Master’s Thesis in Business Administration - track Strategic Innovation Management
BEST VERSUS REST PATTERNS IN R&D INVESTMENTS
Candidate: Riccardo Brusori
S3770354
Supervisor: Dr. Pere Arqué-Castells
Co-assessor: Dr. Pasi Kuusela
Groningen, 21 June 2020
Abstract
1. Introduction
There is growing evidence that nowadays’ economy is pervaded by an increasing disparity across firms, with serious repercussions on market structure and innovation activities. In recent years, economic scholars started to investigate some phenomena that can primarily explain the occurring shift, such as the decline of the labor share, growing productivity dispersion within firms and the rise of market power among certain established firms (Autor, Dorn, Katz, Patterson & Van Reenen, 2020; Loecker & Eeckhout, 2018; Loecker, Eeckhout & Unger, 2020; Berlinghieri, Criscuolo & Blanchenay, 2017). These trends are growing and reinforcing over time, especially in Western economies, creating a widespread concern that structural imbalances in the competitive environment can drive a wedge between a restricted circle of firms, commonly labelled as “superstars”, and the rest. As a result, growing inequality across firms is creating winners and laggards, or a best vs rest pattern in the economy (Arqué-Castells, 2020). When it comes to analyze innovation activities and their consequences on inequality, attention has been mainly addressed to innovation outputs. For instance, the OECD recently reported that the top 2000 R&D investors worldwide were responsible for 64% of the patent families issued at the five largest intellectual property offices in 2014 (OECD, 2017). Academic literature has put scant attention on the distribution of innovation inputs- such as R&D expenditures- among companies and their evolution over time. The aim of this paper is to fill this gap by performing a fine-grained analysis of the distribution of R&D investments across different samples of companies, showing their evolution. Since inequality across firms is rising, it is important to verify if an increasing dispersion of R&D between top investors and the rest of companies is also occurring. This constitutes the research goal of this paper. Several reasons provide a solid foundation for carrying out this analysis.
(Appelt, Galindo-Rueda & González Cabral, 2019), and it is important to verify their impact, if any, on the distribution of R&D across firms.
This paper will not address causality, but it aims to provide stylized facts regarding the existence of a best vs rest pattern in R&D investments. Moreover, it wants to open the debate about possible implications for a firm’s strategy, R&D management and policy makers. This analysis was carried out using three large longitudinal datasets with different characteristics. As expected, this paper shows that R&D is very concentrated in the hands of few companies. Surprisingly, this phenomenon is not recent, and this constitutes an interesting insight. However, since R&D investments are growing and more companies are trying to use R&D for competing worldwide, the repercussions of this high level of concentration are likely more severe nowadays. Moreover, a “superstar effect” is at play for what concerns R&D investments, with few companies able to allocate abnormal and increasing investments over time, with serious consequences for firms aiming to catch up.
This paper is organized as follows: section 2 presents the Literature Review that lays the ground for section 3, Hypothesis Development. Section 4 incorporates all the information regarding the data used in the analysis and section 5 explains the Methodology followed. Section 6 describes the Analysis and Results and section 7 presents the Discussion and the Conclusion. The tables and figures related to the analysis have been placed in a separate section after the References in order to improve the readability of the paper. A separate Appendix provides several tables with annual data for every dataset analyzed.
2. Literature Review
Growing inequality across firms and the rise of “superstars”
Some important changes in the competitive environment are tied to the rise of “superstar firms, and authors investigating them have not reached a common agreement about their causes, although their occurrence is evident.
Loecker et al. (2020) suggest that the fall of the labor share depends on the rise of markups among large firms.
At the same time, concentration across firms is rising, favored by “winner take most” dynamics in platform markets (Autor et al., 2020). To explain rising concentration, Bessen (2019) focused on IT investments in proprietary software, finding that concentration started to increase from the 1980s when some firms began to commit large IT investments, capturing larger shares of intra-industry sales. For instance, when Walmart implemented its own IT proprietary system for its supply chain, its share of sales jumped from 3% in 1982 to 52% in 2012 (Bessen, 2019).
Productivity dispersion is another important element for explaining growing inequality. Some authors claimed that the productivity gains from increasing investments in intangible assets (Crouzet & Eberly, 2018) or the ICT revolution favored firms with sound technological and managerial capabilities (Van Reenen, 2018). Therefore, it is not IT per se the main contributor of this productivity gap, but how IT is managed (Van Reenen, 2018) or appropriated (Bessen, 2019).
Lastly, market power, traditionally associated with markups’ increase, is another major element for explaining the increasing disparity across firms. Loecker and Eckenhout (2018) studied markups at a global level, depicting their steady rise in the U.S., Europe and Asia, while Calligaris, Criscuolo & Marcolin (2018) showed their remarkable growth in the digital sector. Some scholars considered the rise of markups as driven by few firms with substantially higher markups than a few decades ago (Autor et al., 2020; Loecker et al., 2020).
In sum, several trends in today’s economy are having an impact upon the competitive landscape. The fall of the labor share, rising concentration within sectors, “winner take most” dynamics, productivity dispersion and the rise of markups can offer a comprehensive explanation for the increasing inequality between firms, leading to the rise of superstar firms.
The characteristics of superstar firms
technology is combined with imperfect substitution, an output can be highly concentrated among the most talented earners, labelled as “superstars”.
Data provide clear-cut explanations to understand the superstar effect among businesses. For instance, the share of GDP from the 100 biggest companies in the U.S. rose from 33% in 1994 to 46% in 2013 (The Economist, 2016). Moreover, the McKinsey Global Institute discovered that 10% of the world’s public companies generated 80% of all profits in 2016 (McKinsey Global Institute, 2018). In the academic literature, there is no common agreement on how to measure superstar firms. As a consequence, different authors suggested to look at returns on invested capital (Furman & Orszag, 2015; Ayyagari, Demirguc-Kunt & Maksimovic, 2018), profit shares (McKinsey Global Institute, 2018), productivity or markups (Autor et al., 2020; Loecker et al., 2020).
Autor et al. (2020) found that superstar firms are more productive, more innovative, and are substantially increasing their market share, markups and market power. Moreover, they often experience the “scale without mass” effect when their sales and market value grow abnormally respect to the number of people they employ, or when IT enables the replication of their business processes (Brynjolfsson, McAfee, Sorell & Zhu, 2008). In contrast to the McKinsey report, which found the superstar status contestable due to the high churn rates, Autor et al. (2020) found an opposite stabilizing effect among the top 500 Compustat firms (by sales).
Even though the debate about the importance of superstars in the economy is gaining momentum, some authors stressed that the superstar status should be considered with caution and its economic impact is not always clearly proved (Ayyagari et al., 2018; Traina, 2018).
Why R&D investments matter
Since nowadays’ economy is laying the ground for the rise of superstar firms, it is important to link this topic with R&D investments, which have been validated by scholars as a consistent metrics for studying innovation in private companies (Acs & Audrescht, 1988).
When evaluating scale and scope economies tied to R&D, the general impact seems positive, especially due to their “dynamic efficiency gains” (Henderson & Clockburn, 1996, p. 55). For example, Artz, Norman, Hatfield and Cardinal (2010) found important scale effects of R&D with respect to patents, while Henderson and Clockburn (1996) advocated for larger effects of economies of scope in drug discovery activities. Not all authors agreed upon the existence of these effects, with some showing the existence of diseconomies of scale or scope (see Macher & Boerner, 2006; Zenger, 1994). Therefore, is seems that the impact of economies of scale and scope from R&D is context-dependent.
R&D expenditures also face barriers. First, R&D processes are surrounded by uncertainty, which makes eventual payoffs hard to be predicted (Rosenberg, 1990; Zenger, 1994). Moreover, the vital information about R&D projects are usually embedded in a company’s scientists and researchers, who are responsible for them at the operational level. This provokes an asymmetry inside companies that can lead to adverse selection, choosing inappropriate projects to finance (Hall, 2002; Zenger, 1994). As a result, R&D investments often deal with agency problems and moral hazard (Hall, 2002; Henderson & Clockburn, 1996), mainly due to their long timeframes, while managers tend to focus on short-term gains (Gentry & Shen, 2013; Honoré, Munari & Van Pottelsberghe De La Potterie, 2015). Finally, R&D activities account for large fixed costs (Henderson & Clockburn, 1996), especially for scientists’ salaries (Hall, 2002; Zenger, 1994).
Therefore, while all companies can potentially be affected by the pros of R&D, not all are equally affected by the cons. In fact, due to large fixed costs and uncertainty of these investments, firms need substantial cash flows to finance R&D (Bakker, 2013; Hall, 2002).
3. Hypothesis Development
The role of public policies supporting innovation
& Galindo-Rueda, 2016; Appelt et al., 2019). Nevertheless, governments have different cards to play for sustaining innovation.
Public support to innovation can be direct, through R&D grants, or indirect, through R&D tax incentives, which are conceived as a more market-based instrument (Appel et al., 2016) and gaining popularity, with OECD countries with R&D tax incentive systems in place passing from 19 in 2000 to 30 in 2018 (Appel et al., 2019).
Regarding the support to patent protection, research showed that its effectiveness varies across contexts and the likelihood of patenting depends on the expected costs of infringement (James, Leiblein & Lu, 2013). Therefore, the country’s appropriability regime is a key indicator. Park (2008) listed 122 countries according to their patent protection index, showing that these are notably improving worldwide.
In general terms, R&D policies do not always favor young and small firms. In fact, the more generous the R&D tax credits, the more favored slow-growing incumbents, especially in the absence of ceilings and thresholds when claiming them (Appel et al., 2016). Moreover, much of the government support for R&D is appropriated by larger firms, which would have invested in R&D anyway (Becker, 2015). According to Arqué-Castells (2020), one of the problems in the evaluation of R&D tax incentive system regards the fact that full take-up (i.e. when all the eligible firms claim tax credits) are taken for granted, but large fixed costs in the take-up penalize smaller firms vis-à-vis larger firms. Regarding patents, due to the expected infringement costs, James et al. (2013) concluded that firms are more likely to patent if they already detain a large patent portfolio and are well-equipped to face legal disputes. Consequently, established incumbents can benefit more from both R&D incentives and patent protection.
Globalization and knowledge creation
intensity and productivity among European companies. Moreover, they showed that firms from low-tech sectors, or simply not productive enough, were more likely to be disbanded from the market. As a consequence, businesses need to find alternative ways to compete other than only focusing on costs. Bogner and Bansal (2007) evidenced that two necessary ingredients concern the creation of new internal knowledge, followed by building “new knowledge on new knowledge” and appropriating it. Consequently, the authors consider R&D fundamental to maintain industry leadership. Similarly, Zack (1999) found that building a knowledge base is necessary for organizational learning and knowledge recombination. Most of these arguments are linked to ideas of absorptive capacity (Cohen and Levinthal, 1990) and knowledge-based view of the firm (Grant, 1996). In general, due to structural changes mainly favored by globalization, the nexus between knowledge creation, innovation and competitive advantage is gaining momentum in the last decades.
The “burden of knowledge” and research productivity
Recent research demonstrates that knowledge creation is far from being a simple task. With the term “burden of knowledge”, Jones (2009) highlighted the increasing difficulty in producing new knowledge over time, because innovators have to contend with an increasing stock of knowledge. This “burden” is responsible for pushing the knowledge frontier more distant in several fields, increasing specialization and teamwork.
By the same token, Bloom, Jones, Van Reenen and Webb (2017) debunked the idea that research productivity is constant, as endogenous growth models usually assume. Analyzing research productivity in various industries, they showed its general declining. For instance, analyzing the R&D activities of major semiconductor firms, the authors evidenced that only a consistent increase of research efforts (risen by a factor of 18 from 1971 to 2011) permitted to achieve the exponential growth implied by Moore’s law1.
To sum up, public policies supporting R&D are spreading among countries, and patent protection indices are generally improving worldwide. However, there is a risk that large incumbents would be more
1 Bloom et al. (2017) calculated the research effort as the amount of R&D investments deflated by the average salary of
advantaged by these pro-innovation policies. On the other side, globalization is forcing firms to compete trough innovation, investing in knowledge creation. However, the burden of knowledge is making the generation of innovation more demanding for companies. Finally, positive cash-flow are essential to finance R&D projects (Bakker, 2013; Hall, 2002) and all these effects are likely to have grown over time, due to the rising inequality across firms, as previously shown. Consequently, all these compelling arguments suggest that:
H1a: R&D investments are highly concentrated H1b: R&D concentration has increased over time
When accounting for a general increase in R&D concentration, the role of top R&D investors is likely to be fundamental. As stated by Autor et al. (2020) superstar firms are more innovative than the rest, and market power can facilitate R&D investments (Rosenberg, 1990). Moreover, high profitability can favor top investors in committing more resources to R&D, as the McKinsey Global Institute (2018) recently demonstrated, with the top 10% more profitable firms accounting for an increasingly larger share of R&D expenditures in the last 20 years. For all the above reasons, I expect the existence of a “superstar effect” in R&D investments, with top investors accounting for a large share of R&D and capturing most of the of the total growth in R&D investments. Moreover, I expect this share to have increased overtime, due to the market power that top R&D investors detain, which makes them hardly displaceable.
H2a: Top R&D investors account for a large share of R&D investments, capturing a high proportion of the total R&D growth
H2b: The share of R&D investments by top investors has increased over time H2c: The category of top R&D investors is hardly contestable
4. Data
retrieve information about innovation inputs and outputs at the firm-level (Sofka, de Faria & Shehu, 2018). Moreover, CIS data mainly regard SMEs, constituting a valuable tool to study concentration. However, these data are not suitable for this project, since not every country complies with the survey on a regular basis, impeding cross-country analyses. Additionally, CIS data are only available from 1994. Therefore, other feasible sources were used.
Data sources
Compustat. This database contains various financial information at a granular level for publicly held companies in the United States and Canada. Due to its reliability, it is widely used among scholars. Annual data are available from 1950 and are accessible through the Wharton Research Data Services, University of Pennsylvania. Compustat’s key advantages derive from its popularity, reliability, coverage, and long time series available, making it extremely useful for historical analyses. On the other side, it only contains information about public firms in North America, restricting the sample to traditionally large firms of a specific location.
Orbis. It represents the company database of Bureau van Dijk, a Moody’s Analytics company. It contains data from 365 million companies (both public and non-public) worldwide, retrieving information from more than 160 different sources. As reported by Gal (2013), Orbis enables a comparative cross-country analysis at the firm level. Moreover, the geographical variety of company-level data from Europe, United States and many emerging economies represents a great strength of this dataset. However, the quality of the variables varies from country to country, being notably lower for the U.S. (Gal, 2013). Nevertheless, this aspect was addressed by using Compustat, which covers that specific geographical area with extreme reliability. Finally, data are only available from 2010. Therefore, Orbis database can only provide information about recent dynamics, but not about their historical evolution.
consistency. The JRC-EC COR&DIP project (henceforth COR&DIP) consists of three different waves, with data available for the period 2009-2016. One of the key strengths of this database relates to its reliability, since both the OECD and the European Commission are important public institutions. However, this sample only contains information from the major investors, and it does not allow for comparisons with smaller investors or non-investing firms, as Orbis and Compustat.
Final samples
Since the first years are more likely to contain discrepancies, data on Compustat were collected from 1975 to 2018. The variables of interest were company name, location of the headquarter, standard industrial classification, and R&D expenditures (in $ million). Missing values (65% of total observations) were considered as zero values (i.e. R&D expenditure = 0), since public companies normally have to disclose these data.
The final sample from Orbis included all companies with a known value related to R&D expenditures for at least one year between 2010 and 2018. The variables retrieved included the company name, the country ISO code, the NACE2 industrial classification and the amount of R&D expenditures (in $ million). Missing values in Orbis (25% of the total observations) were discarded because, due to heterogeneity of the dataset, they were likely not available.
The section of interest in the COR&DIP database was the “Top Corporate R&D Investors – Financial”. This section contained the company name, R&D investments, net sales, capital expenditures, operating profits (all in $ millions), number of employees, company ranking (from 1 to 2000 in terms of R&D investments), the ISO2 country code (location of the headquarter), and the NACE2 and ISIC4 industrial classification. All these variables were presented for every year. The three different waves available were merged together, forming a unique dataset for the period 2009-2016. The few missing values (1.5% of the total observations) were discarded from the analysis.
Naturally, some overlapping of same companies’ data for the period the three datasets had in common (2010-2016) occurred. However, due to the different periods considered in the datasets, this issue will not represent a problem for this paper, and it was not addressed during the analysis.
series, being the most representative for studying the historical evolution of concentration in R&D. However, it only displays data for public firms in North America. Orbis perfectly complements this issue by providing data from all over the world, with a limited coverage of American companies. However, its time series is limited (2010-2018), which does not allow for tracing any historical evolution, despite offering a recent picture about a consistent number of heterogenous companies. Finally, COR&DIP can be considered as a sort of control sample for this analysis. It represents an already concentrated sample, with data from top 2000 R&D investors worldwide in recent years. If concentration appears to be high even in this dataset, the magnitude of the concentration problem will be well and truly clear.
Table 1: Information about the datasets
Compustat Orbis COR&DIP
Period analyzed (number of years) 1975-2018 (44) 2010-2018 (9) 2009-2016 (8)
Final sample size in n. of observations 486,838 311,823 15,753
Number of companies 38,734 46,461 2,792
Observations reporting a positive value for R&D investments
(% over total) 128,068 (26.31%) 132,190 (42.39%) 15,753 (100%)
Variables
Due to the use of panel data, all the variables presented below were calculated per year. In addition, the
annual growth rate for every variable is accessible in the Appendix.
Variables created for the first set of hypotheses:
- R&D intensive margins: it concerns the amount of R&D investments out of the total number of firms investing in R&D, as in Arqué-Castells (2018), or, more simply, the average R&D investments per firm.
- R&D per percentile: this variable shows the amount of R&D investments for all the investors within a certain percentile in each dataset. The 1st, 5th, 10th, 25th, 75th, 90th, 95th, 99th percentile and the median were considered. The amount was calculated as the share of R&D, in percentage, over total. The absolute values are accessible in the Appendix.
- Gini coefficient: this index is a common measure of inequality. It relates the cumulative proportion of a population to the cumulative proportion of the value it accounts for. The Gini coefficient is widely adopted for income or wealth distributions and it ranges between 0 and 1. At 0, the distribution is perfectly equally distributed, with every observation accounting for the same value, while a value of 1 indicates maximum inequality, with one observation accounting for the entire value of the distribution. Consequently, the closer to 1, the more concentrated the distribution. R&D investors in every dataset were considered in the analysis of the Gini index. Variables created for the second set of hypotheses:
- Top R&D investors: it accounts for the top 10, 25, 50 and 100 firms in terms of amount of R&D investments per year in each dataset. The numerical selection preserves in large part the differences in size among the three datasets and was made to provide clear stylized facts. The correspondent share of R&D investments by top investors was extrapolated from the total and presented in percentage (absolute values accessible in the Appendix).
- Share of R&D growth captured by top investors: this measure is frequently used regarding inequality2, and it explains how much of the annual total R&D growth in a sample is linked to top investors.
- R&D spells by top investors: as in Higón et al. (2011), it indicates the number of successive years in which a company invests in R&D. I will here compare the R&D spells of top investors and the rest of the companies.
Finally, for both the hypotheses the following variable was created:
- R&D intensive sectors: companies were categorized according to the potential R&D intensity of their sectors. The selection was based on the following two digits SIC industrial classification: 28 (Chemicals, including pharmaceutical); 35 (Machineries, including computers); 36 (Electronic and Electrical Equipment); 37 (Transportation); 38 (Instruments, including medical devices); 73 (Business Services, including R&D); 87 (Engineering). This categorization followed in large part Jones (2007) and Kile and Phillips (2009), with few changes made since not all sectors mentioned by these authors were feasible for the analysis. The Compustat sample presented SIC four digits codes, while Orbis and COR&DIP were based on NACE codes. Therefore, the SIC codes were first transformed in 6 digits NAICS codes (based on the United States Census Bureau website) and then into NACE (2nd revision) based on the RAMON scheme from Eurostat. Only the codes resembling the initial SIC two digits codes were considered for the transformation. The choice of picking two digits SIC codes can attenuate the possible missing of any related codes from NAICS and NACE. However, since both the NAICS and NACE classifications have expanded the range of subcategories respect to SIC, it is likely that newly created subcategories in the 2nd NACE
revision were omitted. However, the revisions that normally apply to the same classification do not include major changes. Therefore, the codes attributed to the R&D intensive sectors in Orbis and COR&DIP are not expected to be largely different from SIC, albeit a slightly conservative estimation could occur. Anyway, the analysis demonstrated that this problem did not bias the results, presenting almost identical results regarding R&D intensive sectors among the three datasets.
Descriptive statistics
Table 2 reports descriptive statistics of the variables used in the study. In general, descriptive statistics of Compustat variables has to be taken with more caution, due to the long period analyzed in this dataset, especially regarding absolute values. Moreover, the average percentage of firms active in R&D sectors per year are here displayed. Interestingly, companies active in the Transportation and Engineering sectors account for a small portion of the total companies presented in the datasets, although this proportion augments in the COR&DIP sample.
year-, interesting insights appear when comparing Orbis and Compustat. The first shows a higher percentage of firms investing less than $ 1 million (38.46%) and less than $ 10 million (75.95%) with respect to Compustat (27.40% and 63.92%, respectively). This gap lowers when considering companies investing more than $ 50 million and $ 100 million.
Finally, for acknowledging the geographical location of the companies, three main variables were created: “North America”, if the company was headquartered in the United States or Canada; “Europe”, if headquartered in a European country, excluding Russia; “Rest of the World”, if headquartered anywhere else. European and American companies prevail in Compustat, while Orbis account for much more companies from the rest of the world. COR&DIP presents a very balanced geographical distribution of the companies.
5. Methodology
Despite the difference in terms of size and composition of the datasets, it was decided to apply the same techniques and procedures to all of them. This would strengthen the results of this study in case the results across the three datasets appear to be similar.
In order to test the first set of hypotheses, three different techniques to measure concentration were applied:
1. Comparison between R&D extensive and intensive margins: this figure compares the evolution of the number of firms investing in R&D with the evolution of the average R&D investments per firm. This analysis does not provide any index or numerical output from the comparison, only showing their respective evolution, as in Higón et al. (2011). The COR&DIP sample presents a fixed number of firms per year, making it impossible to calculate the extensive margins, and this analysis was performed in Compustat and Orbis only. The inflation rate was not accounted for the intensive margins of R&D investments due to different time lags.
2. Share of R&D per percentile: this measure presents the share of R&D investments in terms of percentage over the total amount of R&D performed by all the firms within the 1st, 5th, 10th, 25th, 75th, 90th, 95th, 99th and the median percentiles. If this metrics presents a low percentage at the
3. Gini coefficient: calculated for every year, its evolution is here displayed. The Gini coefficient is graphically represented by the difference between the area under the equidistribution line where 𝑦 = 𝑥, and the Lorenz curve. As a first step, it is important to find out the Lorenz function. Following Bellù and Liberati (2005), the x-axis represents the cumulative proportion of the population of firms ranked by their amount of R&D investments; the y-axis represents the cumulative proportion of R&D investments for a given proportion of the population of firms, divided by the total amount of R&D investments. The Lorenz function is thus calculated as:
𝐿 (
𝑘
𝑃
) =
∑
𝑘𝑖=1𝑦𝑖
𝑌
where k=1,….n is the position of each firm in the R&D investments distribution; i=1,….k is the position of each firm in the R&D investments distribution; P is the population representing the total number of R&D investors in the distribution; yi is the R&D investments per ith firm in the
distribution; ∑𝑘𝑖=1𝑦𝑖 represents the cumulated R&D investments up to the kth firm. Once the
Lorenz function 𝐿(𝑋) is obtained, it is possible to calculate the Gini coefficient with the following formula:
𝐺 = 1 − 2 ∫ 𝐿(𝑋)𝑑𝑋
1
0
The final output ranges between 0 (maximum equality) and 1 (maximum inequality).
For testing the second set of hypotheses, a subsample containing the top 100, 50, 25 and 10 R&D investors per year was created in every dataset and the following measures were applied:
1. Share of R&D by top investors: this measure presents the amount of R&D investments in terms of percentage of R&D by top investors over the total amount of R&D performed by all the firms. 2. Share of R&D growth captured by top investors: to calculate this measure, it is first necessary to disaggregate the total R&D performed by top investors and the rest of the firms in the sample. Then, the year-by-year difference for every part (total annual change in the dataset and total annual change among top investors) is needed:
𝑇𝑜𝑡𝑎𝑙𝐶ℎ𝑎𝑛𝑔𝑒 = 𝑇𝑜𝑡𝑎𝑙𝐶ℎ𝑎𝑛𝑔𝑒𝑡 − 𝑇𝑜𝑡𝑎𝑙𝐶ℎ𝑎𝑛𝑔𝑒𝑡−1
Finally, in order to capture the share of growth from the top, one should calculate the ratio between TotalChange and TopChange in every specific year and multiply the result by 100:
𝑆ℎ𝑎𝑟𝑒𝑇𝑜𝑝 = (𝑇𝑜𝑝𝐶ℎ𝑎𝑛𝑔𝑒 / 𝑇𝑜𝑡𝑎𝑙𝐶ℎ𝑎𝑛𝑔𝑒) ∗ 100 for every year t
3. R&D spells: to retrieve this measure, it is first necessary to order the distribution of top investors by name in every dataset and then generate a counting variable which reports the number of years in which those company appear in the selected sample of the top investors. Then, the mean for this newly created variable was computed. This same procedure was followed for computing the R&D spells for the rest of the companies investing in R&D and the outputs were confronted.
6. Analysis and Results
R&D concentration
Extensive and intensive margins. Analyzing the extensive margins in Compustat, it is evident how the number of companies followed well-known economic trajectories, with their number increasing until the dot-com bubble (fig. 1). This event elicited a deflation in the number of companies investing in R&D that is still persisting nowadays. Surprisingly, the percentage of companies investing in R&D was higher at the beginning of the period analyzed (29.77%), it reached its peak in 1998 (31.01) and then started its decline, reaching its historical low in 2011, and accounting for the 23.52% in 2018.
When analyzing the extensive margins by sector (fig. 2), it is immediately clear that more and more companies from R&D intensive sectors have started investing in R&D over time- especially chemical companies-, and the total share of R&D intensive companies passed from 59.14% in 1975 to 82.34% in 2018. On the contrary, companies from non-R&D intensive sectors (“Rest”) halved in relative terms over time, from 40.86% to 17.66%.
18
average company in the sample, with instruments and engineering sectors presenting annual R&D investments lower than the average. Detailed annual data can be found in the Appendix.
The Orbis dataset displays interesting results. Regarding the extensive margins, the number of companies investing in R&D marked a consistent increase, in contrast to Compustat, passing from 36.68% in 2010 to 46.85% in 2018 (fig. 5), and more than doubling in absolute numbers. The analysis of R&D extensive margins by sector displays results very similar to Compustat, with no sector showing any particular growth in the 9-year period analyzed (fig. 6).
The most surprising data relates to the intensive margins, which diminished over time (fig. 7), especially in the first years (from $ 70 million in 2010 to $ 41 million in 2016), for partially bouncing back in 2017 and 2018 only. This result contrasts with the one in Compustat, but it is mainly due to the higher number of small investors in Orbis. A detailed analysis in table 3 shows that the percentage of firms investing less than $ 1 and $ 10 million is not only higher in Orbis-as previously mentioned- but it also increased substantially in the same period where the average R&D investments decreased.
Regarding the sectorial segmentation (fig. 8), the analysis shows a declining trend for every sector, with companies from various R&D intensive sectors investing less in R&D than the average, except for transportation, electronics and chemicals.
R&D per percentile. The second measure of concentration entails the share of R&D investments per percentiles in every dataset for every given year. Detailed annual data can be found in the Appendix. Starting from the lowest percentiles, we can notice that a contribution above 1% of total R&D is not even present before the median in Compustat. Basically, the bottom half of the distribution only accounts for ̴1% of the total R&D investments, therefore a high level of concentration is expected. These results are even sharper in Orbis, with the median’s share ranging from a maximum of 0.71% in 2013 to a minimum of 0.57% in 2014. Moreover, even the COR&DIP dataset reports a high concentration, with the median accounting for only 6.44% of the total R&D investments, on average (see table 4).
19
16.41%; COR&DIP 44.85%) are in line with this trend. The most important growth occurs at the 99th percentile, with a mean value of 57.18% for Compustat, 41.35% for Orbis, and 76.81% for COR&DIP, although remaining very low, especially for Orbis and Compustat.
Table 4 reports low variation for all these variables, suggesting that high concentration has been stable over time. Especially for what concerns the Compustat sample, this gives a robust indication that concentration is not a recent occurrence. Moreover, Orbis presents the highest levels of concentration, showing that R&D investments are still largely performed by few firms. Finally, even the COR&DIP dataset presents high levels of concentration, with the bottom 5% of investors accounting for more than half of the total R&D in the sample.
The graphs can help to better grasp the dimension of these results. Fig. 9 shows the evolution of the percentages for every percentile over time in Compustat. Again, it is worth noting the impact of the dot-com bubble, with the amount of R&D per percentile reaching its lowest around the late 1990s, for resurging thereafter. Fig. 10 mirrors fig.9, showing the amount of R&D by the top 10%, top 5% and top 1% of investors, which have always reaped very high proportion over time3. The figures regarding
Compustat show that high concentration of R&D investments has been a stable phenomenon, being even higher in the initial period, diminishing after the dot-com bubble, and growing again afterwards. Looking at Orbis figures, concentration is higher than in Compustat, showing a slight increase in the last years (fig. 11), with the top 1% of investors passing from 52.57% in 2010 to 60.79% in 2018. Finally, COR&DIP figures show stable trends in the period analyzed.
Figures 15-17 display the amount of R&D coming from all the companies in R&D intensive sectors at the 90th, 95th and 99th percentile. Disaggregated data for every single sector can be found in the Appendix. In Compustat, the amount of R&D from firms in R&D intensive sectors at all the highest percentiles increased after the early 2000s, especially at the 99th percentiles. Therefore, the bottom 99% of the firms active in R&D intensive sectors has substantially increased its contribution to the total R&D from that period on, which passed from 44.20% in 1998 to 53.13% in 2018. Nevertheless, as fig. 15 demonstrates, the 1% of the top R&D investors has remained responsible for around 25-30% of the total R&D coming from R&D intensive sectors over time. Moreover, the amount coming from firms within the 95th and 90th
3 The number of firms represented by the top 10%, top 5% and top 1% of investors in every dataset is reported in the
20
percentile is considerably lower, suggesting that R&D intensive sectors are representative of big R&D investors. Orbis shows more concentrated patterns in relation to the sectorial segmentation. In particular, fig. 16 demonstrates that the top 1% is responsible for a vast amount of R&D coming from R&D intensive sectors (58.58% on average), around the double than in Compustat. Finally, even in the COR&DIP sample, we can notice that the top 1% of R&D investors is responsible for a great amount of R&D coming from R&D intensive sectors (24.78% on average, a similar level as in Compustat). However, the differences between the 90th, 95th and 99th percentiles are not particularly remarkable, showing that only the top 1% has a considerable impact.
Gini Coefficient. The last measure of concentration is represented by the Gini Coefficient. Not surprisingly, Orbis detains the highest Gini coefficient, with a mean value of 0.93 over time. Compustat follows with a mean value of 0.90 and COR&DIP presents a lower value (0.76), which slightly diminished over the analyzed period. The measures of Gini coefficients are fairly stable over time and this strengthen this section of results, showing that concentration is high in all three datasets analyzed (see table 5 and fig. 18) without linearly increasing over time.
Superstar effect
For testing the second set of hypotheses, firms were ranked in terms of their amount of R&D investments in the three datasets. The final samples include the top 100, 50, 25 and 10 R&D investors per year.
Share of R&D by top investors. As fig. 19-21 show, the impact of the top 100 is easily recognizable, accounting for more than half of the total R&D investments. However, these results are not surprising. In fact, recalling table 4, we can see that the average number of firms among the top 1% R&D investors in Compustat and COR&DIP is smaller than some of these top categories. For this reason, the analysis of the top 100, 50 and 25 investors will be particularly important for Orbis, while the analysis of the top 10 is clearly interesting for all the three datasets. However, taking a fixed number of top investors is useful for better shedding light on the superstar effect.
21
in 2018. This strengthens the previous results about concentration, which was higher in the first decades analyzed in this dataset than nowadays.
In Orbis, the top 50 investors (which represent the 0.33% of the total) account for around 40% of the total R&D investments. This figure provides a clear signal about the existence of the superstar effect. In this sample, the top 25 and top 10 investors account, respectively, for 26.46% and 13.99% of the total of R&D investments, on average. Regarding the evolution of these contributions, we can notice that the share of the top 100 and top 50 has slightly declined in Orbis, while the one from the top 25 and top 10 remained stable.
Finally, it is interesting to notice that the top 10 investors also constitute, on average, 14.58% of the total R&D investments among the top 2000 R&D investors worldwide represented in COR&DIP. All these figures clearly explain that top investors account for a substantial proportion of the total R&D investments in every dataset, although their share remained stable over time, or even diminished. Breaking down the analysis of the top 100 R&D investors by sector show similar results across all three datasets (fig. 25-27). First, the amount of investments coming from R&D intensive sectors is very high, accounting for more than 90% in all three datasets. Second, the two most intensive sectors among the top 100 are represented by the chemical and transportation industries in all three cases. The sectorial segmentation was only provided by the top 100, since it would be less representative for the other categories.
Share of R&D growth captured. As table 6 shows, the top investors in Compustat display the highest shares of growth captured, particularly for the top 100. This means that the top 100 investors have been, on average, responsible for almost 72% of the total annual growth in R&D investments. Orbis and COR&DIP present high figures as well (32.04% and 49.79%), with all the shares flowing into similar values in the top 10 categories of every dataset. This shows that top investors have a profound impact on the total growth of R&D investments, being responsible for a consistent part of this growth, especially in Compustat. The top 10 investors captured, on average, around 20% of the total growth across all three datasets and annual data can be accessed in the Appendix.
22
the top 100, this measure is way higher than the for the rest both in Compustat and in Orbis, and slightly lower in COR&DIP. Due to the longest period analyzed, the Compustat dataset is more representative for the R&D spells, and it also provides the strongest results, with higher R&D spells (in years) for the top 100 (27.90 vs 17.50), top 50 (27.56 vs 17.81) and top 25 (22.30 vs 17.99) than for the rest.
Summary of the results. Table 7 provides a summary of the results. Even though the comparison between extensive and intensive margins in Orbis did not suggest a high concentration, the two other measures for H1a (i.e. the analysis of R&D per percentile and the Gini coefficient) largely validated the hypothesis also for this dataset, which presented the highest level of concentration. Also Compustat and COR&DIP data reported a high level of concentration, therefore H1a is confirmed. However, concentration has not generally increased over time, except for a slight increase in Orbis, which is way less significant than in Compustat. Therefore, we can reject H1b, asserting that concentration has not increased over time.
The analysis showed that the superstar effect exists and is remarkable for every top category, with top R&D investors (especially top 100 and top 50) accounting for a large part of the total investments in all three datasets (between 52% and 72% for the top 100; between 39% and 56% for the top 50), with even the top 10 accounting for a share between 14% and 24% of the total R&D investments, on average. Moreover, these top investors were also able to capture a large share of the total R&D growth, especially in Compustat, in which the top 100 captured, on average, almost 73% of the total growth. These data largely confirm H2a. On the other side, as in the case of H1b, since the contribution of top investors has not increased over time, H2b is rejected.
The R&D spells in Compustat and Orbis were higher for the top 100, 50 and 25 R&D investors with respect to the rest. In particular, the top 100 and top 50 investors in Compustat were able to remain in that top category for 10 years more with respect to the firms remaining in the “rest” category, showing that this ranking is hardly contestable. Although the R&D spells were slightly lower with respect to the rest in COR&DIP, Compustat is the most feasible dataset for this analysis and it presents the highest levels. Therefore, we can confirm H2c.
23
performed. Nonetheless, the link between R&D intensive sectors and top investors was fairly strong both in the analysis of the percentiles, as well as in testing the superstar effect.
Table 7: summary of the results
Compustat Orbis COR&DIP
H1a: R&D investments are highly
concentrated Confirmed Confirmed Confirmed
Extensive/Intensive margins
Extensive margins decreased over time, while intensive margins steadily increased
Extensive margins increased over time, while intensive margins decreased
Not tested
R&D per percentile Very high Very high High
Gini Coefficient Very high Very high High
H1b: Concentration has increased
over time Not confirmed Confirmed Not confirmed
H2a: Top R&D investors account for a large share of total R&D and
R&D growth Confirmed Confirmed Confirmed
H2b: Top R&D investors’
contribution increased over time Not confirmed Not confirmed Not confirmed H2c: Top R&D investors
category is hardly contestable Confirmed Confirmed Not confirmed
7. Discussion and Conclusion
This paper aimed to provide evidence about the existence of a best vs rest pattern in R&D investments by showing stylized facts regarding the evolution of concentration in R&D investments. Three large datasets with different but complementary characteristics related to the nature of the firms, geographical coverage and historical periods were analyzed, involving a total of 814,414 observations. This research entails three main macro results.
24
the top 1% of R&D investors still account for 58% of the investments in Orbis and 42% in Compustat. Data from the COR&DIP dataset confirmed the unequal distribution of R&D even when considering the top 2000 R&D investors worldwide. This proportion remains consistent for the top 100, top 50, top 25 and top 10 investors in every dataset. In general, due to the longer time series, Compustat can be considered as the most reliable tool for enquiring over inequality in R&D investments, and the clear results from this dataset confer robust evidence to these general findings.
Moreover, top investors are also able to capture a substantial part of the total growth in R&D. For instance, the top 100 investors captured on average 32% of the total growth in Orbis, 49% in COR&DIP and 73% in Compustat. This metric suggests the importance of disaggregating data when analyzing R&D annual growth, since this growth can be largely due to top investors’ contribution. These results, combined with the high share of R&D by top investors, clearly prove the existence of a superstar effect in R&D investments.
The second macro result shows that top R&D investors exert an enormous market power in relation to R&D investments, being able to continuously allocate enormous resources to R&D. In fact, if a company becomes a “superstar investor”, it will be hardly displaceable from the top category, in line with Autor et al. (2020) who claimed that the superstar status is hardly contestable. In Compustat, I found that the top 100 investors remained in this ranking for 27.90 years in contrast to the rest of the firms remaining in the “rest” category for only 17.50 years. Similar results accrued for the top 50 and top 25, and even in Orbis the R&D spells were higher for the superstar in comparison to the rest. This shows the difficulty for the rest of the firms to reach the same R&D scale of the superstar investors.
25
Nonetheless, although R&D concentration has not increased in relative terms, the amount of R&D allocated nowadays by top investors is way higher than 40 years ago in absolute terms. Even though the impact of R&D in explaining the growing inequality across firms was not the subject of this study, this does not rule out the possibility that it can have an important impact.
Implications
These results suggest that R&D investments do not always activate economies of scale in a linear way. In particular, once engaged in research activities, firms are not able to increase R&D investments in the same way. This problem can have different sources, which I could not properly investigate in this paper. However, two general recommendations to overcome this barrier deserve mention.
First, firms need to balance scientific and technological expertise with sound managerial capabilities when organizing their R&D activities. In fact, technical entrepreneurs founding high-tech startups have a common tendency to believe that business management skills can be self-thought, advocating many of these functions for themselves (Oakley, 2003). However, insufficient attention to the management of R&D can deter companies to sufficiently capitalize on innovation. Therefore, young research-intensive companies should attribute equal importance to managerial aspects that are often underestimated in R&D, focusing on recruiting skilled and talented people also in these areas.
Second, as underlined by Teece (1986), young innovating companies should develop complementary assets once they possess enough liquidity for doing so, in particular competitive manufacturing, complementary technologies, service and distribution. This would possibly enable a young firm to better capitalize on innovation and to commit increasing investments in R&D, scaling them up over time. Limitations and future research
26
27
References
Appelt, S., Bajgar, M., Criscuolo, C., & Galindo-Rueda, F. (2016). R&D Tax Incentives: Evidence on
Design, Incidence and Impacts. OECD Science, Technology and Industry Policy Papers, No. 32.
Appelt, S., Galindo-Rueda, F., & González Cabral, A.C. (2019). Measuring R&D Tax Support: Findings
from the new OECD R&D Tax Incentives Database. OECD Science, Technology and Industry
Working Papers.
Arqué-Castells (2018, June). What drives differences in R&D across countries? Insights from the
intensive and extensive margins. Working Paper.
Arqué-Castells (2020, March). Take-up of R&D Tax Credits. Unpublished manuscript.
Arthur, B. W. (1996). Increasing returns and the new world of business. Harvard Business Review, 74(4), 100–109.
Artz, K., Norman, P., Hatfield, D. & Cardinal, L. (2010). A Longitudinal Study of the Impact of R&D, Patents, and Product Innovation on Firm Performance. Journal of Product Innovation
Management, 27(5), 725-740.
Autor, D., Dorn, D., Katz, L. F., Patterson, C., & Reenen, J. V. (2020). The Fall of the Labor Share and the Rise of Superstar Firms. The Quarterly Journal of Economics, 135(2), 645–709.
Ayyagari, M., Demirguc-Kunt, A., & Maksimovic, V. (2018). Who are America’s Star Firms? (World Bank Group. Policy Research Working Paper No. 8534).
Bakker, G. (2013). Money for nothing: How firms have financed R&D-projects since the Industrial Revolution. Research Policy, 42(10), 1793–1814.
Becker, B. (2015). Public R&D Policies and Private R&D Investments: A Survey of the Empirical Evidence. Journal of Economic Survey, 29(5), 917-942.
28
Berlinghieri, G., Criscuolo, C., & Blanchenay, P. (2017). The Great Divergence(s). OECD Science, Technology and Innovation (Policy Paper No. 39).
Bessen, J. (2019). Industry Concentration and Information Technology (Boston University School of Law. Law & Economics Paper No. 17-41).
Bloom, N., Draca, M., & Reenen, J. V. (2011). Trade Induced Technical Change? The Impact of Chinese
Imports on Innovation, IT and Productivity (NBER Working Paper No. 16717).
Bloom, N., Jones, C. I., Reenen, J. V., & Webb, M. (2017). Are Ideas Getting Harder to Find? (NBER. Working Paper No. 23782).
Bloom, N., Reenen, J.V., & Williams, H. (2019). A Toolkit of Policies to Promote Innovation. Journal of economic Perspectives, 33(3), 163-184.
Bloom, N., Schankerman, M. & Reenen, J.V. (2013). Identifying Technology Spillovers and Product Market Rivalry. Econometrica, 81(4), 1347-1393.
Bogner, W., & Bansal, P. (2007). Knowledge Management as the Basis of Sustained High Performance.
Journal of Management Studies, 44(1), 165-189.
Bound, J., Cummins, C., Griliches, Z., Hall, B., & Jaffe, A. (1984). Who Does R&D and Who Patents? In Griliches, Z. (Ed). R&D, Patents and Productivity. University of Chicago Press.
Brynjolfsson, E., McAfee, A., Sorell, M., & Zhu, F. (2008). Scale without Mass: Business Process
Replication and Industry Dynamics (Working Paper).
Calligaris, S., Criscuolo, C., & Marcolin, L. (2018). Mark-ups in the digital era (OECD Science, Technology and Industry Working Papers).
Chung, K. H., & Cox, R. A. K. (1994). A Stochastic Model of Superstardom: An Application of the Yule Distribution. The Review of Economics and Statistics, 76(4), 771–775.
29
Crouzet, N., & Eberly, J. (2018). Intangibles, Investment, and Efficiency (AEA Papers and Proceedings). Furman, J, & Orszag, P. (2015, October 16). A Firm-Level Perspective on the Role of Rents in the Rise
in Inequality. Presentation at “A Just Society” Centennial Event in Honor of Joseph Stiglitz.
Gal, P. (2013). Measuring Total Factor Productivity at the Firm Level Using OECD-ORBIS (OECD Economics Department. Working Paper No. 1049).
Gentry, R. J., & Shen, W. (2013). The impacts of performance relative to analyst forecasts and analyst coverage on firm R&D intensity. Strategic Management Journal, 34(1), 121–130.
Grant, R. (1996). Toward a knowledge-based theory of the firm. Strategic Management Journal, 17(2), 109-122.
Griliches, Z. (1985). Chapter Title: Productivity, R&D, and Basic Research at the Firm Level in the 1970s, 82-99. In Griliches, Z. (1998). R&D and Productivity: The Econometric Evidence. University of Chicago Press.
Hall, B. (2002). The Financing of Research and Development. Oxford Review of Economic Policy, 18(1), 35–51.
Hall, B., Mairesse, J. & Mohnen, P. (2010). Chapter 24: Measuring the Returns to R&D, 1033-1082. In Hall, B., & Rosenberg, N. (Ed), Handbook of the Economics of Innovation, Volume 2. North-Holland.
Henderson, R., & Cockburn, I. (1996). Scale, Scope, and Spillovers: The Determinants of Research Productivity in Drug Discovery. The RAND Journal of Economics, 27(1), 32–59.
Higón, D., Máñez, J., & Sanchis-Llopis, J. (2011). The role of extensive and intensive margins in
explaining corporate R&D growth: Evidence from Spain. Ivie Working Papers.
30
James, S. D., Leiblein, M. J., & Lu, S. (2013). How Firms Capture Value from their Innovations. Journal
of Management, 39(5): 1123–1155.
Jones, B. (2009). The Burden of Knowledge and the “Death of the Renaissance Man”: Is Innovation Getting Harder to Find? The Review of Economic Studies, 76(1), 283-317.
Jones, D. (2007). Voluntary Disclosure in R&D-Intensive Industries. Contemporary Accounting
Research, 24(2), 489-522.
Kile, C., & Phillips, M. (2009). Using Industry Classification Codes to Sample High-Technology Firms: Analysis and Recommendations. Journal of Accounting, Auditing and Finance, 24(1), 35-58. Loecker, J. D., & Eeckhout, J. (2018). Global Market Power (NBER Working Paper No. 24768). Loecker, J. D., Eeckhout, J., & Unger, G. (2020). The Rise of Market Power and the Macroeconomic
Implications. The Quarterly Journal of Economics, 135(2), 561–644.
Macher, J., & Boerner, C. (2006). Experience and Scale and Scope Economies. Trade-offs and Performance in Development. Strategic Management Journal, 27(9), 845-865.
Mansfield, E. (1980). Basic Research and Productivity Increase in Manufacturing. The American
Economic Review, 70(5), 863-873.
McKinsey Global Institute. (2018, September). Superstars: The dynamics of firms, sectors, and cities
leading the global economy.
Mowery, D. C. (1983). Industrial Research and Firm Size, Survival, and Growth in American Manufacturing, 1921–1946: An Assessment. The Journal of Economic History, 43(4), 953–980. Oakley, R.P. (2003). Technical entrepreneurship in high technology small firms: some observations on
the implications for management. Technovation, 23(8), 679-688.
31
OECD (2020). OECD Main Science, and Technology Indicators. R&D Highlights in the February 2020
Publication. OECD Directorate for Science, Technology and Innovation.
Pierce, J., & Schott, P. (2012). The surprisingly swift decline of U.S. manufacturing employment (NBER Working Paper No. 18655).
Reenen, J. V. (2018). Increasing Differences Between Firms: Market Power and the Macro-Economy (CEP Discussion Papers No. 1576).
Rosen, S. (1981). The Economics of Superstars. The American Economic Review, 71(5), 845–858. Rosenberg, N. (1990). Why do firms do basic research (with their own money)? Research Policy, 19(2),
165–174.
Sofka, W., de Faria, P., & Shehu, E. (2018). Protecting knowledge: How legal requirements to reveal information affect the importance of secrecy. Research Policy, 47(3), 558-572.
Teece, D.J. (1986). Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing and Public Policy. Research Policy, 15(6), 285–305.
The Economist (2016, Sept. 17). The rise of the superstars. Special Report Companies.
Traina, J. (2018). Is Aggregate Market Power Increasing? Production Trends using Financial Statements (Stigler Center for the Study of the Economy and the State. University of Chicago Booth School of Business. New Working Paper Series No. 17).
Zack, M. H. (1999). Developing a knowledge strategy. California Management Review, 41(3), 125–145. Zenger, T. R. (1994). Explaining Organizational Diseconomies of Scale in R&D: Agency Problems and the Allocation of Engineering Talent, Ideas, and Effort by Firm Size. Management
32
Tables and Figures
Table 2: Descriptive statistics of the variables
Compustat Orbis COR&DIP
Mean S.D. Mean S.D. Mean S.D.
Observations per year 11,064.5 2,250.91 34,647 7,544.49 1,969.13 24.82 Extensive margins (n. of companies) 2,910.64 665.98 14,687.78 4,543.25 - - Intensive margins (in $ mln) 83.30 63.93 52.91 11.54 291.59 50.12 Total R&D investments (in $ mln) 257,874.2 195,994.5 733,674 117,441.5 574,996.9 104,374.3 Share of R&D (all figures in %)
1st Percentile 0.00014 0.00007 0.000049 0.000017 0.025 0.030 5th Percentile 0.004 0.001 0.0025 0.00079 0.22 0.120 10th Percentile 0.017 0.005 0.0130 0.0033 0.59 0.207 25th Percentile 0.146 0.031 0.108 0.013 2.15 0.370 Median 1.007 0.191 0.65 0.060 6.44 0.593 75th Percentile 4.45 0.67 2.90 0.34 15.53 0.823 90th Percentile 12.65 1.69 9.24 0.87 30.86 0.850 95th Percentile 22.58 3.19 16.41 1.39 44.85 0.697 99th Percentile 57.18 3.60 41.35 3.95 76.81 0.716 Gini Coefficient 0.9042 0.0093 0.9313 0.0062 0.7592 0.0104
R&D spells (in years) 18.27 11.83 6.97 2.21 6.68 1.97
R&D intensive sectors: % over total number of firms
Chemicals 17.21 7.06 15.28 1.07 17.77 1.67 Machineries 11.76 3.36 10.84 0.29 8.24 0.46 Electronics 14.07 1.21 14.92 1.31 15.81 1.18 Transportation 3.74 0.64 4.33 0.38 8.07 0.22 Instruments 12.26 1.25 7.52 0.18 8.03 0.26 Business Services 14.30 5.63 13.55 2.07 13.38 0.35 Engineering 1.25 0.38 5.70 0.98 4.81 0.65
33 Figure 1: R&D extensive margins (Compustat)
34 Figure 3: R&D intensive margins (Compustat)
35 Figure 5: R&D extensive margins (Orbis)
36 Figure 7: R&D intensive margins (Orbis)
37
Table 3: Evolution of observations per investment size, in percentage over total (Orbis)
Year Below $ 1 mln Over $ 1 mln Below $ 10 mln mln Over $ 10 Below $ 50 mln Over $ 50 mln Below $ 100 mln Over $ 100 mln
2010 29.52 70.48 68.99 31.01 87.30 12.70 92.18 7.82 2011 29.44 70.56 69.87 30.13 87.85 12.15 92.41 7.59 2012 32.78 67.22 72.11 27.89 89.19 10.81 93.19 6.81 2013 33.19 66.81 73.04 26.96 89.89 10.11 93.77 6.23 2014 42.96 57.04 78.65 21.35 92.11 7.89 95.21 4.79 2015 44.63 55.37 79.38 20.62 92.41 7.59 95.41 4.59 2016 43.35 56.65 79.34 20.66 92.27 7.73 95.39 4.61 2017 39.85 60.15 77.57 22.43 91.54 8.46 95.11 4.89 2018 37.37 62.63 75.56 24.44 90.46 9.54 94.44 5.56
38 Figure 10: Share of R&D from top investors (Compustat)
39 Figure 12: Share of R&D from top investors (Orbis)
40 Figure 14: Share of R&D from top investors (COR&DIP)
Table 4: Concentration of R&D investments in the datasets: descriptive statistics
41 Figure 15: R&D per percentile by sector, in percentage (Compustat)
42 Figure 17: R&D per percentile by sector, in percentage (COR&DIP)
Table 5: Gini Coefficients in the various datasets: descriptive statistics
Dataset Mean S.D. Min Max
Compustat (1975-2018) 0.9043 0.0093 0.8907 0.9209
Orbis (2010-2018) 0.9313 0.0062 0.9225 0.9388
43 Fig. 18: Gini coefficients in the datasets
44 Figure 20: Evolution of Total R&D by Top 100 (Orbis)
45 Figure 22: R&D contribution by top investors (Compustat)
46 Figure 24: R&D contribution by top investors (COR & DIP)
47 Figure 26: sectorial composition of the top 100 investors (Orbis)
48
Table 6: Share of R&D by top investors, R&D growth captured and R&D spells (mean values)
Top 100 Top 50 Top 25 Top 10
Compustat
Share (%) of R&D investments
over total 71.94 56.57 40.72 24.29
Share (%) of the total R&D
growth captured 72.98 57.90 39.89 24.14
R&D spells (in years) 27.90 27.56 22.30 16.98 R&D spells by the rest (in years) 17.50 17.81 17.99 18.14 Orbis
Share (%) of R&D investments
over total 52.68 39.77 26.47 13.99
Share (%) of the total R&D
growth captured 32.04 29.76 26.81 20.38
R&D spells (in years) 7.57 7.96 7.60 7.33
R&D spells by the rest (in years) 6.95 6.96 6.97 6.97 COR&DIP
Share (%) of R&D investments
over total 55.58 41.57 27.80 14.58
Share (%) of the total R&D
growth captured 49.79 40.31 28.68 17.90
R&D spells by top investors (in
years) 6.32 6.47 6.21 6.25
