000 # of households (theoretical) 170090 N/A # of households (real) No Partial Panel component Content # of questions 60 15 (Labor Market) Unemployment Yes Yes Underemployment Yes Yes Employment (industry, occupation, job status) Yes Yes Institutional sector (informal sector) Yes Partial Business size Yes No Monthly income Yes No Second job Yes No Source: Roubaud (2008)

One major improvement of the 2007 LFS by GSO is the inclusion of informal sector information. A set of questions on job status, job characteristics (type of contract, social security, and paid leave), and enterprise characteristics (institutional sector, size, industry, business register, type of accounts) was included to measure informal employment and identify informal sector production units.1

1.4. NON-FARM INDIVIDUAL BUSINESS ESTABLISHMENT CENSUS/SURVEY (1995, 2002-2006, GSO) The first establishment census was in 1995. It is unknown whether any establishment survey was collected between 1996-2001. Table 1, taken from Razafindrakoto (2008), show the census/survey schedule and sample sizes from 2002 through 2006. Other than the three-volume publications of the 2002 census results, I have not been able to find or see the results or publications of other years’ censuses or surveys. More information is needed from GSO to understand this census/survey better.

1 The 2007 LFS cannot provide information on the informal sector’s contribution to national accounts. For this, one should refer to the Household Business and Informal Sector Survey also by GSO. Table 2.4.1: Sample design of establishment survey/census 2002 2003 2004 2005 2006 Total # of provinces 61 61 64 64 64 Total # of communes (primary sampling 10,533 10,533 10,964 10,964 10,964 units) % of selected 100% 100% 40% 20% 20% communes (census) (census) % of establishments surveyed in selected 100% 100% 100% 100% 100% communes Source: Figure 3 in Razafindrakoto 2008

An establishment is defined as a single physical location where economic activities took place for at least three months within the past year. The establishment must have an owner/manager/leader. All agricultural establishments are excluded, as well as all businesses that do not have a fixed location (this means a large part of the informal sector would be excluded).

The questionnaires of the Establishment Census/Survey have poor content. It includes questions on the establishment’s identification, its manager’s or director’s information, type of enterprise onwership, business activity, # of employees (by gender, education/professional training), turnover, and IT applications. Compared with the Enterprise Survey, the Establishment census has two advantages: it has employment data by skill, and it covers more (but still incomplete) of the informal sector. But these advantages cannot make up for the inavailability of raw data. Furthermore, no information on labor income is available. So overall, this data set is of little use.

1.5. ENTERPRISE SURVEY (2000-2007, GSO) This survey is much better documented than the LFS and the establishment survey/census, and raw enterprise- level data are also available, making it much more popular than the other two. It is an annual survey, started in 2000,2 covering all enterprises with independent accounting systems that are established under and governed by the Law on SOEs, the Law on Cooperatives, the Law on Enterprises, and the Law on Foreign Investment in Vietnam. All sectors/industries (including agriculture) are covered. Because the criterion to be included in the survey is the establishments and governance under the law, the informal sector is largely ignored, even more so than the establishment survey/census. The enterprise survey’s qusetionnaire content is much more substantial than the establishment survey/census, including various business and production activities: labor and employment, incomes

2 This survey existed in the 1990s, but in a somewhat different form and quality was considered lower (info was taken straight from firms’ financial statement). I have yet to collect more information about this survey in the 1990s. of employees, # of establishments and equipments, assets and liabilities, investments, capital stock, production costs, turnover, products, profits, inventories, taxes, R&D investments, IT application, and others.

Year Size 2000 42,288 2001 51,680 2002 62,908 2003 72,012 2004 91,709 2005 112,795 2006 131,163 2007 155,607

Certain parts of the questionnaires (e.g., production costs) will be collected for a selected sample whose size is about 15% of total # of all enterprises as defined earlier. This sample is representative at the province and city level. Sampling design sounds complicated.3

Overall, the enterprise survey is probably the “best available” (but not necessarily good) source for constructing a time-series (8 years so far) of formal labor market indicators for Vietnam, such as employment’s growth and changing structure (by industry and province), wage growth and wage inequality across sector/industry. Studies of labor market demand at either enterprise or industry level are possible using this data set. There are many problems with this data set nevertheless. The first problem is the comprehensiveness and representativeness of the data. Aggregate variables produced by this data set must be considered indicative only, not representative of the Vietnamese economy, for two reasons.4 First, the survey only covers the formal sector in Vietnam. The informal sector, which could be as large as half of the whole economy, is excluded. Second, even within the formal sector, investigation of the data sets suggests that not all registered enterprises are interviewed each year. We cannot assess for certain how serious this issue is, but it has been found that the trend of some industries (e.g., utility or finance) can be majorly changed or reversed because of incomplete coverage.

Another problem with the Enterprise Survey is that information on workers’ characteristics, particularly education and skill levels, are usually not collected, so more in-depth analysis of labor demand and wage by skill would not be

3 The sampling will follow a random and equal distribution method in combination with a priority given on the law of large number. In particular: In each group of the cell of the survey frame matrix, enterprises will be divided into 2 categories, namely those above and those below the intersection. The intersection will be determined by the enterprise whose number of employees is closest to the group’s average number of employees. The category of enterprises lying above the intersection will be distributed with 50% of the group’ samples while the other category of enterprises will be distributed with the remaining 50% of the group’s samples. After a certain number of samples has been determined for each category, each category will then be sampled by a random of equal distribution with the distance of adjacent samples being K. (K = the number of the category’s enterprises divided by the number of the category’s samples).

4 This point is particularly relevant when, say, we estimate elasticity parameters using these data and try applying the parameters to a CGE model of Vietnam possible. An exception is the year 2001, in which data on enterprise headquarters’ labor by skills are available. Also, the most recent survey in 2007 did have detailed employment by age and skill/education for the whole enterprise, not just the headquarter.

1.6. VIETNAM (HOUSEHOLD) LIVING STANDARD SURVEY (1993, 1998, 2002, 2004, 2006, 2008, GSO) This is a standard World-Bank-type income and expenditure household survey, designed to study living conditions and poverty and inequality issues. It has been collected every few years since the early 1990s (1993, 1998, 2002, 2004, 2006, 2008). The earlier rounds (1993 and 1998, called Vietnam Living Standard Survey, or VLSS) were smaller in size (4,800 and 6,000 households respectively) and were representative at the national, rural/urban, and regional levels. These two surveys are usually considered high quality, especially in comparison to most other surveys in Vietnam. Later rounds (called Vietnam Household Living Standard Survey, or VHLSS) were much larger in size (75,000 in 2002 and 45,000 in 2004 and 2006) and were representative down to the provincial level. However, the surveys’ quality, compared with those in the 1990s, might have been sacrificed to a large extent in exchange for larger sample size.

While the VHLSS is a reasonably good survey and it has an employment section (employment status, working hours, and wages of all family members), it was not designed to be a labor force survey. As a result, it cannot provide the best labor market indicators. For example, an attempt to study employment and wage at industry- level soon reaches a dead-end, because the survey was not designed to be representative at the industry level (there were 99 industries in the survey). Nevertheless, it still provides an excellent source to study wage and wage determinants (see review of this topic in section 2 of this report). It makes possible the analysis of wage determinants and skill premium at the individual level, of wage growth, wage inequality and its relation to poverty and income inequality. The survey also offers indicators on labor force participation overall and by income activity (farm vs. non-farm, wage vs. non-wage, and by industry and occupation for those in wage employment). Its advantage is that the survey does not exclude the informal sector as in the other surveys/censuses, but then the questionnaires do not allow us to properly distinguish informal and formal sector employment.

1.7. AGRICULTURAL CENSUS (1996, 2001, 2006, MARD) This is a large census, covering all agricultural households, farms,5 enterprises, and cooperatives in Vietnam, collected every 5 years and started in 1996. There are separate questionnaires for each type of unit, but they all cover similar information: employment (by skill and other characteristics), land owed and rented (in or out) and their usage, cultivating area of different crops, livestocks, aquaculture, fisheries, forestry, farm equipments and other productive assets (e.g., fishing ships). In the household questionnaire, there are also brief questions on housing. In the farm, enterprise, and cooperative questionnaires, there is more detailed information on business

5 A farm is defined as an agricultural household meeting certain production size criteria. performance (revenue and expenses for each type of crop or agricultural activity), labor cost, capital, investments, credit, and information technology. Note that the farm/enterprise/cooperative questionnaires share several similarities with the Enterprise Survey’s questionnaire, so there is a potential to combine the two data sets if one worries that the Enterprise Survey does not cover enough the agricultural sector.

This census provides a good source for agricultural sector’s employment data. The original data set, while available (apparently), is quite big (80GB). But similar to the LFS or the Population and Housing Census, the Agricultural Census is better for counting purposes, providing indicators of employment (by skills) and wages. For deeper analyzing purposes, a more extensive data set such as the VHLSS is still a better alternative.

1.8. DANIDA’S SME SURVEYS This is an enterprise survey, collected in 2005, 2007 and 2009.6 The sample sizes for the three years are 2603, 2492, and 2492, respectively. The same enterprises were collected in all years. The 2005-2007 panel has 2,298 firms.7 The survey covers small and medium non-state manufacturing enterprises8 in 10 selected provinces: Ha Noi, Phu Tho, Ha Tay, Hai Phong, Nghe An, Quang Na, Khanh Hoa, Lam Dong, HCMC, and Long An. A stratified random sample was chosen such that the data set is representative at the provincial level.

The questionnaire is rich. It has the following modules: general characteristics; enterprise history; household characteristics of owner or manager; production characteristics; sale structure and export; indirect cost, raw materials, services; fees, taxes and informal payments; employment; investments, assets, liabilities, and credit; networks; economic environment, constraints, and potentials. The employment module asks information on employment by gender, full vs. part-time status, occupation, workers’ training, labor union, workers’ benefit, HIV/AIDS program, recruiting and firing process, wage setting process. Unfortunately, the survey does not specifically ask about educational or skill levels of the employees, only their occupation, i.e., managers, professionals, sales, service or production workers. Professionals are defined as those having university/college degree.

Since this survey creates a panel data set and since the questionnaire is quite intensive, the survey can be used for rigorous econometric analysis. For example, one can examine the determinants of employment and employment growth, of labor productivity and labor productivity growth, or of hiring/firing decision. But like other surveys and census, this survey does not ask much information on workers’ characteristics and work hours. So we still cannot study labor demand by skill. Furthermore, the survey is limited to small and medium manufacturing enterprises in a few selected provinces only, so generalization must be treated with caution.

6 2009 data are currently being collected on the field. The survey was also done in 2002 and in the 1990s, but the earlier surveys have not been used much, partly because they have lower quality compared with the later rounds, partly because they are becoming oudated. 7 The attrition rate from 2005 to 2007 was 12%. 8 Small enterprises are defined as having less than 50 employees. Medium enterprises are defined as having between 50 and 300 employees. 1.9. SUMMARY In summary, labor market data in Vietnam, particularly data on labor supply/demand and wages, are available but they come from various sources with varying quality, coverage, and representativeness. Especially lacking are labor market data that are inclusive of the informal sector and migration activities. All current surveys/censuses mostly overlook these issues, yet these are important aspects of the economy in general and the labor market in particular.

But the future does seem brighter, for improvements are being either proposed or made in various fronts. Starting in 2009, the LFS is to be collected as a quarterly survey, and its quality and content are expected to improve much compared with the earlier LFS. Informal sector information will be collected in the future LFS, or potentially in a new survey called the Household Business and Informal Sector Survey (read Razafindrakoto 2008 for more details). Regarding migration data, the VHLSS, starting in 2006, include questions that allow us to clearly identify who is a migrant or non-migrant, such as residence status, length living in a particular city/province, etc.

2. LITERATURE REVIEW9

2.1. THE DEMAND SIDE OF THE LABOR MARKET Studies of wage (to be reviewed in a later section) tend to link changes in relative demand for different types of labor to changes in wage inequality. But what are the underlying causes of changes in relative demand for different types of labor? To understand this, we need to study the factors influencing employment growth and the structure of labor demand. The literature on the demand side of the labor market in Vietnam has explored this topic to some extent. In particular, existing studies have (i) described employment trends over time by sector/industry, etc. (ii) related employment growth to output growth, structural change, and productivity change, (iii) related employment at the industry level to trade/openness or other policy variables. 10 Most of these studies use aggregate data at the economy or industry levels. Very few are at the firm or micro levels.

At the economy or industry level, several studies (Pham et al 2006, Jenkins 2004, Belser 2000) estimate growth- employment elasticities and discuss employment trends in varying level of details. Jenkins (2004) for example examines employment data for the years 1990-2000 using all available sources (MOLISA’s Labor Force Survey, GSO’s Enterprise Survey, GSO’s VLSS, and GSO’s Population and Housing Census). The author notes that the various sources turn out to yield similar trend: the rate of employment growth in Vietnam was in the range of 1.8 to 2.4 percent per year, which was well behind both GDP growth and labor supply growth. The growth- employment elasticity is about 0.2 in the 1990s, much lower than that for other neighboring countries (Indonesia 0.73, Malaysia 0.65, Thailand 0.77, China 0.44). Employment growth during the 1990s were also very unevenly

9 In this report, I review only papers written in English, due to inaccessibility of papers written in Vietnamese. I hope to expand the review to the Vietnamese literature in the future. 10 I am not aware of papers studying the impact of technology on labor demand in Vietnam. distributed, concentrating largely on the service sector. As a result, the large share of laborers continue to stay in the agricultural sector despite its slower growth and declining share in GDP. At the same time, the industrial sector generated very few jobs relative to its fast growth and increasing share in GDP.

Belser (2000) examines household data from the VLSS 1993 and 1998 and arrives at a similar finding: industrial employment grew 4% per year from 1993 to 1998, which is low compared with industrial growth at 11-13% per year. He attributed this slow employment growth to capital-intensive, import-substituting nature of the state sector and foreign investments, which dominate industry. Comparing Vietnam’s endowments with neighboring countries, the author concludes that Vietnam has high endowment in skilled labor and relatively little land, forming the basis for strong comparative advantage in manufactures. Using earlier statistical results from Wood and Mayer (1998) who regressed countries’ share of manufactures relative to primary exports on its human capital and land per worker, Belser estimates that Vietnam could triple its manufacturing exports and create about 1.6 million more jobs in exports sector in the near future. Overall, for the case of Vietnam, Belser recommends labor- intensive manufacturing growth based on comparative advantage, and is against capital-intensive manufacturing growth.

Pham et al (2006) estimate growth elasticity of employment, its levels as well as changes over time, for the period 1986-2001, using 2 methods: (i) compute arc elasticity as in Jenkins (2004), and (ii) econometrically estimate a double-log linear equation relating employment to GDP. Again, the authors find low growth elasticity of employment, confirming the results in the other two papers. They also share a similar view with Belser that the promotion of import-substituting heavy industries dominated by SOEs is to be blame.

Jenkins (2004) nevertheless disagrees with this conventional view. He instead argues that it is primarily the changes within industries resulting in increases labor productivity from very low level in the early 1990s, that have prevented industries from absorbing more workers. In proving this point, Jenkins decomposes employment growth into output growth, structural change (i.e., between industry changes) and productivity growth (i.e., within industry changes). The author finds evidence that productivity changes play a far more significant role in limiting employment growth than sectoral changes.

Beyond computing growth-employment elasticities and descriptive analyses, there are not many papers that study factors impacting labor demand in Vietnam, except Jenkins (2002) and Tran and Heo (2009). Jenkins (2002) explores the impact of trade and FDI on employment in Vietnam in the 1990s. To study the impact of trade, the author uses 3 methods. First, factor content analysis reveals that unskilled women are the main beneficiaries of increasing trade openness. Second, growth decomposition analysis suggests that export growth contributed significantly to employment, while increased import competition had a negative effect both directly and indirectly through rationalization of producers facing foreign competition (see Jenkins 2004 for more details on this method). Third, the paper also regresses employment at industry level on a number of variables including indicator of openness (output, import share, export share, capital-intensity, state ownership share) for the years 1994-1999 and using simple fixed effect OLS.11 Imports are found to have a negative and small impact on employment. On the other hand there is some evidence that exports have a positive effect on employment (over and above that which can be attributed to the additional output that comes about as a result of exports).

To study the direct impact of FDI on employment, Jenkins simply examines employment of foreign firms vs. Vietnamese firms. Within almost every industry, foreign firms create less employment per unit of output than local firms, mostly due to technological/productivity difference. The author then regresses growth of employment at the 2-digit industry level on the change in the share of foreign-invested firms in gross output. The results again indicate limited impact on employment: industries where foreign ownership has increased significantly between 1995 and 1999 have tended to lag behind in terms of employment growth. Regarding indirect effects through linkages, the positive effects on local producers tend to be small because foreign firms tend to import most their inputs. At the same time there is evidence of crowding out of local firms, and of rationalization by state firms in response to increased competition, both of which tend to reduce employment.

Tran and Heo (2009) uses econometric method to estimate the determinants of employment at industry level, similar to Jenkins (2002), but for more recent years, 1999-2004. The estimates by Tran and Heo are major improvements compared with Jenkins’ estimates because: the set of independent variables include wages (in addition to output, import share, and export share); they take into account the dynamic nature (due to adjustment process) of labor demand; and they use a system GMM estimator which is more appropriate for short panel data sets. Despite differences in time period and estimation method, Tran and Heo find a similar result to Jenkins: higher export share is associated with higher labor demand. A 1% increase in export share would expand employment by 0.11%, which is smaller but perhaps more reliable than the larger range of estimates by Jenkins (0.15-0.27%). But unlike Jenkins (2002), Tran and Heo find no statistically significant impact of import share on employment.

As mentioned earlier, studies of labor demand at the firm level are rare in Vietnam. The only paper we are aware of is Belser and Rama (2002), which analyzes the determinants of labor employment at the firm level, but for a very limited sample and an ad hoc purpose: to predict the number of workers who will lose their jobs if SOEs are privatized or restructured. Using a non-random sample of 338 enterprises whose state share of capital ranges from zero to one hundred percent, the authors regress (log of) enterprises’ employment on the state share of capital, controlling for a variety of enterprise characteristics such as age, sector of activity, region, export orientation, capital stock.12 The results reveal substantial labor redundancy (defined as the percentage of workers

11 The standard method for estimating labor demand involves regressing employment on output and the cost of labor relative to capital (assuming capital and labor are the only two factors of production). Jenkins had to ignore the relative cost of labor however due to lack of data. 12 Note that the regression would ideally include real producer wage as an explanatory variable, but such data are not available, so the authors resort to using replacing this variable by exogenous proxies such as dummy variables for sector of activity and region, or enterprise’s age, capital stock. reduced if state share of capital were brought down to zero) in SOEs: about 50%. Labor redundancy varies dramatically across sector of activity, and is only weakly related to indicators of profitability, productivity, and labor cost.

To summarize, the current literature on the demand side of the labor market in Vietnam has focused on studying factors influencing employment growth and the structure of labor demand at the aggregate level. But not all topics have been covered. For example, while the impact of trade and openness on labor demand has received much attention, the impact of technological change (which might also be a result of trade and openness) has not been studied at all. Also, analysis of employment trends or labor demand by workers’ skills and other characteristics are rare in Vietnam, probably because of lack of data. The current LFS does provide employment data by skill, but as mentioned in the data diagnosis, the quality of this data set is highly questionable and raw data are not available. Furthermore, the survey’s content is quite poor, limiting its use in econometric analysis. The higher quality Enterprise Survey unfortunately does not collect employment data by skill, except for the year 2001 and 2007. Even then, the 2001 and 2007 surveys only count the number of employees by skill, and no information on wage by skill is available. There is an urgent need to collect employment and wage data by workers’ characteristics, at least in smaller-scale surveys.

Future research can continue in the current direction, i.e., studying employment growth and structure of labor demand at the aggregate level, updating data to more recent years. But there are limitations to aggregate studies, and more effort should be directed toward utilizing micro/enterprise level data to study firms’ labor demand behavior. There are still no estimates of fundamental parameters that describe how firms or industries will react to changes in external economic environment: (a) cross- and own-price elasticities of labor demand, and (b) elasticities of substitution of labor for other inputs such as capital and energy (and vice versa), (c) elasticities of substitution between types of workers, and (d) elasticities of substitution between workers and hours. In addition, one might also be interested in studying how labor demand varies over time, or how it varies with firms’ characteristics and with changing economic environment. In other words, how do the above elasticity parameters change over time? What are the roles of firms’ characteristics and exogenous shocks such as trade policies, labor regulations, technological change, etc. in inducing such changes in parameters and hence in employment?

Estimation of (a) and (b) are possible employing data sets such as the Enterprise Surveys, although it must be emphasized again that the Enterprise Surveys only represent the formal sector in Vietnam. Estimation of (c) and (d) might not be possible using the current large-scale surveys or censuses, because these surveys/censuses do not collect detailed information on workers’ characteristics or work hours.

2.2. THE SUPPLY SIDE OF THE LABOR MARKET At the aggregate level, studies of the supply side of the labor market simply involve describing the availability of labor supply (over time and by skill level), the average labor participation or unemployment rates (by gender and age), and the average working hours or underemployment rate. In Vietnam, data from the LFS as well as from the VLSS/VHLSS can be used for this purpose. A number of papers have done so at varying levels of details. For instance, Gallup (2002), using VLSS 1993 and 1998, documents that there is a high rate of labor force participation in Vietnam in the 1990s. Even when household work is excluded, 81% of all women and 85% of all men aged 16- 60 were working in 1993, generally higher in the rural areas than in the urban areas. Over the period 1993-1998 this rate increased by 2% for women and unchanged for men. The author also documents that unemployment is very low in Vietnam (0.5% in rural areas and 1.5% in urban areas) and fell by a tenth of 1% from 1993 to 1998. Hours worked also increased rapidly (18%) from 1993-1998, contributing to the large rise in wage income. The highest working hours are in the major cities (HCMC, Hanoi) as well as medium cities, but the increase in working hours in major/medium cities seem more modest than rural and small urban areas.

In studying the supply side of the labor market, the most interesting analyses are at the micro level: studies of household or individual labor supply behavior, which determine the availability of labor supply observed at the macro level. A reliable estimate of elasticity of labor supply in general, or the decision whether or not to work and how much to work, is critical in predicting the likely impact of of changing employment opportunities and various policy changes.13 Yet, such studies are rare in Vietnam, for the supply side of the labor market in Vietnam has received little attention, even less than the demand side. To our knowledge, there are only two papers that have investigated household labor supply behavior: Edmonds and Pavcnik (2005) and Tran (2008).

Edmonds and Pavcnik (2005) first discuss the theoretical channels (the substitution effect vs. the income effect) through which changes in product prices following international product market integration could affect an individual’s labor supply. They then relate changes in participation and hours worked overall and in various economic activities to changes in rice prices across communities and time, using the panel VLSS 1993-1998. The main results are that higher rice prices are associated with lower participation in wage work by boys, girls, and young adults, lower participation in household production by adults, less time devoted to household production for all age groups, and more time devoted to wage work by adults.

Tran (2008) studies a very particular issue: how public transfers might impact the labor supply of the elderly, or their decision to retire. He formulates a simple overlapping generation model to investigate this issue, which show that the interaction between transfers received and labor supplied by the elderly result in a decline in the opportunity cost of retirement, magnifying the crowding-out effects of public transfers. He uses the VLSS 1998 and finds evidence in support of the crowding-out hypothesis (controlling for endogeneity of public and inter-

13 For example, in November 2007, Vietnam introduced its first ever personal income tax law, to be effective starting Jan 2009, but delayed further to May 2009 to stimulus spending in response to the economic crisis. What is the effect of this law on household welfare, income distribution, and government revenue? To answer this question, one needs to know how individuals would react to the introduction of the tax, i.e., one needs to know elasticity of labor supply with respect to after-tax wages. In microsimulation method, one would use a representative data set such as the VHLSS to first estimate this elasticity, then apply the elasticity to first estimate changes in labor supply in response to the tax, then changes in individual’s incomes. household transfers): an increase in public pension income or inter-household transfer income by $100 increases the probability to retire by 7% or 13%, respectively.

To my knowledge, there are no studies of long-run labor supply in Vietnam, which will answer questions such as how individuals’ labor supply fluctuate over time with age and changing income/savings. This is due to the fact that there are no data sets that track labor supply over a long period of time.

2.3. THE DIS-EQUILIBRIUM: WAGE DETERMINANTS & WAGE INEQUALITY Experience of NIEs in the 1960s and 1970s show that following openness there should be a decreased wage inequality. The main theoretical explanation for this phenomenon is the Hecksher-Ohlin model, which maintains that relatively abundant factors (unskilled labor in the case of Vietnam or NIEs during the 1960s or 1970s) would benefit from trade liberalization and skill premium would decrease, leading to reduced wage inequality. As countries specialize in producing goods that they have comparative advantages in, i.e., as sectors employing the relatively abundant factors expand, returns to low-skilled labor would increase relative to the returns of high- skilled labor, and so the skill premium and wage inequality should decrease.

Yet the experience of more recent liberalizers, mainly those in Latin America, contradicted this hypothesis: skill premium and wage inequality increased for these countries (see Arbache et al 2004 for a review). Transition economies such as Poland, Hungary, and China also had a similar rise in wage inequality as in Latin American countries. Several theoretical explanations have been proposed to explain this alternative labor market outcome of trade liberalization. One popular explanation is the technology hypothesis: trade increases the flow of FDI and imports, bringing new technologies into the liberalizing country. If skills are complementarity, this inflow of technology raises the relative demand for and returns to skilled labor (Feenstra and Hanson 1995). Another popular explanation is that the Hecksher-Ohlin model is still valid, and that middle-income countries like those in Latin America are in fact abundant in skilled labor compared with countries like China and India. As China and India enter the world economy, Latin American countries turn to the production of semi-skilled goods, causing a rise in the returns to skilled premium.

Within this context, prior expectation about changes in wage level and wage inequality in Vietnam is not obvious. Can the data tell us what happened in the past 20 years? Fortunately, thanks to the availability of wage data from the VLSS/VHLSS, this is a relatively well-researched area, although more research still need to be done before we can reach firm conclusions, as will be argued later. Papers that have studied wage levels and wage inequality in Vietnam include Gallup 2002, Moock 1997, Pham and Barry 2007a, Pham and Barry 2007b, Liu 2005, Liu 2006, and Cong et al 2006. Most of these papers estimate variations of the Mincerian earnings equation using different estimation methods: Gallup (2002), Liu (2005) and Cong et all (2006) employ simple OLS, Liu (2006) employs Hay’s two-stage method (a generalization of Heckman’s approach), Pham et al 2007a and 2007b employ mean and quantile regression approaches. These papers also apply various decomposition methods to explore characteristics that explain the wage gaps, and study factors that contribute to changes in wage gaps. Despite variation in methodology, all of them use VLSS/VHLSS data, and so they arrived at relatively similar results, with some exceptions. 1. Wage levels increased rapidly in Vietnam since doi moi (the papers used up to VHLSS 2002 data, so most of the results apply up to the year 2002 only). Gallup (2002) documented that average hourly wages grew by 10.5% per year in real terms between 1993 and 1998, considerably faster than the 8% growth rate of household income per person or the 6% growth rate of output per person in the whole economy. Descriptive analysis in Cong et al (2006) shows that while mean hourly wages increase for most education levels, the increase is larger for higher education, indicating a rise in skill premium. 2. Returns to education is found to be low in Vietnam, but increasing over time, consistent with the statistics in Cong et al (2006) just presented.14 This result was agreed by both Gallup (2002) and Liu (2006), who both used VLSS 1993 and 1998 to estimate Micerian earnings equation for wage workers in Vietnam. While the two papers’ results agree qualitatively, they differ quantitatively. Gallup (2002) estimated that the internal rate of returns to one additional year of schooling was 1.9% in 1993, and increased to 3.5% in 1998. Liu (2006) finds that this rate was higher: 5.9% for males and 4.2% for females in 1993. By 1998, the situation has reversed, as the rate for males decreased to 3.5% for males and increased to 4.8% for females. Such differences in results might be explained by the fact that even though the two authors used the same data set, their estimation procedures are different. Gallup used simple OLS, while Liu used Hay’s two-step approach. Liu included many more independent variables in the equation than Gallup did (regional dummies, industry dummies, employers’ sector by ownership type). Finally, Liu estimated two equations separately for male and female workers. 3. There is strong evidence of wage differential and labor market segmentation in various dimensions: gender, ethnicity, and regional. o Gender: as elsewhere, men on average earn more than women in Vietnam, even after controlling for differences in age, education, and experience etc., but this gap has been reducing over time. Pham and Barry (2007b) find that this earning gap was 31% in 1993, but reduced significantly to 19% by 1998 and exhibited stability until 2002. Liu (2006) documented that from 1993-1998, the mean real wage increased by 68% for males but 84% for women. Cong et al (2006) find that females get paid 15% less than men on average, controlling for differences in human capital (they use VLSS 1998 and VHLSS 2002). The gender wage gap was found to narrow down with primary and secondary education but not higher levels of education. This is because female workers with lower educational levels benefited dis-proportionately from labor-intensive export growth, a result found in many other studies of trade liberalization impacts on Vietnam (e.g., Jenkins 2002).

14 Arcand et al (2005) use VLSS data and estimate returns to education in Vietnam. They explore different sets of instruments that rely on demand and supply side sources of variation in schooling, and find that IV estimates of returns to education are substantially higher than OLS estimates. But this paper is in French and so I cannot provide further details. o Ethnicity: It is estimated by Gallup (2002) and Pham et al (2007b) that ethnic minority earns 10% (in the 1990s) and 11% (in 2002) less than the Kinh majority and the Chinese, respectively. Pham et al (2007b) use “index number” decomposition method suggested by Oxaca (1973) to find that ethnic wage gap is largely attributable to differentials in returns to endowments (i.e., treatment effect, not endowment effect, is the dominating force). o Regional: Gallup (2002) found that HCMC and Hanoi had higher wages than the rest of the country in 1993, and their wages grew much faster up until 1998, widening the rural-urban wage gap. Note that these two cities account for 25% of the whole country’s wage jobs even though they account for only 8% of the country’s population. But outside of these two cities, wages are surprisingly similar across regions in 1993, and became even more so in 1998. That is, regional wage gap actually declined outside of HCMC and Hanoi. Controlling for differences in workers’ characteristics, HCMC and Hanoi workers had a 80% and 47% wage premium. Regional differences in returns to education are strong. The rates of return to schooling are esp. low in the Central Highland, in small urban areas, and in the rural South where levels of education are also low. The rate of return to schooling in Hanoi and the rural North is more than double the level in the three lowest regions. Interestingly, Liu (2006) did not find the same result that the urban dummy has a statistically significant effect on wage although the coefficient sign is positive. But she did find that there are significant wage gaps across rural regions of Vietnam. In particular the Mekong River Delta and the Southeast have high wage premium compared to other rural regions. 4. Overall wage inequality seem to fall modestly in the 1990s. Gallup (2002) found that wages of the poor, middle, and rich grew at 14%, 10%, and 12% per year, respectively. Different measures of inequality all pointed to a decrease in wage inequality. But for Hanoi and HCMC, wage inequality actually increased quickly in the 1990s, while it stayed the same in medium and urban areas, and fell sharply in the rest of the country. 5. Gallup (2002) also estimates contribution of each income source to total income inequality and finds that wage employment contributes almost a third of total income inequality, about on par with household enterprise income and other income.

Cong et al (2006) analyze the impact of trade liberalization on the wage structure in Vietnam, in particular the rate of returns to education and gender wage gap by using two strategies. The first one is to follow a difference-in- difference method to examine relative shifts in the wage structure between 1998 and 2002 in traded and non- traded sectors respectively, but they find no significant evidence that trade liberalization has either increased or decreased the returns to education or the gender pay gap. This is probably because their distinction between trade and non-traded sectors are too coarse (the fomer includes agriculture, mining, and manufacturing, and the latter includes the rest). The second strategy is to estimate the Mincerian earnings equation again but include a measure of openness (imports and exports share) for industries at the 2-digit level. Thanks to the panel analysis which helps control for much of industry-level heterogeneity, they find that openness has a positive coefficient, and this coefficient is higher at higher levels of education. This result indicates the increase in skill premium seems smaller in industries with higher imports and exports share.15 Furthermore, openness when interacted with the dummy female has a positive coefficient, suggesting that gender wage gap declines with openness. While the results of this paper are interesting, their reliability can be improved much further in two feasible ways. First, the authors chose to use repeated cross sections of the 1998 VLSS and the 2002 VHLSS, which form a semi-panel data set that helps reduce industry-level heterogeneity. But real panel data are available (1993 and 1998, or 2002 and 2004, 2002 and 2006) and should be used to reduce individual-level unobserved heterogeneity. Second, the literature has widely cited sample selection as a serious issue in estimating wage equation. Sample selection is caused by the fact that in an OLS regression, unemployed people or those having non-wage employment are not included because their wages are not observed. VLSS/VHLSS have data on all types of workers, so there is no reason to not employ some variation of the Heckman two-step approach to account for this sample selection (see Liu 2006 for example).

In summary, Vietnam’s wage level and skill premium have been increasing, so one might say that Vietnam’s experience is similar to Latin America, China, and other transition countries. But this increase in skill premium in Vietnam has not necessarily led to higher overall wage inequality. Indeed, wage inequality has declined modestly and wage employment has not been identified as a major contributor to inequality, at least in the 1990s and in areas outside of HCMC. This might be because in the early 1990s, wage employment was very limited in Vietnam, and richer households started out with most of their income from sources not in wage employment. All of this might have changed in the 2000s however, as the share of the agriculture declines in national GDP and wage employment and wage income become more and more important. To the author’s knowledge, no paper has updated data on wage in Vietnam using the most recent VHLSS 2004 and 2006. A more comprehensive study of wages in Vietnam is called for. Results from this study should ideally be linked to studies on both the demand side and supply side of the labor market (reviewed in previous sections), so that not only do we understand the changes in wage level and wage inequality, but we also understand the mechanism explaining these changes.

3. CURRENT SITUATION OF THE VIETNAMESE LABOR MARKET

3.1. LABOR SUPPLY16 There is no doubt that Vietnam has an abundant and growing labor force, as can be seen in table 4.1.1. Each year, there are about one million new entrants to the labor market, so the pressure to creating new jobs is high. The

15 The author drew the conclusion that increases in openness tend to reduce skill premium. However, I think this is quite a stretch from the available evidence. All that one can say is that the increase in skill premium is less in industries with higher level of openness. 16 I mostly present data from the VLSS/VHLSS for this section because I believe they are of higher quality than LFS data. Differences and inconsistencies between the two sources will be discussed in footnotes. skill level is low, albeit increasing. Table 4.1.2 and figure 4.1.1 show that the majority (more than 40%) of the Vietnamese labor force is unskilled (primary school degree or no degree at all). From 1993 to 2006, the proportion of unskilled workers decreased by only 5%, from 49% to 44%. Most of this decrease in unskilled workers was replaced by semi-skilled workers (lower or upper secondary school degrees). 17 The proportion of high-skilled workers (college degree and above) increased from 1.8% in 1993 to 4.2% 2006. Despite this 100% increase in 13 years, 4.2% is still very low, too slow to catch up with other countries in the region, which partly explains Vietnam’s low labor productivity (see section 4.3). However, could this increase be enough to slow down the increase in skill premium (see section 4.4)?

Table 4.1.3 similarly shows the gradual increase in average years of schooling of the Vietnamese labor force. Table 4.1.4, using data from the LFS, shows that the proportion of workers with college degree and above increased slowly, consistent with VLSS/VHLSS data. The table also reveals that the proportion of workers with vocational training decreased over the years, causing the proportion of all skilled workers to also decrease. This is a particularly alarming trend.

Tables 4.1.5 and 4.1.7 suggest that Vietnam’s overall labor force participation rate decreased while the average number of hours worked increased over time.18 At 81% in 2006, Vietnam’s labor force participation rate is still high. The declining trend of participation rate is consistent with the theory that as income rises, more people stay out of the work force, since the income effect dominates the substitution effect. But this trend is not the same everywhere. While labor force participation rate dropped considerably and average number of hours worked increased significantly for the rural areas, an opposite pattern happened in the urban areas. This can be explained by the fact that rural workers are mostly self-employed farmers and hence are often under-employed. As more of them move into wage employment, full employment should take over under-employment, causing the average work hours to increase. At the same time, rising hourly wage as well as rising number of work hours implied an increase in income per worker. Assuming income effect dominates substitution effect, participation rate would decrease since more household members choose to stay out of the labor force as income rises. As for urban workers, as hourly wage increases, they seem to substitute hours with more worker, i.e., decrease the work hours while increase the number of household members in the labor force. It would be interesting to study why this is the case and to understand household behavior regarding their propensity to substitute work hours with workers and vice versa. This is an area that has not been studied by the literature. Data from the living standard surveys would seem sufficient for this purpose.

17 LFS data in MOLISA (2006) show that from 1995 to 2005 the percentage of employed workers with upper secondary school degree decreased, not increased, while those with primary school degree increased. This doesn’t seem sensible. But LFS data in MOLISA (2009) noted an increase in the percentage of upper secondary school and higher degree, from 14.1% in 1997 to 23.6% in 2007. This is much more in line with VLSS/VHLSS data. 18 LFS data show a decrease in participation rate overall and for both rural and urban areas. Participation rate is defined in LFS as the percentage of employed and unemployed workers in the past 7 days in the total population aged 15 and above, so those rates are lower than the rates reported in table 4.1.4, which are based on employment in the past 12 months and also exclude unemployed workers. Finally, table 4.1.6 shows the increasing trend of overall wage employment, an expected fact. What is more interesting is that while the rate of wage employment participation increased significantly for rural workers, it decreased for urban workers. Perhaps, rural workers have been diversifying from self-employed work to wage work, while urban workers have been moving from wage work to self-employed nonfarm businesses. Also, while ethnic minorities always had lower wage employment participation rate than the majority (ethnic Kinh and Chinese), the increases in this participation rate were parallel for both groups.

Table 4.1.1: Vietnam's labor force avg. 1996-2007 annual 1996 2007 growth rate Population aged 15 and above total 47,620,139 63,305,882 3.0% urban 11,026,793 17,964,868 5.7% rural 36,593,346 45,341,014 2.2% male 22,391,531 30,424,965 3.3% female 25,228,608 32,880,917 2.8% Labor force total 36,082,273 47,144,091 2.8% urban 7,243,053 11,895,757 5.8% rural 28,839,219 35,248,334 2.0% Source: Author’s calculation using Labor Force Survey data Note: Labor force includes employed and unemployed workers

Table 4.1.2: Education structure of working age population 1993 1998 2002 2004 2006 No degree & primary school 49% 65% 51% 46% 44% Lower secondary school 26% 23% 30% 33% 33% Upper secondary school 14% 10% 16% 17% 19% Junior College and above 1.8% 2.5% 3.3% 4.0% 4.2% Source: Author’s calculation using VLSS and VHLSS

Table 4.1.3: Average years of schooling of working age population 1993 1998 2002 2004 2006 All sample 7.43 7.42 7.48 8.13 8.30

Male 7.83 7.85 7.78 8.47 8.61 Female 7.05 7.02 7.18 7.78 7.99

Urban 8.77 8.92 8.96 9.81 9.84 Rural 6.96 6.94 7.00 7.51 7.72

Other minorities 6.1 6.0 4.9 5.7 5.8 Kinh & Chinese 7.6 7.6 7.9 8.5 8.7 Source: Author’s calculation using VLSS and VHLSS Table 4.1.4: proportion of workers with skill (vocational training or college degree and above) 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Total 11.4% 12.5% 12.8% 7.6% 8.2% 7.4% 8.0% 8.4% 9.2% 10.0% Vocational training 9.1% 9.9% 9.7% 4.2% 4.7% 3.8% 3.9% 4.1% 4.2% 4.7% College & above 2.3% 2.7% 3.1% 3.4% 3.4% 3.6% 4.1% 4.4% 4.9% 5.3% Source: Author’s calculation using LFS data

Table 4.1.5: Labor force participation rate 1993 1998 2002 2004 2006 All sample 0.90 0.86 0.83 0.82 0.81

Male 0.91 0.87 0.84 0.84 0.82 Female 0.90 0.85 0.83 0.81 0.80

Rural 0.94 0.91 0.75 0.74 0.74 Urban 0.79 0.75 0.86 0.85 0.83

Other minorities 0.95 0.94 0.90 0.90 0.88 Kinh & Chinese 0.90 0.85 0.82 0.81 0.79 Source: Author’s calculation using VLSS and VHLSS Note: Participation rate = percentage of working-age (15-60) population having employment in the past 12 months; MOLISA calls this employment rate

Table 4.1.6: wage employment participation rate

1993 1998 2002 2004 2006

All sample 0.31 0.34 0.39 0.43 0.44

Male 0.39 0.43 0.49 0.52 0.54

Female 0.23 0.26 0.29 0.33 0.35

Rural 0.27 0.29 0.53 0.56 0.57

Urban 0.46 0.49 0.35 0.39 0.40

Other minorities 0.21 0.25 0.27 0.31 0.34

Kinh & Chinese 0.32 0.36 0.41 0.46 0.47 Source: Author’s calculations using VLSS & VHLSS Note: this is percentage of employed workers having wage employment, which could be primary or secondary job, and could be agricultural or non-agricultural job.

Table 4.1.7: Average hours worked per year 1993 1998 2002 2004 2006 All sample 1729 1928 1841 1828 1884

Male 1771 1921 1862 1836 1912 Female 1691 1935 1821 1820 1855

Rural 1635 1861 2180 2219 2263 Urban 2073 2188 1748 1703 1756 Source: Author’s calculations using VLSS & VHLSS; this is average hours worked in both primary and secondary jobs in the past 12 months Source: VHLSS and VLSS data

3.2. LABOR DEMAND AND EMPLOYMENT STRUCTURE Figures 4.2.1 and 4.2.2 present the changing structure of production and employment in Vietnam. The figures show that there have been no changes in the trend of employment structure documented in earlier studies. As a result, I will keep this section very brief. In particular, the growth in employment and changes in employment structure are not keeping pace with those in production. Vietnam has low arc elasticity of employment (annual average of 0.33 during period 1990-2007). Figure 4.3.3 shows that there is a brief period from 1998 to 2001 during which the elasticity rose up. Because employment growth stayed very stable through the whole study period, this brief rise in elasticity is due to the drop in GDP during the Asian Financial Crisis.

Vietnam has relatively low and stable unemployment rate (between 2-3%) by international standards, as shown in table 4.2.1. The unemployment rate jumped from 1.9% in 1996 to to 2.9% in 1997, when the Asian Financial Crisis started, and stayed high until 2001 when the Vietnamese economy started a new boom period as the regional economy recovered. While this low and stable unemployment rate is an encouraging sign, one would be too hasty to conclude that the rate of employment creation is keeping up well with the rate of growth of labor force. Low unemployment rate is rather typical in low income countries where many workers are self-employed and are too poor to afford unemployment. For such countries, underemployment is usually the real issue. Data on unemployment in table 4.2.2 suggest that this is indeed the case for Vietnam. Also, unemployment is more prevalent in the urban areas than in the rural areas. This is not surprising: rural workers tend to be self-employed; hence underemployment rather than unemployment is more common in the rural areas, as table 4.2.2 demonstrates. In the Socio-Economic Development Plan (SEDP) 2006-2010, the Vietnamese government sets the goal of reducing agricultural sector’s employment to 50% and increasing the share of the insdustrial sector to at least 23-24% and of the service sector to 26-27%, all by 2010. It also sets the goal of keeping the urban unemployment rate below 5%. Comparing the presented data with these targets, we can see that these goals have been or will be achieved soon by the deadline. Despite the current global crisis, the urban unemployment rate in 2008 is 4.6%, still below the 5% target. However, this rate might soar in 2009, for the projected job loss figure for 2009 is 150,000 (MPI 2009). Furthermore, the real challenge is not in generating more jobs and changing the structure of employment; it is to create productive and equitable employment. This will be discussed in the next section. Table 4.2.1: unemployment rate 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 whole country 1.9% 2.9% 2.3% 2.3% 2.3% 2.8% 2.1% 2.3% 2.1% 2.1% n.a. 2.2% urban 5.7% 5.8% 6.6% 6.5% 6.3% 5.4% 5.8% 5.6% 5.4% 5.1% n.a. 4.4% rural 1.0% 2.1% 1.1% 1.1% 1.1% 2.0% 0.9% 1.2% 1.1% 1.1% n.a. 1.4% Source: author’s calculation using LFS data from MOLISA 2006 Note: Unemployment rate = percentage of workers who were not employed in the past 7 days and were looking for work

Table 4.2.2: underemployment rate 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 whole country 10.2% 17.3% 13.1% 10.6% 8.6% 14.4% 13.7% 11.8% 9.2% 8.1% n.a. 4.9% urban 9.2% 9.3% 9.0% 8.1% 6.6% 8.4% 8.6% 7.7% 5.7% 4.5% n.a. 2.1% rural 10.4% 19.5% 14.2% 11.6% 9.1% 16.2% 15.2% 13.1% 10.3% 9.3% n.a. 5.8% Source: MOLISA 2006 Note: Under-employment rate = percentage of workers who would like to work more hours, are able to work more hours, or work less than 8 hours per week

3.3. LABOR PRODUCTIVITY Labor productivity, defined as GDP per worker, has been increasing in Vietnam at a rate of about 5% per year (see table 4.3.1, table 4.3.3 and figure 4.3.1). As table 4.3.2 suggests, the growth rate of Vietnam’s labor productivity is relatively high compared with other ASEAN economies, but certainly behind that of China. Despite a relatively high growth rate, labor productivity is still very low in Vietnam, because it started from a very low base. As of 2006, Vietnam’s labor productivity is only about half of China’s level, and about one third of other ASEAN countries. Thus, Vietnam still has a very large gap to fill regarding its labor productivity. Table 4.3.3 and figures 4.3.1 and 4.3.2 examine labor productivity by sector. It is not surprising to find that the industry sector has the highest labor productivity, followed by the service sector, and that agriculture has the lowest labor productivity (see figure 4.3.1). What is more alarming is the low average growth rate of the service sector’s labor productivity, driven largely by a period of slow or even negative growth from 1995 through 2000 (see figure 4.3.2). Given that this sector makes up 40% of GDP, its slow productivity growth is creating a drag on the country’s aggregate productivity performance. Traditionally, the service sector has less potential for productivity growth. But according to APO (2009), this is changing fast in developed economies, whose service sectors are becoming more intensive in information and communication technology. APO (2009) also documented that this trend is being matched in several Asian countries, including India and the Asian Tigers (except Korea). In this regard, Vietnam is certainly falling behind, for its service sector is still not the driving force behind aggregate productivity growth.

It is also interesting to note that the growth rate of agriculture’s labor productivity has been quite stable over the years (indeed “too” stable), while those of industry and services are more volatile. The latter two sectors had increasing growth rates in the early 1990s, followed by a period of declining growth rates from 1995 through 2000, then started to pick up again since 2000. Note that the decline period started even before the Asian Financial Crisis in 1997-98, but the crisis clearly induced a deep drop. It is of interest to research why the growth of labor productivity in industry and services decelerated in the latter half of the 1990s even before the onset of the crisis.

Improvement in aggregate labor productivity is a result of improvement within each sector (intra-sectoral effect) or of reallocation of labor from lower productivity sector to higher productivity sector (inter-sectoral effect). According to APO (2009), the intra-sectoral effect dominates inter-sectoral effect for most Asian countries, Vietnam included. But the trend varies depending on the period (Nguyen 2008). As table 4.3.4 shows, from 1990 through 2000, productivity growth within sectors is the major contributor to overall productivity growth, contributing more than twice that of sectoral shift. But since 2001, intra-sectoral productivity growth has declined while inter-sectoral productivity growth increased. As a result, from 2001 onward, the effect from sectoral shift takes over, until 2006 when productivity growth within sectors picked up again and the two effects balanced.

The material presented in this section lends further support to Jenkin (2002)’s claim that sluggish employment growth in Vietnam is due to increases in labor productivity. The Vietnamese government has recently emphasized the goal of generating productive and equitable employment. But this goal has an inherent dilemna because of Vietnam’s high growth rate of labor force and low level of labor productivity. For a given increase in output, if labor productivity is increased, then the number of jobs created must be sacrificed. Even as Vietnam’s productivity is growing at a relatively high rate, it will take many years to catch up with other countries in the region, let alone more advanced countries. Moreover, in the initial years as the population’s skill level gradually increases, as more high value-added jobs are created, and as more workers take up these jobs, those who are still left behind in the less productive sector will lose out in a relative sense. In other words, it is likely that wage inequality will rise during the transition period when Vietnam is transforming itself into a high productivity economy, unless proper social policies are in place. In short, in the coming years, the goal of generating productive and equitable employment will continue to be a tremendous challenge.

Table 4.3.1: Labor productivity level

US$ per worker, current price Relative to Singapore (%) Vietnam China ASEAN 8 Vietnam China ASEAN 8

1995 1,753 2,494 5,905 3.6 5.2 12.2

2000 2,457 3,598 6,641 4.3 6.2 11.5

2005 3,300 5,851 8,576 4.4 7.7 11.3

2006 3,617 6,802 9,212 4.5 8.4 11.3 Source: APO (2009)

Table 4.3.2: Labor productivity growth (%)

Vietnam China ASEAN 8 Singapore

1990-1995 5.9 11 5.1 5.5

1995-2000 4.3 7 0.5 2

2000-2006 5.1 8 2.9 2.5 Source: APO (2009)

Table 4.3.3: Labor Productivity Level & Growth by Sector 1990 2007 Average annual growth rate level level (1990-2007) Total 4.5 10.4 5% By sector Agriculture 2.0 3.5 3% Industry 10.0 21.8 5.5% Services 12.3 16.1 1.5% Source: Author’s calculation using data from various Statistical Yearbooks Note: Productivity level is measured in million VND per worker, constant price

Table 4.3.4: inter-sectoral vs. intra-sectoral effect of labor productivity growth inter-sectoral intra-sectoral Year total effect effect 1991-1995 5.7% 4.3% 1.4% 1996-2000 4.8% 3.3% 1.5% 2001 4.3% 0.5% 3.8% 2002 4.5% 1.2% 3.3% 2003 4.5% 1.3% 3.2% 2004 5.2% 2.4% 2.8% 2005 6.0% 3.0% 3.0% 2006 6.1% 3.3% 2.9% Source: Nguyen (2008) table 13 Inter-sectoral effect = effect from productivity growth within sectors Intra-sectoral effect = effect from reallocation of resources from low-productivity to high-productivity sectors

3.4. WAGES Table 4.4.1 presents wage level and growth, using data from the living standard surveys. Three main points stand out. First, average hourly wage has been rising over the years, consistent with the fact that labor productivity has been rising (figure 4.4.1). But this growth has slowed down since 2002. Second, over the entire 1993-2006, wage growth was higher for higher levels of education, indicating a rise in the skill premium (figure 4.4.2). This confirms the trend documented by previous studies which observed the data up to the year 2002. But notice that since 2002, the trend had reversed: from 2002-2006 skill premium tended to decrease lightly rather than increase. This can also been seen in table 4.4.1 (shaded figures) which show that from 2002 to 2006, the lowest education group is the only group experiencing positive wage growth. Third, the year 2002 not only marks a reverse in the trend for the skill premium, but also a reverse in the trend for other measures of wage inequality, including overall wage Gini, female/male wage ratio, rural/urban wage ratio, and minority/Kinh wage ratio (figure 4.4.3). In particular, the Gini coefficient measuring hourly wage inequality increased sharply to 0.45 in 2002, but by 2004 it already dropped back to 1998’s level, 0.36. Regarding gender wage gap, male workers on average earns more than female workers by about 10-30%. But the gap narrowed down significantly from 1993 through 2002, although from 2002 to 2004 it seems to increase slightly again. In constrast to the improving gender wage gap, rural urban wage gap worsened over the years. In the early 1990s, although HCMC had a wage level much higher than the rest of the countries, this was not true for Hanoi and other cities. As a result, in 1993, the rural-urban wage gap was hardly significant. From 1993 to 2002, this gap worsened dramatically. But since 2002 it started to improve again, though not yet to the 1993 level. This is a trend that is mirrored by the ethnic wage gap, which worsened significantly from 1993 until 2002 then turned around and improved.

What can explain the slow down of wage growth and the decrease in the skill premium and wage inequality starting in 2002?19 From the supply side, we saw in section 4.1 that the supply of skilled labor in Vietnam is low but increasing. Could it be that by 2002 the supply of skilled labor started to keep up with the increase for skilled labor, causing the increase in skill premium to slow down? From the demand side, what is the impact of trade and trade liberalization on wage outcomes in Vietnam? Is Vietnam more similar to NIEs in the 1960s and 1970s, or is it more similar to Latin American countries and other recent liberalizers in the 1980s and 1990s? Looking at the 1990s’ data, we can clearly see that this was a the transition period for Vietnam. There were hardly any skill premium in 1993. This cannot exist in a market economy. So as Vietnam transformed itself in the 1990s, skill premium rose rapidly to reflect the new market economy. But starting in 2002, once the transition period is over, the economy start to exhibit NIEs’ phenomenon: decreasing skill premium as the country engages in labor- intensive exports. However, these are all speculations. Further research on both the demand and supply sides of the labor market is needed to provide evidence for these speculations.

19 The 2002 VHLSS is often considered to be of questionable quality in the Vietnam’s research community. It is highly possible that the wage estimate in 2002 is biased. But even if this year is dropped, it has been noted in conversations with other researchers that wage growth has been slower in the 2000s than in the 1990s. Table 4.4.1: Wage levels and inequality Mean hourly wage (thousand VND) % change 1993 1998 2002 2004 2006 93-98 98-02 02-06 93-06 Gini 0.38 0.36 0.45 0.36 0.35 Full sample average 1.83 2.77 4.03 3.93 4.37 51% 46% 8% 139%

Male 2.06 2.98 4.18 4.11 4.59 45% 40% 10% 123% Female 1.48 2.42 3.79 3.63 4.04 63% 57% 6% 172% Female/male wage ratio 0.72 0.81 0.82 0.74 0.63

Urban 1.87 3.50 5.65 5.35 5.36 87% 61% -5% 187% Rural 1.81 2.42 3.19 3.24 3.90 34% 32% 22% 116% Rural/urban wage ratio 0.97 0.69 0.51 0.51 0.52

No degree & primary school 1.97 2.54 2.82 3.00 3.44 29% 11% 22% 75% Lower secondary school 1.75 2.64 3.91 3.44 3.85 51% 48% -2% 120% Upper secondary school 1.69 2.98 4.86 4.91 4.91 76% 63% 1% 190% Junior College and above 2.11 5.48 8.43 7.55 8.47 159% 54% 0% 301%

Other minorities 1.72 2.32 2.18 2.78 3.51 35% -6% 61% 104% Kinh & Chinese 1.84 2.82 4.21 4.06 4.48 54% 49% 7% 144% Minority/Kinh wage ratio 0.94 0.82 0.47 0.57 0.56

Red River Delta 1.45 2.26 2.95 3.15 3.54 56% 30% 20% 143% North East 1.42 2.33 4.07 3.41 4.21 64% 75% 3% 197% North West 2.10 2.51 2.19 2.85 3.65 20% -13% 66% 74% North Central Coast 1.63 2.12 2.88 2.98 3.65 30% 36% 27% 123% South Central Coast 1.50 2.45 3.22 3.28 4.26 64% 31% 32% 184% Central Highland 1.91 2.42 1.60 2.88 4.28 27% -34% 168% 124% South East 2.06 2.80 3.42 3.82 4.35 36% 22% 27% 111% Mekong River Delta 2.14 2.41 3.51 3.19 3.74 13% 46% 6% 75% Urban 1.88 3.50 5.65 5.35 5.36 86% 61% -5% 186%

Note: a) Wage data within each survey year have been adjusted for regional and monthly price differences. Thus, comparison across regions or other sub-categories within each survey can be made directly. However, the data are not yet adjusted for regional and yearly price differences, except that 1993 wage data have been inflated to 1998 prices. So comparisons across survey years should be treated as conservative upper bounds only. b) Wage is the weighted average wage of primary and secondary jobs in the past 12 months. Weight is work hours.

4. CONCLUDING REMARKS This report has accomplished three main tasks. It has reviewed the major sources of labor market data for Vietnam. The main conclusion is that current data come from various sources with varying quality, comprehensiveness, and representativeness. There is a serious lack of data on employment and wage by skill or education, and data on migration and the informal sector. Some improvements are already being made, but it will be several years before the results are realized. In particular, the 2007 Enterprise Survey start to collect employment by skill/education, but wage data by skill/education are still not available. Recent VLSS/VHLSS start to include questions on individuals’ residence status in the household roster module, allowing researchers to distinguish migrants and non-migrants to a certain extent. The new Labor Force Survey (starting 2007) collects information on the informal sector, and is expected to be of higher quality.

The report has also reviewed the literature on the Vietnamese labor market, in particular the demand side, the supply side, and wages. Finally, the report briefly described the current situation of the Vietnamese labor market, updating data to more recent years (2006 or 2007).

The ultimate goal of the report is to indentify gaps in the literature and provide direction for future research. Regarding the demand side, it is recommended that more attention be paid to firm-level analysis in studying factors influencing labor demand. In addition to estimating elasticities that describe firms’ labor demand, we are also interested in studying how these elasticities change over time due to influences such as trade and technology. Regarding the supply side, we still do not have good estimates of labor supply elasticities in general, and elasticities of substitution between workers and work hours in particular. There are also no studies of long-run labor supply for Vietnam yet. Regarding wages, interesting observations about wage growth and wage inequality in Vietnam have been pointed out: in the 1990s, wage growth was rapid, and the skill premium and wage inequality also rose sharply; but since 2002, the trend has reversed. Further research on both the demand and supply sides of the labor market are needed to explain these observations. References

Asian Productivity Organization, APO Productivity Databook 2009, Keio University Press Inc., Tokyo

Arcand, Jean-Louis; d'Hombres, Beatrice; Gyselinck, Paul (2005) “Rendement de l'education et choix des instruments: Application sur donnees vietnamiennes” (Instrument Choice and the Returns to Education: New Evidence from Vietnam. With English summary.) Revue Economique, vol. 56, no. 3, May 2005, pp. 563-72

Belser, Patrick (2000) “Viet Nam on the Road to Labor Intensive Growth” Policy Research Working Paper, No. 2389, The World Bank, Washington DC.

Belser, Patrick and Martin Rama (2001) “State ownership and labor redundancy: estimates based on enterprise- level data from Vietnam” Policy Research Working Paper, No. 2599, The World Bank, Washington DC

Cong, Gian; Doan, Hong Quang, Nguyen, Thi Lan Huong, and Remco H. Oostendorp (2006) “Trade Liberalization, The Gender Wage Gap, and Returns to Education in Vietnam” Working Paper

Edmonds, E., and Pavcnik, N., 2004, Product Market Integration and Household Labor Supply in a Poor Economy: Evidence from Vietnam, Dartmouth College and NBER.

Gallup, John L. (2002) “The Wage Labor Market and Inequality in Viet Nam in the 1990s,” World Bank Policy Research Working Paper, No. 2896

Jenkins, Rhys (2004) “Why has employment not grown more quickly in Vietnam?” Journal of Asia Pacific Economy 9(2) 2004: 191-208

Jenkins, Rhys (2002) “Globalization, FDI and employment in Vietnam” Transnational corporation, Vol. 15(1), pp. 115-142, 2006

Liu, A. Y.C. (2006). ‘Changing wage structure and education in Vietnam 1993-1998: The roles of demand”, Journal of Transition, 14(4): 681-706

Liu, A. Y.C. (2005). ‘Impact of economic reform on wage structure in Viet Nam’, East Asian Economic Perspectives, 16(2): 46-57.

Liu, A. Y.C. (2004a). ‘Changes in gender wage gap in Vietnam’, Journal of Comparative Economics, Vol. 32(3): 586- 596.

Liu, Amy (2004b) “Sectoral gender wage gap in Vietnam” Oxford Development Studies, Vol. 32(4), pp. 225-239

Liu, A. (2004c), 'Flying ducks? Girls' schooling in rural Vietnam - A revisit'. Asian Economics Journal, Vol. 18(3): 293- 318

Liu, A.Y.C. (2001). 'Flying ducks? Girls' schooling in rural Vietnam'. Asian Economics Journal, 15(4): 385-403.

Ministry of Planning and Investment (2009) “Results-based Mid-term Review Report: For Implementation of the Five-Year Socio-Economic Development Plan 2006-2010” Hanoi, Vietnam May 2009

Moock, P. R., Patrinos, H. A., and Venkataraman, M., 1998 “Education and Earnings in a Transition Economy (Vietnam), Policy Research Working Paper, The World Bank, May. Nguyễn, Thị Tuệ Anh, Bùi Thị Phương Liên (2008) “Tăng trưởng năng suất lao động Việt Nam 16 năm 1991-2006: từ góc độ đóng góp của các ngành kinh tế và chuyển dịch cơ cấu ngành” Nhà xuất bản lao động, Hà Nội

Pham, Hung and Reilley Barry (2007a) “The Gender Pay Gap in Vietnam, 1993-2002: A Quintile Regression Approach,” Journal of Asian Economics, Vol. 185(5), pp. 775-806

Pham, Hung and Reilley Barry (2007b) “Ethnic Wage Inequality in Vietnam: Empirical Evidence from 2002” Department of Economics, University of Sussex, November.

Pham, Lan Huong, Bui Quang Tuan, Dinh Hien Minh (2006) “Vietnam: Employment-poverty linkages and policies for pro-poor growth,” Fighting poverty: the development-employment link, ed Rizwanul Islam

Razafindrakoto, Mireille, Francois Roubaud and Le Van Duy (2008) “Measuring the Informal Sector in Vietnam: Situation and Prospects” Statistical Scientific Information: Special Issue on Informal Sector, Institute of Statistical Science, Hanoi

Roubaud, Francois, Phan Ngoc Tram, and Dang Kim Chung (2008) “The Labor Force Survey (LFS) in Vietnam: Assessment of Past Experience and Proposals for a New Survey Design” Report prepared under the UNDP-GSO Project # 40772

Tran, Chung (2008) “Transfers and Labor Market Behavior of the Elderly in Developing Countries: Theory and Evidence from Vietnam” Working Paper, School of Economics, University of New South Wales

Tran, Nhuan Kien and Yoon Heo (2009) “Impacts of trade liberalization on employment in Vietnam: a system generalized method of moments estimation” The Developing Economies 47(1): 81-103