THE IMPACT OF R&D ON PRODUCTIVITY IN A CROSS SECTION OF COUNTRIES AND SECTORS

THE IMPACT OF R&D ON PRODUCTIVITY IN A CROSS

SECTION OF COUNTRIES AND SECTORS

J.J.R. Kestens

[1]

ABSTRACT

In this paper we investigate the link between R&D and productivity growth. The new- or endogenous growth theory allows for various factors to influence the rate of growth of a specific country. Before, many papers took capital accumulation as the main determinant for economic growth while other factors were determined exogenously. From recent literature we can conclude that there is a link between R&D and productivity growth, as predicted by the endogenous growth theory. We investigated this link and concluded from the results that there are differences between R&D elasticities in sectors, but also between R&D elasticities in countries. This paper relies on a dataset which comprises of company specific data. The dataset consists of 562 companies, which reported their R&D investments. The results strongly suggest that R&D investments can lead to different productivity increases in sectors and countries. The results have implications on the way we think about productivity growth.

[2] CONTENTS INTRODUCTION 3 THEORETICAL BACKGROUND 7 EMPIRICAL MODEL 20 RESULTS 30

DISCUSSION AND CONCLUSIONS 35

APPENDIX 38

[3]

INTRODUCTION

When I was doing an internship at a major investment bank in the summer of 2006, I noticed their reliance on information technology. Most employees have two or more screens in front of them, which are constantly displaying new information. The availability of information and the speed it comes available to these banks is essential for doing business. It is well known that Bloomberg and Reuters, two companies that provide news wires to the financial services industry, are competing on how fast they can deliver new information to their clients. New economic data on unemployment, housing, or inflation usually becomes available to the rest of the world in a split second, and many people rely on these to make investment decisions. Internal programs make the latest research available to other departments. Microsoft Excel is a must for financial professionals to make their complicated analysis and Microsoft PowerPoint is used to make presentations to clients. Most recent university graduates who start working for such a firm will spent their first years working with these programs until they become experts. I will not even start on the use of email in these firms. The rise of the Blackberry, a device that enables you to access email anywhere in the world, makes people connected 24/7. You can answer an important message late at night, already at home, or just before you go to bed, making the speed of decisions and the information flow within a firm higher.

[4]

The introduction of information technology (IT) is the latest major technological change in the world economy. The introduction of this new technology was associated with the rise of completely new industries and the reconfiguration of old industries, meaning; the implementation of this new technology in existent industries. The impact of this new technology has been enormous. During the implementation of this technology in industries many scholars discovered that after years of stagnation productivity began to grow. The link between these two observations was easily established and proven (Jorgenson, Ho, and Stiroh, 2002).

Not only long established industries benefited from this new technology. New firms were established to service these firms technology needs. New products based on information technology were developed. Furthermore, it offered opportunities for emerging economies such as India and China. The heavy investment in IT lowered the communication costs with distant countries significantly, and meant that Western countries suddenly had access to the labor force in countries with lower labor costs. This, for example, meant that many western firms outsourced their call centers to India. In ‘The world is flat’ Thomas Friedman calls this heavy investment in IT as one of the forces that flattened the world, which resulted in many new opportunities for Western firms. Many of these opportunities ultimately resulted in productivity increases, making these firms more profitable.

Banks need a lot of financial data to do their day to day business, which is not instantly available. Since it is too costly to let their highly paid employees search for this information, they outsourced much of this work to India. So when I needed some specific information I just sent an email with a request for specific information and the next morning I received the requested information in an email from an employee located in India. A decade ago this kind of outsourcing was not possible. But now, because of the rise of IT, these kinds of opportunities are existent not only in the financial services industry but in almost every industry.

[5]

think about economic growth. Could they still create sustainable economic growth and compete with these nations? Many politicians argue that we should not focus on producing products that require low paid, unskilled labor. Since this production factor is so abundant in China and India the Western world cannot compete with them on these products, but they can compete on capital and research intensive products. Many have argued that in order to compete with these new economic powers, countries should focus on creating a economic system where much attention is given to innovation (Jacobs & Theeuwes, 2004). Innovation of course, heavily depends on the investment in research and development, the level of education of a country’s citizens, etc.

Research Questions

Until now, many papers have focused on the relationship between research and development (R&D) and productivity itself. They focused on a particular industry or country and consequently tried to estimate the influence of R&D on productivity. Research will be presented indicating that much research effort is being made into the relationship between R&D and productivity itself in a later section. Both the theoretical reasoning and the empirical data supporting this relationship seems very obvious and valid. But the positive relationship between R&D and productivity raises a lot of new questions. It seems only logical to derive new research areas from this relationship, namely how the intensity of these two variables can differ. Some papers already started exploring the relationship of R&D and productivity in datasets by splitting firms into innovating and non-innovating.

[6]

some countries invest substantially more in R&D than others. Therefore some countries (or firms) might be better equipped in exploiting the link between R&D and productivity. Meaning, that they have some characteristics which are very helpful in facilitating effective innovation. Thus, differences between countries can lead to diverse outcomes in different countries and industries when investing in R&D.

This study will investigate if there are observable differences in the way R&D contributes to productivity increases in a set of countries. Furthermore, we will also analyze this relationship on a sector basis. This will increase the value of our results, while we can observe in which industry sectors R&D contributes the highest to productivity increases. If we consequently link this to data which includes country sector weights, we can conclude that countries have different growth potentials. A country with a large portion of a particular industry in which R&D has a big impact on productivity, would (as predicted by new growth theory) have a large growth potential. Many papers already included new variables in order to dispose off firm and industry effects, but in this paper we also propose to account for country effects. Namely; the institutional setting in which a firm or industry operates, can have a effect on the relationship between R&D and productivity.

Structure of the Paper

[7]

THEORETICAL BACKGROUND

The purpose of this chapter is to introduce the theoretical background of this paper. It focuses on the theoretical contributions several authors have made to the research area of economic growth. In order to get a clear overview of the most important contributions we will discuss these in a chronological order. Our goal is to end with the so called “new growth theory” on which we heavily depend in the research part of this paper. This chapter can therefore be seen as a theoretical road to “the new growth theory”.

Economists have always been very interested in translating phenomena observable in the real economic world to a model which describes that phenomenon. Naturally observable economic differences are very interesting to investigate. One of the largest economic differences is the gap between rich and poor countries. Many scholars have asked themselves the question what the origin of these gaps are. Why do certain countries grow faster than others? Can countries have sustainable economic growth in the long run? And if so, what determines this long-run rate of growth? Can, for example, the government influence economic growth?

One of the most important goals of a country’s government is to create economic growth of which its population can profit. The process of creating sustainable economic growth is often a difficult policy issue on which many have an opinion. Of course, academia are not satisfied with unfounded opinions, and many are investigating the concept of economic growth and what its main determinants are.

Capital Accumulation

[8]

level, decreasing returns to scale was not observable. In fact, economies kept on growing and did not stop as predicted by microeconomic models.

Solow solved this problem through adding another variable. He agreed that labor and capital have decreasing return characteristics, but believed that there was a third factor that contributed to the continued economic growth observed in the western world. In his paper, he showed that taking the rate of saving and population growth as exogenous, these two variables determine the steady-state level of income per capita (Mankiw, Romer, Weil, 1992).

In the model, productivity growth, and thus economic growth, stems from increases in the amount of capital per workers. If the amount of capital per workers (or capital labor ratio) increases, the marginal productivity of capital declines, which results in the next unit of capital having less impact on worker productivity. Ultimately, the capital labor ratio reaches a constant where labor and capital grow at the same rate. This long-run rate is determined exogenously and Solow named this the rate of technological progress (Fagerberg, 1994). To put it very simple, capital was the most important variable determining economic growth in Solow’s model, and to be more specific; capital accumulation. In the short run, capital accumulation determines the rate of economic growth, but the Solow model predicts a “steady state” level, where a country’s economy grows at the same rate as the technological progress. Especially this point is important to consider. In the long run, economies will grow at the same rate as the rate of technological progress, which according to Solow is an exogenous variable. According to Solow, economic entities could not influence technological progress

Total Factor Productivity

[9]

even more, resulting in a widening gap between the rich and the poor. Another fact is that capital accumulation is relatively stable over time, but economic growth is not. Factor accumulation and economic growth are not linked as closely as predicted by Solow (1956, 1957). Labor and capital are also concentrating in certain regions of the world, which suggests that factors not included in the Solow model might also be important. A final fact is that national policies influence long-run growth. All these facts where discovered through growth accounting, a process which tries to explain the residual of the Solow model. The main point of these facts is that many variables are not included in the Solow model while having a significant impact on economic growth

Through growth accounting, economists discovered that capital accumulation could roughly explain between one-eight and one-fourth of GDP growth in industrialized countries, whereas TFP growth could explain more than half of GDP growth (Levine & Easterly, 2001). Total Factor Productivity (TFP) is a term which measures the joint effectiveness of all inputs combined in producing output. The TFP concept can include many factors that could have an impact on growth and is rather vague. However, research shows that TFP explains much of the differences of GDP growth rates, so in that sense it is an important concept. It is therefore very important to understand what influences TFP and consequently what factors influence differences between countries.

New Growth Theory

Technological change is an important determinant of TFP (Helpman, 2004). This was also put forward by Solow in his paper (1956, 1957). However, Solow sees technological change as an exogenous variable, something that is available to all and is equal across all countries. After a period where many economists focused on the impact of monetary policy on economic growth, scholars such as Romer (1986) and Lucas (1988) became very interested in growth theory as an explanation for economic inequality. In the 1980s this new stream of research was given much attention by economists.

[10]

growth. Lucas (1988) and Romer (1986) believed that the two central assumptions of the neoclassical model were incorrect. They indicated that technological change requires intentional investment by profit seeking entities or the government, e.g. they believed that technological change is not an exogenous variable but endogenous. Firms can choose to invest in research projects they think will bring benefit to them in the future, and the same applies for the government, but on a more macroeconomic scale. Furthermore, Lucas (1988) and Romer (1986) did not believe in the excludability of technological innovations. Not all technological opportunities are equally available in the world. Even within a country technology may not be available to all, for example, because it is not available to the public e.g. secret or heavily incorporated in a small cluster of firms or research institutions.

[11]

If we analyze these assumptions we can predict several phenomena to occur. First of all, the main determinant of economic growth is the improvement of products . For example, the product itself or the production technique of the product can be improved. This consequently can result in productivity growth in a particular firm, but also in a certain economy if we aggregate all these small increases in productivity on a firm level. In this model we can observe a continuous need for firms to invest in R&D as old products and production methods become obsolete. Firms are in a constant race with each other on the speed of releasing new products. A perfect example is the gaming industry. Producers of game consoles such as Sony, Nintendo and Microsoft are in a endless battle to bring out a better console to the market earlier than its competitors. Every two or three years, each producer introduces a completely redesigned game console in order to satisfy customers and keep them from buying a console from a competitor. It is a never ending battle. However it is not a battle on price, as the neoclassical model predicts, but on quality and product improvements. As Grossman and Helpman (1994) write in their paper: “the innovation process has a distinct Schumpeterian flavor, inasmuch successful innovators displace previous industry leaders and snatch from them a share of industry profits”. The speed of which new products are being introduced and production methods are being improved depends largely on industry characteristics. It is important to consider that the costs and benefits of research and development are an important determinant of long-term growth. As a result, specific industry characteristics can play a big role in the R&D investment decision. For example, if knowledge is easily diffused among competitors, a firm might be unable to extract all rents from its investment and consequently decide not to invest in a specific project. Frequently, the debate starts at this point whether governments should play a role in these cases because the investment might not be beneficial to the individual firm but nevertheless beneficial to the society as a whole.

[12]

The paper of Coe and Helpman (1993) discusses the extent to which a country’s TFP depends not only on R&D capital but also on foreign capital. To investigate this, they use cumulative R&D expenditures as a proxy for the stock of knowledge. This stock of knowledge is a direct input into the endogenous growth model. This research by Coe and Helpman (1993) is very influential because it investigates the degree of which R&D is available to all. This concept is often described as the spillover effect of R&D. It tries to incorporate the essentials put forward by Solow (1956, 1957), namely that knowledge is something available to all and that individual entities have no impact on the stock of knowledge. The paper of Coe and Helpman (1993) tries to incorporate both views of the Solow model and the new growth theory. That is, some part of the knowledge stock is determined endogenously and some part not. In this case the part that is not determined endogenously is the foreign R&D level from which a domestic entity can benefit. Coe and Helpman (1993) conclude that both the degree of domestic R&D and foreign R&D can explain the growth in a country’s TFP. They indicate that foreign R&D has beneficial effects on domestic productivity, and that the benefits for an open economy are even more substantiate.

[13]

Caballero and Jaffe (1993) also investigate knowledge spillovers as a determinant of economic growth, but extend this with the annual rate of technological obsolescence. They conclude that endogenous technological indeed change plays a large role in productivity growth. We will continue to discuss the link between R&D and productivity in a later section. These papers use the insight of the new growth theory in order to investigate how knowledge spills over to other countries and how much this can explain international productivity differences. What these papers do not provide is direct evidence that industrial innovation is the engine of economic growth. Here economic historians come into play, who focus much more on particular events that had a important effect on an economic situation.

[14]

widespread use of steam power in the nineteenth century and America’s innovative activities that eventually lead to what the author calls the “American System”. Furthermore, Rosenberg notices that the relationship between technological change and economic growth is a bit of a “black box” (Rosenberg 1982). This because we intuitively know that there must be a linear link between the two, but what exactly happens inside the black box, e.g. what influences these factors, is rather vague. This is exactly why the combination of the historical accounts and the more “growth accounting” research approach make a strong case for the new growth theory. Alone they show a relationship that seems plausible but which is difficult to prove, but together they are very convincing.

Implications of the New Growth Theory

In this paragraph we will shortly mention the implications of new growth theory on the way we think of economic development. Together with the material presented above, it can help us understand the past and current views on the items which influence economic development. This section also incorporates economic theories developed by other research streams which together with the new growth theory offer new causes and effects on how economies can develop.

[15]

investments leads to a sustainable economic advantage in that particular industry. Furthermore, because the new growth theory implies a more oligpolistic competition in markets, one must also expect a somewhat different environment in which firms operate than the neoclassical model predicts. Romer indicates that the economy is more a evolutionary system, and not the equilibrium seeking economy that the neoclassical model predicts. This statement also fits perfectly in the path dependence nature of the economy put forward by the new growth theory. Firms produce in a certain way, but also try new variations of production methods or products due to competition or other factors that influence the firm directly. Some firms will succeed in introducing this new production method or product and thus will be better adapted to the environment in which its operates. Other firms will not succeed and will ultimately disappear.

A second implication of the new growth theory is that institutions have a significant influence on economic development. Douglas North is one of the pioneers in this field and argued that institutions play a large role in economic growth, but these statements did not fit into the neoclassical model. As North (1987) states: “There are no institutions in the neoclassical world; and indeed in such a world, growth is not a problem, its rate being simply being a function of the number of children people have and the rate of saving”. North indicates that the neoclassical model is too restrictive in describing which factors play a role in creating economic growth and that it fails to explain why certain economies are more successful. The new growth theory is better equipped in explaining why institutions matter than the neoclassical model, because of the role that technological change plays in the model. North (1987) argues that, because the use of modern technology is very capital-intensive, characterized by large-scale production and continuous output, complex institutions are needed to support such a economic structure. In the new growth theory, institutions can have a direct impact on economic growth because they influence technological change directly. As stated before, this technological change is an important variable in the new growth theory. Examples of institutions that have a direct impact on innovation, can be the education system, or the patent and copyright law, etc.

[16]

A third implication of the new growth theory that can be developed from the arguments made by North is that location matters. The institutions of a particular country can be a key success factor in a firm’s survival and ultimately the whole economy of a certain country. Due to path dependence and the subsequent development of institutions, some regions might be more economically developed than others. Romer (1992) also indicates that location is very important in this sense. He states that the most successful areas of development will be the ones most effective in the production of new ideas supported by good public institutions. As written in the introduction of this paper, some parts of firms, such as banks’ IT departments, are being transferred to countries with lower labor costs. But why are firms not moving completely to these countries with lower labor costs? Partly, because there are some benefits to the specific area where the firm is located. Cluster theory, developed by Porter (1990), describes many of the advantages of being located in a specific region. He argues that firms clustering together might benefit from knowledge overflows, a theme also discussed in the endogenous growth models. The main point of Porter (1990) and the endogenous model is that not all knowledge is available to everyone, meaning that firms can earn monopolistic rents on new innovations, as stated before. Because the knowledge developed while creating the new innovation is not available to all, firms can potentially benefit from being close to the origin of the new innovation. The main benefits being; spillovers from the innovator.

[17]

their respective countries, while the neoclassical model predicts a limited impact of government policies towards economic development. The question whether governments are good decisions makers is another issue, but the fact is that their decisions can have an impact is important to consider.

Previous research

Many economists believe that in order to create valuable knowledge for innovation or technological change, firms or governments must invest in R&D. The main goal of performing R&D is to raise productivity in a specific firm, industry, or country, which in turn leads to economic growth on a macro level. The theory introduced above, repeats the reasons why innovation and thus performing R&D is so important. Especially the new growth theory puts much emphasize on R&D, because in contrary to the neoclassical mode it can have a direct impact on the economic development of a country. In this section we will introduce the main economic papers that previously investigated the link the between R&D and productivity. It comprises of research being done in different sectors and countries.

One of the most important scholars in the field of R&D and the link to productivity is Zvi Griliches. His early work focused on productivity increases in the agricultural sector, before turning his attention to the relationship between R&D investment and productivity increases. When looking at research which appeared thereafter, one can divide these into two different approaches. One takes a cross-sectional approach and the other takes a time-series approach to investigate the link. Therefore this literature review will be split accordingly, to make the difference more clear.

Cross-sectional Literature

[18]

[19]

Time-series Literature

Time-series data is collected over discrete intervals of time, according to Hill, Griffiths & Judge (2000). There are distinct differences if we compare this kind of data with the previous section. The biggest differences is that instead of looking at differences in R&D investment and the impact it has on productivity between firms, we are now looking how investment in R&D over a longer period within a specific firm will impact productivity. When reading this kind of literature, one must be aware of these differences because it is not always clearly mentioned in papers. Especially, when considering that much of the authors focusing on cross-sectional data, are also publishing papers using time-series data.

If you look at papers studying time-series data such as Griliches and Mairesse (1984), Cuneo Mairesse (1984), Sannsenou (1988), you can observe that their conclusions do not differ very much from cross-sectional literature. However, as we will discuss later in the empirical methods section of this paper, time-series research suffers from some econometrical problems. In particular these studies suffer from collinearity and simultaneity problems, as noted by Griliches and Mairesse.

Hall and Mairesse (1995) also evaluated the robustness of methods currently used to measure the private returns to R&D. They find that datasets with a longer history of R&D investments help improve the quality of the R&D elasticity estimates.

[20]

EMPIRICAL MODEL

In this section we will discuss the model that we will use to investigate the link between productivity and R&D. The paper will depend on a dataset which comprises of company specific data. The dataset consists of 562 companies, which reported their R&D investments1. Along with the R&D investment data, the dataset includes a number of other observations that we will use in our research. A more detailed discussion of the dataset is provided in another section.

There are two methods of analyzing the impact of R&D on productivity. The first one is the method of case studies, which is often used by business economists. Usually, these case studies discuss a particular innovation that has had a major impact on the economy or a business that has been successful in a product or service innovation. These studies are very helpful in investigating the link between R&D and productivity because they are very detailed and offer an unique insight in the innovation process. A disadvantage of this approach is that it is very time consuming and results in very narrow studies, e.g. they focus on one particular innovation or company. Furthermore, because these studies are very narrow, they are not representative for a whole economy or industry. This is why many scholars frequently use a second, more representative method, in order to investigate the link between R&D and productivity. This second method is an econometric model which estimates the impact of R&D on productivity and can be performed on both a micro- or macroeconomic level. The advantage of the econometric approach is that it does not focus on one particular innovation or firm, but on a whole set of innovations and firms. However, the level of detail is usually less since econometric models focus more on numerical values, which cannot explain as much as case studies. In this paper we will use the econometric model in order to investigate the contribution of R&D to productivity, because we specifically want to investigate the difference in R&D contribution between various industries.

Most econometric studies investigating the relationship between R&D and productivity rely on the Douglas production function as their basic analytical framework. The Cobb-Douglas function of production is widely used to represent the relationship of output on input.

[21]

It was proposed by Knut Wicksell (1851-1926), and tested against statistical evidence by Paul Douglas and Charles Cobb in 1928. The basic form is:

β α

_K

AL

Y

=

Where Y is output, L is labour input, and K is capital input. A, α and β are constants determined by technology. In this paper we extent this basic formula with the investment in R&D. We hypothesize that R&D input has a direct effect on output, thus the R&D input variable will simply be another value in the basic Cobb-Douglas production function.

λ β α

_K

_R

AL

Y

=

In this function R represents the R&D. The R variable in this function will be discussed later in more detail. For now, we estimate a linear equation from the Cobb-Douglas production function. We take the logarithms and obtain the following linear regression equation:

i i i i i a l k r y = +

α

β

λ

ε

= log( ) log( ) log( ) )

log(y_i a l k r (4)

In this equation a, α, β, and λ are the unknown parameters to be estimated. The error term ε, frequently called the Solow residual, captures the total factor productivity. Later on, we will discuss this error term of this equation in more detail.

The equation derived above is used in almost all studies with the goal of estimating the attribution of R&D to productivity. It is used in various forms by Griliches (1980), Schankerman (1981), Grilishes and Mairesse (1984, 1990), Jaffe (1986), Cuneo and Mairesse (1984), Griliches (1986), Sassenou (1988), Halle and Mairesse (1995), Husso (1997) and Bartelsman et al (1996).

[22]

(r) variable. Several papers assume that the R&D variable is a stock of R&D, or all the previous investments made by a firm in R&D. The reason for assuming this is that investments made in the past can still have an impact on current productivity growth.

1 1

(

1

)

+

−

=

R

RS

RS

δ

In the equation above, RS is the R&D stock at the beginning of period t. The depreciation _t rate (δ ) is often assumed to be 0.15. Now we have to generate the equation for the first year when the firm invested in R&D,

δ

+

=

g

R

RS

Where g is the long term growth rate of R&D. Bond et al (2002) argue that since, in steady state, R_t =(g+δ)RS_t−1 and also RS_t =(1+g)RS_t−1, we can write:

t t t t t

_RS

g

g

R

g

R

RS

g

RS

+

+

=

⇒

+

=

=

+

1

)

1

(

δ

δ

[23]

expenditures. Because estimating the appropriate depreciation rate while using the production function is especially difficult. For these reasons we will focus on the level of R&D expenditures of the previous year in this research.

Limitations of the model

This section will discuss the various problems that could arise when conducting our research and which we should take into account. These problems can refer to the variables of the model we specified in the previous section or to general econometric problems, which refer to the model as a whole.

Problems in measuring output:

One of the problems when measuring output is that in some industries output is difficult to quantify. Then it becomes even more difficult to attribute R&D to productivity increases. Especially in industries with a large proportion of public research and development, such as defence and health where output is measured based on inputs. For example, in the health sector output is measured by the number of treatments. It would be very difficult to incorporate these output numbers into our model, especially because most firms report their output as total sales. However in this paper we do not have to deal with this problem since all firms in our database report their output in a value of total sales or to be more precise, in a currency value of sales.

[24]

short of what could have been realized if they could discriminate perfectly. For example, some firms can pass these benefits directly on to the consumer. In this way the actual productivity increase will not appear in the statistical data. One can easily observe the impact on the whole dataset, because firms can choose who will benefit from the productivity increase. How the firm will choose depends largely on the market structure in which he operates and other considerations. The same applies for cost reductions due to new production methods and other innovations. Summarizing the points above, it can be stated that in some industries R&D leads to quality improvements or cost reductions, but these are not always reflected in productivity improvements

Problems in measuring research and development capital:

As discussed previously, measuring research and development capital, or the R variable in equation (4), is very difficult. Although we already explained how we will measure this variable, we should take into account all the problems that could arise when measuring R&D capital.

Measuring the exact capital stock of research and developments is a challenging task since it captures concepts which are not easily identified. The contribution of general science to a particular industry is already very difficult to observe because some aspects of science cannot be measured in numerical values. Case studies can partially solve this problem, but in the case of econometric studies we have to focus on identifiable variables. In our research these identifiable variables are the investments firms make in R&D to advance their state of knowledge. Moving forward, the measurement of these investments in R&D has some problems. The paper of Grilliches (1979) discusses three of these problems.

[25]

and new product line coexist next to each other. One can think of the aircraft industry as an example. In recent years the aircraft industry is using more and more composite materials in their airplanes. A great innovation because it is stronger and lighter than aluminium. But although this innovation is ready for implementation for all aircraft models, airplane makers only use this new material in completely new aircraft models while they still continue to build other models using the old material. Before composite materials will be incorporated into every new aircraft a decade or more will be over. Griliches (1979) argues that this leads to a flat but somewhat bell-shaped lag structure connecting firm research and development to its subsequent productivity growth.

The second problem mentioned in the paper of Griliches (1979) is the issue of depreciation of the R&D capital stock. Depreciation is the decrease in the economic value of the capital stock of a firm. In this case, the capital stock is the R&D capital stock. It is very obvious that an investment in R&D twenty years ago is not as valuable as an investment in R&D made today. Over time, new products and processes become available elsewhere and competitors catch up with R&D, so that the value of the R&D investment decreases. The industry around a particular firm does not sit still and R&D spillovers make that R&D investment can depreciate fairly quick. However, if we take a step back and look to this phenomenon from a macro perspective, depreciation is not a major issue. The knowledge stock of a firm can indeed lose its value because other firms obtain more knowledge. However this is a private loss for the specific firm, but it does not have to be a social loss. This because the value of the R&D investment may be lost for that particular firm but the society as a whole can still benefit from it. One should remember that the social benefits of R&D also depreciate over time.

[26]

we could be able to construct an equation which can measure these spillovers effects. In fact, several author have tried to estimate the influence of these spillovers effects (Dilling-Hansen, Eriksson, Madsen and Smith. 1999). However, it is not the objective of this paper to estimate this relationship, but we should know that this phenomenon occurs and will influence our results somewhat. As argued by Griliches, it is difficult to incorporate these effects into the model due to the lack of theory and complexity of these spillover effects.

Econometric problems:

This section will discuss somewhat more general problems associated with doing economic analysis. The three problems discussed above are specific to our research and cannot be generalized. The two econometric problems we have to deal with in our research are multicollinearity and simultaneity.

[27]

proportions. This could very well be the case in certain industries we will examine. Therefore, when analysing our results we should pay particular attention to this problem. As argued by Griliches (1979), focussing on micro data is probably the best way to dispose of multicollinearity, because there is more variability in the R&D histories of individual firms than industry or economy-wide data. Consequently, we will use micro-data in our research in order to construct industries, which we can compare to each other. Another way to solve the collinearity problem is to include better data in the sample. However as simple this solution may sound, most researchers know that obtaining extra data is very expensive and time consuming.

[28]

Data

Most of the data we use in our research we extracted from the 2005 R&D Scoreboard of the UK Department of Trade and Industry (DTI) , which contains extensive data on the top 1000 Global R&D companies2. The database itself contains details of R&D investment, capital expenditure, sales, profits, employee numbers, market capitalisation. Furthermore, it includes changes in these variables from previous years and key ratios for firms. All data is extracted directly from the companies’ annual report and key ratios are calculated for each company and sector. Companies are classified by FTSE sector and by country. The R&D investments included in the Scoreboard is the cash investment which is funded by the companies themselves. It excludes R&D undertaken under contract for customers such as governments or other companies. It also excludes the companies’ share of any associated company or joint venture R&D investment. All of R&D costs have been capitalised. In order to calculate the cash investment, the additions to the appropriate intangible assets are included in the Scoreboard as R&D investments, and any amortisation is eliminated (explanatory notes 2, DTI

website). We used total sales of the database as an output measure. Furthermore, we used the number of employees as a measure of labour input and for the R&D investments value as our R&D input. In the notes of the DTI database some extra information is given on the R&D investment values. They indicate that, for many diversified firms, the R&D investment disclosed in their accounts arises from only part of their activities, whereas sales, operating profit and market capitalisation are in respect of all their activities. Unless all such firms disclose their R&D investment with the other information in segmental analyses, it is not possible to relate the R&D more closely to the results of the individual activities which give rise to it. The impact of this is that some statistics for these companies, e.g. R&D as a percentage of sales, are possibly understated (explanatory notes 2, DTI website). Firms in

some countries are less likely than others to disclose R&D investment, or to disclose it consistently. As a result, the global Scoreboard cannot set out to systematically capture all firms with R&D activity. There is evidence to suggest that the distribution of R&D activity is highly skewed towards larger firms. The dataset captures the most significant R&D investing firms, and in any case, the minimum R&D needed for inclusion in the global Scoreboard is over £22m.

[29]

The last variable left in our research equation is the capital input measure. As argued in several papers, this measure indicates the physical capital of the firm. In other words, the tangible assets which are used in the firm’s production. This measure was not included in the original database. We decided to use the Property, Plant, and Equipment value stated in each firm’s annual report on the balance sheet, because this value is most representative to the physical capital of the firm. Furthermore, this value is available for each publicly traded firm which makes it comparable across a set of firms. All publicly listed firms have to file an annual report which makes it relatively easy to search for. We used different methods to find the Property, Plant, and Equipment measure in the company’s annual report. Firstly, for all European firms we used the AMADEUS database, which includes financial data of most European firms. Secondly, for firms located in the United States we used the EdgarScan database of PricewaterhouseCoopers. This database includes all SEC (Security and Exchange Commission) filings of firms listed on an US stock exchange. EdgarScan is an interface to the United States Securities and Exchange Commission Electronic Data Gathering, Analysis and Retrieval (SEC EDGAR) Filings. EdgarScan pulls filings from the SEC's servers and parses them automatically to find key financial tables and normalized financial data to a common format that is comparable across companies. The EdgarScan database subsequently makes the financial data more accessible in the same way that the AMADEUS database does. The third way is the most time consuming method for looking up the necessary data. We used this method for firms not located in Europe or the United States, e.g. most Asian countries. It involved going to the website of the respective firm, downloading the annual report, and locating the Property, Plant, and Equipment value on the balance sheet. In total, we looked up the capital input measure for 562 firms using these different measures.

All foreign currency amounts have been translated at the UK sterling exchange rates at 31 December 20043. For the data that we looked up in order to include them in the database, we also used the 31 December 2004 exchange rate. We found these historical prices using a Bloomberg Terminal. Because the DTI database uses exchange rates instead of purchasing power parity when referring to currency values, we used exchanges rates.

[30]

RESULTS

Using the empirical model described above, we regressed our data in two ways. In the first analysis we divided the whole dataset up according to each firm’s location. We used Eviews to perform the actual regressions. The results of the regressions are summarized in table 1. If we look at the coefficient covariance matrix given in the appendix, it is important to notice that our variables are not correlated among each other. So the problem of multicollinearity we explained in the econometric problems section is not observable in our dataset. Adjusted R-Squared is also very high in our result, meaning that the model and the found coefficients can explain a large amount of the data.

Table 1. Country analysis (Coefficients and probabilities)

Intercept Capital Labour R&D Adj. R-Squared

All Countries (N=562) -1.7893 0.2518 0.7947 0.0611 0.8927 (0.0000) (0.0000) (0.0000) (0.0221) USA (N=252) -2.5539 0.1571 0.924 0.0844 0.8814 (0.0000) (0.0021) (0.0000) (0.0950 Japan (N=112) 0.1368 0.4059 0.4816 0.1084 0.8687 (0.7049) (0.0000) (0.0000) (0.0291) Europe (N=179) -1.2782 0.2675 0.7207 0.0767 0.9666 (0.0000) (0.0000) (0.0000) (0.0173) France (N=27) -0.8799 0.4004 0.5744 0.1093 0.9403 (0.2218) (0.0000) (0.0001) (0.0984) Germany (N=42) -1.0241 0.2249 0.6854 0.1391 0.9623 (0.0401) (0.0014) (0.0000) (0.0009 Sweden (N=12) -2.6786 0.0919 0.8994 0.2084 0.9793 (0.0014) (0.2587) (0.0000) (0.0261) Finland (N=9) -2.0612 0.0322 0.9260 0.2073 0.8784 (0.2884) (0.7824) (0.0113) (0.1705) Netherlands (N=11) 1.1877 0.3275 0.3937 0.1412 0.8650 (0.5595) (0.1367) (0.2296) (0.4309) UK (N=26) -1.1652 0.3598 0.6708 0.0388 0.8959 (0.1894) (0.0069) (0.0003) (0.7808) Switzerland (N=18) -1.2856 0.0624 0.7699 0.2422 0.9780 (0.0071) (0.5196) (0.0000) (0.0050)

[31]

insignificant coefficients), makes the argument even stronger. However from this country analysis we cannot conclude that the relationship between R&D and productivity is not a direct relationship and that a firm’s location influences the impact R&D has on productivity. Although we can observe differences between countries R&D coefficient, this may be the result of the kind of firms we have included in our dataset.

The second analysis we conduct on our dataset is a sector analysis. Industry variables were already included in our dataset, so we only had to divide the data into the correct industry sub-datasets. After this, we used Eviews for the regressions, of which the results are published in table 2. Again, most industry regressions are significant. However, Food Producers, Forestry & Paper, and Pharma & Biotech are not significant for the R&D variable. For Food Producers and Forestry & Paper the low number of observations is the obvious explanation. Still, this explanation does not hold for the Pharma & Biotech industry since the number of observations is relatively large (N=116).

[32]

Table 2. Sector Analysis

Intercept Capital Labour R&D Adj. R-Squared

All Sectors (N=562) -1.7893 0.2518 0.7947 0.0611 0.8927

(0.0000) (0.0000) (0.0000) (0.0221)

Automobiles & parts (N=39) 0.7017 0.2739 0.388 0.3535 0.9385 (0.3087) (0.0061) (0.0003 (0.0010)

Chemicals (N=41) 0.3133 0.5931 0.2944 0.1468 0.9127

(0.5636) (0.0000) (0.0058) (0.0672)

Construction & Building (N=16) 1.4642 0.5537 0.2009 0.2107 0.7444 (0.2859) (0.0051) (0.3460) (0.2107)

Diversified Industrials (N=19) -0.3322 0.3638 0.6183 -0.0459 0.9574 (0.6652) (0.0012) (0.0003) (0.0515)

Electronic & Electrical (N=38) 1.0964 0.1485 0.3768 0.4364 0.9379 (0.0358) (0.0179) (0.0000) (0.0000)

Engineering & Machinery (N=50) -0.8405 0.1819 0.6724 0.2249 0.8773 (0.1524) (0.0039) (0.0000) (0.0073)

Food Producers (N=15) -0.3017 0.7323 0.3669 -0.042 0.9347

(0.7608) (0.0029) (0.0229) (0.8131)

Forestry & Paper (N=9) 4.3296 0.1866 0.3713 -0.2272 0.5676 (0.0673) (0.6914) (0.1796) (0.5602)

Health (N=40) -0.6936 -0.0597 0.7347 0.3194 0.9216

(0.1425) (0.1795) (0.0000) (0.0000)

IT Hardware (N=81) -0.8927 0.0002 0.6969 0.3647 0.8962

(0.0154) (0.9967) (0.0000) (0.0000)

Oil & Gas (N=26) 2.7259 0.7298 -0.1034 0.3617 0.9122

(0.0212) (0.0000) (0.5291) (0.0713)

Personal Care (N=13) 0.5013 0.0852 0.5966 0.2724 0.9448

(0.4766) (0.4003) (0.0002) (0.0085)

Pharma & Biotech (N=116) -4.1694 0.0664 1.3685 -0.0961 0.8478 (0.0000) (0.5737) (0.0000) (0.4265)

Software & Computer Services (N=36) -0.4325 0.1448 0.4528 0.5214 0.9423 (0.3656) (0.0551) (0.0000) (0.0000)

In table 3 the results are published of the regression with large and long established Pharma & Biotech firms.

Table 3. Adjusted Pharma & Biotech Sector

Intercept Capital Labour R&D Adj. R-squared

Pharma & Biotech (N=36) 0.015378 -0.03274 0.560142 0.466309 0.957756

[33]

Computer services sector. To put it very simply, one euro of R&D investment in one industry can be more productive than one euro in another industry.

Until now, we have seen that the impact of R&D investment can be different in countries or industries. The analysis we have done above is very crude and the conclusion which can we drawn on it is very limited. The only thing we can say is that we can observe differences in the impact of R&D on industries and countries. However, if we could analyse one particular industry in different countries we could make our case stronger. Namely that the impact of R&D on productivity is not a simple one to one relationship. As always this research is limited by the dataset on which it draws. We simply do not have enough observations to research one particular industry in different countries.

To solve our dataset problem we assume that there are only two kinds of industries in the world economy, namely: innovative and non-innovative. Wakelin (2001) also used this classification method in her research which investigates productivity growth and R&D expenditure in UK manufacturing firms. However the way we define innovative will be different in our research. Wakelin (2001) has actual data on which firms produced innovations and which not, but we do not have such data. Therefore, we take a somewhat different approach. We assume that the firms who have the highest R&D coefficient are the most innovative. The reasoning behind this is logical. A firm or sector has exhibits a strong link between R&D and output must produce innovations, otherwise output would not have risen. Firms where this link is not as strong are not as innovative since R&D expenditure does not lead to increased output. With this in mind we rank our sectors according R&D intensity (see table 5 in the appendix) and consequently split the dataset in two. One with high R&D elasticity sectors and one with low R&D elasticity sectors.

We used the same regression method as before and now have two industries in each country. All results are significant, except France. Before performing the actual regression we already excluded some countries due to the lack of data. However, these countries are included in the European regression.

[34]

comparable sectors in the U.S. and Europe, while the low elasticity sector is relatively less elastic.

Table 4. Country high/low R&D elasticity sector analysis (Coefficients and probabilities)

Intercept Capital Labour R&D Adj. R-Squared

U.S. high -0.7491 0.2755 0.6227 0.1638 0.8984 (-0.0157) (0.0000) (0.0000) (0.0001) U.S. low 0.8297 0.3800 0.4445 0.0946 0.9327 (0.1380) (0.0000) (0.0000) (0.0478) Japan high 0.7655 0.3843 0.3509 0.2638 0.8866 (0.1420) (0.0000) (0.0004) (0.0021) Japan low 0.0348 0.4314 0.5067 0.0326 0.7302 (0.9679) (0.0001) (0.0003) (0.0148) Europe high -0.3373 0.3884 0.5727 0.1466 0.8890 (0.6027) (0.0000) (0.0000) (0.0450) Europe low -0.8657 0.2653 0.6689 0.1020 0.9347 (0.0275) (0.0000) (0.0000) (0.0104) France high -0.2559 0.4643 0.4816 0.1068 0.9333 (0.8249) (0.0006) (0.0257) (0.3438) France low -1.9489 0.3060 0.7208 0.1350 0.9789 (0.1025) (0.0984) (0.0107) (0.4163) Germany high -0.9365 0.2314 0.6322 0.2017 0.9644 (0.1649) (0.0212) (0.0000) (0.0114) Germany low -1.8049 0.1413 0.8161 0.1571 0.9525 (0.0511) (0.1920) (0.0000) (0.0271)

[35]

DISCUSSION AND CONLUSIONS

This section will discuss the results of our research and will try to interpreted them with the use of literature we introduced earlier. The study investigates if there are observable differences in the way R&D contributes to productivity increases in a set of countries and on a sector basis.

We have seen from our country analysis regression results that countries differ in the degree R&D investments lead to productivity growth. However, what we need to establish is if these results directly lead to the conclusion that these differences come from other factors that influence this relationship. Meaning that these differences are not simple variances between countries. We tried to accomplish this by including a sector analysis and a two industry analysis in our results.

If we look at the sector analysis we see that industries differ substantially in R&D elasticities. In some industries, such as forestry and paper, conducting more R&D has even a negative effect on productivity. If we subsequently rank the results from high elasticity to low elasticity we see a picture that is very plausible. In the top we see industries already known for technological change. The regressions confirm that the research performed by these firms indeed contribute a lot to their productivity growth. The results of this sector analysis can be used in conjunction with the country analysis. The difference of R&D elasticities in sectors will show up in country data. The bigger part a specific sector plays in a country’s GDP, the bigger influence it will have on the elasticity of R&D in a country.

[36]

The new growth theory introduced in earlier chapters only supports our hypothesis that the relationship between R&D and productivity is not a direct one. The new growth theory predicts and allows for country specific differences, such as institutional, location, or historical differences. An explanation for the observed variations in R&D elasticity could be linked to the factors introduced in the theory part. However, our dataset is too limited to analyse and conclude which specific factors influence this relationship and how much. We can only state that our results point in the direction of country and sector differences.

This brings us at the limitations of our results and the conclusions we derive from them. As with all research we are limited with the dataset we have. A larger dataset with more variables would lead to more robust findings, but unfortunally we have to do with the data we have. The results lead to the conclusion that we indeed can observe differences in the relationship between R&D and productivity and that “other variables” influence this relationship. This paper therefore can be seen as an exploratory study into this relationship. We now know that there are observable differences and a next step would be to investigate what determines these differences. A more detailed investigation could lead to interesting implications. For example, if we continue on the path of doing research on the differences among sectors and we indeed find that in some industries small R&D investments can lead to substantional productivity increases, a country could adapt its innovation policy. A country could encourage specific firms in conducting R&D.

The problems encountered during this research and its results make us better equipped in determining what kind of data we need to explore this research area further and what of problems we might encounter. Potentially, a next step in this research area would be to focus on one particular industry over a range of countries, with the inclusion of country variables that show the characteristics of the area in which the firms operates.

[37]

[38]

APPENDIX

Table 4. Coefficient Covariance Matrix (full dataset)

A B C M

A 0.032101 0.002523 -0.004754 -0.000442

B 0.002523 0.000755 -0.000739 -0.000052

C -0.004754 -0.000739 0.001136 -0.000285 M -0.000442 -0.000052 -0.000285 0.000710

Table 5. Sector R&D Elasticity Ranking

Ranking Industry R&D elasticity 1 Software & Computer Services 0.5214

2 Pharma & Biotech 0.4663 3 Electric & Electrical 0.4364

4 IT Hardware 0.3647

5 Oil & Gas 0.3617

6 Automobiles & parts 0.3535

7 Health 0.3194

8 Personal Care 0.2724

9 Engineering & Machinery 0.2249 10 Construction & Building 0.2107

11 Chemicals 0.1468

12 Food Producers -0.0420

13 Diversified Industrials -0.0459 14 Forestry & Paper -0.2272

[39]

REFERENCES

Barro, R.J, 1991. Economic Growth in a cross section of countries. The Quarterly Journal

of Economics, 106: 407-443.

Bartelsman, E., van Leeuwen G., Nieuwenhuijsen H. and Zeelenberg K. 1996. R&D and

Productivity Growth: Evidence from Firm-Level Data for the Netherlands.

Statistics Netherlands, Department of Statistical Methods. Conference paper.

Bond, S., Harhoff, D., & van Reenen, J. 2002. Corporate R&D and Productivity in Germany and the UK. CEP Discussion Papers dp0599. Centre for Economic Performance, London School of Economics.

Caballero, R.J., & Jaffe, A.B. 1993. How high are the giants’ shoulders: an empirical

assessment of knowledge spillovers and creative destruction in a model of economic growth. NBER working papers, no. 4370.

Cobb, C. W., & Douglas, P. H., 1928. A theory of production. American Economic Review, 18: 139-165.

Coe, D.T., & Helpman E. 1995. International R&D spillovers. European Economic Review, 39: 859-887.

Cortright, J. 2001. New growth theory, technology and learning: a practitioners guide.

Reviews of Economic Development Literature and Practice, no. 4.

Cuneo, P. and Mairesse J. 1984. Productivity and R&D at the firm level in French manufacturing, in Z. Griliches (ed.), R&D, Patents and Productivity, Chicago. University of Chicago Press, 375-392.

David, P. 2000. At last, a remedy for chronic QWERTY-scepticism! Paper prepared for the European Summer School in Industrial Dynamics.

Dilling-Hansen, M., & Eriksson, T., & Madsen, E.S., & Smith, V. 1999. The impact of R&D

on productivity: evidence from Danish manufacturing firms. Danish Institute for

studies in research and research policy. Working paper.

Easterly, W., & Levine, R. 2001. It’s not factor accumulation: stylized facts and growth Models. The World Bank Economic Review, 15: 177-219.

Eaton, J., & Kortum, S. 1999. International technology diffusion: theory and measurement. International Economic Review, 40: 537-569.

Fagerberg, J. 1994. Technology and international differences in growth rates. Journal of

Economic Literature, 32: 1147-1175.

Fogel, R.W. 1993. Railroads and American Economic Growth : Essays in Econometric

[40]

Friedman, T.L. 2005. The World is Flat. New York: Farrar, Straus and Giroux. Griliches, Z. 1964. Research expenditures, education and the aggregate agricultural Production function. American Economic Review, 6: 961-974.

Griliches, Z. 1979. Issues in assessing the contribution of research and development to Productivity growth. The Bell Journal of Economics, 10: 92-116.

Griliches, Z. 1980. Returns to reserach and development expenditure in the private sector, in J.W. Kendrick and B. Vaccara (eds), New Developments in Productivity

Measurement, vol. 44 of Studies in Income and Wealth, Chicago, Chicago University

Press, 419-54.

Griliches, Z. and Mairesse J. 1984. Productivity and R&D at the firm level, in Zvi Griliches

R&D, Patents and Productivity. Chicago, University of Chicago Press, 339-374.

Griliches, Z. 1986. Productivity, R&D, and basic research at the firm level in the 1970’s.

The American Economic Review, 76 (1): 141-154.

Griliches, Z. 1988. Productivity Puzzles and R&D: Another nonexplanation. The

Journal of Economic Perspectives, 2: 9-21.

Griliches, Z. and Mairesse J. 1990. R&D and productivity growth: Comparing Japanese and US manufacturing firms. NBER working paper no. 1778.

Grossman, G.M., & Helpman, E. 1994. Endogenous innovation in the theory of growth.

Journal of Economic Perspectives, 8: 23-44.

Hall, B.H., & Mairesse, J. 1995. Exploring the relationship between R&D and

productivity in French manufacturing firms. Journal of Econometrics, 65: 263-293. Hill, R.C., Griffiths, W.E., Judge, G.G. 2000. Undergraduate Econometrics (2nd _{ed.). New}

York: John Wiley & Sons Inc.

Husso, K. 1997. The Impact of R&D on Productivity: Evidence from Firm-Level Panel Data for Finland, in Seppo Laaksonen (ed.), The Evolution of Firms and Industries,

Research Reports, 223: 311-339.

Jacobs, B., & Theeuwes, J.J.M. 2004. Innovatie in Nederland: De markt draalt en de

overheid faalt. Koninklijke Vereniging voor de staatshuishoudkunde, Preadviezen

2004.

Jaffe, A. 1986. Technological opportunity and spillovers of R&D: evidence from firms’ patents, profits and market value. American Economic Review, 75(6): 984-1002. Jones, C.I. 1995. R&D-Based models of Economic growth. The Journal of Political

[41]

Jorgenson, D. W., & Ho, M., & Stiroh K.J. 2002. Projecting Productivity Growth: Lessons from the U.S. Growth Resurgence. Federal Reserve Bank of Atlanta Economic

Review, 87: 1-13.

Koopmans, T. 1945. Statistical estimation of simultaneous economic relations. Journal of the

American Statistical Association, 40: 448-446.

Landes, D.S. 1969. The unbound prometheus: Technological change and industrial

development in western Europe from 1750 to the present. Cambridge University

Press.

Lichtenberg, F.R. 1992. R&D investment and international productivity differences. NBER Working papers, no. 4161.

Ljungberg, J., Smits, J.P. 2004. Technology And Human Capital In Historical Perspective. Palgrave Macmillan.

Lucas, R.E. 1988. On the mechanics of economic development. Journal of Monetary

Economics, 22: 3-42.

Mairesse, J., & Sassenou, M. 1991. R&D and productivity: a survey of econometric

Studies at the firm level. NBER working paper series, no 3666.

Mankiw, N.G., & Romer, D., & Weil N.W. 1992. A contribution to the empirics of economic Growth. The Quarterly Journal of Economics, 107: 407-437.

Mansfield, E. 1965. Rates of return from industrial research and development. American

Economic Review, 55: 310-322.

Martin, R., & Sunley, P. 1998. Slow Convergence? The new endogenous growth theory and regional development. Economic Geography, 74: 201-227.

Minisian, J. 1969. Research and development, production functions and rates of return.

American Economic Review, 59: 80-85.

Mokyr, J. 2002. The Gifts of Athena: Historical Origins of the Knowledge Economy Princeton University Press.

Nickell, S., Wadhwani, S., Wall, M. 1992. Productivity growth in U.K. companies, 1975- 1986. European Economic Review, 36: 1055-1091.

North, D. 1987. Institutions, transaction costs and economic growth. Economic Inquiry, 25: 419-428.

Odagiri, H., & Iwata, H. 1986. The impact of R&D on productivity increase in Japanese manufacturing companies. Research Policy, 15: 13-19.

[42]

Rogers, M. 2006. Estimating the impact of R&D on productivity using the BERDARD

data. Oxford Intellectual Property Research Centre, working paper.

Romer, P.M, 1986. Increasing Returns and Long-run Growth. Journal of Political Economy, 94(5): 1002-1037.

Romer, P.M. 1990. Endogenous technological change. The Journal of Political Economy, 98: 71-102.

Romer, P.M. 1994. The origins of endogenous growth. The Journal of Economic

Perspectives, 8: 3-22

Rosenberg, N. 1972. Technology and American Economic Growth. New York: Harper and Row Publishers.

Rosenberg, N. 1982. Inside the Black Box. Cambridge. Cambridge University Press. Sassenou, M. 1988. Recherche-développement et productivité dans les entreprises

japonaises: une étude econométrique sur données de panel. Doctoral dissertation,

Ecole des Hautes Etudes en Sciences Sociales, Paris.

Schankerman, M. 1981, The Effects of Double-Counting and Expensing on the Measured Returns to R&D. Review of Economics and Statistics, 63: 454-458.

Shaw, G.K. 1992. Policy Implications of Endogenous Growth Theory. The Economic

Journal, 102: 611-621.

Solow, R.M. 1956. A Contribution to the Theory of Economic Growth. Quarterly Journal of

Economics, 70: 65-94.

Solow, R.M. 1957. Technical Change and the Aggregate Production Function. Review of

Economics and Statistics, 39: 312-20.

Wakelin, K. 2001. Productivity Growth and R&D Expenditure in UK Manufacturing

firms. Economics Department, University of Nottingham. Paper presented at the 25th