Corporate Science, Innovation and Firm Value

Markus Simeth a, b Michele Cincera c

This version: May 2015

Paper forthcoming in Management Science

Abstract

Many firms actively disclose research findings in scientific peer-reviewed journals. The literature
highlights several potential benefits of such scientific boundary-spanning activities, including privileged
access to academic information networks. However, scientific disclosure may lead to unintended
knowledge spillovers. It remains unclear whether active engagement in science leads to higher returns.
This paper investigates the impact of scientific activities on the firm’s market value, using accounting
data for US firms and matched patent and scientific publication data. We find evidence for the positive
impact of scientific publications on a firm’s market value beyond the effects of R&D, patent stocks and
patent quality and also document heterogeneity with respect to this impact between different industrial
sectors.

Keywords: R&D, Industrial science, Scientific disclosure, Tobin’s Q, Knowledge disclosure,

Innovation

______________________

a
KU Leuven, Department of Managerial Economics, Strategy & Innovation, Belgium
b
Universidad Carlos III de Madrid, Department of Business Administration, Spain.
c
Université Libre de Bruxelles (ULB), Solvay Brussels School of Economics and Management, iCite,
Belgium

Corresponding author: Markus Simeth, KU Leuven, Department of Managerial Economics, Strategy

and Innovation, Naamsestraat 69, B-3000 Leuven. Email: markus.simeth@kuleuven.be

Electronic copy available at: https://ssrn.com/abstract=2317166

1 Introduction

Research and Development (R&D) is an important source of competitive advantage for companies. In
particular, investments in upstream research lead to the creation of new knowledge inside a firm and
also facilitate the absorption of academic institutions’ and researchers’ latest knowledge (Cohen and
Levinthal 1989; Rosenberg 1990). Consequently, firms investing in research may produce radical
inventions which promise greater financial returns. Interestingly, many companies involved in R&D not
only absorb external academic knowledge but actively contribute to open science by disclosing their
own findings in scientific peer-reviewed journals (Hicks 1995).

The literature highlights several mechanisms through which firms who adopt academic disclosure
practices might benefit. Broadly summarizing, advantages include increased capacity to absorb external
knowledge, the realization of signaling benefits which facilitate access to valuable research inputs such
as hiring of PhD graduates, and the promotion of science-based products among professional customers
(Hicks 1995; Stern 2004; Polidoro and Theeke 2012). However, the dissemination of research findings
in scientific journals may also lead to unintended knowledge spillovers that facilitate imitation by
competitors and other specific costs associated with the disclosure process (Arrow 1962; Cockburn et
al. 1999; Kinney et al. 2004; Stuart and Liu 2014). Although empirical studies document that firms with
scientific publications lodge higher quality patents (Cockburn and Henderson 1998; Gittelman and
Kogut 2003; Fabrizio 2009), these studies disregard the potential spillover costs caused by scientific
disclosure. It remains unclear therefore whether increased R&D productivity by a firm genuinely
translates into higher financial profitability.

In this paper, we address this question and investigate the profitability implications of scientific
disclosure. In order to capture all positive and negative effects, we apply a market value approach
(Griliches 1981; Jaffe 1986; Hall 1993; Hall et al. 2005). Financial markets react immediately to new
information and anticipate future commercial implications. Thus, this approach avoids making
assumptions about time lags between knowledge creation, disclosure effects, and the realization of
financial returns. We separate - conceptually and empirically - the creation of scientific outcomes from
the voluntary disclosure of these outcomes in scientific peer-reviewed journals. We also provide initial
insights regarding the broad mechanisms through which scientific publications may create value. A
further distinctive contribution is that unlike the majority of studies dealing with science in the corporate
context which refer to the biotechnology and pharmaceutical industries, the present paper considers all
high-technology sectors. Since sectors differ according to their appropriability conditions (Levin et al.
1987; Cohen et al. 2000) which may moderate spillover risks of disclosure, we examine sector
heterogeneity in greater detail.

Our empirical analysis uses firm-level information from the US Compustat database, and matched
scientific publication and patent data for a representative sample of 1,739 stock-market listed firms

-2-

Electronic copy available at: https://ssrn.com/abstract=2317166

(9,920 firm-year observations) from all high-technology sectors, covering the period 1996-2006. The
results of our econometric analysis suggest a positive relationship between the scientific publications of
firms and their Tobin’s Q. We find evidence that disclosure in high-quality journals, which indicates
active involvement in science, is especially valuable in terms of higher stock market premium. However,
we detect a certain degree of heterogeneity across high-technology sectors: whereas scientific
publications are associated with positive returns in the instruments sectors, and to some extent also in
pharmaceutical and biotechnology, we find a negative but non-significant tendency in the ICT sectors.

2 Theoretical background

This paper builds on the already large body of literature examining the relationship between R&D and
market value, and on recent work investigating the scientific activities of profit-oriented firms. With
regard to the former, several studies have examined the impact of R&D expenditures, patent stocks and
patent quality indicators on firm value (see Hall 2000; Czarnitzki et al. 2006, for surveys; Hall et al.
2005). The majority highlights a positive effect of R&D and patent stocks on the firms’ market value,
including a premium for high-quality patents. More recent studies have taken into account the strength
of the appropriability regime, complementary assets and knowledge spillovers as moderating factors on
this impact (Cockburn and Griliches 1988; MacGahan and Silverman 2006; Belenzon 2012). To date
however, few studies have considered heterogeneity of R&D activities across firms. One exception to
this in accounting literature is the study by Deng et al. (1999) which found a positive effect of executing
scientific research on firm value. However that study did not analyze voluntary contributions of firms
to the stock of open science. Instead, the authors relied on non-patent literature references in a firm’s
list of patents as a proxy for scientific research. We build on their work and offer a more complete
perspective, with a special focus on the impact of the active participation of firms in the scientific
community, tangibly reflected in scientific publications.

From a conceptual point of view, there are several reasons why the scientific activities of firms
may have implications for firm profitability. Firms engaging in (basic) research may develop superior
capabilities, in that they may be able to combine technologically distant knowledge to create inventions
which are more valuable than those created by other firms. Such firms are potentially able to absorb
external knowledge and identify promising trajectories for applied research and experimental
development, leading to superior inventive outcomes (Rosenberg 1990; Cohen and Levinthal 1989;
Gittelman and Kogut 2003; Fleming and Sorenson 2004; Fabrizio 2009). While these studies also
suggest a positive relationship between research investments and firm profitability, the potential costs
of basic research are not considered. In particular, basic research findings do not necessarily lead to high
commercial success since translating findings into concrete products can be challenging (see Nelson
1959; Rosenberg 1990; Cockburn et al. 1999; Pisano 2006).

-3-

Electronic copy available at: https://ssrn.com/abstract=2317166

 Beyond the creation of scientific knowledge, many firms also voluntarily disclose their research
results in scientific journals. Such involvement in open science may create value through several
mechanisms associated with access to upstream knowledge and appropriation (see Penin 2007 for a
review). With regard to upstream knowledge, disclosing research results may not only increase the
absorptive capacity (Cockburn and Henderson 1998), but may facilitate knowledge sourcing from
academic scientists, thanks to reciprocity in knowledge exchange. More specifically, academic scientists
may be reluctant to interact with scientists from firms unless the latter share valuable research results
(Hicks 1995; Simeth and Raffo 2013). Moreover, scientific publications can be leveraged as a human
resource instrument with regard to PhD graduates. Many scientists who consider the private sector as a
career option value the possibility of continuing publication activities, and may seek out firms engaging
in this practice (Stern 2004; Sauermann and Roach 2013). Scientific publications are also considered a
potential appropriability device. Disclosure may encourage the adoption of science-based products such
as medical drugs or instruments (Azoulay 2002; Polidoro and Theeke 2012). Clinicians, as users of
medical instruments, influence procurement decisions in university hospitals and are keen to understand
about the technological characteristics of the products. Manufacturers of such devices might gain greater
credibility among these clinicians by disclosing related information in scientific peer-reviewed journals.
Another appropriation-related use of scientific documents is the establishment of prior art through
disclosure, the aim being to hamper patenting strategies of competitors working on similar inventions
(De Fraja 1993; Parchomovsky 2000; Della Malva and Hussinger 2012).

However, the publication process has related costs. The most obvious of these are unintended
knowledge spillovers that may enable competing firms to learn from the disclosed knowledge (Arrow
1962). Knowledge spillovers may reduce the cost of imitation for competitors, offer the latter insights
into future trends and facilitate their exploration of alternative technological trajectories. In addition, the
publication process itself can lead to opportunity costs since a firm’s researchers have to prepare
documents which meet the publishing requirements set down by journals, interact with referees and
codify tacit knowledge (Kinney et al. 2004). Furthermore, firm scientists who publish are more visible
to competing firms, which may impose the need for retention strategies as their external employment
options increase (Stuart and Liu 2014; Kim and Marschke 2005). In summary, the positive and negative
effects of the relationship between corporate science and firm value underline the need for further study
in this area.

3 Data and Methodology

3.1 Data sources

Our analysis was based on a sample of US public companies from high-technology industries, as defined
by the OECD (OECD 2011), including pharmaceuticals and biotechnology, telecommunication

-4-

Electronic copy available at: https://ssrn.com/abstract=2317166

equipment and semiconductors, aircraft, as well as scientific and medical instruments. We also included
the chemicals sector since it is often recognized as science-oriented. These industries were a natural
choice as they draw on scientific knowledge as an input factor, while their publication output is varied
(Cohen et al. 2002; Simeth and Raffo 2013). Firm-level data came from Standard and Poor’s Compustat,
patent data from the Worldwide Patent Statistical Database (Patstat, version April 2014) provided by
the European Patent office, while publication data came from Elsevier’s Scopus database. As coverage
of Scopus changed considerably in 1996, we were obliged not to include earlier periods in our analysis.1
Moreover, the patent information concerning the United States Patent and Trademark Office (USPTO)
patents contains the caveat that applications filed before 29th November 2000 but not granted are
unobservable.2 In order to keep the numbers of patents comparable over time, we therefore only
considered patent applications where we observed the grant within seven years after the application date.
This allowed for the inclusion of patent data until 2006. Consequently, the sample covers the time period
1996-2006, and firms were considered for selection based upon the criterion that they invested in R&D
in at least one of the years during this period.

Combining firm-level data with publication and patent information may lead to errors arising from
name-based matching procedures. To achieve high recall rates of publication and patent numbers, we
carefully pre-cleaned all firm names by correcting misspellings and removing abbreviations denoting
corporate structure (e.g. “Inc.”). The publication and patent matching processes differed as scientific
publications were retrieved manually from the Scopus online database, whereas patent data was matched
using an offline source. The former is a more time-consuming process than “automatic” offline
matching, and the latter requires extensive manual cleaning of the algorithm-based matching results.
Manual retrieval of publication data enabled us to directly inspect hits and immediately exclude other
institutions with an identical name. The process is explained in further detail in Appendix A.1. For the
offline matching procedure with patent data from the Patstat database we first pre-tested all names to
detect problematic cases. Firms with ambiguous names were either excluded directly or marked for
detailed manual inspection once the actual matching took place. Second, we applied name
standardization routines both to the names of our sample firms and the applicant field in the Patstat
database. Third, after testing several matching algorithms, we identified the one which achieved the
highest recall rate while simultaneously limiting false positive hits (see Raffo and Lhuillery 2009). The
resulting matches were checked manually with particular focus on firms with atypical input-output ratios
and those identified in the pre-tests as being uncertain. In order to cross-check the quality of our patent-

1
Scopus extended its journal coverage considerably in 1996, potentially inducing bias to the econometric
estimations. See http://www.info.sciverse.com/UserFiles/2508.SciVerse.Scopus_Facts_Figures%28LR%29.pdf.
2
The policy change was introduced by the American Inventors Protection Act (AIPA) and concerns all patents
that seek protection in the United States (except patents that enter into the PCT procedure for worldwide protection
and “opt-out” cases). As of 29th November 2000 all patent applications are disclosed with the search report of the
patent examiner after 18 months.

-5-

Electronic copy available at: https://ssrn.com/abstract=2317166

matching process, we also combined our data with the Compustat-NBER dataset (see Hall et al. 2001),
performing regressions on the patent numbers based first on our matching and then on Compustat-NBER
matching. The results were consistent. Since the Compustat-NBER dataset also considers static
ownership structure (for the year 1989), this exercise suggests that our results were not affected by our
inability to take into account subsidiaries of the studied companies in the present study. However, it is
still theoretically possible that we missed a relevant number of publications when not considering
subsidiaries. To mitigate such concerns, we searched for information on subsidiaries for some of the
investigated sectors (Biotech SIC 2845 & 2846, Semiconductors SIC 3674, Instruments SIC 3841 &
3842) and downloaded publications taking into account the names of the subsidiaries using the SEC
security filings (report “10-K”). The corresponding regressions suggested that no bias arose from the
non-consideration of subsidiaries in our analysis (see Appendix A.2).

To reduce biases originating from Merger & Acquisition (M&A) activities, measurement errors
and “atypical” firms (e.g. specialized R&D firms with a majority of shares owned by business groups),
several filters were applied to the sample. First, to limit potential biases from M&A events, firms with
large changes in book value were identified based on the criteria of an increase of more than 300% or
decrease of more than 75% over a period of two consecutive years (Griliches 1981; Hall and Oriani
2006; Aldieri and Cincera 2009). We only dropped the firm-year observation when a large change
occurred. In such cases, we treated the firm in subsequent years as a new firm (Griliches and Mairesse
1984). Excluding firms involved in M&A activities entirely could have led to selection bias since M&A
activities are presumably often based on successful R&D operations. To address measurement errors
and inaccurate initial knowledge stock computations, we also dropped observations with extremely high
(the top 1%) knowledge-to-asset ratios.3 Finally, we excluded firms with fewer than 10 employees,
imposed a minimum sales amount of 500,000 USD, and removed firms with an R&D/Sales ratio higher
than one. A firm with no sales generates a high degree of uncertainty among investors in terms of its
survival prospects. Such firms are therefore at a different structural stage than those already generating
a cash flow, the result likely being heterogeneous treatment on the stock market. In our study, since the
R&D/Sales ratio filter affected a notable number of observations, we removed it as a robustness test (see
Section 4.4). The final sample size contained 9,920 firm-year observations, composed as follows:

-- Insert Table 1 here --

Finally, all financial amounts were adjusted for inflation using the Gross Domestic Product (GDP)
deflator.

3
The three measures concerned are R&D/A, PAT/R&D, PUB/R&D which are explained in detail in section 3.2.

-6-

Electronic copy available at: https://ssrn.com/abstract=2317166

3.2 Variables

To construct our dependent variable, we applied the methodology followed by previous related research
and used the Tobin’s Q which represents the ratio of a firm’s market value to its book value. Market
value is the addition of the market value of the equity and the market value of the debts. The former is
calculated by the number of outstanding shares multiplied by the stock price at the end of the fiscal year,
whereas the latter is approximated using the book value of liabilities (Blundell et al. 1999; Hall and
Oriani 2006, Ceccagnoli 2009). The firm’s book value is represented by its assets at the end of the fiscal
year. With respect to the independent variables, one core measure is the firm’s R&D investments which
reflect overall commitment to knowledge production. The investments in R&D take place in a particular
period, but the returns on these investments may last much longer. Therefore, we introduced a stock
measure of R&D. Since knowledge becomes obsolete because of ongoing technological development,
we applied the frequently used perpetual inventory method with an annual 15% depreciation ( ) rate
(Griliches and Mairesse 1984; Hall et al. 2005). In the absence of a strong theoretical justification for
assuming different rates for the other knowledge-related measures, we not only computed the R&D
stock but also the publication and patent stock indicators with the 15% rate as follows:

& = & + 1− ) & (1)

Although the computation of the R&D stocks is technically straightforward, assumptions have to
be made regarding the initial stock ( & ) which remains partly unobserved. In our study,
we applied a standardized growth rate (g) for R&D and the other knowledge stock measures of 8%,
given that our sample contained only high-technology firms (Hall and Oriani 2006; Hall et al. 2007).4
All knowledge-related measures discussed below are accordingly constructed as stocks. We constructed
three measures directly related to the firms’ scientific operations. First, we use a patent-based indicator
to measure outcomes from basic research, and thus indirectly also measure a firm’s potential to create
scientific publications (SCIPAT). It is assumed that the existence of a scientific document in the
backward references of a patent indicates a science-based patent (Narin et al. 1997; Deng et al. 1999).
We identified scientific documents in the Non-Patent-Literature (NPL) section of patents’ backward
references using specific keywords and character combinations collected via extensive screening of the
raw NPL information. Since this measure captured research outcomes but did not depend on the
observation of scientific contributions to academic journals, the use of SCIPAT, in combination with the
publication-based indicators, enabled us to differentiate between research outcomes and disclosure

4
This applies specifically to firms that do not have a long pre-sample record in Compustat. The stocks can be (at
least partly) computed with observed values for firms with IPO’s before 1996. In formal terms, the initial R&D
stock is approximated as follows: & = & / + )

-7-

Electronic copy available at: https://ssrn.com/abstract=2317166

effects.5 Although this measure of basic research relies on patented outcomes, we nonetheless
considered it to be a reasonable measure, given that our sample comprised only firms from high-tech
sectors with a high propensity to patent.

Second, we captured the number of voluntary scientific contributions by computing a stock

measure based on the number of scientific papers published by each firm (PUB). Given our use of
SCIPAT to measure basic research outcomes, the publication stock should not only reflect research
productivity but also effects deriving from the disclosure of results. Moreover, to consider the
heterogeneity of scientific contributions, we introduced a second measure reflecting their academic
quality (TOPPUB). If firms are able to publish in prestigious journals, it is likely that the degree of
involvement in the scientific community will be higher. Publishing in highly ranked journals is also less
likely to reflect appropriation motives since the publication process is riskier with higher likelihoods of
rejections and delays imposed by referees’ requests for further experiments before accepting the
submitted article. In order to overcome potentially imperfect corrections for field effects and inaccurate
weightings of articles by impact factors, we decided to rely on a simpler dichotomous distinction by
identifying the top 10% of journals based on the impact factor distribution within the five meta-
disciplines: Life Sciences, Physical Sciences and Engineering, Social Sciences, Health Sciences and
General/Interdisciplinary. The computed stock variable therefore only considered scientific documents
published in these top journals (see Appendix A.3).

In line with previous literature, we included two measures reflecting the amount of inventive
outputs and their quality. The absolute amount of inventive outcomes is represented by patent stocks
(PAT) and the quality of the inventive outcomes (FWDCIT) is measured using forward citation counts
(Hall et al. 2005). To deal with the truncation problem of citation counts, we considered only those
citations that occurred within a five-year window after the priority date of our focal patents (Lanjouw
and Schankerman 2004; Marco 2007). Finally, we included sector and year dummy variables to take
into account heterogeneous market valuations across industries and time (Cockburn and Griliches 1988;
Hall et al. 2005).

3.3 Model and estimation techniques

In this paper, we analyze the relative market value of firms (Tobin’s Q) as a function of their knowledge
stocks. We rely on the well-established market value function (Griliches 1981, Hall et al. 2005) which

5
Taking into account the possibility that SCIPAT might not exhaustively capture the basic research outcomes, we
also included a stock of “original” patents based on the patent originality measure proposed by Trajtenberg et al.
(1997). The results of this specification are reported in the appendix A.4 and are consistent with those of the main
model.

-8-

Electronic copy available at: https://ssrn.com/abstract=2317166

regards tangible (Ait) and intangible assets (Kit) as additives (“hedonic model”). The function can be
formalized as follows:

, )= , ) (2)

In this equation, qit represents the valuation coefficient of a firm’s assets, and the parameter γ allows
for the eventuality that knowledge assets are valued differently from physical assets. The valuation
coefficient qit may vary across time and industries and also contains a firm-specific component. The
factor σ represents scale effects and is assumed to equal 1, which is also confirmed in our data (Hall et
al. 2005). Applying logarithms on both sides and moving tangible assets to the left-hand side of the
equation yields the following expression:

# *$%
Log "&$% ' = log ) = log + log "1 + '++ (3)
$% &$%

Following our theoretical discussion, we separate the knowledge intangibles of firms into R&D,
patent stocks, and publication stocks. Since patents and scientific publications can be regarded as direct
outcomes of R&D inputs, we introduce these additional measures as ratios denominated by R&D
expenditures (Hall et al. 2005; Hall and Oriani 2006; Hall et al. 2007), formalized in Equation (4):

# ,&-$% /&0$% /23$%

Log " $% ' = log ) = log + log "1 + + . ,&- + 1 ,&- '+ + (4)
&$% &$% $% $%

In our empirical setting, R&D, PAT and PUB represent the respective stock measures computed
using the perpetual inventory method with an assumed depreciation rate of 15%. In the full specification,
we distinguish between the impacts of (basic) research and disclosure effects using further measures
that are introduced as ratios and which are orthogonal to the main variables. These further variables
reflect science-based inventions (SCIPAT/PAT), the quality of the inventive output (FWDCIT/PAT), and
the academic impact of scientific output (TOPPUB/PUB). Consequently, the extended model of
Equation (4) can be written as:

# ,&-$% /&0$% /23$% 789/&0$%

45 "&$% ' = log ) = log + log "1 + + . ,&- + 1 ,&- + 6 +
$% &$% $% $% /&0$%

;<-890$% 0>//23$%
+ : + = '+ + (5)
/&0$% /23$%

-9-

Electronic copy available at: https://ssrn.com/abstract=2317166

Equations (4) and (5) can be directly estimated using Non-Linear Least Squares (NLLS). Instead, OLS
regression models can be applied only by using the approximation log (1+x) ~ x. Nevertheless, higher
values of x lead to imprecisions in the estimated outcome. An important concern is the potential presence
of unobserved firm heterogeneity. However, independent variables are potentially predetermined since
past firm valuations may influence current investment decisions related to R&D strategies. This
eventuality violates the strict exogeneity assumption for the “within” estimator. Moreover, unlike the
rather volatile Tobin’s Q indicator, the innovation-related variables change slowly over time and are
thus highly correlated with the firm-specific effect. Consequently, a “within” fixed-effects estimator
may rather exacerbate estimation problems than allow valid estimates to be obtained (see also Hall et
al. 2005). A potential solution, suggested by Blundell et al. (1999), is the introduction of a pre-stock
average of the firm’s market value as an additional regressor. We tested this approach, but it led to a
substantially decreased sample since many firms do not have a sufficiently long stock market history.
Consequently, like most of the related empirical literature, we decided to focus on NLLS regression
models.

3.4 Descriptive statistics

In Table 2 we provide an overview of the mean and median values as well as the standard deviation of
the regression variables and selected additional measures that provide information about firm
characteristics.

-- Insert Table 2 here --

Since all firms were stock market-listed, the sample consisted predominantly of medium- and
large-sized firms. However, median values suggest considerable heterogeneity among our sample of
firms. The median values for annual R&D expenditures were 8.36 million USD, 79.41 million USD for
sales and 340 for the number of employees, whereas the mean values were much higher, indicating the
presence of both large but also medium-sized and small firms. The average Tobin’s Q ratio was quite
high at 2.65. In 12.3% of the firm-year observations, the market valuation was below the book value.
With regard to trends in publication and patent outputs, the share of both publishing and patenting firms
increased over time. Time trends are shown in Figures 1 and 2.

-- Insert Figure 1 & 2 here --

The average number of publications and patents by a firm both increased over time. At the
beginning of the sample period the number of patents continuously increased while the number of

-10-

Electronic copy available at: https://ssrn.com/abstract=2317166

publications remained stagnant, before starting to increase in 2002. Interestingly, the average level of
publication and patent activity did not differ greatly, which is remarkable given that firms are, by
definition, not concerned with contributions to the stock of scientific knowledge per se. With regard to
the share of firms with at least one publication or patent in a given year, we obtained the following
picture as reported in Figure 2: the share of publishing firms increases from 37% in 1996 to 53% in
2006, whereas the share of patenting firms increased from 46% in 1996 to 60% in 2006. Output types
appeared to be complementary since the majority of publishing firms in a given year also filed a patent
application. Over time, there was an increase in the strategy to patent and publish simultaneously. In
2006 however, 14% of all firms only published, 17% only created patents, and 26% of firms neither
created patents nor published.

4 Econometric results and discussion

4.1 Full sample estimations

The core specifications of the econometric analysis are shown in Table 3. Considering the elements
highlighted in the previous section, we estimated Tobin’s Q with both non-linear and linear regression
models, but ultimately focused on the interpretation of the former since non-linear models enable the
theoretical model to be directly represented. In order to evaluate the quantitative impact of our variables
in the NLLS regressions, we also computed semi-elasticities (at the mean values of the variables) which
are reported in complementary columns for the baseline (2) and full (6) NLLS regression models.6

-- Insert Table 3 here --

In columns (1)-(2), we estimated baseline specifications using NLLS and OLS estimators. All three
knowledge measures representing R&D (R&D/A), patent (PAT/R&D) and publication (PUB/R&D)
stocks had a positive and significant effect on Tobin’s Q, the latter two having similar magnitudes. One
additional patent per million dollars of R&D was associated with an increase of 9% in Tobin’s Q, and
10% for an additional publication per million dollars of R&D. This result provided a first indication that
scientific activities have an important impact on the market value of firms. In order to obtain a more
detailed picture, we introduced the additional measures described in Section 3.3. In columns (3)-(5), we
first separately added the three variables FWDCIT/PAT, SCIPAT/PAT and TOPPUB/PUB to the
baseline model. All three measures exhibited a value premium for Tobin’s Q. The main variables
remained statistically significant when the supplementary patent-based variables were included.

6
Semi-elasticities can be obtained by calculating the derivative of the estimated market value equation with
regard to the variable of interest. See Hall et al. (2005) for further details.

-11-

Electronic copy available at: https://ssrn.com/abstract=2317166

However, when adding the measure capturing the share of publications in prestigious journals
(TOPPUB/PUB) in column (5), the magnitude of the publication stock PUB/R&D decreased, and the
variable only remaining significant at the 10% level. In columns (6)-(7), we report the regression results
with the full variable set. In the NLLS specification (6), it can be seen that R&D stocks, patent stocks
and patent citation stocks (FWDCIT/PAT) had a positive impact, which is in line with findings from
previous studies (e.g. Hall et al. 2005). While publication stocks (PUB/R&D) were only significant at
the 10% level, we detected a strong positive and highly significant effect of publications in top journals.
The increase of the variable by one unit – which is equivalent to a change from none of a firm’s
publications being in top journals to all of the firm’s publications being in top journals - increased
Tobin’s Q by approximately 12%.7 Interestingly, the impact of science-based patents (SCIPAT/PAT)
became substantially weaker in comparison to column (4), with the positive effect of a one unit change
decreasing from 11% to 4%, and the variable becoming statistically insignificant once the share of
publications in top journals was taken into account.

Based on these regression results, the question arises as to whether the positive effects of scientific
activities which we found reflect (i) the successful creation of science-based knowledge outcomes or
benefits that derive from the disclosure of these outcomes, (ii) signaling benefits to academic audiences
or the use of publications as a device to support appropriation. We would like to stress that we cannot
provide conclusive evidence concerning these mechanisms with our measures but can only offer some
indications which must be treated with caution. In the full model presented in column (6), positive effects
for scientific publication stocks and especially top journal publications were observed despite the
presence of science-based patents. While being highly significant and exhibiting a notable magnitude in
column (4), the measure of science-based patents was no longer significant in the full specification. This
indicates that the use and absorption of basic research has a positive impact on a firm’s market value.
However, active involvement in science which is especially represented by publications in top journals,
generates a considerable market value premium beyond the successful absorption of scientific
knowledge. The relevance of publication quality also has implications with regard to point (ii) above
and the underlying value-creating mechanisms of scientific disclosure. Firms that disclose knowledge
for defensive purposes have an interest in keeping control over the timing of disclosure. As a
consequence, there are incentives for firms to target lower impact journals as, presumably, the likelihood
of acceptance in high-impact journals is lower and the time taken to publish longer. Therefore, if this
mechanism is value-enhancing, we would expect a stronger effect of the main publication stock
PUB/R&D in the full model. However, the particularly strong effect of publications in high-impact
journals points more to value deriving from facilitated interactions with academic partners. In other

7
Caution must be exercised when comparing the magnitudes between PUB/R&D and TOPPUB/PUB due to the
different units of the variables. Therefore we also calculated the impact of a change by one standard deviation.
The result of this exercise also suggested that publications in prestigious journals are associated with a particularly
strong increase in Tobin’s Q (3.7% vs. 1.6%).

-12-

Electronic copy available at: https://ssrn.com/abstract=2317166

words, firms actively participating in the scientific community seem to have easier access to valuable
knowledge and to highly qualified graduates, which results in a higher Tobin’s Q.

4.2 Sector analysis

Since the association between scientific activities and firm value may differ across industries, especially
in the light of the relative risks of knowledge spillovers, we analyzed the individual sectors in greater
detail. In Table 4, we report the results of subsample regressions with the meta-sectors Biotechnology
and Pharmaceuticals (hereafter BIO-PHARM), Information & Communication technologies (ICT),
Scientific and Medical instruments (INST), and Chemicals (CHEM).

-- Insert Table 4 here –

Starting with the baseline specification, it can be seen that the effect of scientific publication stocks
differs across the meta-sectors. Although we detected a positive but non-significant sign in BIO-
PHARM and CHEM, a negative and non-significant effect was found in ICT, while a positive and
significant effect was observed for the INST sectors. Estimating the regression model with the full set
of variables revealed interesting patterns. In BIO-PHARM, the publication quality indicator was
significant and of quite a strong magnitude, whereas the variables reflecting the main publication stock
and science-based patents did not have such an impact. This result suggests that as long as a firm is not
seriously committing to Open Science, scientific activities will not be sufficient to generate a value
premium. We observed a similar result for the INST subsample, where the TOPPUB/PUB measure had
a strong positive effect too. However, the main publication variable PUB/R&D also showed a rather
strong magnitude in comparison to the other subsamples, and remained significant. This finding suggests
that in addition to upstream signaling, publications may create value through mechanisms other than
establishing knowledge flows with the scientific community. One plausible interpretation for this
finding lies in marketing effects, since firms in instrument sectors sell to scientifically-trained
professional customers. In order to establish credibility among this audience with regard to the latest
science-based products, publishing in high-impact journals is presumably less crucial. In the ICT
domain, we saw a different picture. Publication stocks were negative and the share of top impact journal
publications also had a negative sign. Interestingly, in addition to the positive effects of patent stocks
and patent quality, the share of science-based patents was positive and significant. In other words,
absorbing scientific knowledge and achieving science-based inventions may create value, whereas
active dissemination of results and participation in the scientific community do not. For the CHEM
sector, we only observed a positive effect of R&D on the market value. However, there was no premium
for patent and publication stocks or for science-based patents.

-13-

Electronic copy available at: https://ssrn.com/abstract=2317166

 These obtained differences across sectors indicate that appropriability conditions have a notable
influence on the returns of scientific disclosure strategies. In sectors with strong legal appropriation, like
the medical instrument and biopharmaceutical industries, we observed positive net returns from
scientific publications, whereas in the ICT domain, which can be characterized as having a weaker
appropriability regime due to its complex and cumulative nature, no premium was observed (see Levin
et al. 1987; Cohen et al. 2000; Hall and Ziedonis 2001). This finding indicates that in the ICT sectors,
scientific disclosure leads to knowledge spillovers which outweigh any benefits. The result for the
Chemicals sector, where appropriability conditions are stronger than those in the ICT sector, can be
explained by the fact that in the former, academic knowledge is only of moderate importance as an input.
Therefore, the potential benefits of scientific disclosure in terms of signaling might be limited in the first
place.

4.3 Robustness tests

Beyond considering subsidiary publications for a subsample of firms, as reported in Appendix A.2, we
performed further robustness checks, the most important of which are displayed in Appendix A.4. These
robustness tests did not change our main models’ results. Consequently, we will only describe them very
briefly. We list several tests and for ease of reading we shall assign them numbers that correspond to
the respective columns in Appendix A.4. (1) We ran fixed-effect regression models. As discussed in
Section 3.3, such models must be interpreted with caution in our context. However, the results suggested
that scientific activities do indeed lead to higher firm values. (2) We removed the filter which initially
excluded firms whose R&D/Sales ratio was larger than one since this filter was perhaps overly
restrictive. However, our results after removing this filter did not change. (3) In order to control for
unobserved selection mechanisms for publication in terms of local spillovers and competition, we
included state dummies. (4)-(5) Since scientific backward references may not exhaustively capture basic
research outcomes in firms, we included the patent originality variable as suggested by Trajtenberg et
al. (1997). This variable behaves very similarly to science-based patents, and supports the interpretation
that the results for publication stocks do not only reflect research productivity effects but also benefits
deriving from the active involvement of firms in science. (6)-(8) We introduced additional control
variables, namely the amount of sales and the growth of sales, sector-level patent and publication
propensities, and the share of firm’s profit on the industry profit. These controls should capture the effect
of unobserved intangibles potentially correlated with our knowledge intangibles, of the abilities of firms
to appropriate returns from R&D, and in the case of the sector-level variables, of incentives to engage
in publication. Our results did not change when these additional controls were included. (9) We used
the patent counts from the Compustat-NBER matching (Hall et al. 2001; Cockburn et al. 2009) to verify
the quality of our own matching. The regression outputs were very similar. (10)-(11) We modelled the

-14-

Electronic copy available at: https://ssrn.com/abstract=2317166

market value function in a different way by including the publication stock to patent stock ratio (i.e.
instead of publication stock to R&D stock ratio) for the subsample of patenting firms. Such a
specification might be an adequate measure of openness if publications and patents reflect the same
knowledge. The variable PUB/PAT was found to be positive and significant, supporting the
interpretation of the core models that publication provides additional value beyond patented knowledge.
(12) We included an additional measure of the share of publications co-authored with academic
institutions. This variable had a positive but not significant effect when added to the model with the full
set of variables.

5 Conclusion

This study examines the impact of scientific activities of firms on their stock market valuation using a
dataset of firm-level information for US high technology companies, combined with scientific
publication and patent data. Although scholars have recently started to conceptually and empirically
address the determinants of scientific activities by firms, in particular disclosure, there is very little data
for firm performance effects. Thanks to its design, our analysis provides an empirical contribution not
only to the growing literature on boundary-spanning activities of firms, but also to more mainstream
studies assessing the performance implications of R&D in firms.

Our analysis documents the positive impact of science-related indicators on a firm’s market
valuation beyond the effects of R&D and inventive outcome indicators. Our measures enabled us to
differentiate between the profitability implications of performing research and a more active
involvement in open science. While we cannot entirely rule out the possibility that we also capture
research productivity effects, our findings suggest that active involvement in science, as reflected by
disclosure in scientific journals, results in higher stock-market values. Furthermore, our results indicate
in particular that the positive effects of such disclosure stem from scientific signaling to upstream
stakeholders, which results in superior knowledge flows from these persons and institutions. In other
words, publishing allows firms to become members of the scientific community and to establish formal
and informal interactions, which in turn provide the former with access to state-of-the-art developments
and research techniques. On the other hand, we found little evidence for a value-enhancing use of
publications for appropriation purposes. Overall, our study provides support for the increasingly strong
view among scholars that minimizing knowledge outflows may not be an optimal choice for firms and
that they should instead strategically disclose knowledge.

Some heterogeneity concerning the impact of scientific activities was found. A negative but non-
significant sign was found for ICT sectors. Instead, a strong positive relationship was observed in the
instruments sectors. Therefore, appropriability conditions would seem to matter, which implies that
R&D managers need to carefully consider the benefits and potential costs caused by related spillovers.

-15-

Electronic copy available at: https://ssrn.com/abstract=2317166

To obtain a better understanding of the mechanisms involved, future work could focus on contextual
conditions that potentially moderate the relationship between scientific activities and the firm’s market
value. In conclusion, the strengths and weaknesses of the market value approach and corresponding
empirical limitations should be kept in mind when interpreting results. Financial markets can be very
volatile, and do not behave completely rationally. Thus, future research could complement our present
study by using direct financial performance measures. This however, would entail dealing explicitly
with the time lags between knowledge creation and commercial returns.

Acknowledgements

The authors thank Dirk Czarnitzki, Dominique Foray, Stuart Graham, Karin Hoisl, Stéphane Lhuillery,
and Christian Sternitzke for valuable comments. The authors also thank audience members at
conferences and university seminars in Torino, Mannheim, Barcelona, Bordeaux, Seville, Lisbon, Nice,
and Madrid for helpful suggestions. The usual disclaimer applies. Markus Simeth acknowledges
financial support from the Swiss National Science Foundation, grant number P2ELP1-151980.

-16-

Electronic copy available at: https://ssrn.com/abstract=2317166

References

Aldieri L, Cincera M (2009) Geographic and technological spillovers within the Triad: Micro evidence
from US patents. Journal of Technology Transfer 34(2): 196-211.
Arrow, KJ (1962) Economic welfare and the allocation of resources for innovation,’ in: HM Groves
(ed.) The rate and direction of inventive activity: Economic and social factors. Princeton University
Press, Princeton.
Azoulay P (2002) Do pharmaceutical sales respond to scientific evidence? Journal of Economics and
Management Strategy 11(4): 551-594.
Belenzon S (2012) Cumulative innovation: Evidence from patent citations. Economic Journal 122(559):
265-285.
Blundell R, Griffith R, Van Reenen J (1999) Market share, market value and innovation in a sample of
British manufacturing firms. Review of Economic Studies 66(3): 529-554.
Ceccagnoli M (2009) Appropriability, preemption, and firm performance. Strategic Management
Journal 30(1): 81-98.
Cockburn IM, Griliches Z (1988) Industry effects and appropriability measures in the stock market’s
valuation of R&D and patents. American Economic Review 78(2): 419-423.
Cockburn IM, Henderson RM (1998) Absorptive capacity, coauthoring behavior, and the organization
of research in drug discovery. The Journal of Industrial Economics 46(2): 157-182.
Cockburn IM, Henderson RM, Stern S (1999) Balancing Incentives: The tension between basic and
applied research. NBER Working Paper 6882.
Cockburn IM, Agrawal A, Bessen J, Graham JHS, Hall BH, MacGarvie M (2009). The NBER patent
citation datafile update. https://sites.google.com/site/patentdataproject/Home
Cohen WM, Levinthal DA (1989) Innovation and learning: the two faces of R&D. The Economic
Journal 99(397): 569-596.
Cohen WM, Nelson RR, Walsh JP (2000). Protecting their intellectual assets: Appropriability conditions
and why U.S. manufacturing firms patent (or not). NBER Working Paper No. 7552.
Cohen WM, Nelson RR, Walsh JP (2002). Links and impacts: the influence of publish research on
industrial R&D. Management Science 48 (1), 1-23.
Czarnitzki, D, Hall BH, Oriani R (2006). Market Valuation of US and European Intellectual Property.
In Bosworth, D. and E. Webster (eds), The Management Of Intellectual Property, Cheltenham, UK,
Edward Elgar.
De Fraja G (1993) Strategic spillovers in patent races. International Journal of Industrial Organization
11(1): 139-146.
Della Malva A, Hussinger K (2012) Corporate science in the patent system: an analysis of the
semiconductor technology. Journal of Economic Behavior and Organization 84(1): 118-135.
Deng Z, Lev B, Narin F (1999) Science and Technology as predictors of stock performance. Financial
Analysts Journal 55(3): 20-32.
Fabrizio KR (2009) Absorptive capacity and the search for innovation. Research Policy 38(2): 255-267.
Fleming L, Sorenson O (2004) Science as a map in technological search. Strategic Management Journal
25(8-9): 909-928.
Gittelman M, Kogut B (2003) Does good science lead to valuable knowledge? Biotechnology fims and
the evolutionary logic of citation patterns. Management Science 49(4): 366-382.
Griliches Z (1981) Market value, R&D, and patents. Economic Letters 7(2): 183-187.

-17-

Electronic copy available at: https://ssrn.com/abstract=2317166

Griliches, Z, Mairesse J (1984) Productivity and R&D at the Firm Level, NBER Chapters, in: R & D,
Patents, and Productivity, pages 339-374, National Bureau of Economic Research.
Hall BH (1993) The stock market’s valuation of R&D investment during the 1980’s. American
Economic Review 83(2): 259-264.
Hall BH (2000). Innovation and Market Value. In Barrell, Ray, Geoffrey Mason, and Mary O’Mahoney
(eds.), Productivity, Innovation and Economic Performance, Cambridge: Cambridge University
Press, pp. 175-198.
Hall BH, Jaffe, AB, Trajtenberg, M (2001). The NBER Patent Citation data file: Lessons, Insights and
Methodological Tools. NBER Working Paper 8498.
Hall BH, Jaffe A, Trajtenberg M (2005) Market value and patent citations. RAND Journal of Economics
36(1): 16-38.
Hall BH, Oriani R (2006) Does the market value R&D investment by European firms? Evidence from
a panel of manufacturing firms in France, Germany, and Italy. International Journal of Industrial
Organization 24(5): 971-993.
Hall BH, Thoma G, Torrisi S (2007) The market value of patents and R&D: Evidence from European
firms. NBER Working Paper 13426.
Hall BH, Ziedonis RH (2001) The patent paradox revisited: An empirical study of patenting in the U.S.
semiconductor industry, 1979-1995. RAND Journal of Economics 32(1):101-128.
Hicks D (1995) Published papers, tacit competencies and corporate management of the public/private
character of knowledge. Industrial and Corporate Change 4(2): 401-424.
Jaffe AB (1986) Technological opportunity and spillovers of R&D: Evidence from firms’ patents,
profits, and market value. American Economic Review 76(5): 984-1001.
Kim J, Marschke G (2005) Labor mobility of scientists, technological diffusion, and the firm’s patenting
decision. RAND Journal of Economics 36(2): 298-317.
Kinney AJ, Krebbers E, Vollmer SJ (2004) Publications from industry - Personal and Corporate
Incentives. Plant Physiology 134(1): 11-15.
Lanjouw JO, Schankerman M (2004) Patent quality and research productivity: Measuring innovation
with multiple indicators. Economic Journal 114(495): 441-465.
Levin RC, Klevorick AK, Nelson RR, Winter SG (1987). Appropriating the returns from industrial
research and development. Brookings Papers on Economic Activity 1987(3), 783-831.
Stuart TE, Liu C (2014) Positions and rewards: The allocation of resources within a science-based
entrepreneurial firm. Research Policy 43(7), 1134-1143.
Marco AC (2007) The dynamics of patent citations. Economic Letters 94(2): 290-296.
McGahan AM, Silverman BS (2006) Profiting from technological innovation by others: The effect of
competitor patenting on firm value. Research Policy 35(8): 1222-1242.
Narin F, Hamilton KS, Olivestro D (1997). The increasing linkage between US technology and public
science. Research Policy 26, 317-330.
Nelson RR (1959) The simple economics of basic scientific research. The Journal of Political Economy
67(3): 297-306.
OECD (2011) ISIC Rev. 3 Technology intensity definition: Classification of manufacturing industries
into categories based on R&D intensities. OECD Directorate for Science, Technology & Industry:
[http://www.oecd.org/sti/industryandglobalisation/48350231.pdf]; Last access: 17th October 2012.
Parchomovsky G (2000) Publish or perish. Michigan Law Review 98(4): 926-952.
Penin J (2007) Open knowledge disclosure: an overview of the evidence and economic motivations.
Journal of Economic Surveys 21(2): 326-347.

-18-

Electronic copy available at: https://ssrn.com/abstract=2317166

Pisano GP, 2006. Science Business. The promise, the reality, and the future of biotech. Harward
Business Review Press, Boston.
Polidoro F, Theeke M (2012) Getting competition down to a science: The effects of technological
competition on firms’ scientific publications. Organization Science 23(4): 1135-1153.
Raffo J, Lhuillery, S (2009) How to play the “Names Game”: Patent retrieval comparing different
heuristics. Research Policy 38(10): 1617-1627.
Rosenberg, N (1990) Why firms do basic research (with their own money). Research Policy 19(2): 165-
174.
Sauermann H, Roach M (2013) Not all scientists pay to be scientists: PhD’s preferences for publishing
in industrial employment. Research Policy 43(1), 32-47.
Simeth M, Raffo JD (2013) What makes companies pursue an Open Science strategy? Research Policy
42(9): 1531–1543.
Stern S (2004) Do scientists pay to be scientists? Management Science 50(6): 835-853.
Trajtenberg M, Henderson R, Jaffe A (1997). University versus corporate patents: A window on the
basicness of invention. Economics of Innovation and New Technology 5 (1): 19-50.

-19-

Electronic copy available at: https://ssrn.com/abstract=2317166

TABLES AND FIGURES

Table 1: Observations according to included sectors

Firm-year
M eta-Sector SIC included observations

Biotechnology & Pharmaceuticals 2834, 2835, 2836 1 602

Chemicals
2800, 2810, 2820, 2821, 2833 389

3570, 3571, 3572, 3575, 3576,

Information and Communication Technologies
3577, 3578, 3579, 3661, 3663,
(ICT)
3669, 3670, 3672, 3674, 3677,
3678, 3679, 4812, 4813, 4822 4 373

Aircraft & Aerospace 3721, 3724, 3728 166

Navigation, Scientific, M edical, and Optical 3812, 3822, 3823, 3824, 3825,
instruments 3826, 3827, 3829, 3841, 3842,
3843, 3844, 3845, 3851, 3861 3 390

Sum 9 920

Table 2: Summary statistics

Variable N M ean Std. Dev. M edian M in M ax

TOBIN'S Q 9920 2.65 2.39 1.89 0.25 23.02

A (BOOK VALUE) 9920 1548.56 7172.64 94.33 0.55 273007.30

R&D (FLOW) 9920 104.29 468.25 8.36 0.00 12942.19

R&D (STOCK) 9920 491.47 2082.16 39.96 0.00 41408.40
R&D / A 9920 0.61 0.71 0.39 0.00 7.42

PAT (STOCK) 9920 114.33 559.85 5.57 0.00 9969.30

PAT / R&D 9920 0.31 0.46 0.15 0.00 3.68
SCIPAT / PAT 9920 0.27 0.31 0.17 0.00 1.00
FWDCIT / PAT 9920 5.59 7.50 4.38 0.00 198.65

PUB (STOCK) 9920 69.54 347.81 1.77 0.00 6440.20

PUB / R&D 9920 0.15 0.29 0.04 0.00 2.85
TOPPUB / PUB 9920 0.20 0.29 0.00 0.00 1.00

SALES 9920 1229.41 5248.79 79.41 0.50 88784.06

EM PLOYEES 9738 4076.93 14324.04 340.00 10.00 238000.00
Monetary amounts in Mio USD (2005, GDP deflated)

-20-

Electronic copy available at: https://ssrn.com/abstract=2317166

 Table 3: Regression outputs
(1) (2) (3) (4) (5) (6) (7)
LOG TOBIN'S Q OLS NLLS NLLS NLLS NLLS NLLS OLS
Coeff (SE) Coeff (SE) Semi-elast. Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Semi-elast. Coeff (SE)

R&D/A 0.048*** 0.055*** 0.051 0.055*** 0.052*** 0.051*** 0.050*** 0.045 0.042***
Electronic copy available at: https://ssrn.com/abstract=2317166

(0.011) (0.014) (0.015) (0.014) (0.014) (0.015) (0.011)

PAT/R&D 0.083*** 0.097*** 0.090 0.071*** 0.083*** 0.097*** 0.072*** 0.064 0.066***
(0.021) (0.025) (0.026) (0.025) (0.025) (0.026) (0.021)
PUB/R&D 0.095*** 0.110*** 0.102 0.108*** 0.091** 0.066* 0.065* 0.057 0.056*
(0.032) (0.039) (0.039) (0.039) (0.038) (0.039) (0.032)
FWDCIT/PAT 0.007*** 0.006*** 0.005 0.004**
(0.002) (0.002) (0.002)
SCIPAT/PAT 0.124*** 0.046 0.041 0.052
(0.035) (0.037) (0.032)
TOPPUB/PUB 0.168*** 0.142*** 0.126 0.125***
(0.040) (0.041) (0.034)

Firm-Year observations 9,920 9,920 9,920 9,920 9,920 9,920 9,920

Firm-IDs (cluster) 1.739 1.739 1.739 1.739 1.739 1.739 1.739
R2 0.129 0.130 0.136 0.134 0.136 0.141 0.141
Standard errors (clustered by firm) reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

All regression models contain year and SIC-4 digit industry dummies. The semi-elasticities reported in the table are computed at the mean values of the variables.
The semi-elasticities of the added variables in columns (3)-(5) are as follows: FWDCIT/PAT 0.007, SCIPAT/PAT 0.113, and TOPPUB/PUB 0.153.

-21-
 Table 4: Heterogeneity by sectors
(1) (2) (3) (4) (5) (6) (7) (8)
BIO-PHARM ICT INST CHEM
LOG TOBIN'S Q NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS
Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE)
Electronic copy available at: https://ssrn.com/abstract=2317166

R&D/A 0.126*** 0.108*** 0.022 0.022 0.086*** 0.075** 0.395* 0.389*

(0.040) (0.041) (0.017) (0.018) (0.029) (0.030) (0.208) (0.206)
PAT/R&D -0.040 -0.028 0.142*** 0.110** 0.103*** 0.079** 0.067 -0.010
(0.067) (0.075) (0.049) (0.048) (0.036) (0.038) (0.115) (0.094)
PUB/R&D 0.133 0.045 -0.009 -0.014 0.218*** 0.149** 0.115 0.058
(0.088) (0.086) (0.049) (0.050) (0.075) (0.075) (0.169) (0.140)
FWDCIT/PAT 0.006 0.003 0.009* 0.022
(0.007) (0.002) (0.005) (0.020)
SCIPAT/PAT -0.037 0.129** 0.021 -0.098
(0.074) (0.058) (0.070) (0.141)
TOPPUB/PUB 0.225** -0.015 0.243*** 0.026
(0.090) (0.054) (0.077) (0.137)

Firm-Year observations 1,602 1,602 4,373 4,373 3,390 3,390 389 389
Firm-IDs (cluster) 331 331 762 762 565 565 57 57
R2 0.071 0.088 0.120 0.127 0.129 0.156 0.136 0.150
Standard errors (clustered by firm) reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1
All regression models contain year and SIC-4 digit industry dummies.
Regression models represent baseline and full model specification for meta-sectors biotechnology and pharmaceuticals (BIO-PHARM ) in columns (1)
and (2), for information and communication technologies (ICT) in (3) and (4), for instruments (INST) in (5) and (6), and for Chemicals (CHEM ) in (7)
and (8). See Table 1 for information on SIC codes included in the meta-sectors. The regression results for the Aircraft sector are omitted in this table
due to the very low number of observations. The semi-elasticities of the publication-based measures are as follows: PUB/R&D : 0.039 in column (2), -
0.014 in (4), 0.125 in (6), 0.058 in (8). TOPPUB/PUB : 0.195 in column (2), -0.014 in (4), 0.204 in (6), and 0.022 in (8).

-22-
Figure 1a: Average publication & patent outputs
30
25
20
15
10

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
Data Year
Average patent count by firm
Average publication count by firm

Figure 1b: Share publishing and patenting firms

.6
.4
.2
0

1996 1998 2000 2002 2004 2006

% PATENTING FIRMS % PUBLISHING FIRMS

% Patent & Publication % Patent only
% Publication only

-23-

Electronic copy available at: https://ssrn.com/abstract=2317166

APPENDIX

Appendix A1: Matching firm-level data with publication data

In the absence of any direct linking identifiers, the firm-level data from COMPUSTAT and data from
SCOPUS have to be combined manually using the firm name (and potential abbreviations) and the
affiliation field in the publication database. The firm names – as listed in the COMPUSTAT database –
contain legal identifiers and other elements which may lead to too few matches (“false negatives”).
Therefore, as a general rule, we reduced the firm names to the most distinctive name components,
removing frequent name elements and legal identifiers in order to achieve high recall rates. Frequent
examples included the words “Pharmaceuticals”, “Technologies”, or “International” which typically
have little identifying power and lead to false negatives when the publication authors do not mention
the full official firm name. Moreover, although the Scopus engine is not sensitive to name capitalisation
and language-specific characters, we harmonized firm names correspondingly since this is a useful step
for post-download processing. It has to be considered that reducing the firm names to a minimum may
have increased the number of false positive search results. A common source of false positives are, for
example, university units or independent firms with a common name component. Thanks to the manual
downloading process, all search results were inspected directly and in the case of ambiguity, a
complementary web search was used to select only eligible publications. Finally, we performed post-
download verification checks by transforming the affiliation information of the different authors of a
particular publication into a “long” format, harmonizing the affiliation field with regard to name
capitalizations and language-specific characters, and applying queries to detect whether the exact field
that created the link between the firm and publication record simultaneously contained an element
pointing to an academic institution. In these rare cases, the publication records were manually checked.

Example: ADHEREX TECHNOLOGIES INC

This name contains with “Technologies” and “INC” two name components that potentially lead to too
few search results. The identifier INC was therefore directly removed. Whereas the search using
“ADHEREX TECHNOLOGIES” led to 21 publications in the period between 1996 and 2010, using
only “ADHEREX” returned one additional publication. After manual inspection of the search results,
we concluded that all 22 publications were eligible and we subsequently downloaded the corresponding
information. Query: AFFILORG (“ADHEREX”) AND PUBYEAR > 1995 AND PUBYEAR < 2007

-24-

Electronic copy available at: https://ssrn.com/abstract=2317166

 Appendix A2: Subsidiary publications
(1a) (1b) (2a) (2b) (3a) (3b) (4a) (4b)
NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS
LOG TOBIN'S Q SUB NO SUB SUB NO SUB SUB NO SUB SUB NO SUB
Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE)
Electronic copy available at: https://ssrn.com/abstract=2317166

R&D/A 0.065* 0.065* 0.080* 0.082* 0.342*** 0.340*** 0.117*** 0.116***

(0.033) (0.033) (0.044) (0.043) (0.088) (0.087) (0.029) (0.029)
PAT/R&D 0.057 0.056 -0.049 -0.051 0.140** 0.139** 0.052 0.051
(0.071) (0.071) (0.093) (0.094) (0.061) (0.061) (0.041) (0.041)
PUB/R&D 0.128 0.136 -0.025 -0.002 0.064 -0.020 0.104* 0.106
(0.092) (0.097) (0.076) (0.104) (0.132) (0.083) (0.062) (0.066)

Firm-Year observations 1,126 1,126 706 706 798 798 2,630 2,630
Firm-IDs (cluster) 186 186 160 160 145 145 491 491
R2 0.123 0.124 0.056 0.055 0.144 0.143 0.069 0.069
Standard errors (clustered by firm) reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1
All models contain year dummies.
Column (1a) and (1b): SIC 3674 (Semiconductors); (2a) and (2b): SIC 2835 & 2836 (In vitro & in ivo diagnostic substance;
Biological products); (3a) and (3b): SIC 3841 & 3842 (Surgical & M edical Instruments, Orthopedic and surgical appliances); (4a)
and (4b): All sectors combined from columns (1)-(3).

-25-
Appendix A3: Definition of top-impact journals

General information

The distinction as to whether a firm publication was of a high quality was made using journal-impact factor
information. The data source for journal impact factors was the SCIMAGO classification, which contains
the SCImago Journal Rank (SJR) indicator based on the records in the Elsevier’s SCOPUS database. The
SCIMAGO ranking is propagated by Elsevier as an official indicator for SCOPUS. Since the journal
rankings are dynamic but change in realty slowly over time, we chose the reference year 2000 which is the
centre year in our sample period.

URL: http://www.scimagojr.com and http://www.journalmetrics.com

Construction of indicator

In order to normalize the indicator for field effects potentially not entirely considered in the SJR indicator,
we created Journal Impact factor rankings specifically for the meta-fields Life Sciences, Physical Sciences,
Health Sciences, General journals, Social Sciences. We created our top-impact journal indicator based on
the distribution of the SJR indicator within the five meta-disciplines, and classified the top 10% of all
journal sources as top-impact journals. The meta-fields contain the following number of journals:

* LIFE SCIENCES: 4,646 sources

* PHYSICAL SCIENCES: 8,916 sources
* HEALTH SCIENCES: 8,483 sources
* GENERAL JOURNALS: 95 sources
* SOCIAL SCIENCES: 6,687 sources

Examples of journal classification in the Innovation Research field

To illustrate the outcome of the classification for journals known in the Management field, some examples
are presented below:

Classified as Top Impact Journals: Management Science, Journal of Industrial Economics, Journal of
Economics and Management Strategy, RAND Journal of Economics; Journal of Economic Behaviour and
Organization, Research Policy

Not classified as Top Impact Journals: Industrial and Corporate Change, Journal of Technology Transfer,
Technovation, Industry and Innovation, Small Business Economics

-26-

Electronic copy available at: https://ssrn.com/abstract=2317166

 Appendix A4: Robustness tests
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
FE R&D Filter LOC D ORIGPAT ORIGPAT Controls Controls Controls NBER PUB/PAT PUB/PAT ACADPUB
LOG TOBIN'S Q WITHIN NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS NLLS
Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE) Coeff (SE)

R&D/A 0.026*** 0.078*** 0.046*** 0.056*** 0.051*** 0.051*** 0.058*** 0.062*** 0.059*** 0.062*** 0.056*** 0.050***
Electronic copy available at: https://ssrn.com/abstract=2317166

(0.009) (0.014) (0.014) (0.015) (0.015) (0.015) (0.015) (0.015) (0.017) (0.017) (0.018) (0.015)
PAT/R&D 0.073*** 0.074*** 0.076*** 0.084*** 0.069*** 0.074*** 0.083*** 0.064*** 0.075** 0.057** 0.075** 0.071***
(0.019) (0.026) (0.025) (0.025) (0.026) (0.028) (0.026) (0.025) (0.030) (0.028) (0.030) (0.026)
PUB/R&D 0.058** 0.044 0.077* 0.102*** 0.064 0.072* 0.036 0.042 0.113*** 0.060
(0.027) (0.034) (0.040) (0.039) (0.039) (0.040) (0.039) (0.039) (0.043) (0.039)
PUB/PAT 0.269** 0.176
(0.111) (0.120)
FWDCIT/PAT 0.001 0.005*** 0.005** 0.005*** 0.006*** 0.007*** 0.006*** 0.006** 0.005***
(0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002)
SCIPAT/PAT -0.039 0.016 0.037 0.036 0.041 0.054 0.050 0.055 0.043
(0.027) (0.034) (0.037) (0.039) (0.040) (0.040) (0.039) (0.042) (0.038)
TOPPUB/PUB 0.033 0.116*** 0.135*** 0.139*** 0.155*** 0.173*** 0.155*** 0.215*** 0.111**
(0.027) (0.037) (0.041) (0.041) (0.044) (0.043) (0.042) (0.049) (0.046)
ORIGPAT/PAT 0.097*** 0.038
(0.032) (0.034)
ACADPUB/PUB 0.052
(0.039)

Firm-Year observations 9,920 11,420 9,920 9,920 9,920 9,154 9,920 9,920 7,392 7,407 7,407 9,920
Firm-IDs (cluster) 1,739 1998 1739 1.739 1.739 1593 1,739 1,739 1.575 1.312 1.312 1.739
2
R 0.089 0.158 0.154 0.132 0.141 0.152 0.112 0.120 0.140 0.140 0.156 0.142

Standard errors (clustered by firm) reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1. M odels contain year and SIC-4 digit industry dummies.

Column (1): FE-"Within" estimator. (2): Filter that excludes R&D/Sales > 1 is deleted. (3): Location dummies (US State level) are added. (4) and (5): Patent originality measure
(ORIGPAT/PAT) is introduced. The threshold for considering a patent as “original” has been set at the mean originality value of all patents filed by the sample firms. (6): Sales and Sales
growth are included as control variables (see Belenzon 2011). The sample size is reduced since Sales growth requires observation of t-1. (7) contains sector-level patent and publication
-27-
intensities (computed at NACE-3 level) as additional control variables and "metasector" industry controls. (8) contains a control variable that captures the share of firm's profits on
industry profits (at SIC-3 digit level). (9): Patentcounts from the combined Compustat-NBER dataset is used instead of the variable obtained by our own matching. Sample limited to the
period 1996-2002 since the NBER patent counts are increasingly incomplete for more recent years. (10) and (11): Alternative consideration of publications as ratio of publication stocks
to patent stocks, for subsample of patenting firms. (12): Share co-authored publication stock with academic institutions is introduced.