Professional Documents
Culture Documents
net/publication/228957392
CITATIONS READS
31 154
2 authors:
Some of the authors of this publication are also working on these related projects:
PRICE SETTING BEHAVIOUR AND PRODUCT MARKET INTEGRATION IN DEVELOPING COUNTRIES View project
All content following this page was uploaded by Stephen S Golub on 31 May 2014.
Revised
April 2002
Edwards
School of Economics
University of Cape Town
South Africa
Email: ledwards@commerce.uct.ac.za
Golub
Department of Economics
Swarthmore College
USA
Email: sgolub1@swarthmore.edu
We would like to thank Matthew Stern and Chris Loewald for inviting us to write this
paper and their support and encouragement. All the views expressed are personal.
i
Summary
With the end of Apartheid and the new government, South Africa’s trade policy
regime shifted in the 1990s from one of import substitution towards one of export
orientation. This process has taken the form of an accelerated tariff liberalization
program since 1994, the adoption of export orientation policies that ranged from direct
support (GEIS) to marketing related assistance, and a macroeconomic strategy (GEAR)
that was explicitly expected to transform South Africa into a “competitive, outward
orientated economy” (GEAR, 1996).
The successes of these policies in South Africa have been mixed. While export
growth has risen, this has not been significantly greater than in other dynamic emerging
economies. Import penetration, particularly within labor intensive sectors, has also risen,
raising questions about the cost competitiveness of domestic labor. Finally, employment
has continued to fall, despite the recovery in output growth. Although the evidence
suggests that this is not due to trade (Edwards, 2000, Fedderke et al., 1999), the public
perception of the correlation between tariff liberalization and continued losses in
employment make this a contentious debate.
This paper extends this debate and focuses on the changing cost competitiveness
of South African labor and the effect this has had on South African export performance.
In doing so the paper develops Golub’s (2000) original analysis of South African
competitiveness in manufacturing in two main directions. First, we have extended the
analysis of relative unit labor costs, productivity and wages, and their effects on trade
flows to the industry level. Second, we have extended the productivity calculations to
total factor productivity, rather than limiting them to labor productivity. Total factor
productivity is a composite indicator capturing the efficiency of all factor inputs,
including capital, in the production process. Because labor productivity depends on
capital input, labor productivity alone can be a misleading indicator of efficiency to the
extent that rising labor productivity could be due to capital-labor substitution rather than
total factor productivity growth.
ii
In the 1980s, South African labor and total factor productivity both grew at about
1 percent annually, as the capital-labor ratio changed little. In the 1990s, total factor
productivity stagnated despite modest annual growth of labor productivity of about 2
percent, as the South African capital-labor ratio increased. While most of the increased
capital-labor ratio reflects capital accumulation, some of it is also attributable to labor
shedding. These findings are disconcerting insofar as sustainable long-run output and
employment growth depends on raising productive efficiency — total factor productivity
— rather than through capital-labor substitution and labor shedding.
Relative to other countries in our sample, South African labor and total factor
productivity have both declined, but relative TFP generally declined at a lesser rate than
relative LP, as other countries experienced more capital-labor substitution than South
Africa. In short, South Africa’s low relative growth of labor productivity reflects both
iii
slow growth of TFP and a lesser rate of capital accumulation than other newly
industrializing countries.
For the most part, labor and total factor productivity rankings by sector are fairly
similar, so that the choice of productivity measure has little impact on the measurement
of comparative advantage, notwithstanding the overall differences in growth rates of LP
and TFP in the 1990s.
Relative Unit Labor Costs. We then turn to calculations of relative unit labor
costs (RULC) which reflect the labor cost of producing one unit of output in South Africa
relative to other countries measured in a common currency. The overall levels of RULC
indicate absolute advantage whereas sectoral rankings are measures of comparative
advantage. A South African sector is “competitive” in absolute terms if its RULC is less
than 1.0. It is competitive in relative terms if the sector has a favorable ranking relative
to other sectors.
Relative unit labor costs can be decomposed into relative productivity and relative
wages converted into a common currency. Thus South Africa’s competitiveness vis-à-vis
other countries could improve when some combination of the following conditions hold :
1) labor productivity in South Africa rises relative to other countries, 2) South Africa’s
relative wages fall, or 3) the rand depreciates. As South African cost competitiveness
improves, exports are predicted to increase and imports decline for the affected sectors.
particularly large declines in RULC include industrial chemicals, non-ferrous metals and
professional equipment.
The structure of South Africa’s comparative advantage does not differ markedly
between developed versus less developed countries. Further, no clear pattern emerges as
to the characteristics of South African comparative advantage. Those sectors with the
highest levels and improvements of competitiveness include a variety of labor intensive,
capital-intensive, and natural resource-intensive goods. Few sectors have shown a
consistent improvement or deterioration in competitiveness over all periods, indicating
that the structure of competitiveness has changed substantially since the 1970s. There is
also no consistent relationship between these changes and sector characteristics such as
factor content, skill intensity, size, and trade orientation.
Relative Unit Labor Costs and Exports. Both simple charts and sophisticated
econometrics show that South African exports and to a lesser extent trade balances
respond strongly to relative unit labor costs.
competitive and hence export less. A clear negative relationship appears for all time
periods, including the 1990-98 period shown in Figure 1. In addition, some long-run
elasticities derived from the estimated export demand functions vis-à-vis developing and
developed countries are presented in Table 1.
Figure 1
South African Sectoral Relative Unit Labor Costs and Exports to Output Ratios
Exports/Output, 1990-98
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0.60 0.80 1.00 1.20 1.40 1.60 1.80
RULC average
Table 1
South African real exports as a function of relative unit labor costs and other variables,
by destination to developing and developed countries
Notes: Developed countries consist of: Canada, France, Italy, Japan, Netherlands, Spain, UK, USA, Hong
Kong.
Developing countries consist of: India, Korea, Singapore, Chile, Hungary, Poland, Turkey, Kenya, and
Zimbabwe.
vi
We also decompose the effects of relative unit labor costs into relative
productivity and relative wages. In general we find that relative wages are more
important determinants of trade than relative productivity. A 1 % rise in relative
productivity vis-à-vis developing and developed countries increases total real exports by
1.1 % and 1.3 %, respectively. In contrast, a 1 % rise in wages vis-à-vis developing and
developed countries reduces total exports by 1.6 % and 1.7 %, respectively. Relative
wages can in turn be decomposed into the effects of exchange rate changes and wages in
local currency. In a number of country specific estimations we find that the exchange
rate is also an important determinant of export performance.
Relative unit labor costs generally explain the 1990s less well than the earlier
periods. This may be due to the huge structural changes of the last decade. Overall,
however, our results are strongly supportive of the theoretical presumption that RULC
matters importantly for trade flows.
Fifth, the improvements in competitiveness during the 1990s have arisen largely
from the depreciation of the exchange rate and not from wage moderation and
productivity improvements. The process of improving competitiveness and export
growth through the depreciation of the exchange rate cannot be sustained indefinitely
without risking an upsurge in inflation.
Sixth, there are several areas where public policy can play a role in encouraging
long term export growth, while stopping short of “picking winners”. One such area is
education. Although the rewards may only be realized in the long run, improvements in
education can substantially raise the productivity of South Africa’s labor force and
through this export performance (Fedderke 2001).
This lowers the cost of access to foreign technology as well as induces productivity gains
through increased international competition. Work by Jonsson and Subramanian (2000)
has shown a positive relationship between improved TFP growth and trade liberalization.
The necessity of liberalizing trade has been given further impetus by the significant real
depreciation of the currency during late 2001. This will have substantially raised the cost
of imported machinery.
I. Issues
With the end of Apartheid and the new government, South Africa’s trade policy
regime has shifted in the 1990s from one of import substitution towards one of export
orientation. This process has taken the form of an accelerated tariff liberalization
program since 1994, the adoption of export orientation policies that ranged from direct
support (GEIS) to marketing related assistance, and a macroeconomic strategy (GEAR)
that was explicitly expected to transform South Africa into a “competitive, outward
orientated economy” (GEAR, 1996).
The successes of these policies in South Africa have been mixed. While export
growth has risen, this has not been significantly greater than other dynamic emerging
economies. Import penetration, particularly within labor intensive sectors, has also risen,
raising questions about the cost competitiveness of domestic labor. Finally, employment
has continued to fall, despite the recovery in output growth. Although the evidence
suggests that this is not due to trade (Edwards, 2000, Fedderke et al., 1999), the public
perception of the correlation between tariff liberalization and continued losses in
employment make this a contentious debate. This paper extends this debate and focuses
on the changing cost competitiveness of South African labor and the effect this has had
on South African export performance.
As the South African economy liberalizes its trading regime, the ability of
domestic producers to effectively participate in the global economy will increasingly be
defined by their cost or price competitiveness vis-à-vis the rest of the world. This
importance placed on international competitiveness has given rise to a wide literature
covering a diverse range of competitiveness measures. In South Africa these include real
effective exchange rates (IMF, 1998, Kahn, 1998, Tsikata, 1999, and Golub, 2000), unit
labor costs (Nordas, 1996, and Golub, 2000), revealed comparative advantage measures
(Edwards, Mlangeni, Van Seventer, 2000, and Valentine and Krasnik, 2000) and other
composite measures. Two conclusions generally arise from these analyses. First, South
2
Secondly, the bulk of the work has treated manufacturing as an aggregated sector.
This tends to hide much of the variation in competitiveness at the sectoral level.
International trade theory predicts non-uniform effects on the competitiveness of
manufacturing arising from trade liberalization. 1 Thus, it is not evident which sectors are
driving the improvements in relative labor cost competitiveness shown by Golub (2000)
during the 1990s. Analyses of unit labor costs in South Africa at the sectoral level are
unfortunately not widely available. One sectoral level study by Nordas (1996) compares
relative unit labor costs between SA and USA and finds that South Africa is competitive
in medium wage, low technology and resource intensive industries. No link to the
structure of bilateral trade flows between SA and USA was made. Further, the analysis
uses data from the early 1990s making extrapolation to recent years tenuous.
1
If all trade is intra-industry trade, then a uniform impact is possible.
3
In summary, this project extends the work by Golub (2000) on South African
international competitiveness and unit labor costs in two directions. First, instead of
aggregate manufacturing, this paper analyzes international competitiveness at the sectoral
level. Second, the analysis of productivity is extended to total factor productivity rather
than only labor productivity.
reviews the literature on econometric estimation of export functions for South Africa.
Section VIII discusses the econometric methodology. Section IX looks at the cross-
sectional evidence and section X examines the time-series evidence. Section XI
concludes. An Appendix discusses the sources and methods of our data analysis.
Unit labor cost is the cost of labor per unit of output, i.e. the ratio of wages to
productivity. In addition to its clear intuitive appeal, relative unit labor cost is the key relative
price in a Ricardian model of trade. Furthermore, in a world where capital is mobile and
production is footloose between countries, it is the relative price of non-tradable inputs, notably
labor, rather than outputs that matters. 2 Further, technological advancements have enabled the
fragmentation of the production process into smaller distinct steps, the physical location of which
can be spread around the globe without losing control of the production process (Gourevitch et
al. 2000). In this milieu labor costs have become an important determinant in the location of
these production processes. For example, much of the assembly of consumer electronics has
shifted to low wage, labor intensive economies within the South East Asian region.
It is therefore of interest to compare both levels and rates of change of labor costs and
labor productivity between countries. For these reasons and others, Turner and Van’t Dack
(1992) and Turner and Golub’s (1997) surveys of the literature conclude that relative unit labor
costs in manufacturing are the best single indicator of competitiveness. Where data are
2
Jones (1980) analyzes the implications of footloose factors in a Ricardian model. See also Caves,
Frankel, Jones (1999, Ch. 9).
5
available, Hinkle and Nsengiyumba (1999) also endorse the use of unit labor costs, both for
analysis of levels and rates of change of competitiveness.
At a disaggregated level, the Ricardian model provides an appealing framework for the
analysis of trade flows (see Carlin et al (2001), Golub (1994) and Golub and Hsieh (2000) for
further discussion). This model provides an integrated framework for understanding the macro-
and microeconomic factors determining trade flows, as most elegantly shown in Dornbusch,
Fischer and Samuelson (1977) (DFS). The basic idea is that relative unit labor cost is influenced
both by sector-specific variables (productivity and wages) as well as the real exchange rate. 3 On
a micro level, the Ricardian model has both advantages and disadvantages compared to the
Heckscher-Ohlin-Samuelson (HOS) model, in which comparative advantage is derived from
factor endowments. The main advantage is that the Ricardian model allows for technological
differences between countries, which in practice seem very significant. There are large and
persistent gaps in labor and total factor productivity by industry across countries (e.g. Harrigan
1999). Moreover, the HOS model has received lukewarm empirical support (Bowen, Leamer,
and Sveikauskas, 1987). It has recently been shown that a key reason for the failure of the HOS
model is the failure to incorporate international differences in technology and that when such
differences are allowed for the HOS model improves (Trefler 1995, Davis and Weinstein 1998).
The main disadvantage of the Ricardian model is that it implies that countries specialize
completely in tradable goods production. In practice, import-competing industries contract but
rarely disappear completely in the face of foreign competition. To allow for incomplete
specialization in a Ricardian context, one would have to introduce other considerations, such as
differentiated products. Also, unit labor cost may be an imperfect gauge of competitiveness if
quality differences are not measured accurately or if labor is not the only non-tradable factor of
production.
Let ai represent the unit labor requirement (the inverse of productivity), for sector i:
3
Most expositions of the Ricardian model assume perfectly competitive labor markets and hence a uniform
wage rate across sectors. But this assumption is easily relaxed in empirical work.
6
Li ,
(1) ai =
Qi
where Q is value-added, and L is labor employment. Marginal productivity and hence ai are
assumed to be constant with respect to variations in Li.
Let w denote the wage and e the exchange rate between home and foreign currencies. If
labor is the only factor of production (or that other factor costs do not differ across countries),
home (South African) average costs of production are equal to unit labor costs wi.ai. Expressed
in domestic currency, foreign unit labor cost is e.wi*.ai*. International competitiveness in sector
i then depends on relative unit labor cost (RULC) ci :
aiwi
(2) ci = .
ai * wi * e
ai . wi .
(3) ci =
ai * wi * e
Equation (3) illustrates the decomposition of relative unit labor costs into relative productivity
and relative wages converted into a common currency. Thus South Africa’s competitiveness vis-
à-vis other countries could improve when some combination of the following conditions hold : 1)
labor productivity in South Africa rises relative to other countries, 2) South Africa’s relative
wages fall, or 3) the rand depreciates. As South African cost competitiveness improves,
exports are predicted to increase and imports decline for the affected sectors. 4 Notice that the
exchange rate affects all sectors simultaneously, while the competitiveness of each individual
sector also depends on wages and productivity in that sector vis-à-vis other countries.
4
Strictly speaking, in a Ricardian setting, countries will specialize completely in those sectors in which
they have comparative advantage, and an improvement in RULC will only affect trade flows if it causes
RULC to fall from above 1.0 to below 1.0. In practice, however, because of product differentiation and
other discrepancies from the model, countries do not specialize completely and the response to changes in
RULC is likely to be smoother. See Golub and Hsieh (2000) for further discussion.
7
The home country will have an absolute advantage in good i when ci > 1, i.e., South
African unit labor costs are below those of its trading partners, or equivalently when relative
South African productivity exceeds relative the relative South African relative wage, measured
in a common currency, i.e.,
ai * > wi
(4) .
ai w *i e
This model implies both cross-section and time-series relations between RULC and trade
flows. Declines in RULC over time are predicted to raise export volumes and lower import
volumes. We can also assess South African comparative advantage by comparing RULC across
industries. Those industries with lowest RULC are predicted to have the largest export shares
and net exports.
Our empirical analysis will examine both levels and changes over time in relative
productivity, relative wages, and unit labor costs and their effects on the composition and time
pattern of trade flows.
The Ricardian model, sketched above, assumes a single factor of production, labor. Other
factors of production, particularly capital, are also important determinants of output and costs,
however. The focus on labor can be partially justified by the argument that labor is largely
immobile internationally but capital is mobile, so that labor costs are likely to diverge much
more across countries than costs of capital. Nonetheless, allowing for capital could still be
important in interpreting labor productivity, because labor productivity depends on capital input.
In particular, labor productivity could rise either because of an increase in efficiency or because
of capital-labor substitution. For example, if wages rise exogenously, due to union pressure,
firms will shed labor and the marginal and average product of labor (labor productivity) will rise,
but it would be misleading to interpret this a rise in “true” productivity. The rise in labor
8
productivity reflects an increase in the capital labor ratio. Total factor productivity which
captures the efficiency of all factor inputs, including capital, in the production process, however,
would be largely unchanged. It is therefore important to separate changes of labor productivity
into changes in total factor productivity and changes in capital/labor ratios.
α 1−α
(5) Qi = Ai Ki Li .
Total factor productivity (TFP) is A and can be computed as (suppressing the subscripts)
Q
(6) A= α 1−α
.
K L
Q AK α L1−α K α
(7) = = A
L L L
Thus, changes in labor productivity are due to changes either in TFP or in the capital-labor
ratio. Letting q be labor productivity and k be the capital-labor ratio, (7) can alternatively
be expressed as
∆q ∆A ∆k
(7’) = +α
q A k
That is, labor productivity growth equals TFP growth plus the growth of the capital-labor
ratio weighted by capital’s share of income. Using equation (7’) we can see that if labor
9
productivity growth exceeds TFP growth, the difference can be accounted for by rising
capital intensity of production.
Relative TFP of South Africa relative to another country is consequently, for any
given sector
Q L * K *
α 1− α
A
(8) = × ×
A* Q* L K
Thus, if the capital-labor ratio in South Africa (K/L) is less than the capital-labor ratio of
another country (K*/L*) then relative labor productivity is less than relative TFP (A/A*).
We will compute relative TFP for South Africa against a range of countries, as well as
relative capital labor ratios. We will then in turn look at how changes in capital labor ratios
break down into capital accumulation and employment changes.
Exports. As noted by others (e.g. Tsikata 1999, Golub 2000), South Africa’s exports
have shifted toward manufacturing and away from primary products, although the latter remain
important. The share of manufactures in total exports rose substantially from around 25 percent
in 1970 and 15 percent in 1980 to above 40 percent in the late 1990s (bottom row of Table 3.1).
Table 3.1 shows the commodity composition of South African manufactured exports, by three
and two-digit SIC category for selected years. It can be observed that the share of food products
(SIC 31) has declined substantially over the period 1970-1998 as a whole, while both base
metals (SIC 37) and fabricated metals and machinery (SIC 38) have risen sharply. But
10
expansion of the share of base metals occurred prior to 1990, while the share of fabricated
metals has expanded sharply since 1990. Particularly strong increases have occurred in
machinery and transport equipment. The shares of labor intensive sectors (textiles and clothing,
wood) have grown modestly.
Figure 3.1 shows export to output ratios for selected sectors. The labor-intensive sectors
show a surprisingly strong general upward trend that has continued into the 1990s, with a
particularly marked increase for leather products. Natural resource intensive sectors display a
mixed pattern. The export to output ratio for food and non-metallic mineral products has
declined, but this ratio has increased steadily for non-ferrous metals, and increased sharply and
then decreased again since 1985 for iron and steel. For the fabricated metals and machinery
industries, there has been a dramatic increase in the export to output ratios in the 1990s for all
sectors. The chemical industries also display an increase, particularly for industrial chemicals.
Generalizing, the post-Apartheid period has been characterized by rising openness, with
export to output ratios increasing in most sectors. The most rapid increases appear to be in the
higher technology and capital intensive machinery industries, but there has also been growth in
labor intensive sectors.
This greater openness has also led to greater import penetration, however. Despite the
rapid growth of manufactured exports shown in Table 1 and Figure 1, South Africa continues to
be a net importer of manufactured goods. It is therefore also instructive to examine net exports.
Figure 3.2 displays the ratio of net trade (exports minus imports) to output for the same
industries as Figure 3.1. Again, somewhat surprisingly, the labor-intensive sectors show a
marked upward long-term movement in net exports, in some cases moving from deficit to
surplus around 1990. For most of these sectors, particularly textiles, this upward trend went into
reverse in the 1990s. Leather products continue to show widening trade surpluses relative to
output, although they remain small in absolute terms. South Africa continues to have a revealed
comparative advantage in natural resource intensive sectors, particularly base metals, with large
net surpluses, although the iron and steel industry surplus has declined. There are sharply
11
increasing deficits in the machinery sectors in the 1990s, notwithstanding the rapid export
growth observed in Figure 3.1. This is not necessarily undesirable, however, as it may reflect
increased investment and improved future productivity. Indeed, the fabricated metals industry
shows continued declines in trade deficits while the transport equipment sectors shows moderate
deterioration. The chemical industry is also mixed, with a substantial decline in the deficit
position of industrial chemicals.
Labor Productivity. Our primary source of data is the United Nations Industrial
Development Organization (UNIDO) Industrial Statistics Database (INSTAT) at the 3-
digit SIC level. This database contains information on production, value added, labor
compensation and employment for a large group of countries. The advantage of this
database is the comprehensive country coverage. The main deficiencies of the data are
12
that UNIDO does relatively little to ensure international consistency and there are gaps in
the data. To fill these gaps and to provide a consistency check we also made extensive
use of the OECD’s Structural Analysis Industrial Database (STAN). The OECD strives
for greater consistency and completeness than UNIDO but only covers OECD countries,
which now includes Korea and Mexico. Due to the importance of comparing South
Africa to other developing countries, we chose to work with UNIDO. Where there was
overlap between the two data sets, however, we used STAN in some cases to adjust the
levels of the results obtained with the UNIDO data. We obtained South African data from
TIPS rather than using UNIDO as the South African INSTAT data was incomplete and
behaved peculiarly in the 1990s.
To compare levels of real outputs and labor compensation across economies, they
must be converted to a common currency. Productivity is calculated as manufacturing
value added per employee, deflated by the manufacturing value-added deflator and
converted to $U.S. at an equilibrium or purchasing power parity exchange (PPP) rate. In
principle, it would be desirable to have PPPs at a disaggregated level, as in Golub and
Hsieh (2000). Unfortunately, there are no available estimates of absolute PPP exchange
rates for manufacturing for South Africa. As an alternative, we used the average real
exchange rate over the sample period, using manufacturing value added deflators for
South Africa and its trading partners, as a rough indicator of the equilibrium real
exchange rate. For this reason as well as international differences in data construction for
value added, level comparisons must be made with great caution. Given the interest and
importance of level comparisons, we chose nonetheless to continue to report levels, as in
Golub (2000).
Price deflators by industry, necessary to obtain real value added for productivity
calculations, were obtained in two alternative ways. No value added deflators are
available at a disaggregated level in the UNIDO data set. However, output deflators can
be calculated by taking the ratio of nominal output and indexes of industrial production.
Alternatively, we also use the aggregate manufacturing value added deflator to deflate
each industry. Although the sectoral output deflators were sometimes erratic and
13
T
(9) K it = ∑ I i ,t − n (1 − δ) n −1
n= 1
For some of the developing countries, the INSTAT database did not contain
investment data. The sample of developing countries with adequate capital stock data
consists of Zimbabwe, Korea, Hong Kong, Singapore, India, Chile, and Turkey.
14
For all countries the capital share was assumed to be 0.4. Thus, the production
functions by sector are assumed identical across countries except for differences in the
TFP parameter A. Harrigan (1999) adopts a more general specification which allows for
sectoral and international differences in factor shares, but this our assumption of equal
factor shares is unlikely to significantly affect the results and allows for clearer
comparisons between countries.
IV.2. Results
Table 4.1 shows average annual TFP and LP growth by sector for South Africa
for 1980-1990 and 1991-1997, using the sector-specific output deflators to calculate real
value added. Table 4.2 presents the same information, except that it uses the aggregate
manufacturing value added deflator. Disconcertingly, the method of deflation makes an
important difference to the results. In almost all cases, the growth rate of TFP is higher
in the case of the output deflators, especially in the 1980s. The gap between TFP and LP
tends to be similar between the two methods, however. Also, both methods of deflation
yield similar rankings for most sectors. 5 In the earlier period, unweighted average
sectoral annual LP and TFP growth rates are quite similar to each other, using either
method: 3 percent using the output deflators and 1 percent using the aggregate
manufacturing value added deflator.
In the 1990s, LP growth exceeds TFP growth by the order of 2 percent, again
using either method. Drawing on equation (7’) this implies that there was little capital-
labor substitution in the 1980s, but that much of the growth of labor productivity in the
1990s reflected capital-labor substitution. There are, however, considerable variations
between sectors. Overall, however, sectoral rankings according to TFP are generally
quite similar to sectoral rankings according to LP, for both periods and both methods of
5
The Spearman’s rank correlation coefficients for the variables using different deflators exceed 0.5 during
the 1980s and 0.9 during the 1990s. The 2-tail test value at a 5 % significance level for 24 sectors is 0.4061.
15
deflation. 6 This suggests that competitiveness rankings are not likely to be greatly
affected by the choice of LP versus TFP as measures of productivity.
We also compared South African relative TFP and relative LP for aggregate
manufacturing vis-à-vis a number of countries. 7 These are shown in Figure 4.1. In most
cases, the movements over time of the two series are fairly similar, although South
African relative TFP tends to rise relative to South African relative LP beginning in the
late 1980s. Throughout the sample period, relative LP and TFP generally trend
downward, indicating that South African labor and total factor productivity tended to
grow more slowly than those of other countries over the sample period. For the
developed countries and some newly industrialized countries such as Korea and
Singapore, South African relative TFP exceeds relative LP, indicating that South African
capital-labor ratios were less than those of these other countries (see equation 8). For the
less developed countries in the sample (India and Zimbabwe), however, the reverse is
true, although the gap between relative LP and relative TFP dwindles over the sample
period in these cases. In short, South African capital-labor ratios are higher than those of
the least developed countries in the sample and lower than those of developed and newly
industrialized countries. This is confirmed in Figure 4.2 which displays relative capital-
labor ratios for selected countries. Figure 4.2 also shows that South African capital-labor
ratios declined relative to those of all these countries in the late1980s. In the 1990s,
6
For example, in the case of the output deflators, the Spearman’s rank correlation coefficient between
growth in TFP and LP exceeded 0.8 during both the 1980s and the 1990s.
7
These figures and tables use the common value added deflator as there is no aggregate manufacturing
deflator for output in the INSTAT database.
16
South African capital-labor ratios increased relative to developed countries, but not
relative to developing countries.
In summary, South African labor and total factor productivity tended to grow
more slowly than those of other newly industrializing countries over the sample period.
For the most part, labor and total factor productivity rankings by sector are fairly similar,
so that the choice of productivity measure has little impact on the measurement of South
African comparative advantage. Also, until the late 1980s, relative LP and TFP tended to
grow at similar rates. In the 1990s, however, total factor productivity in South Africa
grew more slowly than labor productivity due to capital-labor substitution. This capital
labor substitution underlies part of the lack of growth of employment in South Africa in
the post-Apartheid period.
8
Aggregate manufacturing in Table 4.3 differs from the sectoral average in Table 4.2, as the latter is an
unweighted average of sectoral productivities shown in Table 4.2. Aggregate manufacturing productivity
will reflect the different sizes of sectors and also includes some sectors excluded in Table 4.2 such as
petroleum refining and tobacco, where data appeared unreliable for some countries.
17
V.1 Data
Wages are defined as total wages and salaries divided by total employment,
converted to $U.S. at the market exchange rate. 9 As is accepted in the literature on
international labor productivity and unit labor cost comparisons, PPP exchange rates are
used for international productivity comparisons to eliminate the effects of exchange-rate
volatility on measures of real output, which should be invariant to such exchange-rate
fluctuations. But deviations of exchange rates from PPP do affect relative labor costs, so
it is appropriate to use the market exchange rate in converting wages. 10 Currency
depreciation consequently tends to improve international competitiveness by reducing
labor costs relative to labor productivity.
A problem, however, is that the TIPS and UNIDO data are not fully consistent.
TIPS data for labor compensation, like STAN, includes employer-paid fringe benefits,
whereas UNIDO data only includes wages and salaries. Moreover, there seems to be no
available breakdown for South Africa of labor compensation into wages and salaries
versus fringe benefits. 11 Consequently, the South African data was adjusted in an attempt
to make it consistent with the UNIDO definition. The adjustment factors were
determined by comparing UNIDO and STAN measures of labor compensation per
worker for those countries where the two data sets overlap. 12 Where there was no data in
STAN, we assumed that the discrepancy between wages and total compensation was
equal to the average for those countries where data are available.
9
The UNIDO INSTAT database used for these calculations generally defines wages and salaries as all payments in
cash to employees but excludes social insurance payments by employers. In contrast STAN adopts a more
comprehensive measure of labor costs which includes social insurance payments.
10
See Hooper and Larin (1989) for further justification of the use of PPP exchange rates for productivity
comparisons but market exchange rates for labor-cost comparisons.
11
Also, like STAN, the South African data is based on national-accounts measures of value added whereas
UNIDO data generally follows the census method. Census measures of value added tend to exceed those
of national accounts measures.
12
Those countries with overlap are the United States, United Kingdom, France, Italy, Japan, Canada, the
Netherlands, and Korea.
18
Relative unit labor costs (RULC) are the “bottom line” here so we focus on these.
Both levels and rates of change are of interest. As explained in section II, overall levels
of RULC indicate absolute advantage whereas sectoral rankings are measures of
comparative advantage. A South African sector is “competitive” in absolute terms if its
RULC is less than 1.0. It is competitive in relative terms if the sector has a favorable
ranking. Given the uncertainty about absolute PPPs and consistency of data concepts
across countries, considerable caution is in order in assessing the overall level of
competitiveness. Rates of change and sectoral comparisons are less subject to these
concerns since all of them are likely to be affected similarly by any such biases.
13
Weights for each country are determined by their total trade (exports plus imports) to the world.
19
they fell below 1.0 in almost all cases by the late 1990s. Tables 5.1.a and 5.1b show that
this improvement derives mostly from declining South African relative wages, rather than
improving South African relative productivity, until the late 1990s. In the 1995-98
period, however, most of the improvement derived from higher South African relative
labor productivity. In turn, much of the rise in South African labor productivity in the
1990s reflected higher capital-labor ratios rather than improved efficiency, as noted
earlier. Much of the decline in relative wages is due to depreciation of the exchange rate
and consequently this improvement is shared by almost all sectors. Labor intensive
sectors had large improvements, particularly for leather and footwear. Other sectors with
particularly large declines in RULC include industrial chemicals, non-ferrous metals and
professional equipment.
RULC competitiveness rankings (bottom half to Table 5.1c) show a decline for
some natural resource based sectors, including food, paper, and iron and steel. Non-
ferrous metal remains the most competitive industry for most of the period, however.
Metal products and machinery do not display a clear pattern with some sectors improving
in rank and others declining. Labor intensive sectors are also mixed with leather and
footwear improving sharply but others losing ground.
14
Tobacco and petroleum are known to be highly capital intensive industries with low employment. This
makes value added per worker highly volatile and unreliable.
20
in Spearman’s rank correlation coefficients of more than 0.5 for all periods. There are
some exceptions. South Africa is more competitive in relative terms against developing
countries than developed countries in transport equipment, but less so in textiles and
apparel. Overall, however, South Africa’s pattern of comparative advantage does not
differ markedly between developed versus less developed countries.
Figure 5.1 presents some of this same data graphically. South African relative
wages and productivity are shown for selected sectors both against developed and
developing countries. The bold line depicts relative wages, while the broken lighter line
shows relative productivity. Competitiveness can be inferred by comparing the levels of
these two lines, i.e. whether or not relative wages exceed relative productivity. For most
sectors, it can be seen that South African relative productivity and wages tend to decline,
but wages decline more, hence the improvement in competitiveness. The greater level
and improvement of competitiveness against developed relative to less developed
countries is again clear in most of the charts, although the improvement relative to less
developed countries is also clear for many industries.
A closer look at the rankings and levels of relative wages, relative productivity and
RULC in Tables 5.1 and 5.2 shows substantial diversity in competitiveness across sectors
over time. Few sectors show a consistent improvement or deterioration in
competitiveness over all periods. This is reflected in a very weak correlation between the
level and rank of RULC across sectors over the aggregated time periods. The Spearman’s
rank correlation coefficients, which capture the similarity in the ranking of sectors, are
extremely low (less than 0.27) vis-à-vis both developed and developing countries for the
periods 1980-89 and 1995-98. The inconsistency in growth of RULC across sectors
implies that the ranking of sectors according to the level of competitiveness has changed
substantially since the 1970s.
Since 1990 the ranking of sectors according to RULC levels has remained relatively
constant. Spearman’s rank correlation coefficients in excess of 0.8 were obtained when
comparing the average ordering of sectors according to the level of RULC in 1990-94
21
and 1995-98. Electrical machinery and rubber products were amongst the worst 5
performing sectors (as determined by growth in RULC) in the periods 1990-94 and 1995-
98. Other chemicals and paper and paper products were amongst the 10 worst performing
sectors in both periods. Sectors that have consistently been in the top 5 performing
sectors (reduction in RULC) are: leather products and non-ferrous metals. The
competitiveness of fabricated metal products and industrial chemicals has also improved
substantially during the 1990s.
There are also no consistent relationships between changes in RULC and sectoral
characteristics such as sector size, labor intensity and skill intensity. The relationship
between sectors ranked according to trade related variables (export orientation and import
penetration) and changes in competitiveness measures were also weak for most periods.
As noted in section II, theory predicts both a time series and a cross-sectional
relationship between relative unit labor costs and trade flows. Given the large data sets
involved here, these relationships are best examined using econometric methods. To get
some sense of the strength of these relationships we begin with a graphical analysis
before proceeding to the econometrics.
Figure 6.1 displays the cross sectional relationship between industry-level average
RULC (against all countries) and export to output ratios and net exports to output ratios,
for 10 year averages, using the weighted averages for all countries. A negative
22
correlation between RULC and exports is clear for all three decades. The relationship
between RULC and net trade is less strong but also negative in the 1980s and 1990s. The
large outlier in the export/output graphs in the 1970s and 1980s is non-ferrous metals.
The outlier in the net exports/output graphs in the 1980s and 1990s is professional and
scientific equipment.
Figure 6.2 displays the time series relationship between RULC and trade, using
the same variables, for selected industries. For ease of visual comparison, RULC is
inverted so that an increase is now an improvement in competitiveness, and a positive
correlation is now expected with exports or net exports.
The improvement in competitiveness over time for most sectors noted in the
previous section is apparent in these graphs. Labor intensive sectors (textiles, apparel,
leather), paper, industrial chemicals, fabricated metal, non-electric machinery and
transport equipment display large gains in competitiveness.
In summary, both the time series and cross section charts suggest strong effects of
relative unit labor costs on trade flows by sector. In addition, the rapid export growth in
the 1990s for a number of sectors appears to exceed the effects of improved
competitiveness, and could be due to the end of sanctions and the change in economic
policies.
23
The specification of the relevant equations that are estimated are first presented. A
cursory overview of South African evidence is then presented. This is followed by a
discussion on the econometric methodology and the results.
Various specifications have been used to analyze the relationship between RULC
and the structure of trade. MacDougall (1951) and Stern (1962), in their studies on US-
UK trade, analyze the relationship between total export ratios, relative labor productivity
and/or relative unit labor costs. Balassa (1963), also focuses on US-UK trade, but
critiques the use of total trade on the grounds that bilateral trade between the countries is
strongly affected by the relative size of US and UK tariff barriers. 15 He uses exports to
third markets instead, but the use of this has been questioned by Golub and Hsieh (2000).
Golub and Hsieh (2000) more appropriately uses bilateral trade flows as the dependent
variable and relative productivity and relative unit labor costs as explanatory variables.
Drawing on the Ricardo theory and international literature the following cross
section equations estimating the relationship between RULC and export performance
across sectors can be defined:
Xisa is total South African exports of commodity i, Xij is total country j exports of
commodity i, Xisaj and Misaj are exports and imports of commodity i to and from country j,
and RULCisaj, RelProdisaj and RelWageisaj are unit labor costs, labor productivity and
wages in South Africa over those of country j. Equation (10a) uses the ratio of total
exports as the dependent variable and is similar to the approach by MacDougall (1951)
and Stern (1963). Equations (10b) and (10c) follow Golub and Hsieh (2000) and use
bilateral trade balances as the dependent variable. In equations (10d) and (10e) the
15
MacDougall (1951) also identified this as a reason for the low import content of consumption within
these countries.
25
emphasis is on the structure of exports and not net trade, and real bilateral exports are
used as the dependent variable.
The above section on RULC and the structure of trade is cross-sectional in nature,
i.e. during a particular time period are sectors with high RULC also sectors with low
exports? We are also interested in identifying whether changes in RULC within a
particular sector over time influence export performance. In specifying the equation used
to estimate this relationship, the literature on export demand functions was drawn upon.
The standard export model, as discussed by Goldstein and Khan (1985), can be
represented as a system of equations dealing with export supply (Xs) and export demand
(Xd ) equations 16
Y*, P*, P, Px, and s are foreign income, foreign domestic price, domestic price, domestic
price of exports and subsidies, respectively. Export demand is positively affected by
foreign income (Y*) and the price of competing foreign goods (P*), but is negatively
affected by the domestic price of exports (Px).
On the supply side domestic firms produce export products if the return to export
production rises via a rise in the price of exports or export subsidies, s, but declines if the
price of domestic competing goods rise. Often variables to account for trends and cyclical
effects are included in the export supply equations. Trend income or trend export
16
This model is an imperfect substitutes model where imperfect substitutability between domestic and
export products enables domestic and export prices to differ from one another (Goldstein and Khan, 1985).
26
One problem with estimated coefficients from the reduced form functions is they
are not estimates of the structural coefficients in the export demand equation. To obtain
these it is necessary to estimate the export demand and supply equations simultaneously,
27
or in the case of fully identified equations these can be ‘retrieved’ from the reduced form
coefficient estimates directly.
The expected signs of the coefficients for RULC and Y* are negative and positive
respectively.
The analysis of RULC and their relationship to international trade has been
extremely limited in South Africa. This is particularly the case in the cross section
structural analysis. Nordas (1996) analyses RULC for 22 industrial sectors using 1990
data and finds that the traditional natural resource intensive export sectors are relatively
competitive vis-à-vis the US. No formal link to the structure of trade is drawn.
The body of empirical research on the export functions in South Africa is larger
and the key results are summarized in Table 7.1.
A number of shortcomings are evident. Firstly, the analysis is largely done at the
aggregate level with the exception of Bhorat (1998) and Wood (1995) who estimate
export functions at the sectoral level. As Wood (1995) argues “modeling at the aggregate
28
level is likely to be misleading unless the trends in all the disaggregate categories are the
same”. In his analysis of 22 industrial sectors he finds large inter-industry variations in
responsiveness of the export share, but finds very few significant coefficients. Bell,
Farrell and Cassim (1999) also show that trends in broad sector specific real effective
exchange rates for natural resource based products and non-commodity-manufacturing
products differ substantially and can explain the changing composition of South African
exports. This diversity across sectors suggests that sector specific relationships need to be
directly incorporated into the analysis. 17
Existing sector specific results are extremely poor with no conclusive evidence
that relative prices affect export demand or export supply. This may be due to the very
short time period in Bhorat (1998) or due to the specification in Wood (1995). Wood
(1995) uses the share of South African exports in world exports as the dependent variable
rather than real South African exports. One of the problems in the Wood (1995) results is
that the estimation techniques do not take into consideration potentially homogeneity in
coefficients for the exogenous variables across sectors. Panel data techniques, as used by
Golub (1994), both enlarge the sample size and enable assumptions of homogeneity to be
tested and will potentially yield more efficient estimators.
Secondly, only Golub (2000) and Golub and Ceglowski (2001) focus on the
impact of unit labor cost derived relative price indices on export performance, albeit at
the aggregate manufacturing level. They find the correct negative coefficient between
real exports and RULC with estimates of the elasticity ranging between –1.08 and –0.78.
Thirdly, there is no consistency in the use of level variables or growth rates in the
regressions. Many of the estimates using levels may be spurious as stationarity of the
variable is assumed and no testing for a long run cointegrated relationship is done. Bhorat
(1998) in his analysis of export supply functions uses cointegration techniques, but
17
The relationship between relative price and export performance may be similar across all sectors.
However, the aggregated variable does not capture differences in relative prices across sectors and thus
cannot capture the sectoral impact on export performance. Panel data techniques enable one to capture the
29
appears to apply the Engle-Granger two step approach to a multivariate model. A more
satisfactory approach is to use Johansen’s maximum likelihood procedure.
Using log difference growth rates solves for the problem of non-stationarity if
variables are I(1). However, the regression results only provide information on the short
run dynamics around a possible long run relationship. A model that captures both the
long relationship and the dynamics of this relationship is more appropriate. The export
price elasticities using growth rates range between –0.8 and –1, while the level based
elasticities are slightly higher in the case of Tsikata (1999). A wider range of estimates is
found by Golub (2000).
sector specific variables as well as common relationships between relative prices and export performance
(in fixed effect model).
30
appropriate the Mean Group estimator results will be discussed. In particular, the
individual sector regressions used in the Mean Group estimations will be presented.
We also estimate the cross section equations (10a-e) for a number of sub-periods
between 1970-98. 18 As noted by Pesaran and Smith (1995) the cross section regression
can provide consistent estimates of the long run relationships. We use the cross section
regressions to test the stability of the relationship between RULC, relative productivity,
relative wages and export performance over time.
VIII.2 Data
Two panel data sets were constructed for the estimation of the cross section and
time series equations. A large panel data set consisting of data for 24 sectors and 19
countries over the period 1970-98 was constructed. The 19 countries included are:
Canada, France, Hong Kong, Italy, Japan, Netherlands, Spain, United Kingdom, USA,
Chile, Hungary, India, Kenya, Korea, Mexico, Poland, Singapore, Turkey and
Zimbabwe. 19 An aggregated data set was also constructed summing across all countries
other than Mexico for which a complete data set was not available. All variables were
aggregated using share of total group trade (exports + imports) as weights. 20 The
aggregated panel data set is used to analyze total South African trade. The disaggregated
panel data set is used to analyze bilateral South African trade.
The results of equations (10a-e) analyzing the structure of South African trade are
presented first. These equations were estimated using the large disaggregated panel data
set as well as the aggregated panel data set. In the disaggregated case the equations were
estimated using all countries in the sample and then for each country separately.
18
Golub and Hsieh (2000) utilize seemingly unrelated regression approach to analyzing the cross sectional
regressions. Golub (1994) applies OLS to the cross section data, but the low number of sectors limits the
power of these regressions.
19
The panel is not balanced, i.e. there are data missing for some countries (Mexico).
31
Because the aggregated data set covered the period 1970-98, it was possible to
estimate the impact of RULC on the structure of trade for a number of sub-periods. For
brevity only the aggregated panel data set results analyzing the structure of South African
exports (equation 10d and 10e) are presented here, although reference will be made to
alternative results where appropriate. These results are presented in Table 9.1.
The coefficients for relative wages and relative productivity are of the expected
sign and are both significant. Sectors with low relative wages and/or high relative labor
productivity export more. The absolute value of these coefficients are not significantly
different from each other and a percentage rise in relative productivity or a percentage
decline in relative wages have the same impact on exports. This is expected as these
variables affect RULC equally.
The analysis of sub-periods suggests that the influence of relative unit labor costs
on the structure of exports has diminished over time. During the periods 1970-79 and
1980-89 RULC, relative wages, and relative labor productivity were significant in
explaining the structure of exports. However, during the 1990s these variables are
incorrectly signed and fail to explain the structure of trade as shown by the low F-
statistic. However, as shown in Figure 6.1 when using exports as a share of output a
20
See the data Appendix for further information on the sources and methods used.
32
strong negative relationship still emerges during the 1990s. Cross sector regressions using
exports as a share of output as the dependent variable find a significant relationship for
1990-94 and 1995-98, but only when using RULC. 22 The coefficients on relative
productivity and relative wage are insignificantly different from zero. The weaker results
during the 1990s suggest that other factors not related to labor costs are becoming more
important in influencing the sectoral composition of South African exports. Possible
causes of the weaker results during the 1990s are discussed in the following section.
More detailed insight in the structure of trade vis-à-vis trading partners can be
obtained by repeating the cross section regressions for each country. Data limitations,
however, constrained the number of possible export functions that could be estimated for
developing countries. Cross-section regressions were estimated for developed countries
using real bilateral exports (equation 10d and 10e), bilateral net exports (equation 10b
and 10c) and the ratio of total South African exports to total foreign country exports
(equation 10a) as the dependent variable. Bilateral trade data for developing countries
was not available prior to 1988 limiting the estimated export function to equation (10a)
for these countries. Although bilateral HS trade data were available for developing
countries after 1988, the results using this data were poor, as were the case for most
estimates during the 1990s. For complete coverage of all countries the results using the
log of total exports ratio (Xisa/Xij) as the dependent variable (equation 10a) are presented
in Table 5. These estimates are thus similar to the original research by MacDougall
(1951) and Stern (1962).
Three sets of cross section regression results are presented. The first set of
regressions (columns 1 & 4) has RULC as the explanatory variable. The second set of
regressions (columns 2 & 5) replaces RULC with relative wage and relative productivity.
The final estimations (columns 3 & 6) replace relative productivity with relative TFP.
Relative TFP is expected to capture aspects of productivity (such as capital productivity)
21
The coefficients are significantly higher than those found by Golub and Hsieh (2000).
33
that are not reflected in labor productivity. A positive coefficient is expected. Because
relative TFP data were not available for all periods, the averages for this variable may not
correspond to the averages for the entire time periods in Table 9.1. For illustrative
purposes only the significant coefficients of each country specific regression are
presented in Table 9.2. Empty cells reflect coefficients insignificantly different from zero.
Looking first at the developed country results it is clear that the results are
stronger for the 1980s than the 1990s. For the 1980s the coefficients on RULC are
negative and significant in 6 out of 8 countries, but none were significant in the 1990s.
The impact of relative wages in determining relative exports appears to be greater than
that of relative productivity.
The results for developing countries were weaker than those for developed
countries, although when significant (at or below the 10 % level) these were mostly of the
correct sign. 23 The coefficient on relative wage in Zimbabwe and Poland were
incorrectly signed. This may reflect relatively large exports of high wage-high
productivity goods by South Africa to these developing countries. Tsikata (1999) shows
that the structure of South African exports is relatively capital and human capital
intensive when compared to other middle income countries. Some of this is explained by
the regional comparative advantage South Africa has in the production and export of
high-tech products to neighboring countries.
The coefficient on relative TFP was generally very weak. Significant results with
the correct sign were only found for the UK, Chile and India during the 1980s. When
compared to these countries South African exports to the world were relatively high in
sectors also experiencing relatively high TFP. The results for Japan during the 1990s and
22
A coefficient of -1.66 is obtained for the 1990s. This is significant at the 5% level.
23
One reason for the weaker results is that the dependent variable is the ratio of total country exports
according to sector and not bilateral exports according to sector. According to the Ricardo model trade
between similar countries is expected to be low. Because bilateral trade is small between South Africa and
most other developing countries, changes in RULC vis-à-vis these countries will not have a substantial
direct affect on total SA exports. Changes in RULC will, however, affect competitiveness in third countries
which, depending on the substitutability of exports, will affect the ratio of total exports. Because other
34
for Zimbabwe during the 1980s were negative, but were only significant at the 10 %
level.
A clear result from the disaggregated and the aggregated analysis is that the
importance of RULC, relative wages and relative productivity in explaining the level of
real exports across sectors has diminished during the 1990s. Yet, as is shown in Figure
6.1 RULC are still important in determining the export orientation of South African
industry. Sectors with high RULC tend to export low shares of their output, even in the
1990s.
There are a number of possible reasons for the poor results during the 1990s. The
1990s were a period characterized by significant structural breaks such as the ending of
sanctions, the election of a new government, a new macroeconomic policy, new labor
legislation and the initiation of tariff liberalization. The impact of these structural breaks
may affect exports in a way that is not necessarily related to RULC. For example, the
reintegration of South Africa into the world economy since 1994 led to rapid increases in
exports to various regions. Some of this export growth is due to increased market access
and not changes in labor cost competitiveness. Structural changes in the economy have
also affected RULC across sectors. This has been shown by the poor Spearman’s rank
correlation coefficients between the 1980s and 1990s. These changes in the sectoral
structure of exports and RULC may obscure the long run relationship between the two.
As shown by Pesaran and Smith (1995) the cross-section estimator is biased for low T.
The 1990s may be too short a period for the cross-section estimator to give an unbiased
estimate of the average long run coefficient, particularly given the substantial changes in
the policy environment.
Non-traditional exports have also grown rapidly during the 1990s and have
increased the diversity of South Africa’s export bundle (Black and Kahn, 1998). Because
factors (market access agreements, product differentiation) also affect exports, these products are unlikely
to be perfect substitutes; hence the relatively poor results for developing countries.
35
these changes are not adequately captured in the broad industrial categories used within
this analysis, potential relationships between trade and RULC will be obscured.
Finally, the decline in importance of RULC may also reflect the rising importance
of human capital and technology in determining export competitiveness. Edwards
(2001b) shows a positive relationship between skill intensity and export growth between
1993-97, some of which is due to growth of non-traditional exports, particularly to
neighboring countries. Using a factor classification system Edwards and Schoer (2001)
and Lewis (2001) also show a bias in South African exports of technology intensive and
human capital intensive sectors that has continued into the late 1990s. This bias may
reflect a shift in competitiveness and trade away from natural resource and labor
intensive products towards more technology and skill intensive products. The use of
microelectronics and automation processes in production, the development of new
products and production processes through innovation and R&D and the provision of
after sales service are increasingly being seen as important determinants of
competitiveness. This is the view that lies behind the recent Department of Trade and
Industries (DTI) industrial strategy framework document (DTI, 2001).
To conclude, the cross section results provide substantial support for the
Ricardian theory prior to 1990. During this period, sectors characterized by high RULC
or high relative wages and low relative productivity exported significantly less than
sectors with low RULC or low relative wages and high relative productivity. When using
the aggregated data set, the impact on exports of a 1 % rise in relative productivity
appears to have the same impact as a 1 % decline in relative wages. In contrast the
country specific regressions suggest that relative wages have a greater impact. Since 1990
the evidence is weaker, particularly at the country level. Significant structural breaks, the
role of innovation and technology in competitiveness, and rising diversity of exports
within industrial sector classifications may give rise to the weaker relationship between
RULC and export levels. Unit labor costs are, however, still significant in explaining the
degree to which a firm is export orientated. Further evidence is provided in the time-
series analysis.
36
In this section the time series results are presented. The reduced form equation
(13) derived from the export demand and supply of equations is estimated. Although the
estimated coefficients for world income and relative prices in the reduced form equation
(13) do not necessarily represent income and price elasticities of export demand, they
give insight into the relationship between exports and these variables. The export demand
function equation (14) with RULC used as a proxy for relative prices is also estimated.
This function implicitly assumes an infinite export supply elasticity. All the export
functions were estimated with RULC and RULC substituted by relative productivity,
relative wage and exchange rates. Dynamic versions as specified in the appendix
equation (A9) are also estimated using the dynamic fixed effects estimator.
These equations were estimated using the large disaggregated and the aggregated
panel data sets. The latter is more appropriate for the dynamic estimates as it covers the
period 1970-98 providing a larger sample for the estimation of the ‘within sector’
variation (as is done by the fixed effects estimator).
Within the aggregated panel data set weighted averages of the exogenous
variables were constructed for developed and developing countries within the sample.
The equations were estimated for these broad regions separately. The export functions
were also estimated for the following manufacturing sub-categories: natural resource
intensive, labor intensive, chemical intensive and machinery & metal products sectors.
This may overcome problems of inconsistent estimates if the slope coefficients differ
across sectors. As these manufacturing sub-sectors are more homogenous than the entire
sample, the relationship between the exogenous variables and export performance is
expected to similar. Further insight into the sectoral variations in response is provided by
the individual sector regressions used to calculate the Mean Group estimates of the
average long run coefficients.
37
The large disaggregated panel was also used to estimate export functions for the
entire sample as well as for specific countries. In the entire sample estimate a fixed effect
was assumed for each sector within each country. The intercept in these cases should
capture any country and sector specific effect that is constant over time. Unlike the
aggregated panel data set it was possible to estimate country export functions with
relative wages in local currency units, relative productivity and nominal exchange rates
separately. This should give insight into the relative importance of each of these in
influencing export performance.
Table 10.1 presents the estimated coefficients of the various export functions
using the aggregated panel data set under the assumption of homogeneity of coefficients
across sectors.
As shown in Table 10.1 the coefficients for RULC, relative wage, relative
productivity and world manufacturing GDP (World VA) are statistically significant and
of the correct sign for all export functions. In the export demand function (column 1) the
24
Carlin et al. (2001) in their study of OECD constructed weighted relative wage and exchange rate indices
from nominal wage and nominal exchange rate indices. Although they sum variables in different units, this
is unlikely to be too great a problem in a sample of countries with similar inflation rates. Given the broader
38
short run price elasticity (coefficient of RULC) is –0.37 which is statistically not different
from the reduced form equation coefficient –0.35 in column (4). The long run elasticities
are considerably larger ranging between –2.34 and – 2.45. A 1 % rise in RULC reduces
real exports by approximately 0.36 % in the short run, but by approximately 2.4 % in the
long run. 25 This reflects a far greater responsiveness to RULC changes than is found by
Golub (2000).
The results of export demand and reduced form equations using relative wages
and relative productivity as explanatory variables are shown in column (2) and column
(5), respectively. The coefficients of these variables are also not significantly different
from one another when compared across the different specifications of the export
functions. A 1 % rise in relative productivity raises real exports by 0.32 % in the short
run and by 1.9 % in the long run. Relative wages appear to have a more significant
impact with a 1 % rise in relative wages resulting in a short run negative impact of
between 0.5 % and 0.54 % on real exports. The impact in the long run lies between -3 %
and -3.18 %. As mentioned, it is not possible to separate out the impact arising from
exchange rate movements versus nominal wage movements from these coefficients.
World income has the correct sign and ranges between 0.27 and 0.35 in the
dynamic fixed effects model. This gives rise to a long run income elasticity of between
1.64 to 2.35. These values are similar to those found in the international survey by
Goldstein and Khan (1985), but are slightly higher than those found in Fallon and Pereira
de Silva (1993), Tsikata (1999), Golub (2000) and Golub and Ceglowski (2001). The
positive sign found in this study as well as many other studies on South Africa also
question the negative coefficient found by Wood (1995). Capacity utilization has the
expected negative sign, although it is not significant in the reduced form export equation
coverage of this paper as well as the inclusion of high inflation economies, it is not possible to pursue a
similar approach here.
25
When estimating a dynamic fixed effects ARDL(1,1,1,1) model, the long run coefficients for RULC,
capacity utilisation and world value added were –2.19, -7.18 and 2.14, respectively. These closely
approximate those found in the ARDL(1,0,0,0) model in column (4) of Table 4. The average long run Mean
Group estimates for RULC, capacity utilisation and world value added when using the ARDL(1,0,0,0)
model were –1.71, -3.63 and 2.37, respectively. The Mean Group estimates using the ARDL(1,1,1,1) model
were –1.64, -4.41 and 2.02, respectively.
39
in column (3). As domestic demand rises and capacity utilization falls, firms redirect
production away from the export market and concentrate on the domestic market. The
short run coefficient is high ranging between –0.69 to –0.93. These values exceed the –
0.3 to –0.5 range found in the survey by Goldstein and Khan (1985). The relatively high
value may reflect the use of the export market by South African firms as a vent-for-
surplus arising from the recessions during the 1980s and early 1990s.
The coefficient on the sanctions dummy was significant and negative for most
export functions and suggest that sanctions reduced real exports by approximately 0.06 %
between 1985 and 1993.
Other variables were also included in the analysis. South African investment in
machinery and equipment as a share of total capital stock of machinery and equipment
according to sector was included as an exogenous variable. As in Carlin et al. (2001) this
was used as a proxy for innovation, R&D and embodied technological change. No
significant results were found. Effective rates of protection according to sector were
obtained from Fedderke and Vase (2001) and included as explanatory variables. Declines
in effective protection are expected to capture reductions in the anti-export bias. Unlike
Tsikata (1999) who found a negative relationship, no significant results were found.
Insufficient data and the exclusion of non-tariff protection in calculating these effective
rates of protection are likely to partly explain the poor results.
In the process of estimating the average long run coefficients using the Mean
Group Estimator, the reduced form export function (as shown in column (4) Table 10.1)
were estimated for each sector. The long run results for each sector and the error
correction term are presented in Table 10.2. Care must be taken in interpreting these as
the correct procedure outlined by Pesaran and Smith (1995) and Pesaran and Shin (1996)
for testing for the existence of long run relationships and the subsequent estimation
40
thereof using a correctly specified ARDL model has not been followed. However, as in
Pesaran, Shin and Smith (1999) the sector specific results are still presented.
The sector specific results were mixed. As discussed in Pesaran and Smith (1996)
for a long run relationship to exist the error correction coefficient φi ≠ 0. This exceeded –
0.1 for all sectors reaching –0.68 for paper products. Other sectors displaying rapid
convergence to the long run equilibrium (φi < -0.4 ) were textiles, wood products,
printing, glass products, other non-metallic products, non-electrical machinery, transport,
professional and scientific equipment and other manufactures. 26
The long run elasticity for RULC is correctly signed for 21 of the 24 sectors and
significant in 9 of them. The largest significant coefficient is for furniture (-5.14) and the
lowest is –1.27 for other manufactures. Labor intensive sectors appear most affected by
RULC with 4 of the 6 labor-intensive sectors displaying negative coefficients. These are
textiles, wearing apparel, wood products and furniture. No coefficients for RULC were
significant in chemical intensive sectors. Of the remaining sectors, significant and
negative results were found for food, paper products, fabricated metal products,
professional and scientific equipment and other manufacturing. These findings generally
accord well with Figure 6.2. As found by Goldstein and Khan (1985) the elasticities of
sub-sectors generally exceed those for manufacturing as a whole.
Capacity utilization was not significantly different from zero in any of the sector
equations. World Value added was significant (at the 10 % level) in 12 of the sectors,
although leather products had the incorrect sign. The long run coefficient ranged from
1.562 for non-electrical machinery to 7.599 for rubber products. Other sectors strongly
affected by foreign income are fabricated metal products, transport, other non-metallic
mineral products and glass products. Sanctions were significant and negative for
26
Evidence of serial correlation is found in 4 sectors (paper products, other chemicals, transport, and
professional and scientific equipment). Heteroskedasticity is found in furniture, plastic, non-ferrous metal,
electrical machinery and transport sectors. 3 sectors are found to be mis-specified (food, industrial
chemicals and other manufactures).
41
Further differences emerge when RULC is broken down into relative wages and
relative productivity. Relative productivity is positive and significant for all sub-sectors
apart from chemical intensive sectors. The short run coefficients range from 0.31 to 0.42
and are similar to those for manufacturing as a whole. The long run coefficients are larger
and range between 0.83 (chemicals) and 2.44 (natural resources). As found earlier
relative wages have a more severe impact on export performance than relative
productivity. In the long run natural resource intensive and labor intensive sectors are
strongly affected with a 1 % rise in relative wages reducing exports by over 3 %.
Capacity utilization and sanctions are only significant for the natural resource
intensive sectors and may reflect the greater impact of sanctions on these sectors. World
income, on the other hand, is significant and positive for the remaining sectors
42
The reduced form export demand functions were repeated for developed and
developing countries separately. In these cases the relevant explanatory variables were
weighted using country share of total developed or developing country trade in the
sample. The dependent variable in both cases is total real sectoral exports and not
regional real exports due to data limitations. The results are presented in Table 10.4.
Given the similarity in Table 10.1 of the results between the lagged export
demand function (specification 2) and the reduced form export function (specification 5)
in the case of the country-aggregated data, it was decided to apply the former to each
country using the large disaggregated panel data set. 28 The advantage of the
27
Rich countries (defined as developed OECD countries) account for over 50 % of South African exports
and over 70 % of South African imports (Edwards and Schoer, 2001).
28
This is also theoretically more consistent with the notion of export demand functions as the coefficients
can be interpreted as elasticities under the assumption of infinite supply of exports. In the context of
drastically declining domestic demand during the 1980s and the rise in excess capacity, this assumption
may not be too extreme. When capacity utilisation was included to form the reduced form equation, the
43
disaggregated panel data set is that it increases the number of observations, enables
country specific regressions and permits a separation of exchange rate and relative wage
effects in the country specific regressions.
Unfortunately bilateral trade data from 1970 are not available for the full set of
countries. However, South African exports to developed countries for the period 1980-98
can be obtained from the Canadian World Trade Analyzer. 29 We use this data to estimate
export functions for trade with each developed country in the sample. Foreign income is
proxied by total imports (in US$) by each country from the world. The results are
presented in Table 10.5a-c.
The results are generally favorable (Table 10.5a). Although the entire sample
estimate yields results consistent with the aggregated results, the coefficient for RULC
are smaller giving rise to lower long run elasticities. A 1 % rise in RULC negatively
affects real exports, but only by –0.09 % in the short run and –0.2 % in the long run. The
coefficient signs for sanctions and world income (proxied by country imports from
world) are negative and positive, respectively, and are similar (if not larger) than the
aggregated results.
The country specific regression results are mixed. Note that because of the
smaller time period over which the equations were estimated, the results may differ from
the aggregate results. Of the 9 country specific equations estimated, all the short run
coefficients on RULC are correctly signed and three of them are significant. Their values
ranged between –0.26 and –0.59. The long run equivalents ranged from –0.53 (Italy) to –
1.43 (France). The poor results may reflect the exclusion of the period 1970-79 where
RULC appeared to be more significant in explaining the structure of trade than in the
1990s.
results did not change, although capacity utilisation was significant and negative for the entire sample
estimate and for the Netherlands and UK regressions.
44
Breaking down RULC into relative wage (Table 10.5b), exchange rates and
relative productivity did not improve the results substantially. The short run coefficient
on relative productivity was positive and significant in only two of the country specific
regressions and ranged between 0.5 and 0.69. The long run coefficient for these two
significant variables ranged between 1.22 to 1.33. Relative wages were negative and
significant in four of the country specific regressions and ranged between –0.94 and –
2.25 in the long run. As is being consistently found, the impact of wages on exports
exceed the impact of productivity. Exchange rate movements also have a significant
effect on export performance. The short run coefficient was significant and positive in 5
of the 9 country regressions and ranged between 0.52 and 1.23. In the long run a 1 %
depreciation of the rand is estimated to boost exports by between 0.77 % (Hong Kong)
and 2.4 %(Italy and France).
World demand plays a prominent role in influencing South African exports. The
short run coefficient on world imports was significant in all 9 country regressions with
RULC as an explanatory variable. However, it was significant in only 3 (but of the
correct sign in 7) of the country regression where RULC was broken down into its
constituent parts.
Time series versions of the McDougall (1951) model (equation 10a) with the log
total export ratio (Xisa/Xij ) as dependent variable were also estimated for all countries
between 1980-98 (see Appendix Table A1a-c). This enabled the inclusion of developing
countries for which trade data are available. RULC was significant in 8 of the 19
countries, of which 6 were of the correct sign (Chile, Hungary, India, Mexico, Poland,
Singapore). Interestingly, no significant coefficients were found for developed countries.
When relative productivity and relative wages were included, three countries had
significant coefficients with the correct sign on relative productivity and nine had the
correct sign on relative wages.
29
Bilateral data classified according to the Harmonised System was available for all countries from 1988.
The time period, however, was seen to be too short to yield reliable estimates. When estimated the results
45
The country specific equations also enabled the inclusion of relative TFP as an
exogenous variable (Table 10.5.c). Equations were estimated in which relative TFP
replace relative productivity. The results were extremely poor. When real South African
exports to the country were used as the dependent variable only Japan had a significant
and positive coefficient on relative TFP (Table 10.5). The long run coefficient indicates
that a 1 % rise in relative TFP increases South African exports to Japan by 1.15 % in the
long run. When the log total export ratio was used as the dependent variable significant
and positive coefficients were found for Hungary, India, Italy, South Korea and
Zimbabwe (Appendix Table A1).
XI. Conclusions
This paper has sought to extend Golub’s (2000) analysis of South African
competitiveness in manufacturing in two main directions. First, we have extended the
productivity calculations to total factor productivity, rather than limiting them to labor
productivity. Labor productivity can be a misleading indicator of efficiency to the extent
that rising labor productivity could be due to capital-labor substitution rather than total
factor productivity growth. Secondly, to assess South African comparative advantage we
have extended the analysis of relative unit labor costs, productivity and wages, and their
effects on trade flows to the industry level.
South African Relative Productivity. Our calculations show that South African
relative productivity growth in manufacturing in the 1980s and 1990s has been relatively
weak, measured either by labor productivity or total factor productivity, especially in
comparison to other newly-industrializing countries in Asia and others such as Chile.
In the 1980s South African labor and total factor productivity grew at about the
same rate, as the capital labor-ratio changed little. In the 1990s, however, the South
African capital-labor ratio increased so total factor productivity stagnated despite modest
were poor.
46
growth of labor productivity. While most of the increased capital-labor ratio reflects
capital accumulation, some of it is also attributable to labor shedding. These findings are
disconcerting insofar as sustainable long-run output and employment growth depends on
raising productive efficiency rather than through capital-labor substitution and labor
shedding.
Relative to other countries in our sample, however, South African labor and total
factor productivity have both declined at roughly similar rates, as other countries also
experienced some capital labor substitution. That is, many of these countries experienced
much faster labor and total productivity growth than South Africa, as noted above.
We have also evaluated sector competitiveness rankings and their evolution over
time. No clear pattern emerges as to the nature of South African comparative advantage.
Those sectors with the highest levels and improvements of competitiveness include a
variety of labor intensive, capital-intensive, and natural resource intensive goods.
47
Firstly, sustainable growth of output, exports and employment, all depend on total
factor productivity growth, together with wage moderation. While this paper has not
directly investigated the sources of South Africa’s poor productivity growth, some of the
requisite policy measures are well known.
One such area is education. Although the rewards may only be realized in the
long run, improvements in education, particularly the quality thereof, can substantially
raise the productivity of South Africa’s labor force and through this export performance.
The positive impacts on economic growth of improved educational quality have been
shown by Fedderke (2001).
48
Second, over time, productivity growth enables real wage growth. It is important,
however, to ensure that growth of labor costs does not outstrip productivity growth, and
that unit labor costs remain international competitive. Given that South African relative
unit labor costs, although much reduced, remain high compared to a number of newly
industrializing countries with which South Africa competes, further reductions in relative
unit labor costs would be beneficial.
Third, South African exports respond strongly to market forces and particularly
relative unit labor costs, both overall and at the industry level. Sectoral rankings of
competitiveness, however, are not readily identified with observable characteristics and
change substantially over time. These findings suggest that there is no easy way to pick
winners based on industry characteristics and that the government should not attempt to
favor some sectors over others. Instead, as indicated above, a general environment which
fosters productivity growth and wage moderation will entail rapid export growth. Market
forces will then determine which industries are particularly successful in exporting.
49
References
Balassa, B., 1963. An empirical demonstration of classical comparative cost theory. Review of
Economics and Statistics, 4: 231-8.
Balassa, B., 1979. Export composition and export performance in the industrial countries, 1953-
71. The review of Economics and Statistics, vol. 61: 604-607.
Bell, T., Farrell, G. and Cassim, R., 1999. Competitiveness, international trade and
finance in a minerals-rich economy: The case of South Africa. Paper presented at
the Trade and Industrial Policy Secretariat annual forum, 19-22 September, 1999.
Black. A, and Kahn, B., 1998. The performance of South Africa’s non-traditional exports
since 1980. Trade and Industry Monitor, vol. 7: 9-13.
Bhorat, H., 1998. Income and price elasticities in manufacturing exports. Trade and
Industry Monitor, vol. 5: 9-13.
Bowen, H., Leamer, E. and Sveikauskas, L., 1987. Multicountry, multi-factor tests of the
factor abundance theory. American Economic Review, 77, 5: 791-809.
Carlin, W., Glyn, A. and Van Reenen, J., 2001. Export market performance of OECD
countries: An empirical examination of the role of cost competitiveness. The
Economic Journal, 111 (January): 128-162.
Caves, R., Frankel, J. and Jones, R., 1999. World Trade and Payments: An Introduction.
Addison Wesley, New York.
Davis, D. and Weinstein, D., 1998. An account of global factor trade. NBER Working
Paper 6785.
Dornbusch, R., Fischer, S., and Samuelson, P.A. 1977. Comparative advantage, trade and
payments in a Ricardian Model with a Continuum of Goods. American Economic
Review 65: 297-308.
DTI. 2001. Driving competitiveness in South Africa: The integrated industrial strategy
for sustainable employment and growth. Pretoria.
Edwards, L. and Schoer, V., 2001. The structure and competitiveness of South African
trade. Paper presented at the TIPS Annual Forum, September 10-12.
Edwards, L., 2001a. Globalisation and the skill bias of occupational employment in South
Africa. South African Journal of Economics, 69, 1: 40-71.
50
Edwards, L., 2001b. Trade and the structure of South African production, 1984-97.
Development Southern Africa, vol. 18, 4: 471-491.
Edwards, L., Van Seventer, D. and Mlangeni, T., 2000. Revealed Comparative
Advantage in SADC Countries. Southern African UPDATE, vol. 5.
Fallon, P. and Pereira de Silva, L., 1994. South Africa: Economic performance and
policies. World Bank Discussion paper, 7. The Southern Africa department, The
World Bank.
Fedderke, J., 2001. Technology, Human Capital and Growth: Evidence from a Middle
Income Country Case Study Applying Dynamic Heterogeneous Panel Analysis.
Paper presented at the TIPS Annual Forum, September 10-12.
Fedderke, J., Shin, Y. and Vase, P., 1999. Trade and labour usage: An examination of the
South African manufacturing industry, Econometric Research Southern Africa
Woking paper, No. 15.
Fedderke, J., and Vase, P., 2001. The nature of South Africa’s trade patterns by economic
sector, and the extent of trade liberalisation during the course of the 1990’s. South
African Journal of Economics, 69, 3: 436-473.
Goldstein, M. and Khan, M., 1978. The supply and demand for exports: A simultaneous
approach. The Review of Economics and Statistics, vol. 60: 275-286.
Goldstein, M. and Khan, M., 1985. Income and price effects in foreign trade. In Jones, R.
and Kenen, P. Handbook of international economics, vol. II.
Golub, S. 1994. Comparative advantage, exchange rates, and sectoral trade balances of major
industrial countries. IMF staff papers, vol. 41, June 1994.
Golub, S., and Ceglowski, J., 2001. South African real exchange rates and manufacturing
competitiveness. Unpublished paper.
Golub, S., 2000. South Africa’s International Cost Competitiveness. TIPS working paper,
14.
Hayter, S., Reinecke, G. and Torres, R., 1999. South Africa: Studies on the social
dimensions of globalisation. Task force on country studies on globalization,
International Labor Office, Geneva.
Hinkle L. and Nsengiyumva, F., 1999. External Real Exchange Rates: Purchasing Power Parity,
the Mundell Fleming Model, and Competitiveness in Traded Goods. In Hinkle, L. and
Montiel, P. (Eds), Exchange Rate Misalignment: Concepts and Measurement. World
Bank.
IMF., 1998. South Africa: Selected issues. IMF Staff Country Reports No. 98/96.
Jones, R., 1980. Comparative and absolute advantage. Swiss Journal of Economics and
Statistics, 3: 235-60
Jonsson, G. and Subramanian, A., 2000. Dynamic Gains from Trade - Evidence from
South Africa. IMF Working Paper No. 00/45.
Kahn, B., 1998. Assessing South Africa’s Competitiveness: Is the Reserve Bank’s
Measure Misleading. LSE Centre for Research and Finance in Southern Africa
Quarterly Review, no. 4.
Khan, M. and Ross, K., 1977. The functional form of the aggregate import demand
equation. Journal of international economics, 7: 149-160.
Lewis, J., 2001. Reform and opportunity: The changing role and patterns of trade in SA
and SADC. Africa Region Working Paper Series, no. 14. World Bank.
Lowe, N. and Kenney, M. 1999. Foreign investment and the global geography of
production: Why the Mexican consumer electronics industry failed, World
Development, vol. 27, 8: 1427-1443.
MacDougall, G., 1951. British and American exports: A study suggested by the theory of
comparative costs: Part 1. The Economic Journal, vol. 61: 697-723
Marquez, J. and McNeilly, C., 1988. Income and price elasticities for exports of
developing countries. Review of Economics and Statistics, vol. 70: 306-314.
52
Pesaran, M.H., Shin, Y., and Smith, R. J., 1999. Pooled mean group estimation of
dynamic heterogeneous panels. Journal of the America Statistical Association, 94:
621-634.
Pesaran, M.H., and Smith, R.P., 1995. Estimating long run relationships from dynamic
heterogeneous panels. Journal of Econometrics, 68,79-113.
Pesaran, M.H., Shin, Y., 1999. An autoregressive distributed lag modelling approach to
cointegration analysis. In Strom, S. (ed), Econometrics and Economic Theory in
the 20th Century: The Ragnar Frisch Centennial Symposium. Cambridge
University Press, Cambridge.
Pesaran, M.H., Shin, Y., and Smith, R. J., 1996. Testing for the existence of a long run
relationship. University of Cambridge DAE Working Paper No.9622.
Riedel, J. 1988. The demand for LDC exports of manufactures: Estimates from Hong
Kong. The Economic Journal, 98, March: 138-148.
Smith, R., 2001. Estimation and inferences with non-stationary panel time-series data.
Paper prepared for the RC33 conference, Cologne, October 2000.
Stern, R., 1962. British and American productivity and comparative costs in international
trade. Oxford Economic Papers, 14, 3: 275-303.
Trefler, D., 1995. The case of missing trade and other mysteries. American Economic
Review, 85: 1029-46.
Tsikata, Y., 1999. Liberalisation and trade performance in South Africa. World Bank
informal discussion papers on aspects of the South African economy No. 13, The
Southern African department. Washington DC: The World Bank.
Turner, A. and Golub, S., 1997. Towards a System of Unit Labor Cost-Based
Competitiveness Indicators for Advanced, Developing and Transition Countries.
Staff Studies for the World Economic Outlook, IMF..
Valentine, N and Krasnik, G., 2000. SADC trade with the rest of the world: Winning
export sectors and revealed comparative advantage ratios. The South African
Journal of Economics, vol. 68, 2: 266-285.
Verbeek, M., 2000. A Guide to Modern Econometrics. John Wiley & Sons, Singapore.
Wood, E., 1995. Essays on trade and manufacturing industry in South Africa. Unpublished PhD,
University of Cambridge.
53
54
The world demand for South Africa’s exports of sector i in period t is specified in
log-linear form as30
P
(A1) log X itd = β0 + β1 log x + β2 log Yt *
ULC * it
where Xitd is quantity of exports Pxit is price of exports, ULCit* is unit labor cost of
foreign firm and Yt* is real income of South Africa’s trading partners. 31 This model
assumes a perfectly competitive market in which labor is the only factor of production.
The domestic price of a product, P, thus equals ULC. 32 Different demand elasticities in
the home and export market, different cost structures for home and export production,
market distortions and product differentiation give rise to differences in export and
domestic prices (Goldstein and Khan, 1985). Because the variables are in logs, the
coefficients represent elasticities, the expected signs of which are:
β1 < 0; β2 > 0 (although if the product is inferior then we may find β2 < 0)33 .
Thus a decline in price competitiveness of South Africa arising from an increase rise in
export price or a decline in foreign domestic price (ULC*) reduces export demand.
30
Khan and Ross (1977) find that the log-linear specification performs better than a standard linear
function.
31
A common β1 is assumed as it is generally assumed that the demand function is homogenous of degree
zero in prices, i.e. a doubling of Px and of ULC * has no effect on demand as relative prices have not
changed.
32
It is possible to incorporate other costs, or mark-ups, which raise the market price above ULC by a
constant fraction λ, i.e. P = λULC , without affecting the results.
33
Wood (1995) finds a negative coefficient in his estimation for some sectors in South Africa.
34
A problem with this functional form is that it fails to take into account possible interdependence between
capacity utilisation and export performance (Balassa, 1979).
55
markets are substitute products. A rise in Px or a decline in domestic price (ULC) raises
the relative profitability of export causing a shift of resources into export producing
sectors. If ULC also serves as a proxy for overall production costs, the decline in ULC
will also reflect a reduction in the cost of producing exports. Capacity utilization has been
included to capture the cyclical impact of domestic demand on export behaviour. In
South Africa this variable is expected to capture the impact of the volatile domestic
demand conditions during periods of political upheaval. As domestic demand declined
firms turned to the export market as a ‘vent-for-surplus’. Not included in the equation is a
variable to capture domestic production that is generally included to proxy a country’s
capacity to produce (Goldstein and Khan, 1978). The expected signs of the elasticities
are:
α1 > 0, α2 < 0.
Imposing the equilibrium condition, Xs = Xd = X, the two equations solve for the
two endogenous variables X and Px. If equation (A2) is normalised for the price of
exports, Px, and then substituted into equation (A1) we obtain the reduced form
equations:
−1
β βα βα
(A3) log X it = 1 − 1 (β 0 − 1 0 ) + β1 log RULCit − 1 2 log CU it + β 2 log Yt*
α1 α1 α1
w + log 1
log RULCit = log
a.PPP
(A4) + log e
a * it w * it R / For it
where a and a* reflect labor requirements per unit real output in domestic and foreign
industries, respectively, w and w* reflect domestic and foreign wage per worker,
respectively, PPP is the PPP exchange rate, and eR/For is the Rand value of foreign
currency. The first variable on the right hand side measures the productivity of foreign
workers relative to South African workers while the second measures the relative wage of
South African workers compared to foreign workers. Thus a 10 % rise in RULC can be
brought about by either a 10 % rise in relative productivity of foreign workers, a 10 %
56
rise in the relative wage of South African workers or a 10 % appreciation of the currency.
This relationship can be tested directly by substituting equation (A4) into regression
equation (A3).
Given the signs of the structural parameters in the export supply and demand
functions the expected signs of the reduced form equation (A3) coefficients are as
follows
Note, however, that these are not necessarily estimates of the structural
coefficients in the export demand equation (A1). To obtain these it is necessary to
estimate the export demand and supply equations simultaneously, or in the case of fully
identified equations these can be ‘retrieved’ from the reduced form coefficient estimates
directly.
Frequently studies ignore the export supply equation and estimate the export
demand function in equation (A1) directly. The problem with this approach is that these
models implicitly assume infinitely elastic export supply or stable demand functions with
a supply function that shifts around it (Goldstein and Khan, 1985; Riedel, 1988). The
estimates for β1 and β2 are interpreted as elasticities, but because these are weighted
averages of the true demand and supply elasticities they tend to be biased downwards
(Goldstein and Khan, 1985). More problematic is the correlation between the endogenous
variable on the right hand side and the error terms in the regression as these violate the
conditions for OLS and will yield inconsistent estimates (Gujarati, 1995). Nevertheless,
because these models have dominated research in this area, and particularly research in
South Africa, it is useful to estimate the export demand function (A1), if only for
comparative purposes.
Often the presence of adjustment costs and incomplete information induces a lag
in the response of demand to changes in the exogenous variable. This adjustment process
57
is frequently introduced by imposing various distributed lag structures on the model. The
standard adjustment process used in the literature is the partial adjustment model where
the actual change in exports in time period t equals a fraction of the desired change in
exports. That is:
where X̂ it is desired level of exports and λ is the coefficient of adjustment, 0 < λ < 1.
Assume an export demand function similar to equation (A1), but where Px /P* is
proxied by RULC:
ULC
(A6) log X itd = β0 + β1 log + β2 log Yt *
ULC * it
If we substitute (A6) into the export demand equation (A5) and simplify we obtain
ULC
(A7) log X itd = λβ0 + λβ1 log + λβ2 log Yt * + (1 − λ) log X it −1
ULC * it
The coefficients λβ1 and λβ2 are the short run price and income elasticities, but become
long run elastiticies when divided by (1 – λ). 35 It is also possible to substitute equation
(A4) for RULC in equation (A7).
B. Econometric methodology
35
Note that this model assumes a geometrically declining lag structure. The model also assumes that the lag
in the response of exports is the same irrespective of whether the change in exports is due to changes in
world income or prices (Goldstein and Khan, 1985. See also Marquez and McNeilly (1988) who use
alternative dynamic specifications). Polynomial distributed lag structures have also been used.
58
Drawing on Pesaran, Shin and Smith (1999) and equation (A3) we assume that
the long run export function is given by
(A8) X it = θ0i + θ1i RULC it + θ2 i Yit* + θ3i CU it + µit
where Xit is the logarithm of export demand, RULCit is the log of relative unit labor cost
(a proxy for relative prices), Yit* is log of real foreign income and CUit is logarithm of
capacity utilization. i = 1,…,N, stand for industrial sectors and t = 1,2,….T, stand for time
periods. The expected signs of the long run coefficients are
θ1i < 0, θ2i > 0 and θ2i < 0.
To estimate the long run properties of this model we impose the dynamic
structure as obtained from the Partial Adjustment Model
(A9) X it = αi + λi X i ,t −1 + γ1i RULC it + γ2iYit* + γ3iCU it + εit .
where
(A11) θ0i = αi , θ1i = γ 1i , θ2i = γ 2i , θ3i = γ 3i , φi = − (1 − λi ) .
1 − λi 1 − λi 1 − λi 1 − λi
φi is the error correction coefficient and measures the speed of adjustment towards long
run equilibrium. The larger the value the greater the convergence to long run equilibrium.
For stability we require φi < 0.
Pesaran and Smith (1995) identify a number of procedures for estimating the long
run effects and emphasize the potential inconsistencies arising from using these under
different conditions of heterogeneity across sectors. One approach is to estimate a
59
Dynamic Fixed Effects model where homogeneity for all parameters except for a fixed
effect (α i) for each sector is assumed. 36 In this model for all i = 1,2,…N
λi = λ; γ1i = γ1 ; γ2i = γ2 ; γ3i = γ3
The estimated long run coefficients are calculated from the estimates of these coefficients
(A12) θˆ1 = γˆ1 , θˆ2 = γˆ2 , θˆ3 = γˆ3
1−λ ˆ 1−λ ˆ 1−λ ˆ
As shown by Pesaran and Smith (1995) this estimator is consistent for large T and N.
A more appropriate approach in this case is to use the Mean Group Estimator
(MG) of Pesaran and Smith (1995). Equation (A9) is estimated separately for each sector
and the long run coefficients are computed as averages of the individual sector
coefficients.
For example, the long run elasticities for RULC, Y* and CU for each industry i
are defined as
(A13) θˆ1i = γˆ1i , θˆ2i = γˆ 2i , θˆ3i = γˆ3i .
1− λ ˆ
i 1− λ ˆ
i 1 − λˆi
36
It is also possible to assume αi = α for a common constant across all sectors and αi = α1 + η1i for sector
60
N N N N
(A14) θˆ1 = ∑ θˆ1i / N , θˆ2 = ∑ θˆ2i / N , θˆ3 = ∑ θˆ3i / N , φˆ = ∑ φˆi / N .
i =1 i =1 i =1 i =1
As shown by Pesaran and Smith (1995) these averages are consistent estimates of the
parameters. 37 Because a separate equation for each sector is estimated it is possible to use
different ARDL specifications in each case. This is particularly useful when the short run
dynamics differ from sector to sector. As is evident in the examples of Pesaran, Shin and
Smith (1999) the choice of lag structure appears to be more important when T is small.
In cases where some coefficients are equal across sectors, the MG estimator will
provide consistent, but inefficient estimates of the long run coefficients. Pesaran, Shin
and Smith (1999) suggest that frequently the long run effects are homogenous and
propose a Pooled Mean Group (PMG) estimator that constrains the long run coefficients
to be the same for each sector, i.e.
(A15) θˆ1 = θ1i , θˆ2 = θ2i , θˆ3 = θ3i for i = 1,2,…N.
The short run coefficients, the γ’s, are allowed vary across sectors. The PMG estimator
for the short run coefficient, γ1 , is
N
(A16) γˆ 1 = ∑ γ~1i / N
i =1
where γ~1i is the coefficient for sector i computed by the pooled maximum likelihood
estimation. The PMG estimators for γ2 and γ3 are calculated in a similar manner. A
further benefit of the PMG over the MG is that it appears to be less sensitive to outliers, a
common problem with the MG estimator (Pesaran, Shin and Smith, 1999). It also appears
to be relatively robust to choice of lag order.
A final approach to estimating the average long run relationship is to average the
data over time and then estimate cross-section regressions on group means. Pesaran and
Smith (1995) show that for large N and T the cross-section estimator will give consistent
estimates of the average long run coefficient even if the coefficients differ randomly
across sectors. Strict exogeneity of the regressors is required. In this approach the
following function is estimated
X it = θ 0i + θ 1 RULCt + θ 2Yt * + θ 3 CU i + ε i
In this paper the results of the Dynamic Fixed Effects estimator are presented. To
cater for problems associated with heterogeneity across sectors, the equations will be
estimated for sub-groups. Where appropriate the Mean Group estimator results will be
discussed. In particular, the individual sector regressions used in the Mean Group
estimations will be presented. Finally, the cross section equations (10a-e) are estimated
for a number of sub-periods between 1970-98. As noted by Pesaran and Smith (1995) the
cross section regression can provide consistent estimates of the long run relationship
between RULC, relative productivity, relative wages and export performance. The
estimation over various periods will enable an analysis of the stability of the relationship
over time.
C. Data
The calculation of RULC, relative wages, relative productivity and relative TFP
for the vast range of countries and sectors against which South Africa is compared is a
data intensive process. This section briefly outlines the data sources and some of the
changes made to the data in cases of missing values and clear inconsistencies.
with incomplete data, 9 developed countries and 10 developing countries were chosen.
The developed countries are Canada, France, Hong Kong, Italy, Japan, Netherlands,
Spain, United Kingdom and USA. The developing countries are: Chile, Hungary, India,
Kenya, Korea, Mexico, Poland, Singapore, Turkey and Zimbabwe. These countries cover
most geographic regions with at least one developed country selected from Europe, North
America and the East. Complete data were not available for Australia or New Zealand.
Developing countries were selected from South America (Chile), the South East Asian
region, Eastern Europe and Africa. Although these countries have been classified as
developing, they show substantial differences in level of development.
For the purpose of the study two consistent data sets were constructed. A large
panel data set consisting of 27 sectors (Table C.1) between 1970-98 for all countries.
Three sectors, tobacco (314), petroleum refineries (353) and miscellaneous petroleum
products (354) were dropped as the data were either missing or highly irregular. The
analysis is thus based on 24 industrial sectors. An aggregated data set was also
constructed using all countries other than Mexico for which a complete balanced data set
was not available. All variables were aggregated using share of total group trade (exports
+ imports) as weights. The data used to construct the panel data sets were obtained from
various sources. The variables and sources are briefly discussed below.
The calculation of RULC required sector level data on wages and salaries,
employment and value added. For the conversion of output and wages into comparable
currencies data on nominal exchange rates and PPP exchange rates were also required.
Depending on how value added data is deflated sector specific deflators or common
manufacturing level deflators for each country are required.
set. These data sets contain, amongst other variables, values for employment, real and
nominal output, value added and wages and salaries. As the OECD STAN data set does
not have data on many developing countries (other than Mexico and Korea) and data
were at times not available during the 1970s it was decided to focus on the UNIDO data
set. A comparison of RULC using the UNIDO and STAN data sets indicates that the
UNIDO based values over–estimate RULC compared to the STAN based values by
between 50 and 100 % (Table C.2).
These differences arise from different methods of obtaining the data. In particular,
the STAN data set includes employer-paid fringe benefits which gives rise to lower
estimates of RULC. For most countries these differences remained constant over time,
although the USA and Japan show rising differences in RULC. To test the robustness of
the econometric results, both the STAN and UNIDO based variables were used in the
estimations. The sign and level of the coefficients were similar, suggesting that the fixed
effects in the regressions capture much of the difference.
The UNIDO data for South Africa from 1994 appeared to have been estimated as
for most sectors the share of wages and salaries in total value added was constant. As
result value added, employment and wage data obtained from the Trade and Industrial
Policy Strategies (TIPS) were used.
Relative wages were constructed by dividing the total wage bill by employment.
The wage bill was first converted into a common currency using nominal exchange rates
obtained from the World Development Indicators.
The relative productivity variable was constructed by dividing real value added by
employment. This was then converted to a common currency using PPP exchange rates
for 1987. Unfortunately, PPP exchange rates are not available for South Africa
necessitating the construction thereof. PPPt is represented as:
P$ Pd
PPPt = e*0 0d t $
P0 Pt
64
where * P0$ is the equilibrium real exchange rate and Ptd and Pt$ are domestic and
e0 d
P0
foreign prices at time t. Given the equilibrium real exchange rate it is possible to calculate
PPPt for all t using the correct domestic and foreign prices. However in SA case we do
not have this information. Golub (2000) overcomes this problem by using the average
Manufacturing value added deflators calculated from nominal and real manufacturing
value added obtained from the World Development Indicators (WDI) were used as the
price indices. Ideally, one should utilize sector specific PPPs, but as these were not
available a common PPP for manufacturing as a whole was assumed. 38 While this may
distort level comparisons at the sector level, the trends in RULC will not be affected. In
the econometric analysis level differences are captured by the inclusion of sector specific
intercepts leaving the estimates of the slope coefficients unaffected.
Two approaches to deflating value added were pursued. In the first case sector
value added was deflated to base year 1987 using a common value added deflator for
each country derived from the WDI. In the second case, sector value added was deflated
using sector specific deflators calculated from the nominal and real output values and
indices in the UNIDO data set. This approach assumes that the output and value added
deflators are the same. The sector specific deflators were highly variable at times, and
various procedures were used to correct the data (see later). The correlation coefficient
between the common value added deflator and the sector specific deflator exceeded 0.8 in
most cases. The exception was Italy were the correlation coefficient was below 0.3 for
almost all sectors. 39 RULC values calculated using the value added deflator were lower
than those calculated using the sector specific deflator prior to 1980, but higher during
38
As Golub and Hsieh (2000) show the use of different PPPs give rise to different results. The signs
however were not reversed.
39
Other exceptions were Japan: Electrical machinery (-0.03) and Professional goods (-0.66), Netherlands:
Rubber products (-0.86), and UK: Miscellaneous petroleum products (-0.01).
65
the 1990s (Table C.3). The sector specific deflator based RULC depict a greater decline
in RULC and thus improvement in competitiveness over the period.
It was decided to utilize the sector specific deflators, despite their greater
variability, as it was felt that price trends at the sector level deviated substantially from
the aggregate for manufacturing as a whole. Further, the estimated sign of RULC and
relative productivity in the econometric analysis using both deflators were similar. The
results were, therefore, robust to changes in the deflator used.
International trade data were required for the export functions and as weights for
the purpose of aggregating all countries into an aggregated panel data set. Trade data
were obtained from a number of sources. Customs and Excise bilateral trade data
classified according to the 8 digit Harmonised System was downloaded from the TIPS
Online data site. This was converted into the ISIC classification system consistent with
the UNIDO data set using a concordance file obtainable from the same site.
Unfortunately, this data are only available from 1988 precluding its use in any
econometric analysis to estimate the long run relationship between exports and RULC. It
is also only available for the Southern Africa Customs Union (SACU) and not South
Africa alone.
Total country exports and bilateral trade data between South Africa (essentially
SACU) and each country were then obtained from Canada World Trade Analyzer. This
data are available from 1980-98. The Canadian World Trade Analyzer data are classified
according to SITC and were converted to the UNIDO ISIC classification using a
concordance file constructed from one available on the Jon Haveman site
(http://www.eiit.org/ ) (as used by Edwards and Schoer, 2001). Unfortunately the bilateral
trade data were only complete for the developed countries. A further problem is that
South Africa did not report any data between 1974-91 (Tsikata, 1999). The data for this
period are then obtained from partner countries.
66
Total South African imports and exports according to sector between 1970-98
were obtained from TIPS and were used in the aggregate panel data set. The countries
within the sample account for 55 % of total SA exports and 55 % of total SA imports.
Although the countries sampled do not account for the entire trade, much of the missing
trade can be located within Europe where it is expected that RULC will move
concurrently with those of other developed countries.
In cases where real South African exports were used as the dependent variable,
the nominal data were converted into real values using output deflators. Export and
import price indices are not currently available for South Africa over the entire period
and for all sectors.
Tariff data at the sectoral level for entire the period 1970-98 are not available.
Aggregate tariff rates were calculated from customs revenue and imports for the entire
period. Fedderke and Vase (2001) have calculated effective rates of protection according
to sector using customs revenue for the period 1988-98. The values used in this study
have been taken directly from their paper.
C.4 Other
Other variables such as real investment in machinery and equipment, real capital
stock of machinery and equipment, real and nominal output, employment and capacity
utilization have been used in the study. This data were obtained from TIPS.
Firstly, the classification of industrial sectors changed over the years. This
resulted in large once off changes in the level of output or value added measured between
years. This was problematic if the real index and employment had not been similarly
adjusted. In most cases employment, value added and output (real and nominal) were
adjusted to be consistent with the reclassification of industrial sectors. At times the value
added variable had not been adjusted correctly. Changes in gross output were then used
to adjust the level of value added for that sector. The remainder of the series was adjusted
using value added growth rates derived from the original data.
Secondly, real output indices used for the construction of sector level deflators
were frequently only available for aggregated sub-sectors. Nominal output was similarly
aggregated in order to calculate a common deflator for all the sectors within the
aggregated group.
Thirdly, with the reclassification of industrial sectors the output, value added and
employment of certain sectors was aggregated (e.g. iron & steel (371) with non-ferrous
metal (372)). The aggregated sectors were disaggregated using the average share in the 3
to 5 years prior to the reclassification of sectors. Where possible alternative data sources
were used (STAN data for OECD countries). Employment levels were adjusted using
relative average wages prior to the reclassification.
Fourthly, data were sometimes missing for a number of sectors and countries for a
number of years. In cases where these covered periods of up to 3 years average annual
growth rates were used to adjust the data when no alternative sources were available. If
these periods exceeded 3 years then the country was either dropped or alternative data
were used to update the data. For example, the missing value added and output data for
France from 1993 was estimated using the STAN data set. Wages and salaries prior to
1977 were estimated in a similar manner.
In order to further check the data for outliers and level changes the following
assessment procedure was used:
68
(a) Sector level deflators for each country were analyzed for sudden shifts or
outliers. When nominal output and the real index clearly diverged as the result of
once off shifts, the initial level of the real index was adjusted and the original growth
rates were used to adjust the trend in the real index. Compound growth rates were
used to replace outlying deflators.
(b) Figures of RULC for each country were then analyzed for outliers and
sudden shifts in levels. The raw data was re-analyzed in problematic cases. This
process was useful in identifying errors in the value added data.
69
Table 3.1
South African Exports by Industry as a Share of Total Exports
Table 4.1
Growth Rates of South African Labor Productivity and Total Factor Productivity
(annual averages), Output Deflators
1980-90 1991-97
LP TFP LP TFP
Table 4.2
Growth Rates of South African Labor Productivity and Total Factor Productivity
(annual averages) Aggregate Manufacturing, Value Added Deflator
1980-90 1991-97
LP TFP LP TFP
Table 4.3
Growth Rates of Productivity and Capital-Labor Ratio,
Aggregate Manufacturing, Annual Average
1980-90
Labor Total Factor
Capital Stock Employment Capital/Labor Productivity Productivity
Canada 5.6% 0.1% 5.5% 1.4% -0.7%
France 2.0% -1.6% 3.6% 2.3% 0.9%
Italy -0.3% -1.9% 1.6% 2.0% 1.4%
Japan 3.3% 0.9% 2.4% 3.6% 2.7%
Spain 2.2% -1.9% 4.2% 2.9% 1.3%
UK 0.2% -3.2% 3.5% 3.5% 2.2%
USA 1.7% -1.1% 2.5% 2.7% 1.6%
1991-97
Labor Total Factor
Capital Stock Employment Capital/Labor Productivity Productivity
Canada 1.5% -1.1% 2.6% 3.8% 2.7%
France -0.2% -2.3% 2.1% 4.5% 2.8%
Italy 2.2% 0.2% 2.0% 3.2% 2.8%
Japan 2.6% -1.5% 4.2% 2.5% 0.8%
Spain 5.6% 1.8% 3.7% 1.7% -0.4%
UK 2.0% -2.1% 4.2% -0.8% -2.4%
USA 3.2% 0.1% 3.2% 5.2% 3.9%
Table 5.1.a
South African Relative Wages Against Developed Countries, Weighted Averages of
Countries
Table 5.1.b
South African Relative Productivity Against Developed Countries, Weighted Averages of
Countries
Table 5.1.c
South African Relative Unit Labor Cost Against Developed Countries, Weighted
Averages of Countries
Table 5.2.a
South African Relative Wages Against Developing Countries, Weighted Averages of
Countries
Table 5.2.b
South African Relative Productivity Against Developing Countries, Weighted Averages
of Countries
Table 5.2.c
South African Relative Unit Labor Cost Against Developing Countries, Weighted
Averages of Countries
Table 7.1
South African price, income and other export elasticities for manufacturing
Table 9.1
Labor cost competitiveness and the structure of South African real exports
RULC -4.66 0.03 -5.74 0.00 -3.71 0.10 0.20 0.87 0.47 0.60
2.00 1.13 2.18 1.24 0.87
constant 20.87 0.00 21.85 0.00 20.36 0.00 19.90 0.00 20.54 0.00
0.72 0.66 0.80 0.33 0.27
R2 0.20 0.54 0.12 0.00 0.01
obs 696
F(1,22) 5.42 0.03 25.72 0.00 2.89 0.10 0.03 0.87 0.30 0.60
lxreal Coef. P>t Coef. P>t Coef. P>t Coef. P>t Coef. P>t
Rel Prod 11.53 0.00 5.59 0.00 9.88 0.01 -0.12 0.95 -0.44 0.67
2.77 1.21 3.49 1.92 1.00
Rel Wage -10.44 0.01 -5.00 0.01 -10.49 0.05 1.44 0.56 2.67 0.08
3.91 1.76 5.07 2.45 1.44
constant 14.81 0.13 17.16 0.02 22.81 0.07 13.91 0.09 10.56 0.03
9.49 6.84 11.89 7.72 4.67
R2 0.51 0.51 0.30 0.03 0.19
obs 240
F(2,21) 10.74 0.00 10.76 0.00 4.55 0.02 0.31 2.45 0.11
Note: Dependent variable is log total real exports. All variables are in logs. The standard errors are in
Italics below the estimated coefficients.
81
Table 9.2
Labor cost competitiveness and the structure of trade according to country, log total
exports ratio (Xisa/Xij) as dependent variable.
1980-89 1990-98
(1) (2) (3) (4) (5) (6)
RULC Rel Rel RTFP Rel Wage RULC Rel Rel RTFP Rel
Prod Wage Prod Wage Wage
Developed countries(1)
France -3.24* 3.32* -4.43* -1.82***
Hong Kong -2.51*** 2.68***
Italy -4.79* 4.74* -5.69*
Japan -3.73** -3.65** -2.37***
Netherlands -3.57* 2.45* -3.00* -2.45**
Spain
UK -3.17* 3.16* -4.77* 3.16** -4.16*
USA -2.32** 1.88** -3.80* -3.56* 1.08**
*
Table 10.1
Export function using aggregated panel data set
Long run
RULC -2.45 -2.34
Rel Prod 1.93 1.95
Rel Wage -3.18 -2.99
Capacity -6.18 -4.16
World income 2.35 1.67 2.16 1.64
F_test 1503 0.00 1211 0.00 202 0.00 1230 0.00 1028 0.00
obs= 672
Sectors = 24
Notes: 18 countries included in analysis. Variables are in logs. The standard errors are in Italics below the
estimated coefficients.
(1) Dynamic export demand function (Equation 14).
(2) Dynamic export demand function (Equation 14) with decomposed RULC.
(3) Static reduced form export function (Equation 12).
(4) Dynamic reduced form export function (Equation 12).
(5) Dynamic reduced form export function with decomposed RULC (Equation 12)
83
Table 10.2
Sector specific estimation and diagnostic results based on ARDL(1,0,0,0) specification.
Other coeff -0.422 -1.268 -2.743 0.515 -0.333 0.124 0.221 5.661 2.110 0.962
manufacturing Prob [.000] [.000] [.188] [.541] [.032] [.643] [.027] [.158]
85
Table 10.3
Reduced form export functions for broadly defined manufacturing sectors, log real
exports as dependent variable
Table 10.4
Export functions according to developed and developing country classification
Table 10.5a
Country specific export demand functions for developed countries in sample, 1980-97.
Relative Unit Labor Costs
Total Canada France Hong Kong Italy Japan Netherlands Spain UK USA
Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t|
RULC
Export(-1) 0.56 0.00 0.45 0.00 0.59 0.00 0.48 0.00 0.51 0.00 0.62 0.00 0.53 0.00 0.57 0.00 0.65 0.00 0.60 0.00
RULC -0.09 -0.04 -0.28 -0.24 -0.59 0.00 -0.06 -0.63 -0.26 -0.04 -0.40 -0.02 -0.12 -0.42 0.14 -0.36 -0.09 -0.31 -0.11 -0.45
Sanctions -0.09 0.00 -0.39 0.00 -0.17 -0.03 0.13 -0.13 -0.12 -0.14 -0.05 -0.55 -0.09 -0.24 0.06 -0.59 -0.04 -0.43 -0.36 0.00
World 0.44 0.00 0.51 0.00 0.50 0.00 0.41 0.00 0.34 -0.01 0.21 0.00 0.80 0.00 0.46 0.00 0.44 0.00 0.39 0.00
imports
Long run
RULC -0.20 -0.50 -1.43 -0.11 -0.53 -1.06 -0.26 0.33 -0.25 -0.26
World 0.99 0.92 1.21 0.78 0.70 0.56 1.69 1.07 1.25 0.96
imports
R2 within 0.50 0.30 0.58 0.49 0.45 0.53 0.59 0.55 0.70 0.53
between 0.94 0.89 0.95 0.93 0.97 0.98 0.92 0.94 0.97 0.94
overall 0.84 0.73 0.84 0.81 0.86 0.91 0.79 0.80 0.90 0.84
F(4,3533) 881 0 41 0 139 0 94 0 79 0 109 0 140 0 103 0 248 0 117 0
obs 3762 405 433 425 412 408 424 361 450 444
88
Table 10.5b
Country specific export demand functions for developed countries in sample, 1980-97.
Relative productivity, relative wage and exchange rate
Total Canada France Hong Kong Italy Japan Netherlands Spain UK USA
Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t|
Export(-1) 0.55 0.00 0.43 0.00 0.58 0.00 0.47 0.00 0.50 0.00 0.62 0.00 0.53 0.00 0.56 0.00 0.66 0.00 0.57 0.00
Rel Prod 0.01 0.88 0.69 0.02 0.29 0.21 0.03 0.84 0.11 0.46 0.50 0.02 -0.11 0.57 -0.15 0.44 -0.04 0.71 -0.03 0.86
Rel Wage -0.31 0.00 -1.00 0.01 -0.95 0.00 -0.26 0.31 -0.34 0.15 -0.96 0.00 -0.11 0.62 -0.04 0.91 -0.11 0.50 -0.40 0.03
Erate 0.56 0.00 0.58 0.11 1.01 0.00 0.40 0.07 1.23 0.00 0.75 0.00 0.61 0.02 0.50 0.17 0.52 0.00 0.83 0.00
Sanctions -0.13 0.00 -0.43 0.00 -0.20 0.02 0.15 0.08 -0.21 0.02 -0.24 0.02 -0.09 0.25 0.01 0.90 -0.06 0.22 -0.39 0.00
World 0.19 0.00 0.87 0.00 0.33 0.21 0.26 0.03 -0.37 0.11 0.34 0.01 0.12 0.64 0.17 0.38 0.01 0.95 -0.01 0.94
imports
Long run
Rel Prod 0.02 1.22 0.68 0.05 0.22 1.33 -0.23 -0.35 -0.12 -0.07
Rel Wage -0.69 -1.77 -2.25 -0.49 -0.69 -2.54 -0.23 -0.09 -0.31 -0.94
Erate 1.25 1.03 2.40 0.76 2.44 1.98 1.29 1.14 1.53 1.94
World 0.42 1.53 0.79 0.48 -0.74 0.89 0.26 0.39 0.03 -0.03
imports
R2 within 0.51 0.31 0.59 0.49 0.47 0.55 0.60 0.56 0.71 0.55
between 0.83 0.80 0.97 0.96 0.84 0.97 0.97 0.96 0.99 0.99
overall 0.75 0.67 0.85 0.82 0.69 0.90 0.78 0.80 0.90 0.84
F(6,3531) 607 0 29 0 96 0 63 0 57 0 76 0 96 0 69 0 172 0 84 0
obs 3762 405 433 425 412 408 424 361 450 444
89
Table 10.5c
Country specific export demand functions for developed countries in sample, 1980-97.
Relative TFP, relative wage and exchange rate
Total Canada France Hong Kong Italy Japan Netherlands Spain UK USA
Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t|
Export(-1) 0.50 0.00 0.43 0.00 0.50 0.00 0.38 0.00 0.48 0.00 0.60 0.00 0.46 0.00 0.48 0.00 0.66 0.00 0.54 0.00
TFP 0.02 0.72 0.39 0.19 -0.06 0.85 0.18 0.20 0.04 0.82 0.46 0.06 -0.08 0.76 0.05 0.83 -0.01 0.93 -0.05 0.77
Rel Wage -0.36 0.00 -0.73 0.04 -0.67 0.03 0.08 0.76 -0.33 0.18 -0.82 0.00 -0.24 0.42 -0.40 0.27 -0.29 0.04 -0.53 0.00
Erate 0.61 0.00 0.43 0.26 0.87 0.01 1.08 0.00 1.19 0.00 0.61 0.01 1.29 0.00 0.62 0.12 0.61 0.00 1.06 0.00
Sanctions -0.11 0.00 -0.41 0.00 -0.13 0.19 0.09 0.31 -0.21 0.03 -0.17 0.12 -0.59 0.02 0.08 0.49 -0.06 0.17 -0.39 0.00
World 0.18 0.00 0.72 0.01 0.32 0.31 0.04 0.79 -0.32 0.20 0.27 0.07 0.02 0.96 0.16 0.47 0.04 0.80 -0.06 0.72
imports
Long run
TFP 0.05 0.67 -0.11 0.29 0.08 1.15 -0.15 0.09 -0.03 -0.12
Rel Wage -0.73 -1.27 -1.33 0.13 -0.63 -2.06 -0.44 -0.77 -0.87 -1.15
Erate 1.24 0.74 1.72 1.73 2.28 1.54 2.38 1.18 1.80 2.32
World 0.36 1.25 0.64 0.06 -0.61 0.68 0.04 0.30 0.11 -0.14
imports
R2 within 0.45 0.27 0.46 0.51 0.43 0.53 0.45 0.46 0.73 0.52
between 0.80 0.81 0.96 0.98 0.86 0.97 0.96 0.97 0.99 0.97
overall 0.72 0.68 0.83 0.76 0.70 0.90 0.72 0.78 0.89 0.81
F(6,3046) 407 0 22 0 47 0 54 0 45 0 65 0 37 0 42 0 169 0 69 0
obs 3267 376 356 337 382 382 298 324 408 404
Notes: Bilateral export data taken from Canadian World Trade Analyzer. Converted from SITC to SIC using concordance file adapted from J. Haveman.
Exports are converted to Rands using R/US$ exchange rate and then converted into real values by deflating with sectoral output deflators.
90
Appendix
Table A1a
Country specific export functions using total export ratios (Xisa/Xij) as dependent variable, 1980-98.
Relative Unit Labor Costs
Total Canada Chile France Hong- Hungary India Italy Japan Kenya South Mexico Netherla Poland Singap Spain Turkey UK USA Zimbab
Kong Korea nds ore we
X ratio(-1) 0.66 0.80 0.34 0.78 0.77 0.82 0.70 0.79 0.78 0.27 0.84 0.72 0.76 0.82 0.83 0.78 0.64 0.78 0.72 0.39
RULC -0.06 -0.06 -0.12 0.08 0.04 -0.19 -0.11 -0.06 0.10 0.20 -0.08 -0.29 0.13 -0.15 -0.13 0.00 0.00 0.03 0.09 0.11
Sanctions -0.06 0.05 -0.31 -0.09 -0.09 0.09 -0.03 -0.07 -0.05 -0.30 -0.05 0.09 -0.14 0.12 -0.04 -0.04 0.13 -0.07 -0.07 -0.03
Capacity -0.34 -0.29 -0.53 -0.51 -0.20 -0.42 -0.01 -0.32 -0.61 1.26 0.01 1.11 -0.31 -0.15 -0.78 -0.31 0.00 -0.50 -0.81 0.49
World 0.01 0.15 -0.40 0.24 0.04 -0.03 0.08 0.13 0.25 0.03 0.11 -0.06 0.23 -0.01 0.04 0.07 0.05 0.18 0.11 -0.14
imports
Long run
RULC -0.18 -0.27 -0.18 0.38 0.19 -1.06 -0.39 -0.28 0.43 0.27 -0.46 -1.02 0.53 -0.83 -0.74 -0.02 0.00 0.14 0.31 0.18
World 0.01 0.72 -0.61 1.07 0.17 -0.14 0.26 0.61 1.13 0.03 0.66 -0.21 0.96 -0.05 0.23 0.30 0.14 0.84 0.40 -0.23
imports
R2 within 0.47 0.65 0.76 0.69 0.72 0.69 0.51 0.68 0.80 0.10 0.76 0.61 0.68 0.63 0.68 0.63 0.58 0.67 0.58 0.24
between 1.00 0.96 0.72 0.91 1.00 0.99 0.98 0.98 0.93 0.92 0.97 0.96 0.88 0.98 0.99 0.99 0.99 0.95 0.97 0.62
overall 0.95 0.90 0.73 0.86 0.96 0.91 0.91 0.94 0.91 0.41 0.93 0.84 0.83 0.88 0.94 0.93 0.86 0.90 0.91 0.39
F(5,8028) 1447 154 262 190 216 184 88 182 339 10 264 117 181 146 177 144 116 171 117 26
obs 8507 450 450 450 450 450 450 450 450 450 450 407 450 450 450 450 450 450 450 450
91
Table A1b
Country specific export functions using total export ratios (Xisa/Xij) as dependent variable, 1980-98.
Relative productivity, relative wage and exchange rate
Total Canada Chile France Hong- Hungary India Italy Japan Kenya South Mexico Netherla Poland Singap Spain Turkey UK USA Zimbab
Kong Korea nds ore we
X ratio(-1) 0.65 0.80 0.29 0.78 0.78 0.80 0.68 0.78 0.78 0.26 0.81 0.65 0.75 0.80 0.81 0.77 0.63 0.79 0.72 0.44
Rel Prod 0.00 -0.09 0.01 -0.10 -0.09 0.22 0.15 0.07 -0.05 -0.13 0.06 0.13 -0.17 0.08 0.01 0.00 -0.07 -0.08 -0.18 0.34
Rel Wage -0.09 -0.02 -0.87 0.02 0.06 -0.27 -0.33 -0.31 -0.04 0.18 -0.42 -0.21 0.10 -0.14 -0.19 -0.15 -0.12 0.08 0.06 -0.74
erate 0.11 0.23 0.03 -0.01 0.11 0.26 -0.03 0.03 -0.01 -0.50 0.04 0.35 -0.14 0.10 0.34 -0.01 0.11 -0.05 0.05 -0.35
Sanctions -0.06 0.04 -0.26 -0.09 -0.09 0.09 0.01 -0.09 -0.08 -0.26 -0.03 0.21 -0.14 0.07 -0.05 -0.05 0.12 -0.07 -0.07 0.24
World 0.01 -0.03 -0.56 0.21 -0.04 -0.05 0.11 0.20 0.28 -0.07 0.11 0.03 0.28 -0.10 -0.12 0.12 0.01 0.19 0.02 -0.36
imports
Long run
Rel Prod 0.00 -0.45 0.02 -0.46 -0.39 1.14 0.47 0.30 -0.23 -0.17 0.34 0.38 -0.67 0.40 0.04 0.01 -0.18 -0.36 -0.63 0.61
Rel Wage -0.25 -0.10 -1.21 0.07 0.25 -1.40 -1.03 -1.42 -0.17 0.25 -2.19 -0.60 0.39 -0.70 -0.96 -0.66 -0.33 0.35 0.21 -1.32
erate 0.33 1.16 0.04 -0.03 0.49 1.34 -0.10 0.12 -0.03 -0.68 0.21 1.00 -0.54 0.49 1.78 -0.04 0.29 -0.26 0.19 -0.62
World 0.03 -0.14 -0.78 0.95 -0.19 -0.25 0.35 0.92 1.27 -0.09 0.60 0.10 1.12 -0.51 -0.60 0.52 0.04 0.87 0.06 -0.64
imports
R2 within 0.48 0.65 0.80 0.69 0.72 0.69 0.51 0.69 0.80 0.11 0.77 0.62 0.68 0.64 0.69 0.63 0.59 0.67 0.59 0.29
between 0.99 1.00 0.60 0.93 1.00 0.98 0.95 0.97 0.93 0.89 0.97 0.99 0.84 0.95 0.98 0.98 1.00 0.95 1.00 0.07
overall 0.94 0.93 0.65 0.88 0.96 0.91 0.89 0.93 0.91 0.37 0.93 0.85 0.80 0.86 0.93 0.92 0.85 0.91 0.93 0.15
F(6,8024) 1212 132 281 158 182 154 74 157 280 8 229 103 151 123 155 121 99 141 99 29
obs 8504 450 450 450 450 450 450 450 450 447 450 407 450 450 450 450 450 450 450 450
92
Table A1c
Country specific export functions using total export ratios (Xisa/Xij) as dependent variable, 1980-98.
Relative TFP, relative wage and exchange rate
Total Canada Chile France Hong- Hungary India Italy Japan Kenya South Mexico Netherla Poland Singap Spain Turkey UK USA Zimbab
Kong Korea nds ore we
X ratio(-1) 0.77 0.79 0.27 0.73 0.74 0.66 0.64 0.77 0.75 0.76 0.42 0.61 0.79 0.75 0.61 0.77 0.70 0.42
TFP 0.00 0.06 0.06 -0.05 -0.14 0.33 0.25 0.12 0.06 0.19 0.03 0.07 0.07 0.07 -0.05 0.05 -0.10 0.34
Rel Wage 0.01 -0.08 -0.83 0.14 0.01 -0.47 -0.36 -0.28 -0.09 -0.40 -0.32 0.06 -0.20 -0.14 -0.13 0.00 -0.02 -0.13
erate 0.12 0.27 0.04 -0.01 0.08 0.23 -0.14 0.00 0.02 -0.01 0.83 0.13 0.31 -0.03 0.13 0.01 0.09 0.01
Sanctions -0.07 0.01 -0.24 -0.13 -0.10 0.08 0.02 -0.11 -0.08 -0.04 0.99 -0.50 -0.05 -0.07 0.09 -0.07 -0.07 0.11
World 0.04 -0.01 -0.57 0.14 -0.04 -0.03 0.03 0.20 0.28 0.11 0.18 0.33 -0.10 0.11 0.05 0.16 0.07 -0.32
imports
Long run
TFP 0.01 0.26 0.08 -0.20 -0.52 0.97 0.70 0.51 0.23 0.79 0.05 0.17 0.33 0.26 -0.12 0.22 -0.33 0.59
Rel Wage 0.05 -0.36 -1.14 0.50 0.03 -1.37 -1.01 -1.23 -0.36 -1.64 -0.56 0.14 -0.94 -0.56 -0.33 0.01 -0.06 -0.23
erate 0.50 1.27 0.06 -0.02 0.29 0.67 -0.39 0.00 0.07 -0.03 1.44 0.34 1.48 -0.11 0.34 0.05 0.30 0.02
World 0.17 -0.04 -0.78 0.53 -0.14 -0.09 0.08 0.89 1.11 0.44 0.30 0.85 -0.50 0.43 0.13 0.69 0.24 -0.55
imports
R2 within 0.63 0.63 0.81 0.57 0.68 0.50 0.48 0.66 0.76 0.73 0.65 0.49 0.67 0.62 0.57 0.64 0.56 0.33
between 0.91 1.00 0.65 0.96 1.00 0.96 0.98 0.96 0.93 0.97 0.90 0.79 0.99 0.98 0.99 0.97 0.99 0.19
overall 0.87 0.93 0.69 0.90 0.96 0.90 0.92 0.93 0.91 0.93 0.78 0.74 0.94 0.92 0.85 0.92 0.93 0.25
F(6,3240) 912 108 264 74 119 48 39 125 196 169 47 44 126 98 84 111 79 27
obs 3461 408 408 360 359 312 288 408 408 408 172 312 408 390 408 408 408 360
Notes: Export data taken from Canadian World Trade Analyzer. Converted from SITC to SIC using concordance file adapted from J. Haveman. Exports are in
current US$. This implicitly assumes a common deflator for each sector across countries.
Bold variables are significant at the 10 % level.
93
Table C.1
Industrial sectors
Table C.2
Comparison of differences in RULC between STAN and UNIDO databases
(RULC UNIDO /RULC STAN)
Figure 3.1
Exports and Net Exports, Selected Industries
0.5 1
0.45 0.9
0.4 0.8
0.35 0.7
0.3
0.6
0.25
0.5
0.2
0.4
0.15
0.3
0.1
0.2
0.05
0.1
0
1970 1974 1978 1982 1986 1990 1994 1998 0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
Textiles Wearing apparel
0.4 0.6
0.35
0.5
0.3
0.4
0.25
0.2 0.3
0.15
0.2
0.1
0.1
0.05
0
0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1974 1978 1982 1986 1990 1994 1998
Figure 3.2
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
-0.25
-0.30
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
0.20 0.00
0.00 -0.10
-0.20 -0.20
-0.40 -0.30
-0.60 -0.40
-0.80 -0.50
-1.00 -0.60
-1.20 -0.70
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
Fabricated metal products Machinery, except electrical Industrial chemicals Other chemicals
Figure 4.1
South African Total Factor Productivity and Labor Productivity
Aggregate Manufacturing, Relative to Other Countries
Canada Italy
0.70 0.70
0.60 0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
Japan
France
0.50
0.60
0.45
0.50 0.40
0.35
0.40
0.30
0.30 0.25
0.20
0.20 0.15
0.10
0.10
0.05
0.00 0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
0.45 0.70
0.40
0.60
0.35
0.30 0.50
0.25
0.40
0.20
0.30
0.15
0.10
0.20
0.05
0.10
0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
relative TFP
relative labor productivity relative TFP relative labor productivity
98
Korea Singapore
2.00
1.20
1.50 1.00
0.80
1.00
0.60
0.40
0.50
0.20
0.00 0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
Spain Chile
0.80
0.80
0.70
0.70
0.60
0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10
0.10
0.00
0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
1979 1981 1983 1985 1987 1989 1991 1993 1995
Zimbabwe India
2.50
12.00
2.00 10.00
8.00
1.50
6.00
1.00
4.00
0.50 2.00
0.00 0.00
1979 1981 1983 1985 1987 1989 1991 1993 1995 1979 1982 1985 1988 1991 1994 1997
Figure 4.2
South African Relative Capital-Labor Ratios, Total Manufacturing
3.5
2.5
1.5
0.5
0
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
SA/India SA/Korea SA/Zimbabwe SA/Chile
1.2
0.8
0.6
0.4
0.2
1979 1981 1983 1985 1987 1989 1991 1993 1995 1997
Figure 5.1
South African Relative Wages and Productivity, by Industry
8.00 0.40
6.00 0.30
4.00 0.20
2.00 0.10
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
3.00
2.50 0.60
2.00 0.50
0.40
1.50
0.30
1.00
0.20
0.50
0.10
0.00
0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
3.50
3.00 0.60
2.50 0.50
2.00 0.40
1.50 0.30
1.00 0.20
0.50 0.10
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
0.80
4.00
0.70
3.00 0.60
0.50
2.00 0.40
0.30
1.00 0.20
0.10
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
8.00
7.00 1.20
6.00 1.00
5.00 0.80
4.00
0.60
3.00
0.40
2.00
1.00 0.20
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
6.00
1.00
5.00
0.80
4.00
0.60
3.00
0.40
2.00
0.20
1.00
0.00
0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
4.00
3.50 0.60
3.00 0.50
2.50 0.40
2.00
0.30
1.50
0.20
1.00
0.50 0.10
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
5.00
0.60
4.00
0.50
3.00 0.40
0.30
2.00
0.20
1.00
0.10
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1974 1978 1982 1986 1990 1994 1998
10.00
1.50
8.00
1.00
6.00
4.00 0.50
2.00
0.00
0.00
1970 1974 1978 1982 1986 1990 1994 1998
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
6.00
5.00 1.20
1.00
4.00
0.80
3.00
0.60
2.00
0.40
1.00 0.20
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1974 1978 1982 1986 1990 1994 1998
8.00
0.80
6.00 0.60
4.00 0.40
2.00 0.20
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 1970 1974 1978 1982 1986 1990 1994 1998
Figure 6.1
Relative Unit Labor Costs and Trade Flows, Cross Section
0.45
0.40 1.00
0.35 0.50
0.30
0.00
0.25
-0.50
0.20
-1.00
0.15
-1.50
0.10
0.05 -2.00
0.00 -2.50
0.80 1.00 1.20 1.40 1.60 1.80 2.00 2.20 2.40
0.80 1.00 1.20 1.40 1.60 1.80 2.00 2.20 2.40
0.60
1.00
0.50 0.50
0.00
0.40 -0.50
-1.00
0.30
-1.50
0.20 -2.00
-2.50
0.10 -3.00
-3.50
0.00 0.80 1.00 1.20 1.40 1.60
1.00
0.60
0.50
0.50
0.00
0.40 -0.50
-1.00
0.30
-1.50
0.20
-2.00
0.10 -2.50
-3.00
0.00
0.60 0.80 1.00 1.20 1.40 1.60 1.80
0.60 0.80 1.00 1.20 1.40 1.60 1.80
RULC average
RULC average
104
Figure 6.2
Relative Unit Labor Costs and Trade, Selected Industries
1.00 1.00
0.25 0.20
0.95 0.95
0.85 0.85
0.15 0.10
0.80 0.80
0.70 0.70
0.05 0.00
0.65 0.65
1.20 0.60
1.20 0.60
1.10 0.50
0.50
1.10
1.00 0.40 0.40
1.00
0.30
0.90 0.30
0.90 0.20
Exports/Output, Non-Ferrous
Metals Net Trade/Output, Non-Ferrous
0.20 0.10
0.50
0.50
0.00
0.10
0.00 -0.10
0.00 0.00 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1.20 0.25
1.20 0.00
1.10
0.20 1.10 -0.05
1.00
1.00 -0.10
0.15
0.90
0.90 -0.15
0.10
0.80
0.80 -0.20
0.05
0.70
0.70 -0.25
1.00 0.10
1.00 0.00
0.45 0.10
2.50 2.50
0.40 0.05
0.35
2.00 2.00 0.00
0.30
-0.05
1.50 0.25 1.50
-0.10
0.20
1.00
0.15 1.00 -0.15
0.10 -0.20
0.50
0.50
0.05 -0.25
0.00 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 0.00 -0.30
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1.40 0.30
1.40 0.20
1.30
0.25 1.30 0.15
0.90
0.10 0.90 -0.05
1.10 -0.20
0.40 1.00
1.00 -0.30
0.30 0.90
0.90 -0.40
0.80
0.20 -0.50
0.80
0.70
0.10 -0.60
0.70
0.60 -0.70
0.60 0.00 1970 1972 19741976 1978 19801982 1984 19861988 1990 19921994 1996 1998
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998
1.50 0.18
1.50
0.02
1.40 0.16
1.40
0.00
1.30 0.14
1.30
-0.10
0.70 0.02 0.70
0.60 -0.12
0.60 0.00
1970 1972 19741976 1978 1980 1982 19841986 1988 1990 19921994 1996 1998
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998
inverted RULC average exports/output
inverted RULC average net trade/output
107
1.00 0.18
1.00 0.00
0.16
0.90 -0.20
0.90
0.14
-0.40
0.80 0.12 0.80
-0.60
0.10
0.70
0.70
-0.80
0.08
0.60
0.60 0.06 -1.00
0.04 0.50
-1.20
0.50
0.02
0.40 -1.40
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998
0.40 0.00
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998
-0.20
0.35 1.10
1.10
0.30 -0.40
1.00
1.00
0.25
-0.60
0.70
0.05 0.70
-1.20
0.60 0.00
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 0.60 -1.40
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998
1.30 0.20
1.30 0.00
0.18
1.20 -0.10
1.20
0.16
-0.20
1.10 0.14 1.10
-0.30
0.12
1.00 1.00 -0.40
0.10
0.90 0.90 -0.50
0.08
-0.60
0.80 0.06 0.80
-0.70
0.04
0.70 0.70
-0.80
0.02
0.60 -0.90
0.60 0.00
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998