Empirical Evidence From Swedish Manufacturing Firms Relationship between R&D and Productivity doc

Masters of Economics of Innovation and Growth.



Thesis

Empirical Evidence From Swedish Manufacturing Firms

Relationship between R&D and Productivity.






Prepared by:
Mohammed Najim Uddin




Supervisor:
Hans Lööf
Associate Professor,
Economics of Innovation and Growth,

CESIS.


Royal Institute of Technology (KTH).



1





Acknowledgement

This dissertation is the partial fulfilment of Master programme in Economics of Innovation
and Growth at The Royal Institute of Technology (KTH), in Stockholm Sweden. The paper
was the result of a series of meeting with my supervisor Professor Hans Lööf. I am grateful
to him for kind direction. I am also grateful to Professor Borje Johansson, Almas
Heshmati, Martin Anderson and other professor in the department of Economics of
Innovation and Growth during my programme I got help from them. I would like to thank
you from my heart for the support, guidance, invaluable suggestions throughout this study.
My special thank to Professor Hans Lööf for giving me an opportunity to work on the firm
level data set and his valuable instruction. Mr Lööf strongly impressed me during this
programme. I highly appreciate to Sofia Norlander and Joanna Wasilewska for their great
help in my whole study period in KTH, Sweden. Lots of thanks to my classmates in this
program.

2





Content Page

1. Introduction …………………………………………………… 5-7

2. Literature review … 7-15

2.1. The history of economic growth theory ………………… 7-11


2.2. Empirical evidence of R&D and productivity relationship … 11-15


3. Objectives and hypothesis ………………………………………… 15-16

4. Data and methodology analysis ……………………………………. 16-24

4.1. Data and variable analysis ……………………………… 16-19

4.2. Research methodology analysis ….……….……………… 19-24

5. Empirical result analysis ………………………………………… 24-26

6. Limitation ……………………………………………………… 26-27

7. Conclusion ………………………………………………………… 27-28

8. References ………………………………………………………. 29-33

9. Appendix ………………………………………………………… 34-40





3




Abstract
Many empirical studies have been introduced to show the relationship among R&D and
productivity at the firm level. The motivation of this paper is to extend the literature of the
relationship between R&D and productivity level at the firm level. This research paper

deals only firms direct benefits that actually gains from conducting the research. The paper
is based on 6665 Swedish manufacturing firms’ unbalanced panel data set during the
period 1992-2000. To estimate the R&D impact on productivity level this study uses the
autoregressive model or dynamic model. This model works on where large number of
firms and small number of period’s data are observed. The study employs the econometrics
tools OLS, fixed effect and generalized method of moment (GMM) estimators. The
research is conducted with and without industry dummy in the model. The empirical
results confirm that the R&D expenditure has significant impact on the firm level
productivity at the 5% level of significance. The empirical result suggests that industry
specific effect has no impact on significant level.









_________________________________________________________________
Key words: OLS, fixed effect, generalized methods of moment (GMM)







4

1. Introduction:

Research and development (R&D) spending have been increasing all over in the
world. It increases the stock of useful knowledge through research and development
activity. It is one of the crucial determinants of productivity and economic growth. The
neo-classical growth model argues that the long run growth is exogeneously determined by
either saving rate or rate of technological change. Neoclassical economics state that
technological progression and other external factors are the main sources of economic
growth. On the other hand, endogenous growth theory agrues that economic growth is
generated from within a system as a direct result of internal processes. Endogenous growth
theory explains that growth is usually determined by the production of new technologies
and human capital given to the production. More specifically, the theory refers the
enhancement of a nation's human capital that will lead to economic growth through new
forms of technology, process of specialization, efficient and effective means of production.
According to the endogenous growth theory the most important factor for determining the
economic growth rate is the rate of advance of a country's use of knowledge stock and its
important determinant is R&D productivity. Because R&D expenditure is the source of
valuable knowledge that leads to improve innovation, specialization and productivity.
Innovation increases product market competition and stimulates the process of creative
destruction which brings new business opportunities of the firms. Innovation induces firms
to enter and exit from the market.

R&D provides benefits two ways: one is direct productivity benefits and another
is indirect benefits. The direct productivity benefit occurs through conducting research
such as automobile or aircraft manufacturers industry and the indirect benefits come
through new technology spreading to others parts of the economy. In the present
monopolistic competition market, firms achieve monopolistic power through product
differentiation. R&D has a crucial role to obtain product variety. R&D investment result is

invention, new ideas, design and productivity increment which can be a source of
competitive advantages in the global market economy. The firms can sustain growth
through investment in R&D. Productivity is one of the key driving forces of economic
growth. The term productivity refers to measures the output from production process per
unit of input.


5



To measure the gains from R&D spending the researchers rely on firm level
productivity measurement analysis. The available empirical studies have generally
confirmed the significant role of R&D investment on productivity level at the firm level.
Most of the studies can be divided into two categories. Production function based studies
and cost function based studies. The production function based studies show the impact of
R&D on productivity level that is R&D elasticity and the cost function based studies show
the R&D impact on production cost. Some studies have estimated the rate of return in
R&D (CBO -2005). From the literature of productivity measurement studies it can be
observed that there is no single measurement of productivity. Broadly, productivity
measurement can be classified into two categories: - single factor and multi-factor
productivity (MFP) measure. The single factor productivity measure is also called partial
productivity measure. Multi-factor productivity again can be two forms one is value added
based capital-labour MFP measure and another one is gross output based capital, labour,
energy and material (KLEMS) MFP productivity measures. In case of value added concept
where value added is considered as firms out put (OECD Manual-2001). The most
frequently used productivity measurement methods can be expressed by the following tree
diagram:
Figure-1: productivity measurement methods:

Productivity Measurement



production function based


Cost function based




Single factor productivity
Multifactor productivity

measurement
measurement




Value added or sales based
capital-labour MFP measurement

Gross output based capital,

labour, energy and material

(
KLEM

)
MFP


Source: Measurement of aggregate and industry level productivity growth: OECD manual (2001).
6



Among those measurements, most of the studies uses value added based
productivity measure. To explore the relationship between R&D and productivity level this
research paper uses the value added concept. The research paper will be organized by
using seven sections where section two provides literature review, section three present the
objectives and hypothesis of this paper, section four explains the data and methodology
analysis of the study, section five contains empirical result, Section six focuses on
limitation of this study. Finally, in the last section gives some conclusions.

2. Literature review:

2.1. The evolution of growth theory:
The growth theory and growth empirics are attractive subjects in economics. The
modern concept of economic growth started with the critique of Mercantilism, especially
by the physiocrats. During the 16
th
to 18
th
century the mercantilist believed that nation
wealth and power were best served by increasing export and collecting precious metals in
return. The mercantilism focused on ruler’s wealth, accumulation of gold or the balance of
trade. Physiocracy is a school of thought founded by François Quesnay (1694-1774). This

theory originated in France and was most popular during the second half of the 18
th

century. Physiocracy is the first well developed theory in economics. This doctrine was
dominated by Marquis de Mirabeau, Mercier de la Riviere, Dupont de Nemours, La
Tronse, the Abbe Baudeau and others. The main theme of this doctrine was Francois
Quesnay’s (1759) axiom that only agriculture yielded a surplus- what he called a net
product. The physiocrats believed that the wealth of nations was derived solely from the
value of land agriculture or land development. From the viewpoint of modern economics
the main weakness is that they only consider agricultural labor to be valuable. Physiocrats
viewed the production of goods and services as consumption of the agricultural surplus,
while modern economists consider these to be productive activities which add to national
income.The most important contribution of the physiocrats was emphasis on productive
work as the source of national wealth. The productive capacity itself allows growth and
increment of national wealth.

The classical economics is the first modern school of economics of thought. Adam
Smith “The Wealth of Nations” in 1776 is considered the beginning of this school. Adam
Smith, David Ricardo, Thomas Malthus and John Stuart Mill are founders of this school.

7



The "Classical" school is sometimes called the "Surplus" school. Often classical economics
school expanded up to William Pretty, Johann Heinrich Von Thunen and Karl Marx.
Classical economists explained the growth and development. Adam Smith explained a
supply side driven model of growth. This model can be expressed by simple production
function:

Y = f (L,K,T)

Where Y is the output, L is labour, K is capital and T is land. The out put growth (g
Y
) was
driven by population growth (g
L
), investment (g
K
) and land growth (g
T
) and increases
overall productivity (g
f
). Smith suggested that growth was related to capital accumulation,
technological progress and institutional and social factors.

g
Y
= φ(g
f,
g
L,
g
K,
g
T
)

Where time was endogenous it depends on the sustenance available to accommodate the

increasing workforce. Investment (g
K
) was also endogenous: determined by the rate of
savings; land growth (g
T
) was dependent on the conquest of new lands or technological
improvements of fertility of old lands. Technological progress could also increase growth
overall. Smith also emphasized improvements in machinery and international trade as
engines of growth as they facilitate further specialization. He stated that the division of
labour improves growth, was a fundamental argument.

David Ricardo (1817) modified Smith idea including diminishing return to land.
His most important assumption was that economic growth must decline and end due to the
scarcity of land and its diminishing marginal productivity. He showed that how
distributional changes between wages, rent, interest and profit affected the prospects for
long run capital accumulation and growth. According to Ricardo’s idea output growth
requires growth of factor input. But land supply is limited. Land can not be produced or
created. As a result two effects occur on growth. Firstly, landowner rents increases
overtime reducing the profit of capitalist and secondly, wage goods will be rising in price
and this reduces profit as workers require higher wages. As the economy continued to
grow, then, by his theory of distribution, profits would be eventually squeezed out by rents
and wages. In the limit, Ricardo argued, a "stationary state" would be reached where
capitalists will be making near-zero profits and no further accumulation would occur.

8



Ricardo suggested that this decline can be checked by technological improvements and
foreign trade.

Ricardo first claimed that technical improvements would help push the marginal
product of land cultivation upwards and thus allow for more growth. But in his later third
edition of his principles, Ricardo modified his position on machinery. He noted that
machinery displaces labour and that ‘set free’ might not be reabsorbed elsewhere and thus
merely generate downward pressure on wage and thus lower labour income. In order to
reabsorb this extra labour without this effect, then the rate of capital accumulation must be
increased.

Malthus (1796) in his ‘Essay on the Principle of Population." In essence, Malthus
said that any growth in the economy would translate into a growth in population. He
claimed in his hypothesis that population growth exceed the growth of mean of
subsistence. Actual population growth is kept in line with food supply growth by positive
checks. If the population growth was not easily checked and would quickly outstrip growth
and increasing misery all around.

John Stuart Mill (1806-1873) improved little upon Ricardo, Mill first adopted
Ricardo's view that the average wage is determined by a fixed amount of capital divided by
the number of workers, he stated that other factors play a role in determining wages,
among them workers' expectations as well as various institutional factors.

Karl Marx (1818-1883) modified the growth theory through his reproduction
scheme. He believed that all production belong to labour because workers produce all
value within society. The market system allows capitalists, the owner of machinery and
factories, to exploit workers denying them a fare share of what they produce. He predicted


_______________________________________________________________________
Note:

9



that capitalist compete for profit that led capitalist to adopt labour saving machinery. He
used the multi sectoral context and provided the steady state growth equilibrium. Marx did
not believe that labour supply was endogenous to wage. Marx claimed that wage is not
determined by necessity or natural factors but rather by bargaining between capitalist and
workers. Marx also saw that the profit and raw material as the determinant of savings and
capital accumulation.

John Maynard Keynes (1936) distinguishes his theory from classical economics.
He developed theory that explains determinants of saving, consumption, investment and
production. In that theory, the interaction of aggregate demand and aggregate supply
determines the level of output and employment in the economy. According to Keynes
investment demand is one of the determinants of aggregate demand and that demand is
linked to output via the multiplier. He argued that without market imperfections, aggregate
demand might fall short of the aggregate productive capacity of its labour and capital.

Keynes theory of demand determined equilibrium first extended Sir Roy F.
Harrod (Harrod in 1939, Domar in 1946) into a theory of growth that is the ‘Harrod-
Domar’ model of growth. The Harrod –Domar model was initially created to help analysis
the business cycle. Harrod-Domar (1946) model is used in development economics to
explain an economy’s growth rate in terms of the level of savings and productivity of
capital. It suggests that there is no natural reason for an economy to have balanced growth.
Its implication was that growth depends on the quantity of labour and capital; more
investment leads to capital accumulation, which generates economic growth.

The neo-classical model was an extension of Harrod-Domar model (1946) that
includes new term, productivity growth. The most important contribution was given Robert
Solow (1956). Solow and T.W. Swan developed neo-classical model of economic growth
which is also called Solow-Swan model or exogenous growth model. The exogenous
model shows how economies will naturally tend to steady state. This model explains the
long-run economic growth. Solow extended the Harrod-Domar model by including: (a)
labour as a factor of production; (b) Diminishing return occur to labour and capital
separately. And constant return to scale for both factor combined; and (c) A time varying
technology distinct from capital and labour. Capital output ratio is not fixed as they are in

10



the Harrod-Domar model. In neo classical growth model the long run growth rate is
exogenously determined, in other words it is determined out side of the model. The
common prediction of this model is that an economy will always converge towards steady
state rate of growth which depends on the rate of technological progress and the rate of
labour force growth.

The endogenous growth theory is also called new growth theory that was
developed in the 1980s as a response to criticism of the neo-classical growth model. Romer
(1986) developed the model of increasing returns in which there was a stable positive
equilibrium growth rate that resulted from endogenous accumulation of knowledge. The
most important work of this model that distinguishes itself from neoclassical growth by
emphasizing that economic growth is an endogenous outcome of an economic system, not
the result of forces that impinge from outside. The importance is usually given to the
production of new technologies and human capital. Endogenous growth theory argues that
economic growth is generated from within a system as a direct result of internal processes.
This was an important break with the existing literature, in which technological progress

had largely been treated as completely exogenous. More specifically, the basic mechanism
of endogenous growth theory is productive externalities. The enhancement of a nation's
human capital will lead to economic growth by means of the development of new forms of
technology and efficient and effective means of production. Endogenous growth theory
demonstrate that policy measure have an impact on the long run growth rate of an
economy. Subsidies on research and development or education increase the growth rate by
increasing the incentive to innovation. R&D has a special economic significance apart
from its conventional association with scientific and technological development. R&D
investment generally reflects a government and organization willingness to forgo current
operation or profits to improve future performance or returns and its ability to conduct
research and development. This is necessary due to technology change and development as
well as other competitors and the changing preference of customers.

________________________________________________________________________
Note:




11



2.2. Empirical evidence of R&D and productivity relationship:

Many studies have consistently proved that R&D investment and firm’s
performance has positive correlation. A large number of empirical studies have estimated
the effect of R&D investment on productivity. Most of the studies are based on production
function.

Griliches (1980) investigated the R&D elasticises on the 883 U.S. firms during the
period 1957 to 1965. Author has found the R&D elasticity is positive and approximately
0.08. In another study, author (Grillchese-1986) explored the relationship between R&D
investment and productivity on 1000 largest US manufacturing company during the period
1957-1977. Later study used the Cobb-Douglas production function and found that output
elasticity with respect to R&D investment is 0.086. Griliches and Mairesse (1984) worked
on 133 US manufacturing firm during the period 1966-1977 and found the R&D elasticity
is 0.08(without year dummies) and 0.09 (with year dummies). Minasian (1969) conducted
the study on 17 US manufacturing firms during the periods 1948-1957. This study uses the
multiple regression models and estimated the R&D elasticity on value-added and found
R&D coefficient is 0.08 and statistically significant.

Jaffe (1986) explored the relationship between R&D and productivity on 432 US
firm for the period 1973-1979. This study uses patent data to classify the firms in
technology- based categories and found significant and positive R&D coefficient (0.098).
Hall & Mairesse (1995) examined the R&D elasticity on 197 French manufacturing firms
for the period 1980 to 19870. The study found the productivity of R&D spending for
French manufacturing firms is positive (0.07). Kafouros (2004) focused on the 78 UK
manufacturing firms for the period (1989-2002). This study examined the contribution of
R&D is approximately 0.04. The elasticity for large firm (0.44) is and for small firm is
0.035. The R&D-elasticity is considerably high for high-tech sectors (0.11), but
statistically insignificant for low-tech sectors. Beneito (2001) investigated the R&D
elasticity on 501 Spanish manufacturing firms. The empirical result suggests that R&D
elasticity of productivity is 0.07 and statistically significant.

Klette & Kortum (2002) use the dynamic model and showed that R&D as a
fraction of revenues is strongly related to firm productivity and largely unrelated to firm
size or growth. Mogens et al (1999) have used a production function approach to estimate

12

the effects of the R&D capital on output. They use 684 Danish manufacturing firms during
the period 1987 to 1995, the empirical findings was a positive and output elasticity of
R&D capital is 0.08 and statistically significant. Heshmati and Lööf (2006) using the
multivariate vector autoregressive model have found the significant heterogeneity between
the firms’ investment and performance behaviour by their size. The study was conducted in
Swedish manufacturing firm it also refers that relationship between investment and firm
performance is existed in both case of cross-sectional and time series analysis. The
empirical result shows that a significant and positive elasticity of productivity with respect
to human capital, physical capital and knowledge capital.

Wakelin (1997) worked on 170 UK manufacturing firms. Using Cobb-Douglas
production function the study demonstrated that a positive and significant relationship
between R&D expenditure and productivity when no control is made for sector effects. But
the relationship is insignificant when sector dummy variables are introduced (fixed effects
are included). The study also showed that innovative firms spent more on R&D
expenditure relative to sales than non-innovating firms (2.3% against 0.8% in the period
1988 to 1992).

Wang and Tsai (2002) estimated the impact of R&D on productivity within the
private sector, Based on a sample of 136 large manufacturing firms listed in the Taiwan
Stock Exchange (TSE) during the period 1994-2000. In this research authors uses the
Cobb-Douglas production function and estimated R&D output elasticity was around
0.18;but, when the whole sample is classified into two categories, high-tech and other
firms, The study found a statistically significant difference in R&D elasticity between the
two samples. The R&D elasticity for high-tech firms is around 0.3, but only 0.07 for other
firms. The study also demonstrated that the Schumpeterian hypothesis, that the impact of
R&D on productivity is an increasing function of firm size, is also tested within this study;

however, the empirical results do not support the proposition at the 5 per cent significance
level.

Lesley et al (2008) have used an unbalanced longitudinal database on 532 top
European R&D investors over the six-year period 2000-2005. They divided the panel data
into three subgroups high, medium and low- tech group. This study compared between the

13



POLS (pooled ordinary least square) and the random effect model. The empirical results of
random effect model suggest that knowledge stock has a significant positive impact on a
firm's productivity, with an overall elasticity of about 0.125. The estimated elasticity
coefficients of the high, medium and low- tech firm are 0.160, 0.146 and 0.068
respectively. Where as the POLS estimated elasticity for overall sample is 0.123. The
estimated elasticity for high, medium and low tech is 0.18, 0.138 and 0.048 respectively.

Bond et al (2003) use the dynamic production function approach and estimated the
impact of R&D on productivity between German and UK firms. The study conducted on
more than 200 firms German and UK in each country during the period 1987 -1996. After

Table-1: Firms R&D elasticity in different studies:
Study R&D
elasticity
Firm and periods
Griliches (1980)
0.08
883 U.S. firms; 1957 to 1965
Griliches and Mairesse (1984)

0.09
133 U.S. firms; 1966 to 1977
Jaffe (1986)
0.098
432 U.S. firms; 1973 to 1979
Griliches (1986)
0.086
1000 largest US manufacturing company
1957-1977.
Minasian (1969)
0.08
17 U.S. firms; 1948 to 1957
Hall and Mairesse (1992)
0.07
197 French manufacturing firms; 1980 to
1987
Wang and Tsai(2002)
0.18
136 large manufacturing firms listed in the
Taiwan Stock Exchange (TSE) during the
period 1994-2000,
Beneito(2001)
0.07
501 Spanish manufacturing firms 1990-
1996
Potters et al (2008)
0.13
532 top European R&D investors over the
six-year period 2000-2005.
Kafouro(2004)

0.04
78 firms in UK, (1989–2002),
Mogens et al (1999)
0.08
684 Danish manufacturing firms (1987-
1995)
Kwon and Inui (2003)

0.093

3830 Japanese manufacturing
Firms for the period between 1995 and
1998.
Bond, et al (2003) 0.079(Ger
many)
0.065(UK)
200 firms in German and UK
In each country, 1987 -1996.


14



controlling for firm size and industry effects. This study has found that the R&D output
elasticity is approximately the same in both countries. In UK, they obtain a preferred
estimate of 0.065 (SE 0.024) while the estimate for the German sample is 0.079 (SE
0.042).

Doraszelski, and Jaumandreu, (2006) uses the dynamic model on 1800 Spanish

manufacturing firms in nine industries during the 1990s. Empirical findings indicate that
the link between R&D and productivity is subject to a high degree of uncertainty,
nonlinearity, and heterogeneity across firms. R&D expenditures stimulate productivity.
The growth in expected productivity corresponding to observations with R&D
expenditures is often higher than the growth corresponding to observations without R&D.
they estimate this difference in expected productivity to be around 5% in most cases and
up to 9% in some cases. The following table-1 provides the empirical evidence of some
studies about the R&D elasticity on productivity level.

Kwona and Inui (2003) estimated a Cobb-Douglas production function, this paper
investigated on 3,830 Japanese manufacturing firms for the period between 1995 and 1998.
A positive and significant role is found on R&D expenditure and productivity. The R&D
elasticity is 0.09. This study also showed that the roles are different by the firms’ sizes and
characteristics of technology. The empirical result showed that the effects of R&D on
productivity improvement are larger for the large sized and high-tech firms than they are
for the smaller sized and low-tech firms.

The evidence from these studies indicates that R&D expenditure has a positive
and significant impact on firm level productivity in US, UK, Spain, Germany, French,
Denmark and Japan.

3. Objectives:

Since many studies in different countries has confirmed the significant relationship
between R&D and firm level productivity. Hence the motivation of this research work is to
examine the relation between R&D and productivity level using dynamic model in case of
Swedish manufacturing industry. The research will be conducted on 6665 Swedish firms
during the period 1992-2000. This research paper investigates the following hypothesis:

15

a) Ho: The null hypothesis is that R&D spending has no impact on productivity.
b) Ha: The alternative hypothesis is that R&D spending has a positive role on
Productivity.

4. Data and methodology analysis:

4.1. Data and variable analysis:

The data set contain different kinds of information of Swedish manufacturing
firms for the period 1992-2000. The data set include many firms among many industries.
The following table-3 provides the summary statistics for the variables those are used in
this study analysis.

Table 3: Number of observations and variables:
Variable Observations Mean Standard
deviation
Minimum
Maximum
Value added 6665 6.06479 0.5937051 1.038458 11.14649
Capital 6665 5.404803 1.159983 .6931472 12.71501
Employment 6665 5.309587 1.122408 0.6931472 10.63099
R&D 6665 2.682535 1.701679 4.364372 8.208708
Sales 6665 7.238846 0.5867245 3.840305 11.84665
EBIT 6665 4.009222 1.466744 3.958907 10.62835
Human
Capital
6665 0.0982907 0.0905808 0 00.7

Note: EBIT: Earning before income tax. R&D: Research and Development expenditure.


It is unbalanced panel data and consists of various information such as value added, sales,
wage, employment, R&D expenditure, gross investment, labour cost, production cost,
earning before income tax (EBIT), human capital, subsidy and others. Some of the extreme
value of the variable can affect result of the research. Hence the data set has been trimmed
outliers using different techniques. The data set contain the negative wage value that is
required to be positive. The study also trimmed out the variables those which are not
necessary for this model analysis. After shorting the data set it contains 6665 Swedish

16



manufacturing firms for the period 1992-2000. All the variables in the table-3 have been
reported in the log form. The brief descriptions of the variables are as follows:

Value added (Per employee log value added):
The original data set was given in total value added form. The total value added of
every firm in each period is divided by their corresponding employee number and then log
value has been taken.

Capital (log value of per employee net capital):
The data set report the physical capital of every firm which is transformed into net
capital through perpetual inventory method and the constant rate of depreciation is applied.
In this case I follow the same procedure conducted by Hall and Mairesse (1995), Lööf
(2008), Hyeog Ug Kwon and Tomohiko Inui (2003). The perpetual inventory method as
follows:

NECA= K
t-1
(1-δ) + I
t

Where, NECA is consist of the last year depreciated tangible assets and the gross
investment. K
t-1
is the last year net capital investment; δ is the depreciation rate and I
t
is the
gross investment. The depreciation rate is constant overtime (0.15). The net capital has
been transformed into per employee net capital by divided with employee number of this
firm.

Employment (log value of employee):
The preferred measure of labour input is hours worked, but some researchers use
number of people employed if hours are unavailable. For the firm level data most of the
uses hours as measures of labour unit. If working hours is not available researcher uses the
researcher number of people employed in these establishments. Here, the data set report
the number of employee according to the establishment. This paper uses log value of
employee.

________________________________________________________________________
definition:(The perpetual inventory method (PIM) IMF, OECD, Eurostat, ILO): The perpetual inventory method (PIM)
produces an estimate of the stock of fixed assets in existence and in the hands of producers by estimating how many of
the fixed assets installed as a result of gross fixed capital formation undertaken in previous years have survived to the
current period.

17

R&D( log value of per employee R&D spending):
The data set covers every firms every firm which reported positive R&D
expenditure and the number of employees are more than 49. This study has trimmed
outsome of the extreme observation of R&D spending that may influence result. The
author observed that a few number of firms per employee spending was more than 2000
million and these observations have been removed from the data set. Then log value of
R&D spending has been taken.

Human capital(engineering knowledge and general knowledge):
The human capital is consisted of engineering knowledge and general knowledge.
The employees who have university degree are counted as engineering knowledge and the
employees those who are undergraduate are counted as general knowledge.

The table-4 provides the correlation result of the main variables which have been
used in this study. The auto correlation table have been presented in the appendix part. A
correlation is a single number that describes the degree of relationship between two
variables. The correlation coefficient indicates the strength and direction of a linear
relationship between two random variables. The closer the coefficient is either-1 or 1, the
stronger the correlation between the variables.

Table-4: correlation matrix:

Value
added
Capital Employment R&D Sales EBIT Human
Capital
Value

added
1.0000


Capital 0.3719

1.0000
Employ
ment
0.1201 0.2341 1.0000
R&D 0.2037

0.0131 0.1818 1.0000
Sales 0.5952

0.4459 0.1496 0.2422 1.0000
EBIT 0.6230

0.3603 0.1074 0.2256 0.5719 1.0000
Human
Capital
0.1399 -0.1165 0.0624 0.3353 0.0624 0.0582 1.0000


18



The correlation matrix -4 report that the sales and EBIT is highly correlated with
value added. In case of productivity measurement value added, sales and earning before

income tax (EBIT) are considered as productivity of firms. The correlation between
productivity and R&D is 0.2037, in case of capital and employment these coefficients are,
0.3719 and 0.1201 respectively.


4.2. Research methodology analysis:

Most of the research paper uses Cobb-Douglas production function approach to
estimate the contribution of R&D on productivity that is a mathematical equation which
explain how inputs are combined to produce output (See for example, Griliches (1986);
Mairesse and Sassenou (1991), Lichtenberg and Siegel (1989); Goto and Suzuki (1989),
Wang and Tsai (2002), Mogens et al (1999), Kwon and Inui (2003), Wakelin (1997)). The
extended form of Cobb- Douglas production function is:

Q
it
= Ae
λτ
K
α
it-1
L
β
it
R
γ
it-1
e
Өit +

εit
(i)

The definition of the variables differs depending on the study. The studies that use
firm-level data tend to use firms’ revenues, production or sales, or value-added a measure
of output (Q). L is the labour input, K is the physical capital input, R is the R&D capital
investment. Where Α is the level of technology of firms that is constant (TFP). α, β and γ
are the elasticity of production with respect to capital, labour and R&D expenditure
respectively and e is the disembodied technological change and λ is the rate of
disembodied technological change and λt means the time trend is usually replaced by time
dummies in the estimation that capture common technological change, shocks, such as
strikes and weather, and allow such aggregate shock to have flexible effects on Q and free
correlation with K and L (Kwona and Inui, 2003). Ө
it
is the firm specific variables. The
notation i-represents the firms or sectors and t- represent the period or year. The subscript
t-1 is the lagged values on the physical capital and R&D capital and ε
it
is the disturbance
term. The logarithmic form of the equation (i) is as follows:

Log (Q
it
)=log (A) + λt+ Ө
it
+ α log (K
it-1
) + β log (L
it
) + γ

1
log (R
it-1
) +

ε
it
(ii)



19



Within this framework the studies focus mainly on the estimated elasticity γ of R&D
capital. The coefficient of the R&D variable γ measures the elasticity of output with
respect to R&D effort.

In order to examine the R&D impact on productivity this paper uses the
autoregressive model. In the time series regression analysis it is playing an important role
to explain how the dependent variable and independent variables are related. The model
works well when panels are large and time period is small. The impact of independent
variable on dependent variable can not occur immediately the total effect is distributed
over serial periods of time. R&D capital has been viewed as a measurement of the current
state of technical knowledge that is determined by current and past R&D expenditure
(Griliches, 1979). But it is difficult to determine the appropriate lag structure. Most of the
studies consider R&D has a significant impact on productivity two years later. This paper
uses autoregressive model with three inputs such as physical capital, labour and knowledge
capital (R&D). The model is as follows:

y
it
= αy
it-1
+ βx
it
+ (η
i
+ v
it
) |α|<1 ; i = 1,2 N; t = 1, 2 T (iii)



The above equation contains a lagged dependent variable as an explanatory
variable. Where the certain time period is t, the previous period is t-1. This model specifies
that in the period t, y is determined by the value of previous year y and by the t previous
values of x. Therefore, the effect of x variable is distributed over t+1 periods of time. y
it
is
the log value of output for firms i in period t, y
it-1
is log value of previous year output of
the firms i. The variable x
it
is the log value of explanatory variables. α, and β are the
parameters to be estimated. The error component is consisted of two parts; one (η
i
) that is

unobserved individual time invariant effects and another (v
it
) which is common disturbance
term. In the equation (iii) this study assumes that (η
i
) is stochastic term and the disturbance
term (v
it
) is serially uncorrelated. The number of firms (N) are sufficient large and time
period (T) are small.

The coefficient α and β can be estimated by the method of ordinary least squares
(OLS). In the presence of the lagged dependent variable the OLS estimation rule does not
give a linear unbiased estimator. The ordinary least square (OLS) of the dynamic model

20



can be biased and inconsistent. Because the OLS method ignores the firm specific
individual effects (η
i
) which are potentially correlated with the lagged value of previous
year output (y
it-1
). When the individual effects are unobserved then there is an endogeneity
problem. In order to eliminate endogeneity problem one of the most commonly used
solutions is fixed effect estimation technique. Fixed effect estimation requires the
assumption that the heterogeneity is constant across time. This heterogeneity is typically;
managerial quality and structure, brand image, and market networks, the ethnicity, the year

and location etc (Wang and Tsai- 2002, Kwon and Inui-2003). The conventional fixed-
effects treatment of time-invariant individual effects is called the within estimator, which is
equivalent to least squares after transformation of the data to deviations from individual
means(Seung Chan Ahn, Young Hoon Lee, Peter Schmidt).

The disadvantage of fixed effects is a large loss of degrees of freedom and too
many dummies may aggravate the problem of multicollinearity among the regressors
(Baltagi, 2001). Besides, Nickell (1981) also shows that the estimator of the fixed effect
model is not consistent as the number of individuals goes to infinity but the time period
remain finite.
If there are many individuals (firms) there are mathematically equivalent models
which achieve the same effect. This model is called the random effect model. A random
effect model makes the assumption that the individual effects are randomly distributed. To
choose between fixed and random effect the Hausman test is applied. The Hausman test
checks the more efficient model against a less efficient one. In the Hausman test analysis
the null hypothesis can not be rejected if Probability is greater than Chi
2
(Probability>
Chi
2
). That means random effect model is more efficient than the fixed effect estimator and
safe to use in the productivity analysis. Random-effects estimator considers weighted
average of the fixed and between estimates. But random effect estimator is inconsistent in
case of time series data.

Another econometric problem associated with time series estimates of production
function is simultaneity bias which arises when one or more of the explanatory variables in
a regression equation are jointly determined with the dependent variable. This is the
violation of OLS independent assumption. This assumption is satisfied when the flow of
causality runs in one direction, from the explanatory variables to the dependent variable.

21



R&D investment and productivity are likely to be mutually dependent. Growth of output is
a function of R&D investment and R&D investment is a function of past output growth. If
the explanatory and dependent both variables are determined simultaneously OLS based
estimates of the coefficient will be biased (CBO-2005). Several statistical techniques are
used to solve the simultaneity bias problem instrumental variables (IV) and GMM are two
of them. The conventional IV estimator (though consistent) is, however, inefficient in the
presence of heteroskedasticity. The usual approach today when facing heteroskedasticity of
unknown form is to use the Generalized Method of Moments (GMM), first introduced by
Hansen (1982). GMM makes use of the orthogonality conditions to allow for efficient
estimation in the presence of heteroskedasticity of unknown form (Christopher et al 2003).

In case of time series data endogeneity problem arises when the independent
variable is correlated with error term in a regression model. This implies OLS is biased.
The GMM technique controls the endogeneity problem of the model. Linear dynamic
panel-data models include p lags of the dependent variable as covariates and contain
unobserved panel-level effects, fixed or random. By construction, the unobserved panel-
level effects are correlated with the lagged dependent variables, making standard
estimators inconsistent. Arellano and Bond (1991) derived a consistent generalized
method-of-moments (GMM) estimator for this model. The Arellano and Bond estimator
can perform poorly if the autoregressive parameters are too large or the ratio of the
variance of the panel-level effect to the variance of idiosyncratic error is too large.
Building on the work of Arellano and Bover (1995), Blundell and Bond (1998) developed
a system GMM estimator that uses additional moment conditions (xtdpdsys: implements
this estimator). This estimator is designed for datasets with many panels and few periods.
This method assumes that there is no autocorrelation in the idiosyncratic errors and

requires the initial condition that the panel-level effects be uncorrelated with the first
difference of the first observation of the dependent variable (Stata). Since the data set
covers the period 1992-2000. So the time period is small and data set is large. Therefore,
system GMM estimator is more appropriate for this R&D and productivity analysis. The
system GMM estimator uses the levels equation to obtain a system of two equations: one
differenced and one in levels equation. By adding the second equation additional
instruments can be obtained. Thus the variables in levels in the second equation are
instrumented with their own first differences. This usually increases efficiency.

22



The study also reports the two tests; serial correlation and sargan over
identification test. Auto-correlation generally occurs when we use time series.
Autocorrelation is a special case of correlation, and refers not to the relationship between
two or more variables, but to the relationship between successive values of the same
variable. One of the assumptions of regression analysis is that the error terms are
independent from one another. Formally, this assumption is expressed as E(
ε
i
ε
j
) = Cov(
ε
i
ε
j
)
=0 for all i ≠ j. The violation of this assumption gives rise to auto correlation. If this

assumption is not satisfied it means that the values of the error term are not independent,
that is, the error in some period influences the error in some subsequent period next
period or beyond. Wind Meijer (2004) has shown that the estimated asymptotic standard
error of the two step GMM estimator can be severely biased downward in case of small
sample. Hence the auto correlation test in the dynamic panel model is very important
together with the parameter estimations.

On the other hand, Sargan test is a test of the validity of instrumental variables. It
is a test of the over-identifying restriction. The hypothesis is being tested with the Sargan
test is that the instrumental variable are uncorrelated to some set of residuals, and there for
they are acceptable, healthy instruments. If the null hypothesis is confirmed statistically
(that is not rejected) the instrument pass the test. They are valid by this criterion.

Model specification:

In this dissertation, the investigation will be conducted using following two
equations where value added is used as a dependent variable. Keith C. Brown (1968)
mentioned that the industry dummy variables are introduced in order to explain industry
difference which can not be explained by the control variables (leverage, growth, and
payout). This research work uses the autoregressive distributed lag model that is described
in the equation (iv) and (v). The equation (iv) represent the model where industry dummy
(NC*) is not included and the equation (v) represents the model where industry dummy is
included in the model.

y
it
= αy
it-1
+β
1

K
it
+β
2
K
it-1
+γ
1
L
it
+γ
2
L
it-1
+ ξ
1
R
it
+ξ
2
R
it-1
+ ξ
2
R
it-2
+ T (iv)


y

it
= αy
it-1
+β
1
K
it
+β
2
K
it-1
+γ
1
L
it
+γ
2
L
it-1
+ ξ
1
R
it
+ξ
2
R
it-1
+ ξ
2
R

it-2
+ T + n
c
(v)

23



In equation (iv) the log form of value added (y
it
) depends on the previous year log form of
value added(y
it-1
), current and one year lagged values of the physical capital input (K
it
, K
it-
1
), current year labor input (L
it
), current (R
it
) and two years lagged values of the R&D
investment (R
it-1,
R
it-2
), year dummies (T). Firm’s innovation is a function of its current
R&D and knowledge generated by its past R&D. The equation (v) shows industry

dummies (n
c
) impact on productivity. Dummy variables are used to represent qualitative
variable in the regression.

5. Empirical result analysis:

In this result section, the tables-5 and 6 provide the partial result of the dynamic
model of equation (iv) and (v) where only R&D impact on productivity is reported with
respect to 1
st
, 2
nd
and 3
rd
year lag. The complete model results have been provided in the
appendix tables 6 and 7 respectively. The table-5 reports the result of equation (iv) where
industry dummy is excluded from the model and table-6 represent the result of equation (v)
where industry dummy is included in the model. The model is estimated using the
statistical tools OLS, fixed effect and system GMM estimator. The column1, 2 and 3 of the
table-5 provides OLS, fixed effect and GMM estimator result respectively.

Table-5: OLS, fixed effect and GMM estimation (value added as a dependent variable):
Without industry dummy
Variables
(1)OLS (2)Fixed effect (3)System GMM
R&D(t)
0.0079**

(0.0050)

0.0013

(0.0058)
0.0041*

(0.0067)
R&D (t-
1
)
-0.0100

(0.0055)
-0.0241***

(0.0057)
-0.0144**

(0.0060)
R&D(t-
2
)
0.0146***

(0.0049)
-0.0032*

(0.0057)
-0.0059**

(0.0060)

Note: legend: * p<0.05; ** p<0.01; *** p<0.001

The parentheses of these rows values show the standard error (SE) of the
coefficient. The star sign indicates the significance level. The GMM estimator is the two
step result of system GMM. The estimation results of the table demonstrate that R&D has
significant impact on productivity level at the 5% level of significance. Hans Lööf (April

24



2008-“The dynamics of firm growth”) referred the Roodman (2006) ‘Rules of thumb’
which a estimate can be good estimates of the true parameter if the first lag of the
dependent variable should lay between OLS and fixed effect coefficient estimation The
appendix table-6 report that the GMM estimation coefficient lay between OLS and fixed
effect coefficient estimation. The GMM estimation result shows that the R&D coefficients
are 0.0041, 0.0144 and 0.0059 in respect to first, second and third years lag respectively
(see table-5).

The first and second order serial correlation tests are reported by the AR (1) and
AR(2) respectively. The first order serial correlation test AR(1) usually rejects the null
hypothesis. The second order test AR(2) is more important because it will detect auto
correlation in levels. The over identification test of the GMM estimation is examined by
the Sargan test which also support the instrument validity of the model. The Sargan test has
a null hypothesis of “the instruments as a group are exogenous”. Therefore, the higher the
p-value of the Sargan statistics the better. The test statistics result do not mislead of the
model.

Table-6: OLS, fixed effect and GMM estimation (value added as a dependent variable):
With industry dummies

Variables (1)OLS (2)Fixed effect (3) System
GMM
R&D(t) 0.0089**

(0.0050)
0.0018

(0.0057)
0.0057**

(0.0069)
R&D (t-
1
) -0.0096

(0.0055)
-0.0245***

(0.0056)
-0.0122**

(0.0059)
R&D(t-
2
) 0.0143**

(0.0050)
-0.0025*

(0.0057)

-0.0044**

(0.0059)
Note: legend: * p<0.05; ** p<0.01; *** p<0.001


To examine the industry dummy variable (n
c
) impact on the estimation result the
study includes it in the model. The table-6 reports the estimation result of equation (v)
where industry dummy is included in the model. A little difference occurs in the R&D
coefficient when industry dummy variable is included in the model. The GMM estimator

25