MSc Thesis International Economics and Business
The Effects of Foreign and Domestic Competition
on Innovation
University of Groningen
Faculty of Economics and Business
Name: Sam Brands Student Number: S2135485
Student email: s.a.m.brands@student.rug.nl Supervisor: Dr. D.H.M. Akkermans Co-assessor: Dr. J. de Haan
Abstract
Building on the inverted U-shape model of Aghion, Bloom, Blundell, Griffith and Howitt (2002) this thesis empirically estimates the effect of foreign and domestic competition on innovation. The inverted U-shape model argues that for low levels of competition innovation is stimulated but when competition rises this effect weakens and every extra competitor has a stronger negative effect on the level of innovation. This thesis hypothesizes competition should be defined separately for foreign and domestic since they both have different effect on innovation. Foreign competition is argued to be especially intense due to distinct home-based advantages, high productivity levels and a broad access to resources, and therefore the negative effects of enhanced competition on innovation are hypothesized to set in at an earlier stage. For domestic competition the reverse effect is theorized. Following the BEEPS dataset for 2005, 2007 and 2009 on transition economies, a logistic panel regression is performed. The findings suggest the importance of differentiating between foreign and domestic competition, however no sufficient evidence is found to support an inverted U-relation between foreign or domestic competition and innovation.
Preface
Table of Contents
1. Introduction………...6 2. Theoretical Framework……….8 2.1 Conceptual Model………...8 2.2 Innovation………...8 2.2.1 Firm-level drivers……….9 2.2.2 Country-level drivers………9 2.3 Competition………...102.3.1 Defining foreign and domestic competition………...10
2.3.2 Foreign and domestic competition in transition economies...…...11
2.4 Innovations and Competition………11
2.4.1 ‘Traditional’ competition-innovation literature……….….11
2.4.2 The inverted U-shape model……….. 12
2.4.2.1 The escape-competition effect………...…..13
2.4.2.2 The Schumpeterian and discouragement effect…………...14
2.4.2.3 Graphical representation………..14
2.4.3 Building upon the inverted U-shape model………15
2.5 Measuring Innovation………16
2.5.1 Input Model………16
2.5.2 Output Model………..16
2.6 Foreign and domestic competition and the Inverted U-shape model…………17
2.6.1 Foreign competition………...……….17
2.6.2 Domestic competition………..……….…..18
2.6.3 Input and Output model………..18
3.3 Robustness Checks………24
3.3.1 Fixed or Random effects and Endogeneity……..………..25
3.3.2 Common Method Bias………...26
3.3.2.1. The Correlational Marker Technique……….26
3.4 Econometric Model………...27 4. Results……….28 4.1 Descriptive Statistics……….28 4.1.1 Summary Statistics………28 4.1.2 Multicollinearity………....28 4.1.3 Hausman test……….29 4.1.4 Correlation Matrix……….30
4.2 Correlational Marker Technique………...30
4.3 Model Estimation………..32
4.3.1 Econometric Model 1: Input Model………...32
4.3.2 Econometric Model 2: Output Model………35
5. Conclusion………..37
6. Limitations and Further Research………...39
7. References………...40
8. Appendices………..43
Appendix 1: Description of variables……….43
Appendix 2: Industry overview dataset………..44
Appendix 3: Results of the Hausman test………...…………44
A: Fixed effects logistic panel regression………44
B: Hausman test results………....45
Appendix 4: Graphical representation (Lowess Curve) of the hypothesized curvilinear relation between R&D Expenditures and Foreign or Domestic competition……….46
1. Introduction
The origins and features of innovation have been much debated since the work of Schumpeter (1942) but a changing world has opened up a new round of discussion. In recent years literature adds to the understanding of innovation by questions that arise due to globalization, and more specifically the global market place with its enhanced competition (e.g. Vives, 2005; Gaubinger, Rabl, Swan and Werani, 2015; Gorodnichenko, Svejnar and Terrell, 2010). Globalization has changed the level of competition, every country or company that desires to do so, can try to claim a share of the market. Competition has intensified, and, following Porter and Stern (2001): “Innovation has become the defining challenge for global
competitiveness”.
This thesis is concerned with the effect competition has on innovation. Current literature is scattered between the belief whether competition has a positive or negative effect on innovation. Most recent literature builds on the inverted U-shape model first introduced by Aghion, Bloom, Blundell, Griffith and Howitt (2002). The Inverted U-shape model integrates both the positive and negative effects of competition on innovation. It argues that for low levels of competition innovation is stimulated but when competition rises this effect weakens and ultimately, in very competitive markets, every extra competitor has a stronger negative effect on the level of innovation. When these findings are presented in a graph an inverted U-shape displays the relation between innovation and competition. (Vives, 2008; Aghion et al., 2002). The inverted U-shape model doesn’t differentiate between the effect of domestic and foreign competitive pressure. Yet, as classic economists like Porter (1990) and Melitz (2003) find, there is a vast difference between the two. Following these classical theories it can be argued foreign competition is fiercer than domestic competition, and considering the Inverted U-shape model of Aghion et al. (2002), it is expected the competition-innovation relation is different for foreign or domestic competition.
What is the effect of foreign and domestic competition on innovation, and to what extent do these effects differ from each other?
In order to best answer this research question the panel data of the Business Environment and Enterprise Performance Survey (BEEPS) 2005, 2007 and 2009 will be used to conduct a logistic panel regression analysis. The BEEPS panel dataset used in this research is conducted of three firm level surveys for the countries in the regions of Central and Eastern Europe and Central Asia. The results of the empirical model estimation are unsatisfactory to draw any meaningful conclusions with regards to the research question.
The following chapter will provide a theoretical framework to define the key terms and clarify the argument made from where the research question is posed. After this the methodology will explain the choice of dataset as well as the explanation of the model and its variables. Hereafter the chapters will show the data obtained and explain what can be derived from the results of the empirical analysis. Ultimately this thesis will conclude and provide an overview of the most important contributions as well as limitations of the research.
2. Theoretical Framework
The purpose of this chapter is to discuss several aspects of competition as well as innovation and more importantly the relation between these two. Following this theoretical framework this chapter builds the hypotheses, which will be tested later on in this thesis.
2.1 Conceptual Model
Before discussing the framework on which the theory of this thesis is built, the following figure gives an overview of the variables that will be modeled and the relations that will be addressed in the continuing of this thesis. The square boxes on the left indicate the independent variables, representing competition. The box on the right represents innovation, the dependent variable. Innovation is subdivided into R&D expenditures that represent the input measurement and new development to represent the output measurement of innovation. The control variables, depicted by the box below (dashed), are present to make sure the purest possible relation between competition and innovation can be identified. Lastly, the interaction effect between domestic competition and the foreign competition-innovation relation is hypothesized which is depicted by the diamond shaped figure on the far left.
Foreign Competition R&D Expenditures H3 Domestic Competition
New product or service development Control Variables Innovation Domestic Competition H1 H2 Input Output
Figure 1: Conceptual Model
2.2 Innovation
In this thesis the following definition of innovation will be used:
“Innovation is the multi-stage process whereby organizations transform ideas into
differentiate themselves successfully in their marketplace” - Baregheh, Rowley and
Sambrook, 2009.
In order to comprehend the origins of innovation and get an understanding of the different effects of domestic and foreign competition on innovation it is important to be aware of the drivers of innovation outside of competition. These drivers can be sub-divided in firm- and country-level drivers.
2.2.1 Firm-level drivers
First of all, the size of a firm can be a driver of firm level innovation (Schumpeter, 1942). Larger firms generally have better access to finance and thereby are better able to invest in innovative activities (Shefer and Frenkel, 2005; Pavitt, Robson and Townsend, 1987). Correspondingly, the OECD (2015) provides evidence that firms categorized as innovative see sufficient access to finance as a main driver of their innovative activities. Thus it is found that the likelihood that innovation will occur increases when a firm increases in size. Secondly, firm age can be a driver of innovation. Young firms, more specifically start-ups, are most likely to radically innovate, meaning introduce a product totally new to the market (EBRD, 2014). This is important since radical innovations can create new industries and can function as a stepping-stone for many (incremental) innovations in the future. Where young firms are found to be very innovative, older firms are considered more reluctant to change and innovations (Goedhuys and Veugelers, 2012; Hansen, 1992). For these reasons firm age is negatively related to innovation. Lastly, the ownership structure of a firm can influence the level of innovation. When a firm is privately owned, returns on investments are for a large part to be kept by the owner of the firm, which causes the incentive to innovate to be relatively high since innovations can yield high returns. However, when a firm is state owned the incentive is significantly lower since the public manager doesn’t receive the returns of the risky and time-consuming investment that has to be made to innovate (Shleiger, 1998; EBRD, 2014). Innovation isn’t necessarily driven by the private ownership of a firm but when a firm is state-owned the likelihood of innovations to occur can be hampered (EBRD, 2014).
2.2.2 Country-level drivers
property must be well protected to stimulate firms to innovate. Though, the relation between intellectual property rights and innovation is non-linear (e.g. Lai, 1998; Chen and Puttitanun, 2005). Full restrictions on intellectual property rights are not always better than partial restrictions. In most cases intellectual property protection can be a driver of innovation but it must be considered too much protection has the opposite effect (Auriol, Biancini and Paillacar, 2014).
2.3 Competition
Competition basically is the fighting for market share in the business context. In order to best answer the research question posed in this thesis it is important to understand the difference between foreign and domestic competition.
2.3.1 Defining foreign and domestic competition
2.3.2 Foreign and domestic competition in transition economies
Most literature on competition and innovation has investigated developed or developing nations. This thesis adds to the existing literature by investigating transition economies. A transition economy is an economy that is in transition of being a planned economy towards a market economy (Cambridge Dictionary, 2016). Transition economies present an interesting focus since, as the EBRD (2014) finds, transition economies have joined the globalized world of today in terms of trade liberalization, but simultaneously the integration process of an adequate institutional context has proven itself to be challenging. Considering the previously discussed theories of Porter (1990) and Melitz (2003), the difficulties with the institutional context together with the high degree of changes at the firm and country level when transitioning, are likely to magnify the differences between foreign and domestic competition when considering transition economies.
2.4 Innovations and Competition
Competition and innovation are often considered endogenous terms; meaning there is causality loop between them (Martin, 2012). When competition increases, firms are likely to start innovating to keep up with competitors and ensure their survival, but simultaneously innovation is a tool to compete and thereby increases competition, hence the terms partially originate in each other. This thesis is interested in the effect of competition on innovation; the effects of innovation on competition will not be directly addressed in this thesis but are important to bear in mind. In most cases literature takes a firm stand on whether the effects of enhanced competition on innovation are positive or negative but in recent years a theory called ‘the Inverted U-shape model’ has combined these two stances.
2.4.1 ‘Traditional’ competition-innovation literature
presence of many rivals. Kamien and Schwartz (1975) model technological rivalry against innovation and find an intermediate level of rivalry is more likely to stimulate innovation than a monopoly situation. Another stream of literature argues competition has a positive effect on innovation since fierce competition will motivate firms to innovate because they have to do so in order to survive, no matter at what cost (Ahn, 2002). To summarize, since Schumpeter introduced the subject, literature has been dispersed about the relation between competition and innovation ever since (Gilbert, 2006).
2.4.2 The inverted U-shape model
Most traditional literature about competition and innovation takes a definite stand on whether the relation is positive or negative. In an exceptional case an intermediate effect is found. However there is a theory that has been able to integrate traditional perspectives by acknowledging the relation between competition and innovation is dependent on the intensity level of competition. Aghion, Bloom, Blundell, Griffith and Howitt introduce the inverted U-shape model in 2002. A model that argues competition and innovation are curvilinear related.
The inverted U-hypothesis argues that neither perfect competition nor a monopoly can provide the perfect circumstances for innovation. The research focuses on product market competition and innovation and argues that for low levels of competition the so-called ‘escape competition’ effect dominates, and for higher levels of competition the Schumpeterian effect is most evident. Aghion et al. (2002) find very convincing evidence for an inverted U-shape relation between competition and innovation where the number of patent citations measures innovation, and product market competition is measured using the Lerner Index1. The data used is drawn from a firm-level panel dataset from the UK spread out over a range of industries.
The logic of the model is build upon the motivation of a firm whether or not to innovate. This motivation is linked to the technological abilities of the firm compared to the technological frontier of its industry. Aghion et al. (2002) call the distance of a firm to the technological frontier of an industry its technology gap, denoted by m, where 0 indicates no gap at all and 1
1 Measures the firm’s market power using the price, marginal costs and price elasticity.
Source: http://www.policonomics.com/lp-monopoly1-lerner-index/
2 The Wald test can be used to test the equality of two coefficients. Source: Stata.com
the largest technology gap possible. This gap also corresponds to the competitive situation of an industry. If the firms in an industry are neck-and-neck, which means at the same competitive level with little difference in their technological abilities, the industry is leveled and m = 0. But when one, or a few firms are leading the other firms in the same industry, and the competitive abilities of the firms in the same industry are thus very dispersed, the industry is labeled as unleveled and m = 1.
The inverted U-shape model interprets firm-level data at the industry level. Dependent on size, age or other firm level characteristics, firms have a different position opposed to their industry’s technological frontier (Syverson, 2014). Initially Aghion et al. (2002) include a parameter in their model that reflects the fact that firms perform distinctive to be able to interpret their model at the industry level. However, in the same 2002 paper they find that the inverted U-shape theory also holds at the firm level. The only addition is that when firms are characterized as highly competitive the inverted U-shape relation between competition and innovation is steeper. Accordingly Inui, Kawakami and Miyagawa (2012) have investigated the appliance of the Inverted U-shape theory at the firm level for Japanese firms in the manufacturing industry. They confirm the inverted U-shape relation between competition and innovation at the firm level holds similarly to the original model. Their only addition is that when the firms’ distances to the technological frontier of an industry are very dispersed the effect is somewhat lessened.
All firms react differently when competition increases dependent on how far they are located from the technological frontier. But, translating these different reactions into an industry context, the sizes of these differences together determine whether an industry classifies as leveled or unleveled. Hence the model shows where firm performances meet. As Inui et al. (2012) and Aghion et al. (2002) find, the effects of competition on innovation depend on to what extent the industry is characterized as (un)leveled (or how dispersed the firm level performances within an industry are). The following two sections will discuss the working of the escape-competition effect that occurs in the leveled state and the Schumpeterian and discouragement effect that arise in the unleveled state of an industry.
2.4.2.1 The escape-competition effect
firms gain additional motivation to innovate when they have the possibility to ‘escape’ its competitors and instead of being neck-and-neck with them strive by them. This means that when competition is relatively low every firm-level innovation will rapidly increase the average innovation level of the industry. Competition and innovation are thus strongly positively related for low levels of competition when the industry is characterized as leveled. However, since the firms in a leveled industry are neck-and-neck, a significant innovation to strive by competitors is difficult to achieve and the industry will spend most of its time remaining a leveled industry. Nonetheless firms will still try to innovate, even more when competition enhances and the yield of successful innovation increases. Ultimately, at some point, a firm does succeed to escape competition, the technology gap increases, m becomes closer to 1, and the industry will classify as unleveled.
2.4.2.2 The Schumpeterian and discouragement effect
For an unleveled industry the opposite relation to innovation exists. Aghion et al. (2002) find that when competition is very high, m is close to 1, the Schumpeterian effect, and as Ding, Su and Jiang (2016) find, the ‘discouragement’ effect occurs. To recap what was mentioned before, Schumpeter finds very high competition causes firm profits to be diminished and for this reason no excess capital to invest in innovation will remain, hence the average innovation in the industry will decrease when competition increases. This effect is found to be strongest for firms who are further from the technological frontier (Aghion et al., 2002; Bloom, Draca and Van Reenen, 2015). Aghion et al. (2002) label these types of firms laggards. Laggard firms are likely to get discouraged to innovate because even if they manage to innovate while dealing with price wars, this innovation will still not get them to the frontier of the industry and classify them to be a ‘neck-and-neck-firm’. This is called the ‘discouragement effect’ (Ding et al. 2016). For firms closer to the technological frontier the same effects occur, though less extreme and sudden. Since Aghion et al. (2002) investigate the industry level, laggards and leaders come together and, on average, the relation between competition and innovation is slowly decreasing for an unleveled industry. Following the Schumpeterian effect an industry that is unleveled will spend most of the time remaining unleveled since price wars only become more forceful when competition enhances.
2.4.2.3 Graphical representation
Aghion et al. (2002) investigate the industry level drawn from firm level panel data and thus give an average depiction of the relation between competition and innovation for leveled and unleveled industries. The top of the inverted U-shape marks the transition of leveled into unleveled, or from the escape-competition effect into the Schumpeterian or discouragement effect. Competition In n ov at io n Unleveled Leveled H ig h L ow Low High
Figure 2: Graphical representation of the Inverted U-shape model of Aghion et al. (2002)
2.4.3 Building upon the inverted U-shape model
reason it is difficult to draw any conclusions. This thesis will attempt to integrate the inverted U-shape model with the findings of Bloom et al. (2015) to be able to address the effects of different forms of competition on innovation. To be able to achieve this in a transition economy context it is first important to understand the measurement implications that characterize a broad term like innovation.
2.5 Measuring Innovation
Innovation is a difficult term to grasp; the most widely accepted empirical measure is the measurement of the number of patents, or more specifically the number of citations of patents (Klein and Sorra, 1996). However this tool isn’t suited for this research because the tendency to apply for patents differs per country, and in the case of transition economies this tendency is relatively low (EBRD, 2014). The EBRD (2014) finds that in transition economies universities hold most of the patents, which is a persistent heritage of the planned economy past, and makes them an unsuitable measure for the firm-level focus of this research. Also this information is not openly available. Besides, this paper attempts to capture innovation in the broadest sense possible and patents are more useful when considering inventions instead of innovations (Gorodnichenko and Schnitzer, 2010). For these reasons this thesis approaches the measurements of innovations by, similar to Crepon, Duguet and Mairessec (1998), capturing innovation by assessing both the inputs and outputs of innovation. Meaning the resources that are needed for innovation to occur will be investigated as well as the actual creation of ‘something new’.
2.5.1 Input model
The input side of the model refers to the resources that are considered to be beneficial for innovation to occur. In existing literature one specific input is often used as an indicator of innovation; the amount a firm spends on research and development (R&D). The more a firm invests in research and development the higher the chance of innovations to be formed (e.g. Bilbao-Osorio and Rodriguez-Pose, 2004). This also means R&D expenditures are no guarantee for innovation. Nevertheless, R&D expenditures give insight in the innovation efforts of a firm and are available for many years, countries and industries.
2.5.2 Output model
‘new’, it is a valuable measure of innovation that is not always available. Nonetheless, even this rather straightforward measure of innovation has a downside one must be aware off. The measurement of new products and services leaves out any other form of innovation like, for instance, process innovation or innovative alterations to existing products or services. It is for these reasons important to be attentive towards the limitations of the measurements of innovation, but combining input and output measures gives a valuable estimation (Crepon et al., 1998).
2.6 Foreign and Domestic competition and the Inverted U-shape model
The theory discussed up to this point has shown competition and innovation have a curvilinear relation that is either positive or negative dependent on the competitive structure of a market. Since foreign competition is argued to be fiercer than domestic competition, foreign competition is theorized to have a different effect on the competitive structure of a market. For this reason it could be argued identifying foreign and domestic competition separately causes the dynamics of the inverted U-shape model to change.
2.6.1 Foreign Competition
Foreign competitors bring different, valuable knowledge along with them. This knowledge is new and innovative to the domestic firms in that market and the entrance of foreign firms can thereby create large differences in the competitive positions of the firms active in the market, and quickly ‘unlevel’ the industry.
earlier stage and the financial possibilities for domestic firms to innovate will slowly diminish with every enhancement of foreign competition.
2.6.2 Domestic Competition
When modeling domestic competition, the reverse effect is expected. Domestic competitors have the same resources available and therefore it can be difficult to escape from a neck-and-neck position. For this reason it is expected the leveled situation of an industry is prolonged and the motivation to innovate will slowly increase until domestic competition is relatively fierce and the yield of innovating, and thereby the return on investment on innovations, is high enough. When this happens, as Aghion et al. (2002) find, the industry becomes unleveled and the Schumpeterian effect sets in. Since, the breaking point of the industry from leveled to unleveled is expected to be positioned at a high level of domestic competition, the firm-level finances and market prices are already on edge and the investments in innovative activities are expected to decrease at a relatively high rate.
2.6.3 Input and Output model
Considering the input and output measurements of innovation adopted in this thesis it is expected the development of new products or services has a more severe effect on the technology gap than R&D expenditures. This is because when a firm is able to introduce something entirely new, the gap to the other firms in an industry or market is suddenly profoundly increased. With R&D expenditures this possibility exists, but since investing in R&D isn’t a guarantee for innovation, the effects of investing extra on R&D are expected to be less influential. Since both the input and output model measure innovation, the same relations for foreign and domestic competition on innovation are expected. However for new development the foreign and domestic competition innovation curve is expected to be slightly steeper.
2.6.4 Graphical representation
However in the hypothesized foreign and domestic curve, the division between leveled and unleveled also shifts to the left or right dependent on the location of the top of the hypothesized curve. Competition In n ov at io n H ig h L ow Low High
Figure 3: Hypothesized relations foreign and domestic competition for the Inverted U-shape model
2.7 Interaction effect
Up until this point it has been assumed foreign and domestic competition are two separate forms of competition that have no influence on each other. However, in reality, they occur jointly and moreover, they are expected to interact with each other.
2.8 Hypotheses
After discussing the theoretical framework surrounding the main research question the hypotheses that will be empirically tested can be formulated. The main goal of this thesis is to investigate the separate effects of foreign and domestic competition on innovation. Following the theoretical framework presented it is hypothesized both forms of competition have a nonlinear relation with innovation like in the original inverted U-shape model. In the case of foreign competition it is expected the inverted U shape has a short but steep positive effect, meaning for relatively low levels of competition innovation increases quickly, but at an early stage a lengthy downturn is expected where the competition-innovation relation becomes negative. This results in the following hypothesis:
H1: Innovation and foreign competition have an inverted U-shape relation with a short positive effect and a lengthy negative downturn
For domestic competition the reverse effect is theorized. As addressed previously, escaping domestic competition is more difficult due to the available resources, which prolongs the escape-competition effect. This also means a long build-up of competition is expected before a leveled industry becomes unleveled. This results in a long positive and short but steep negative relation between competition and innovation from which the next hypothesis can be formulated:
H2: Innovation and domestic competition have an inverted U-shape relation with a lengthy positive effect and a short negative downturn.
Lastly, foreign and domestic competition occur simultaneously and so it can be expected the relation to innovation is sensitive to the interaction of both forms of competition. It is argued that when domestic competition is high, firms or industries are better able to withstand foreign competition and innovation will be less effected by enhanced foreign competition. Therefore it is expected foreign and domestic competition negatively interact with each other which results in the last hypothesis:
3. Methodology
This chapter explains the methodology of this thesis. First of all it will discuss the sample and its characteristics, secondly the variables will be extensively discussed by separately identifying the dependent, independent and control variables to then discuss the robustness checks performed before building the econometric model.
3.1 Sample
For this study data from the Enterprise Survey initiative of the World Bank and the European Bank for Reconstruction and Development (EBRD) is used. This dataset, labeled as the Business Environment and Enterprise Performance Survey (BEEPS), is comprised of face-to-face interviews with managers and business owners and entails over 17.000 firms’ interviews. The interviews are conducted in over 135 countries and are available for several years. The dataset covers firm characteristics like size or ownership structure, but even more variables on the particular business environment. Examples of this are questions about political stability, innovation activities or the degree of competition.
From these series of surveys I use the panel dataset of 2005, 2007 and 2009, available on Central and Eastern Europe as well as Central Asia and the Caucasus since these are the largest transition regions in the world. The countries included in the sample are Albania, Belarus, Bosnia, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Kosovo, Latvia, Lithuania, Macedonia, Moldova, Montenegro, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Ukraine for Central and Eastern Europe and Armenia, Azerbaijan, Georgia, Kazakhstan, Kyrgyzstan, Mongolia, Tajikistan, and Uzbekistan for Central Asia and the Caucasus. The countries China, Vietnam, Korea or Cambodia are not included in the dataset and are deliberately left out since, even though they are transition economies, they present a different region of the world. For the same reason the observations of Turkey are excluded from the existing sample. Since 2008 the BEEPS datasets have adopted the Enterprise Surveys Global Methodology and use stratified random sampling. Obtaining panel data, hence interviews with the same firms across multiple years, is a priority of the World Bank and EBRD (www.ebrd-beeps.com/methodology/).
2005 and 2007 10,915 observations are included in the sample and from this sample 2,626 observations are matched with 2009 as well.
3.2 Variables
3.2.1 Dependent variables
The dependent variable in this research is innovation. As previously discussed this research attempts to best measure innovation by including an input and output function. In the empirical context this means two models will be tested each with its own dependent variable. The dependent variables are whether or not the firm has made any R&D expenditures over the last three years and; whether or not the firm has developed new products or services in the last three years. This means both variables are binary measures and the outcome is either ‘yes’ or ‘no’. The BEEPS dataset does include a continuous measure of both dependent variables, however, due to the large amount of missing observations, the reliability of their outcome has strongly decreased and therefore the binary measure will be included in the empirical model. Both dependent variables will be included in the empirical model as dummy variables, meaning they are assigned a 1 if the answer is ‘yes’ R&D expenditures or new developments were made, and 0 if this wasn’t the case and the answer was ‘no’. See Appendix 1 for more elaborate specification of the variables.
3.2.2 Independent variables
domestic and foreign competition will be added as an independent variable to be able to test hypothesis 3.
3.2.3 Control variables
In order to be able to say something about the exclusive effect of competition on innovation in transition economies and distinguish between the effects of foreign and domestic competition I want to control for variables outside the scope of this thesis that can influence this relation. The controlling for externalities makes sure the tested relation between foreign or domestic competition on innovation isn’t biased.
control for firm-level innovation drivers like firm size, firm age and the ownership structure of the firm. An overview of all variables and their measurements can be found in Appendix 1. Additionally I want to see what the effects are of the financial crisis of 2008 on the found results. Since the dataset contains measurements of 2005, 2007 and 2009, the amount of R&D expenditures or the development of new products or services is likely to be comparatively lower in 2009 than in the previous measurement-years. Since innovation is considered an important but difficult and costly investment it is to be expected the financial crisis has caused a cut back in innovations regardless of the level of competition, and can thereby create a bias in the findings of the effect of competition on innovation. I want to control for this effect by including a dummy variable that takes a value of 1 if an observation is also matched with the 2009 data, and 0 if not. Lastly I want to control for industry. The BEEPS dataset identifies eighteen types of industries: that can be subdivided in Manufacturing, Retail and Other Industries (for an overview of the specific industries and the frequency they occur in the dataset consult Appendix 2). Since a large majority of the observations falls in the category manufacturing little additional knowledge can be derived from the industry data concerning the competition-innovation relation. For this reason no industry specific effects will be tested but instead I want to control for industry effects since the average innovation rate of one industry can differ from another. Often manufacturing accounts for more than half of a country’s business and to keep this position manufacturing companies are often considered highly innovative (Bloomberg innovation index, 2015). Following the Bloomberg Innovation Index (2015) it can also be concluded retail is not considered an innovative industry. These differences between industries can bias the overall results of competition on innovation and for this reason I control for the industry of an observation.
3.3 Robustness Checks
Before composing the final model and testing the hypotheses alongside it, it is important to test the data used for other causes that can bias the results. This section is dedicated to the tests performed to ensure the robustness of the model.
first the difference between foreign and domestic competition will be tested via a Wald test, then quadratic terms of the independent variables will be added to the logistic regression as well as an interaction term of domestic and foreign competition.2 Moreover, the data used is collected for more than one year, which requires a method to ensure possible estimation bias by correlation between years is filtered out. For this, panel analysis can be used.
3.3.1 Fixed or Random effects and Endogeneity
A logistic panel regression with random effects will be used. Random effects are selected since it is expected different entities influence the dependent variable to some extent since the data collection is dispersed among different firms, industries and countries (Schmidheiny, 2016). Moreover, a Hausman’s specification test was performed to determine empirically whether to adopt fixed or random effects. The idea behind this test is that no matter whether fixed or random effects are adopted, the estimators of both models should be consistent. If this is the case both estimators converge to the true parameter β and for large samples they thus should be identical. When this is not the case, which happens when the independent variables are correlated to the individual specific error component, the random effects estimator is inconsistent while the fixed effects estimators remain constant (Hill, Griffiths and Lim, 2012). The null hypothesis of the Hausman test is ‘all estimators yield identical results’, and when rejected a fixed effects model should be used and when accepted a random effects model (Hill et al., 2012). For this thesis the null hypothesis is accepted, and random effects are adopted in the panel logistic regression (the test results can be found in Appendix 3).
Unfortunately endogeneity is a common problem in random effect models because the error component of random effects is in most cases related with some of the independent variables (Hill et al. 2012). As mentioned before theory also finds competition and innovation to be endogenous variables. In terms of the empirical model this means the endogenous independent variable is correlated with the error term, which can create estimation biases. To the best of my knowledge, there is no fitted solution to solve for this estimation bias in a panel logistic regression with random effects. Therefore the possibility of endogeneity and thereby estimation biases must be taken into account when interpreting the results of the empirical model.
3.3.2 Common Method Bias
Another statistical technique used to check for robustness of the data is the Correlational Marker Technique (CMT). This technique serves to test for the effects of Common Method Bias (CMB) in the dataset and can be defined to be the effect of something external that has influenced the response given by the respondent (Lindell and Whitney, 2001). CMB especially occurs when the same technique for predictor and outcome variables is used (Heppner, Kernis, Lakey, Campbell, Goldman, Davis and Cascio, 2008). CMB is often present in surveys where the participants have to answer subjective questions because external effects can influence the responses given. An example of such an externality is answering differently due to social desirability or laziness. This kind of bias can account for 40-80% of the correlations in the model and thus be very misleading (Heppner et al., 2008). In the case of the BEEPS dataset, and especially for the competition variables used in this study, the same problem arises. For this reason I want to test for the presence of Common Method Bias by using the Correlational Marker Technique of Lindell and Whitney (2001).
3.3.2.1. The Correlational Marker Technique
The Correlational Marker Technique is the latest approach to detecting CMB and implements a methodological and analytic framework (Vishwanath, Egnoto, and Ortega, 2012). Even though the CMT still has disadvantages like that it doesn’t control for common rater problems3, it is considered to be the best measure and statistical solution to CMB (Sharma, Yetton and Crawford, Appendices 2009). The Correlation Marker Technique adds an extra variable to the regression that serves as a marker variable. This has to be a variable that is theoretically unrelated to at least one of the variables in the model (Lindell and Whitney, 2001). The correlation between the marker variable and the theoretically unrelated variable can be interpreted as an estimate of CMB. If CMB is detected you can solve this by partialling out the correlation of the marker variable (Lindell and Whitney, 2001).
Following the correlation matrix of the desired model plus the marker variable, which in this study will be the length of the interview in minutes (‘how many minutes did the interview
last’), the smallest, positive correlation can be identified as rs. The coefficient of each
independent variable with the dependent variable can be identified as rYi. As Lindell and
3 Occur when an individual observes and evaluates another and a systematic personal judgment error is present. Source:
Whitney (2001) describe, the shared variance of the marker variable, representing the influence of CMB on the model, can be removed from the model with the following formulas:
(1) 𝑟𝑌𝑖𝑀= (𝑟(1−𝑟𝑌𝑖−𝑟𝑠) 𝑠)
(2) t-statistic = !𝑌𝑖𝑀
(!!!𝑌𝑖𝑀)(!!!)
Where N is the sample size and rYi.M represents the values of the model corrected for Common
Method Bias. If the coefficients that are significant from the correlation matrix remain significant following equation (2) we can conclude that the results cannot be accounted for by CMB. The results of the Correlational Marker Technique and other robustness checks can be found in Chapter 4 Results.
3.4 Econometric model
Taking into account the above mentioned variables and theory this results in two econometric models: an input model and output model. The models will be tested separately, but as mentioned, together represent innovation.
Econometric Model 1: Input Model
R&D Expenditures = β0 + β1 Foreign competitive pressure + β2 Domestic competitive pressure
+ β3 Foreign competitive pressure2+ β4 Domestic competitive pressure2+ β5 Domestic competitive
pressure*Foreign competitive pressure + β6 Knowledge spillovers+ β7 Firm size + β8 Firm age
+ β9 Ownership structure + β10 Financial Crisis + β11 Industry + ε
Econometric Model 2: Output Model
Development of new products or services = β0 + β1 Foreign competitive pressure + β2 Domestic
competitive pressure + β3 Foreign competitive pressure2+ β4 Domestic competitive pressure2
+ β5 Domestic competitive pressure*Foreign competitive pressure + β6 Knowledge
spillovers+ β7 Firm size + β8 Firm age + β9 Ownership structure + β10 Financial crisis +
4. Results
This chapter is concerned with the empirical testing of the hypotheses proposed in chapter 2. First of all an overview of the dataset will be provided and its statistics, secondly the calculations to omit common method bias are presented to ultimately present and discuss the model estimation.
4.1 Descriptive Statistics
4.1.1 Summary Statistics
Table 1 describes the variables that will be studied in this thesis. The number of observations N, mean, standard deviation and the minimum and maximum value of the variables are portrayed. As can be read from the table all variables have the same number of observations, this is because the missing values for the dependent variables were dropped due to the missing 2009 data. This has cut the total number of observations of the BEEPS dataset more or less in half but this has left a more pure sample where observations were made in 2005, 2007 and in some cases these observations were matched with 2009 as well. These matched observations are indicated by the variable ‘Financial Crisis’. Moreover, Table 1 shows the majority of the variables have a minimum value of zero and a maximum value of one. This indicates they are dummy variables. Considering the dependent variables at the top of the table we can see firms, in the last three years, in 24.8% of the time have developed new products while 54.8% of the firms in the sample have invested in R&D. The summary statistics in Table 1 also tells us the domestic pressure from competitors is considered more influential than foreign competitive pressure. Foreign pressure is in most cases ranked second or ‘slightly important’ while domestic pressure is more close to the third rank: ‘fairly important’. This is contrary to the theoretical argument that foreign competition is fiercer than domestic competition.
4.1.2 Multicollinearity
Table 1: Summary Statistics
Variable N Mean Std. Dev Min Max
New Development 10,915 0.248 0.432 0 1 R&D Exp. 10,915 0.548 0.498 0 1 Foreign 10,915 2.084 1.135 1 4 Domestic 10,915 2.730 1.049 1 4 K. Spillovers 10,915 0.100 0.230 0 1 Small firm 10,915 0.630 0.483 0 1 Large firm 10,915 0.098 0.230 0 1 Firm Age 10,915 1993.999 13.532 1825 2008 Gov. Own 10,915 1.635 10.063 0 99 Financial Crisis 10,915 0.231 0.421 0 1 Manufacturing 10,915 0.373 0.484 0 1 Retail 10,915 0.414 0.492 0 1
Table 2: Multicollinearity test
Variable VIF 1/VIF
Manufacturing 1.79 0.557 Retail 1.74 0.574 Small firm 1.29 0.776 Large firm 1.26 0.793 Foreign 1.12 0.895 Firm Age 1.10 0.909 Domestic 1.09 0.918 K. Spillovers 1.06 0.941 Gov. Own. 1.03 0.972 Financial Crisis 1.00 0.996 Mean VIF 1.25 4.1.3 Hausman test
Table 3: Hausman specification test
Hausman test Fixed effects = Random effects
Input Model Output Model
[R&D Exp.] Fixed – [R&D Exp.] Random = 0 [New Dev.] Fixed – [New Dev.] Random = 0
Chi2=17.08 Chi2=7.86
Prob>Chi2=0.196 Prob>Chi2=0.853
4.1.4 Correlation Matrix
Table 4 shows the Pearson Correlation Matrix of all variables and indicates when this correlation is significant at the 5% level. Noticeable is that in general all coefficients are low (<0.3), only the correlation between retail and manufacturing is relatively high (-0.648). The closer the correlation coefficient is to 0, the less likely a linear relation exists between the independent variables. The relatively low correlations correspond with the findings of the VIF test presented above. Moreover, the Correlation Matrix in table 4 shows an extra variable called Minutes. This variable was added to the matrix to test for Common Method Bias. The following section will elaborate on the results of this test.
4.2 Correlational Marker Technique
