The effect of M and A activity in the pharmaceutical industry on innovation performance : a poison pill for innovation? : Is there a relationship between M and A activity and R and D intensity and patent intensity in th

(1)

The effect of M&A activity in the pharmaceutical

industry on innovation performance; a poison pill

for innovation?

Is there a relationship between M&A activity and R&D intensity and patent intensity in the European and American pharmaceutical industry for the time period 1995 – 2010?

Bachelor’s Thesis Frida Buck

Faculty Economics and Business

Supervised by L. Zhao

July 15

th

₂₀₁₅

(2)

Abstract

This paper empirically investigates the effect of M&A activity on innovation performance in the European and American pharmaceutical industry for the period 1995-2010. A panel data fixed effects regression is performed by using R&D intensity and patent intensity as the main innovation indicators while additionally examining the influence of firm size on innovation performance. This study finds significant evidence for a negative relationship between M&A activity and R&D expenses and a positive relationship between firm size and M&A activity. Several possible explanations are presented in this paper for the relationship between M&A activity and innovation performance, however, inconclusiveness remains and further research must be performed.

Key words: innovation performance, M&A, pharmaceutical industry, fixed effects regression

(3)

Table of Content

1. Introduction 1

2. Theoretical framework 3

2.1 The pharmaceutical industry: a challenging business model 3

2.1.1 The process of developing a NCE 4

2.1.2 Research and development: key step in the value chain 6

2.1.3 Measuring innovation performance 7

2.2 Mergers and acquisitions in the pharmaceutical industry 7 2.2.1 Empirical studies on the impact of M&A activity 8 2.2.2 Rationale for M&A 8

2.2.3 Influence of firm size on innovation performance 10

3. Data & Methodology 12

3.1 Data description 12

3.2 Empirical model 14 3.3 Threats to the OLS regression 15

4. Results & Analysis 16

4.1 Results R&D intensity model 16

4.2 Results R&D expenses model 17

4.3 Results patent intensity model 18

4.4 Results patent applications model 18

5. Discussion 19

6. Summary & Conclusion 22

Bibliography 24

Tables 27

Appendices 33

Appendix 1: Summary statistics 33

Appendix 2: Tests for heteroskedasticity 34

Appendix 3: Tests for significance of time fixed effects (Wald test) 36

(4)

1

1. Introduction

“Nobody holds out for organic growth anymore. Sure, we can all exist by ourselves, but the question is: Where do I want to be in ten years’ time? We’re going to end up, as everybody knows, with a handful of really serious research-based pharmaceutical companies.” Sir Richard Sykes’ statement, executive director of Glaxo Wellcome at the time, shows the thought behind the wave of mergers and strategic alliances of pharmaceutical companies, where the large merger between Glaxo Wellcome and Smith Kline Beecham was part of in 2000. After a productivity shock in 1990, where after the number of new molecular entities launched were decreasing despite an increase in research and development expenditures, the pharmaceutical industry went through a structural change. Technological patents began to expire and the price pressure on medicines increased. As a response, many pharmaceutical mergers and acquisitions occurred. For the years between 1989 and 2004 at least eight of the top ten ranked pharmaceutical companies in 2004 have all been part of a merger or acquisition (Grabowksi & Kyle, 2007). While M&A tend to occur in waves, the number of mergers and acquisitions in the pharmaceutical industry continued to occur constantly over time. Many reasons exist to engage in a merger or acquisition for a firm. Cutting in costs, expanding to other markets or gaining a greater knowledge base, for example. However, the role of M&A activity in gaining a greater knowledge base has been under discussion. Although M&A activity can be projected as an improving situation for the whole company in a managerial perspective, it remains the question whether a company can reach the scope and scale advantages of M&A and really become more innovative. Being a research-intensive industry innovation plays a key role to survive in the pharmaceutical market. Theoretical arguments and empirical studies have studying the role of M&A activity in improving the innovation performance. By examining this role a clearer view can be created about how a research-intensive firm can absorb and make use of external knowledge. Previous research have tried to isolate the factors of M&A whether the innovation performance will be improved, yet contrasting conclusions have been made with regard to the relationship between M&A and innovation performance in the pharmaceutical industry.

To address this ambiguity on the impact of M&A on innovation performance, this paper will examine this relationship by performing a panel data fixed effects regression

(5)

2

for M&A activity in the European and American pharmaceutical industry for the time period 1995-2010. The research question underlying this thesis is as follows: Is there a relationship between M&A activity and R&D intensity and patent intensity in the European and American pharmaceutical industry and if so what relationship can be found? This paper aims to examine the effect by controlling for M&A fixed effects and time fixed effects to isolate the effect exclusively caused by M&A activity on R&D intensity and patent intensity.

The remainder of this paper is organized as follows. The second section contains a theoretical framework, which gives background information about the pharmaceutical industry, the research and development process and M&A activity. The third section discusses the data and methodology used to assess the impact on innovation performance followed by the results in section four. In section five the results will be discussed and evaluated. Finally, in section six a summary of the results and a conclusion will be given as well as recommendations for future research.

(6)

3

2. Theoretical framework

2.1 The pharmaceutical industry: a challenging business model

“A business model of discovery, development, manufacture and marketing of new chemical entities (NCEs) and bio products designed to enhance human health”. While Northrup (2005) gives a clear definition of the main goal of the pharmaceutical industry, a contradiction arises with the current status of the global pharmaceutical industry today. The industry has created new life-saving medicines and innovative medical procedures, which improved global health, quality of life and life expectation. However, many factors have been driving the industry to structural changes. One crucial factor remains the conflict of interest between the stakeholders’ profit-maximizing goals of the pharmaceutical firms and the medical, social and economic need for rationality in bringing effective medicines to the market (Dukes, 2006). Additionally, the rising costs of R&D, increased risk in drug development and intense regulation of governments bring the pharmaceutical sector under pressure. Industry-specific characteristics have a determinant role in facilitating these changes. Important characteristics are the interplay of high risk, long-term phases for drug development and a requirement of high investment returns to attract investors (Northrup, 2005).

Walker (1977) has set four main challenges, which the global pharmaceutical industry has been facing: profit, innovation, efficiency and regulatory challenges. While producing new and effective medicines, pharmaceutical companies must secure revenues to be able to pay the costly R&D of developing a new drug. The costs of R&D are rising rapidly, because of the larger and more complex clinical studies and new expensive technologies (Gassmann et. al, 2008). Secondly, pharmaceutical firms are expected to market innovative products for the cure of diseases, which have no treatment yet or inadequate treatment. Therefore, innovation plays a key role in the pharmaceutical industry and serves as an underlying strength, as it is one of the world’s most research-intensive industries. In 2013 the pharmaceutical sector had the highest ratio of R&D investments to net sales with 15,1% compared to other high-technology sectors (EFPIA, 2013). A third challenge is to bring new medicines, efficiently and quick, to the market. These goals can be reached by maintaining a short time for development and quick approval of medicines. Here, the fourth challenge, the influence of regulatory forces,

(7)

4

plays a big role. An increased necessity of multiple testing before a new drug can be launched, affects the costs for drug development and thus the efficiency of a pharmaceutical company (Ebbers, 2005).

2.1.1 The process of developing a NCE

New chemical entities (NCEs) are molecules that can bind to a target in the human body and cause a biological process to stop or start (Northrup, 2005). While the pharmaceutical industry creates innovation in many forms, the continuous discovery and development of NCEs remains the most significant (DiMasi et al., 2003). The process of developing a NCE can take several years with an average time period of twelve years. After pharmaceutical companies have requested approval for investigation of a new drug, clinical testing follows, which can be divided into three different phases: phase I, II and III. In Phase I small amounts of the drug are tested on healthy volunteers to examine the first effects on a human body. Phase II continues with testing on humans, but now with a higher amount of volunteers. These volunteers are usually patients for whom the drug may be intended to develop. After successful testing in phase I and II, a large-scale trial takes place on the targeted patients, after which a final execution plan will be made and the marketing of a new drug begins (Northrup, 2005).

Figure 1: Success rates per phase in the pharmaceutical business model1

____________________

1_{Source: J.A. Dimasi, “Uncertainty in Drug Development: Approval Success Rates for New Drugs,” in Clinical Trials and}

(8)

5

With the described process of clinical testing, pharmaceutical companies are trying to fill their R&D pipeline with as many as possible new drug development trials to eventually produce a ‘blockbuster drug’, a drug that sells $1 billion or more per year. However, as Figure 1 depicts, the success rates of the different phases are sufficiently low. This creates a costly combination with the high R&D investments already incurred. As an alternative to focusing on the production of high-cost blockbusters, pharmaceutical firms also rely on protecting their patents on drugs that are already developed and in this way avoid incurring large investments for new drug development (Gassmann et al., 2008). A patent will exclude other pharmaceutical firms to bring the same NCE to the market, which brings an indirect gain to the company, without the need for high investments in drug development trials.

Figure 2: Number of New Chemical Entities by Area (1993 - 2013)2

As depicted in Figure 2, the number of new chemical and biological entities launched has been decreasing since the mid-1990s until 2004 (Gassmann et al., 2008). A first explanation for the large amount of NCEs in the 1990’s was the large amount of pharmaceutical firms at the time, being an unconcentrated industry (LaMattina, 2011). Munos (2009) proved with his analysis that the number of new drugs launched during the past 60 years were correlated with the number of firms. This indicates that the increasing concentration may have caused a decline in the number of NCEs launched.

____________________

2_{Source: SCRIP – EFPIA (According to Nationality of Mother Company), EFPIA Key Data 2013}

0 10 20 30 40 50 60 70 80 90 100 1994 - 1998 1999 - 2003 2004 - 2008 2009 - 2013 N umb er of n ew N CE s Time

Number of NCEs by area (1993 - 2013)

Europe USA Others

(9)

6

2.1.2 Research and development: key step in the value chain

Being a highly research-intensive industry, pharmaceutical companies are dealing with a high proportion of R&D expenditures on their total expenditures. Since the 1990’s the industry-wide R&D expenditures of pharmaceutical firms in Europe and the United States of America have been increasing over time, as depicted in Figure 3. The exact allocation of these increased R&D expenditures is difficult to discover. Despite this fact, the increase indicates that investors expected high revenues over the lifetime of developing a new drug and thus invested a higher amount, which resulted in an increase of R&D investments.

Figure 3: Pharmaceutical R&D expenditures in Europe and USA (1990 - 2013)3

When combining the two trends of the number of launched NCEs and pharmaceutical R&D expenditures a contradiction can be seen between the trends for both Europe and the United States of America. As the R&D expenditures (R&D inputs) increase over time and the number of NCEs marketed (R&D outputs) decreases, the R&D processes of many pharmaceutical companies of the last year seem to have an inefficient innovation process. As a consequence the development costs for a new drug per marketed NCE have been increasing. Although a direct comparison between R&D expenditures and R&D productivity is not completely valid, the tendency of the higher R&D cost per new drug indicates another challenge for the companies active in the pharmaceutical industry. ____________________

3_{Measured in millions of national currency units (Europe in €, USA in $). Source: EFPIA member associations, PhRMA}

0 10000 20000 30000 40000 50000 1990 2000 2010 2012 2013 R & D ex pe nd itu re s ( mi lli on s) Time

Pharmaceutical R&D expenditures in Europe

and USA (1990 - 2013)

Europe USA

(10)

7

2.1.3 Measuring innovation performance

The method for measuring innovation performance remains a well-discussed topic. However, no consensus has been found yet on the most efficient measure. This is the case, because the innovation process of a firm, which may include tangible and intangible elements, is a complex process to define and quantify (Archibugi, 1988). One important and basic measure of innovation input is the investment in R&D. The allocation of these R&D investments and the efficiency of how an R&D investment is turned into an R&D output may differ by firm and is therefore difficult to identify. As a result, and despite the information these R&D investments provide, it cannot describe the innovation process fully on its own.

After the digitalization of patent statistics it was made possible to follow behaviour of patenting by firms. This made patent counts an easy accessible quantitative innovation indicator of the amount of inventions made by research-intensive firms (Pakes & Griliches, 1980). It is therefore one of the most direct measures of innovative output currently available. However, patent measures remain to have difficulties such as how to determine the importance of a patent or renewing patent legislation. Changing patent legislation, for example, may cause a misinterpretation of data on patents. To reduce this problem, data for patent applications should be used, when applying a patent count, instead of data for granted patents. The interest of a company to obtain innovation protection can be measured by using data on patent applications, whereas the data of granted patents may be subject to misinterpretation from new patent legislation and a possible lag created between the application and granting of a patent due to processing of patent offices (Basberg, 1987).

Summarizing, using R&D expenditures and patent counts as indicators of R&D inputs and R&D outputs, respectively, will improve the quality of measurement of innovation performance of a company (Hagedoorn & Cloodt, 2003).

2.2 Mergers and acquisitions in the pharmaceutical industry

Facing the aforementioned challenges, a trend of M&A activity in the pharmaceutical industry emerged since the 1990’s. This M&A activity mostly happened in waves, which occurred in the early 1990’s, the mid-1990’s and the beginning of the 21st_{century (Koenig}

(11)

8

industry, which used to be a stable and unconcentrated industry. The frequency of M&A activity in the pharmaceutical industry today implies there are still several reasons why these M&A occur. In the remainder of this section existing research on M&A activity and innovation performance will be discussed and rationale for M&A will be evaluated.

2.2.1 Empirical studies on the impact of M&A activity

While studies about the effects of M&A activity on a firm’s performance are numerous, the specific effects on innovation performance are examined by a handful of empirical studies. This paragraph will discuss a selection of studies performed and summarize their findings.

Danzon et al. (2007) used a sample of 534 M&A firms for the time period 1988-2001 to examine the determinants and effects of M&A activity in the pharmaceutical and biotechnical industry. Gaps in product pipelines, patent expirations and financial trouble were findings of motives for the M&A examined. They found a slower growth on R&D expenditures in the three following years of the merger, although the effect was small. In the study of McCarthy et al. (2012) a sample of 112 M&A were examined on their effect on their patent intensity after M&A activity for the time period 2000-2007. Their findings indicated that there was a chance of more than 60% of a decrease in patent intensity in the post-M&A period.

Desylas & Hughes (2010) studied a panel data set of 2624 M&A between high-technology firms for the time period 1984-1998, whether they improved innovation after M&A. They found a neutral to positive effect of acquisitions on R&D intensity and a negative to neutral effect of acquisitions on R&D productivity in the three years following M&A.

Hitt et al. (1991) investigated the effect of M&A on R&D intensity and patent intensity on a sample of 191 manufacturing acquisitions in the US for the time period 1970 and 1986. A negative effect was found on both R&D intensity and patent intensity following the fours years after M&A activity. Also Hitt et al. (1996), who used a sample of 250 manufacturing acquisitions in the US for the time period 1985-1991, found a negative effect on both R&D intensity and patent intensity after M&A activity.

Summarizing, the findings from empirical studies about the effect of M&A activity on innovation performance in the pharmaceutical industry give ambiguous results. However, a heavy-weighted consensus can be found that there is a negative

(12)

9

impact on innovation performance of the subsequent entity after M&A. Taking into account theoretical arguments and previous empirical studies, the following hypothesis is formed:

Hypothesis 1:

A negative relationship exists between M&A activity and innovation performance

As discussed before, innovation performance can be measured by R&D intensity and patent intensity, together representing R&D inputs and R&D outputs, respectively, thus the following sub hypotheses can be formed:

Hypothesis 1a:

A negative relationship exists between M&A activity and R&D intensity, measured as annual R&D expenditures divided by annual sales (_𝛽𝛽1 < 0)

Hypothesis 1b:

A negative relationship exists between M&A activity and patent intensity, measured as the annual patent applications divided by annual sales (_𝛽𝛽₁ < 0)

While working further on previous empirical studies, factors as different research assumptions, time periods and data samples and M&A motives must be taken into account.

2.2.2 Rationale for M&A

Burns et al. (2005) argued that motives for M&A activity could be broadly categorized into adaptive and defensive motives versus proactive or offensive motives. Factors as economic recessions, new governmental reforms and price pressures are exogenous factors that stimulate defensive M&A activity. As governmental decisions and price pressures can directly affect the pharmaceutical industry, these may be important motives for pharmaceutical companies to merge or acquire other businesses. Another key objective in pharmaceutical M&A is maintaining growth of earnings. As discussed before, in the case of a declining R&D productivity caused by rising R&D expenditures and a decrease in new marketed drugs, the M&A strategy would be a strategic choice to

(13)

10

meet the companies’ challenges. Besides the defensive motives, an offensive motive for M&A activity could be to gain access in a foreign pharmaceutical market (Burns, 2005). An efficient way to enter a new market is to cooperate with local operating firms. Next to entering new geographical pharmaceutical markets, a pharmaceutical company might benefit from entering pharmaceutical markets working in different therapeutic areas. Besides these motives for M&A, Danzon et al. (2007) describe another motive for the pharmaceutical- and biotechnical M&A activity, being the ability to reach a greater innovation performance. Chesbrough (2003) discusses that M&A can play a role in filling product line gaps and enhancing a firm’s internal capability. However, Burns (2005) argues that a lack of focus on the merger of conflicting agendas may cause a problem. Also, when firms merge or acquire they need to be able to absorb the new capacity of gained knowledge in an efficient way.

2.2.3 Influence of firm size on innovation performance

The consensus on the relationship between firm size and innovation shifted over the years. Schumpeter (1942) hypothesized that larger firms are more innovative compared to small firms. First, it can be argued that through capital market inefficiencies large firms have the advantages of securing cash flows for the financing of R&D processes, due to the availability of internally generated funds (Cohen & Levin, 1989). Secondly, larger firms could gain by having economies of scale and scope. Other researchers believed that the argument ‘more is better’ applies: a higher scope of R&D will increase the bets on drug development trials and thus will increase the chance of reaching a success with drug development trials (Burns, 2002).

As Schumpeter emphasized this positive relationship, empirical studies proved otherwise throughout the years and casted doubt on the consensus that was set. Researchers executed empirical research and improved empirical studies by, for example, taking particular business units as observations instead of the whole firm. This resulted in different findings, namely a negative effect of firm size on the innovation performance of the subsequent entity. Others argued, while larger firms may undertake much innovation investment due to their financial advantage, they do not have to be the initial source of their innovations. Also, as firms grow, efficiency in innovation can be undermined through misallocation by management (Cohen & Levin, 1989). Several studies have

(14)

11

complemented this argument by showing firm size had a small and negative relationship with innovation performance (Burns, 2002).

Inconclusiveness remains in explaining the relationship between firm size and innovation performance. Several explanations can be given for this ambiguity. Empirical studies may have not taken into account different firm characteristics or did not take control for different industries. The shift of the consensus on this relationship could also be explained by a real change in the relationship itself. Changes in the pharmaceutical industry as for example changing patent legislation, different industry structure or increasing competition may be factors that might have affected the relationship. However, the shift of the consensus, mostly hold today, towards a negative relationship between firm size and innovation will be followed in this study. This gives the following expectation about the effect of firm size on innovation performance in this study:

Hypothesis 2:

A negative relationship exists between firm size and innovation performance, measured by R&D intensity and patent intensity (_𝛽𝛽₂ < 0)

(15)

12

3. Data & Methodology

In the following section the data and regression method will be discussed. The objective is to determine for a various amount of completed M&A deals within the pharmaceutical industry whether there is an effect of M&A activity on innovation performance and if so what relationship can be found. A panel data fixed-effects model will be used to account for M&A fixed effects and time fixed effects. In this way the effect on innovation performance, exclusively caused by M&A activity, can be determined. Additionally, the possible influence of firm size on innovation performance will be discussed.

3.1 Data description

In order to examine the effect of M&A on innovation performance a panel dataset is being used which combines time series and cross-sections. The dataset consists of multiple variables measured over multiple years, each relating to 47 M&A deals in the European and American pharmaceutical industry, consisting of 37 acquisitions and 10 mergers. European and American pharmaceutical M&A are chosen for this study, because they together count for the highest percentage of new drugs launched in the chosen time period. Observations are taken for nine years in the time period 1995 – 2010. Due to a possible lag of innovation input into a firm’s output observations are taken in the four years after an M&A deal has been completed. As a result of inadequate data, nine M&A deals have observations for eight years, which creates an unbalanced panel data set.

M&A deals in the pharmaceutical industry have been defined by using deals completed by companies that fall under the SIC-codes 2833 – 2836, defined by the U.S. Standard Industry Classification, where the 3-digit ‘283’ is classified as the ‘Drugs’ industry. The relevant M&A deals have been collected through the databases Thomson One and Zephyr. Several assumptions have been made regarding the sampling of the pharmaceutical M&A deals. The sample consists of M&A deals with a minimum transaction value of $1 million. In this way a distinction is being made between small value transactions and large value transactions whereby an emphasis is being placed on large firms operating in the pharmaceutical industry. Companies who were involved in a

(16)

13

merger or acquisition had to be public companies. At last, for all the acquisitions the acquirer owned 100% of the stake after the acquisition.

To measure innovation performance most efficiently, the indicators R&D expenditures and patent counts are being used. Argued in the theoretical framework, there could be a relationship between firm size and innovation performance. Therefore, R&D expenditures and patent counts will be divided by annual sales to correct for this relationship. This leads to the two dependent variables, which together define a firm’s innovation performance in this study: R&D intensity and patent intensity.

The dependent variables are conducted by adding the observed units of both companies. In the post-M&A period the observed units of the merged entity are taken, in case of a merger, and the observed units of the acquirer, in case of an acquisition. In this way, a comparison can be made for the companies taken together, before and after M&A activity. To completely perform the analysis concerning the two dependent variables, separate regressions with R&D expenditures and patent applications as dependent variables will be executed, which will result in four models explaining the relationship between M&A activity and innovation performance.

R&D expenditures have been collected through COMPUSTAT, Orbis, Osiris, annual reports and SEC fillings. The number of patent applications was obtained through the European Patent Register, hold by the European Patent Office (EPO). All money-related observations are measured in U.S. dollars to standardize the panel data sample. Concerning the European M&A deals historical exchange rates have been used to convert European currencies (EUR, GBP, PLN, CHF) to U.S. dollars.

The independent variable, which assesses the relationship between both R&D- and patent intensity and M&A activity, is the size of the involved companies. To estimate firm size, the number of employees is measured for each time period and is conducted by adding the observations for both companies. To make the positively skewed distribution of the number of employees more normal, the natural logarithm of the observations is taken.

Besides the included independent variable there may be other factors that have an impact on innovation performance of the pharmaceutical companies in the sample. A fixed effects regression will be performed to control for these omitted variables, considering M&A fixed effects and time fixed effects. M&A fixed effects will control for M&A specific effects that change over time, but are constant across the M&A deals. Time fixed effects will control for changes that affect all the M&A deals at the same time.

(17)

14

3.2 Empirical model

To test the aforementioned hypotheses a multiple ordinary least squares (OLS) regression will be executed. Four models are being analysed, depending on R&D intensity, patent intensity, R&D expenditures and patent applications. The empirical analysis focuses on the relationship of R&D intensity and patent intensity between the M&A activity dummy variable, explained by the influence of firm size. M&A fixed effects and time fixed effects, where i is the M&A deal index and t the time-index. The multiple OLS regressions will be run with either including only M&A fixed effects or including both M&A fixed effects and time fixed effects.

The fixed effects model assumes that individual specific effects are correlated with the independent variable and thus individual specific effects are being excluded from the regression (Wooldridge, 2013). Therefore, in this study it is assumed that the occurrence of M&A activity is uncorrelated with the controlled variables by the fixed effects. Examples of unobserved M&A specific effects can be M&A-specific characteristics as being domestic or foreign and the fact of being a merger or acquisition. Considering R&D intensity as the dependent variable, the following regression model will be applicable, when including both M&A fixed effects (µ_𝑖𝑖) and time fixed effects (_𝛾𝛾_𝑡𝑡):

𝑅𝑅&𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝑀𝑀&𝐴𝐴𝑖𝑖𝑡𝑡+ 𝛽𝛽2𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝑖𝑖𝑡𝑡+ µ𝑖𝑖 + 𝛾𝛾𝑡𝑡+ 𝜀𝜀𝑖𝑖𝑡𝑡

Considering patent intensity as the dependent variable, the following model will be applicable, when including both fixed effects:

𝑙𝑙𝐷𝐷 (𝑃𝑃𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝑀𝑀&𝐴𝐴𝑖𝑖𝑡𝑡+ 𝛽𝛽2𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝑖𝑖𝑡𝑡+ µ𝑖𝑖+ 𝛾𝛾𝑡𝑡+ 𝜀𝜀𝑖𝑖𝑡𝑡

𝑅𝑅&𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖𝑡𝑡 represents the ratio of annual R&D expenditures over annual sales of M&A deal i at time t, 𝑀𝑀&𝐴𝐴𝑖𝑖𝑡𝑡 is a dummy variable which equals 0 in the pre-M&A period and equals 1 in the post-M&A period and _{𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷}_{𝑖𝑖𝑡𝑡} denotes the natural logarithm of the sum of the number of employees for both firms for M&A deal i at time t. 𝑙𝑙𝐷𝐷 (𝑃𝑃𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 denotes the natural logarithm of the ratio of annual patent applications over annual sales of M&A deal i at time t. Due to positively skewed distributions of patent intensity, R&D expenditures and patent applications, these variables are transformed into natural logarithms.

(18)

15

The two separate regressions with R&D expenditures and patent applications as the dependent variables are described by the following regression models, when both M&A fixed effects and time fixed effects are included:

𝑙𝑙𝐷𝐷(𝑅𝑅&𝐷𝐷 𝐷𝐷𝑒𝑒𝑒𝑒𝐷𝐷𝐷𝐷𝑒𝑒𝐷𝐷𝐷𝐷𝑒𝑒𝑒𝑒𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝑀𝑀&𝐴𝐴𝑖𝑖𝑡𝑡+ 𝛽𝛽2𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝑖𝑖𝑡𝑡+ µ𝑖𝑖+ 𝛾𝛾𝑡𝑡+ 𝜀𝜀𝑖𝑖𝑡𝑡

𝑙𝑙𝐷𝐷 (𝑒𝑒𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝑀𝑀&𝐴𝐴𝑖𝑖𝑡𝑡+ 𝛽𝛽2𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝑖𝑖𝑡𝑡+ µ𝑖𝑖+ 𝛾𝛾𝑡𝑡+ 𝜀𝜀𝑖𝑖𝑡𝑡

ln(𝑅𝑅&𝐷𝐷𝐷𝐷𝑒𝑒𝑒𝑒𝐷𝐷𝐷𝐷𝑒𝑒𝐷𝐷𝐷𝐷𝑒𝑒𝑒𝑒𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 is the sum of the annual R&D expenditures of both companies for M&A deal i at time t and 𝑙𝑙𝐷𝐷 (𝑒𝑒𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷)𝑖𝑖𝑡𝑡 is the sum of the number of patent applications filed by both companies for M&A deal i at time t.

The coefficient _𝛽𝛽₁ can be interpreted as the change in the four dependent variables starting the year of M&A activity. The coefficient _𝛽𝛽2 can be interpreted as the change in the four dependent variables influenced by the size of the firms. Lastly, the estimated coefficients of the time effect dummies can be interpreted as the change in the dependent variable in each year.

3.3 Threats to the OLS regression

Several conditions must be met to predict unbiased, valid and consistent coefficient estimates (Wooldridge, 2013). Concerning this panel data study two threats will be discussed: heteroskedasticity and multicollinearity. Heteroskedasticity fails the assumption for an OLS regression that the variance of the unobserved error, conditional on the explanatory variable, is constant. To make sure heteroskedasticity does not take place the heteroskedasticity-robust standard errors will be used. Multicollinearity means that two or more independent variables are highly correlated (Wooldridge, 2013). When multicollinearity was found the specific variables are being omitted from the regression, which can be seen in the regression outputs (Table 5 – Table 12). The omitted fixed effects dummies and time effect dummies are used as a reference value.

(19)

16

4. Results & Analysis

In this section the results from the empirical study will be discussed. The outcomes of the regressions are being compared and it will be determined whether an effect of M&A activity on innovation performance can be found and if so what relationship exists. To determine the relationship between the four dependent variables and M&A activity multiple OLS regressions were run.

4.1 Results R&D intensity model

The regression with R&D intensity as the dependent variable was run which resulted in Model 1 and Model 2 (Table 1). Comparing the adjusted R-squared of both models, Model 1 has the highest adjusted R-squared (0.5792) and therefore shows the best fit for the data in the R&D intensity model, when only M&A fixed effects are included. Time effects appear not to improve the R&D intensity model, because of the lower adjusted R-squared (0.5792) of Model 1 compared to Model 2, although the difference is small. This can be confirmed when testing the significance of the time effect dummies included in Model 2 by using the Wald test (Appendix 3). The null hypothesis that the coefficients for all years are jointly equal to zero has not been rejected, which means there is no significant difference in the coefficients of the time fixed effect dummies. This indicates no time fixed effects are needed in Model 2 to explain the effect of M&A activity on R&D intensity. The coefficient of the M&A activity dummy variable (0.7963046) indicates that M&A activity has a positive effect on R&D intensity. However, the results show that the p-value of the coefficient is not statistically significant, which makes the relationship between R&D intensity and M&A activity not very robust. The coefficient of the dummy variable denoting the size of the involved companies (-3.282675) in Model 2 indicates a negative relationship between firm size and R&D intensity. Again, however, the p-value of the coefficient is not statistically significant, so it remains inconclusive concerning a possible effect. Overall, no significant result of M&A activity on R&D intensity can be concluded for this panel data sample.

(20)

17

4.2 Results R&D expenses model

Running the R&D expenses regression results in Model 3 and Model 4 (Table 2). By comparing these two models of the R&D expenses regression, Model 4 seems to have the best fit by having the highest adjusted squared (0.9504) compared to the adjusted R-squared of Model 3 (0.9443). In this case, time effects improve the quality of the R&D expenses model due to a higher percentage of the variation in R&D expenses that is explained. This can be confirmed by testing the significance of the time effect dummies, where the null hypothesis of the coefficients of all years jointly being zero is rejected (Appendix 3). The time fixed effect dummies for the years 1995 until 2003 are significant at either a 1% level or 5% level with a negative coefficient. The negative values of the time effect dummies found expresses a decrease in R&D expenses in every year. The coefficient of the M&A activity dummy variable (-0.2991973) in Model 4 indicates a negative relationship between the R&D expenses and M&A activity. This is the case, because the coefficient of the dummy variable indicating M&A activity is negative and the coefficient is significant at a 5% level. The relationship between the logged R&D expenses variable and the dummy variable can be interpreted as after the M&A deal, their R&D expenses, on the average, decreased with 25.85%. This percentage is determined by applying the following formula: _𝐷𝐷(𝛽𝛽1−1) = percentage change in the logged

dependent variable, on average. The coefficient of the firm size variable (0.4441077) indicates a positive effect of the firm size on its R&D expenditures with a significance level of 1%. The positive relationship indicates that a larger firm, measured by a greater amount of employees, would lead to an increase in R&D expenditures after M&A activity. This contradicts with the findings in Model 1, where the coefficient of the dummy variable indicates a negative effect. However, the result in Model 1 was insignificant and thus still can be seen as inconclusive.

Overall, there seems to be evidence in the R&D expenses model for a negative relationship of M&A activity on R&D expenses. Furthermore, a positive influence can be found between the size of a firm and the M&A activity.

(21)

18

4.3. Results patent intensity model

Model 6 seems to be the model with the best fit in the patent intensity model due to the higher adjusted R-squared (0.8679) compared to the adjusted R-squared (0.8490) of Model 5 (Table 3). The time effects seem to improve the quality of the patent intensity model, which can be confirmed by testing the significance of the time effect dummies jointly (Appendix 3). The coefficient of the M&A activity dummy variable (0.0347871) indicates a very small and positive effect of M&A activity on patent intensity. The coefficient of the firm size variable (-0.1538047) indicates a negative effect of firm size on M&A activity. The signs of the two independent variables in Model 6 show a similarity to Model 1 and a contradiction to Model 4. The fact that Model 6 gives insignificant results means no distinct conclusion yet can be made on the effect of M&A activity on patent intensity.

4.4 Results patent applications model

The regression of the patent applications model resulted in Model 7 and Model 8 (Table 4). Model 8, including both M&A fixed effects and time fixed effects, seems to have the best fit for the patent applications model due to the higher adjusted R-squared (0.8466) compared to the adjusted R-squared of Model 7 (0.8359). Time fixed effects are improving the quality of the model, which is also proven by testing the significance of the time fixed effects for the patent applications model (Appendix 3). Only the time effect dummies in the years 2006 until 2008 are not significant. Continuing with Model 8 as the best fit for the patent applications model the coefficient of the M&A activity dummy variable (0.0474244) indicates a small and positive effect of M&A activity on patent applications of the firms. This is similar to the result concerning the M&A dummy variable of Model 1 and Model 6, but in contradiction with Model 4. The coefficient of the firm size variable (0.1622984) indicates a small and positive effect of firm size on M&A activity. The two independent variables, however, are insignificant and cannot be used directly to make a distinct conclusion about the M&A activity effect.

(22)

19

5. Discussion

Before taking final conclusions about the results from the four regressions, each regression must be evaluated by its results and the significance or insignificance of the results must be discussed.

Looking at the results of the four regression models, several findings can be evaluated. Each model was run including only M&A fixed effects and also with both including M&A fixed effects and time fixed effects. For the R&D expenses-, patent intensity- and patent applications model, when time effects were included, the model reached a better fit for the data. For the R&D intensity model the adjusted R-squared was lower, when time effects were added, but the difference is small. An explanation for the insignificance of the time effect dummies for the R&D intensity model specifically is hard to find. However, the significance of the time effect dummies in the other three models may be explained. A possible explanation could be the crisis in 2007-2008 and a resulting possible lag, which may have affected the efficiency condition of firms and thus their R&D process efficiency.

For the four different regressions with each dependent variable representing a part of the innovation performance, only the R&D expenses regression gave significant results, namely Model 4 where both M&A fixed effects and time fixed effects were included, which showed the best fit for the data used in the R&D expenses regression. The results of the other three regressions were insignificant, which implies the used data set was not sufficient in size. Considering the size of the data set used, this could indicate a very small effect of M&A activity on the examined parameters. Therefore, insignificant results will be evaluated and compared to earlier similar research, besides the evaluation of significant results.

For the R&D intensity regression, Model 1 seemed to have the best fit for the data sample used. The results of the regression output showed insignificant results for the M&A activity dummy variable and firm size variable. The positive sign of the coefficient on the M&A activity dummy variable and the negative coefficient on the firm size would indicate a positive and negative effect on R&D intensity, respectively. The insignificance, however, could mean that there is no relationship or a minor effect of M&A activity and firm size on R&D intensity. Earlier empirical research done by Desylas & Hughes (2010) also found a neutral effect first of M&A activity on R&D intensity for the time period

(23)

20

1984-1998, which did have a significant result. Comparing this study to the one of Desylas & Hughes, differences can be found in the time period taken, the industry specification and the data set used. The M&A sample of 2624 deals used by Desylas & Hughes, in particular, may explain the insignificant result in this study.

For the R&D expenses regression, Model 4 had the highest fit for the data sample. The significant results of Model 4 indicate a negative effect of M&A activity on R&D expenses and a positive effect of firm size on R&D expenses. First, Hypothesis 2 expected a negative effect of firm size on innovation performance. However, this hypothesis cannot be used to compare this result, because the hypothesis concerns the total innovation performance. An explanation for this effect on R&D expenses would be the economies of scope and scale a large firm reaches. The availability of more cash flow for investment for larger firms results in a higher R&D investment compared to smaller firms. Hypothesis 2 expected a negative effect of firm size on R&D inputs and R&D outputs. The result of Model 4 gave evidence for a positive impact of firm size on R&D inputs solely. Secondly, Model 4 showed a negative effect of M&A activity on R&D expenses. Hypothesis 1a expected a negative effect of M&A activity on R&D intensity. This hypothesis cannot be sued for a comparison, as it describes an effect on R&D intensity, but the theory behind it can explain the result found. As mentioned in the theoretical framework, M&A activity can affect managerial commitment to the innovation process. A possible effect could be that, in this data sample, the managerial commitment decreased and less R&D investments were incurred in the post-M&A period. Comparing this result to earlier research, Danzon et al. (2008) found a slower growth in R&D expenses, which indicates a similar result.

For the patent intensity regression, Model 6 had the best fit for the data sample in the patent intensity regression. The estimated coefficients would indicate a positive effect of M&A activity on patent intensity and a negative effect of firm size on patent intensity. Due to the insignificance of these results, comparable empirical studies must be found and similarities discussed. Several studies, Hitt et al. (1991), Hitt et al. (1996) and McCarthy et al. (2012), discussed earlier have found a significant negative effect of M&A activity on both R&D intensity and patent intensity. Comparing the sample size of these studies which are 191, 250 and 121 M&A deals, respectively, with a sample size of 47 M&A deals in this study, indicates that the insignificant results in this study are the result of an insufficient size of the data set used. Furthermore, the negative effect found in the studies mentioned above, coincides with the significant negative effect of M&A activity

(24)

21

on R&D expenses found in this study, assuming a positive relation between R&D expenses and patent intensity. Based on the literature and other findings in this paper, one would expect to be able to verify hypothesis 1c by future research with a sample size of at least 120 M&A deals.

For the patent applications regression, Model 8 had the best fit for the data sample in the patent intensity regression. The estimated coefficients would indicate a positive relation of both M&A activity and firm size on patent applications in the model. Due to the insignificance of the results, the result needs to be compared to similar research and their findings. Studies made by Hitt et al. (1991), Hitt et al. (1996) and McCarthy (2012) found a negative effect on patent intensity. No comparison, however, can be made with these findings and the positive effect found in this model.

(25)

22

6. Summary & Conclusion

The pharmaceutical industry remains an industry with ongoing challenges, facing a conflict of interest between the stakeholders’ business goals and the social, economic and medical needs. The productivity decline, triggered in the 1990’s, made the challenges fiercer. Rising R&D expenditures, due to new technologies and more complex innovative procedures and decreasing NCEs launched, created a change to the structure of the industry. It went from a stable and unconcentrated industry to a concentrated industry as a response to the fear of failure. Part of this change was the M&A activity, which occurred between pharmaceutical companies. While this M&A activity often is projected as an increase in overall company value, the effect of M&A on innovation performance remains ambiguous.

To address this ambiguity this study examined the relationship between M&A activity and innovation performance with a data sample of 47 M&A deals in the European and American pharmaceutical industry for the time period 1995-2010. Innovation performance was defined by R&D intensity and patent intensity. In total four models were regressed, where R&D expenses and patent applications were added to complete the innovation performance analysis.

A literature review was conducted to evaluate theory and make expectations about this study. Several motives exist for a pharmaceutical firm to engage in M&A, categorized in adaptive and proactive motives. Improving innovation performance remains a critical motive. It was hypothesized that M&A activity had a negative effect on the innovation performance in the post-M&A period. Additionally, the influence of firm size on innovation performance was being examined. Based on theoretical arguments and empirical studies, it was hypothesized, that firm size had a negative effect on innovation performance.

The four regression outputs showed many insignificant results. By comparing the insignificant results of each model with earlier research, it often seems the case that the sample size in this study is too small to reach a significant result. This is one crucial factor for performing further research on this specific effect. Only the R&D expenses model showed a significant result. The negative effect of M&A activity on R&D expenses and the positive effect of firm size on R&D expenses were evaluated and explained. This result can be taken for further research on M&A effects on R&D inputs. Summarizing,

(26)

23

the main results of this study were a negative effect of M&A activity on R&D expenses and a positive effect of firm size on innovation performance for a sample of 47 M&A deals in the European and American pharmaceutical industry.

Improvements can be made to the model, which was used to examine the relationship between M&A and innovation performance. The fixed effects panel data model assumes that individual specific effects are correlated with the independent variable and thus are being excluded from the regression. By using another type of model for the regression, individual effects can be included, which may give other interesting views on this topic.

At last, the term ‘innovation performance’ has been defined by measuring R&D intensity and patent intensity. Although this is a well-used definition, it could be the case that certain aspects of the innovation process were not included. Besides making assumptions on innovation performance, measuring a firm’s R&D inputs and R&D outputs has several difficulties. This is the case, mainly caused by the lags of innovation inputs on innovation outputs. When further research is being executed, longer lags of innovation should be taken in consideration.

(27)

24

Bibliography

Archibugi, D. (1988). “In Search of a Useful Measure of Technological Innovation,” Technological Forecasting and Social Change 34, 253-277

Basberg, B.L. (1987). “Patents and the Measurement of Technological Change: A Survey of the Literature,” Research Policy 16, 131-141

Burns, L.R., Nicholson, S., Evans, J. (2005). ‘Mergers, acquisitions and the advantages of scale in the pharmaceutical industry’ in: Burns, L.R. (ed.), The Business of Healthcare, Cambridge: Cambridge University Press, 223-270

Cohen, W.M., Levin, R.C. (1989). Empirical studies of innovation and market structure. In Schmalensee, R., Willig, R. (Eds). Handbook of Industrial Organization (2). Amsterdam, The Netherlands: North-Holland

Chesbrough, H. (2003). Open Innovation: The New Imperative for Creating and Profiting

from Technology. Harvard Business School Press: Boston, MA

Danzon, P., Epstein, A. and Nicholson, S. (2007). ‘Mergers and acquisitions in the pharmaceutical and biotech industries’, Managerial and Decision Economics 28 (4/5), 307-28

Desyllas, P., Hughes, A., (2010). Do high technology acquirers become more innovative. Research Policy, 39 (8), 1105-1112

DiMasi, J.A., Hansen, R.W., Grabowski, H.G. (2003). ‘The price of innovation: new estimates of drug development costs’, Journal of Health Economics 22 (1), 141-85.

Dukes, G., (2006). The law and ethics of the pharmaceutical industry. Amsterdam: Elsevier.

(28)

25

Efpia.eu. (2013). EFPIA – The Pharmaceutical Industry in Figures 2013 [online] Available at: http://www.efpia.eu/uploads/Figures_Key_Data_2013.pdf [Accessed: 24 April 2015]

Gassmann, O., Reepmeyer, G., von Zedtwitz, M. (2004). Leading Pharmaceutical

Innovation: Trends and Drivers for Growth in the Pharmaceutical Industry.

Springer: Heidelberg.

Grabowksi, H., Kyle, M. (2007) Mergers and alliances in pharmaceuticals: Effects on innovation and productivity. In Gugler, K. and Yurtaglu, B.B. (eds.) The

economics of corporate governance and mergers. Cheltenham, UK: Edgar Elgar.

Hagedoorn, J., Cloodt, M. (2003). Measuring innovative performance: is there an advantage in using multiple indicators? Research Policy 32 (8), 1365-1379.

Hitt, M., Hoskisson, R., Johnson, R. A., Moesel, D. D. (1996). The market for corporate control and firm innovation. Academy of Management Journal 39 (5), 1084-1119.

Hitt, M., Ireland, R., Harrison, J. and Hoskisson, R. (1991). Effect of acquisitions on R&D inputs and outputs. Academy of Management Journal, 34 (3), 693-706

Koenig, M.E.D., Mezick, E.M. (2004). “Impact of Mergers & Acquisitions on research productivity within the pharmaceutical industry,” Scientometrics 59 (1), 157-169

LaMattina, J.L. (2011). “The Impact of Mergers on Pharmaceutical R&D,” Nature

Review Drug Discovery 10, 559-560

McCarthy, K.J., Dolfsma, W.A. & Fijlstra, J. (2012). Fusies en overnames in de farmaceutische industrie in: Economische Statistische Berichten 97 (4634), 269-270

Munos, B. (2009). Lessons from 60 years of pharmaceutical innovation. Nature Review Drug Discovery 8, 959-968

(29)

26

Northrup, L.R., The Pharmaceutical Sector in: The Business of Healthcare Innovation, ed. 2005 Cambridge Univ. Press.

Pakes, A., Grichiles, Z. (1980). Patents and R&D at the firm level: A first report. Economic Letters 5, 377-381

Schumpeter, J.A. (1942). Capitalism, socialism and democracy. New York: Harper.

Walker, S.R. (1997). Global Responses: The Search for Cures in the Development of Pharmaceuticals: Indiana Journal of Global Legal Studies 5 (65), 65-83

Wooldridge, J.M. (2006). Introductory econometrics: A modern approach. Mason, OH: Thomson/South-Western

(30)

27

Tables

Table 1: Results regression analysis with the dependent variable R&D intensity1

1 _{R&D intensity is measured as annual R&D expenses divided by annual sales. M&A activity is a dummy variable, which equals 0 for}

the pre-M&A period and 1 for the post-M&A period. Firm size is measured by the number of employees of both firms. The robust-standard error of a variable is given between brackets.

* p < 0.10 ** p < 0.05 *** p < 0.01

Table 2: Results regression analysis with the dependent variable R&D expenses2

2 _{R&D expenses are measured in units of U.S. dollars. M&A activity is a dummy variable, which equals 0 in the pre-M&A period and}

1 in the post-M&A period. Firm size is measured by the number of employees of both firms. The robust-standard error of a variable is given between brackets.

* p < 0.10 ** p < 0.05 *** p < 0.01

R&D intensity Model 1 Model 2

M&A fixed effect Yes Yes

Time fixed effect No Yes

Constant 30.20167 (28.49113) 44.96095 (37.52029) M&A activity 0.7963046 (2.310221) 0.5200579 (3.310946) Size -3.282675 (3.515611) -5.02072 (4.456252) R2 _0.6296 _0.6455 Adjusted R2 _0.5792 _0.5769 N 402 402

ln (R&D expenses) Model 3 Model 4

Constant 14.73467*** (1.185726) 16.58527*** (1.259716) M&A activity 0.2296747** (0.0950299) -0.2991973** (0.1332747) Size 0.5537796*** (0.147349) 0.4441077*** (0.1477757) R2 _0.9512 _0.9587 Adjusted R2 _0.9443 _0.9504 N 402 402

(31)

28

Table 3: Results regression analysis with the dependent variable patent intensity3

3 _{Patent intensity is measured as annual patent applications divided by annual sales. M&A activity is a dummy variable, which equals}

0 in the pre-M&A period and 1 in the post-M&A period. Firm size is measured by the number of employees of both firms. The robust-standard error of a variable is given between brackets.

* p < 0.10 ** p < 0.05 *** p < 0.01

Table 4: Results regression analysis with the dependent variable patent applications4

4 _{Patent applications are measured in units of the number of applications per year. M&A activity is a dummy variable, which equals 0}

in the pre-M&A period and 1 in the post-M&A period. Firm size is measured by the number of employees of both firms. The robust-standard error of a variable is given between brackets.

* p < 0.10 ** p < 0.05 *** p < 0.01

ln (Patent intensity) Model 5 Model 6

Constant -16.53961*** (1.749349) -18.61381*** (1.909397) M&A activity -1.039559*** (0.1627682) 0.0347871 (0.2170934) Size -0.0983364 (0.21224) -0.1538047 (0.2204125) R2 _0.8709 _0.8937 Adjusted R2 _0.8490 _0.8679 N 402 402

ln (Patents) Model 7 Model 8

Constant 1.151859 (0.8736326) 0.5083326 (0.8905855) M&A activity -0.4058882*** (0.1294828) 0.0474244 (0.154662) Size 0.2197218** (0.1044914) 0.1622984 (0.1012362) R2 _0.8594 _0.8764 Adjusted R2 _0.8359 _0.8466 N 402 402

(32)

29

Table 5: Regression output R&D intensity model without time effects (Model 1) Fixed-effects (within) regression Number of obs = 402 Group variable: ma Number of groups = 47

R-sq: within = 0.0133 Obs per group: min = 5 between = 0.0005 avg = 8.6 overall = 0.0000 max = 9 F(2,46) = 0.74 corr(u_i, Xb) = -0.4555 Prob > F = 0.4805 (Std. Err. adjusted for 47 clusters in ma) --- | Robust

rho | .67034957 (fraction of variance due to u_i)

---

Table 6: Regression output R&D intensity model with time effects (Model 2) Fixed-effects (within) regression Number of obs = 402 Group variable: ma Number of groups = 47 R-sq: within = 0.0556 Obs per group: min = 5 between = 0.0007 avg = 8.6 overall = 0.0005 max = 9 F(18,46) = . corr(u_i, Xb) = -0.6077 Prob > F = . (Std. Err. adjusted for 47 clusters in ma) --- | Robust

rdintens | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- dummyma | .5200579 3.310946 0.16 0.876 -6.144531 7.184647 lnsize | -5.02072 4.456252 -1.13 0.266 -13.99069 3.94925 t1 | -5.989697 8.231831 -0.73 0.471 -22.55951 10.58012 t2 | -7.473206 8.036096 -0.93 0.357 -23.64903 8.702616 t3 | -5.618734 7.134295 -0.79 0.435 -19.97932 8.741856 t4 | -2.997496 5.821717 -0.51 0.609 -14.716 8.721012 t5 | -.5233929 5.258495 -0.10 0.921 -11.1082 10.06141 t6 | -1.579199 4.313125 -0.37 0.716 -10.26107 7.102672 t7 | -.6295534 3.512966 -0.18 0.859 -7.700786 6.441679 t8 | -1.297485 2.894588 -0.45 0.656 -7.123989 4.529019 t9 | -3.253087 3.756225 -0.87 0.391 -10.81397 4.307801 t10 | 6.250245 6.275842 1.00 0.324 -6.382369 18.88286 t11 | 1.671531 3.597351 0.46 0.644 -5.569562 8.912624 t12 | 1.209031 3.88576 0.31 0.757 -6.612599 9.03066 t13 | -1.222086 3.488204 -0.35 0.728 -8.243476 5.799303 t14 | -1.847959 3.496112 -0.53 0.600 -8.885267 5.189348 t15 | -2.343952 3.467162 -0.68 0.502 -9.322986 4.635083 t16 | -4.187129 2.967542 -1.41 0.165 -10.16048 1.786224 t17 | -3.090352 2.245613 -1.38 0.175 -7.610537 1.429833 t18 | 0 (omitted) _cons | 44.96095 37.52029 1.20 0.237 -30.56348 120.4854 ---+--- sigma_u | 22.63736 sigma_e | 14.08777

(33)

30

Table 7: Regression output R&D expenses model without time effects (Model 3) Fixed-effects (within) regression Number of obs = 389 Group variable: ma Number of groups = 47 R-sq: within = 0.2028 Obs per group: min = 4 between = 0.7467 avg = 8.3 overall = 0.7556 max = 9 F(2,46) = 8.96 corr(u_i, Xb) = 0.5314 Prob > F = 0.0005 (Std. Err. adjusted for 47 clusters in ma) --- | Robust

---

Table 8: Regression output R&D expenses model with time effects (Model 4) Fixed-effects (within) regression Number of obs = 389 Group variable: ma Number of groups = 47 R-sq: within = 0.3250 Obs per group: min = 4 between = 0.7079 avg = 8.3 overall = 0.7296 max = 9 F(18,46) = . corr(u_i, Xb) = 0.5995 Prob > F = . (Std. Err. adjusted for 47 clusters in ma) --- | Robust

lnrdexpens | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- dummyma | -.2991973 .1332747 -2.24 0.030 -.5674652 -.0309293 lnsize | .4441077 .1477757 3.01 0.004 .1466507 .7415647 t1 | -1.803363 .4288577 -4.21 0.000 -2.666609 -.9401175 t2 | -1.589076 .4289997 -3.70 0.001 -2.452608 -.7255446 t3 | -1.987638 .5756334 -3.45 0.001 -3.146328 -.8289482 t4 | -1.605676 .3960461 -4.05 0.000 -2.402875 -.8084763 t5 | -1.51506 .3662004 -4.14 0.000 -2.252183 -.7779364 t6 | -1.145537 .2943134 -3.89 0.000 -1.737959 -.5531148 t7 | -.9250919 .2860603 -3.23 0.002 -1.500901 -.3492824 t8 | -.9022864 .2664655 -3.39 0.001 -1.438654 -.3659191 t9 | -.7448017 .3297946 -2.26 0.029 -1.408644 -.0809596 t10 | -.575602 .2915546 -1.97 0.054 -1.162471 .011267 t11 | -.4479602 .2482562 -1.80 0.078 -.947674 .0517536 t12 | -.293185 .2143454 -1.37 0.178 -.7246398 .1382699 t13 | -.1303557 .1828282 -0.71 0.479 -.4983698 .2376585 t14 | -.0409422 .1661111 -0.25 0.806 -.3753064 .2934221 t15 | -.199113 .1140842 -1.75 0.088 -.4287526 .0305266 t16 | -.3521538 .1721521 -2.05 0.047 -.6986781 -.0056295 t17 | -.2336612 .1205302 -1.94 0.059 -.4762759 .0089535 t18 | 0 (omitted) _cons | 16.58527 1.259716 13.17 0.000 14.0496 19.12095 ---+--- sigma_u | 1.7058098 sigma_e | .57196103

(34)

31

Table 9: Regression output patent intensity model without time effects (Model 5) Fixed-effects (within) regression Number of obs = 333 Group variable: ma Number of groups = 47 R-sq: within = 0.2387 Obs per group: min = 2 between = 0.3488 avg = 7.1 overall = 0.2473 max = 9 F(2,46) = 20.71 corr(u_i, Xb) = 0.3067 Prob > F = 0.0000 (Std. Err. adjusted for 47 clusters in ma) --- | Robust

---Table 10: Regression output patent intensity model with time effects (Model 6) Fixed-effects (within) regression Number of obs = 333 Group variable: ma Number of groups = 47 R-sq: within = 0.3736 Obs per group: min = 2 between = 0.0888 avg = 7.1 overall = 0.1727 max = 9 F(18,46) = . corr(u_i, Xb) = -0.0371 Prob > F = . (Std. Err. adjusted for 47 clusters in ma) --- | Robust

lnpatentin~s | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- dummyma | .0347871 .2170934 0.16 0.873 -.4021993 .4717735 lnsize | -.1538047 .2204125 -0.70 0.489 -.5974719 .2898626 t1 | 3.934722 .7227441 5.44 0.000 2.479914 5.389531 t2 | 3.627081 .7555426 4.80 0.000 2.106253 5.14791 t3 | 3.147413 .7641841 4.12 0.000 1.60919 4.685636 t4 | 3.205521 .6676245 4.80 0.000 1.861663 4.54938 t5 | 3.464781 .6354943 5.45 0.000 2.185598 4.743965 t6 | 3.075955 .6141078 5.01 0.000 1.83982 4.312089 t7 | 2.608222 .5671745 4.60 0.000 1.466559 3.749885 t8 | 2.512236 .5542428 4.53 0.000 1.396603 3.627869 t9 | 2.31811 .5870294 3.95 0.000 1.136481 3.499739 t10 | 2.223989 .5102989 4.36 0.000 1.19681 3.251168 t11 | 1.799342 .4709717 3.82 0.000 .8513255 2.747359 t12 | 1.243343 .4789432 2.60 0.013 .27928 2.207405 t13 | 1.070437 .4736282 2.26 0.029 .1170725 2.023801 t14 | .1986508 .5777212 0.34 0.733 -.9642417 1.361543 t15 | .3360807 .4025611 0.83 0.408 -.4742327 1.146394 t16 | .52915 .3060199 1.73 0.090 -.0868362 1.145136 t17 | .6478788 .2768053 2.34 0.024 .0906987 1.205059 t18 | 0 (omitted) _cons | -18.61381 1.909397 -9.75 0.000 -22.45723 -14.77039 ---+--- sigma_u | 2.1730626 sigma_e | .90907751