R&D Expenditure and Economic Performance: A German Panel Analysis

R&D Expenditure and Economic

Performance: A German Panel

Analysis

MASTER’s THESIS WITHIN: Economics NUMBER OF CREDITS: 30

PROGRAMME OF STUDY: Economic Analysis AUTHOR: Siddhi Shailen Salvekar

Master Thesis in Economic Analysis

Title: R&D Expenditure and Economic Performance: A Regional German Panel Analysis

Authors: Siddhi Shailen Salvekar Tutor: Charlotta Mellander Date: 2020-05-18

Key terms: Innovation, R&D Expenditure, Regional GDP, Business R&D expenditure, Public R&D expenditure

Abstract

There has been a long line of studies concerning the nexus between R&D expenditure and Economic Performance. However, there has been little research on this nexus on a regional level in Germany. This paper aims to analyse the relationship between R&D expenditure (total, business sector and public sector each) and GDP per capita at a regional level in Germany for the time period 2000-2017. The method of estimation employed in fixed effects panel regression analysis. It is found that total R&D expenditure and regional GDP per capita have an insiginificant relationship, but a negtiavely significant one when a lagged value of R&D expenditure is considered. The relationship between public R&D expenditure and regional GDP per capita is significantly negative, whereas the relationship between business sector R&D expenditure and regional GDP per capita is insignificant. Further scope for research could include analysing the effect of regional innovation clusters and the role of R&D strategy in improving the economic performance of regions in Germany.

Table of Contents

1 Introduction ... 1

2 Background ... 3

3 Literature Review ... 7

4 Data Description and Methodology ... 15

4.1 Data Description ... 15

4.2 Fixed Effects Models for Panel Data ... 17

4.3 Descriptive Statistics ... 20

5 Results and Discussion ... 22

5.1 Fixed Effects Estimation Results ... 22

5.2 Fixed Effects Estimation with corrected standard errors ... 26

6 Conclusion ... 30

References ... 32

Figures

Figure 1 Map of Total Internal R&D Expenditure per region in Germany (2017) ... 4

Figure 2 Business to Public R&D Ratio per province in Germany ... 5

Tables

Table 2 Descriptive Statistics ... 20

Table 3 Correlation Matrix ... 21

Table 4 Regression Results – FEM Estimation ... 23

Table 5 Regression Results – Driscoll Kraay Corrected Standard Errors Estimation ... 27

Appendix

1. Introduction

_____________________________________________________________________________________

The purpose of this section is to introduce the reader to the aim and objectives of this paper, research question and the background regarding the context of the research objectives.

______________________________________________________________________ The importance of industrial innovation, or the creative contribution of technological advancement was first introduced by Schumpeter (1942). The idea that technological advancement i.e. increasing stocks of knowledge have a positive impact on economic growth, was later delved deeper into by Romer (1986) and Lucas (1988), and thus was coined as the New Growth Theory. Since then, the relationship between knowledge creation and economic growth has been analysed to a great extent by social scientists all over the world. Technological advancement is treated as a source of economic growth in the endogenous growth theory, and it is imperative to identify and analyse the factors affecting technological advancements in countries and regions within countries alike (Grossman & Helpman, 1994). Research and development activities carried out by private firms, governmental institutions and higher educational institutions contribute to the stock of knowledge accumulated in a country/region (Guellec & van Pottelberghe, 2004).

Increased accumulation of the knowledge borne by R&D activities, will eventually contribute to an increase in the productivity and economic growth in that country/region, by producing products of higher quality (Lichtenberg, 1992). In this paper, I aim to address this concept by testing the relationship between total R&D expenditure and the gross domestic product (GDP) per capita in the 16 provinces (Bundeslander) in Germany for the time period of 2000-2017, by adopting a multivariate panel regression approach. The analysis further examines the relationship between business sector R&D expenditure and public sector R&D expenditure, with the per capita GDP at a regional level.

Innovation and technological changes have been a driver for economic growth across regions, and increasing the standard of living in regions (Grossman & Helpman, 1994; Bilbao-Osario & Rodriguez-Pose, 2004). Analysing the relationship between levels of innovation in a region and economic growth can affect policymakers’ decisions regarding the commitment towards the R&D effort through subsidies and planned investments. Spillovers from R&D investments made by firms and governments alike can contribute to sustained economic growth in the long-term (Brautzsch et al., 2015).

The German government has focused on maintaining considerable amounts of expenditure regarding R&D activities, along with motivating firms and small and medium enterprises to increase innovativeness through internal R&D expenditures (Almus & Czarnitzki, 2012). It is crucial to observe the positive externalities that this R&D expenditure generates on the overall economic growth of regions. Germany was a pioneer in introducing high levels of R&D funding in the time period of the 1980s, and has continued to do so till date (Czarnitzki & Fier, 2003). This raises an important question, that sums up the purpose of this paper: does internal R&D expenditure within a region affect the overall regional economic growth in a country?

The concept regarding regional disparities in growth and the lack of convergence thereafter, in EU regions is rather interesting to study. The idea that knowledge is ‘sticky’, and concentrated in one region, which then causes certain regions to perform better than others in terms of productivity and innovation is imperative to focus on (Bilbao-Osario & Rodriguez-Pose, 2004). Investment in R&D has been known to affect the standard of products and services manufactured in a region, which in turn leads to higher levels of income and growth. The New Growth Theory put forth by Romer (1990), certainly points to the fact that there exists a positive correlation between investment in better technology i.e. technological advancement and increased growth rates. Lucas (1988) has also stated that countries (regions in the case of this analysis) having higher levels of initial stock of knowledge will fare better in terms of economic growth than countries/regions who do not. However, certain institutional settings will also affect this postulate, such as focused policy decisions regarding innovation and productivity within a region (Andersson & Karlsson, 2007). The European Commission Data (2002) showcases that publicly funded R&D activities have a major share in explaining growth in countries such as Spain, Portugal, Finland and Ireland (Bilbao-Osario & Rodriguez-Pose, 2004). Publicly funded research and policy in these countries play a crucial role in speeding up growth.

This analysis aims to observe the following questions: Does total R&D expenditure at a regional level affect regional economic performance? Do public and private sector R&D expenditures affect regional economic performance and to what extent? Section 2 of this paper will provide an extensive literature review of the area undertaken for this research. Section 3 will describe the data and variables used as well as the methodology employed in the paper. Section 4 will deliver a summary of the analysis and results of the analysis. Section 5 will conclude the paper with a discussion of the entire analysis as well as some

relationship between total R&D expenditure and regional GDP per capita is insignificant, but becomes negatively significant when lagged value to R&D is considered. The relationship between public R&D expenditure and regional GDP per capita is significantly negative and the relationship between business R&D expenditure and regional GDP per capita is insignificant.

2. Background

Science and Research in technological advancements is the keystone of innovative capabilities in any country. Germany ranks 4th _{in terms of R&D spending in the world, with}

the expenditure being 123.22 billion USD in 2019 (Statista, 2020). In case of OECD countries, the average R&D spending as a proportion of GDP is 2.4% (OECD, 2020). Germany is well above this average, with their R&D spending as a proportion of GDP reaching up to 3.13% (OECD, 2020). R&D expenditure when expressed as a proportion of GDP, explains the R&D intensity of any particular region or country, and the commitment to innovative activities. Around two-thirds of the R&D expenditure in Germany comprises of R&D expenditure done by the business sector. Most of the innovative capabilities in Germany can thus be attributed to the business sector. This is a result of fast-paced structural changes in high-performing sectors post the 1990s, especially the automobile sector (Legler, Rammer & Grenzmann, 2006).

Given below is the map depicting the total R&D expenditures of every region in Germany in 2017. The regions (Bundeslander) having the highest R&D expenditures are Baden-Wurttemberg, Bayern (Bavaria), Nordrhein-Westfalen and Niedersachsen. All of these regions, before the fall of the Berlin wall, belonged to West Germany. As is evident from the map, the regions with the highest levels of R&D expenditure belong to the western part of Germany, and the R&D expenditures are on the lower side of the scale in the eastern part of Germany.

The motivation for studying the relationship between R&D expenditure and economic growth at a regional level chosen in this study, which is the provincial level, is that these regions provide a uniformity in several institutional characteristics, such as being under a shared umbrella of shared federal policies. However, there are significant regional differences in the level of R&D spending and industrial structures. Some regions are manufacturing heavy or ‘tech-heavy’, while some regions lack the presence of high-tech industries within the region. New Economic Geography lays down the concept of increasing returns that are

a product of innovation clusters and agglomerations within a country (Funke & Niebuhr, 2000). These agglomerations can cause large differences and segregations among regions with regard to some regions forming the ‘core’ i.e. the economically advanced regions and the ‘periphery’ i.e. the relatively under-developed regions (Funke & Niebuhr, 2000). In this case, the regions of Baden-Wurttemberg and Bayern could be classified as the rich core. The period after the fall of the Berlin Wall nudged the East German economy into a higher speed of convergence, but the differences in the economic growth rates of East and West Germany remain evident through higher unemployment rates, lower exports and overall weaker economic performance in West Germany (Kirbach & Schmiedeberg, 2008). These differences between the regions of East and West German economies make it noteworthy to analyse the effects of differing levels of R&D expenditures on economic growth.

Figure 1: Map of Total Internal R&D Expenditure per region in Germany (2017) Source: Federal Statistical Office, Wiesbaden (2020)

Significantly large volumes of investment are done in developing new technologies and products in Germany every year, which is evident from the fact that Germany is one of Europe’s top research spenders (Federal Statistical Office, 2019). As Germany continues to attract attention as one of the favourable business locations in the world, the R&D foreign

direct investment stocks rose by 96 percent between the years of 2014 and 2017, to almost 2 billion euros (Germany Trade And Invest, 2019). The research and development capabilities of Germany are not achieved by the industrial sector alone, but through fruitful partnerships between various companies, universities and research institutions as well (German Institute of Economic Research, 2019). Strong regional cluster networks such as the “Go-Cluster” programme has ensured that financial support through innovative services and funding for new ideas creates sub-systems of innovation within the country (Germany Trade And Invest, 2019).

The German government’s support through policies such as the “High-Tech strategy” fosters the growth of new technological advancements through streamlined partnerships between the governmental institutions and industries (HTS Progress Report, 2019). The government’s support through this programme consists of reduced interest rate loans, generous R&D grants as well as special partnership programmes between research institutions and the business sector (HTS Progress Report, 2019).

Figure 2: Business to Public R&D Ratio per province in Germany Source: Data extracted from Federal Statistical Office, Germany

0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 13000 14000 15000 16000 17000 18000 19000 20000 21000 22000 EU R mil lion Provinces

Public and Business R&D Expenditure (2017)

As it is evident from the graph, most of the R&D expenditure in Germany across various regions is contributed by the business sector. Baden-Wurttemberg is one of the most innovative and prosperous regions in Germany, with its GDP share in Germany being that of 15% (Eurostat, 2019). Trade in this region too, accounts for around 14% of the total export share in Germany (European Commission, 2019). This success can be attributed to the high levels of productivity in big corporations and SMEs alike in this region. Baden-Wurttemberg and Bayern have some of the lowest unemployment rates in Germany. The R&D intensity in Baden-Wurttemberg was almost 5.6% in 2017, and around 3% in Bayern respectively (Federal Statistical Office, 2020). Moreover, Baden-Wurttemberg has the highest amount of the workforce employed in the high-technology industries, with its share being 16.3% in 2018 as opposed to the total country average of 9.9% (Eurostat, 2019). Thus, it becomes interesting to study the relationship between R&D expenditure and regional GDP, given the peculiar regional differences in R&D strategy.

3. Literature Review

_____________________________________________________________________________________

The purpose of this chapter is to provide the theoretical background on the relationship between different types of R&D expenditures and economic performance, along with empirical evidence regarding the impact of innovation on economic growth. The section concludes with the hypotheses formulated to test the relationship between different types of R&D expenditures and economic performance in the German regions.

______________________________________________________________________ Economic growth in developed countries is largely centred around the production of new knowledge and technological advancements (Romer, 1986; Lucas, 1988; Jones 1995; Fremaine & Balina, 2016). The intuition of research and development being an accelerator of economic growth in the long run is drawn from the New Growth Theory put forth by Romer (1986), starting with Schumpeter (1942) establishing the importance of organized knowledge creation in explaining economic growth (Mladenovic et al., 2016).

3.1. Theoretical Anchoring

The New Growth Theory put forth by Romer (1986), provides a basis for explaining economic growth in a country based on the level of capital stock or subsequently, the level of investment into technological advancement that a country has. The concept that the rate of investment increases with an increase in capital stock, thus charting a higher growth path for a country is quite relevant to this study, as it examines the level of R&D investments that differ regionally, having a proportional impact on the economic growth in that region. This approach is supported by the concept that external economies contribute significantly to the growth process. When firms or regions add to their stock of capital, they unconsciously have a role to play in the productivity of the capital contributed by others (Grossman & Helpman, 1994). These spillovers are observed in investments in physical as well as human capital (Arrow, 1962; Romer, 1986; Lucas, 1988; Grossman & Helpman, 1994; Stokey, 1995). The formation, diffusion and application of new knowledge are the major drivers for economic expansion (Bayarcelik & Tasel, 2012). This process can be an explanation for the “research and development” efforts for firms and governments on a regional or a country level. Most of the R&D models explaining economic growth, focus on “scale effects”: if the amount of resources utilized towards R&D activities are doubled, then the per capita growth rate of output should also double (Jones, 1995). However, several empirical studies present

a significant relationship between investment in R&D at a regional or local level, and economic growth in that region or country, which will be discussed in subsection 3.2. R&D comprises of creative work conducted consistently, to generate new technologies and products to increase the stock of knowledge in a region. Knowledge spillovers, or spillovers pertaining to technological advancement, are often a result of firms investing in R&D activities, which then lead to superior products and services, followed by the imitation done by other firms (Segerstrom, 1991). When this process takes place iteratively, is when a process of economic growth is observed. This in turn, influences policy decisions, depending on the rate of innovation and imitation taking place in that region, as governments aim to stimulate R&D expenditure to increase economic output in that region (Segerstrom, 1991).

Knowledge, being classified as a public good in neoclassical models, is available everywhere and flows in a frictionless manner, contributing to long-term convergence of countries to the growth path (Rodriguez-Pose & Crescenzi, 2008). However, this is not always the case when there are differing stocks of knowledge across space, which then explain differing growth paths (Lucas, 1988). The linearity of the relationship between investment in R&D activities and economic growth in the long-run however simplistic in nature, is a popular policy-making device (Bilbao-Osario & Rodriguez-Pose, 2004). Thus, investment in R&D gives regions the ability to improve and add to the existing stock of knowledge, increase their productivity and thus chart the path to economic growth (Bayarcelik & Tasel, 2012). However, what the neoclassical models of growth lack, is defining the sources of technological changes, as well as different socio-economic characteristics that regions possess (Mladenovic et al., 2016). Innovation is a territorially entrenched process and cannot be understood without considering social and institutional factors fully (Lundvall, 1992). This has led to studies analysing local innovation systems and the success of industrial districts due to inter-organizational networks, financial and governmental institutions in explaining differing levels of innovation among regions (Rodriguez-Pose & Crescenzi, 2008). This process is said to follow an “evolutionary economic change” that considers the creation, diffusion and generation of innovative activities across regions (Bayarcelik & Tasel, 2012). A higher level of accumulation of innovative activities would lead to increased knowledge spillovers and lead to a growth of a country as a whole, but this concept may not be quite realistic at a regional level. Regions do not only rely on their own internal capabilities to produce technological advancements, but they are a result of innovative units interacting with local financial and governmental institutions as well (Rodriguez-Pose & Crescenzi,

2008). This then leads us to the question of the degree of interaction or the balance of external and internal factors affecting innovation within regions.

It is important to observe the relationship between R&D expenditure (public and private) and economic growth at a regional level, as geographical proximity is important for the transmission of knowledge, as much of the stock of knowledge is localized (Blanco et. al., 2016). There is a constant tension between the two opposite forces of uniformly available and standard ‘codified’ knowledge, and the spatially bounded ‘tacit’ knowledge, which leads to different levels of R&D capabilities and thus different levels of regional growth (Rodriguez-Pose & Crescenzi, 2008). The flow of “tacit” knowledge is subject to high costs and the tendency to wane with increasing distance between regions. The innovation systems within a region are observed to be the strongest as they can derive the maximum benefit from knowledge spillovers (Wu Zhou & Wu, 2017). However, this does not stop the flow of knowledge from region to region, but the intensity of the benefit derived from this flow will be directly proportional to the distance between innovative regions (Wu Zhou & Wu, 2017). Bilbao-Osario & Rodriguez-Pose (2004) characterize regional differences as “social filters”: factors that provide regions the capacity to accumulate external R&D and transform it into economic activity. This leads to “innovation-prone” and “innovation-averse” regions in society (Bilbao-Osario & Rodriguez-Pose, 2004). These factors could consist of various indicators such as the level of unemployment, the level of educational attainment of individuals in that region etc. Jones (1995) also uses the number of scientists and researchers in the US as a variable in his analysis to explain the total factor productivity of firms, as in the absence of a human capital variable, there was an overestimation of the coefficient of the R&D expenditure variable. Thus, this provides evidence regarding the importance of including a human capital variable in the model analysing the relationship between R&D expenditure and economic growth.

There are substantial complementary effects observed between return on R&D investments done by regions and firms, and human capital investments and levels too (Frantzen, 2000). The geographical concentration of highly skilled workforce is often found to be high in regions that exhibit larger levels of innovative capabilities (Fritsch & Stuetzer, 2008). The main reason why high levels of human capital contribute to high levels of innovation is that the concentration of skilled people in a region leads to fruitful generation of knowledge, which eventually has an effect on the increase of the stock of knowledge in a region (Florida, 2005). Increased R&D expenditure i.e. increasing the stimulus provided for innovative activities can lead to attracting skilled labour force, which in turn can provide motivation for

firms to increase technological capabilities, to increase output. Florida (2005) finds evidence of high levels of human capital in a region having an effect on the entrepreneurial and innovative capabilities of that region. The amount of workforce having a high level of tertiary educational attainment was found to be the highest in the cities of Berlin, Munich, Frankfurt and Cologne (Fritsch & Stuetzer, 2008). This is consistent with the dataset used in this analysis, where the regions in which these cities belong are found to have high levels of educational attainment i.e. graduation rate from tertiary education.

A large body of literature showcases that returns to public R&D are much lower than the returns to private R&D (Silaghi et. al., 2014). This finding is consistent with the theory put forth by Romer (1993) that companies utilize R&D to maximize profit whereas public R&D is characterized by firms receiving it at subsidized rates (Harhoff, 2000). Several studies state that there is a positive relationship between R&D subsidies or incentives provided by the government in strengthening the R&D capabilities of firms, thus leading to higher outputs and accelerating economic growth (Davidson & Segerstrom, 1998; Segerstrom, 2000; Gorg & Strobl, 2007). When studying the effects of R&D expenditure on economic growth, it is essential to take into account market failure caused by R&D spending occurring at a level below socially optimal level (Gorg & Strobl, 2007). Hence, the importance of public R&D expenditure is to correct this kind of a market failure by aiming to achieve a movement towards the social optimal level (Gorg & Strobl, 2007). Publicly funded R&D expenditure can provide a stimulus to the R&D capacities in a particular region or country, and further the innovative activities by creating additionality to private R&D expenditures (Beise & Stahl, 1998).

It is important to understand the sources of R&D and their relationship with regional economic growth. Some papers suggest that public and private R&D are substitutional in nature, however more often than not, they are found to be complementary to each other (Coccia, 2010). Beise & Stahl (1998) argue that the government funding for research is instrumental in increasing R&D capabilities of firms and regions alike, but it does not incentivize researchers and scientists to commercialize innovation. Firms will prioritize increased innovation to maximize their profits, and governments will do so to maximize the benefit to the economy of regions and nations. The distinction between the economic motivations of the two, are the reason why the substitutability and complementarity argument is important to analyse. In case of basic research, the government is highly likely to encourage business R&D to reduce the burden of subsidies and grants, or the costs and

efficient, then the public and private R&D expenditure will prove to be complementary rather than substitutable, and an increase in public R&D will only encourage more private spending in R&D.

However, the counterargument for public R&D providing a push for private R&D expenditure, is that increasing productivity on the firm level may ‘crowd out’ public R&D spending as firms better their own technological capabilities to rely on their own research and development activities, which are essentially derived from public R&D spending (David, Hall & Toole, 1999). Firms may increasingly develop their own technologies and innovation, thus contributing to growing productivities that reflect in a higher share of the output at a sectoral level in regions and countries (David, Hall & Toole, 1999; Guellec & van Pottelberghe, 2003). This tends to be the case when there is no “additionality” that public funding adds to the private funding allocated towards R&D activities, which then leads to other firms also depending on their own private support higher than public funding, leading to an aggregate non-additionality, leading to a higher share of public funding in the economy (Guellec & van Pottelberghe, 2003). This result is supported by Silaghi et al (2014)’s study, which analyses that the impact of private R&D spending on the economic growth of central and Eastern European countries is much higher than the impact of public spending. Countries with weak R&D intensities are a result of a heavy dependency on public funding for research and development activities in the private sector, which was found to be inefficient (Silaghi et. al., 2014). Moreover, the economic structure and the type of industry that exists in an area also plays a major role in determining the level of innovation in a region. Agricultural countries or regions may not report high levels of innovation or R&D activities, but those regions that are manufacturing or ‘tech-heavy’ industrial regions may report higher innovative capabilities (Bilbao-Osario & Rodriguez-Pose, 2004). This could lead to the observation of public and/or private funding for R&D activities being higher in these regions. Thus, it evidently becomes important to this analysis to gauge the effect of internal R&D expenditures at a public as well as private level on the regional economic growth in Germany.

3.2. Empirical Evidence

There are several studies that contribute to the analysis pertaining the long-run relationship between R&D and economic growth, for individual firms, sectors, countries as well as regions such as the EU, as will be discussed further in this section. Multidisciplinary innovation plays a significant role in sustained national economic growth. Afonso et al. (2014) conduct a panel study for the countries of Austria, Finland, Sweden, the UK and the USA and find significant positive effects of increased productive public expenditure in R&D on economic growth in the short, medium as well as the long run. The importance of productive public expenditure as well as productive interactions between various public and private institutions is further highlighted by Akinwale (2012) in the analysis focusing on the effect of public and private R&D investment on economic growth in Nigeria. The main finding in his study shows that the coefficient R&D is negatively related to economic growth, which in turn showcases that just a straightforward increase in R&D spending is not enough to stimulate innovation when there are weak institutions, low level of interaction between government institutions and businesses involved which hinder the effect of increased spending. However, in case of the USA, Blanco et. al. (2016) found spillover effects to be present in US states and a significant positive impact of R&D spending on the returns to regional GDP in the states. Many of the studies conducted in the case of developed economies in the EU, or the US, have shown positive relationships between R&D expenditure and economic growth.

The importance of collaboration between enterprises and private non-profit or government institutions is highlighted by Czarnitzki, Ebersberger & Fier (2017) in a comparison study between Germany and Finland at the firm level, where collaborations and public R&D subsidies are found to have a positive impact on economic growth. However, in case of Germany, they find that public R&D subsidies have little to no impact on economic growth at a national level. In one of the German panel data studies, it was found that at a firm level, high technology sectors or firms generate more R&D spending, which leads to increased technological and economic productivity in those regions (Harhoff, 2000). This is evident within different provinces in Germany, that certain provinces perform better as they accumulate larger levels of technological stock, and maintain steady investment patterns, thus experiencing greater productivities. In the Swedish context, Edquist & Henrekson (2015) find that the growth contribution by R&D expenditure is significant in increasing the value added in the non-farm business sector.

In the EU region, a positive significant relationship between public and private R&D spending and economic growth for individual nations as well as the region as a whole is found in several studies (Piras et. al., 2012; Silaghi et. al., 2014; Moutinho et. al., 2015 Freimaine & Balina, 2016; Solokov et. al, 2016; Szarowska, 2017). Bronzini & Piselli (2009) find evidence of a positive relationship between R&D expenditure and the economic growth in the context of Italian regions, along with a strong impact of human capital on economic growth. Frenken, Van Oort & Verburg (2006) perform a regional level analysis for Netherlands, and too find that R&D expenditures cause a significant positive impact on regional productivities. The positive impact of R&D expenditure on regional economic growth has also been analysed and proven by Wang & Wu (2015) in the case of Chinese regions. In central and eastern European countries as well, the impact of innovation on economic growth has been highlighted by Petrariu et. al. (2013). The effects of technological spillovers and increased R&D expenditures on economic growth have been found to be positive in the case of Russian regions as well (Kaneva & Untura, 2018). Some of the studies consider firm level data to analyse sectoral growth within a country, and many consider a panel of multiple countries in their data set. Much research on the differing rates of economic growth and R&D investments in Germany at a regional level has not been conducted yet, and this paper intends to contribute to this gap found in the surveyed literature, by assessing the impact of internal public and private R&D expenditure on the regional economic growth in every province.

3.3. Hypothesis

The fixed effects panel regression models are employed to test the relationships between total R&D expenditure, public and private R&D expenditure and GDP per capita in this paper. Based on the research objectives stated in section 1, these hypotheses come into light in lieu of the analysis to be conducted in this paper:

Hypothesis 1: Total R&D Expenditure has a positive relationship with GDP per capita. Hypothesis 2a: Business R&D expenditure has a positive relationship with GDP per capita. Hypothesis 2b: Public R&D expenditure has a positive relationship with GDP per capita

While Hypothesis 1 takes precedence in terms of the main objective of this paper, it is also interesting to analyse whether public sector and business sector R&D have a significant impact on the regional GDP across Germany, and to what extent this relationship exists. Business sector R&D expenditure, as mentioned in the literature surveyed as well as in the context of Germany, has a significant share in the total R&D expenditure of regions, thus it is quite reasonable to expect business R&D expenditure to have a significant relationship with GDP. Similarly, since Germany has a plethora of governmental and non-profit institutions dedicated to scientific research, it is also of interest to this study to assess the relationship between public sector R&D expenditure and regional GDP. These relationships are captured within hypotheses 2a and 2b.

4. Data Description and Methodology

_____________________________________________________________________________________

The purpose of this chapter is to provide some descriptive statistics, correlation and information about the data used in the analysis, along with an explanation of the models and the method used to test the hypotheses pertaining to the research objective

______________________________________________________________________ This analysis employs a multivariate fixed effects regression model framework, in the case of two different models which test the two hypotheses. A Fixed effects (FE) approach is important to study the relationships between regional GDP per capita and each of the independent variables (namely R&D expenditure, graduation rate and unemployment) within each region individually. Each region has its own features and factors affecting each of these individual relationships, which may affect the relationships between the dependent and explanatory variables, and it is important to explore that in this framework. Further explanation for the motivation to use a fixed effects estimation is discussed in subsection 4.2. The analysis will then test various combinations of variables in each model pertaining to the two different hypotheses, to observe the interplay between the dependent and the explanatory variables, and study individual relationships. Postestimation analysis includes tests for autocorrelation, heteroskedasticity as well as the Hausman test to determine whether a fixed effects or a random effects model will be used1_{. The fixed effects model, with region}

and time fixed effects, will the main estimation technique used in this paper. 4.1. Data Description

The data used for all the variables in this analysis has been collected from the Federal Statistical Database “Genesis” of Germany, for the time period of 2000-2017. This is the longest available time period in terms of data availability. Variables such as trade openness (sum of exports and imports of the country/region expressed as a proportion of the GDP) are often used in the literature studying the impact of R&D expenditure on economic growth, but such data was not available for a long time period in the regional database of Germany, which could be a limitation for this analysis. Nevertheless, the variables chosen in this analysis are described in table 1.

Table 1: Variable Definitions

All of the variables are transformed into the natural logarithmic form. Gross Domestic Product (GDP) is one of the most widely used measures of economic performance, and it measures the extent of economic activity in a particular region or country. The GDP per capita is the variable used to measure real economic growth in every province of Germany. Funke & Niebuhr (2000) use this variable to capture the relationship between R&D intensity and economic growth in the context of West German regions in their paper. The advantages of using this measure are that it allows the researcher to measure the economic performance in various regions, with respect to the population size of the region.

The total internal R&D expenditure (RD) by every province in Germany is expressed as the total internal R&D expenditure (public and private) as a proportion of GDP in that particular region. The business sector internal R&D expenditure is the total R&D spending done by enterprises belonging to every sector in that region. The public sector R&D comprises of the R&D spending done by government and private non-profit institutions in every region respectively. This analysis is focused on the relationship of R&D expenditure and growth, and the measure of R&D intensity is useful for analysing the commitment to innovative activities in every region. Funke & Niebuhr (2000), Silaghi et. al.(2014), Solokov et al. (2016) use R&D expenditure as a proportion of GDP, to capture not just the size of the R&D expenditure, but R&D intensity in each region.

Variable Variable Label Definition

Economic Performance GDP GDP per capita

Total R&D Expenditure RD Total internal R&D Expenditure (as % of GDP)

Business Sector R&D

Expenditure BRD Internal Business Sector R&D expenditure (as % of GDP

Public Sector R&D

Expenditure PRD Internal Public Sector R&D Expenditure (as % of GDP)

Human Capital (Education) HC Graduation Rate from higher education institutions (%)

The human capital variable (HC), is the graduation rate from higher education (tertiary) level institutions in every region. This variable is expressed as percentage share of graduates belonging to the relevant age group. Endogenous Growth theory stresses the importance to high levels of human capital contributing to economic growth in the long run, thus the addition of a human capital variable was theoretically motivated.

The Unemployment Rate (UR) in every region is added as a control variable as it affects both the R&D capacities and expenditures on a firm and regional level, as well as has a relationship with GDP growth in the short and the long-run. This is not one of the main variables of interest, but was added to specify the models and add another factor affecting the economic output in a region.

There are certain variables or characteristics that vary across regions such as industrial structures: certain regions are manufacturing heavy in Germany such as Baden-Wurttemberg and Bayern, thus could possess higher levels of innovative capabilities that translate into higher outputs and productivities. As the sectoral classifications with respect to the variables used in this analysis could not be considered due to lack of open data, this analysis focuses on the regional context alone.

4.2. Fixed Effects Models for Panel Data

A traditional panel regression analysis is employed in this paper, whilst controlling for region and time fixed effects. Different variations of the models are employed as a robustness check for the models used in the hypothesis. Every hypothesis is tested using combinations of various variables to observe the individual relationships of each of the variables with GDP per capita in the provinces of Germany. Moreover, tests for autocorrelation as well as heteroskedasticity are used as a part of the postestimation analysis. If there is a presence of heteroskedasticity and/or autocorrelation in the models, then panel corrected standard errors are used to remedy these issues and to estimate the same model. The general equation used to estimate the model is:

𝑙𝑛𝐺𝐷𝑃𝑖𝑡 = 𝛼1+ 𝛽1𝑙𝑛𝑅𝐷𝑖,𝑡+ 𝛽2 𝑙𝑛𝐻𝐶𝑖,𝑡+ 𝛽3 𝑙𝑛𝑈𝑅𝑖,𝑡+ 𝜇𝑖+ 𝛿𝑡+ 𝜀𝑖𝑡 (1)

Here, 𝜇𝑖 are the region specific effects that are considered in the estimation, and 𝛿𝑡 are the

year specific effects used. The fixed effects regression analysis has several advantages in this case as opposed to a random effects estimation. In a regional setting, fixed effects are

included in the model to control for regional characteristics such as geographical location of the region, past history etc. These time invariant characteristics need to be controlled for when individual relationships between the dependent and independent variables are tested empirically, hence the choice for a fixed effects model. Moreover, in a random effects estimation, it is assumed that specific individual characteristics are random variables that do not have a correlation with the lagged, current as well as future values of the regressors (Solokov et. al., 2016). This is quite a strong assumption to have in this analysis, let alone any economic regression analysis. The fixed effects estimation technique allows specific individual effects to have a correlation with the regressors.

Fixed Effects estimation is specially relevant in the case of a regional study such as this one, due to idiosyncrasies and differences observed in the case of East and West German R&D expenditures as well as economic growth patterns. Differing institutional factors such as region-specific policy framework, governmental institutions, political history etc. are important regional differences that a fixed effects estimation technique can control for. Several unobserved factors, especially in a regional panel framework such as the one considered in this paper, are bound to be common across the regions. This is why controlling for unobserved heterogeneity is paramount to this analysis. Time-invariant variables which could be institutional factors shared across all regions, are absorbed by the fixed effects employed for regions and years, which helps isolate individual relationships between the explanatory and dependent variables. While random effects modelling can control for heteroskedasticity to an extent, it imposes strict restrictions such as the explanatory variables being uncorrelated with time-invariant unobservable heterogeneity across regions, which in this case, is a poor assumption to hold. A simpler explanation could be that the error component, in case of a random effects model, captures disturbances in a random, large sample (Gujarati, Porter & Gunasekar, 2009). For a regional study, the 16 regions in Germany may not necessarily be such a large sample. For analysis that could potentially influence policymaking, FE modelling is far more consistent than the RE approach (Wooldridge, 2009). However, there could be certain time-variant effects that could potentially affect GDP, R&D expenditures or the other explanatory variables used. Factors such as inflation rate, tax rates and consumption expenditures are examples of time-variant effects that could affect the estimated relationship between the different types of R&D expenditures and GDP. Nevertheless, this analysis aims to attempt at testing the relationship

between economic performance of regions as well as innovative capabilities of regions utilizing the available data in the designed framework.

Additionally, the 2008 economic crisis caused certain minor fluctuations in the GDP and R&D investments in the regions. Fixed effects models also allow to control for external shocks that may affect the dependent variable in the given time period. Optimal lags for the R&D variables are also used in the analysis for every model, as the effect of R&D expenditure may not necessarily be reflected in GDP growth in the same time period. The lag criteria for the model is chosen according to the Bayesian Information Criterion as well as the Hannan-Quinn Information Criterion, which are most suitable as they are more parsimonious that the Akaike Information Criteria as they use the least number of parameters to pick the optimal lag length, thus being more parsimonious.

The model used for estimating the second hypothesis is specified as follows:

𝑙𝑛𝐺𝐷𝑃𝑖𝑡 = 𝛼1+ 𝛽1𝑙𝑛𝐵𝑅𝐷𝑖,𝑡+ 𝛽2𝑙𝑛𝑃𝑅𝐷𝑖𝑡+ 𝛽3 𝑙𝑛𝐻𝐶𝑖,𝑡+ 𝛽4 𝑙𝑛𝑈𝑅𝑖,𝑡+ 𝜇𝑖+ 𝛿𝑡+ 𝜀𝑖𝑡 (2)

The second model tests the relationship between business and public R&D expenditure with GDP with the same control variables used in the model used to test the first hypothesis. The objective of this model is to estimate the extent of the impact of each business and public R&D expenditure on the regional GDP, given the levels of human capital and unemployment across regions respectively.

Autocorrelation in panel data models causes a bias in standard errors and can reduce the efficiency of estimators in any estimation. However, The Wooldridge test for autocorrelation is conducted in this analysis, as it is flexible, robust and has a high size and power (Drukker, 2003)2_{. The null hypothesis of this test is that there is no autocorrelation in the idiosyncratic}

error terms3_{. Furthermore, the modified Wald test for heteroskedasticity is used to test for}

heteroskedasticity across various cross-sections. This test is particularly useful for panel data analysis with a moderate N and T (cross-sections and time periods respectively). The null hypothesis of this test is that there is homoskedasticity across cross-sectional units. The

2 _{The results from the Wooldridge Autocorrelation test and the Wald Test for Heteroskedasticity are presented}

in Appendix B.

3_{Wooldridge(2002) conducts a supplementary regression that tests for serial correlation in the model in first}

advantage of the Wald test over tests such as LM test and standard Wald tests is that it can be performed even when the assumption of non-normality of the disturbances is breached (Baum, 2001).

4.3. Descriptive Statistics

Stated below are some descriptive statistics of the variables used in the analysis. The R&D expenditures are expressed in millions (EUR) as proportions of GDP. The GDP per capita is also expressed in millions (EUR). The Graduation Rate (HC) is expressed as a percentage share of graduates from tertiary level educational institutions and the Unemployment Rate is expressed as a percentage value of the unemployment in a particular region as well. The R&D expenditures variables (total, business and public) are expressed as a proportion of the GDP in various regions, hence they have smaller values.

Table 2: Descriptive Statistics

Variable Mean Standard Deviation Min. Max.

GDP 30456 9514.054 16223.29 63586.27 RD 2.256 0.886 0.968 5.637 BRD 1.292 0.850 0.120 4.712 PRD 0.470 0.253 0.130 1.229 HC 25.432 8.525 7.2 57.8 UR 9.947 4.269 3.2 20.5

The presence of correlation among the variables employed in this model is observed through testing the correlation among the variables used in both the models in this analysis. The interplay among the different variables in this analysis will be further tested by using variations of the control variables (with and without) in the analysis. Correlation is tested at 1% level of significance4_.

Table 3: Correlation Matrix GDP RD BRD PRD HC UR GDP 1 RD 0.444* 1 BRD 0.525* 0.897* 1 PRD -0.121* 0.159* -0.234* 1 HC 0.720* 0.561* 0.455* 0.251* 1 UR -0.613* -0.411* -0.615* 0.510* -0.403* 1

Due to the nature of the relationship between unemployment rate and economic growth and R&D variables, the correlation amongst them is negative. The highest correlation is between total R&D expenditure and business R&D expenditure, which can raise issues of multicollinearity in the model. However, I will be estimating the relationships of each of these variables with GDP growth in separate models to control for this issue. The correlation between BRD and GDP is significant and positive, which is expected theoretically. However, the correlation between PRD and GDP is significant and negative, with a rather low value. This could be due to the fact that internal public R&D expenditure, as is evident throughout the data set, does not account for a major share in the region’s total R&D expenditure, thus having a weak relationship with GDP. Nevertheless, the analysis will test out each of the R&D variables’ relationships with regional GDP.

5. Results & Discussion

_____________________________________________________________________________________

This section provides a discussion of the results from the fixed effects estimation analysis for hypotheses 1 and 2, postestimation analyses as well as interpretation and a general summary of these results

______________________________________________________________________ 5.1. Fixed Effects Estimation Results

Hypothesis 1 and 2 are tested using the fixed effects regression analysis with different variable specifications as well as different lag combinations for R&D expenditures, the main variable of interest. The changes in the resulting coefficients and their signs are observed by using different variable combinations in both of the two main models used for testing hypotheses 1 and 2. Four different models, namely 1a, 1b, 1c and 1d are used for testing hypothesis 1, and models 2a, 2b, 2c and 2d are used to test hypothesis 2. A linear regression model framework is used to test all of the aforementioned models, including region and year fixed effects. Models 1d and 2d use lagged values of RD, BRD and PRD to assess whether lagged values of these variables have a significant relationship with GDP in regions.

Model 1 tests the relationship between RD and GDP, with using different variable

combinations in models 1a, 1b and 1c, and a one-period lag for RD is used in model 1d. Table 4 presents the fixed effects regression results for all of the models employed in this analysis.

Model 1a utilizes just the RD variable and GDP as the dependent variable, and RD has a negative significant relationship with GDP in this model. The model gives the result that a 1 percent change in RD will yield a -0.085 percent change in GDP. The expected relationship between R&D expenditure and GDP is often positive, which is theoretically motivated as well as empirically evident through the literature surveyed relevant to this case. However, in this case, the coefficient of RD has a negative sign. A straightforward reason pertaining to this result could be that given the small value of the coefficient, even if the sign is negative, RD has a small, yet negative impact on GDP at a regional level in Germany i.e. increased investments in R&D do not positively influence GDP directly. What was found during estimation, is that after adding year fixed effects, the sign of the RD coefficient changes from positive to negative, with the value of the coefficient remaining more or less similar. This

could very well be due to the possibility of some time trend being captured through the year effects that is not accounted for in this analysis.

Table 4 - Regression Results – FEM Estimation

GDP (1a) (1b) (1c) (1d) (2a) (2b) (2c) (2d) Constant 10.131*** 10.004*** 10.713*** 10.748*** 9.964*** 9.877*** 10.645*** 10.657*** (0.01) (0.055) (0.055) (0.092) (0.023) (0.055) (0.057) (0.058) RD -0.085*** -0.087*** -0.022 (0.02) (0.02) (0.016) RD (-1) -0.031* (0.016) BRD 0.001 0.001 0.003 (0.010) (0.010) (0.006) BRD (-1) 0.002 (0.007) PRD 0.103*** -0.099*** -0.054*** (0.021) (0.021) (0.014) PRD (-1) -0.064*** (0.015) HC 0.046** -0.013 -0.026* 0.033* -0.018 -0.031** (0.02) (0.013) (0.014) (0.019) (0.013) (0.013) UR -0.246*** -0.236*** -0.241*** -0.229*** (0.014) (0.014) (0.013) (0.013)

Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes

R-sq. (overall) 0.131 0.168 0.376 0.336 0.256 0.376 0.375 0.342

Observations 288 288 288 272 288 288 288 272

Groups 16 16 16 16 16 16 16 16

Note: *, ** and *** denote significance at 10 %, 5% and 1% level of significance respectively, and the standard errors are displayed in parentheses.

When year fixed effects are not added, it is often possible for the relationship observed between the dependent and the independent variable to be spurious, or a false relationship,

that could be rather present in the dependent variable and some other variable shadowed in the time trend. Year fixed effects encompass the interpretation of various unobservable time-variant factors that may affect the dependent as well as independent variables over the time period stipulated in the models. Controlling for year effects is vital, as they showcase business cycle trends that exist in the dataset. The overall R-squared value for this model is quite low, but progressively gets better as more variables are added to the model. Model 1b showcases the results after the addition of HC. Here as well, the RD variable has a similar coefficient, and has a negative significant relationship with GDP. HC has a positive significant relationship with GDP, and a 1 percent increase in the graduation rate causes a 0.046 percent increase in the GDP of regions. This result is expected, as according to the Endogenous Growth Theory, regions or countries with higher levels of human capital fare better in their levels of economic welfare (Romer, 1986; Lucas, 1988; Hasan & Tucci, 2010, Bartelsman et. al., 2014).

Model 1c showcases results after the inclusion of UR. This model indicates drastically different results as opposed to models 1a and 1b. Upon the addition of UR, the relationship between RD and GDP is insignificant, along with the relationship between HC and GDP becoming insignificant as well. The sign of the coefficient of RD remains the same, but the value reduces. The sign of the coefficient of HC becomes negative in this model, and it is insignificant as well. UR is negatively strongly significant at the 1 percent level, and has a negative coefficient. A 1 percent rise in unemployment rate leads to a decrease of 0.246 percent in GDP. This relationship as well, is theoretically consistent, as economic performance of a region or a country is negatively affected by a rise in unemployment. However, the peculiar aspect of this model is the effect that inclusion of UR has on the sign and significance of RD and HC. This could be attributed to the high correlation between GDP and UR, and the relationship between these variables being stronger than the other explanatory variables used in this model, namely RD and HC. This could lead to UR statistically drowning out the significance between HC and GDP particularly, as UR and HC too have a strong correlation, thus possibly yielding the discussed result due to multicollinearity issues.

Model 1d includes a one-period lag of RD, along with variables HC and UR in the model. Here, it is observed that after adding a lag, RD becomes negatively significant, as opposed to model 1c where it was negative and insignificant. A 1 percent increase in RD causes the GDP to decrease by 0.031 percent. HC as well becomes negatively significant, whereas UR

significant impact on GDP. This kind of effect is expected, as it is often observed that R&D expenditure tends not to have an immediate impact on the economic performance of regions in the same year, but rather yields an impact after a few years. Technological shocks/advancements, both endogenous and exogenously, tend to have a lagged effect on the growth of an economy, due to the effect of its chained reactions, and linkages to other aspects of an economy that is unobservable, and sometimes observable. However, since the sign of the coefficient of RD is negative, it cannot be concluded that there is a positively significant relationship between RD and GDP at a regional level in Germany.

The negative sign that the coefficient of RD exhibits throughout the models, could also be attributed to the nature of the variables RD and GDP. RD is the R&D expenditure as a proportion of GDP, meaning that mathematically GDP is the denominator in this variable. So as the GDP per capita rises, the denominator in the RD variable rises as well. This could lower the value of the RD variable, thus generating the result that is obtained in the estimation results table. Similarly, this could be the case when the relationships between PRD and GDP, and BRD and GDP are tested as well, as these variables are also calculated as a proportion of GDP.

Model 2 tests the relationships between BRD and PRD, with GDP within the same

framework as applied in model 1, with HC and UR as control variables used in this model. In this case too, different variable combinations and one-period lag in the case of model 2d is used to test the relationship between BRD and PRD with GDP. Models 2a, 2b, 2c and 2d all are tested using region and year fixed effects too.

Model 2a tests the standalone relationship between BRD, PRD and GDP without HC and UR. It is found that the coefficient of BRD is positive and insignificant, and the coefficient of PRD is negative and significant. This implies that a 1 percent increase in PRD causes a 0.103 percent decrease in GDP. The basic interpretation of this model is that public R&D expenditure has a negatively significant relationship with GDP, while business R&D expenditure has no significant relationship with GDP. This could be due to the fact that individual firm characteristics such as factor productivities, allocation of resources within firms and asset values are not taken into consideration in this analysis, thus testing the relationship between business R&D expenditure and GDP becomes a difficult task through this model specification alone. Public R&D expenditure has a significant relationship with

GDP, however, it is negative, just as the relationship between total R&D expenditure and GDP.

Model 2b has the inclusion of the HC variable. The results showcase that PRD still has a negatively significant relationship with GDP, while BRD still has an insignificant relationship with GDP. A 1 percent increase in PRD causes the GDP to decrease by 0.099 percent. However, HC is significant and the coefficient is positive, which is consistent with theory and other empirical findings in the surveyed literature. A percentage increase in the graduation rate causes the GDP to increase by 0.033 percent. Model 2c includes all the control variables (HC and UR). BRD is still insignificant, but PRD remains negatively significant. The values of the coefficients of PRD reduce from model 2a to 2c consistently as more variables are added, but the sign of the coefficient remains constant. It also remains significant at the 1% level of significance. UR is still negatively significant, as expected. Model 2d includes one-period lags for BRD and PRD, to analyse whether there is a lagged effect of any of these variables. The control variables HC and UR are included as well. PRD, consistent with models 2a, 2b and 2c, is negatively significant and BRD, consistent with the earlier models as well, is insignificant when their lags are considered. HC, upon adding lags of BRD and PRD, becomes significant but the coefficient has a negative sign. UR too shows consistent results with respect to earlier models, and maintains a negatively significant relationship with GDP. A possible explanation for the significance of HC in this model could be that the effect of multicollinearity it has with level BRD and PRD reduces when lagged values of BRD and PRD are included in the model, hence it shows a significant relationship. Moreover, throughout all the models, it is noteworthy to mention that the region fixed effects are always present. The core of this regional study of Germany is that different regions of Germany are endowed with different levels of growth and existing capital in terms of technology and business structures. So, region fixed effects are important in accounting for the regional heterogeneity when considering the variables in this paper’s analysis.

5.2. Fixed Effects Estimation with corrected standard errors

Postestimation analyses reveals through the Wooldridge test for autocorrelation and the modified Wald’s test for heteroskedasticity, that there indeed is autocorrelation and heteroskedasticity present in the data. Moreover, the Pesaran’s test for cross-sectional dependence also shows that there is spatial dependence present in the data. Due to these

and consistent estimates. As a potential remedy for this, the fixed effects regression model is run with Driscoll-Kraay (1998) standard errors.

Adjusted standard errors obtained through the Driscoll-Kraay estimation using fixed effects are robust to autocorrelation of residuals, heteroskedasticity as well as spatial dependence. While it is quite obvious that regional studies such as the one considered in this paper are bound to have certain levels of spatial dependence, it must be controlled for to avoid unobservable aspects to have an effect on the estimates in the regression models (Hoechle, 2007). Driscoll & Kraay (1998) provide a solution to these issues by using a nonparametric approach towards the covariance matrix estimator to ensure robust standard errors that control for cross-sectional dependence (Hoechle, 2007).

Table 5 - Regression Results – Driscoll Kraay Corrected Standard Errors Estimation

GDP (1a) (1b) (1c) (1d) (2a) (2b) (2c) (2d) Constant 10.131*** 10.004*** 10.713*** 10.748*** 9.964*** 9.877*** 10.645*** 10.657*** (0.014) (0.098) (0.055) (0.092) (0.033) (0.090) (0.049) (0.058) RD -0.085*** -0.087*** -0.022 (0.02) (0.02) (0.013) RD (-1) -0.031** (0.013) BRD 0.001 0.001 0.003 (0.008) (0.009) (0.004) BRD (-1) 0.002 (0.004) PRD -0.103*** -0.099*** -0.054** (0.031) (0.034) (0.022) PRD (-1) -0.064** (0.029) HC 0.046 -0.013 -0.026 0.033 -0.018 -0.031* (0.036) (0.018) (0.016) (0.037) (0.019) (0.016) UR -0.246*** -0.235*** -0.241*** -0.229*** (0.016) (0.014) (0.018) (0.013)

Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes

R-sq. (within) 0.959 0.96 0.981 0.981 0.961 0.961 0.982 0.982

Observations 288 288 288 272 288 288 288 272

Groups 16 16 16 16 16 16 16 16

Note 1: *, ** and *** denote significance at 10 %, 5% and 1% level of significance respectively, and the standard errors are displayed in parentheses.

Model 1a generates similar results to the results obtained from the fixed effects estimation in table 4, in case of the significance and the coefficient sign of the RD variable. Model 1b also gives similar results for RD, but HC is insignificant after adjusting standard errors. This contrasting result could be due to controlling for autocorrelation and heteroskedasticity. Model 1c again, gives similar results as model 1c in table 4. RD is insignificant, along with HC, and UR has a negatively significant relationship with GDP. Model 1d with the one-period lag for RD exhibits a negatively significant relationship with GDP, while HC in contrast to the fixed effects estimation in table 4 shows insignificant relationship with GDP after adjusting standard errors.

Model 2a yields similar results to model 2a in table 4, with PRD having a negatively significant relationship with GDP. Model 2b, with the inclusion of HC and adjusted standard errors, yields an insignificant relationship between HC and GDP. However, the negatively significant relationship between PRD and GDP remains consistent with the results from table 4. Model 2c, with the inclusion of UR, yields a negatively significant relationship between PRD and GDP and UR and GDP respectively as well. HC remains insignificant. Model 2d includes a one-period lag for PRD and BRD, along with the control variables HC and UR. In this case however, HC is significant at the 10 percent level of significance but still has a negative coefficient. PRD showcases a negatively significant relationship with GDP, still consistent with the fixed effects estimation without robust adjusted standard errors. Throughout this analysis, BRD remains insignificant as well.

Lastly, as an econometric check, a Hausman test is performed to check whether the fixed effects estimation or a random effects estimation approach is suitable in the chosen model specification5_{. The Hausman test determines whether the error term is correlated with one}

or more of the regressors. In this case, the null hypothesis of the test is accepted and the conclusion of the test is that the random effects model is more appropriate i.e. there is no correlation between error terms and the regressors. However, as Gujarati, Porter & Gunasekar (2009) explain, there is no ultimate rule (in this case, the Hausman test) to determine which model should be preferred based purely on a statistical test. While the Hausman test suggests the use of the random effects specification, economic intuition on which model should be employed in the analysis should be of paramount importance. In this

paper, it is important to control for regional heterogeneity, thus by using region fixed effects. Moreover, controlling for year effects, as discussed previously in this section, is important to capture time trends in this analysis. Thus, the model chosen in this paper focuses on a fixed effects model specification.

To summarize the results, it is found that total R&D expenditure across various regions in Germany does not have a positive significant relationship with GDP, and in fact, yields a negatively significant relationship with GDP. Thus, hypothesis 1 is rejected. Similarly, it is observed that business sector R&D expenditure does not have a significant relationship with GDP, but public sector R&D expenditure is found to have a negatively significant relationship with GDP. Thus, hypothesis 2 is rejected as well. The results regarding the relationship between RD and GDP, peculiarly, could me mirroring the relationship between PRD and GDP, which is evident through the estimation results from tables 4 and 5. In another scenario with a different model specification for testing the relationship between RD and GDP or BRD and GDP, the estimation could possibly produce different results. Another potential drawback and a possible reason for getting these results could be that the inclusion of year effects reflect some effects of time-variant variables that are not included in these models. The effects of R&D subsidies provided by the federal government along with grants or funding, differ across regions and could not be captured in this model due to unavailability of this data at a regional level. Concluding remarks and a further discussion of these results is presented in the next section.