Download The conditional labor demand from Leontief production function can

Structural Transformation, Biased Technological Change, and Labor Demand in Vietnam
Philip Abbott*
[email protected]
Department of Agricultural Economics, Purdue University, Krannert Bldg, 403 W State Street,
West Lafayette, IN 47907-2056, U.S.A.
Ce Wu
[email protected]
Department of Agricultural Economics, Purdue University, Krannert Bldg, 403 W State Street,
West Lafayette, IN 47907-2056, U.S.A.
and
Finn Tarp
[email protected]
Department of Economics, Copenhagen University, Copenhagen, Denmark and
Director of the UNU World Institute for Development Economics Research (UNU-WIDER),
Helsinki, Finland
March 2011
________________________
* Corresponding author. Thanks are due to participants at a CIEM seminar in July, 2010 in
Hanoi, who commented on this research while it was in progress. Responsibility for the content
of this paper remains with the authors.
Structural Transformation, Biased Technological Change, and Labor Demand in Vietnam
Abstract
Labor demand in Vietnam has grown much more slowly than GDP over the last two
decades. Structural transformation has moved labor from lower productivity traditional sectors,
especially agriculture, to higher productivity modern sectors including manufacturing and
services. The Vietnamese labor ministry (MOLISA and ILSSA, 2009) believes this restructuring
is not moving rapidly, has involved too little innovation, and exhibits an overly capital intensive
development strategy. Slow labor demand growth has been seen elsewhere in Asia, and some
believe the “Asian miracle” has been the result of high savings rates and capital accumulation
with relatively little technical progress. We use data for 18 aggregate sectors and the overall
Vietnamese economy to examine the roles played by structural transformation, technical change
and institutional bias toward capital intensive development to explain the difference observed
between labor demand and GDP growth. Decomposition of the factors behind labor demand
growth attributes only 30% of this difference to shifts from low productivity sectors to higher
productivity sectors, while the remaining 70% comes from declining labor use per unit output
that is also found in agriculture. We estimate technical progress using several production
function specifications following both accounting and econometric approaches. Somewhat low
TFP growth emerges using standard choices, consistent with past literature. But estimation using
a Leontief production function is consistent with significant labor augmenting technical progress
and better explains historical economic performance in Vietnam. For the overall economy labor
efficiency is found to grow at 5.8 % per year, while capital efficiency grows at 2.0% per year.
This confirms suspicion that there may have been technical progress behind the Asian miracle,
but one must address a labor augmenting bias in technical change to find it. Rapidly rising
minimum wages were found to contribute to a limited extent to more capital intensive
development because market wage trends have followed well behind the minimum wage and
because low elasticities of substitution mean that somewhat higher wage rental ratios contribute
little to declining labor use per unit output. But investment by state owned enterprises appears to
be much more capital intensive, due to both sectoral and technical choices, and factor use has
changed less rapidly than in the private sector. Restructuring and rapid private sector growth in
1
key sectors mean this factor plays a small role in explaining labor demand evolution. Structural
transformation only partially explains slow labor demand growth in Vietnam. While some of the
difference from GDP growth may be attributed to capital intensive investment by the state, the
majority of the difference is found to be due to technical change when low elasticities of
substitution and labor augmenting bias in technical change are taken into account.
Keywords – labor demand, structural transformation, biased technological change, minimum
wage, state investment, Vietnam
JEL classification: J08, J21, J23, O33, O47, O53
2
Table of Contents
1. Introduction ................................................................................................................................. 4
2. Background ................................................................................................................................. 7
2.1. GDP and employment .......................................................................................................... 7
2.2. Structural transformation...................................................................................................... 9
2.3. Technological change ......................................................................................................... 11
2.4. Institutional biases .............................................................................................................. 12
2.4.1. Minimum wage policy ................................................................................................. 12
2.4.2. State investment ........................................................................................................... 13
3. Labor demand growth decomposition ...................................................................................... 14
4. Technological change and labor demand .................................................................................. 18
4.1. Cobb-Douglas and translogarithmic production function .................................................. 20
4.2. CES production function .................................................................................................... 25
4.3. Leontief production function .............................................................................................. 29
4.4. Model selection on production functions ........................................................................... 32
4.5. Derived conditional labor demand ..................................................................................... 34
4.5.1. Derived conditional labor demand from Cobb-Douglas production function ............. 34
4.5.2. Derived conditional labor demand from CES production function ............................. 36
4.5.3. Derived conditional labor demand from Leontief production function ...................... 37
4.5.4. Model selection for derived conditional labor demand ............................................... 39
5. Roles of policies in labor demand ............................................................................................. 40
5.1. Minimum wage policy ....................................................................................................... 40
5.2. Role of state investment ..................................................................................................... 45
6. Conclusions ............................................................................................................................... 51
3
1. Introduction
For the past two decades Vietnam has experienced rapid GDP growth, reaching over 8%
every year during 2005-2007. Employment growth has been more sluggish, averaging 2.8% from
2005-2007 (GSO, 2010). The large gap between labor demand and economic growth already
existed before the Asian financial crisis. From 1991 to 1995, for example, the average difference
between real GDP growth and employment growth in Vietnam reached 5.6%, which is
comparable to that in the 2000s (MOLISA and ILSSA, 2009).
Vietnam, like many other developing countries, is undergoing structural transformation, a
process which moves labor out of the agricultural sector into industry. Yet the expansion in
industry can hardly generate enough capacity and appropriate positions to absorb all the labor
squeezed out of agriculture due to increasing productivity. Underemployment and
unemployment are still concerns despite dramatic output growth in Vietnam and elsewhere.
The labor study by MOLISA and ILSSA (2009) tackled the issue of stagnant labor
demand in Vietnam, and attributed slow labor demand growth largely to structural
transformation. They raised the concern that development in Vietnam may have been excessively
capital intensive, which is particularly pronounced in investments made by state-owned
enterprises and foreign-invested firms. Moreover, the minimum wage may have fostered capital
intensive growth, in that minimum wages may have been high enough to induce firms to
substitute labor with capital. Investment accounted for 30-40% of GDP in 2000-2008, but
investment efficiency is relatively low, suggested by the high incremental capital-output ratios.
The labor study by MOLISA and ILSSA suggested that the investment structure should be
adjusted so that high value-added industries, and particularly private firms, should gain more
attention in order to create more jobs. According to this view, economic growth in Vietnam is
now due to capital accumulation much more than technological innovation. In 2007, for example,
total factor productivity growth only accounted for 26% of GDP growth, whereas capital
accumulation contributed to more than 60% of GDP growth. In addition to capital intensive
development lowering labor demand, slow restructuring of the economy from rural to urban
areas, from agriculture to manufacturing, and from public to private firms, resulted in slower
labor migration out of sectors where productivity and performance are low.
4
Unlike the unanimity regarding the role of structural transformation in at least partially
explaining sluggish labor demand, the role of technological progress during the course of
Vietnam development is debated in previous studies. Most of the previous studies on Vietnam, to
our knowledge, found low TFP growth rates. For example, MOLISA and ILSSA (2009)
concluded that the contribution of total factor productivity to economic growth is very limited,
and that GDP growth is mainly a result of factor accumulation. Labor demand is also very
insensitive to total factor productivity growth. According to ILSSA (2008), 1% TFP growth
leads to less than a 0.5% decrease in labor use for most economic sectors.
In a more general set of studies on East Asian development, the role of technological
progress during the course of economic growth is much more widely discussed, but those studies
can hardly reach an agreement. Some economists (e.g., Collins and Bosworth, 1996; Kim and
Lau, 1994; Krugman, 1994; Young, 1995) maintain that the East Asian miracle is mainly due to
capital accumulation and that technological progress is still unimportant. Others (e.g., Freeman,
1995; Kruger et al., 2000; Nelson and Pack, 1999; Pack and Page, 1994; Rodrik, 1997; World
Bank, 1993: 56) argue that technological progress, particularly labor-saving1 technological
change, fuelled the extraordinary economic growth that has constituted the “Asian miracle”.
Rodrik (1997) explicitly pointed out that the low TFP growth rates found in most studies are
problematic, because they failed to incorporate biased technological progress in their calculations.
If technological progress saves labor, then labor demand should grow more slowly than output.
Policy interventions undertaken in the developing countries may further accelerate the
process of structural transformation and promote labor-augmenting technological change,
leading to larger gap between GDP growth and labor demand growth. Institutional biases are
another source that may slow down labor demand. Minimum wages in Vietnam have been
actively adjusted over time. If increasing minimum wages drives up market wages and distorts
the wage-rental ratio, then we would expect lower labor demand growth. However, if the
minimum wage has an unimportant effect on market wage determination, then the impact of
increasing the minimum wage on labor demand would be muted. State investment is also a
potential factor responsible for slow labor demand in Vietnam. If state investment in Vietnam is
We will use the terminology “labor-augmenting technical change”, corresponding with the notion that efficiency of
the labor input is increasing. “Labor-saving” is ambiguous, since an improvement in capital efficiency could reduce
labor demand, but only if the elasticity of substitution exceeds one.
1
5
overly capital intensive, as many argued, then the state-owned enterprises have the tendency to
substitute labor with capital, limiting employment growth in state-invested firms. Sectoral
reallocation of the state investment is also critical for labor demand. Sectors receiving more state
investment may not require high labor intensity for production, and hence the benefit to
employment due to expanding production in those expanding sectors may not be strong enough
to compensate the negative employment effect occurred in shrinking sectors.
In this study, we propose three hypotheses to explain slow labor demand in Vietnam:
structural transformation, technological progress, and excessively capital intensive development.
Two institutional biases that facilitate capital intensive development in Vietnam are studied:
minimum wage policy that may distort the wage-rental ratio and state and foreign investment
that may be overly capital intensive.
As a major result, we find that structural transformation can at most explain 30% of the
difference between slow labor demand growth and rapid GDP growth. The remaining 70% rests
on factors that have been less researched in previous studies on labor demand in Vietnam, as
well as in other parts of the world. We identified strong evidence of biased (labor-augmenting)
technological change, and strong evidence supporting the use of a Leontief production function
in describing production activities in Vietnam. Under a Leontief production function, for the
economy as a whole, the capital efficiency growth rate is 2% on average, and the labor efficiency
growth rate is 5.8% on average during the period 2000-2008. Under this assumption, the entire
economy exhibits labor-augmenting technological progress. We also find downward pressure on
labor demand due to overly capital-intensive state investment, but the effect is weaker than the
labor demand effect caused by the restructuring in non-state firms. Statistical results show that
inflation, rather than the minimum wage, is the major driving force determining market wages
for most sectors. The minimum wage policy seems to have larger impact on wages in state
enterprises. The correlation between the state wage and the minimum wage is stronger than that
between the overall market wage and the minimum wage.
Our investigation proceeds as follows: Section 2 describes the current labor demand
situation in Vietnam, and offers preliminary evidence on the three groups of potential sources of
stagnant labor demand, namely, structural transformation, biased technological change, and
policy interventions, such as the minimum wage policy and state investment policy. Section 3
6
reports labor demand growth decomposition results based on data from 2000 to 2008. This
analysis quantitatively identifies the contributions of structural transformation, state investment
bias, biased technological change, and institutional wage bias to the gap between labor demand
growth and economic growth in Vietnam. Section 4 and Section 5 study the impacts of biased
technological change and institutional biases on labor demand in depth. In Section 4, we
calculate productivity growth rates under a Hicks-neutral technical change assumption and under
biased technical change assumption. We explore four production functional specifications:
Cobb-Douglas, translog, CES, and Leontief. We also derive conditional labor demand functions,
and then choose the determine labor demand function of best fit. Section 5 examines linkages
between the market wage, the minimum wage, and inflation. We then study the role of the
minimum wage bias in lowering labor-output coefficients and hence in lowering labor demand.
We also evaluate the impact of state investment on labor demand through effects on production
intensity and sectoral output shares. Section 6 concludes.
2. Background
2.1. GDP and employment
During the past two decades, Vietnam has witnessed dramatic GDP growth. From 2000
to 2008, for example, the average annual GDP growth rate was 7.6%, which is 5% higher than
the average growth rate of the world (MOLISA and ILSSA, 2009). Rapid GDP growth has not
been achieved through uniformly high rates of input accumulation, however. For the two
conventional factors of production, capital and labor, the growth rates are lower, with capital
growing at 6.2% per year and labor growing at 2.2% annually during the same time period (GSO,
2009). The gap between output growth and labor demand growth is much more pronounced than
the gap between output growth and capital growth. Both the large volume of capital inflows from
foreign countries and increasing domestic investment supported by a high saving rate explain the
high rate of capital accumulation. The share of investment in GDP has averaged 30-40% of GDP
during 2000-2008 (MOLISA and ILSSA, 2009). However, labor demand has been steadily
falling behind. Even for the recent booming years prior to the recent financial crisis, despite over
7
8% GDP growth every year during 2005-2007, the corresponding labor demand only grew at 2.8%
annually on average, only around one-third of the GDP growth rate (GSO, 2010).
[Insert Figure 1 about here]
Figure 1 exhibits the evolution of real GDP growth, employment growth, and labor force
growth by sector from 1986 to 2009 (World Bank, 2011). GDP has been growing more than 4.5%
faster than total employment in most time periods from 1986 to 2009, with exceptions in 1998,
1999, 2001 and 2008. Those years correspond with recession in the world markets.
The phenomenon of stagnant labor demand is more pronounced in the agriculture sector
relative to industry and service sectors. Particularly in the 1990s before the Asian financial crisis
in 1997, GDP grew 6% higher than the employment growth in the agricultural sector on average.
The difference between GDP growth and sectoral employment growth for the agricultural sector
has been about 2-3% higher than that for the manufacturing and service sectors.
The large difference between real GDP growth and employment growth in Vietnam
actually is not a new phenomenon in Asia. Table 1 reports the gap between real GDP growth and
employment growth as well as the elasticity of employment with respect to real GDP in other
Asian countries. The difference in Vietnam is comparable with that found in South Korea,
Thailand, and Indonesia when GDP per capita was also low, and this difference is much lower
than that in China. Among those countries, only the Philippines has maintained relatively low
gaps, hence more rapid labor demand growth. The elasticities of labor demand with respect to
real GDP calculated based on the World Development Indicators (World Bank, 2011) also show
that Vietnam is not very different from other East Asian countries. The elasticity of labor
demand with respect to real GDP had been between 0.27 and 0.32 since 1991for Vietnam, which
is comparable to the elasticities of labor demand in other Asian countries2. The difference
between real GDP growth and employment growth reflects real gains in labor productivity,
implying that Vietnam along with many other Asian countries has accomplished output
expansion through improved labor productivity, raising income per capita.
2
The World Bank (1999) and MOLISA and ILSSA (2009) also calculated the elasticity of labor demand with
respect to real GDP. Their elasticities for other Asian countries are too high, not consistent with the statistics from
World Development Indicators (World Bank, 2010).
8
[Insert Table 1 about here]
2.2. Structural transformation
Structural transformation is a process resulting in increasing proportions of employment
and output of the economy in sectors other than traditional sectors such as agriculture. The
economy becomes less agriculturally oriented in a relative sense, although agriculture may
continue to generate important growth linkages to the rest of the economy (Johnston and Mellor,
1961). Based on Lewis (1954) “Dual Sector Model”, the traditional agricultural sector generates
surplus workers due to nearly zero marginal productivity of labor in the sector. So, output in the
traditional sector would not be affected as labor moves out of the traditional sector into modern
sectors. Yet, the addition of labor into the modern sectors would result in higher production
levels per capita. Overall, the entire economy would see an expanding modern sector and a
shrinking traditional sector. This pattern appears in Vietnam. Figure 2 exhibits the evolution of
employment structure in this recent decade. There has been a steady decrease in the share of
agricultural employment, with an almost constant employment share for the fishing industry.
Service and industry sectors have accounted for increasing proportions in total employment,
even though agriculture and fishing sectors use the most labor in the economy (GSO, 2010).
[Insert Figure 2 about here]
The traditional sector usually features a relatively low productivity level, whereas
modern sectors are characterized by much higher productivity levels. Figure 3 presents the huge
productivity differences observed across sectors in Vietnam. Industry has the highest
productivity level, followed by services. The agricultural sector has a much lower productivity
level, which is on average 13% of productivity in industry, and 20% of productivity in the
service sector over the period 1996-2008. For the economy as a whole, the productivity level has
been consistently staying between the productivity level in the agricultural sector and the
productivity level in the service sector. As the economy transforms, due to different productivity
levels by sector, a one unit decrease in the output produced in the traditional sector releases more
labor than required for a one unit output increase in the modern sectors. Table 2 describes
changes in labor productivity by firm type and by disaggregated sectors based on data from
9
Vietnam from 2000 to 2007 (GSO, 2009). It shows that state-invested and private firms have
outperformed foreign-invested firms in improving labor productivity. The manufacturing sector
has consistently increased labor productivity over the years. However, the trend in labor
productivity is very mixed in the service sectors, with large labor productivity enhancement in
the infrastructure service sector3, and rapid decline in labor productivity in several other service
sectors.
[Insert Figure 3 about here]
Often accompanying structural transformation are redundant labor and underemployment
in the traditional sector. Table 3 presents unemployment and underemployment rates in Vietnam
in 2009 (GSO, 2010). Even though the unemployment rate is not very high, staying at 2.9% on
average for the entire economy, the underemployment rate is much higher, reaching 5.6% in the
same year. The underemployment issue is more severe in the rural area, where traditional sectors
are largely located. The underemployment rate averaged 6.5% in the rural area in 2009, which is
almost three times the unemployment rate in the rural area. Unemployment is more severe in the
urban areas relative to rural areas. But even the urban unemployment rate was not high, only
averaging at 4.6% in 2009, a year still under the impact of worldwide economic recession.
[Insert Table 3 about here]
MOLISA and ILSSA (2009) largely attribute stagnant labor demand to structural
transformation. Due to various barriers and improved labor productivity in the modern sectors,
the expanding sectors cannot create enough appropriate positions for labor moving out of the
traditional sector. The structural transformation is an important factor contributing to sluggish
labor demand, but it is not the sole factor responsible for the large gap between output and labor
demand growth. We will try to estimate through labor demand growth decomposition the
contribution of structural transformation and other factors to slow labor demand growth.
3
The infrastructure service sector includes wholesale and retail trade, hotel and restaurant, and transport, storage and
communications. The detailed aggregations for all the sectors can be found in Appendix I.
10
2.3. Technological change
The role of technological change in Vietnam has been debated in previous studies.
According to MOLISA and ILSSA (2009), total factor productivity growth only accounted for
26% of GDP growth, whereas capital accumulation contributed to more than 60% of GDP
growth. The same issue persists in broader literature on East Asian economic development.
While some economists (e.g., Collins and Bosworth, 1996; Kim and Lau, 1994; Krugman, 1994;
Young, 1995) maintain that the East Asian miracle is mainly due to capital accumulation and that
technological progress is limited, others (e.g., Freeman, 1995; Krauger et al., 2000; Nelson and
Pack, 1999; Pack and Page, 1994; Rodrik, 1997; World Bank, 1993: 56) argue that technological
progress, particularly labor-saving technological change, fuelled the extraordinary economic
growth that has constituted the “Asian miracle”. Rodrik (1997) explicitly pointed out that the low
TFP growth rates found in most studies are problematic, because they failed to incorporate
biased technological progress in their calculations.
As shown in Figure 4, in the past decade, real rent to capital by sector increased almost at
the same rate as the real wage, with the market interest rate, as a proxy for the opportunity cost
of capital, changing more irregularly in Vietnam4. As input prices change altogether, the wagerental ratio has stayed almost constant until 2006, as shown in Figure 5. The wage-rental ratio
increased by about 10% from 2000 to 2008, yet the labor-capital ratio decreased more than two
times faster, by about 35% during the same time period. As shown in Figure 5, both labor and
capital productivities improved, indicating the presence of technological progress during the time
period. Moreover, labor productivity grew at a higher rate, suggesting the possibility of laboraugmenting technological change.
[Insert Figure 4 and Figure 5 about here]
Figure 6 to Figure 8 exhibit changes in input prices and input intensities for the
agriculture, manufacturing, and service sectors. For both the agriculture and manufacturing
sectors, wages climbed faster than the return to capital, partially due to higher marginal labor
4
The real rent to capital is calculated as the ratio between real value added to capital and the amount of calculated
capital stocks (GSO, 2009). The real wage is calculated as the ratio between real value added to labor (GSO, 2009)
and employment (GSO, 2010). The formulae for the real rent to capital and the real wage are given in Section 4.2.
Market interest rates are from IMF (2010).
11
productivity. Both labor and capital per unit of output decline, suggesting the existence of
technological progress in the agriculture and manufacturing sectors. Furthermore, production
became more capital-intensive during the course of development, consistent with laboraugmenting technological change or institutional bias. Only for the service sector, returns to
capital grew faster than wages, and factor intensity did not alter significantly during the period
from 2000-2008. Different from the agriculture and manufacturing sectors, the service sector
started with a much higher level of capital intensity in its production, and the transformation in
input intensity is less dramatic.
[Insert Figure 6 to Figure 8 about here]
2.4. Institutional biases
2.4.1. Minimum wage policy
The minimum wage is the fastest growing wage rate in Vietnam. The nominal minimum
wage increased by about 300% from 2000 to 20085, compared with a 60% increase in real state
wages (GSO, 2010) and about a 50% increase in overall real market wage6, as shown in Figure 4.
The minimum wage followed inflation, but grew much faster than the inflation rate,
demonstrated in Figure 9. Minimum wages grew nearly as fast as nominal GDP. In Figure 4, the
positive correlation between the minimum wage and the market wage can be seen, yet whether
the minimum wage or inflation drives the market wage is not clear. If the minimum wage drives
market wage and indirectly distorts the wage-rental ratio, then labor demand may be lower as a
consequence.
[Insert Figure 9 here]
A considerable gap exists between wages for firms with different forms of ownership.
Table 4 presents the nominal average monthly salary and nominal wage growth rates by firm
5
The minimum wage data are summarized on a Wikipedia webpage:
http://translate.google.com/translate?js=y&prev=_t&hl=zh-CN&ie=UTF8&layout=1&eotf=1&u=http://vi.wikipedia.org/wiki/L%25C6%25B0%25C6%25A1ng_t%25E1%25BB%2591i_thi
%25E1%25BB%2583u_t%25E1%25BA%25A1i_Vi%25E1%25BB%2587t_Nam&sl=vi&tl=en
6
The overall market wage is computed as the ratio between value added to labor and total employment.
12
type (GSO, 1998-2006). For all the years listed, foreign-invested firms paid workers the most,
averaging at about 1,300,000 VND7 per month in 2006. State enterprises follow, averaging at
about 1,100,000 VND per month in 2006. The foreign-invested and state enterprises also showed
the largest salary increases during the period from 1998 to 2006, more than an 8.5% annual
growth rate on average. Household enterprises paid workers the least. The average monthly
salary in 2006 was only about 660,000 VND, half of the salary paid by the foreign invested firms.
The annual growth rate of salary was also the lowest, only 2.3% from 1998 to 2006 and much
lower than the average inflation rate during the same time period. We can infer that wages paid
by household enterprises on average cannot keep up with minimum wages.
[Insert Table 4 about here]
To sum up, even though the Vietnamese government has been actively increasing
minimum wages, those wages influence state-owned enterprises and foreign invested firms more
so than private and especially informal firms. Workers in the household firms and some private
firms are not as well paid as those in state-owned enterprises and foreign invested firms. The bias
in wage-rental ratios would therefore affect state-owned enterprises and foreign-invested firms
more so than private firms. Sectoral wage data are consistent with this expectation.
2.4.2. State investment
State investments tend to favor capital-intensive sectors. For example, very capitalintensive natural resource sectors, such as electricity, gas and water supply, and mining and
quarrying, are usually heavily state invested. 77% of total investments received in the electricity,
gas and water supply sector during 2000-2008 are from the state, and 69% of total investments
received in the mining and quarrying sector are from the state. Investments in those two natural
resource sectors account for an average of 26% of total state investment (GSO, 2009). In
comparison, the manufacturing sector does not heavily rely on state investment as the natural
resource sectors. 70% of the investments received in the manufacturing sector are from non-state
7
VND stands for Vietnamese dong.
13
sources. The manufacturing sector is a very large sector, so state investment in the manufacturing
sector accounts for 13% of total state investment over the period 2000-2008.
Overly capital intensive state investment would bring less labor demand than otherwise.
State investments tend to be concentrated in some special sectors. Those sectors are important to
growth of GDP, but are much less important to employment. The overly capital intensive state
investment does not necessarily translate into similar rates of increase in jobs. In state invested
firms, labor accumulation tends to be slower than capital accumulation, which widens the gap
between labor demand growth and output growth. Relative to private firms, state-invested firms
tend to be more capital intensive and to exacerbate the growth of capital intensity in the
production. The technology employed in production requires higher capital intensity, which
limits the benefit to employment from expanding output.
3. Labor demand growth decomposition
Table 6 presents average growth rates of labor and value added by sector during 20002008 (GSO, 2009). The economy as a whole grew at 7.6% annually, yet it only increases
employment by about 2.2% annually. The traditional agricultural sector even had a 0.4%
decrease in labor demand, while output grew by 3.9% on average every year. The impressive 12%
output growth in manufacturing sector was achieved through only a 7% increase in labor use.
The infrastructure service sector (including trade, hotel and restaurant, and transport, storage and
communication sectors), and the education and health sectors also accomplished large increases
in output without fuelling rapid expansion in labor usage. Slow labor demand increases
accompanying rapid output growth are a norm for the majority of the economic sectors and for
the economy as a whole.
[Insert Table 5 about here]
This section provides to a quantitative analysis of the contributions of biased
technological change, structural transformation, and policies to the sluggish labor demand for
each sector and overall. As mentioned earlier, the possible sources related to stagnant labor
14
demand include structural transformation, biased technological change, institutional wage bias,
and state investment bias.
This decomposition starts with a straightforward relationship that links labor demand,
labor efficiency, sectoral output shares, and economic output, shown in Equation (1):
𝐿𝑡 = ∑𝑖 𝐿𝑖𝑡 = ∑𝑖 𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 𝑌𝑡
(1)
where 𝐿𝑡 is total labor demand in time t, which is the sum of sectoral employment 𝐿𝑖𝑡 across the i
sectors. 𝑎𝐿𝑖𝑡 denotes unit labor efficiency coefficient for sector i in time t, which is the ratio
between labor used in sector i and output in sector i. 𝑆𝑖𝑡 is the proportion of output in sector i in
total output in time t, and 𝑌𝑡 is the output of the entire economy in time t.
Differentiating Equation (1) with respect to time gives:
(2)
𝑑𝐿𝑡⁄
𝑑𝑎𝐿𝑖𝑡⁄
𝑑𝑆𝑖𝑡⁄
𝑑𝑌𝑡⁄
𝑑𝑡 = ∑𝑖 𝑆𝑖𝑡 𝑌𝑡 (
𝑑𝑡) + ∑𝑖 𝑎𝐿𝑖𝑡 𝑌𝑡 (
𝑑𝑡) + ∑𝑖 𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 (
𝑑𝑡)
Defining the growth rate of variable X as 𝑔𝑋 = (𝑑𝑋⁄𝑑𝑡) ∙ (1⁄𝑋), we can turn Equation (2) into
growth rates as follows:
𝑑𝐿𝑡
1
𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 𝑌𝑡
∙ 𝐿 = ∑𝑖
𝑔𝑎𝐿𝑖𝑡 + ∑𝑖
𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 𝑌𝑡
𝑔𝑆𝑖𝑡 + ∑𝑖 𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 𝑎
𝑌𝑡
(3)
𝑔𝐿𝑡 =
where 𝑎𝐿𝑡 =
𝐿𝑡
⁄𝑌 = ∑𝑖 𝑎𝐿𝑖𝑡 𝑆𝑖𝑡 , the overall labor-output ratio for the economy.
𝑡
𝑑𝑡
𝑎𝐿𝑡 𝑌𝑡
𝑡
𝑎𝐿𝑡 𝑌𝑡
𝐿𝑡 𝑌𝑡
𝑔𝑌𝑡
Equation (3) can be further simplified into Equation (4):
𝑔𝐿𝑡 = ∑𝑖
(4)
𝑎𝐿𝑖𝑡 𝑆𝑖𝑡
𝑎𝐿𝑡
𝑔𝑎𝐿𝑖𝑡 + ∑𝑖
𝑎𝐿𝑖𝑡 𝑆𝑖𝑡
𝑎𝐿𝑡
𝑔𝑆𝑖𝑡 + 𝑔𝑌𝑡
A discrete approximation to the continuous Equation (4) is:
(5)
𝐿
𝑙𝑛 ( 𝑡⁄𝐿
𝑡−1
) = ∑𝑖
𝑎𝐿𝑖𝑡 𝑆𝑖𝑡
𝑎𝐿𝑡
𝑙𝑛(
𝑎 𝑆𝑖𝑡
𝑎𝐿𝑖𝑡
𝑆
𝑌
⁄𝑎𝐿𝑖𝑡−1 ) + ∑𝑖 𝐿𝑖𝑡
𝑙𝑛 ( 𝑖𝑡⁄𝑆
) + 𝑙𝑛 ( 𝑡⁄𝑌 )
𝑎𝐿𝑡
𝑡−1
𝑖𝑡−1
Equations (4) and (5) incorporate all of the sources that contribute to slow labor demand
identified earlier. As mentioned in Section 2, biased technical change affects labor demand by
altering input efficiency. Minimum wage policy may distort input-output ratios by altering
15
relative input prices, so relative factor intensities, as well. Thus, these two effects are both
incorporated in the changes in the labor-output coefficient 𝑎𝐿𝑖𝑡 . The first term on the right hand
side of Equation (5) therefore measures the contribution of biased technological change and
institutional wage bias to sectoral labor demand. Structural transformation fundamentally
influences sectoral output shares, with traditional sectors shrinking and modern sectors
expanding. State investment policies direct resources to flow into specific sectors, and hence
change sectoral output shares, 𝑆𝑖𝑡 , as well. Thus, the labor demand effects of both structural
transformation and state investment policies are summarized in the changes in sectoral output
shares, 𝑔𝑆𝑖𝑡 . Therefore, the second term in Equation (5) measures the contribution of structural
transformation and state investment bias to sectoral labor demand. In addition, the third term in
Equation (5) measures the contribution of economic output to labor demand. If labor-out
coefficients, 𝑎𝐿𝑖𝑡 , and output shares, 𝑆𝑖𝑡 , remain constant, then we will observe the same growth
rates for output and employment. The first two terms in Equation (5) are weighted by relative
sectoral labor efficiency and sectoral output. Technological progress, structural transformation
and relevant policies associated with sectors that feature low labor efficiency (i.e., high laboroutput coefficients) and large output shares would generate relatively larger impacts on total
labor demand trends.
We assembled annual data over the years 2000-2008 for Vietnam. Both labor demand
data (𝐿𝑖𝑡 ) and sectoral output data are from the GSO Statistical Yearbook (2010). The proportion
terms, 𝑆𝑖𝑡 , and labor-output coefficients, 𝑎𝐿𝑖𝑡, were then computed based on sectoral employment
and output data. They were originally disaggregated at an 18-sector level, and the sector
activities can be found in Appendix I. Since structural transformation mainly focuses on shifts
between aggregate sectors, such as agriculture, manufacturing and services, we aggregated
eighteen economic sectors into nine sectors according to the concordance provided in Appendix I,
mainly eliminating some details in the service sectors. Among the nine economic sectors, there is
an aggregate traditional sector, a manufacturing sector, an energy and natural resources sector,
and six service sectors. We also added a sector titled “GDP” as the tenth sector, representing the
entire economy. The sum of contributions from each of the nine sectors will be the tenth row.
[Insert Table 6 about here]
16
The results of labor demand growth decomposition analysis are presented in Table 68.
The entire economy has an average of 2.2% annual labor demand growth over the time period
from 2000 to 2008, as shown in the first column for Sector 9, 5.4% lower than GDP growth.
Biased technical change and institutional wage bias slow down labor demand by 3.5%, as shown
in the fifth column for Sector 10, while structural transformation plus state investment bias slow
down labor demand by another 1.5%, as shown in the seventh column for Sector 10. Therefore,
for the entire economy, biased technological change and institutional wage bias are responsible
for 70% of the difference between economic output growth and labor demand growth, and
structural transformation and state investment bias are responsible for the other 30% of the
difference. This result suggests that there are other factors beyond structural transformation
important in explaining stagnant labor demand.
The traditional agricultural sector is significantly impacted in the course of development.
The traditional sector experienced a 3.5% decline in output shares per year. The economy
became less reliant on agriculture, forestry and fishing, and more dependent on manufacturing
and service sectors. Even though the traditional sector is usually characterized by low labor
productivity, the agricultural sector in Vietnam witnessed a 4.3% increase in labor productivity
during 2000-2008. The labor efficiency improvement further reduced labor use in this traditional
sector. The energy and natural resource sector is another shrinking sector in the economy,
contributing only slightly to the overall negative impact due to structural transformation and state
investment bias. It is worth noting that the energy and natural resource sector is a heavily state
invested sector. Different from many other sectors, the energy and natural resource sector had a
decrease in labor productivity over the years, but very low employment per unit of output. The
manufacturing sector along with most service sectors became more important in the economy,
suggested by increases in labor and output shares. Even though larger output shares generate a
positive impact of structural transformation and state investment bias on labor demand in those
sectors, biased (labor-augmenting) technical change and the institutional wage bias slow down
labor demand in the manufacturing and the majority of the service sectors.
8
Due to lack of data, we cannot perform the labor demand growth decomposition analysis by firm type. Thus, the
results do not reflect differences across sectors by ownership. We will explore the role state owned enterprises using
very strong assumption in section 5, however.
17
In general, Vietnam experiences labor efficiency improvement and transformation in its
economic structure. As the economy shrinks the shares of its traditional sectors characterized by
low productivity fall, and it expands the modern sectors characterized by high productivity.
However, structural transformation and state investment bias are only responsible for 30% of the
observed sluggish labor demand. Technological change and wage biases explain the remaining
70% of the gap. We now focus on another source of slow labor demand, biased technological
change. The purpose of performing the analyses related to technological change is to identify the
specific importance of technological progress in production and in determining labor demand.
The results will then be used to sort between the contribution of technological progress and the
contribution of institutional wage bias to stagnant labor demand.
4. Technological change and labor demand
Technological change can be classified into two categories: One is Hicks-neutral and the
other is factor-augmenting. Technological change is Hicks-neutral if the mix of inputs in the
production function is not affected, and the production function only differs by scaling of output.
If all the economic activities follow Hicks-neutral technological change, then we would expect
equal productivity growth rates between inputs at fixed prices. In comparison, factor-augmenting
technological change is featured by unequal productivity growth rates between inputs when there
is technological change. The technological change can be labor augmenting if the productivity of
labor grows faster than the productivity of capital, and it can be capital augmenting if the
productivity of capital grows faster than the productivity of labor. If technological progress is
believed to be Hicks neutral, total factor productivity growth rates (TFPG) would be an
appropriate measure of productivity improvement. If biased technological progress prevails, then
input-specific productivity growth rates are needed.
In the following analyses, we experimented with different forms of production functions
in measuring technical progress, namely, Cobb-Douglas (Section 4.1), translog (Section 4.1),
CES (Section 4.2), and Leontief (Section 4.3). These production functions assume various levels
of elasticity of substitution between the two inputs and different ways to incorporate
technological progress. The purpose of experimenting with multiple functional forms is to
18
explore different possibilities of production behavior in Vietnam. Functional forms such as CES
and Leontief are also necessary to capture biased technical change.
For Cobb-Douglas, translog, and CES production functions, we first assumed Hicksneutral technological change and estimated TFPG. TFPG measures the growth rate of total factor
productivity, which is the portion of economic growth that cannot be explained by capital and
labor accumulation and hence is attributed to technological progress. TFPG is thus a measure of
the rate of technological progress, often expressed as an annual percentage growth rate. To
obtain TFPG under a Cobb-Douglas production function, we employed both accounting and
econometric approaches. The accounting approach directly uses observed data on inputs and
output and computes TFPG as the difference between output growth and the portion of output
growth explained by share-weighted input accumulation. The accounting approach is widely
used in previous studies. A less commonly adopted method of finding TFPG is the econometric
approach. Instead of using actual data on input cost shares, the coefficients of a production
function are econometrically estimated. TFPG is the constant in the estimable equation. If CobbDouglas is the right production function, then we would expect that accounting and econometric
approaches should generate similar results. Large differences in the results between the two
methods invite suspicions as to the robustness of this production function specification. However,
past literature rarely compares the results from these two methods. Most of them only adopt the
accounting approach without further evaluating the validity of the results. Often the results from
an accounting approach appear to be sensible, but may hide potential weaknesses associated with
the assumptions of the Cobb-Douglas production function.
As mentioned earlier, previous studies often assume Hicks-neutral technological change
and only compute TFPG. However, critics of those studies point out that lack of consideration of
biased technological change may lead to underestimation of TFP growth (e.g., Rodrik, 1997).
Therefore, after examining TFPG under Hicks-neutral technological change, we then invoked a
more realistic assumption that technological progress may be factor augmenting. Productivity
growth rates of the two inputs, capital and labor, were estimated for CES and Leontief models.
We did not impose the assumption of biased technological progress on Cobb-Douglas production
function, because biased technological change cannot be identified using this functional form
(Diamond and McFadden, 1965). For the Leontief production function, we did not calculate
19
TFPG under Hicks-neutral technological progress assumption. Instead, statistical tests on
whether the technological progress is Hicks neutral can be performed for this functional form
once we obtain the estimated factor-specific productivity growth rates.
The model selection results (Section 4.4) will supply us with some statistical evidence to
assist in the decision on which model can best represent production activities in Vietnam. Since
labor demand is the overriding theme for this study, we derived conditional labor demand
equations based on various production functions, estimated those functions, and applied model
selection statistics to compare the labor demand model (Section 4.5). All the analyses contain
sectoral details.
4.1. Cobb-Douglas and translogarithmic production functions
Due to unitary elasticity of substitution, productivity growth rates assuming biased
technological change are not identifiable in a Cobb-Douglas production function9. Relative
productivity coefficients of the production function are reflected by cost shares, and are fixed
over time. In this case, the only measure of productivity growth is TFPG assuming Hicks-neutral
technological progress. Large TFPG corresponds with rapid productivity enhancement.
For the Cobb-Douglas production function, both accounting and econometric approaches
were employed when estimating TFPG. The procedures to derive sectoral TFPG are described
below.
The mathematical form of Cobb-Douglas production function with two inputs (i.e. capital
and labor) can be expressed as:
(6)
𝛼
𝛼
𝑌𝑖𝑡 = 𝐴𝑖𝑡 𝐾𝑖𝑡 𝑖𝐾 𝐿𝑖𝑡𝑖𝐿
where 𝑌𝑖𝑡 denotes total output in sector i in time t, 𝐴𝑖𝑡 denotes total factor productivity for sector
i in time t, 𝐾𝑖𝑡 denotes capital stock in sector i in time t, and 𝐿𝑖𝑡 denotes labor use in sector i in
time t. If assuming perfect competition and constant returns to scale, then 𝛼𝑖𝐾 equals the cost
9
The inability to identify biased technological progress in Cobb-Douglas is explained in detail in
Diamond and McFadden (1965) and Sato (1970).
20
share of capital for sector i, and 𝛼𝑖𝐿 is the cost share of labor. Since there are only two inputs into
production and constant returns to scale are assumed in Cobb-Douglas, the input cost shares add
to 1. Thus,
𝛼𝑖𝐾 + 𝛼𝑖𝐿 = 1
(7)
The logarithmic form of this production function is as follows:
𝑙𝑛𝑌𝑖𝑡 = 𝑙𝑛𝐴𝑖𝑡 + 𝛼𝑖𝐾 𝑙𝑛𝐾𝑖𝑡 + 𝛼𝑖𝐿 𝑙𝑛𝐿𝑖𝑡
(8)
Therefore, the total factor productivity (TFP) is:
𝑇𝐹𝑃𝑖𝑡 = 𝑙𝑛𝐴𝑖𝑡 = 𝑙𝑛𝑌𝑖𝑡 − 𝛼𝑖𝐾 𝑙𝑛𝐾𝑖𝑡 − 𝛼𝑖𝐿 𝑙𝑛𝐿𝑖𝑡
(9)
The growth rates of TFP can then be derived through the following equivalent formula:
(10)
𝐴𝑖𝑡
⁄𝐴
)
𝑖𝑡−1
𝑌
𝐾
𝐿
= 𝑙𝑛 ( 𝑖𝑡⁄𝑌 ) − 𝛼𝑖𝐾 𝑙𝑛 ( 𝑖𝑡⁄𝐾 ) − 𝛼𝑖𝐿 𝑙𝑛 ( 𝑖𝑡⁄𝐿
)
𝑖𝑡−1
𝑖𝑡−1
𝑖𝑡−1
𝑇𝐹𝑃𝐺𝑖,𝑡−1,𝑡 = 𝑙𝑛 (
For the econometric approach, we can first assume that the productivity growth follows 𝐴𝑖𝑡 =
𝐴𝑖0 𝑒 𝑔𝑖𝑡𝑡 , where 𝑔𝑖𝑡 is the growth rate of TFP. Then Equation (8) can be rewritten as:
𝑙𝑛𝑌𝑖𝑡 = 𝑙𝑛𝐴𝑖0 + 𝑔𝑖𝑡 𝑡 + 𝛼𝑖𝐾 𝑙𝑛𝐾𝑖𝑡 + 𝛼𝑖𝐿 𝑙𝑛𝐿𝑖𝑡
(11)
Thus, the growth rate of TFP 𝑔𝑖𝑡 can be estimated as the constant from the following equation:
(12)
𝑌
𝑙𝑛 ( 𝑖𝑡⁄𝑌
𝑖𝑡−1
𝐾𝑖𝑡
𝐿
⁄𝐾 ) + 𝛼𝑖𝐿 𝑙𝑛 ( 𝑖𝑡⁄𝐿
)
𝑖𝑡−1
𝑖𝑡−1
) = 𝑔𝑖𝑡 + 𝛼𝑖𝐾 𝑙𝑛 (
As mentioned earlier, for the accounting approach, the average cost shares based on
actual data were used to compute TFPG by sector and by year. Equation (10) is the formula to
compute TFPG. For the econometric approach, the cost share parameters are estimated from the
regression equation based on Equation (12). The constant in the regression equation is TFPG.
Data for annual employment are the same as those used in Section 3 (GSO, 2010).
Physical output levels are used for 𝑌𝑖𝑡 . They are derived by dividing sectoral nominal gross
output by a sectoral producer price index. Both nominal gross outputs and producer price indices
are from GSO (2009).
21
In order to obtain capital stocks by sector, we first assume that the incremental capitaloutput ratio (ICOR) in the first year (i.e., year 2000) is equal to the initial capital-output ratio.
Thus, in the first year, multiplying ICOR with output yields capital stocks in that year. In the
following years, capital stocks equal last year’s capital stock net of depreciation plus investment.
The calculation of ICOR is through the following formula:
(13)
𝑇−1−𝑡
∑𝑇−1
𝑡0 𝐼𝑖,𝑡 (1−𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑎𝑛𝑛𝑢𝑎𝑙 𝑑𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒𝑖 )
𝐼𝐶𝑂𝑅𝑖 = 𝑟𝑒𝑎𝑙 𝑔𝑟𝑜𝑠𝑠 𝑜𝑢𝑡𝑝𝑢𝑡
𝑖,𝑇 −𝑟𝑒𝑎𝑙
𝑔𝑟𝑜𝑠𝑠 𝑜𝑢𝑡𝑝𝑢𝑡𝑖,𝑡0 (1−𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑎𝑛𝑛𝑢𝑎𝑙 𝑑𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒𝑖 )𝑇−𝑡0
where 𝑡0 denotes initial time period, and 𝑇 denotes final time period. 𝐼𝑖,𝑡 is real investment in
sector i in time t (Abbott, Wu and Tarp, 2010). Capital stocks in the first year and in the
following years can thus be derived by following Equation (14) and Equation (15), respectively:
(14)
(15)
𝐾𝑖0 = 𝐼𝐶𝑂𝑅𝑖 ∙ 𝑟𝑒𝑎𝑙 𝑔𝑟𝑜𝑠𝑠 𝑜𝑢𝑡𝑝𝑢𝑡𝑖0
𝐾𝑖𝑡 = 𝐾𝑖𝑡−1 (1 − 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑎𝑛𝑛𝑢𝑎𝑙 𝑑𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒𝑖 ) + 𝐼𝑖,𝑡
Investment and real gross output data used to calculate capital stocks are from GSO (2009).
The shares of capital and labor in total factor payments are calculated by using GSO
(2009) data on value added to capital and value added to labor, respectively. Since employment
data are at 18-sector aggregation level, we re-aggregated all of the other GSO data originally at
112-sector aggregation level into 18 sectors by following the concordance in Appendix II. The
time period for data on capital stocks and cost shares are the same for output and employment
data, which is from 2000 to 2008.
[Insert Table 7 about here]
The results from both the accounting approach and econometric approach are
summarized in Table 7. The average annual TFPG’s by sector are reported. As can be seen, the
observed cost shares are very different from the estimated cost share parameters found using the
econometric approach. As a result, the TFPG’s derived from these two approaches vary greatly.
Also, when using the econometric approach, for nine sectors, the cost share estimates do not lie
between zero and one. For 6 out of 19 sectors, one of the cost shares is negative in the estimation
results. The estimated parameters are not consistent with the definition of cost shares. Even
though the results from the accounting approach appear to be sensible, the results from
22
econometric approach provide evidence to reject Cobb-Douglas as the appropriate production
function. In the area of growth economics, the Cobb-Douglas production function has been
widely assumed in TFPG calculations due to its computational simplicity. The majority of the
previous studies on TFPG only adopted the accounting approach without reporting results based
on an econometric approach. The seemingly reasonable accounting results may hide the
invalidity of the assumption of a Cobb-Douglas production function, just as our results indicate.
According to Young (1995), the translogarithmic value added production function is
more flexible than the Cobb-Douglas production function in that translog production function
allows cost shares to change over time and the elasticity of substitution may deviate from unity.
We next assume a translog production function, following Young’s formula to perform TFPG
computations. The translog production function is written in the following form:
(16)
1
𝑌𝑖𝑡 = exp[𝛼𝑖𝑜 + 𝛼𝑖𝐾 𝑙𝑛𝐾𝑖𝑡 + 𝛼𝑖𝐿 𝑙𝑛𝐿𝑖𝑡 + 𝛼𝑖𝑡 𝑙𝑛𝑡 + 2 𝐵𝑖𝐾𝐾 (𝑙𝑛𝐾𝑖𝑡 )2 + 𝐵𝑖𝐾𝐿 (𝑙𝑛𝐾𝑖𝑡 )(𝑙𝑛𝐿𝑖𝑡 )
1
1
+𝐵𝑖𝐾𝑡 (𝑙𝑛𝐾𝑖𝑡 ) ∙ 𝑡 + 2 𝐵𝑖𝐿𝐿 (𝑙𝑛𝐿𝑖𝑡 )2 + 𝐵𝑖𝐿𝑡 (𝑙𝑛𝐿𝑖𝑡 ) ∙ 𝑡 + 2 𝐵𝑖𝑡𝑡 𝑡 2
where K, L, and t denote capital input, labor input, and time. Time is included to permit technical
change in this specification. The parameters of this functional form are subject to the following
constraints assuming constant returns to scale:
(17)
𝛼𝑖𝐾 + 𝛼𝑖𝐿 = 1, 𝐵𝑖𝐾𝐾 + 𝐵𝑖𝐾𝐿 = 𝐵𝑖𝐿𝐿 + 𝐵𝑖𝐾𝐿 = 𝐵𝑖𝐾𝑡 + 𝐵𝑖𝐿𝑡 = 0
Young (1995) gives the following formula to measure TFPG, assuming this translog production
function, without offering specific derivations in his paper10:
(18)
𝑌
𝐾
𝐿
𝑇𝐹𝑃𝐺𝑖,𝑡−1,𝑡 = 𝑙𝑛 ( 𝑖𝑡⁄𝑌 ) − ̅̅̅̅𝑙𝑛
𝛼𝑖𝐾 ( 𝑖𝑡⁄𝐾 ) − ̅̅̅̅
𝛼𝑖𝐿 𝑙𝑛 ( 𝑖𝑡⁄𝐿
)
𝑖𝑡−1
𝑖𝑡−1
𝑖𝑡−1
Equation (18) differs from the estimable Equation (12) in the cost share parameters, ̅̅̅̅
𝛼𝑖𝐾 and ̅̅̅̅.
𝛼𝑖𝐿
Using a non-econometric method, shares of each input in total factor payments are allowed to
vary across years according to observed data, whereas those parameters are fixed over time in the
Cobb-Douglas production function. The shares can be computed as below:
10
Young only footnoted two references for this equation. We are unable to locate the references or to find the
derivations from any other sources.
23
(19)
𝛼𝑖𝐾 = (𝛼𝑖𝐾𝑡−1 + 𝛼𝑖𝐾𝑡 )/2
̅̅̅̅
(20)
𝛼𝑖𝐿 = (𝛼𝑖𝐿𝑡−1 + 𝛼𝑖𝐿𝑡 )/2
̅̅̅̅
where 𝛼𝑖𝐾𝑡 and 𝛼𝑖𝐿𝑡 represent observed shares of capital and labor in total factor payments in
sector i in time period t. ̅̅̅̅
𝛼𝑖𝐾 and ̅̅̅̅
𝛼𝑖𝐿 denote “average” capital and labor cost shares for each year,
which are averages over the current year and the prior year. Only the accounting approach using
actual cost share data is applied in calculating TFPG, because the translog production function
cannot be econometrically estimated due to too few observations and hence lack of degrees of
freedom.
Table 8 presents the results for this approach. If the economic activities obey a translog
production function and the technological progress is Hicks neutral, then the entire economy
would have an average 4% annual productivity growth for the period from 2000 to 2008. Only
four sectors, namely mining and quarrying (#3), electricity, gas and water supply (#5), real estate
(#12), and activities of Party (#17), experience decreases in total factor productivity on average
over this time period. The rates of decline exceed 1% for the latter two sectors. Service sectors
tend to have relatively high productivity growth rates, and several of the service sectors show
more than 6% productivity growth. The productivity in the manufacturing sector grows at 3%
annually on average, 2% lower than the agriculture and forestry sector.
As shown in the second column of Table 8, TFP growth rates derived from a translog
production function are almost identical to those based on a Cobb-Douglas production function
using the accounting approach, indicating that translog production does not offer a significant
improvement over Cobb-Douglas production function in estimating TFPG. The input cost shares
in Vietnam do not vary greatly over time for most economic sectors. These results calls into
question the adequacy of a translog production function, if it is parameterized as in Young’s
method.
[Insert Table 8 about here]
24
4.2. CES production function
The CES production function has been a challenge for researchers studying growth
accounting. Diamond and McFadden (1965) proposed the “Impossibility Theorem”, stating that
there is no way of identifying the elasticity of substitution if technological progress is biased. For
the following years, the CES function has been consistently avoided in most work on growth
accounting, with some exceptions of attempts to challenge the impossibility. Sato (1970) and
recently Cantore et al. (2009) touched on this issue. Even though neither Sato nor Cantore et al.
fundamentally solved the problem, these two studies along with David and van de Klundert
(1965) are illuminating and lead us to an approach that may finally make the impossibility
possible. The following section reports a method of estimating a CES production function with
biased technological change, followed by an application of this method to the Vietnam market.
David and van de Klundert (1965) proposed the following form of the CES production
function that incorporates biased technological change:
𝑌𝑖𝑡 = [(𝐸𝐾𝑖𝑡 𝐾𝑖𝑡
(21)
)(𝜎𝑖 −1)/𝜎𝑖
+ (𝐸𝐿𝑖𝑡 𝐿𝑖𝑡
)(𝜎𝑖 −1)/𝜎𝑖
]
𝜎𝑖
⁄(𝜎 −1)
𝑖
where 𝜎𝑖 is the elasticity of substitution for sector i. Following the previous notation convention,
𝑌𝑖𝑡 , 𝐾𝑖𝑡 , 𝐿𝑖𝑡 denote output, capital stock and labor for sector i in time t, respectively. 𝐸𝐾𝑖𝑡 and 𝐸𝐿𝑖𝑡
represent the levels of efficiency of the conventional inputs of labor and capital for sector i in
time t, respectively. We adopted the commonly used functional forms for factor-specific
technological progress, e.g., 𝐸𝐾𝑖𝑡 = 𝐸𝐾𝑖0 𝑒 𝜆𝑘𝑡 and 𝐸𝐿𝑖𝑡 = 𝐸𝐿𝑖0 𝑒 𝜆𝐿𝑡 where 𝜆𝑘 and 𝜆𝐿 represent the
growth rate in technological progress associated with capital and labor, respectively; 𝑡 denotes
time; and 𝐸𝐾𝑖0 and 𝐸𝐿𝑖0 are the base value for factor efficiency coefficients. The CES production
function nests Cobb-Douglas when 𝜎𝑖 = 1; and the Leontief production function when 𝜎𝑖 = 0.
For CES production function, we estimate the relevant equation under both Hicks-neutral and
biased technical change assumptions.
To circumvent potential problems associated with non-linear estimation, an estimable
equation in a linear format can be derived. First differentiating Equation (21) partially with
respect L yields:
25
𝑚𝐿𝑖𝑡
(22)
(𝜎 −1)⁄
𝜎𝑖
𝑖
𝜕𝑌
= 𝑖𝑡⁄𝜕𝐿 = 𝐸𝐿𝑖𝑡
𝑖𝑡
𝑌𝑖𝑡
1⁄
𝜎𝑖
(𝐿 )
𝑖𝑡
If perfect competition in all the markets is assumed, then dividing the marginal product by real
relative wage 𝑤𝑖𝑡 =
𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑤𝑎𝑔𝑒𝑖𝑡
⁄𝑃 yields:
𝑖𝑡
𝑚𝐿𝑖𝑡
⁄𝑤𝑖𝑡 = 1 =
(23)
where 𝛼𝐿𝑖𝑡 =
(𝜎𝑖 −1)⁄
𝜎𝑖
𝐸𝐿𝑖𝑡
𝑌𝑖𝑡
1⁄
𝜎𝑖
(𝐿 )
(𝜎𝑖 −1)⁄
(1−𝜎𝑖)
⁄
−1⁄
𝜎𝑖
⁄𝜎
𝜎𝑖 𝑤
𝑖
=
𝐸
𝛼
𝐿𝑖𝑡
𝑖𝑡
𝑤𝑖𝑡
𝐿𝑖𝑡
𝑖𝑡
𝑤𝑖𝑡 𝐿𝑖𝑡
⁄𝑌 , which is the share of labor in total factor payments. Rearranging
𝑖𝑡
Equation (23) gives:
𝜎 −1
1−𝜎𝑖
𝛼𝐿𝑖𝑡 = 𝐸𝐿𝑖𝑡𝑖 𝑤𝑖𝑡
(24)
1−𝜎𝑖
= (𝐸𝐿𝑖0 𝑒 𝜆𝑖𝐿𝑡 )𝜎𝑖 −1 𝑤𝑖𝑡
Similarly, the cost share equation for capital can be written as below:
𝜎 −1 1−𝜎𝑖
𝑖
𝛼𝐾𝑖𝑡 = 𝐸𝐾𝑖𝑡
𝑟𝑖𝑡
(25)
where 𝑟𝑖𝑡 =
1−𝜎𝑖
= (𝐸𝐾𝑖0 𝑒 𝜆𝑖𝐾 𝑡 )𝜎𝑖 −1 𝑟𝑖𝑡
𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑟𝑒𝑡𝑢𝑟𝑛 𝑡𝑜 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑡
⁄𝑃 denotes real rate of return to capital for
𝑖𝑡
sector i in time t, and 𝛼𝐾𝑖𝑡 =
𝑟𝑖𝑡 𝐾𝑖𝑡
⁄𝑌 . Dividing Equation (25) by Equation (24) leads to the
𝑖𝑡
expression for the capital-labor ratio,
𝜎𝑖 −1
𝐾𝑖𝑡
𝐸
𝑟
⁄𝐿 = ( 𝐾𝑖0⁄𝐸 )
( 𝑖𝑡⁄𝑤𝑖𝑡 )−𝜎𝑖 𝑒 (𝜎𝑖 −1)(𝜆𝑖𝐾 −𝜆𝑖𝐿)𝑡
𝑖𝑡
𝐿𝑖0
(26)
An estimable equation is obtained by taking the logarithmic form of Equation (26),
(27)
ln (
𝐾𝑖𝑡
𝐸
𝑟
⁄𝐿 ) =(𝜎𝑖 − 1)𝑙𝑛 ( 𝐾𝑖0⁄𝐸 ) − 𝜎𝑖 ln( 𝑖𝑡⁄𝑤𝑖𝑡 ) + (𝜎𝑖 − 1)(𝜆𝐾 − 𝜆𝐿 )𝑡
𝑖𝑡
𝐿𝑖0
𝑟
In this equation, the two independent variables are relative input price ( 𝑖𝑡⁄𝑤𝑖𝑡 ), and time 𝑡. The
coefficient on the time trend term contains the information on the biases of technological
progress. It is worth noting that if the elasticity of substitution 𝜎𝑖 is equal to one, as in the CobbDouglas case, then the time term associated with biased technological progress drops out, and
biased technical change is not identifiable in this case. If the technological progress is assumed to
26
be Hicks neutral, then the estimation equation is reduced to the following form, and again the
time trend term drops out:
ln (
(28)
𝐾𝑖𝑡
𝐸
𝑟
⁄𝐿 ) =(𝜎𝑖 − 1)𝑙𝑛 ( 𝐾𝑖0⁄𝐸 ) − 𝜎𝑖 ln( 𝑖𝑡⁄𝑤𝑖𝑡 )
𝑖𝑡
𝐿𝑖0
Technological change is labor augmenting if 𝜆𝐾 − 𝜆𝐿 > 0, and is capital augmenting if 𝜆𝐾 −
𝜆𝐿 < 0. When estimating Equation 18 and 18a, we will directly obtain the estimate for 𝜎𝑖 . With
the estimates for (𝜎𝑖 − 1)(𝜆𝐾 − 𝜆𝐿 ) available in Equation 18, we will then be able to calculate
the relative input efficiency coefficient (𝜆𝐾 − 𝜆𝐿 ). Testing whether or not the coefficient on the
time trend term is significantly different from zero gives us further information on whether or not
the technical progress is biased. Furthermore, with the elasticity of substitution available, the
individual input efficiency coefficients can be calculated by adopting Equations 9 and 10 derived
in Sato (1970). Specifically, the efficiency coefficient for capital 𝜆𝐾 is:
𝑟̇
𝑦̇
(𝜎𝑖 𝑟𝑖𝑡 − 𝑦𝑖𝑡)
𝑖𝑡
𝑖𝑡 ⁄
𝜆𝐾 =
(𝜎𝑖 − 1)
(29)
where 𝑦𝑖𝑡 =
𝑌𝑖𝑡
⁄𝐾 .
𝑖𝑡
In a similar fashion, we can write the efficiency coefficient for labor 𝜆𝐿 as:
𝑤̇
𝑧̇
(𝜎𝑖 𝑤𝑖𝑡 − 𝑧𝑖𝑡)
𝑖𝑡
𝑖𝑡 ⁄
𝜆𝐿 =
(𝜎𝑖 − 1)
(30)
where 𝑧𝑖𝑡 =
𝑌𝑖𝑡
⁄𝐿 .
𝑖𝑡
In estimating the derived capital-labor equation, we explored three variants. We firstly
directly estimated Equation (27) and calculated the efficiency coefficients, and estimated
Equation (28) assuming Hicks-neutral technological change. Then, we regressed capital-labor
ratios on lagged relative input price, instead of contemporaneous relative input price, to reflect
the possibility of making investment and employment decisions before contemporaneous inputs
prices are known and assuming a naïve form of expectations. In order to solve the potential
endogeneity problem related to the relative input price, we applied an instrument variable
estimation approach with population, state wage and market interest rate as instruments for the
27
relative input price term. In addition, we tried various other combinations of instruments.
However, none of the modified models give qualitatively and quantitatively different results than
a direct estimation of the derived equation. Therefore, we only report the estimation results based
on the direct estimation. Both Hicks-neutral [Equation (28)] and biased technological change
[Equation (27)] were considered.
Data for capital stocks, labor, and output are the same as described in the previous
sections. The input price data are calculated based on the GSO data. The formulae for real rate of
return to capital and real relative wage are shown in Equations (31) and (32) below:
(31)
(32)
𝑣𝑎𝑙𝑢𝑒 𝑎𝑑𝑑𝑒𝑑 𝑡𝑜 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡
⁄(𝐾 ∙ 𝑃 )
𝑖𝑡
𝑖𝑡
𝑣𝑎𝑙𝑢𝑒 𝑎𝑑𝑑𝑒𝑑 𝑡𝑜 𝑙𝑎𝑏𝑜𝑟𝑖𝑡
𝑤𝑖𝑡 =
⁄(𝐿 ∙ 𝑃 )
𝑖𝑡
𝑖𝑡
𝑟𝑖𝑡 =
where 𝑃𝑖𝑡 denotes price index for sector i in time t.
Data on valued added to inputs and sectoral price indices are from GSO (2009). The time
period for all the data is from 2000 to 2008.
[Insert Table 9 about here]
Table 9 reports the estimation results based on a direct estimation of Equation (27) and
Equation (28), and subsequent calculations of input efficiency coefficients using Equations (29)
and (30). The results indicate that in the absence of biased technological change, input efficiency
(i.e., TFP) increases by 4.5% annually for the entire economy from 2000 to 2008. The elasticity
of substitution is 1.54, higher than unity. If biased technological progress is allowed, then
efficiency would increase by 4.9% annually for capital, and increase by 10.4% annually for labor.
The elasticity of substitution is 0.36 for aggregated results on the entire economy in this scenario.
The economy features labor-augmenting technological change according to the
estimation results, evidenced by a significant difference between labor efficiency growth and
capital efficiency growth. The sectoral details demonstrate that when Hicks-neutral technological
progress is assumed, several sectors have dramatic decreases in efficiency use. When biased
technological progress is allowed, many sectors have more than a 10% increase in the rates of
input productivity growth, and three of them have huge reductions in the efficiency of input use.
28
In addition, four sectors have insignificant coefficient on the relative price term, suggesting that
the elasticity of substitution is very close to zero. Five of the sectors have significant coefficient
on the time trend term, but the elasticity of substitution is less than 0.5. Those sectors with
insignificant or small elasticities of substitution may favor a Leontief function as their production
specification.
The overly high input productivity growth rates and large efficiency reductions invite
suspicion. Moreover, the alternative models with lagged input prices and with instruments on
relative input price generate somewhat different regression results. These estimation results are
consistent with the changes in output and input prices exhibited in Figure 5 to Figure 8. Wagerental ratios change much more slowly than labor-capital ratios, hinting at relatively low
elasticities of substitution. The unsatisfactory results based on CES production function
motivated us to continue investigating alternative functional forms.
4.3. Leontief production function
The Leontief production function can be considered as a special case of CES with an
elasticity of substitution equal to zero, which implies that the factors of production are used in
fixed proportions (Leontief, 1941). In this study, since we use actual employment and capital
stock data, factors of production in the model are assumed to be fully used. The mathematical
form of the Leontief production function used in this study is presented below:
(33)
𝑌𝑖𝑡 =
𝐾𝑖𝑡⁄
𝐿𝑖𝑡
𝑎𝑖𝑡 = ⁄𝑏𝑖𝑡
where 𝑎𝑖𝑡 and 𝑏𝑖𝑡 are input-output coefficients for capital and labor, respectively. These measure
how many units of capital and labor are needed to produce 𝑌𝑖𝑡 units of output using the technique
in time t. We also assume that there may be technological progress over time in sectoral
production. The changes in input-output coefficients follow these two functions:
(34)
𝑎𝑖𝑡 = 𝑎𝑖0 𝑒 𝜃𝐾𝑡 and
𝑏𝑖𝑡 = 𝑏𝑖0 𝑒 𝜃𝐿𝑡
where 𝑎𝑖0 and 𝑏𝑖0 are the base values for capital-output coefficient and labor-output coefficient.
𝜃𝐾 and 𝜃𝐿 denote rates of increases in the capital-output coefficient and labor-output coefficient,
29
respectively. Thus, negatives of those two coefficients are the growth rates of capital efficiency
and of labor efficiency, which are equivalent to the estimates of 𝜆𝐾 and 𝜆𝐿 in the CES case. If we
are able to obtain estimates of 𝜃𝐾 and 𝜃𝐿 , then we can use statistical tests to examine if there is
labor-augmenting technological change.
In order to estimate 𝜃𝐾 and 𝜃𝐿 , we will first compute 𝑎𝑖𝑡 and 𝑏𝑖𝑡 by rearranging the terms
in Equation (33) into the following form:
𝑎𝑖𝑡 =
(35)
𝐾𝑖𝑡
𝐿
⁄𝑌 and 𝑏𝑖𝑡 = 𝑖𝑡⁄𝑌
𝑖𝑡
𝑖𝑡
Taking the logarithmic form of Equation (35) immediately leads to estimation equations for the
efficiency parameters 𝜃𝐾 and 𝜃𝐿 :
(36)
𝑙𝑛𝑎𝑖𝑡 = 𝑙𝑛𝑎𝑖0 + 𝜃𝐾 𝑡
(37)
𝑙𝑛𝑏𝑖𝑡 = 𝑙𝑛𝑏𝑖0 + 𝜃𝐿 𝑡
The hypotheses for biased technological change are:
𝐻0 : −𝜃𝐾 ≥ −𝜃𝐿 𝑐𝑎𝑝𝑖𝑡𝑎𝑙 𝑠𝑎𝑣𝑖𝑛𝑔 𝑡𝑒𝑐ℎ𝑛𝑖𝑐𝑎𝑙 𝑐ℎ𝑎𝑛𝑔𝑒
𝐻𝑎 : −𝜃𝐾 < −𝜃𝐿 𝑙𝑎𝑏𝑜𝑟 𝑠𝑎𝑣𝑖𝑛𝑔 𝑡𝑒𝑐ℎ𝑛𝑖𝑐𝑎𝑙 𝑐ℎ𝑎𝑛𝑔𝑒
The hypotheses state that if the efficiency of capital grows faster than the efficiency of labor,
then the new technology is capital augmenting. If the reversed relationship is true, then the new
technology saves labor. A t test is built according to the following equation:
𝑡=
(38)
(𝜃𝐾 − 𝜃𝐿 )
⁄
√
2
𝑆𝐾
𝑁𝐾
𝑆𝐿2
+𝑁
−1
𝐿 −1
Since the time period for the data is from 2000 to 2008, the number of observation is 9, i.e. 𝑁𝐾 =
𝑁𝐿 = 9. 𝑆𝐾 and 𝑆𝐿 stand for the standard errors of the estimated efficiency parameters 𝜃𝐾 and 𝜃𝐿 ,
respectively.
Data needed for this estimation include capital stocks, employment, and physical output
by sector and by year. All the data are the same as those used in the previous analyses.
[Insert Table 10 about here]
30
Table 10 presents the estimation results for the efficiency parameters under a Leontief
production function, and the testing results regarding which factor is saved by the new
technology. The entire economy, i.e. sector 19, uses 5 units of capital and 7 units of labor to
produce 1 unit of output on average. The economy overall experiences 2.0% capital efficiency
growth and 5.8% labor efficiency growth annually. Both capital and labor efficiency parameters
are significantly different from zero at a 1% level, indicating that there is technological progress
in the economy. Also, the null hypothesis of capital-saving technological change is rejected for
the entire economy, and technological change in the whole economy favors labor. The
agriculture and forestry sector (#1) has relatively high capital efficiency growth at 4.1% annually,
yet the labor efficiency grows even faster at 5.7%. The differences in growth between capital
efficiency and labor efficiency are more pronounced for the fishing (#2) and manufacturing (#4)
sectors. Their capital efficiency growth rates are 1.5% for the fishing sector and 1.6% for the
manufacturing sector, less than 1/3 of their corresponding labor efficiency growth. The mining
and quarrying sector (#3) does not have significant input efficiency improvement, and neither
does the electricity, gas and water supply (#5) sector, both of which are typically heavily stateinvested sectors. Those sectors may have limited adoption levels of new techniques targeting
enhancing production efficiency. The service sectors overall tend to have larger input efficiency
improvements than traditional and industrial sectors. Factor efficiency in all the service sectors
significantly grows over this time period, except for the capital efficiency in the real estate sector
(#12), the labor efficiency in the financial intermediation sector (#10) and in the activities of
Party (#17). Most service sectors exhibits labor-augmenting technological progress, with the
exceptions of construction (#6), financial intermediation (#10), real estate (#12), public
administration and defense (#13) sectors, and activities of Party (#17). Among the five sectors
with capital-saving technological progress, two of them do not have statistically significant
improvement in labor efficiency.
Compared to the estimates from CES production function, sectoral estimations based on
the Leontief production function are much more reasonable, with most annual input efficiency
growth rates locating within the range between 0.4% and 10%. The economy-wide input
efficiency growth rates are also comparable to findings from other studies [see Leon-Ledesma et
al. (2009) for a summary]. The Leontief production function seems to be the most reasonable
among all the production functions we have studied so far.
31
4.4. Model selection on production functions
After applying all the production function specifications to the production data in
Vietnam, we now select the best model among Cobb-Douglas, CES and Leontief production
functions. Even though we estimated total factor productivity growth rates and biases based on
the translog production function, as mentioned before, the number of parameters to be estimated
in the translog production function is more than the number of observations. Thus, the estimation
of a translog production function is not feasible. Therefore, we exclude the translog production
function at this stage of model selection. In addition, since the results from translog production
function are also very similar to those from Cobb-Douglas production function using the
accounting approach, we can infer that the selection results should be very similar to CobbDouglas under the accounting approach. The three production functional forms examined in this
section represent different levels of elasticities of substitution, featuring constant unitary
elasticity of substitution in Cobb-Douglas, perfect complements with zero elasticity of
substitution between factors in Leontief, and an indeterminate constant as the elasticity of
substitution in the CES production function.
We adopted the root mean squared error (RMSE) method for model selection. For CobbDouglas production function, we reported the RMSE from the accounting approach as well as
the econometric approach. For the accounting approach, we used the actual average cost shares
to estimate output, and then compute the RMSE through Equation 32 below. For the econometric
approach, we directly obtained estimated output in logarithmic form from Equation (8)
constrained by Equation (7). We then removed the logarithm by exponentiating the estimates and
computed the RMSE through Equation (39).
(39)
2
̂
̂
𝑅𝑀𝑆𝐸[𝑌
𝑖𝑡 ] = √𝐸[(𝑌𝑖𝑡 − 𝑌𝑖𝑡 ) ]
̂
where 𝑌
𝑖𝑡 denotes estimated output in sector i in time t, and 𝑌𝑖𝑡 denotes actual output in sector i in
time t. Hence, we ask how well each functional form and estimation method fit observed output
data.
32
For CES and Leontief production functions, we obtained the RMSE indirectly. For CES
production function, we first inserted the estimated parameters derived in Section 4.2 into the
production function to get estimated output by sector, and then computed the RMSE through
Equation (39) above. For the Leontief production function, we inserted the estimated efficiency
coefficients derived in Subsection 4.3.2 into the Leontief production function to obtain estimated
output by sector over time, and then computed the RMSE through Equation (39) listed above. It
is worth noting that we assumed fully employment for both capital and labor, and hence different
RMSE results were derived based capital and labor usages for the Leontief production function.
We further divided the RMSE by actual average output to show the relative estimation errors.
[Insert Table 11 about here]
Table 11 presents the model selection results. For all the sectors, including sector 19
representing the entire economy, results favor the Leontief production function as the best
description of the production activities. For the economy as a whole, the root mean squared error
is 1% of actual output, compared with 11% for the econometric approach and 43% for the
accounting approach using a Cobb-Douglas production function, and with 7% for the CES
production function. These patterns are present for the other sectors as well. CES nests Leontief,
but the estimable equation for a CES production function is not a relaxed form of the Leontief
production function. This is because the estimable equations for CES production function are
derived. We can directly estimate Leontief production function, but cannot directly estimate a
CES production function. Instead, in order to make the CES production function estimable, we
need to impose some assumptions to derive simpler forms that are estimable. The predictability
for output for CES production function may be weakened as we take these additional steps to
derive those estimable equations. Leontief is better than the CES, and is also obviously superior
to the Cobb-Douglas production function in explaining output. For the Cobb-Douglas production
function, the econometric approach is better than the accounting approach for the reason that the
econometric approach can find the production function parameters that fit the data better than
actual cost shares. However, as noticed in Section 4.1.1, many of the cost share parameters from
the econometric estimations do not lie between zero and one, which is not sensible for share
terms under assumption of Cobb-Douglas. For the Leontief production function, the levels of
estimation accuracy between capital and labor equations are very close.
33
From these analyses, we find that there is biased technological change, specifically laboraugmenting technological change in the economy, which may in turn contribute to slow labor
demand. Among all the production functions, the Leontief production function strongly
outperforms the others, which implies low substitutability between factors of production in
Vietnam.
Since labor demand is the overriding theme of this study, in the next section, we will shift
our focus from production to labor demand. We will examine the estimation results from derived
conditional labor demand equations, and then integrate the model selection results for the
production function with the model selection results from labor demand equations to draw the
final conclusions on which model we will use to examine institutional biases.
4.5. Derived conditional labor demand
4.5.1. Derived conditional labor demand from Cobb-Douglas production function
The conditional labor demand function sets up a relationship between labor demand,
output and input prices. Conditional labor demand functions can be derived based on the
production function. Below, we present the derivations of the conditional labor demand equation
from the Cobb-Douglas production function.
To start with, the Cobb-Douglas production function incorporating the constraint on cost
share terms is rewritten below:
𝛼
1−𝛼𝑖𝐾
𝑌𝑖𝑡 = 𝐴𝑖𝑡 𝐾𝑖𝑡 𝑖𝐾 𝐿𝑖𝑡
(40)
First differentiating Equation (40) partially with respect to each input factors leads to:
(41)
(42)
𝑚𝐾𝑖𝑡 =
𝑚𝐿𝑖𝑡 =
𝜕𝑌𝑖𝑡
𝛼 −1 1−𝛼
⁄𝜕𝐾 = 𝐴𝑖𝑡 ∙ 𝛼𝑖𝐾 ∙ 𝐾𝑖𝑡 𝑖𝐾 𝐿𝑖𝑡 𝑖𝐾
𝑖𝑡
𝜕𝑌𝑖𝑡
𝛼
−𝛼
⁄𝜕𝐿 = 𝐴𝑖𝑡 ∙ (1 − 𝛼𝑖𝐾 ) ∙ 𝐾𝑖𝑡 𝑖𝐾 𝐿𝑖𝑡 𝑖𝐾
𝑖𝑡
The optimality condition for production states that the marginal rate of technological substitution
between two inputs equals the relative input price. Therefore,
34
𝑚𝐾𝑖𝑡
𝛼
𝐿
𝑟
⁄𝑚𝐿𝑖𝑡 = 𝑖𝐾⁄(1 − 𝛼 ) ∙ 𝑖𝑡⁄𝐾 = 𝑖𝑡⁄𝑤𝑖𝑡
𝑖𝑡
𝑖𝐾
(43)
Combining Equations (43) and (40) yields
𝑌𝑖𝑡 = 𝐴𝑖𝑡 ∙ [
(44)
𝛼𝑖𝐾
𝛼𝑖𝐾
𝑤
⁄(1 − 𝛼 ) ∙ 𝑖𝑡⁄𝑟𝑖𝑡 ] ∙ 𝐿𝑖𝑡
𝑖𝐾
Thus, the conditional labor demand is as follows:
−𝛼𝑖𝐾
𝑌
𝛼
𝑤
𝐿𝑖𝑡 = ( 𝑖𝑡⁄𝐴 ) [ 𝑖𝐾⁄(1 − 𝛼 ) ∙ 𝑖𝑡⁄𝑟𝑖𝑡 ]
𝑖𝑡
𝑖𝐾
(45)
Taking the logarithmic form of Equation (45) gives:
(1 − 𝛼𝑖𝐾 )⁄
𝑟
𝑙𝑛𝐿𝑖𝑡 = 𝑙𝑛𝑌𝑖𝑡 − 𝑙𝑛𝐴𝑖𝑡 + 𝛼𝑖𝐾 ln( 𝑖𝑡⁄𝑤𝑖𝑡 ) + 𝛼𝑖𝐾 ln [
𝛼𝑖𝐾 ]
(46)
(1 − 𝛼𝑖𝐾 )⁄
𝑟
= 𝑙𝑛𝑌𝑖𝑡 − 𝑙𝑛𝐴𝑖0 − 𝑔𝑖 𝑡 + 𝛼𝑖𝐾 ln( 𝑖𝑡⁄𝑤𝑖𝑡 ) + 𝛼𝑖𝐾 ln [
𝛼𝑖𝐾 ]
Equation (46) is the equation for conditional labor demand. The independent variables are output
𝑟
𝑙𝑛𝑌𝑖𝑡 , time trend 𝑡, and relative price ratio 𝑙𝑛( 𝑖𝑡⁄𝑤𝑖𝑡 ). The total factor productivity term
1 − 𝛼𝑖𝐾⁄
𝛼𝑖𝐾 )] becomes the constant. The coefficient on output is constrained
[−𝑙𝑛𝐴𝑖𝑡 +𝛼𝑖𝐾 ln (
to be one. Thus, the only parameters to be estimated are (−𝑔𝑖 ) and 𝛼𝑖𝐾 . The data sources for the
variables are the same as those used in production function estimations.
[Insert Table 12 about here]
Table 12 reports the estimation of the conditional factor demand function based on CobbDouglas production function. As can be seen, most sectors, including sector19 representing the
entire economy, do not have a significant coefficient on the relative input price term. Even for
those three sectors with significant coefficients on relative input prices, the sign is not correct for
one of them, the real estate sector (#12). The coefficient on time trend is significant and correctly
signed for most sectors. This suggests that there is improving labor productivity over time for
most sectors, generated by technological progress. The regression results also indicate that the
time trend is a more important factor than relative input prices in determining labor use per unit
of output.
35
4.5.2. Derived conditional labor demand from CES production function
The conditional labor demand function can also be derived from the CES production
function. As derived earlier in Equation (24), the following relationship holds:
(24)
𝜎 −1
1−𝜎𝑖
𝛼𝐿𝑖𝑡 = 𝐸𝐿𝑖𝑡𝑖 𝑤𝑖𝑡
where 𝛼𝐿𝑖𝑡 is the share of labor in total factor payments. 𝐸𝐿𝑖𝑡 is the level of efficiency of labor.
Therefore, Equation 24 can be rewritten as:
(47)
𝑤𝑖𝑡 𝐿𝑖𝑡
𝜎 −1 1−𝜎
⁄𝑌 = 𝐸𝐿𝑖𝑡𝑖 𝑤𝑖𝑡 𝑖
𝑖𝑡
Thus, the labor demand can be written as:
(48)
𝜎 −1
−𝜎
𝐿𝑖𝑡 = 𝐸𝐿𝑖𝑡𝑖 𝑤𝑖𝑡 𝑖 𝑌𝑖𝑡
Since 𝐸𝐿𝑖𝑡 = 𝐸𝐿𝑖0 𝑒 𝜆𝐿𝑖 𝑡 , Equation (48) can be rewritten as follows:
(49)
𝜎 −1
−𝜎
𝑖
𝐿𝑖𝑡 = 𝐸𝐿𝑖0
𝑒 𝜆𝐿𝑖 𝑡(𝜎𝑖 −1) 𝑤𝑖𝑡 𝑖 𝑌𝑖𝑡
Taking logarithmic form of Equation (49) gives:
(50)
𝑙𝑛𝐿𝑖𝑡 = 𝑙𝑛𝑌𝑖𝑡 − 𝜎𝑖 𝑙𝑛𝑤𝑖𝑡 − (1 − 𝜎𝑖 )𝜆𝐿 𝑡 − (1 − 𝜎𝑖 )𝑙𝑛𝐸𝐿𝑖0
Equation (50) is the estimable conditional labor demand function from the CES production
function. This equation says that labor is dependent on output and real relative wage. The last
term associated with the base value of the efficiency coefficient of labor [−(1 − 𝜎𝑖 )𝑙𝑛𝐸𝐿𝑖0 ]
becomes the constant. The coefficient on time trend [−(1 − 𝜎𝑖 )𝜆𝐿 ] shows the impact of labor
efficiency improvement on labor demand. This is a constrained estimation, since the coefficient
on output 𝑙𝑛𝑌𝑖𝑡 is set to be one. The parameters to be estimated are the coefficient on real relative
wage 𝑙𝑛𝑤𝑖𝑡 , which should be equal to the opposite of the elasticity of substitution (−𝜎𝑖 ), and the
coefficient on time trend. The data for all the variables in this econometric estimation have the
same source as that used when estimating the CES production function.
[Insert Table 13 about here]
36
Table 13 presents the estimations of the parameter on real relative wage. Except for the
manufacturing sector (#4), and recreational, cultural and sporting activities (#16), labor demand
in all the other sectors shows significant negative correlation with the real relative wage. For the
economy as a whole, represented by sector 19, labor demand is also significantly correlated with
the real relative wage, with a price elasticity of 0.75. Labor uses in the traditional sector, natural
resource and energy sectors, and most service sectors are very sensitive to the change in real
relative wages and have price elasticity of labor demand greater than one. There is also a slow
decreasing trend in the labor use per unit of output for eight sectors and the economy as a whole,
indicated by negative and significant coefficients on the time term. The values of the estimated
elasticities of substitution estimated from the conditional labor demand equation based on the
CES production function in Table 13 are very different from the estimated elasticities of
substitution based on CES production function (see Table 9), which challenges the robustness of
the CES production function to describe production activities in Vietnam.
4.5.3. Derived conditional labor demand from Leontief production function
The conditional labor demand from Leontief production function can be directly obtained
by rearranging the terms in the production function. Since it assumes fixed proportions for
factors of production, the labor demand is not a function of input prices in this case. The
derivations of conditional labor demand equations based on the Leontief production function can
be obtained through the following equations. The labor usage portion of the Leontief production
function is rewritten below:
(33)
𝑌𝑖𝑡 =
𝐿𝑖𝑡
⁄𝑏
𝑖𝑡
Equation (34) presents the growth of labor efficiency coefficient:
(34)
𝑏𝑖𝑡 = 𝑏𝑖0 𝑒 𝜃𝐿𝑡
Combining Equations (34) and (33) leads to
(51)
𝐿𝑖𝑡 = 𝑏𝑖𝑡 𝑌𝑖𝑡 = 𝑏𝑖0 𝑒 𝜃𝐿𝑡 𝑌𝑖𝑡
37
The logarithmic form of Equation (51) is
𝑙𝑛𝐿𝑖𝑡 = 𝑙𝑛𝑏𝑖0 + 𝜃𝐿 𝑡 + 𝑙𝑛𝑌𝑖𝑡
(52)
Equation (52) is the estimable conditional labor demand equation based on the Leontief
production function.
As can be seen, the derived conditional labor demand equation based on Leontief
production function is very similar to the derived conditional labor demand equation from CES,
since Leontief can be seen as a special case of CES when 𝜎𝑖 = 1. As the elasticity of substitution
equals unity in CES derived conditional labor demand equation [Equation (50)], the coefficient
on time trend is reduced to the opposite of the efficiency growth rate (−𝜆𝐿 ). It is worth noting
that by definition 𝜆𝐿 is positive and increasing with labor productivity, and 𝜃𝐿 is negative and
decreasing with labor productivity. Therefore, 𝜆𝐿 is directly comparable with (−𝜃𝐿 ). Similar to
the conditional labor demand equations introduced before, the coefficient on output 𝑙𝑛𝑌𝑖𝑡 is
constrained to be one. The coefficient on time trend 𝜃𝐿 is just the rate of change in the laboroutput coefficient. The base value for the labor efficiency coefficient 𝑙𝑛𝑏𝑖0 is the constant term in
this estimation equation. Data used for this estimation have the same source as that used in
estimating production functions.
The estimation results of the conditional labor demand function using Leontief
production function are virtually identical to the estimation results of the Leontief production
function. For the entire economy (sector 19), labor demand is significantly correlated with the
trend term at 5% level, with a coefficient equal to -0.06. Except for the natural resources and
energy sectors (#3 and #5), and activities of Party and of membership organizations (#17), all the
other sectors show a significant decline in the share of labor in total output, i.e., labor
productivities improve over time. These results are the same as described earlier for the Leontief
production function.
From the estimation results of the conditional labor demand functions, we find that real
relative wage and the time trend are important determinants of labor demand. We will next apply
the RMSE to choose which derived conditional labor demand model best fit the data.
38
4.5.4. Model selection using derived conditional labor demand
The RMSE method is adopted for model selection among these three classes of
conditional labor demand functions based on different production functions. The formula to
compute the statistic for RMSE is Equation (39), with replacements of predicted and actual
output with predicted and actual labor demand for this analysis. RMSE’s for each type of
conditional labor demand functions are summarized in Table 14. As can be seen, the conditional
labor demand function from Cobb-Douglas production function has the worst fit among the three.
The conditional labor demand function based on Leontief production is the best for the
manufacturing sector, where RMSE of Leontief is only smaller than RMSE of CES by 0.0001,
whereas the conditional labor demand function based on CES is the best for all the other
seventeen sectors. For the economy as a whole, CES is the best fit.
[Insert Table 14 about here]
However, these model selection results should be used with caution. The Leontief
function can be seen as a special case of a CES function by constraining the elasticity of
substitution to be zero. Thus, in the derived conditional labor demand function based on Leontief,
the real relative wage term disappears from the equation. Illustrated in Figure 4 to Figure 8, the
real relative wage has a time trend, and so does the labor demand, suggesting a correlation
between labor demand and the real relative wage term. Therefore, as real relative wage, an
independent variable that is somewhat correlated with labor demand, is removed from the
regression model, the constraining Leontief conditional labor demand model fits worse than the
full CES conditional labor demand model. As the relative wage term is removed from the model,
due to multi-collinearity between the relative wage variable and the time trend variable, at least
part of the information contained in the real relative wage term is picked up by the trend term in
the derived conditional labor demand function based on the Leontief production function. Thus,
the goodness of fit does not differ significantly between these two types of conditional labor
demand functions for several sectors. Even though conditional labor demand function based on
CES production function fits the data better, we cannot ignore the lack of robustness of the
model, evidenced by very different estimated elasticities of substitution between conditional
labor demand estimation and the CES production function estimation.
39
The derived labor demand equations allow us to quantitatively evaluate existing or
potential policies that affect market wages and outputs. In the following section, we will
investigate if minimum wage policy affects the market wage. If so, then we can gauge the effect
of minimum wage policy on labor demand through the derived labor demand equations. We will
use the CES conditional labor demand function, but must be cautious in interpreting results.
5. Roles of policies in labor demand
In Vietnam, there are two policies under focus that may be biased toward capital
intensive development. They are the minimum wage policy and state investment policy. In this
section, we will investigate if these two policies induce capital intensive production, and how
they affect labor demand in various key economic sectors.
5.1. Minimum wage policy
The Vietnamese government has been actively influencing the market wage by adjusting
the minimum wage. The nominal minimum wage increased by 300% from 2000 to 2008.
However, the role of minimum wage policy in determining the market wage and thus labor
demand is still a controversial issue in Vietnam. The institutional wage setting mechanism in
Vietnam gives rise to various levels of association between the minimum wage and the market
wage across different firm types. Since state-owned enterprises are legally bound by the
minimum wage policy, their wages tend to follow the minimum wage more closely. However, in
informal sectors, the minimum wage policy is less respected. Since the informal and private
formal sectors account for a considerable portion of the entire Vietnamese economy, the overall
importance of the minimum wage in determining market wage is weakened. Vietnam applies
different levels of minimum wages to different types of firms. Foreign-invested firms have
higher minimum wages. Therefore, without details on ownership, we can only examine mixed
pass-through results. Also, since the number of employees required in the state-owned
enterprises is usually pre-determined and cannot exceed a certain limit, labor movement between
different types of firms is not entirely free. Wages can hardly be equalized across different firms.
Thus, higher state wages may not imply a similar level of wages paid in other types of firms. All
40
of those factors present in institutional wage setting are responsible for the mixed levels of
linkages between minimum wages and market wages.
If minimum wage drives the market wage, then the affected market wage would translate
into changes in labor demand. On the other side, however, if minimum wage policy is not an
important determinant of the market wage, then the role of the minimum wage policy in labor
demand is trivial. In the following section, we attempt to explore if the minimum wage is an
important factor in setting market wages. If minimum wage is an important driver of the market
wage, then we will evaluate to what degree the minimum wage would affect the labor-output
coefficient and the labor demand using the previously estimated conditional labor demand
function based on CES.
The first regression model we employed only has the real minimum wage and the CPI as
independent variables to explain the variation in the nominal market wage. The goal of this
estimation is to determine whether the real minimum wage or inflation is a more important
determinant of nominal market wages in Vietnam. The model is written as follows:
(53)
ln(𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑚𝑎𝑟𝑘𝑒𝑡 𝑤𝑎𝑔𝑒)it = αi ln(𝑟𝑒𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑤𝑎𝑔𝑒)t + βi ln(𝐶𝑃𝐼) + γi
The coefficient on the real minimum wage 𝛼𝑖 measures the level of correlation between nominal
market wage and real minimum wage, and the coefficient on CPI 𝛽𝑖 measures the level of
correlation between nominal market wage and inflation. In addition, we also constrained the
coefficient on CPI to be one, assuming inflation pass-through is neutral, and directly regressed
the real market wage on the real minimum wage. The mathematical expression of this model is
as follows:
(54)
ln(𝑟𝑒𝑎𝑙 𝑚𝑎𝑟𝑘𝑒𝑡 𝑤𝑎𝑔𝑒)it = δi ln(𝑟𝑒𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑤𝑎𝑔𝑒)t + θi
Figure 4 in Section 2 implies that state wages are more sensitive in response to minimum wage.
Hence, we regressed the real state wage on real minimum wage to examine if different levels of
41
responsiveness to changes in the minimum wage exist between state enterprises and enterprises
of other ownerships11. The regression model is written as:
(55)
ln(𝑟𝑒𝑎𝑙 𝑠𝑡𝑎𝑡𝑒 𝑤𝑎𝑔𝑒)it = ηi ln(𝑟𝑒𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑤𝑎𝑔𝑒)t + ζi
Nominal market wages are imputed by removing the price index from the formula used in
computing real relative wages [Equation (31)].
(56)
(𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑚𝑎𝑟𝑘𝑒𝑡 𝑤𝑎𝑔𝑒)𝑖𝑡 =
𝑣𝑎𝑙𝑢𝑒 𝑎𝑑𝑑𝑒𝑑 𝑡𝑜 𝑙𝑎𝑏𝑜𝑟𝑖𝑡
⁄𝐿
𝑖𝑡
Data sources for value added to labor and employment are the same as those introduced in
Section 4.2. Data on the minimum wage are based on the Labor Law by the National Assembly
of Vietnam, which are accessible online12. Data on state wage are from GSO Statistical
Yearbook (2010). CPI values are from IMF (2010) and GSO (2010). Deflating the nominal
wages by price indices yields real relative wages.
[Insert Table 15 about here]
The estimation results are summarized in Table 15. The results indicate that for most
sectors, inflation is a more important factor than the minimum wage in driving the market wage,
indicated by positive and relatively large correlation between the nominal market wage and
inflation for most sectors. For five out of eighteen sectors, namely sectors of fishing (#2),
manufacturing (#4), transport, storage and communications (#9), education and training (#14),
and recreational, cultural and sporting activities (#16), the minimum wage plays a significant
role in driving the market wage, still with less influence than inflation on nominal market wages.
These five sectors have positive and significant coefficients on the real minimum wage. Yet, all
of the elasticities with respect to real minimum wage are small and less than unity. For the
economy as a whole, both the real minimum wage and inflation are significant variables
determining nominal wages, but the correlation between the nominal wage and inflation is much
11
Due to lack of wage data for other types of firms, we did not run this regression model for private and foreigninvested firms.
12
The minimum wage data are summarized on a Wikipedia webpage:
http://translate.google.com/translate?js=y&prev=_t&hl=zh-CN&ie=UTF8&layout=1&eotf=1&u=http://vi.wikipedia.org/wiki/L%25C6%25B0%25C6%25A1ng_t%25E1%25BB%2591i_thi
%25E1%25BB%2583u_t%25E1%25BA%25A1i_Vi%25E1%25BB%2587t_Nam&sl=vi&tl=en
42
stronger, with a coefficient of 1.15 on inflation versus a coefficient of 0.32 on the real minimum
wage.
In the case of inflation constrained to one, i.e., neutral inflation pass-through for all the
sectors, similar patterns appear. Only five sectors have a significant but low correlation between
the real market wage and the real minimum wage. For the economy as a whole, the real
minimum wage is a significant variable in explaining the real market wage, but the elasticity is
only 0.35, similar to the unconstrained result above. As mentioned earlier, wage setting
mechanisms vary greatly across firm types. Thus, various levels of correlation between minimum
wages and market wages across firm types are found. Since the overall economy is a mix of all
firm types, the estimation results of Equations (53) and (54) based on data of the overall sector
are also only a reflection of the averaged sector.
Consistent with Figure 4, the real state wage is more closely correlated with the real
minimum wage relative to wages in enterprises of other types of ownership, which is consistent
with the fact that state-owned enterprises are more tightly bound by the minimum wage policy.
The minimum wage is important for most sectors, and the coefficients are largely above 0.5. For
the economy as a whole, we would expect that a 1% increase in the real minimum wage would
lead to 0.5% increase in the real state wage.
Minimum wages may affect labor demand by influencing market wages. The more
sensitive the market wage is in response to the minimum wage, the larger the impact on labor
demand would be. To quantitatively evaluate the impact of the changes in the minimum wage on
labor demand, we first used the estimates from Equation (54), a direct regression of real
minimum wage on real market wage, to obtain predicted changes in the real market wage. Then,
we adopted the estimates from the derived conditional labor demand equation based on the CES
production function to assess the changes in labor-output coefficients. From the magnitude of the
changes in the labor-output coefficient, we can assess the contribution of institutional wage bias
to labor demand based on the labor demand growth decomposition analysis. It is worth noting
that we only performed this analysis for three major sectors, namely, agriculture, manufacturing
and infrastructure services. These sectors account for the vast majority of changes in labor
demand partially attributable to institutional wage bias in the labor demand growth
decomposition. Also, since the estimated coefficients on real wage from Equation (54) tend to be
43
bigger than the estimates from Equation (53), the minimum wage effect we assess here would be
an upward bound.
Table 16 reports the percentage changes in the labor-output coefficients due to
institutional wage bias in the agriculture, manufacturing and infrastructure service sectors,
compared with total percentage changes in labor-output coefficients in those sectors. As can be
seen, the changes in labor-output coefficients due to institutional wage bias are not large for any
of the three sectors. In the agriculture sector, institutional wage bias leads to an 1.05% decrease
in labor use per unit of output, which is about a quarter of the total decrease in labor-output
coefficient in the agriculture sector. The institutional wage bias is even less significant in the
manufacturing sector. The labor-output coefficient only decreases by 0.11% due to the
institutional wage bias, which is responsible for only 3% of the total decline in the labor-output
coefficient in the manufacturing sector. The impact of institutional wage bias is revealed to be
the strongest in the infrastructure service sector, where the labor-output coefficient decreases by
1.37% due to institutional wage bias. The contribution of institutional bias to lowering laboroutput coefficients is about half of the estimated effect of biased technological change in the
infrastructure sector. Since the effect on labor demand is proportional to the effect on the laboroutput coefficient in the labor demand growth decomposition analysis, we can infer that the
effect of minimum wages on labor demand is also insignificant, and may account for less than
one-third of total percentage change, i.e., a 1% decrease in labor demand growth. The passthrough of the minimum wage is muted by the low correlation between the minimum wage and
the market wage, in conjunction with the low correlation between the real market wage and
labor-output coefficients in the CES-based conditional labor demand functions. The overall
outcome is small effects of institutional wage bias on slowing down labor demand, at least in the
short run. Biased technological change is still the major explainer of stagnant labor demand.
In summary, all the estimation results obtained above offer evidence that inflation, rather
than the minimum wage, is a more important factor driving market wages. Only for five sectors
is the coefficient on the minimum wage statistically significant in wage determination, but the
impact on the market wage is still lower than that of inflation. State wages tend to follow the
minimum wage more closely, consistent with the state wage setting mechanism, and hence the
minimum wage plays a more significant role in explaining changes in state wages relative to
44
explaining changes in the overall market wage or the market wage offered by enterprises of other
types of ownership. The labor demand growth decomposition analysis shows that the effect of
institutional wage bias on labor demand is greatly muted. This reinforces the notion that laboraugmenting technical change may be an important factor explaining slow labor demand growth.
Even in the case representing an upward bound, the downward pressure on the labor-output
coefficient due to institutional wage bias is only responsible for less than one-third of the total
changes in the labor-output coefficient. The low correlation between minimum wage and market
wage, combined with the low correlation between the real wage and labor-output coefficient in
the conditional labor demand function based on CES, lead to small percentage changes in laboroutput coefficient and hence in labor demand, as minimum wage is adjusted.
5.2. Role of state investment
According to MOLISA and ILSSA (2009), investment in state-owned firms tends to be
overly capital intensive, and they could absorb more labor if that investment were less capital
intensive. In this section we will partially explore if state investment is more capital intensive
than other firm types, and how that investment bias would affect the structure of the economy. In
order to evaluate the role of state investment, we first calculated labor-output coefficients for
state and non-state firms, and for the overall sector in 2003. We then analyze the relative
incremental labor-output ratio to the incremental capital-output ratio to investigate how input
intensity has evolved over time for state firms and for the overall sector. Lastly, we examined
how the sectoral shares of output are adjusted for state and non-state firms to assess the role of
state investment on labor demand growth. Due to lack of data on output and value added by firm
type, we cannot use data from the same sources to analyze the role of state investment. Rather,
we combined the GSO data with estimated sectoral output shares by ownership for 35 sectors in
2003 based on the Enterprise Survey and other assumptions by Boys (2008). Since the output
share data are necessary to compute the initial capital stock and output levels by firm type, we
chose 2003 as the base year to start this analysis. We performed these analyses for the major
45
sectors, including the agriculture13, manufacturing, construction, and infrastructure service
sectors. We computed the averaged output shares for those aggregated sectors based on the
estimated shares by Boys (2008), weighting by the sectoral output levels. Since we aim to
evaluate the role of state investment, all the analyses were implemented by firm type.
The first analysis involves labor-output ratios for state firms, non-state firms and for the
sector overall. As used earlier, total output in 2003 by sector are available from GSO (2010), and
hence output for state firms in 2003 can be derived by multiplying total sector output by output
shares of state-owned enterprises. Total employment data in 2003 are also available from GSO
(2010), and state employment data in 2003 are available from the GSO Statistical Yearbook
(2010). Once total and state output and employment data by sector are ready, taking the
difference between total values and state values gives non-state output and employment levels in
2003. With labor and output levels for state firms, non-state firms and the overall sector in 2003,
we can then directly obtain labor-output coefficients in 2003 by firm type. Table 17 presents the
results. The labor-output coefficients in state sectors are clearly lower than those in non-state
sectors, indicating less labor use per unit of output in the state sectors. The sharpest contrast
between state and non-state firms is in the agriculture sector, where labor-output coefficient is
only 0.010 in state firms and is 0.507 in non-state firms. However, it is worth noting that the
output shares are based on the Enterprise Survey, in which only formal sectors are included. The
formal sector is only a subset of each sector, and state firms may be overrepresented in this
subset relative to non-state firms. Therefore, the output share of state-owned enterprises by Boys
(2008) may be higher than the actual shares, and state output based on the estimated shares may
be overestimated. Therefore, the labor-output coefficients for state sector we derived serve as
lower bounds for this statistic.
[Insert Table 17 about here]
To evaluate the evolution of technological progress and input intensity, we also
calculated the incremental labor-output ratio relative to the incremental capital-output ratio for
state-owned firms and overall economy from 2003 to 2008.
13
It is worth noting that results on the agricultural sector should be taken with caution. State-owned enterprises in
the agricultural sector are mainly concentrated in food processing, marketing and distribution, but have very limited
involvement in primary production, which is the major employment hub in the agricultural sector.
46
The incremental labor-output ratio is defined as how much more labor is needed for one
unit of output increase. The mathematical expression is as follows:
𝑎̂
𝐿𝑖𝑡 =
(57)
∆𝐿𝑖𝑡
⁄∆𝑌
𝑖𝑡
where 𝑎̂
𝐿𝑖𝑡 denotes the incremental labor-output ratio, ∆𝐿𝑖𝑡 denotes the change in labor demand,
and ∆𝑌𝑖𝑡 denotes the change in output.
Similarly, the incremental capital-output ratio is defined as how much more capital is
needed for one unit of output increase. The mathematical expression is as follows:
𝑎̂
𝐾𝑖𝑡 =
(58)
∆𝐾𝑖𝑡
⁄∆𝑌
𝑖𝑡
where 𝑎̂
𝐾𝑖𝑡 denotes incremental capital-output ratio, and ∆𝐾𝑖𝑡 denotes the change in capital used.
Thus, the ratio between the incremental labor-output ratio (𝑎̂
𝐿𝑖𝑡 ) and the incremental
capital-output ratio (𝑎̂
𝐾𝑖𝑡 ) becomes the ratio between the change in labor and the change in
capital, shown as follows:
𝑎̂
∆𝐿
𝐿𝑖𝑡⁄
= 𝑖𝑡⁄∆𝐾
𝑎̂
𝑖𝑡
𝐾𝑖𝑡
(59)
We calculated the relative incremental ratios for state firms and the overall sectors. Total
employment and capital data by sector are the same as those used in previous analyses, and the
changes can then be directly computed. State employment data for the entire time period are
available from GSO Statistical Yearbook (2010), and hence the change in state employment can
be easily computed. Data on capital stocks for the state sector are not directly available, so they
need to be computed. The analysis is on incremental basis, and is built on strong assumptions.
As in Equation (15), the state capital stocks in each time period can be computed as
follows:
(15)
𝐾𝑠𝑡𝑎𝑡𝑒,𝑖𝑡 = 𝐾𝑠𝑡𝑎𝑡𝑒,𝑖𝑡−1 ∙ (1 − 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑎𝑛𝑛𝑢𝑎𝑙 𝑑𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒𝑖 ) + 𝐼𝑠𝑡𝑎𝑡𝑒,𝑖,𝑡
We assume that state and non-state firms have the same depreciation rate. We also assume the
same capital-output coefficient for state and non-state firms in the base year 2003. Therefore, the
47
state capital stocks in 2003 account for the same shares in total capital stocks as the
corresponding output shares. The state capital stocks in the base year can then be computed by
multiplying total capital stocks by state output shares. Since state investment data and annual
depreciation rates are available from GSO (2009), state capital stocks for the following years can
be calculated through Equation (15). As a result, the change in state capital stocks can then be
obtained.
Figure 10 to Figure 14 depict evolutions of relative input incremental ratios for state
firms and for the sector overall for the agriculture, manufacturing, construction, and
infrastructure service sectors, and for the entire economy from 2004 to 200814. The relative
incremental labor to capital ratios are lower for the state firms than the overall sector. State firms
have slower changes in labor demand relative to the changes in capital demand than the average
sector. The pattern also appears for the economy as a whole, except for 200515. The low relative
incremental input ratio in the state sector might be due to slow increase in labor demand, and/or
due to rapid capital accumulation in state invested firms. The state investment exacerbates
growth rates of capital intensity in state-owned firms and so in the overall economy. Also, by
examining the rate of change in the relative incremental ratios, we find that the overall sectors
tend to exhibit faster changes than the state firms. This might be caused by relatively static
technology in state firms. Non-state firms may experience faster technological progress than
their state counterparts, and the technological progress induces more rapid changes in input
intensity. Since differences between overall sectoral results and state results are significant in all
the figures, we can infer that private enterprises exhibit very different patterns in terms of the
evolution of relative incremental ratios over time from state-owned enterprises.
[Insert Figure 10 to Figure 14 about here]
We lastly evaluated the impact of state investment bias on labor demand growth for those
key sectors. As concluded in the previous section, the Leontief function best represents
production activities in Vietnam, particularly for predicting the output. Therefore, we assumed a
14
The variations of the ratios over time are the focus, but the absolute values of the starting points may be more
sensitive to the assumptions.
15
2005 seems to be a very special year according to Figure 14. The state relative incremental labor to capital ratio is
much larger than in all the other years. This is because of a huge increase in state employment in that year. But the
reason why state employment increased so much in 2005 is unknown.
48
Leontief production function to compute output levels in state-owned enterprises for the period
2004-2008. From the previous analysis, we find that technology in state-owned enterprises tend
to be static. Thus, we fix the labor-output ratio at the 2003 level for the state-owned enterprises,
and compute the output for the following years through the formula below:
(60)
𝑌𝑠𝑡𝑎𝑡𝑒,𝑖𝑡 =
𝐿𝑠𝑡𝑎𝑡𝑒,𝑖𝑡
⁄𝑎𝐿𝑖,2003
where 𝑌𝑠𝑡𝑎𝑡𝑒,𝑖𝑡 denotes total output in state-owned enterprises in sector i in time t, and 𝐿𝑠𝑡𝑎𝑡𝑒,𝑖𝑡
denotes state employment in sector i in time t. 𝑎𝐿𝑖,2003 denotes labor-output ratio for sector i in
2003. Equation (60) describes that state output is equal to the ratio between sectoral state
employment and the labor-output ratio for state firms at the initial level. With output by state
enterprise available, we can then compute sectoral output produced by non-state enterprises by
subtracting state output from total output. Since we have information on employment by
ownership, we can then compute labor-output ratios by ownership. The estimated output and
labor-output ratios by ownership permits us to recalculate the second term in the labor demand
decomposition equation [Equation (5)] by firm type to evaluate to what extent state investment
bias affects labor demand growth.
Table 18 shows the results of the labor demand impact of state investment bias. The
negative values in the first column indicate that the state share has been falling for all those
sectors on average during the period 2003-2008. In fact, for most years during the time period,
those modern sectors witnessed decreasing state output, not just declining shares as private firms
rapidly expanded. The booming sectors, such as the manufacturing, construction, and
infrastructure services, have mainly expanded due to increased investment by both domestic
private and foreign-invested firms. The change in the state output shares is most dramatic in the
manufacturing sector, averaging 12.5% decline every year. Private and foreign-invested firms
quickly replace those state-owned enterprises, and this leads to an overall 3.7% increase in the
share of output in the manufacturing sector. State-owned enterprises also exit relatively quickly
from the construction sector, averaging an annual 12.4% decline in the sectoral output share.
Private and foreign-invested firms increase their output shares by an average of 8.7% annually in
the construction sector. The overall increase in the construction output share in the entire
economy is 0.6%. Infrastructure service sector also experiences a large scale reduction in state
49
output shares. The sectoral output share in state-owned enterprises declines by 9.8% every year
in the infrastructure service sector. With expanding production in the private and foreigninvested firms, the overall sectoral output share increases by 1% annually. The decline in state
output shares in those sectors may have been largely achieved through privatization.
[Insert Table 18 about here]
The impact of the restructuring in state investment on labor demand also depends on
sectoral relative labor productivity levels and output shares. The last column in Table 18 reports
such effects. The results show that state investment bias has a small negative impact on labor
demand for all of these key sectors. State output shares are decreasing, but the magnitude of the
impacts are small relative to the impacts generated by non-state investments, mainly because of
low labor requirement in production, evidenced by relatively low labor-output coefficient. For
the manufacturing sector, declining state investment lowers labor demand by 0.02% annually on
average from 2003 to 2008. By contrast, non-state firms increase employment by 0.98% annually
during the same time period. The overall employment in the manufacturing sector increases by
0.51% per year on average due to structural transformation and state investment bias. In the
construction sector, even though employment decreases by 0.15% per year due to reduction in
state investment in the construction sector, private and foreign-invested firms expand
employment by 0.37% every year. The overall impact on labor demand due to changes in output
shares is very small, a 0.03% increase in labor demand due to changes in output shares. The role
of state investment on labor demand in infrastructure services is similarly small. State investment
bias causes 0.09% decrease in labor demand. The decline in employment is replenished by
expanding private and foreign-invested firms. The non-state sectors generate a 1.46% increase in
employment in the infrastructure service sector. The overall infrastructure service sector
experiences a 0.18% increase in labor demand due to the output restructuring. The results
suggest that restructuring into the private and foreign-invested firms, which are relatively less
capital intensive, is an important factor driving labor demand growth. If state firms were
expanding in output shares, then the labor demand growth rate in those modern sectors would be
lower due to relatively low labor requirement in state enterprises. The declining importance of
state production in those key modern sectors weakens the impact of an overly capital-intensive
production strategy pursued by state firms on slowing down labor demand growth for the entire
50
economy. But it may be the case that the restructuring could occur more rapidly, making the gap
between GDP and labor demand growth smaller as private firms expand more rapidly.
All these analyses show that state investment tends to exacerbate the growth rate of
capital intensity in the economy. Yet, technological progress in the state enterprises may not be
as fast as in private and foreign-invested firms. The increase in capital intensity in state-owned
enterprises is more a result of capital accumulation, not labor augmenting technical progress.
Due to privatization in those sectors, sectoral state output declined in most years during the
period 2003-2008. As private and foreign-invested firms expanded, state output shares in GDP
decreased rapidly. Even though state investment tends to be more capital intensive, because of its
declining role in production in many important modern sectors, the impact on slowing down
labor demand is limited. Restructuring into non-state sectors generates more significant labor
demand impacts.
6. Conclusions
Labor demand in Vietnam has grown much more slowly than GDP over the last two
decades. Since 2000 until 2008 GDP has grown on average 7.6% per year while employment
grew at only 2.2% per year. While that difference reflects improvements in labor productivity it
also raises concerns that economic development is not creating enough new jobs. Structural
transformation has moved labor from lower productivity traditional sectors, especially
agriculture, to higher productivity modern sectors including manufacturing and services. As of
2008, agriculture accounted for 18.3% of GDP but it still employed 48.9% of the workforce
(GSO, 2010). The Vietnamese labor ministry in its recent assessment of the labor situation in
Vietnam (MOLISA and ILSSA, 2009) believes this restructuring is not moving sufficiently
rapidly, has involved too little innovation, and exhibits an overly capital intensive development
strategy.
Slow labor demand growth has been seen elsewhere in Asia. While there are some cases
where the elasticity of employment demand with respect to output growth reached 0.7 to 0.8,
according to data from the World Bank (2010) in many Asian countries when GDP capita was at
51
a level similar to that found in Vietnam recently, this employment elasticity has been in the range
of 0.28 found in Vietnam over the last decade. Moreover, some believe the “Asian miracle” has
been the result of high savings rates and capital accumulation with relatively little technical
progress (e.g. Young, 1995). Others have argued that estimates of TFP growth in Asia are
underestimates, and that correcting for bias toward labor augmenting technical change would
result in higher estimates of TFP growth (e.g. Rodrik, 1997). The controversy on estimating TFP
growth is found for Vietnam, as well. MOLISSA and ILSAA (2010) believe that only a quarter
of Vietnam’s recent rapid economic growth can be attributed to technical progress, and some
earlier estimates find even slower technical change.
We used data for 18 aggregate sectors and the overall Vietnamese economy from 2000 to
2008 provided by GSO (2009, 2010) to examine the roles played by structural transformation,
technical progress and institutional bias toward capital intensive development to explain the
difference observed between labor demand growth and GDP growth. Decomposition of the
factors behind labor demand growth attribute only 30% of this difference to shifts from low
productivity sectors to higher productivity sectors, while the remaining 70% comes from
declining labor use per unit output that is also found in agriculture. That simple decomposition
cannot separate technical progress from effects of higher wage-rental ratios driving choice of
capital intensive techniques. Nor can it distinguish between sectoral composition changes due to
structural transformation versus choices toward capital intensive sectors made by state owned
(and possibly foreign invested) enterprises. We attempted to sort between these competing
factors by investigating estimation of TFP growth by sector and by examining institutional biases
that might lead to more capital intensive development.
We estimated technical progress using several production function specifications
following both accounting and econometric approaches. Accounting approaches based on CobbDouglas production functions are most common in the literature, and their application to
Vietnamese data lead to reasonable but somewhat low estimates of Hicks-neutral technical
change. Econometric estimation of Cobb-Douglas production functions gives vastly different
results and parameters are so unreasonable as to call into question this description of production
functions. Use of Young’s (1995) accounting method based (he claims) on a translog production
function gives results indistinguishable from Cobb Douglas. Assumption of a unitary elasticity of
52
substitution must be relaxed to examine biased technical change. Our attempts to apply a CES
production function provide evidence of low elasticties of substitution, but results are not robust
and sectoral parameters are found in implausible ranges. Estimation using a Leontief production
function provides the most robust results, and estimates of TFP growth that are not unreasonably
large but larger than found using the more standard approaches. For the overall economy labor
efficiency is found to grow at 5.8 % per year, while capital efficiency grows at 2.0% per year.
Comparisons of the predictive ability of alternative production function and derived conditional
labor demand specifications confirm the robustness of the Leontief production function results.
Those results are consistent with significant labor augmenting technical progress and better
explain historical economic performance in Vietnam. This confirms suspicion that there may
have been significant technical progress behind the Asian miracle, but one must address a labor
augmenting bias in technical change to find it.
Rapidly rising minimum wages were found to contribute to a limited extent to more
capital intensive development when we looked at the two key links between the minimum wage
and labor intensity – the effect of minimum wages on overall wages and the effect of wages on
technical choices. Market wage trends have followed well behind the minimum wage. From
2000 to 2008 the nominal minimum wage rose 300% while the overall market wage rose only
50%. Wages in state owned enterprises followed minimum wages more closely, but they only
rose 60% over this period. We also found that inflation better explained wage evolution than did
minimum wages. Low elasticities of substitution found using a CES production function mean
higher wage rental ratios contribute relatively little to declining labor use per unit output. Trends
corresponding to technical progress captured more of change in input use per unit output than did
wage changes. Thus, the 3.5% differential between GDP and labor demand growth due to falling
labor output ratios is mostly due to technical progress. Institutional wages driving capital
intensive technique choices explains at most about one-third of the difference. Undoubtedly, low
wages have played a role in sectoral and technical choices made in Vietnam, but over the short
run institutional biases in wage setting are only a limited factor explaining this data.
Investment by state owned enterprises appears to be much more capital intensive, due to
both sectoral and technical choices, and factor use patterns have changed less rapidly than in the
private sector. Estimates of input-output ratios for state owned firms are much different from
53
those found for private firms. These ratios appear to change more slowly than in the private
sector, and exhibit less technical change. Data limitations prevent us from looking at technology
adopted by foreign invested enterprises. The state is also more heavily invested in capital
intensive sectors, such as energy and mining. Strong assumptions lead us to find declining roles
played by state-owned enterprises in the manufacturing, construction, and infrastructure service
sectors. The overall output expansion in those sectors are brought by increasing production levels
of private and foreign-invested enterprises. The impact of overly capital intensive state
investments on slowing down labor demand is weakened by the declining role of state
enterprises in production. If restructuring occurred faster, however, enabling private firms
expanded more rapidly, the differential might narrow faster as well. Changes in output shares in
non-state sectors generate more significant labor demand impacts than do investments in stateowned enterprises.
Structural transformation only partially explains slow labor demand growth in Vietnam.
While some of the difference from GDP growth may be attributed to capital intensive investment
by the state, a significant share of the difference is found to be due to technical change when low
elasticities of substitution and labor augmenting bias in technical change are taken into account.
54
List of Tables
Table 1. Real GDP growth and employment growth in selected Asian countries from 1986 to
2008............................................................................................................................................... 56
Table 2. Sectoral labor productivity from 2000 to 2008 ............................................................... 57
Table 3. Unemployment and underemployment rates in rural and urban Vietnam in 2009 ......... 58
Table 4. Average monthly wages and annual wage growth rates from 1998 to 2006 .................. 59
Table 5. Average growth rates of labor and value added by sector from 2000 to 2008 ............... 60
Table 6. Labor demand growth decomposition by sector, 2000- 2008 ........................................ 61
Table 7. Estimated cost shares and average rates of total factor productivity growth (TFPG) by
sector assuming a Cobb-Douglas production function ................................................................. 62
Table 8. Average TFPG by sector and by year assuming a translog production function,
compared with average TFPG from Cobb-Douglas accounting approach ................................... 64
Table 9. Estimation of elasticities of substitution, input efficiency coefficients and directions of
factor augmentation from derived equations based on a CES production function ..................... 65
Table 10. Average unit input coefficients and estimation of growth rates of input efficiency and
directions of factor augmentation from a Leontief production function ...................................... 67
Table 11. Model selection between Cobb-Douglas, CES and Leontief production functions based
on the root mean squared error (RMSE) method .......................................................................... 69
Table 12. Estimation of derived conditional labor demand function from Cobb-Douglas
production function ....................................................................................................................... 70
Table 13. Estimation of derived conditional labor demand function from a CES production
function ......................................................................................................................................... 71
Table 14. Model selection between conditional labor demand functions from Cobb-Douglas,
CES and Leontief production functions based on the root mean squared error (RMSE) method 72
Table 15. Estimation of wage determination equations ................................................................ 73
Table 16. The impact of institutional wage bias on labor-output coefficients, 2000-2008 .......... 74
Table 17. Labor-output coefficients in 2003 by firm type ............................................................ 75
Table 18. Impact of state investment bias on labor demand from 2003 to 2008 .......................... 76
55
Table 1. Real GDP growth and employment growth in selected Asian countries from 1986
to 2008
Country
China
Korea, Rep.
Thailand
Indonesia
Philippines
Vietnam
China
Korea, Rep.
Thailand
Indonesia
Philippines
Vietnam
1986-90
1991-95
1996-2000
2001-05
2006-08
Gap between real GDP growth and employment growth (%)
5.7
11
7.5
8.6
10.5
6.4
5.6
3.4
3.4
3.5
7.7
8.3
-0.4
3.4
3.1
4.6
5.5
-1.9
2.7
4.2
2.0
-0.9
1.7
2.0
3.3
2.7
6.4
4.7
5.7
5.0
Elasticity of labor demand with respect to real GDP
0.28
0.11
0.13
0.10
0.06
0.33
0.29
0.25
0.24
0.16
0.25
0.04
1.58
0.33
0.25
0.35
0.30
2.91
0.42
0.30
0.57
1.42
0.58
0.55
0.40
0.53
0.28
0.30
0.27
0.32
Source: Calculated using data from WDI (2010)
56
Table 2. Sectoral labor productivity from 2000 to 2008
2000
2005
2007
2008
7,276
9,242
10,444
10,906
5.2%
31,854
39,575
45,095
45,985
5.0%
3,929
4,972
5,703
6,031
5.3%
79,203
41,888
39,285
39,596
-10.0%
2,428
2,926
3,183
3,339
3.9%
Mining and quarrying
72,048
66,981
55,104
48,852
-3.8%
Manufacturing
14,504
17,022
19,083
19,840
3.9%
Electricity, gas and water supply
76,626
74,287
68,452
66,336
-1.6%
Construction
19,852
17,223
18,906
17,841
-0.7%
Wholesale and retail trade
11,457
12,963
14,274
14,965
3.1%
Hotels and restaurants
12,931
17,553
20,993
22,338
6.9%
Transport, storage and communications 9,137
12,678
15,437
17,407
7.5%
Financial intermedation
75,133
52,444
45,979
46,756
-7.0%
Scientific activities and technology
83,564
96,653
101,784
108,030
2.8%
Real estate, renting and business
191,408
97,860
73,481
64,684
-13.7%
Public administration and defence; compulsory
21,327 social
16,158
security
15,363
14,966
-4.7%
Education and training
9,207
10,640
11,408
11,932
3.1%
Health and social work
17,491
15,680
17,101
17,801
-0.3%
Value added per worker
State
Private
Foreign investment sector
Agriculture and forestry
Source: GSO, 2009
57
Growth 2000-07
Table 3. Unemployment and underemployment rates in rural and urban Vietnam in 2009
2009
Unemployment rate (%)
Underemployment rate (%)
General Urban
2.90
4.60
5.61
3.33
Source: GSO, 2010
58
Rural
2.25
6.51
Table 4. Average monthly wages and annual wage growth rates from 1998 to 2006
Form of ownership
Average monthly salary of a worker, 1000
VND
1998
2002
2004
2006
Households
552
606
649
664
2.34
Private and collective
554
771
852
936
6.77
State
572
1002
1077
1103
8.55
Foreign investment
680
1037
1044
1316
8.60
Wage gap between FDI and
household sectors
1.2
1.7
1.6
2.0
Annual
growth rate
per cent
Source: MOLISA and ILSSA, 2009. Based on GSO and VHLSS, 1998-2006
59
Table 5. Average growth rates of labor and value added by sector from 2000 to 2008
Code
Activity
1
2
3
4
5
6
7
8
9
10
Growth rate (%) of:
Labor
Value added
-0.44
3.91
8.27
4.79
7.18
11.72
10.42
9.58
3.18
8.16
14.37
5.48
4.86
6.22
11.31
7.74
7.23
6.38
2.22
7.57
Agriculture, forestry and fishing
Energy and natural resources
Manufacturing
Construction
Infrastructure services
Professional services
Education and health
Public services
Other services
GDP
60
Table 6. Labor demand growth decomposition by sector, 2000- 2008
Code
2
3
4
5
6
7
8
9
10
Activity
Agriculture, forestry and
fishing
Energy and natural
resources
Manufacturing
Construction
Infrastructure services
Professional services
Education and health
Public services
Other services
GDP
Labor
demand
growth
rate (%)
gLi
Relative
unit
labor
use
aLi/aL
GDP
share
Si
Biased
technical
Growth
change and
rate of unit institutional
labor use
wage bias
(%) gaLi
(%)
Growth
rate of
GDP
shares
gSi
Structural
transformation
and state
investment bias
(%)
-0.44
2.91
0.20
-4.27
-2.48
-3.46
-2.01
8.27
7.18
10.42
3.18
14.37
4.86
11.31
7.23
2.22
0.14
0.54
0.52
0.68
0.12
1.31
0.37
0.74
1.00
0.08
0.22
0.09
0.24
0.07
0.03
0.05
0.03
1.00
3.62
-3.90
1.33
-4.66
9.04
-1.16
3.85
1.05
-5.07
0.04
-0.47
0.06
-0.75
0.07
-0.04
0.07
0.02
-3.48
-2.65
3.79
1.79
0.55
-1.96
-1.27
0.16
-1.11
0.00
-0.03
0.46
0.08
0.09
-0.02
-0.05
0.00
-0.02
-1.50
61
GDP growth
rate (%) gY
7.57
Table 7. Estimated cost shares and average rates of total factor productivity growth (TFPG) by sector assuming a CobbDouglas production function1, 2
Code Activity
1
Agriculture and forestry
Accounting Approach
𝜶𝑲 3 𝜶𝑳 3 Average TFPG
0.17 0.83 0.054
2
Fishing
0.32 0.68 0.042
3
Mining and quarrying
0.57 0.43 -0.006
4
Manufacturing
0.51 0.49 0.034
5
Electricity, gas and water supply
0.49 0.51 -0.008
6
Construction
0.51 0.49 0.043
7
Wholesale and retail trade; repair of motor vehicles,
motor cycles and personal and household goods
0.48 0.52 0.070
8
Hotels, restaurant
0.46 0.54 0.096
9
Transport, storage and communications
0.51 0.49 0.106
10
Financial intermediation
0.45 0.55 0.070
11
Scientific activities and technology
0.23 0.77 0.097
12
Real estate, renting and business activities
0.49 0.51 -0.034
62
Econometric Approach
Average TFPG
𝜶𝑲 3
𝜶𝑳 3
-1.19
2.19
0.076
(-0.32) (0.59)
(1.20)
0.73
0.27
0.033
(2.21) (0.83)
(1.39)
0.67
0.33
-0.004
(0.40) (0.20)
(-0.06)
0.26
0.74
0.046
(0.41) (1.19)
(1.50)
0.72
0.28
-0.003
(1.27) (0.50)
(-0.10)
1.75* -0.75
0.040
(3.62) (-1.55) (1.89)
-3.01
(-0.43)
3.19
(1.89)
-2.24
(-1.04)
1.15
(1.61)
0.86*
(2.85)
0.45**
(5.10)
4.01
(0.57)
-2.19
(-1.30)
3.24
(1.51)
-0.15
(-0.21)
0.15
(0.48)
0.55***
(6.27)
0.250
(0.68)
-0.037
(-0.43)
0.328
(1.88)
0.121
(1.82)
0.082*
(3.18)
-0.037**
(-4.25)
Table 7 continued
Code Activity
Accounting Approach
𝜶𝑲 3 𝜶𝑳 3 Average TFPG
Econometric Approach
Average TFPG
𝜶𝑲 3
𝜶𝑳 3
-0.34
(-0.48)
-0.11
(-0.05)
0.56
(0.67)
-0.002
(-0.00)
0.38
(0.96)
1.34
(1.92)
1.11
(0.48)
0.44
(0.52)
1.002
(1.28)
0.62
(1.58)
0.84
(2.02)
0.74
(1.68)
0.16
0.019
(0.38) (1.22)
0.26
0.028
(0.60) (1.60)
13
Public administration and defense; compulsory
social security
0.11 0.89 0.026
14
Education and training
0.18 0.82 0.081
15
Health and social work
0.21 0.79 0.068
16
Recreational, cultural and sporting activities
0.34 0.66 0.095
17
Activities of Party and of membership organizations
0.08 0.92 -0.025
18
Community, social and personal service activities
and private household with employed persons
0.22 0.78 0.027
19
GDP
0.39 0.61 0.041
Notes: 1 Numbers in parentheses are t statistics.
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
3
𝛼𝐾 and 𝛼𝐿 denote the share of capital and labor in total factor payments, respectively.
63
0.005
(0.12)
0.088
(1.40)
0.063
(2.18)
0.121
(1.93)
0.003
(0.06)
Table 8. Average TFPG by sector and by year assuming a translog production function,
compared with average TFPG from Cobb-Douglas accounting approach
Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Activity
Agriculture and forestry
Fishing
Mining and quarrying
Manufacturing
Electricity, gas and water supply
Construction
Wholesale and retail trade; repair of motor
vehicles, motor cycles and personal and
household goods
Hotels, restaurant
Transport, storage and communications
Financial intermediation
Scientific activities and technology
Real estate, renting and business activities
Public administration and defense; compulsory
social security
Education and training
Health and social work
Recreational, cultural and sporting activities
Activities of Party and of membership
organizations
Community, social and personal service
activities and private household with
employed persons
GDP
64
Average TFPG from
translog accounting
approach based on
Young (1995)
0.054
0.042
-0.006
0.034
-0.008
0.043
Average TFPG from
Cobb-Douglas
accounting approach
0.054
0.042
-0.006
0.034
-0.008
0.043
0.069
0.096
0.106
0.070
0.097
-0.035
0.070
0.096
0.106
0.070
0.097
-0.034
0.026
0.026
0.081
0.068
0.095
0.081
0.068
0.095
-0.026
-0.025
0.027
0.040
0.027
0.041
Table 9. Estimation of elasticities of substitution, input efficiency coefficients and directions of factor augmentation from
derived equations based on a CES production function1, 2
Code
Activity
Hicks-neutral
technological change
3
1
Agriculture and forestry
2
Fishing
3
Mining and quarrying
4
Manufacturing
5
Electricity, gas and water
supply
Construction
6
7
8
9
10
Wholesale and retail
trade; repair of motor
vehicles, motor cycles
and personal and
household goods
Hotels, restaurant
Transport, storage and
communications
Financial intermediation
𝝈𝒊
-0.38
(-0.86)
0.00***
(0.00)
0.97***
(9.45)
1.52***
(24.93)
0.57***
(5.35)
0.28
(1.41)
0.00***
(0.00)
0.00***
(0.00)
1.31***
(27.16)
0.75***
(33.66)
𝝀𝒌 = 𝝀𝑳
0.817
0.015
-0.418
0.020
-0.001
0.053
0.470
-0.002
0.040
1.850
4
𝝈𝒊
0.13***
(3.34)
1.41***
(13.08)
1.21***
(4.67)
0.87**
(2.35)
0.51***
(3.96)
0.69***
(10.69)
0.09
Biased technological change
Capital
or
5
4
4
(𝝈𝒊 − 𝟏)(𝝀𝒌 − 𝝀𝑳 ) 𝝀𝒌
𝝀𝑳
augmenting
0.017***
0.050
0.070 Labor
(33.04)
-0.020**
-0.273
-0.239 Capital
(-4.18)
0.006
-0.584
-0.589 Labor
(1.00)
0.019
0.363
0.520 Labor
(1.78)
-0.003
0.061
0.044 Capital
(-0.93)
0.022***
0.028
0.101 Labor
(10.16)
0.052***
0.065
0.125 Labor
(1.68)
0.47***
(-0.08)
0.15
(1.57)
0.69***
(6.24)
(46.05)
0.039***
(18.53)
0.072***
(12.26)
-0.007
(-0.55)
3
65
0.099
0.178
Labor
0.073
0.167
Labor
0.209
0.185
Capital
labor
Table 9 continued
Code
11
12
Activity
Scientific activities and
technology
Real estate, renting and
business activities
14
Public administration and
defense; compulsory
social security
Education and training
15
Health and social work
16
Recreational, cultural and
sporting activities
17
Activities of Party and of
membership organizations
Community, social and
personal service activities
and private household
with employed persons
GDP
13
18
19
Hicks-neutral
technological change
𝝈𝒊 3
0.00***
(7.37)
𝝀𝒌 = 𝝀𝑳 4
0.227
𝝈𝒊 3
0.85***
(12.89)
Biased technological change
Capital or labor
(𝝈𝒊 − 𝟏)(𝝀𝒌 − 𝝀𝑳 )5 𝝀𝒌 4
𝝀𝑳 4
augmenting
0.008**
0.309
0.379 Labor
(4.59)
0.67***
(34.05)
-0.094
0.75***
(19.86)
0.009
(2.26)
-0.065
-0.015
Labor
0.85***
(10.69)
1.01***
(23.64)
1.08***
(5.03)
1.56
(1.79)
0.92***
(22.64)
0.008
0.23
(0.86)
0.56***
(4.19)
0.87***
(5.06)
-0.07
(-0.57)
0.93***
(7.14)
-0.039
(-2.40)
0.012*
(3.40)
0.008*
(2.78)
0.07***
(23.22)
0.002
(0.13)
0.079
0.031
Capital
0.074
0.104
Labor
0.395
0.455
Labor
0.050
0.128
Labor
0.644
0.602
Capital
1.09**
-0.049
0.88***
0.007*
0.414
0.473
Labor
0.045
(3.43)
0.36***
(5.04)
(2.58)
0.035***
(30.06)
0.049
0.104
Labor
(3.30)
1.54**
(2.31)
-0.140
0.092
0.109
0.011
Notes: 1 Numbers in parentheses are t statistics.
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
3
𝜎𝑖 denotes estimated elasticity of substitution.
4
𝜆𝑘 and 𝜆𝐿 denote estimated growth rates of capital efficiency and labor efficiency, respectively.
5
(𝜎𝑖 − 1)(𝜆𝑘 − 𝜆𝐿 ) is the coefficient on time trend.
66
Table 10. Average unit input coefficients and estimation of growth rates of input efficiency and directions of factor
augmentation from a Leontief production function 1, 2
Code Activity
Average unit
capital coeff. ak
1
Agriculture and forestry
5.77
2
Fishing
0.80
3
Mining and quarrying
3.11
4
Manufacturing
0.48
5
Electricity, gas and water
supply
Construction
5.41
Wholesale and retail trade;
repair of motor vehicles,
motor cycles and personal
and household goods
Hotels, restaurant
0.99
6
7
8
9
10
11
12
Average unit Capital efficiency
labor coeff. al growth rate
−𝜽𝒌 = 𝝀𝑲
13.11
0.0408***
(12.35)
3.49
0.0153*
(3.27)
0.48
0.0317
(1.43)
0.78
0.0160**
(4.42)
0.47
0.00793
(0.99)
1.15
0.0488**
(5.08)
3.78
0.0544*
(2.70)
Labor efficiency
growth rate
−𝜽𝑳 = 𝝀𝑳
0.0573***
(16.51)
0.0533***
(7.21)
0.012
(0.60)
0.0597***
(22.81)
-0.00296
(-0.33)
0.0577**
(3.64)
0.107**
(5.17)
Capital or labor
augmenting
1.03
1.87
6.23
2.55
1.66
0.67
Scientific activities and
technology
2.40
0.28
0.131***
(11.00)
0.150***
(27.18)
0.0259
(1.30)
0.113***
Labor
Transport, storage and
communications
Financial intermediation
0.0817***
(8.14)
0.0689***
(12.12)
0.110***
(6.64)
0.0949***
0.30
(10.91)
-0.00405
(13.43)
-0.0631***
(-0.51)
(-14.24)
Real estate, renting and
business activities
0.57
0.85
67
Labor
Labor
Capital
Labor
Capital
Neutral3
Labor
Labor
Capital
Labor
Capital
Table 10 continued
Code
Activity
Average unit
capital coeff. ak
2.06
14
Public administration and
defense; compulsory social
security
Education and training
Average unit
labor coeff. al
1.62
2.58
3.46
15
Health and social work
1.37
2.01
16
Recreational, cultural and
sporting activities
Activities of Party and of
membership organizations
Community, social and
personal service activities
and private household with
employed persons
GDP
3.96
2.08
3.09
8.18
12.11
3.00
5.16
7.51
13
17
18
19
Capital efficiency
growth rate
−𝜽𝒌 = 𝝀𝑲
0.0822***
(10.91)
Labor efficiency
growth rate
−𝜽𝑳 = 𝝀𝑳
0.0303**
(5.32)
Capital or labor
augmenting
0.0698***
(13.56)
0.0716***
(9.84)
0.0475***
(8.42)
0.0713***
(6.85)
0.0188**
0.0952***
(18.12)
0.0846***
(9.98)
0.120***
(26.58)
-0.0209
(-1.92)
0.0292**
Labor
(4.05)
0.0200***
(15.02)
(4.28)
0.0580***
(27.63)
Capital
Labor
Labor
Capital
Labor
Labor
Notes: 1 Numbers in parentheses are t statistics.
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
3
For this sector, the efficiency coefficients are individually significant. The difference between the two coefficients is not
significantly different from zero.
68
Table 11. Model selection between Cobb-Douglas, CES and Leontief production functions based on the root mean squared
error (RMSE) method
Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Activity
Agriculture and forestry
Fishing
Mining and quarrying
Manufacturing
Electricity, gas and water supply
Construction
Wholesale and retail trade; repair of motor
vehicles, motor cycles and personal and
household goods
Hotels, restaurant
Transport, storage and communications
Financial intermediation
Scientific activities and technology
Real estate, renting and business activities
Public administration and defence; compulsory
social security
Education and training
Health and social work
Recreational, cultural and sporting activities
Activities of Party and of membership
organizations
Community, social and personal service
activities and private household with employed
persons
GDP
RMSE/actual output of:
Cobb-Douglas
CES
Econometric Accounting
0.13
0.93
0.15
0.10
0.96
0.59
0.16
0.17
1.41
0.09
0.85
0.24
0.06
0.58
0.33
0.19
0.81
0.61
Leontief
Capital1 Labor2
0.02
0.02
0.03
0.05
0.17
0.15
0.02
0.02
0.06
0.07
0.06
0.09
Model of best fit
Leontief
Leontief
Leontief
Leontief
Leontief
Leontief
0.29
0.30
0.31
0.33
0.26
0.10
0.67
0.84
0.33
0.96
1.07
0.88
19.31
0.53
0.35
553.42
0.39
1.50
0.14
0.05
0.03
0.14
0.08
0.07
0.14
0.07
0.03
0.19
0.05
0.04
Leontief
Leontief
Leontief
Leontief
Leontief
Leontief
0.17
0.22
0.22
0.23
1.06
1.04
1.05
0.94
0.33
0.41
10.44
0.40
0.05
0.04
0.05
0.04
0.03
0.04
0.05
0.03
Leontief
Leontief
Leontief
Leontief
0.08
1.04
1.34
0.08
0.06
Leontief
0.07
0.11
0.99
0.43
0.27
0.07
0.04
0.01
0.05
0.01
Leontief
Leontief
Notes: 1 This column reports the ratio between RMSE and actual output based on the capital usage in the Leontief production function.
2
This column reports the ratio between RMSE and actual output based on the labor usage in the Leontief production function.
69
Table 12. Estimation of derived conditional labor demand function from Cobb-Douglas
production function1, 2
Code
1
Activity
Agriculture and forestry
2
Fishing
3
Mining and quarrying
4
Manufacturing
5
Electricity, gas and water supply
6
Construction
7
8
Wholesale and retail trade; repair of motor vehicles,
motor cycles and personal and household goods
Hotels, restaurant
9
Transport, storage and communications
10
Financial intermediation
11
Scientific activities and technology
12
Real estate, renting and business activities
13
14
Public administration and defense; compulsory social
security
Education and training
15
Health and social work
16
Recreational, cultural and sporting activities
17
Activities of Party and of membership organizations
18
19
Community, social and personal service activities and
private household with employed persons
GDP
-Coefficient
on time gi
0.0588***
(16.62)
0.0517***
(6.03)
0.0384*
(2.49)
0.0600***
(21.19)
0.0232
(1.71)
0.0486
(1.85)
0.106***
(7.48)
0.137***
(6.02)
0.160***
(24.30)
-0.0789
(-1.39)
0.108***
(16.50)
-0.0975***
(-7.78)
0.0217
(1.73)
0.0951***
(13.85)
0.0809***
(6.60)
0.118***
(21.10)
-0.0964
(-1.95)
0.0464*
(3.70)
0.0583***
(26.42)
Coefficient on rentalwage ratio 𝜶𝒊𝑲
-0.157
(-1.24)
-0.0901
(-0.46)
0.695*
(3.21)
0.0102
(0.19)
0.340
(2.29)
-0.205
(-0.44)
0.650*
(2.94)
-0.145
(-0.30)
-0.239
(-2.07)
-0.907
(-1.93)
0.307
(2.26)
-0.190*
(-2.67)
-0.106
(-0.74)
-0.00170
(-0.02)
-0.108
(-0.53)
-0.0534
(-0.58)
-0.560
(-1.56)
0.209
(1.56)
-0.0446
(-0.76)
Notes: 1 Numbers in parentheses are t statistics
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
70
Table 13. Estimation of derived conditional labor demand function from a CES production
function1, 2
Code
1
Activity
Agriculture and forestry
2
Fishing
3
Mining and quarrying
4
Manufacturing
5
Electricity, gas and water supply
6
Construction
7
8
Wholesale and retail trade; repair of motor vehicles,
motor cycles and personal and household goods
Hotels, restaurant
9
Transport, storage and communications
10
Financial intermediation
11
Scientific activities and technology
12
Real estate, renting and business activities
13
14
Public administration and defense; compulsory
social security
Education and training
15
Health and social work
16
Recreational, cultural and sporting activities
17
Activities of Party and of membership organizations
18
Community, social and personal service activities
and private household with employed persons
GDP
19
-Coeff. on real
relative wage 𝝈𝒊
1.052***
(6.97)
1.664**
(4.83)
1.022***
(31.68)
0.321
(0.93)
0.786***
(6.88)
1.358***
(10.42)
0.897***
(57.64)
1.035***
(33.51)
1.055***
(10.73)
1.641*
(3.04)
1.020***
(19.73)
0.950**
(3.98)
0.947***
(16.52)
0.752**
(4.22)
1.092***
(6.50)
0.771
(1.82)
0.953***
(19.07)
1.126***
(6.02)
0.745**
(5.37)
Notes: 1 Numbers in parentheses are t statistics
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
71
Coeff. on time
(𝝈𝒊 − 𝟏)𝝀𝑳
0.00252
(0.29)
0.00764
(0.58)
-0.0184***
(-11.03)
-0.0456*
(-2.95)
-0.00452
(-1.26)
0.0196
(2.34)
-0.0259***
(-15.36)
-0.00286
(-0.73)
-0.00501
(-0.37)
-0.0446*
(-3.00)
-0.0112
(-2.21)
0.00519
(0.34)
-0.00492*
(-2.78)
-0.0319
(-2.09)
-0.0252*
(-2.58)
-0.0486
(-1.23)
-0.0159***
(-6.49)
-0.0235***
(-8.01)
-0.0210*
(-3.02)
Table 14. Model selection between conditional labor demand functions from Cobb-Douglas,
CES and Leontief production functions based on the root mean squared error (RMSE)
method
Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Activity
Agriculture and forestry
Fishing
Mining and quarrying
Manufacturing
Electricity, gas and water
supply
Construction
Wholesale and retail trade;
repair of motor vehicles,
motor cycles and personal
and household goods
Hotels, restaurant
Transport, storage and
communications
Financial intermediation
Scientific activities and
technology
Real estate, renting and
business activities
Public administration and
defense; compulsory social
security
Education and training
Health and social work
Recreational, cultural and
sporting activities
Activities of Party and of
membership organizations
Community, social and
personal service activities
and private household with
employed persons
GDP
RMSE of:
Cobb-Douglas
CES
0.026
0.010
0.061
0.028
0.101
0.013
0.019
0.018
Leontief
0.027
0.058
0.154
0.018
CES
CES
CES
Leontief
0.059
0.130
0.027
0.030
0.074
0.122
CES
CES
0.110
0.099
0.007
0.007
0.159
0.092
CES
CES
0.035
0.132
0.010
0.105
0.043
0.154
CES
CES
0.051
0.008
0.063
CES
0.028
0.021
0.038
CES
0.046
0.044
0.070
0.007
0.022
0.025
0.044
0.040
0.066
CES
CES
CES
0.039
0.031
0.036
CES
0.077
0.012
0.084
CES
0.048
0.017
0.022
0.007
0.053
0.016
CES
CES
72
Model of best fit
Table 15. Estimation of wage determination equations1, 2
Code
Activity
nominal wage
1 Agriculture and forestry
2 Fishing
3 Mining and quarrying
4 Manufacturing
5 Electricity, gas and water
supply
6 Construction
7 Wholesale and retail trade;
repair of motor vehicles, motor
cycles and personal and
household goods
8 Hotels, restaurant
9 Transport, storage and
communications
10 Financial intermediation
11 Scientific activities and
technology
12 Real estate, renting and
business activities
13 Public administration and
defense; compulsory social
security
14 Education and training
15 Health and social work
16 Recreational, cultural and
sporting activities
17 Activities of Party and of
membership organizations
18 Community, social and personal
Dependent variables:
real wage
real state wage
real min
wage 𝜶𝒊
0.0963
(1.36)
0.286**
(3.88)
0.504
(2.04)
0.262*
(2.68)
0.132
(0.79)
-0.0433
(-0.26)
-0.103
CPI 𝜷𝒊
1.679***
(11.99)
0.906***
(6.23)
0.149
(0.31)
0.814**
(4.21)
0.371
(1.13)
1.220**
(3.78)
2.002
real min
wage 𝜹𝒊
0.376***
(6.39)
0.198***
(7.94)
0.0551
(0.62)
0.130**
(4.88)
-0.211
(-3.55)
0.0181
(0.28)
0.330
real min wage
𝜼𝒊
0.472**
(3.79)
0.488***
(7.01)
0.638***
(6.09)
0.392***
(10.75)
0.278**
(4.29)
0.361***
(6.86)
0.532***
(-0.22)
0.206
(1.94)
0.596**
(5.46)
-0.414
(-3.17)
0.440
(2.28)
-0.485
(-4.46)
-0.0442
(2.16)
1.887***
(9.01)
1.106**
(5.14)
0.938*
(3.65)
0.820
(2.16)
0.298
(1.39)
0.704**
(1.90)
0.585***
(7.39)
0.603***
(17.14)
-0.487
(-11.32)
0.311**
(5.25)
-0.863
(-16.88)
-0.228
(10.53)
0.616***
(6.49)
0.416***
(5.63)
0.783***
(6.89)
0.797***
(7.51)
0.321
(2.21)
0.626***
(-0.53)
0.452**
(5.76)
0.229
(1.95)
0.407*
(3.61)
-0.682
(-3.88)
-0.0760
(4.30)
0.218
(1.41)
0.680*
(2.94)
0.483
(2.17)
1.216*
(3.51)
0.707*
(-8.24)
0.0358
(0.79)
0.0330
(0.82)
0.118*
(3.04)
-0.622
(-9.15)
-0.259
(9.35)
0.637***
(12.71)
0.649***
(11.99)
0.621***
(8.92)
0.501***
(12.43)
0.210*
service activities and private
household with employed
persons
(-0.74)
(3.49)
(-8.75)
0.320**
1.148***
0.347***
(5.59)
(10.16)
(13.39)
Notes: 1 Numbers in parentheses are t statistics
2
* represents p<0.05, ** represents p<0.01, and *** represents p<0.001.
19 GDP
73
(2.81)
0.492***
(9.86)
Table 16. The impact of institutional wage bias on labor-output coefficients, 2000-2008
Code
Activity
1
3
5
Agriculture, forestry and fishing
Manufacturing
Infrastructure services
Percentage change in:
unit labor use due to institutional wage
overall unit labor
bias (%)
use (%)
-1.05
-4.27
-0.11
-3.90
-1.37
-4.66
74
Table 17. Labor-output coefficients in 2003 by firm type
Code
Activity
1
3
4
5
9
State
0.010
0.035
0.037
0.010
0.030
Agriculture
Manufacturing
Construction
Infrastructure services
GDP
75
Labor-output coefficient for:
Non-state
Overall
0.507
0.345
0.076
0.064
0.083
0.059
0.162
0.083
0.190
0.122
Table 18. Impact of state investment bias on labor demand from 2003 to 20081
Code
Sector
3 Manufacturing
Overall
State
Non-state
4 Construction
Overall
State
Non-state
5 Infrastructure services
Overall
State
Non-state
Growth rate of output
shares gSi
relative unit labor use
aLi/aL
GDP share
Si
Impact on labor demand
(%)
3.65
-12.49
7.83
0.59
0.04
0.64
0.24
0.04
0.19
0.51
-0.02
0.98
0.55
-12.44
8.71
0.61
0.39
0.72
0.09
0.03
0.06
0.03
-0.15
0.37
1.03
-9.81
8.51
0.75
0.10
1.13
0.24
0.09
0.15
0.18
-0.09
1.46
Note: 1 The values for the overall sectors in this Table are not exactly the same as those in Table 6, because different time periods are
considered. In Table 6, the time period is 2000-2008, and the time period for this Table is 2003-2008.
76
List of Figures
Figure 1. GDP and employment growth rates in Vietnam, 1986-2008 ........................................ 78
Figure 2. Employment structure in Vietnam from 2000 to 2008 .................................................. 79
Figure 3. Labor productivity across sectors in Vietnam from 1986 to 2008 ................................ 80
Figure 4. Input costs in Vietnam from 2000 to 2008 .................................................................... 81
Figure 5. Relative input costs and productivity levels in Vietnam from 2000 to 2008 ................ 82
Figure 6. Input costs and productivity levels in the agricultural sector in Vietnam from 2000 to
2008............................................................................................................................................... 83
Figure 7. Input costs and productivity levels in the manufacturing sector in Vietnam from 2000
to 2008 .......................................................................................................................................... 84
Figure 8. Input costs and productivity levels in the service sector in Vietnam from 2000 to 2008
....................................................................................................................................................... 85
Figure 9. Nominal GDP, minimum wages, and inflation in Vietnam .......................................... 86
Figure 10. Relative incremental labor to incremental capital ratios in the agricultural sector for
state firms and overall from 2004 to 2008 .................................................................................... 87
Figure 11. Relative incremental labor to incremental capital ratios in the manufacturing sector for
state firms and overall from 2004 to 2008 .................................................................................... 88
Figure 12. Relative incremental labor to incremental capital ratios in the construction sector for
state firms and overall from 2004 to 2008 .................................................................................... 89
Figure 13. Relative incremental labor to incremental capital ratios in the infrastructure service
sector for state firms and overall from 2004 to 2008 .................................................................... 90
Figure 14. Relative incremental labor to incremental capital ratios in the entire economy for state
firms and overall from 2004 to 2008 ............................................................................................ 91
77
12
10
Growth rates (%)
8
6
4
2
0
1986
1990
-2
1994
1998
2002
2006
Year
GDP growth
in Agriculture
Employment growth
Labor force growth
in Industry and services
Figure 1. GDP and employment growth rates in Vietnam, 1986-2008
Source: WDI, 2010
78
70
Employment structure (%)
60
50
40
30
20
10
0
2000
2001
2002
Agriculture
2003
2004
Year
Fishing
2005
Industry
2006
2007
Services
Figure 2. Employment structure in Vietnam from 2000 to 2008
Source: GSO, 2010
79
2008
3000
2500
2000
1500
1000
500
0
Value added per worker based on labor force
Agriculture
Industry
Services
Figure 3. Labor productivity across sectors in Vietnam from 1986 to 2008
Source: WDI, 2010
80
3.50
3.00
2.50
2.00
Index
1.50
1.00
0.50
0.00
2000
-0.50
2001
2002
2003
2004
2005
2006
2007
2008
-1.00
-1.50
Year
real wage
real rent to capital
nominal min wage
real state wage
real lending rate
Figure 4. Input costs in Vietnam from 2000 to 2008
Sources: Authors’ calculations based on data from GSO (2009) and IMF (2011)
81
1.20
1.10
Index
1.00
0.90
0.80
0.70
0.60
0.50
2000
2001
capital-output
2002
2003
2004
Year
labor-output
2005
labor-capital
2006
2007
2008
wage-rental
Figure 5. Relative input costs and productivity levels in Vietnam from 2000 to 2008
Source: Authors’ calculations based on GSO (2009)
82
1.50
1.40
1.30
1.20
Index
1.10
1.00
0.90
0.80
0.70
0.60
0.50
2000
2001
2002
2003
2004
Year
2005
real relative wage
real rent to capital
labor-output
wage-rental
2006
2007
2008
capital-output
Figure 6. Input costs and productivity levels in the agricultural sector in Vietnam from
2000 to 2008
Source: Authors’ calculations based on GSO (2009)
83
1.50
1.40
1.30
1.20
Index
1.10
1.00
0.90
0.80
0.70
0.60
0.50
2000
2001
2002
2003
2004
Year
real relative wage
real rent to capital
labor-output
wage-rental
2005
2006
2007
2008
capital-output
Figure 7. Input costs and productivity levels in the manufacturing sector in Vietnam from
2000 to 2008
Source: Authors’ calculations based on GSO (2009)
84
1.90
1.70
Index
1.50
1.30
1.10
0.90
0.70
0.50
2000
2001
2002
2003
2004
Year
real relative wage
capital-output
wage-rental
2005
2006
2007
2008
real rent to capital
labor-output
Figure 8. Input costs and productivity levels in the service sector in Vietnam from 2000 to
2008
Source: Authors’ calculations based on GSO (2009)
85
400
350
300
Indices
250
200
150
100
50
0
2000
2001
2002
CPI
2003
2004
Year
GDP current
2005
2006
2007
Minimum Wage
Figure 9. Nominal GDP, minimum wages, and inflation in Vietnam
Source: GSO, 2010
86
2008
Relative labor to capital incremental ratios in the
agriculture sector
0.000
2004
2005
2006
2007
2008
-0.005
-0.010
-0.015
-0.020
-0.025
Year
state
overall
Figure 10. Relative incremental labor to incremental capital ratios in the agricultural
sector for state firms and overall from 2004 to 2008
87
Relative labor to capital incremental ratios in the
manufacturing sector
0.015
0.010
0.005
0.000
2004
2005
2006
2007
2008
-0.005
-0.010
Year
state
overall
Figure 11. Relative incremental labor to incremental capital ratios in the manufacturing
sector for state firms and overall from 2004 to 2008
88
Relative labor to capital incremental ratios in the
construction sector
0.030
0.025
0.020
0.015
0.010
0.005
0.000
2004
-0.005
2005
2006
2007
2008
-0.010
-0.015
-0.020
Year
state
overall
Figure 12. Relative incremental labor to incremental capital ratios in the construction
sector for state firms and overall from 2004 to 2008
89
Relative labor to capital incremental ratios in the
infrastructure service sector
0.007
0.006
0.005
0.004
0.003
0.002
0.001
0.000
2004
-0.001
2005
2006
-0.002
2007
2008
Year
state
overall
Figure 13. Relative incremental labor to incremental capital ratios in the infrastructure
service sector for state firms and overall from 2004 to 2008
90
Relative labor to capital incremental ratios in the
entire economy
0.012
0.010
0.008
0.006
0.004
0.002
0.000
2004
2005
2006
-0.002
2007
2008
Year
state
overall
Figure 14. Relative incremental labor to incremental capital ratios in the entire economy
for state firms and overall from 2004 to 2008
91
References
Abbott, P., K. Boys, P. L. Huong and F. Tarp. 2008. Trade and Development in Vietnam:
Exploring Investment Linkages. Hanoi, Vietnam.
Abbott, P., C. Wu, and F. Tarp. 2010. Transmission of World Prices to the Domestic Market in
Vietnam. Hanoi, Vietnam.
Boys, K.A. 2008. "Investment, Trade, and Economic Development: Lesson from Vietnam." PhD
Thesis, Purdue University, West Lafayette.
Cantore, C., M.A. León-Ledesma, P. McAdam, and A. Willman. 2009. "Shocking Stuff:
Technology, Hours, and Factor Substitution." European Central Bank, Frankfurt.
Collins, S.M., B.P. Bosworth, and D. Rodrik. 1996. "Economic Growth in East Asia:
Accumulation versus Assimilation." Brookings Papers on Economic Activity
1996(2):135-203.
David, P.A., and van de Klundert, T. 1965. "Biased Efficiency Growth and Capital-Labor
Substitution in the U.S., 1899-1960." The American Economic Review 55(3):357-394.
Diamond, P.A., and D. McFadden (1965) Identification of the Elasticity of Substitution and the
bias of Technical Change: An Impossibility Theorem, University of California Berkeley
Working Paper No. 62.
Freeman, C. 1995. "The ‘National System of Innovation’ in historical perspective." Cambridge
Journal of Economics 19(1):5-24.
General Statistical Office of Vietnam (GSO). 2009. Investment, Trade and Price Data for
Vietnam. General Statistical Office of Vietnam, Hanoi.
---. 1998-2006, 2010. Statistical Yearbook of Vietnam. General Statistical Office of Vietnam,
Hanoi.
---. 1998-2006. Vietnam Household Living Standard Survey (VHLSS). General Statistical Office
of Vietnam, Hanoi.
Institute of Labour Science and Social Affairs (ILSSA). 2008. Forecasting the Relation between
Investment, Economic Growth and Employment and Income of Employees. Institute of
Labour Science and Social Affairs, Hanoi.
International Monetary Fund (IMF). 2010. International Finanacial Statistics. International
Monetary Fund, Washington DC.
Johnston, B.F., and J.W. Mellor. 1961. "The Role of Agriculture in Economic Development."
The American Economic Review 51(4):566-593.
Kaüger, J., U. Cantner, and H. Hanusch. 2000. "Total factor productivity, the east Asian miracle,
and the world production frontier." Review of World Economics 136(1):111-136.
92
Kim, J.-I., and J. L. Lawrence. 1994. "The Sources of Economic Growth of the East Asian
Newly Industrialized Countries." Journal of the Japanese and International Economies
8(3):235-271.
Krugman, P. 1994. "The Myth of Asia's Miracle." Foreign Affairs 73(6):62-78.
Lewis, A. 1954. "Economic Development with Unlimited Supplies of Labour." The Manchester
School 22(2):139-191.
Ministry of Labour, Invalids and Social Affaris (MOLISA) and Institute of Labour Science and
Social Affairs (ILSSA). 2009. Vietnam Labour and Social Trends 2009: In International
Economic Integration Context. Hanoi, Vietnam.
Nelson, R.R., and H. Pack. 1999. "The Asian Miracle and Modern Growth Theory." The
Economic Journal 109(457):416-436.
Pack, H., and J.M. Page Jr. 1994. "Accumulation, exports, and growth in the high-performing
Asian economies." Carnegie-Rochester Conference Series on Public Policy 40:199-235.
Rodrik, D. 1997. "TFPG Controversies, Institutions, and Economic Performance in East Asia."
National Bureau of Economic Research Working Paper No. 5914.
Sato, R. 1970. "The Estimation of Biased Technical Progress and the Production Function."
International Economic Review 11(2):179-208.
World Bank. 1993. The East Asian Miracle: Economic Growth and Public Policy: Summary.
New York: Oxford University Press.
---. 2008. Social Protection: Vietnam Development Report 2008. Washington DC: World Bank.
---. 2009. Taking Stock: An Update on Vietnam's Recent Economic Developments. Washington
DC: World Bank.
--- .2010. World Development Indicators (WDI). Washington DC, World Bank. Available online
at: http://data.worldbank.org/indicator
Young, A. 1995. "The Tyranny of Numbers: Confronting the Statistical Realities of the East
Asian Growth Experience." The Quarterly Journal of Economics 110(3):641-680.
93
Appendix I. Concordance for aggregations from eighteen economic sectors to nine
economic sectors
18-sector
code
1
8-sector
code
1
Fishing
2
1
Mining and quarrying
Manufacturing
3
4
2
3
Electricity, gas and water supply
Construction
Wholesale and retail trade; repair of motor vehicles,
motor cycles and personal and household goods
Hotels, restaurant
Transport, storage and communications
Financial intermediation
Scientific activities and technology
Real estate, renting and business activities
Public administration and defense; compulsory
social security
Education and training
Health and social work
Recreational, cultural and sporting activities
Activities of Party and of membership organizations
Community, social and personal service activities
and private household with employed persons
5
6
2
4
7
8
9
10
11
12
5
5
5
6
6
6
13
14
15
16
17
7
8
8
9
7
Public services
Education and
health
Other services
Public services
18
9
Other services
Description of 18 sectors
Agriculture and forestry
94
Description of 8
sectors
Agriculture,
forestry and
fishing
Energy and
natural resources
Manufacturing
Energy and
natural resources
Construction
Infrastructure
services
Professional
services
Appendix II. Concordance for Selected Aggregations of 2000 SAM
Description of 112 sectors
Paddy (all kinds)
Raw rubber
Coffee beans
Sugarcane
Tea
Other crops
Pig (All kinds)
Cow (All kinds)
Poultry
Other livestock and poultry
Irrigation service
112-sector
code
1
2
3
4
5
6
7
8
9
10
11
18-sector
code
1
1
1
1
1
1
1
1
1
1
1
Other agricultural services
Forestry
Fishery
Fish-Farming
Coal
Metallic ore
Stone
Sand, gravel
12
13
14
15
16
17
18
19
1
1
2
2
3
3
3
3
Other none-metallic minerals
Crude oil, natural gas (except
exploration)
Processed, preserved meat and byproducts
Processed vegetable, and animals oils
and fats
20
3
21
3
22
4
23
4
Milk, butter and other dairy products
Cakes, jams, candy, coca, chocolate
products
Processed and preserved fruits and
vegetables
24
4
25
4
26
4
Alcohol, beer and liquors
27
4
95
Description of 18
sectors
Agriculture and
forestry
Fishing
Mining and quarrying
Manufacturing
Appendix II continued
Description of 112 sectors
112-sector
code
28
18-sector
code
4
Non-alcohol water and soft drinks
Sugar, refined
Coffee, processed
Tea, processed
29
30
31
32
4
4
4
4
Cigarettes and other tobacco products
33
4
Processed seafood and by-products
Rice, processed
34
35
4
4
36
4
Glass and glass products
37
4
Ceramics and by-products
Bricks, tiles
Cement
Concrete, mortar and other cement
products
38
39
40
4
4
4
41
4
Other building materials
Paper pulp and paper products and byproducts
42
4
43
4
Processed wood and wood products
44
4
Basic organic chemicals
45
4
Basic inorganic chemicals
Chemical fertilizer
Fertilizer
Pesticides
Veterinary medicine
Health medicine
46
47
48
49
50
51
4
4
4
4
4
4
Beer and liquors
Description of 18
sectors
Other food manufactures
96
Manufacturing
Appendix II continued
Description of 112 sectors
112-sector
code
18-sector
code
Processed rubber and by-products
Soap, detergents
52
53
4
4
Perfumes and other toilet preparations
54
4
Plastic (including semi-plastic products)
Other plastic products
Paint
55
56
57
4
4
4
Ink, varnish and other painting materials
58
4
Other chemical products
59
4
Health instrument and apparatus
60
4
Precise and optics equipment, meter (all
kinds)
61
4
Home appliances and its spare parts
62
4
Motor vehicles, motor bikes and spare parts
63
4
Bicycles and spare parts
64
4
General-purpose machinery
65
4
Other general - purpose machinery
66
4
Other special-purpose machinery
Automobiles
Other transport means
Electrical machinery
67
68
69
70
4
4
4
4
Other electrical machinery and equipment
71
4
Machinery used for broadcasting, television
and information activities
72
4
97
Description of 18
sectors
Manufacturing
Appendix II continued
Description of 112 sectors
112sector
code
18sector
code
73
4
74
4
Weaving of cloths (all kinds)
75
4
Fibber, thread (all kinds)
Ready-made clothes, sheets (all
kinds)
Carpets
76
4
77
78
4
4
Non-ferrous metals and
products
Ferrous metals and products
(except machinery equipment)
Weaving and embroidery of
textile-based goods (except
carpets)
79
4
Products of leather tanneries
Leather goods
Animal feeds
80
81
82
4
4
4
Products of printing activities
83
4
Products of publishing house
Other physical goods
Gasoline, lubricants (already
refined)
Electricity, gas
Water
Civil construction
Other construction
Trade
84
85
4
4
86
87
88
89
90
91
4
5
5
6
6
7
Repair of small transport
means, motorbikes and personal
household appliances
Hotels
Restaurants
Description of 18 sectors
Manufacturing
Electricity, gas and water supply
Construction
Wholesale and retail trade; repair of
motor vehicles, motor cycles and
personal and household goods
92
93
94
7
8
8
98
Hotels, restaurant
Appendix II continued
Description of 112 sectors
112sector
code
95
96
18sector
code
9
9
Water transport services
Air transport services
97
98
9
9
Communication services
Tourism
Banking, credit, treasury
Lottery
Insurance
99
100
101
102
103
9
9
10
10
10
Science and technology
Real estate
Real estate business and
consultancy services
104
105
11
12
106
12
Transportation
Railway transport services
State management, defense
and compulsory social
security
Education and training
Health care, social relief
107
108
13
14
109
110
15
16
111
17
112
18
Culture and sport
Association
Other services
99
Description of 18 sectors
Transport, storage and communications
Financial intermediation
Scientific activities and technology
Real estate, renting and business
activities
Public administration and defense;
compulsory social security
Education and training
Health and social work
Recreational, cultural and sporting
activities
Activities of Party and of membership
organizations
Community, social and personal service
activities and private household with
employed persons