A),P3 30 23
POLICY RESEARCH WORKING PAPER 3 023
Productivity Growth and Product Variety
Gains from Imitation and Education
Douglas M. Addison
The World Bank
African Technical Families
Poverty Reduction and Economic Management 3
April 2003
l POLICY RESEARCH WORKING PAPER 3023
Abstract
rs there a correlation between productivity and product development (R&D) employment, macroeconomic
variety? Certainly it appears that the rich countries are stability, and domestic security. These results are robust
more productive and have more product variety than the to the addition and subtraction of various explanatory
poor nations. In fact, the relationship is quite strong variables but fragile with respect to an influential data
when measured in levels. Does this same correlation hold point for Venezuela. Industrial nations tend to generate
up when measured in growth rates? If so, can poor most of their productivity gains through R&D
countries imitate the success of the rich? employment in a stable environment that results in better
Addison provides theoretical and empirical reasons to production processes and product quality. In contrast,
believe the answer to both questions is yes. Recent the largest source of productivity growth in developing
economic theory suggests that rising variety in factor countries is product variety imitation while instability
inputs can help avoid diminishing marginal returns. takes away from productivity.
Product variety can also sustain learning-by-doing which Addison tests various explanations for growth in
would otherwise be exhausted in a fixed number of variety. The results show that nations furthest from the
products. Finally, invention or imitation adds to the frontier of observable variety tend to imitate fastest, with
stock of non-rival knowledge. There have been only two the ability to imitate being improved by educational
previous empirical tests of the correlation between attainment and by productivity gains. This could be a
growth in product variety and productivity growth. Both source of hope for small, less developed nations. Growth
were affirmative but neither examined a wide range of in market size was not correlated with growth in variety,
developing countries and neither looked deeper to test though this may be due to a rather short sample period
what might drive product variety. of only eight years.
This research is based on a cross-country sample of 29 In addition to the empirical testing, Addison also
countries (13 rich and 16 poor). The data display a contributes to a general discussion of measurement
statistically significant and positive relationship between concepts and measurement issues related to product
growth in product variety and productivity growth when variety and sets out an agenda for further research.
condition on other variables such as research and
This paper is a product of Poverty Reduction and Economic Management 3, Africa Technical Families. Copies of the paper
are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Douglas Addison, room
J7-114, telephone 202-473-1188, fax 202-473-8466, email address daddison@worldbank.org. Policy Research Working
Papers are also posted on the Web at http://econ.worldbank.org. April 2003. (25 pages)
The Policy Research Working Paper Senes disseminates the findings of work in progress to encourage the exchange of ideas about
development issues. An objective of the series is to get the findiiigs out quickly, even if the presentations are less than fully polished The
papers carry the names of the authors and should be cited accordingly The findings, interpretattons, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Research Advisory Staff
PRODUCTIVITY GROWTH AND PRODUCT VARIETY:
GAINS FROM IMITATION AND EDUCATION
Douglas M. Addison
DAddison(WorldBank.Org
Key words: growth, productivity, imitation, education, variety
JEL Classification: 0140, 0150, 0400.
3
PRODUCTIVITY GROWTH AND PRODUCT VARIETY:
GAINS FROM IMITATION AND EDUCATION
A. Introduction
1. Casual observation suggests there could be a positive correlation between product variety
and productivity in manufacturing. The manufacturing sectors in developed nations are
diversified across thousands of product categories while those in the developing countries (DCs)
are not. For example, UNIDO data show that Venezuela was producing manufactures in 73 out
of 81 broad industrial categories in 1982 while the UK was producing in all categories. This gap
in the variety of product categories is much wider when one looks at the Standard International
Trade Classification (Revision 2) data for manufactured export variety: Venezuela exported in
187 out of 1,600 categories while the UK exported in 1,582. Similarly, total factor productivity
(TFP) in most DCs is generally lower than in the more developed nations. Hall and Jones (1998)
find that TFP in the USA was nearly nine and a half times greater than it was in China in 1988
and almost 13 times higher than in Zambia. On average, TFP in the developed countries is twice
that of the developing nations.
2. In fact, it is easy to find a strong and statistically significant positive correlation between
product variety and TFP across countries in levels. The more difficult question is whether a
similar relationship can be established in growth rates. This paper suggests the answer is yes, if
the relationship is conditioned by other variables.
3. The rest of the text is organized as follows. Section B provides a brief review of the
theoretical and empirical literature. Section C summarizes the new empirical work presented
here. Section D lays out the framework for hypothesis testing. The data are briefly described in
Section E. The correlation between growth in product variety and growth in productivity is
pursued in Section F. The ramifications of the findings for development policy are explored in
Section G. Section H concludes with a general summary of the paper and a list of outstanding
empirical issues.
B. Literature Review
4. Theory. Several models of endogenous growth based on research and development
(R&D) suggest a link between product variety and productivity. Their main attraction is the
explicit acknowledgement that changes in product variety are the result of deliberate decisions to
make costly investments in human capital, and that these investments have an impact on growth.
Investments in R&D generate knowledge, some of which is non-rival. The expanding stock of
non-rival knowledge is an important input in production that can increase TFP.
5. There are a number of R&D based models of endogenous growth to choose from. Many
depict an expansion in the variety of products, or horizontal innovation, as being directly
correlated With TFP growth. An example can be found in Romer (1990) where TFP is a function
4
of variety in intermediate goods. The models by Young (1993) and Aghion and Howitt (1998)
are particularly appealing because they highlight an important link between new products
resulting from R&D and production cost reductions resulting from learning-by-doing (LBD).
Cost reductions from LBD allow producers to maintain some degree of profitability but the gains
from LBD decline with production experience. In order to open up new profit opportunities,
producers will eventually be compelled to introduce new products. These new goods then
provide additional opportunities for LBD. Some of these models are fairly rich in policy
implications, notably for government expenditures on education, the composition of educational
spending, subsidies for research and development, trade policies and the promotion of inward
foreign investment.
6. Finding out what drives variety is also important. It is an interesting topic in its own
right, and all the more so if it is correlated with productivity. In addition, the development of a
good empirical framework for hypothesis testing requires instruments that can explain growth in
variety in case there is contemporaneous correlation between productivity growth and variety
growth.
7. One possibility found in the literature for industrial organization and for international
trade is that national product variety is driven by population size and income per-capita. For
example, if there is imperfect competition within a market due to fixed costs of production
(implying increasing returns to scale) with one product per firm, then cournot competition leads
to a situation where the number of firms is a function of market size. A simple example can be
found in Cabral (2000).' A similar conclusion emerges from the demand side. Variety will be
correlated with income per-capita when preferences are diverse and goods are not infinitely
divisible. For example, with a given budget constraint, a person might choose to own a car but
not a motorcycle because cars and motorcycles are not sold in fractions.2 With a larger income,
that same person might own both. Alternatively, both will be sold if the number of consumers
with diverse tastes is large enough.
8. There are several alternative hypotheses. The imitation hypothesis found in Barro and
Sala-i-Martin (1995) as well as many other texts suggests that growth in product variety should
be easiest and fastest for countries furthest from the frontier of observable variety. The
underlying assumption is that products would be imitated in order of increasing technological
sophistication as a nation's labor force accumulates knowledge and skill. Aghion and Howitt
(1998) assume that variety increases through deliberate investments in R&D without regard to
the initial level of variety. Nelson and Phelps (1966) and Griffith et al. (2000), support the
presence of interactive effects where the rate of imnitation is modified by human capital.
9. Previous empiricaR work. Several models of endogenous growth and horizontal
innovation have been put forth in the last 15 years but, to date, there have been few empirical
The trade literature includes several parallel theories. Linder (1961) argued that countries with similar
consumer preferences ought to trade more manufactured goods with one another than with dissimilar
countries, with the variety of goods rising in income per-capita. More recent models based on increasing
returns and product differentiation such as Krugman (1980) also predict a correlation between market size
and trade in differentiated products.
2 They can, however, be rented over time.
5
tests -- and almost none are pertinent to the issue of product variety. The empirical literature that
tests the link between product variety and productivity is very recent and very thin. To date, the
works by Feenstra, Madani, Yang and Liang (1999) and by Funke and Ruhwedel (2001) are the
most pertinent. Feenstra et al. describe a new measure of growth in product variety and establish
that product variety growth contributed to the lead in productivity growth held by South Korea
over Taiwan. Funke and Ruhwedel expand the analysis to cover 18 OECD countries and also
find support for the variety hypothesis.
10. Very little empirical work has been done on the determinants of product variety. The
work by Hummels and Klenow (2002) is one of only a handful of quantitative papers on product
variety. They find the level of national export variety is robustly correlated with the number of
workers and GDP per worker - both measures of market size - in a sample of 1 10 exporters with
5,000 product categories. They do not, however, investigate changes in variety over time.3
C. New Empirical Work and Key Findings
11. This paper provides a summary of new empirical research conducted by the author
(2002). The critical question is whether the correlation between variety growth and productivity
growth can be found in a larger sample that includes many developing countries (DCs). It would
be disappointing for those interested in economic development if the gains from product variety
were confined only to OECD countries. To explore this question, a measure of product variety
was constructed for a sample of 29 countries covering the period 1979 to 1986. The robustness
of the regressors, including variety growth, was tested by using the new Bayesian robustness test
developed by Sala-i-Martin (1997) and Dopplehofer et al. (2000). Robustness across country
observations was also tested by looking for influential points.
12. The main conclusion is that the variety hypothesis finds support, though this result is not
robust when an influential data point is removed. One also finds an interesting contrast between
sources of growth for the developing and developed nations. While the DCs gain substantially
through imitation of variety, R&D employment in the developed nations is important for
productivity gains beyond those generated by observable product variety. The results also
highlight the importance of macro stability and domestic security. Better performance in these
two areas provide the developed nations with a substantial lead over the DCs. Capital imports
may also contribute to productivity.
13. Another key finding is that the developed nations show a tendency towards convergence
in product variety while the DCs do not. This appears to be explained by the weak educational
performance in some countries. In particular, nations furthest behind the frontier of product
variety find it easy to imitate, and their ability to imitate is increased by educational attainment.
Sadly, not all countries are making effective choices. Colombia, Kenya, Pakistan and Zimbabwe
are examples of countries with growth rates in educational attainment well below what cross-
country convergence in educational attainment would require.
They also find import variety is correlated with country size. Klenow and Rodriguez-Clare (1997) found
import variety increased in Costa Rica in response to tariff liberalization.
6
14. There is a surprise too: growth in product variety does not appear to be correlated with
population growth, despite the finding by Hummels and Klenow of a significant correlation
when the relationship is expressed in levels. This presents a challenge for future researchers to
confirm or refute. One might speculate that there is a long-run correlation and that the sample
period used here was simply too short to detect it.
D. Firamnework for IHypotihesis TIesfing
15. The starting point for the development of the empirical framework is a system of well
identified equations that encompass the key hypotheses found in the literature. A multiple
equations approach is required because of the potential endogeneity between productivity
growth, growth in product variety and R&D employment. Equations for each of these variables
are laid out below.
16. TFP and Variety in Growth Rates. The main goal is to test whether period average
TFP growth, gTFP, in each country i over the period starting in time s and ending in time t, is a
function of period average growth in product variety, gV, over the same period. These gains are
assumed to accrue to all manufacturing products within a country. It is quite likely, however,
that gains from product variety will be conditional on one or more additional variables. In
particular, the imitation and diffusion literature give prominence to the role of R&D in adapting
technology to local conditions. Including R&D will also make allowance for within-product
gains such as improvements in production processes and output quality. This can be proxied by
using R&D employment R in period s. Note that the use of R in levels is conscious even though
gTFP and gV are defined in terms of growth rates. This was done with the recognition that each
scientist and engineer will produce a stream of innovations that will progressively increase
productivity. One can therefore write:
1) gTFP,st = ro + )rglV, + X2 In R,, +
The null hypothesis is that or, is zero, with the alternative hypothesis being that 2r, is larger than
zero. The disturbance term is represented by fli,,,.
17. One way to calculate TFP is to assume that a common (Cobb-Douglas) production
function is applicable for all countries in the sample.4 This allows one to write:
2) gYi st = gTFPj st + [a * gL,,st + / - gKi,st ]
Thus, by substitution, the test equation 1 is rewritten as:
3) gyi,st = [Pf0 + ITI * 9Vi,,s +r2 -*h lR.,, ] + [a * gLi,st + *gK~,st ] + Hi st
4 Another way to generate TFP estimates is to subtract labor and capital growth rates, weighted by observed
factor expenditure shares, from the growth of value-added. This can be done for individual countries and
does not require the assumption of the same production function across countries.
7
18. Three additional regressors may help explain TFP growth. The first is population
density. Marshall (1959) suggested that knowledge spillovers would be more likely as the
density of people increased, especially among agglomerations of people in similar trades.
Ciccone and Hall (1996) rejected the null hypothesis that density has no effect on per-capita
output in levels. On the contrary, they found a positive correlation. The period average growth
rate for population density, gDENS is used because equation 3 is based on growth rates rather
than levels. The second additional regressor is period average growth in telephone lines per
1,000 people or gTELE. Easterly and Rebelo (1993) find that public investment in
telecommunications increases real growth. I assume this effect acts through TFP rather than
factor accumulation: increasing communications capacity should increase TFP by allowing non-
rival knowledge to flow more easily and by facilitating the coordination of productive processes.
The third regressor is the real growth rate of imports of capital goods, gM. Keller (1999) finds
that imported capital goods can help explain productivity growth. New knowledge may be
bundled with imports or imports may induce productivity gains through the competition they
create.
19. Two explanations are tried for negative TFP growth.5 The first is substantial armed
conflict (CONFL) within a country between times s and t. It is reasonable to expect that
productivity would decline due to chaos and knowledge lost through deaths and emigration.6
These effects are proxied by the number of war-related deaths per year, normalized by the
population at time s. The second is the standard deviation of inflation, SDINFL, between times s
and t as a proxy for macroeconomic instability. The presumption is that macroeconomic
instability makes it difficult to allocate resources to maximize productivity.
'20. Following Feenstra et al. (1999), export data are used in the construction of a measure of
product variety. Feenstra et al. included a dummy in their regression equation for the
depreciation of the Taiwanese dollar in 1981. This is important because a real depreciation
would tend to make it easier to export existing product varieties that were previously too costly
for foreign markets. The gains in TFP from depreciation induced variety should be discounted
since they do not represent actual gains in varieties produced. Period average growth in the real
exchange rate, gRER, is therefore included as a control variable. If depreciation induced variety
is to be discounted, then the coefficient on this variable should be positive (depreciation is
signified by negative growth in the real exchange rate). One could argue for the opposite sign as
well. It is possible that depreciation, by making imported inputs more expensive, would induce
greater productivity as producers shed unprofitable products and/or used more cost-effective
inputs.
21. Equation 3 can be further modified to allow for the possibility that each new product
contributes more to productivity than the product that came before. This assumption is a
common ingredient in R&D based models. See, for example, chapter 6 of Aghion and Howitt
(1998). In other words, the marginal contribution to TFP from variety may increase as nations
5 Negative TFP growth is both puzzling and common.
6 It is also possible that conflict can increase productivity as shown by the case of Liberty Ship production.
8
move closer to the observable frontier of product variety.7 This can be captured by multiplying
growth in product variety by the logged ratio of initial variety to total observable variety, VF,
where Tz2>0.8 The final result is:
4) gy,,, = ro +r V,r , +f2gVist .ln(V,M/VF)+,r3 InR,.s + r4gDens+...
..+ r5gTele + ir6gM + r,CONFL + ;8SDINFL +
...+ r9gRER + agLj st + 8gK,5, + i5,u
22. Growth in Product Variety. A review of the literature leaves one without a clear
picture of what drives growth in product variety. An encompassing equation is therefore used to
capture four competing hypotheses. First, the imitation hypothesis found in Barro and Sala-i-
Martin (1995) and in many other texts suggests that growth in product variety should be greatest
for countries furthest from the frontier of observable variety. This is measured using the logged
ratio of initial variety in each country, V5, to the maximum at the frontier, VF. Second, Aghion
and Howitt (1998) assume that variety increases through deliberate investments in R&D where
AB=KR. Third, the work of Nelson and Phelps (1966) and Griffith et al. (2000), supports the
presence of interactive effects where the rate of imitation is modified by some measure of human
capital such as educational attainment, H, or R&D employment, R. The fourth possibility, found
in Jones (1999) and in Chapter 12 of Aghion and Howitt (1998), is that countries with growing
populations, gN, should display an expansion in product variety.9
23. The use of export variety as a proxy for output variety necessitates the introduction of
two more variables. First, when using export-based measures of variety, it is reasonable to
expect that real exchange rate depreciation, gRER, should increase gV as more products become
more affordable and marketable overseas. 10 Second, one needs to account for the impact of
rising productivity. Increased productivity could allow price reductions that would make it
easier to market goods overseas. Productivity growth, gTFP, is therefore used as a control with
respect to export variety. Putting everything together, the final encompassing equation becomes:
5) gVi.,t = go + 61 ln("',, /VF )+62 lnRi,, + 63 lnRi,, .ln(V,,/VF )+ 4gN1,,, --
...+ 5gRER + ±6gTFP + 6es,
where £5 and 63 should be negative because ln(Vj,,,VF) is always negative. The coefficient 65
should also be negative when exchange rate depreciation is indicated by a negative growth rate.
The three remaining coefficients should all be positive. A third equation is added below to
explain R&D employment.
7 Alternatively, this variable could also be used as a correction for the fixity of observable variety where
small observed changes in variety for nations close to the frontier represent larger actual changes.
9 The time subscript is omitted because observable frontier is fixed for any system of product classification.
The author also experimented with using real growth in GDP per capita as the proxy for growth in market
size in addition to, and instead, of population growth.
'° Import variety could contract as the exchange rate depreciates. On a related note, one niight be concerned
about endogeneity between gM(in equation 4) and gRER (which appears in equations 4 and 5). Addison
(2002) finds that the correlation between these period average variables is not statistically significant even
at the 15 percent level of confidence in his sample.
9
24. Research and Development. According to Aghion and Howitt (1998), R&D
employment is a function of population, N, and expected productivity growth which is proxied
by observed TFP growth. In addition, it is reasonable to expect that R&D employment will also
be corTelated with average educational attainment, H. One can therefore write:
6) InRR,, = po + p, InN,1s + p2gTFP+ p3 InH,, + 1,
All coefficients in equation 6 are expected to be positive. This completes the system of
equations.
25. Identification. There is now a system of three equations. Each equation in the system is
identified in the sense that none can be confused with a linear combination of the others. This
satisfies the order condition necessary for identification.11 A summary of the system of
equations is shown in Table 1 below along with signs indicating prior expectations for each
coefficient. The variables have been sorted into three groups: those that are potentially
endogenous according to the theory, those that are pre-determined and those that are maintained
as exogenous.
Table 1: Equations for Hy pothesis Testin
- Description . Equation 4 Equaton 5 - Equation 6
Dependent gY V InR
Endogenous
gTFP TFP growth (from TFP index) +P2
gY Manufacturing value-added growth 1
gV Variety growth 1
InR R&D employment, 1979 +5, 1
ln(V/VF) gV Interactive term
ln(V/VF) lnR Interactive term 4
gL Labor force growth +a
gK Capital stock growth +B_
Pre-determined
In( VA/F) Distance from variety frontier, 1979 -Si
InN Total population, 1979 +PI
InH Avg. educational attainment, 1979 +P3
Exogenous
gDens Growth in population density + 4
gTelo Growth in telephone lines per 1,000 people +M
gM Growth in capital equipment imports
Confl War related deaths/capita, 1979-86
SDINFL Standard deviation, CPI inflation -U
gRER Growth of real exchange rate ±JZb 4
gN Urban population growth rate _ _ _
a ll growth rates are 197986 period averages
26. Endogeneity. Testing the simultaneous system developed above will be made difficult
by the endogeneity between gY, gV and R. Endogeneity creates contemporaneous correlation
11 Necessary but not sufficient. The order condition requires the total number of restrictions, such as setting a
parameter to zero, should be at least as great as the number of equations in the system, less one.
12 It is very likely that real exchange rate depreciation and capital imports would be correlated within a single
country and across time. The same would be true of the growth capital imports and the growth total
imports. This need not be the case for cross-country period averages and, in fact, no significant
correlations between these variables were found.
10
between the regressors and the residual, thus creating biased and inconsistent coefficient
estimates. The usual tactic in such cases is to rely on instrumental variables. For this, one needs
a good set of instruments that are strongly correlated with the independent variables and not
correlated with the disturbance term. Fortunately, the pre-determined and exogenous variables in
a simultaneous system of equations are highly suitable for use as instruments because they drive
the endogenous variables and because they are uncorrelated with the disturbance term. This
suggests the 3 pre-determined and 7 exogenous variables listed in Table I above may be used as
instruments in the formation of instrumental variables. In the case of equation 4, there may also
be endogeneity between gY, gL and gK. Thus, for this equation, the initial (pre-determined)
levels of labor and capital lnL, and lnK, are added to serve as instruments for gLEt and gKs.
27. The efficacy of the instrumental variable approach also depends upon how well the
instrumental variable estimates substitute for the original data in addition to the avoidance of
contemporaneous correlation with the disturbance term. Higher correlations between the
instruments and the explanatory variables leads to greater asymptotic efficiency in the
estimators. Thus, in addition to the variables listed in Table 7.1 above, several other pre-
determined variables are used as instruments in the construction of instrumental variables for the
potentially endogenous regressors. These include LIFE for average life expectancy, URB for the
per cent of population living in urban areas, DENS for initial population density and TELE for
telephone lines per 1,000 people. These are all measured in 1979. Also included are a measure
of openness to trade, OPEN, and a measure of the quality of governance, GADP, which both
overlap the period 1979-86.
E. Tlihe Datsa
28. The data are a cross-section of 29 countries for the period 1979-86. The choice of time
period was dictated by the data for product variety, an issue taken up below. A summary of the
main variables appears in Table 2 below and a list of the countries appears in Table 3. A brief
description of the data can be found in the Data Appendix. An in-depth description and analysis
of the data for variety can be found in Addison (2002).
29. The product variety data are based on a simple count-based measure of variety V and
growth in variety gV. One starts with a data source that divides output into a fixed number of
product categories. Many countries, for example, record output using the International Standard
Industrial Classification (ISIC). The problem with output data, however, is that ISIC includes
only 81 categories and many countries merge those categories over time.
30. Trade data are far more disaggregated and can serve as a proxy for output variety: a well
diversified exporter must be a well diversified producer, although the converse is not necessarily
true. The data used here are organized according to the Standard International Trade
Classification (SITC Revision 2) which divides manufacturing exports into 1,600 categories.
Variety V is defined as the number of non-zero export categories in any given year. The variable
gVis the period average growth rate of V. Highly disaggregated trade data are rather noisy, so
non-zero categories with values below US$1,000 and exports that were not part of contiguous
blocks were both filtered out. The end result was that each product category could be empty,
11
continuously full, represent a new category of trade or represent a category that becomes
obsolete.
Table 2: Data Sources and Descriptive Statistics
-. . . . . j > ~~~~~~-. ~~ -. -~~~~ 'Standard,
-.' , - -'.---'~ -;'-'w1 3- Perioid Mean r bviion maximur' MnlImum
Output
Growth in manufacturing value-added (gY) 1979-86 0 033 0.032 0.108 -0.021
Index of total factor productivity (TFP) 1979 1 094 0.402 1 830 0 279
GroDwh zn total factor productivity (gTFP) 1979-86 0 014 0 023 0.043 -0 040
Variety
Product vanety (V) 1979 902.861 581.824 1,594.000 15.000
Distancefrom frontier of observable variety (In(V/VF)) 1979 -0.984 1.168 -0.004 4.670
Growth in product vartety (gV) 1979-86 0.038 0 041 0 134 0.000
Factor Inputs
Value of capital stock (InK) 1979 10.280 1.667 13.941 7.051
Real growth of capital stock (gK) 1979-86 0.040 0.023 0 096 0 006
Manufacturing tabor force (lnL) 1979 6977 1.567 9.950 3.738
Labor force growth (gL) 1979-86 0.007 0.032 0 114 -0 044
Total populaton (InM 1979 16.722 1.445 20 326 13.766
Urban population growth rate (gN) 1979-86 0.023 0.021 0.081 0 001
Educational attainment (H) 1979 6015 2867 11.503 1.715
Educational attainment growth rate (gH) a 1979-86 0.012 0.012 0 054 -0 003
Research and development employment (InR) 1979-86 8 962 2 299 13.375 3 135
Life expectancy (LIFE) a 1979 68.200 7.645 75.800 53.900
Life expectancy growth rate (gLIFE) b 1979-86 0 005 0 002 0 009 0 001
Trade and Knowledge
Openness to intemational trade (OPEN) a 1950-94 0.530 0.346 1 000 0 000
Real growth rate in capital equipment imports (gM) 1979-86 0.045 0 118 0.567 -0.081
Real growth in manufacturing exports (gX) b 1979-86 0.058 0.094 0 461 -0.027
Telephone lines per 1,000 people (TELE) 1979 176 161 165.461 564.020 2 990
Growth in telephone lines per 1,000 people (gTELE) a 1979-86 0.055 0.036 0.172 -0.012
Governance (GADP) b 1986-95 0 764 0.190 0.988 0 407
Security (CONFL) 1979-86 0 008 0.030 0 158 0 000
Macroeconomic Management
Consumer price inflation (INFL) b 1979-86 0.122 0 085 0.426 0.029
Price volatility (SDINFL) 1979-86 0.063 0.064 0.270 0.022
Growth of real exchange rate (gRER) 1979-86 -0 032 0.077 0.041 -0 407
Other Country Characteristics
Urban share of population (URB) a 1979-86 59 950 22.153 95.300 15.460
Population density (DENS) a 1979 142.588 155.133 536.480 1 890
Growth in population density (gDENS) a 1979-86 0.015 0.011 0 037 0.000
Mmning share of GDP (MINING) b 1988 0 033 0 034 0 111 0.000
a. Used as additional instruments in 2SLS.
b. Used as additional regressors in BACE robustness test.
Source Addison, 2002.
31. Cross-country trade data arranged by product categories must be used with caution.- Not
all countries use the same trade category classification system. Concordances must therefore be
established to translate the data from an originating country's system into the system of choice,
SITC in this case. This would not be a problem except for the fact that some countries switch
from one system to another drop category codes, add new codes and even redefine codes. These
changes can induce artifacts in the data that look like variety changes even though actual product
variety may not have changed at all. Thus, for example, one finds artificial bursts of product
12
variety in 1977 due to the fact that a number of countries switched from to SITC version 2 from
version 1 between 1976 and 1977. Likewise, countries in the European Union switched to the
Common Nomenclature system from NIMEXE in 1988 which affected the concordance to SITC
version 2. For these reasons, the sample period had to be limited to 1978-87. The filtering
process required a further contraction of the sample period to 1979-86.
32. The measure devised by Feenstra et al (1999) is also based on disaggreFated trade data.
Their methodology, however, requires far more complex computational effort.3 There is an
important trade-off: their measure allows for the possibility that the productivity spillovers may
be greater for some new goods than for others while the count based measures require the
contributions from each new product to be the same.
33. The measures used by Feenstra et al and the simpler measure used here are both subject
to the limitations of the data. In particular, the use of pre-established product categories makes
it impossible to measure gains in product variety within any specific category and beyond the
number of pre-established product categories. This puts a premium on the level of
disaggregation in the data. One should expect a greater ability to differentiate between nations as
the data become more detailed. In fact, by experimenting with SITC trade data at three levels of
aggregation (3-digit, 4-digit and 5-digit category codes) one can see that this is indeed the case.
As the level of disaggregation increased, the distinction between low variety and high variety
nations became much sharper. The ability to see changes in product variety also improve as the
data become more detailed. (See Figures 1 and 2 below.)
34. The new data are congruent with the previous empirical work done by Funke and
Ruhwedel and by Hummels and Klenow. They tell the same stories. Funke and Ruhwedel
found a significant and positive relationship between product variety, proxied by export variety,
and GDP per capita for 18 OECD countries. 14 Hummels and Klenow found the same for a much
larger sample of 110 exporters. The simple correlation between GDP per capita and variety in
the sample of 29 countries used here is 0.82 and the slope coefficient is highly significant at the 1
percent confidence level. A similar relationship can be seen in Figure 3 and Table 3 below
which use the new data for TFP and export variety, V. The correlation between TFP in levels
and Vfor 1979 data is 0.65 and the slope coefficient is highly significant at the 1 percent
confidence level.
35. The new data are also supportive of the comparison of Korea and Taiwan in Feenstra et
al. even though they are for the manufacturing sector as a whole rather than for manufacturing
sub-sectors. In particular, one can see that growth in manufacturing TFP is faster in South Korea
than in Taiwan while the same is true for the measure of product variety. (See Table 4 below.)
Funke and Ruhwedel find evidence that this same pattern in growth rates holds true, when
conditioned on country fixed effects, for a wider sample of 14 OECD countries.
13 The author did experiment with the measure developed by Feenstra et al. in Addison (2002). This method
did not generate a strong signal. This could be due to the shorter time period (1979-86 versus 1972-91) or
the more aggregated nature of the data (1,600 versus more than 10,000 product categories) used by
Feenstra et al.
14 Their data are expressed relative to the US which the authors consider to be the technological leader.
13
Figure 1: Data Aggregation and Observable Variety
1,800
1,600 - El SaIvador
2 1,400 - {- India
* -h--~~~France
* 1,200 -
> 1,000 -
800-
600-
O 400
200 -
- 500 1,000 1,500 2,000
Maximum Observable Variety
(199,650 and 1,600)
Figure 2: Data Aggregation and Observable Variety Growth
p V A R
0 1 0
0 0 9
0.08
0 .07
0 086
0 0 5 - D se v e lo p e d
0 .04
O 0- D e v e lo p in g
0 0 3_ _ _ _ _ _ _ _ _ _ _ _ _ _ _
0 0 2
0 .01
0 0 0 -_
-O 0 1-_
50 0 1 ',0 0 0 1 s 5 0
Data are averages for 13 developed and 15 developing countries. Zimbabwe is excluded because it is a
large outlier.
14
36. Before discussing the results reported in Table 5, it would be useful to review the
econometric methodology used here for hypothesis testing. Several competing hypotheses were
advanced for both TFP growth and variety growth. For example, is TFP growth a function of
variety growth, R&D employment, or both? To find out, a simplification search process was
used. In the search process, one systematically imposes and tests zero-restrictions on the
parameters of an equation that encompasses several competing variables and models. One tests
for the redundancy of insignificant variables individually and in blocks. This is also known as
the "general-to-specific" approach.'5 The robustness ofthe regressors, including variety growth,
was tested further by using the new Bayesian robustness test developed by Sala-i-Martin (1997)
and Dopplehofer et al. (2000). Robustness across country observations was also tested by
looking for influential points.
37. The Hausman (1978) test was used to test for contemporaneous correlation generated by
measurement error as well as the potential endogeneity between gTFP, gV and lnR - and
between gY, gK and gL. Addison (2002) provides details on the choice of instruments. Despite
the theoretical reasons for expecting contemporaneous correlation, the data fail to reject the null
hypothesis of no contemporaneous correlation in each of the regression equations. This might be
explained either by the small sample size or by the choice and number of instruments. (The
Davidson and MacKinnon (1993) test of over-identifying restrictions was also used to make sure
none of the instruments actually belonged in the test equations.)
38. White's (1980) test did detect heteroskedasticity in the equations for gVand lnR. Thus,
White's heteroskedastic-consistent estimator was employed for those equations. The Jarque-
Bera (1980) test was used to test the null hypothesis that that the error term is normally
distributed. If not, the estimated standard errors would be biased and inconsistent. In all three
equations, the data failed to reject the null hypothesis. Finally, the Ramsey (1969) RESET test
was used to test the null hypothesis of no mis-specification due to functional form or omitted
variables. The data failed to reject the null in each case.
39. Productivity Growth. The null hypothesis of no correlation between gTFP and gV, that
iz,=0, is rejected by the data. The elasticity of growth in manufacturing value-added to variety
growth is 0.262. In addition to the evidence supporting the variety hypothesis, one can see that
R&D employment statistically significant, implying that there are important gains being made
within product categories as well as across product categories. (Table 6 shows that within
category gains occur mainly in the developed nations.) Real growth in capital imports also
contributes to productivity with an elasticity of 0. 132. It is somewhat puzzling to find that real
exchange rate depreciation, which ought to militate against capital imports, also boosts
15 D.F. Hendry is one of the leading proponents of this technique. Hendry (2001) claims his new software,
PcGets, can sort through an encompassing model with 40 irrelevant variables and locate the correct,,
smaller model 97 per cent of the time. The author has not tried this software yet.
15
Figure 3: Product Variety and Total Factor Productivity, 1979
18 ~
1 6 _ ~~~~~~~ -~~~~~-s~~~ .~ n~~f Frant
~ 2 J0TIC~ - {X * ilan - -fd~
1.6~~~~~~~ ~Epr Coloriety
10 .~~~.~~V~nezueDevlope
Norway 04,719 1,284 102 M r
Canad 1411A139s12rosaaic 382 18-9
0.2 - , __ar.a a--
Japan 9,756 1400158Vneul 80091,200 1600
Sweden 12,261U1,503K1.28 Dvloped
Autralya 12227 1,228 1.33 Zaimbabw 1,105 150 0288
Finland 10,2629 1,229 1.16 Jamaica 2,508 133 1063
No.way 11,719 1,284 1.025Mauritiuy 42 1 0.58
Cehranada 14,11428 1 ,539 1.34Cst Rhicapie 3,821 118 109
J .,5,6 1,412 1.58 Vreeze 8,092 16 1 0386
Denmack 11 ,484 1,544 1.22 Kenyia 931 1824 0.66
Sweden 12.26 1.503 1.28Solombina 2,8478 292 1.59
Belgium 10,67961,579 1.76 Tunisia 2,368 330 0.39
U.K. 10,483 1,581 1.15Turkey, 2h ,955 4748 0.596
N etheWrlads 11,286, 1,583 n itmain rie, 1.45 Phlppnse 1,854 547pl 1orlao 30hvrets082sml
Franel 1166 1,9 1s 7 n830
eqal .00
TableT3abProductlVarietyeandSoth Faca: 3,347 921ea 1 0
.- . P . C-p --_ _- 4 TFPe -- V --:--- -V GD- pIa- V --FPU .
Australia12227 122833 ZimbSouth Korea 3,321 98 1 0 75
Jap_an_9_756_1_412___ _ Ta,nwan C 4,02 1,08 0 0.9
a PenWrld Tablesh,oe i 1985 base yea S c l w a i82 ,si
bItFa idel etoolg from1 Caves, Ch1tnsn 9n D Pekertanb Unegh 1,0gemeri sampl averag
Beqgual 1.00.,7 17 unsa2,63003
Table 4: ~~~~~South Kora vrsu Taiwan China0
Soulh Korea 42 31~~~~~~~out 98e 331 08175 7
Taiwan. Chna 32 28~~~~~ahan,Ch 4281,080 096 6
a. TFP index methodology from Caves, Christensen and Diewert (11 982ab).Uwihe,goercsml vrg
16
productivity. One might speculate that depreciation induces more efficient use of imported
intermediates in production. Finally, productivity is reduced by conflict and by macroeconomic
instability as proxied by volatility in the inflation rate.
40. The robustness of the results in column 1 were tested using Bayesian Averaging of
Classical Estimates (BACE) as documented in Dopplehofer, Sala-i-Martin and Miller (2000).
All of the regressors that survived the simplification search, as shown in column 1, were found to
be robust except for real growth of imports of capital goods. This remained true when extra
variables were added to the equation as another robustness test.16
41. It was also discovered that the observation for Venezuela is an influential point with
regard to the partial correlation between gVand gTFP. Excluding it from the dataset renders gV
insignificant. As Thomas (1997) notes, most economists prefer to retain such influential
observations in their samples unless it can be proved that the observation is wrong or
inappropriate.
42. A case can be made for retaining this observation. Until the late 1980s, Venezuela was
known as a fast growing and stable nation. The period 1979-86 includes the tail end of the
second OPEC oil boom. This boom made possible a rapid expansion of education in many oil
producing nations, including Venezuela. Average educational attainment between 1975 and
1980 grew by 8.4 percent per annum compared to only 2.4 percent per annum between 1970 and
1975. As can be seen from equations 5 and 6, this would have led to increased R&D
employment and increased product variety, both of which would have contributed to productivity
on an ongoing basis. Whether or not one agrees with this argunent favoring the retention
Venezuela, the fragility of the results demands that future research should be aimed at finding
out if the partial correlation between gV and gTFP is increased or decreased by an expanded list
of country observations and/or the use of panel data.
43. Variety Growth. The simplification search was conducted twice in order to determine
whether InR or lnH would be the better measure of human capital - with better implying more
explanatory power. Both worked well but lnH is preferred on the basis of a J-test between the
two model specifications. In addition, it was found that Zimbabwe was a strong outlier
observation that needed a dummy because of it's strong rebound in manufacturing output and
variety at the conclusion of its civil war in 1980.
44. The final results confirm that imitation is easier for nations furthest from the frontier of
product variety, with imitative ability increased by average educational attainment. (Note: the
signs on 81 and 53 are negative because the log of a fraction, such as V/IVF, is always negative.)
Productivity growth contributes to growth in variety with an elasticity of 0.208. Exchange rate
depreciation (denoted by a negative change in gRER) is also significant because output variety is
proxied by export variety: depreciation makes products less expensive overseas and thus easier
to market.
16 As a result of this process, it was found that the average rate of inflation performed better than the standard
deviation of inflation. This result is not shown in Table 5.
17 It is unfortunate that these gains were not maintained in later years.
17
Table 5: Final Results
, . - , - - - gYt' --. . - , . .-- - pPAR MR -
* .--. - -- - - ~ - - EguriEon4u - - .- EOu an5S Equadon 6-
. 0 - - . . . OLS - BACE-- OLS OLS
gTFP 56 P2
Coefficient 0.208 18.058
i-Statistic 2.081 2.155
gy
Coefficient I 1
t-Statistic
gV 'B' Ill
Coefficient 0.262 0.410 1
t-Statistic . 2.137 2.445
InR/1000 7 7T3
Coefficient 5.404 6.519 1
t-Statistic 2.340 2.822
gN Pi
Coefficient 1 078
i-Statistic 9.111
InH/looo PJ
Coefficient 2.491
t-Statistic 8.119
In(V,IVF) 6,
Coefficient -0.020
t-Statistic -4.453
In(VWVF).lnH/1OOO !63
Coefficient -17.062
t-Statistic -3 571
gM i7
Coefficient 0.132 0.083
t-Statistic 2.364 1.227
CONFL , ,
Coefficient -1.357 -1.427
t-Statistic -3.117 -2.913
SDINFL ,
Coefficient -0.172 -0.173
t-Statistic -2.574 -2.678
gRER ff9 n 6S
Coefficient -0.237 -0.254 -0.089
t-Statistic -3.829 -3 117 -7.637
L a a
Coefficient 0.383 0.284
t-Statistic 2.332 1.552
gK 6 6
Coefficient 0.683 0 869
i-Statistic 3.846 2 697
Zimbabwe 67
Coefficient 0.405
t-Statistic 39.066
Constant nb 6o Po
Coefficient -0.052 -0 004 -13 527
t-Statistic -2.135 -1 380 -5.428
Adiusted Rz 0.761 0 994 0.866
Joint Significance (F-statistic) 12.167 987 468 52.704
Heteroskedasficity (White) none I yesv
Normal Disturbance (Jarque-Bera) y yes yes
Mis-specification (RESET) none none none
Contemporaneous Correlation none none none
a. BACE prior assumption is that 8 out of 10 variables regressors should be significanL BACE mixed
t-statistics > 2 can be interpreted in fashion similar to OLS t-statistics: 95 percent of the probability
density associated with the BACE estimates are non-zero.
18
45. The developed nations within the sample showed a strong tendency towards convergence
to the frontier of observable variety with the developing nations do not. This appears to be
explained by the interaction between imitative ability and educational attainment. Most
countries in the data sample show good progress in improving educational performance over
time. This progress is not uniform. Colombia, Kenya, Pakistan and Zimbabwe as examples of
countries with growth rates in educational attainment well below average. By contrast, Tunisia
and South Korea are examples of nations where educational growth rates were exemplary.
46. The simplification search also produced a surprise. The data fail to reject the null
hypothesis that population growth does not contribute to growth in product variety. The same is
true when growth in real GDP per-capita is substituted for population growth - and when both
variables are tried concurrently. One might speculate that these variables would indeed be
correlated with growth in variety if the time period were longer than the 8 years between 1979
and 1986 used here.
47. Research and Development. The results for equation 6 are straightforward. The data
fail to reject the hypothesis that R&D employment is a function of population, average
educational attainment and expected productivity gains - the latter being proxied by actual
productivity gains.
G. Patterns of Growth amid Poicy llmpllicatons
48. Several interesting policy related conclusions concerning the determinants of cross-
country patterns of productivity growth can be deduced from the work that appears in Section F
above. A decomposition of estimated productivity growth, gTFP*, for the developed and
developing countries is provided in Table 6. The average contribution to gTFP* from each
variable is shown for the developed nations and for the DCs - each relative to the full sample
mean for each variable. Thus, the fact that most of the developed nations experienced lower
growth in output than the DCs implies that average gY* will be shown as negative for the
developed nations and positive for the DCs.
49. The top half of the table shows how gTFP* is derived from predicted real growth in
output, gY*, less real growth in factor inputs. There is an interesting contrast here. On average,
and relative to the sample means, the DCs are accumulating factors while their productivity falls.
The opposite situation applies for the developed countries.
50. The bottom half of the table shows the contributions to gTFP* from the key variables in
the model. The developed nations growth (relative to the sample mean) is generated by R&D
employment, the absence of conflict, macro stability and capital imports. Of these, the largest
contribution comes from R&D employment. This suggests there are important productivity
gains coming from sources other than gains in variety such as process innovations and qualitative
improvements. In absolute terms, the contributions to gTFP* in the developed nations from
product variety growth and exchange rate depreciation are approximately zero, hence their
deviations from the full sample means are negative. The contrast with the DCs is quite striking.
Positive contributions to predicted productivity growth in the DCs (relative to the sample mean)
19
come from gains in product variety (0.74 per cent) and from real exchange rate depreciation
(0.64 per cent). Unfortunately, these gains are reversed by low R&D employment, low real
growth in capital imports, macro instability and conflict.
Table 6: Decomposition of Contributions to gTFP
: - : -- - - Growth Rates (%) a
_ . - - - - , - Developed - DCs Difference
Predicted gTFP* e/ 0.46 -0.40 0.86
Predicted output growth ( gY) -1.34 1.16 -2.50
less labor growth ( a. gL ) -0.83 0 72 -1.55
less capital growth (/6. gK) -0.98 0.85 -1.83
Predicted gTFP* f' 0.46 -0.40 0.86
Variety growth ( r,- pVAR) -0.86 0.74 -1.60
Real exchange rate (g - gRER) -0.74 0.64 -1.38
Conflict ( ;r7 CONFL) 0.38 -0.33 0.71
R&D employment ( r3 - InR) 0.79 -0 69 1.48
Macro stability ( r8. SDINFL) 0.52 -0.45 0.97
Capital imports growth ( r.gM) 0.36 -0.31 0.67
a. Growth rates are relative to sample means for each variable. The sample excludes Zimbabwe. Totals are subject
to rounding error.
b. Average for developed nations less average for full sample.
c. Average for DCs less average for full sample.
d. Developed average less DC average.
e. gTFPe = gY* - av gL - D a gK g
f From equation 4.
51. Policy Implications. The policy ramifications of the results for equation 4 can be seen
more clearly if equations 5 and 6 are used to generate the reduced form equation, shown below. 18
Statistically insignificant variables have been omitted.
gTFP* = ro0 + v50± + )r3PO 7± E, lIn(V/VF ) + rt3 p3ln H + ;1, 3 ln(V /VF )ln H+
1-5r166 - f3P2 1-r1S6 - n3P2
7) (,r9 +,ri65 )gRER + 7r6gM
I - gr,56 - ;r3P2 1 - X,6 16 - ;r3 A
'T7CONFL 7T8SDINFL + r3 p, In N
1 - 'r156 - r3P2 1 - X,156 -r3P2 1 - ,6 -r3P2
52. This reduced-from equation is used to create another growth decomposition in Table 7.
The reader is cautioned that the numbers in Table 6 and 7 are similar but not directly
18 Jones (1999) critiques such formulations for individual economies, noting that growth in population or
educational attainment should cause accelerating growth in output. Models that explain why this does not
actually occur are still being proposed and debated.
20
comparable. This is because the reduced form version implies slightly different implicit values
for lnR and pVAR and because all coefficients are multiplied by I/(1-)s- pr32).
53. Equation 7 shows that average educational attainment contributes to productivity directly
and via its interaction with the variety gap, ln(VIVF). The growth decomposition in Table 7
illustrates the magnitude of the contribution. The average developed nation is able to generate
productivity gains 1.3 per cent per annum higher than the DCs because of higher educational
attainment. (See the "difference" column in Table 7.) Yet the DCs gain 1.8 per cent per annum
more than the developed nations through their greater imitative potential and the boost that
education gives to imitation through the interactive term. From this, it is easy to deduce that
policies that increase educational attainment will have the greatest effect on the DCs, though
higher attainment will remain important even for nations close to the variety frontier.
Table 7: Reduced Form Decomposition of Contributions to gTFP
Growth Rates )v
____________________________________________ Developed w Developing Difference
Predicted gTFP e/ 0.35 -0.31 0.66
Predicted Output Growth (gr) -1.46 1.26 -2.72
less labor growth ( a.gL ) -0.83 0.72 -1.55
less capital growth ( b-gK) -0.98 0.85 -1.83
Predicted gTFP* f 0.35 -0.31 0.66
Average Educational Attainment (InH) 0.68 -0.59 1.27
Education and Imitation -0.94 0.82 -1.76
Distance from variety fronfier (ln(VIVF)) -0.47 0.41 -0.88
Interactive term (In(VIVF)IlnH) -0.47 0.41 -0.88
Trade and Exchange Rate Effects -0.53 0.46 -0.99
Capital imports growth (gM) 0.43 -0.37 0.80
Real exchange rate (gRER) -0.96 0.83 -1.79
Instability and Insecurity 1.06 -0.92 1.98
Conflict (CONFL) 0.45 -0.39 0.84
Macro instability (SDINFL) 0.61 -0.53 1.14
Total Population (InN) 0.08 -0.07 0.15
a. Growth rates are relative to sample means for each variable. The sample excdudes Zimbabwe. Totals are subject
to rounding error.
b. Average for developed nations less average for full sample.
c. Average for DCs less average for full sample.
d. Developed average less DC average.
e. gTFP = gY - a. gL -,6. gK.
f. From equation 7.
54. Greater macroeconomic and domestic stability provide the developed countries with
another 2 per cent per annum advantage over the DCs. The developed nations also gain 0.8 per
cent per annum more from real growth in capital imports than do the DCs (though the BACE
results in Table 7.33 suggest this may not be robust). This could be attributed to any number of
21
policies that lead to increasing export output and improving terms of trade. Differences in
population size matter as well, conferring a 0.15 per cent per annum advantage to the developed
nations.
55. The empirical results also suggest that real exchange rate depreciation contributes to
productivity growth. In this regard, it is easy to believe that real depreciation could induce
manufacturers to economize on imported inputs, thus increasing their productivity. In particular,
the average DC generates 1.8 per cent per annum more from this source than do the developing
nations. (The average developed nation did not experience depreciation in my sample period.)
Even so, it is hard to argue for exchange rate depreciation as a development policy for several
reasons. One, it is in direct conflict with the need to increase real capital goods imports. Two,
under most circumstances, depreciation is not correlated with low inflation. Three, greater
productivity ought to lead to real exchange rate appreciation. Four, it is also bad politics:
depreciation by one trading partner could induce others to depreciate in order to remain
competitive.
H. Summary and Outstanding Issues
56. Is gVpositively correlated with gTFP when conditioned by several other variables that
can be controlled through policy choices? The null hypothesis that ;r,=0, meaning that gV does
not affect gTFP, is rejected by the data. This result is robust when tested using BACE.
Unfortunately, the presence of an influential point for Venezuela makes the results fragile. It is
argued that the Venezuelan observation belongs in the data set. At the same time, future research
should seek to test if the partial correlation between gVand gTFP is strengthened when
additional observations are added.
57. The analysis was supported by an investigation into the cross-country determinants of
product variety, one of just a very few found in the literature. I found strong support for
convergence in product variety where the rate of convergence is modified by average educational
attainment. This provides support for the imitation hypothesis. In making this test, I also failed
to find support for the alternative hypothesis that variety growth is a function of national
population growth.
58. Even if population growth is found important, the results published here suggest that
small countries are not doomed to slow growth and low product variety if they can improve
average educational attainment. In particular, an exploration of equation 6 confirmed that R&D
employment is a function of population, average educational attainment and expected
productivity growth. As made clear by the reduced form equation 7, this implies that educational
policy choices can drive productivity directly through R&D employment and indirectly through
the imitation of product variety. Sadly, not all countries are making effective educational policy
choices. Colombia, Kenya, Pakistan and Zimbabwe are examples of countries with growth rates
in educational attainment well below average.
59. The results also highlight the importance of macro stability and domestic security. Better
performance in these areas provide the developed nations with a substantial lead over the DCs.
22
Capital imports may also contribute to productivity, though the BACE tests suggest this result
may not be robust.
60. Many empirical issues remain. It should be acknowledged first that the business of
testing TFP growth is, itself, very tricky. There are many methodological issues and pitfalls
involved. One issue in this regard is the decision to define productivity in terms of value-added
per capital and labor rather than gross output per capital, labor, intermediates and energy. This
may create bias in observed productivity because productivity gains from more efficient use of
intermediates and energy are ignored. The work by Feenstra et al (1999) was the only test of the
variety hypothesis where TFP based on gross output was used and this effort was limited to only
2 countries. Second, it would be ideal to find a data source that allows a direct measure of
product variety for a wide range of rich and poor countries rather than rely on export variety as a
proxy. Third, it would be desirable to use panel data, something that would be possible if only
more R&D observations were available. This would allow a test of Granger causality. Fourth, it
would be very useful to continue a line of inquiry started by Feenstra et al: does input (import)
variety matter more than output (export) variety or are they highly correlated? In this regard, it is
worth noting that export variety can include intermediates and capital. Sorting these out would
be a useful exercise. Finally, it is hard to imagine that gains from product variety are limited to
manufacturing. There may be enough data available from other sectors, perhaps agriculture, to
allow the question to be pursued elsewhere.
lBnbRhogiraph y
Addison, D., 2002. Productivity Growth, Imitation and Product Variety, Master of Philosophy
Dissertation, University of Sussex.
Aghion, P. and P. Howitt, 1998. Endogenous Growth Theory, Cambridge MA: The MIT Press.
Barro, R. and J. Lee, 1994. 'Sources of Economic Growth', Carnegie-Rochester Conference on
Public Policy, 40: 1-46.
Barro, R. and X. Sala-i-Martin, 1995. Economic Growth. New York: McGraw-Hill Inc.,
Cabral, L., 2000. Introduction to Industrial Organization. Cambridge MA: The MIT Press.
Ciccone, A. and R. Hall, 1996. 'Productivity and the Density of Economic Activity', American
Economic Review, 86: 54-70.
Crego, A., D. Larson, R. Butzer, Y. Mundlak, 1 999a. 'A New Database on Investment and
Capital for Agriculture and Manufacturing', Working Paper Series 2013, Washington DC: The
World Bank.
Davidson R. and J. MacKinnon, 1993. Estimation and Inference in Econometrics. New York:
Oxford University Press.
23
Doppelhofer, G., X. Sala-i-Martin, and R. Miller, 2000. 'Determinants of Long-run Growth: A
Bayesian Averaging of Classical Estimates (BACE) Approach', Working Paper 7750,
Cambridge MA: National Bureau of Economic Research.
Easterly W. and S. Rebelo, 1993. 'Fiscal Policy and Economic Growth: An Empirical
Investigation', Journal of Monetary Economics, 32(3): 417-458.
Feenstra R., D. Madani, T. Yang and C. Liang, 1999. 'Testing Endogenous Growth in South
Korea and Taiwan', Journal of Development Economics, 60: 317-41.
Funke M., and R. Ruhwedel, 2001. 'Product Variety and Economic Growth: Empirical Evidence
for the OECD Countries', IMF StaffPapers, 48(2), 225-242.
Griffith R., S. Redding and J. Van Reenen, 2000. 'Mapping the Two Faces of R&D:
Productivity Growth in a Panel of OECD Countries', London School of Economics and Political
Science, Centre for Economic Performance, Discussion Paper 458.
Hall, R. and C. Jones, 1998. 'Why Do Some Countries Produce So Much More Output per
Worker Than Others? ', Working Paper 6564, Cambridge MA: National Bureau of Economic
Research.
Hendry, D., 2001. 'Achievements and Challenges in Econometric Methodology', Journal of
Econometrics, 100: 7-10.
Hummels, D. and P. Klenow, 2002. 'The Variety and Quality of aNation's Trade', Working
Paper 8712, Cambridge MA: National Bureau of Economic Research.
ILO, 2000. LABORSTA, International Labour Organization, Internet. Available from
http://laborsta.ilo.org; accessed September 2000.
International Monetary Fund, International Financial Statistics, various issues, Washington DC:
The International Monetary Fund.
Jones, C., 1999. 'Growth With or Without Scale Effects?' American Economic Association
Papers and Proceedings, 89(2): 139-144.
Keller, W., 1999. 'How Trade Patterns and Technology Flows Affect Productivity Growth',
Working Paper 6990, Cambridge MA: National Bureau of Economic Research.
Klenow, P. and A. Rodriguez-Clare, 1997. 'Quantifying Variety Gainsfrom Trade
Liberalization', available from www.klenow.com.
Krugman, P., 'Scale Economies, Product Differentiation, and the Pattern of Trade', American
Economic Review, 70(5): 950-959.
24
Linder, S., 1961. An Essay on Trade and Transformation, New York: Wiley.
Marshall, A., 1959. Principles of Economics, 8h Edition, London: MacMillan Press Limited.
Martin, W. and D. Mitra, 1999. 'Productivity Growth and Convergence in Agriculture and
Manufacturing', Working Paper 2171, Washington DC: The World Bank.
Nelson, R. and E. Phelps, 1966. 'Investment in Humans, Technological Diffusion and Economic
Growth', American Economic Review, 56(2): 69-75.
OECD, Statistics Directorate, 1999. ISDB '98 International Sectoral Data Base User's Guide,
Paris: Organization for Economic Cooperation and Development.
Sachs, J. and A. Warner, (1995). 'Economic Reform and the Process of Global Integration',
Brookings Papers on Economic Activity, 1: 1-118.
Sambanis, N., 2000. 'Partition as a Solution to Ethnic War: An Empirical Critique of the
Theoretical Literature', World Politics, 52: 437-83.
Romer, P., 1990. 'Endogenous Technological Change', Journal of Political Economy, 98(5):
S71-S102.
Sala-i-Martin, X., 1997. 'I Just Ran 2 Million Regressions', American Economic Review, 87(2):
178-183.
Thomas, R., 1997. Modern Econometrics: An Introduction, Harlow: Addison-Wesley.
Young, A., 1993. 'Invention and Bounded Learning by Doing', Journal of Political Economy,
101(3): 443-72.
25
Appendix A
Data Descriptions and Sources
r ._ . PerIod Units - - Source
Output
Manufacturing value-added _ 1979-86 Constant 1980 dollars valued at PPP OECD ISDB
Manufacturig value-addd V 1979-86exchange rates World Bank
Index of total factor productivity TFP 1979 Index Author
Growth in total factor productivity gTFP 1979-86 Penod average growth rate Author
Count of SITC product categories for
Variety V 1979-86 manufactured goods, mainly in WITS
categories 5 through 8
Factor Inputs_
17-6Constant 1980 dollars valued at PP OECD 1SDB,
Value of physical capital stock K 1979-86 exchange rates Crego, Larson, Butzer and
Mundlak (1999) and author
Manufacturing employees L 1979-86 Total manufactunng employment Marin and Mitra (1999),
Population N 1979-86 Total population World Bank
Average educational attainment H 1979-86 Years of schooling Barro and Lee (1994)
Research and development R 1979 Scientists and engineers per million World Bank
employment"' people
Average life expectancy LIFE 1979-86 Years World Bank
Trade and Knowiedge I
Index bounded by zero and one
Openness to trade OPEN 950-94 Measures aompount o between 1950 and Sachs and Warner (1995)
1994
US dollar imports, SITC Category 7, WITS
Imported capital equipment M 1979-86 deflated by index of manufactures unit World Bank MUV
_____ _____ _____ ____ _____ __ __ __I _ values (M UV)
Exported manufactures X 1979-86 US dollar exports, deflated by index of WITS
Exprte m umanufactures unit values World Bank MUV
Telephone lines per 1,000 people TELE 1979-86 Main lines per 1,000 people World Bank
Govemance & Property Rights" GADP 1986-95 Index bounded by zero and one Hall and Jones (1998)
Security" CONFL 1979-86 Average civil war related deaths per Calculated from data in
annum normalized by 1979 population Sambanis (2000)
Macroeconomic Management
Consurner price inflation INFL 1979-86 Period average growth rate Calculated from IMF IFS 64
Price volatility SDINFL 1979-86 Standard deviation in annual CPI Author
inflation rate Ato
Real exchange rate RER 1979-86 Real exchange rate IMF IFS rec
Other Country Characteristics I
Urban ,share of population URB 1979-86 Percentage World Bank
Population density DENS 1979-86 Population per square kilometer World Bank
Mining MINING 1988 Mining and quarrying share of GDP Hall and Jones (1998)
a. Based on available data for 1980-91 extrapolated to 1979. Observations for Kenya, Taiwan, Tunisia and Zimbabwe are
estimates based on equation 6.
b. Based on Political Risk Services ratings of maintenance of law and order, the quality of govemment bureaucracy and the degree
of perceived government corruption for the penod 1986-95. It is assumed these ratings would not be much different from the
period 1979-86.
c. Only Colombia, Philippines, Turkey, South Africa and Zimbabwe suffered conflict related deaths.
d. All applicable growth rates based on these data are period averages except gLIFE which is defined as the exponential growth
rate between data points in 1977 and 1987
Source Addison, 2002
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS2996 Bank Competition, Financing Thorsten Beck March 2003 K. Labrie
Obstacles, and Access to Credit Asli Demirgu0-Kunt 31001
Vojislav Maksimovic
WPS2997 Financial and Legal Institutions Thorsten Beck March 2003 K Labrie
and Firm Size Asl1 Demirguq-Kunt 31001
Vojislav Maksimovic
WPS2998 Does Micro-Credit Empower Women? Mark M Pitt March 2003 P Kokila
Evidence from Bangladesh Shahidur R. Khandker 31001
Jennifer Cartwright
WPS2999 Explaining Liberalization Philipp Harms March 2003 P Flewitt
Commitments in Financial Aaditya Mattoo 32724
Services Trade Ludger Schuknecht
WPS3000 International Climate Regime beyond Franck Lecocq March 2003 V Soukhanov
2012 Are Quota Allocations Rules Renaud Crassous 35721
Robust to Uncertainty7
WPS3001 An Introduction to Financial and Antonio Estache March 2003 G Chenet-Smith
Economic Modeling for Utility Martin Rodriguez Pardina 36370
Regulators Jose Maria Rodrfguez
German Sember
WPS3002 Information-Based Instruments for Uwe Deichmann March 2003 Y. D'Souza
Improved Urban Management Somik V Lall 31449
Ajay Sun
Pragya Rajoria
WPS3003 The Investment Climate and the Mary Hallward-Driemeier March 2003 P Sintim-Aboagye
Firm Firm-Level Evidence from China Scott Wallsten 37644
Lixin Colin Xu
WPS3004 Institutions, Trade, and Growth David Dollar March 2003 A. Bonfield
Revisiting the Evidence Aart Kraay 31248
WPS3005 Vouchers for Basic Education in Varun Gauri March 2003 H Sladovich
Developing Countries A Principal- Ayesha Vawda 37698
Agent Perspective
WPS3006 Social Rights and Economics. Claims Varun Gauri March 2003 H Sladovich
to Health Care and Education in 37698
Developing Countries
WPS3007 The Impact of Urban Spatial Structure Antonio M Bento March 2003 V Soukhanov
on Travel Demand in the United Maureen L Cropper 35721
States Ahmed Mushfiq Mobarak
Katja Vinha
WPS3008 Technology, Computers, and Wages. Chris N. Sakellariou March 2003 N. Vergara
Evidence from a Developing Economy Harry A Patrinos 30432
WPS3009 Agriculture in the Doha Agenda Patrick Messerlin April 2003 P Flewitt
32724
WPS3010 Vulnerability in Consumption, Edmundo Murrugarra Apnl 2003 E. Murrugarra
Education, and health. Evidence from Jose Signoret 34452
Moldova during the Russian Crisi
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS3011 Renegotiation of Concession J. Luis Guasch April 2003 J. Troncoso
Contracts in Latin America Jean-Jacques Laffont 37826
Stephane Straub
WPS3012 Just-in-Case Inventories A Cross- J. Luis Guasch Apnl 2003 J. Troncoso
Country Analysis Joseph Kogan 37826
WPS3013 Land Sales and Rental Markets in Klaus Deininger Apnl 2003 M. Fernandez
Transition. Evidence from Rural Songqing Jin 33766
Vietnam
WPS3014 Evaluation of Financial Liberalization: Xavier Gin6 Apnil2003 K. Labrie
A General Equilibrium Model with Robert M. Townsend 31001
Constrained Occupation Choice
WPS3015 Off and Running? Technology, Trade, Carolina Sanchez-Paramo April 2003 H Sladovich
and the Rising Demand for Skilled Norbert Schady 37698
Workers in Latin America
WPS3016 Partisan Politics and Stuti Khemani April 2003 H Sladovich
Intergovernmental Transfers in India 37698
WPS3017 Why is Unemployment so High Jan Rutkowski April 2003 J. Rutkowski
in Bulgaria? 84569
WPS3018 Integrating the Least Developed Paul Brenton April2003 L.Tabada
Countries into the World Trading 36896
System The Current Impact of EU
Preferences under Everything but Arms
WPS3019 Foreign Aid, Conditionality, and Ghost Thilak Ranaweera April 2003 P. Rathan Raj
of the Financing Gap: A Forgotten 33705
Aspect of the Aid Debate
WPS3020 Consumption, Health, Gender, Anne Case April 2003 D. Ballantyne
and Poverty Angus Deaton 87198
WPS3021 Institutional Arrangements for Public Elizabeth Currie April 2003 M. Rosenquist
Debt Management Jean-Jacques Dethier 82602
Eriko Togo
WPS3022 Sending Farmers Back to School: Gershon Feder April 2003 P. Kokila
The Impact of Farmer Field Schools Rinku Murgai 33716
in Indonesia Jaime B. Quizon