Content uploaded by Hanna Hottenrott
Author content
All content in this area was uploaded by Hanna Hottenrott on Apr 09, 2018
Content may be subject to copyright.
Cambridge Journal of Economics 2011, 35, 527–544
doi:10.1093/cje/beq038
Advance Access publication 15 October 2010
Industrial research versus development
investment: the implications of financial
constraints
Dirk Czarnitzki, Hanna Hottenrott and Susanne Thorwarth*
Previous literature provided evidence on financing constraints for investment in
research and development (R&D) activities due to capital market imperfections and
special features of R&D investments. Moreover, it has been shown that a shift in
capital structure towards more debt results in a reduction of R&D investments. This
article complements this literature by compartmentalising R&D activities into its
components, ‘R’ and ‘D’. In particular, we distinguish research from development as
these activities not only differ in their nature but also, to a large extent, take place
sequentially. Our results show that ‘R’ investment is more sensitive to a firms’
operating liquidity than ‘D’ indicating that firms have to rely even more on internal
funds for financing their research compared with development activities. Moreover,
we find that (basic) research subsidy recipients’ investment is less sensitive to
internal liquidity.
Key words: Research and development, Liquidity constraints, Innovation policy
JEL classifications: O31, O32, O38
1. Introduction
Research and development (R&D) activities and resulting innovation constitute an
important driver of economic competitiveness and hence sustainable economic growth.
The crucial role of R&D for promoting economic growth has long been emphasised in
economic theory (e.g. Solow, 1957; Romer, 1990; Grossman and Helpman, 1991; Aghion
and Howitt, 1992; Jones, 1995). Empirical studies at aggregate country levels usually
estimate the elasticity of labour productivity, total factor productivity and output per
capital to R&D expenditures. These studies show that R&D effects vary across countries
and time periods, but the effect is generally found to be positive.
1
Manuscript received 10 August 2009; final version received 20 July 2010.
Address for correspondence: Dirk Czarnitzki, K.U. Leuven, Department of Managerial Economics, Strategy
and Innovation, Naamsestraat 69, 3000 Leuven, Belgium; email: dirk.czarnitzli@econ.kuleuven.be
* K.U. Leuven, Belgium and Centre for European Economic Research (ZEW), Mannheim, Germany. We
thank IWT Flanders for providing their ICAROS database. DC gratefully acknowledges financial support
from the Flemish Science Foundation (grant G.0534-07N). We also thank two anonymous referees as well as
conference participants at the EARIE in Ljubljana and the DRUID in London for their helpful comments.
1
See, e.g., Soete and Patel (1985), Lichtenberg (1993) and Nadiri (1993). Moreover, studies by Coe and
Helpman (1995), Park (1995), Lichtenberg and van Pottelsberghe (1998) and van Pottelsberghe and Guellec
(2004) estimated the effects of foreign R&D on productivity.
ÓThe Author 2010. Published by Oxford University Press on behalf of the Cambr idge Political Economy Society.
All rights reserved.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
At the firm level, studies have shown, to large extent unanimously, that the impact of
R&D on productivity stems from the implementation of newly generated knowledge and
technological discoveries into new products, improvement of existing products and
production processes or cost reductions of producing existing products or services
(Stiglitz, 1969; Griliches, 1980; Schankerman, 1981; Griliches and Mairesse, 1984,
1990; Hall and Mairesse, 1995). Consequently, R&D has been recognised as an important
input factor to industrial production.
However, Griliches (1986) points out that basic research is a main driver for productivity
at the firm level. He shows that expenditures for basic research significantly contributed to
productivity growth of US manufacturing firms in the 1970s (see also Mansfield, 1980). In
his cross-sectional analysis, he found that firms that invested a larger fraction of their total
R&D on basic research were more productive, hence stressing the importance of this
component. Recently, this finding has been complemented by Czarnitzki et al. (2009) who
show that research expenditures exhibit a significant premium over development
expenditures with regard to patent productivity in a panel of Belgian firms.
Moreover, economic theory suggests that the social rates of return are larger than the
private rates of return from research activities because of the incompleteness of appropri-
ability of the knowledge that is being created by investment (Arrow, 1962; Usher, 1964).
2
When Arrow (1962) wrote about economic welfare and the allocation of resources for
invention, he was particularly referring to private research activities—or, generally
speaking, the production of new knowledge—that may suffer from financing constraints
due to market failure: ‘Thus basic research, the output of which is only used as an
informational input into other inventive activities, is especially unlikely to be rewarded’.
Moreover he concludes that ‘[...] underinvestment will be greater for more basic research.’
Hence, he was mainly referring to the ‘R’ component of R&D. Thus, especially for basic
research, which is used as an informational input into subsequent inventive activities, firms
may face particular difficulties to attract external investors or receive bank loans for
financing these activities. Firms with limited internal liquidity may therefore have to
constrain their research to currently available funds and postpone or abandon projects that
they would conduct if additional financing was available. This reduces incentives to invest
in long term research projects and may bring overall industrial research in the economy way
below optimal levels.
Hence, underinvestment in ‘R’ due to financing constraints may result in noticeable
slowdown in productivity growth and consequently have particularly detrimental effects on
technological progress and economic development.
Previous literature has provided vast evidence for the conjecture that capital market
imperfections, in particular information asymmetries, influence lending and investment
decisions of firms. As we will also discuss in Section 2, information asymmetries may be
particularly severe in the case of R&D activities. Complexity, specificity and outcome
uncertainty of such investment projects may make it particularly difficult for outsiders to
judge the expected return. Additionally, firms may be reluctant to reveal details of the
projects to potential investors. Besides information asymmetries, the intangibility of the
asset that is being created by the investment and the uncertainty of returns may
make raising funds externally more costly for R&D than for other types of investments
2
Empirically, numerous studies have confirmed positive spillover effects from industrial R&D; see
Griliches (1992) for a survey. However, to date there is no study on social returns that explicitly disentangles
‘R’ and ‘D’ investments with regard to spillover effects.
528 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
(e.g. Mansfield et al., 1977; Berger and Udell, 1998; Harhoff, 1998). Financing R&D
externally may thus be more costly compared to financing other types of investment (e.g.
Meyer and Kuh, 1957; Stiglitz and Weiss, 1981; Myers and Majluf, 1984; Anton and Yao,
2002). This may indicate that the Pecking Order Theory of capital structure developed by
Myers and Majluf (1984) is of particular relevance for financing decisions for R&D.
Yet, most articles, surveys, evaluations and reports concerning allocation of resources to
R&D do not explicitly distinguish between the different components of R&D.
3
Previous
studies on financing constraints for knowledge creating activities have usually analysed
industrial R&D in aggregate form. However, several authors have already stressed that
research activities are not only often separated from development activities organisationally
in industrial practice (Karlsson et al., 2004; Mansfield et al., 1971; Seiler, 1965), but also
important factors like time, originality, organisation and knowledge depth differ sub-
stantially between ‘R’ and ‘D’ projects (Karlsson et al., 2004). Moreover, characteristics
usually attributed to R&D activities in general, such as intangibility and outcome
uncertainty are very likely to be more applicable for ‘R’ compared to ‘D’. While not
empirically discussed, Kamien and Schwartz (1978) developed a theoretical model to
show that firms conducting fundamental or radical R&D projects are more likely to
encounter financial constraints than firms performing predominantly routine R&D.
Although the authors do not explicitly differentiate between ‘R’ versus ‘D’, their concept
of radical versus routine R&D may also apply to the distinction between ‘R’ and ‘D’, since
fundamental innovations usually involve basic research, require significantly more
resources and are much riskier in terms of default and expected returns.
This article aims to complement previous studies by explicitly taking the heterogeneity
of the two components of R&D into account. By compartmentalising industrial R&D
activity into its components, we argue that financing development, ‘D’, externally should
be less critical than it is for industrial research, ‘R’.
4
Our empirical study, indeed, reveals
that ‘R’ investment is more sensitive to the firms’ operating liquidity than ‘D’ indicating
that firms have to rely even more on internal funds for financing their research compared to
development activities. Looking at aggregated R&D expenditures of the firm would not
have revealed this effect.
Moreover, we find that (basic) research subsidy recipients invest more into ‘R’ than
other firms, and that their investment is also less sensitive to internal liquidity.
This paper proceeds as follows. Section 2 outlines the motivation for this research as well
as the conceptual framework of our analysis and Section 3 describes the data. The
econometric set-up and our model specifications are described in Section 4. Section 5
presents the results before we conclude in Section 6.
2. Conceptual framework
Although the presence of financing constraints for R&D has been shown in numerous
empirical studies,
5
R&D has usually been analysed in aggregate form. If external financing
3
Link (1982) is one of the few studies to analyse determinants of inter-firm differences in the composition
of R&D expenditures. Also Mansfield (1981) studies the effects of firm size on the composition of R&D
investments and finds that increases in size of firm—measured in firms’ sales—are associated with more than
proportional increases in the amount spent on basic research. This may already hint at financing constraints
on firms with lower levels of internally available funds.
4
See page 7 for the definition of firms ‘R’ versus ‘D’ expenditure according to the OECD Frascati Manual
(1993, 2002).
5
See Hubbard (1998) and Hall (2008) for extensive surveys of the literature.
Industrial research versus development investment 529
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
for R&D is constrained, firms engaging in R&D may have to rely to a larger extent on
internal financial resources than firms that mainly invest in capital goods (e.g. Carpenter
and Petersen, 2002; Chiao, 2002; David et al., 2008). Theoretical and empirical literature
has illustrated that firms foremost use internal funds to finance innovation projects as
compared to debt indicating such a gap in the cost of capital (Leland and Pyle, 1977;
Bhattacharya and Ritter, 1983; Hall, 1990, 1992; Himmelberg and Peterson, 1994;
Harhoff, 1998; Bougheas et al., 2003; Czarnitzki and Hottenrott, 2010).
However, research activities that usually precede development activities may show
fundamentally different characteristics. Attributes typically related to R&D activities in
general, such as intangibility and uncertainty are very likely to be more applicable for
research ‘R’ compared to development ‘D’. Research projects are usually characterised by
being ‘far from the market’ and may suffer from stronger externalities decreasing the
expected profitability. In many cases, results obtained from research activities are not even
planned to reach the market at all, but are rather used to build up individual or group
knowledge, which might be beneficial for future research (Mansfield et al., 1971). By
contrast, development projects aim at commercialising products to fit customer demands.
Research can be labelled as a more discontinuous process, which may result in solutions
displaying the possibilities in a technology area whereas development is a more continuous
solution of existing ideas (Karlsson et al., 2004).
To illustrate this, think of, for instance, the recently emerging technology named DNA
computing. This technology uses genetic material to create nano-computers, i.e. DNA-
based microprocessors, which, compared to traditional silicon-based computers, are
capable of storing considerably more information on an even smaller surface. Moreover,
they promise to allow the solving of complex ‘fuzzy logic’ problems, solutions to which
cannot be found with existing technologies. Suppose a semiconductor chip producing
company such as Intel works on this new technology. Projects that have the objective of
increasing Intel’s knowledge on molecular computing in general would be a clear case for
what we would call ‘R’-projects. This may involve basic research on thinking whether
molecular computing is possible at all. If so, applied research will address the challenge of
potential implementation of this new technology. In contrast, current development
activities of Intel would clearly revolve around introducing a new generation or improved
version of a silicon-based chip on the market. Of course, if research ever leads to a feasible
implementation of DNA-computing, the prototyping of such chips would then become
a development project in the (distant) future.
Thus, although ‘D’ projects usually follow some form of ‘R’ project, not every ‘R’ project
is followed by a corresponding ‘D’ project, because not every ‘R’ project may deliver results
that can immediately serve as input for the development phase. Hence, ‘R’ and ‘D’ projects
can be either entirely independent of each other or, of course, results from ‘R’ may enter
‘D’ projects. Nevertheless, development activities may build on previously generated
knowledge and may be tangible to a certain extent, for example in the form of patents or
other intellectual property rights. Thus, the gap in financing costs may be larger in the case
of research activities as this is the part of the R&D process that is more prone to
information asymmetries.
Consequently, firms may be limited in the financing of their R&D by the funds that are
internally available. At the same time, internal financial resources are affected by the firms’
debt payment obligations. The higher the obligations relative to the internal funds of the
firm, the less liquidity remains for activities that have to be financed internally such as
R&D. Thus, increases in firms’ levels of debt may put pressure on the firm to use its cash
530 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
flow to service interest and repayment at the expense of long-term investments such as
R&D (Hall, 1990, 1992; Long and Ravenscraft, 1993; Bhagat and Welch, 1995; Ogawa,
2007). Moreover, high leverage may reduce access to further credit due to increasing
default risk; therefore, we also control for debt.
6
This article therefore explicitly aims at disentangling the effects of liquidity and debt on
R&D using firm-level panel data. We estimate separate equations for investment in ‘R’ and
investment in ‘D’ to identify differences in the reliance of firms on internal funds for both
components of R&D reflecting constraints in the access to external financing.
We expect a positive relationship between firms’ liquidity and the firms’ expenditures in
R&D. For ‘R’, however, we expect the effect to be stronger due to the nature of ‘R’
activities versus ‘D’ activities. The former is likely to be more prone to information
asymmetries and secrecy issues, as well as uncertainty of outcomes. These factors may
make it more costly to finance externally if not impossible. Thus, a firm has to rely more on
internal funds for ‘R’ than for ‘D’ and therefore may show a larger sensitivity to the
availability of internal funds.
While the differences in funding gaps for ‘R’ versus ‘D’ have not yet been uncovered
explicitly in an empirical study on financial constraints at the firm-level, this phenomenon
has long been understood by technology policy. In Flanders, for instance, different subsidy
schemes exist. Firms can submit project proposals for subsidy requests, and possible grants
are based on a percentage of the eligible costs, which consist of project-related payroll
costs, direct costs (operating costs and depreciation of the equipment) and, to a limited
extent, indirect (general) costs. The percentage to which total project cost can be
subsidised depends on the type of research project: industrial ‘basic and strategic’ research
may receive up to 50% of the accepted total costs, ‘experimental development and
prototyping’ up to 25% of the accepted total costs and ‘mixed research projects’ up to 38%
of the accepted total costs.
Consequently, we supplement our initial analysis with a policy variable. We investigate
whether firms that received grants for ‘basic and strategic’ research projects are less
constrained in their investment into ‘R’. For ‘D’ we do not expect an impact of public basic
research funding on the level of constriction.
3. Data
The data for our analysis stems from the Belgian part of the Organisation for Economic
Cooperation and Development (OECD) R&D survey. The survey is harmonised across
OECD countries and is conducted every second year in order to compose the OECD Main
Science and Technology Indicators with the collected data. This R&D survey is
a permanent inventory of all R&D-active companies in Flanders. Our analysis is based
on six consecutive waves of the R&D survey data covering the period from 1999 until 2007.
Each wave provides information on firm level data for two consecutive years. The
definition of research and development used in most economic research encompasses
several kinds of activities. According to the definition of the OECD Frascati Manual
6
In Belgium, as in many other European countries, raising equity for financing investment projects does
not generally appear to be a preferred option. Only a few firms of the entire population are listed at the stock
exchange. In particular, small (family owned) firms view issuing of new equity as not particularly desirable,
e.g. Deloof and Jegers (1999) point out there was not one single public issue of a straight bond by a Belgian
company between 1990 and 1995. Borrowing from banks is the most common form of raising funds for
investment besides internal sources such as intra-group loans (see for example Deloof, 1998).
Industrial research versus development investment 531
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
(1993, 2002), which frames the methodology for collecting and using statistics about R&D
in OECD countries, the term R&D covers three activities: basic research, applied research
and experimental development. The first two can be summarised to be the ‘R’ of R&D,
while the latter constitutes the ‘D’ component of R&D. According to the international
definition, basic research is ‘experimental or theoretical work undertaken primarily to
acquire new knowledge of the underlying foundation of phenomena and observable facts,
without any particular application or use in view’. Whereas applied research also is original
investigation undertaken in order to acquire new knowledge, it is, however, directed
towards a specific practical outcome. Development activities are rather ‘systematic work,
drawing from existing knowledge gained from research and/or practical experience, which
is directed to producing new materials, products or devices [...]’ (OECD, 2002, p. 30).
In order to construct the financial indicators we supplemented our data with information
on the firms’ financial background with accounting and balance sheet data from the
BELFIRST database. For the policy analysis, we supplement our panel data with public
R&D project funding information that has been provided by the IWT Flanders, which is the
innovation and technology policy agency of the Flemish government administering
innovation subsidies.
Our sample comprises firms that at least once invested in R&D and are observed at least
twice in the reference period as we estimate panel data models that allow controlling for
unobserved heterogeneity. On average, we observe each firm about four times in our panel.
After eliminating data with missing values in the variables of interest, our final sample
consists of 3,686 unique firm-year observations referring to 952 different firms.
7
3.1 Variables
Firms’ R&D expenditures serve as dependent variables for our analysis. The peculiarity of
the survey, however, provides information not only about aggregate R&D spending, but also
about ‘R’ and ‘D’ separately. Thus, we are able to disentangle a firm’s R&D expenditures
(R&D) into its components research (RESEARCH) expenditures on the one hand and
development (DEVELOPMENT) expenditures on the other hand.
We derive indicators for the capital structure of the firms in our sample from balance
sheet information accounted according to local Belgian GAAP on an annual basis.
We employ the firms’ stock of working capital as a measure of operating liquidity to
overcome limitations of cash flow as indicator for firms’ liquidity as suggested by Hall and
Kruiniker (1995). The appropriateness of cash flow as an indicator for the availability of
internal funds and the interpretation of sensitivity of R&D investment to changes in cash
flow has been seriously questioned in the literature (Fazzari and Petersen, 1993; Hao and
Jaffe, 1993; Hall and Kruiniker, 1995; Kaplan and Zingales, 1997, 2000). Especially in the
case of large firms, free cash flow levels may be determined by accounting as well as
dividend policies aimed at mitigating moral hazard problems (Jensen and Meckling, 1976;
Jensen, 1986; Dhanani, 2005). Additionally, a positive relationship between investment
and cash flow may simply reflect that both of them correlate with promising market
demand.
Firms use working capital for day-to-day financial operations and it is therefore an
important indicator of the liquidity of the firm. By retaining cash inflows, firms accumulate
the financial funds needed for investment as reflected in the stock of working capital. The
7
Table A1 in the appendix provides details on the distribution of firms across industries.
532 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
advantage of working capital over cash flow is therefore that it is an accumulation rather
than a flow parameter. Although working capital is affected by cash flow working capital is
a more precise liquidity indicator when it comes to investment decisions as it includes not
only cash but also values that can be converted into cash relatively easily.
Working capital (WCAP) is the net amount of short term assets, i.e. the difference
between current assets minus current liabilities of a firm. The higher the working capital
the more secured is a firm’s liquidity and accordingly its financial flexibility. This variable
can take positive or negative values. A positive working capital means that short term
liabilities are covered by current assets (cash, accounts receivable and inventory) whereas
a negative working capital indicates that a firm’s current assets are not sufficient to cover its
current liabilities. Hence, it serves as indicator for a firm’s short-term liquidity.
In addition, we use firm’s debt as further variable controlling for credit market access.
The overall liabilities of the firm (DEBT) consist of current liabilities payable within one
year and non-current liabilities payable later than one year. The overall debt-to-equity ratio
(leverage) of the firms in our sample is about 60%. However, we employ debt-to-tangible
assets ratios rather than debt-to-equity as the former type is more meaningful in terms of
reflecting the firms’ liabilities relative to the firms’ collateral value.
As is common in the financial constraints literature, we scale the dependent variables as
well as WCAP and DEBT by firms’ assets (Fazzari et al., 1988; Fazzari and Petersen, 1993;
Harhoff, 1998). We use lagged tangible assets (K) calculated as the value of a firm’s total
fixed assets minus current assets and financial assets (as already captured by WCAP),
goodwill and other intangible assets.
As is common in firm level studies, we also control for firm size. Larger firms may be able
to realise economies of scope while conducting research and development activities.
Therefore we include the log of the firms’ tangible assets ln(K) as well as its squared value
[ln(K)]
2
.
The dummy variable GROUP taking the value 1 if a firm belongs to a group (0,
otherwise) controls for different governance structures. Group members may conduct
more R&D activities since firms associated with a group can make use of intra-firm
spillovers, internalise externalities as well as fund R&D from intra-group sources. To
control for age-related effects, since younger and newly established firms may invest
relatively more into research and development than older firms, we also use the log of age,
ln(AGE ). We also allow for a non-(log)-linear relationship by including [ln(AG E )
2
].
For the supplemental regression including the policy variable, we create a dummy
variable indicating that the firm received government funding for basic or strategic research
(SUBR). This variable will be interacted with WCAP to evaluate whether possible financial
constraints are alleviated by basic research funding.
Furthermore, a set of time dummies is included to take business cycle effects into
account. Finally, 16 sector dummies on the basis of the European standard industry
classification NACE are included to capture different technological opportunities.
All variables in monetary units are measured in million Euros in prices of the year 2000.
We used the GDP deflator for price adjustment. To avoid a simultaneity bias, which can
arise if there are feedback effects from the dependent variable to current explanatory
variables, we use lagged values of all time variant exogenous variables (except AG E ).
8
8
For the variable DEBT, we use a two period lag as debt is measured at the end of the year t–2, so that its
signal to lenders is effective in t–1. This two-period lag then indicates credit market access in t–1, when also
the available working capital is measured.
Industrial research versus development investment 533
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Table 1 shows the descriptive statistics for all variables. Average R&D expenditures are
around e3 million for each firm per year where roughly two thirds are spent for
development and one third for research. On average, liabilities amount to about e77
million. The median, however, is much lower (about e5 million).
Firms in our sample have about 270 employees on average. However, the size
distribution of firms is skewed and at the median the number of employees is only 54.
The sample comprises very young firms of 2 years as well as established firms of up to 125
years of business activity. The average firm age is about 26 years. Further, 60% of the firms
in our sample are part of a group.
4. Econometric analysis
We estimate censored panel regression models as not all of our firms in the sample perform
R&D in each period. Small firms in particular may conduct R&D only on an irregular
basis. In particular, we estimate random effects models that can be written as
yit 5maxð0;xit b1ci1uit Þ;i51;2;...;N;t51;2;...;T
uit jxi;ci,N0;s2
u
where ydenotes the dependent variable, xthe set of regressors, ca firm-specific time-
constant effect and uthe usual random error term. The parameters to be estimated are
denoted by the vector b. We first estimate the model as random effects Tobit that requires
the assumption of no correlation between cand x(see Wooldridge, 2002, pp. 540–1, for
further technical details).
9
In addition, we follow Wooldridge (2002) and relax the
assumption of uncorrelatedness between xand cby modifying the model. Let
ci5c1
xij1ai:, then we can write
Table 1. Descr iptive statistics (3,686 obser vations)
Variable Unit Median Mean SD Min Max
R&D
i,t
Million e0.16 2.86 18.36 0 471.35
RESEARCH
i,t
Million e0.04 1.07 7.33 0 217.58
DEVELOPMENT
i,t
Million e0.07 1.79 13.39 0 424.21
DEBT
i,t–2
Million e5.33 76.94 386.6 0.01 7,764.43
WCAP
i,t–1
Million e2.00 12.54 42.65 –155.02 634.90
K
i,t–1
Million e1.39 15.94 85.69 0.01 1,805.05
EMP
i,t–1
Headcount 54 271.34 745.79 1 11,575
AGE
i,t
Years 19 25.75 19.25 2 125
GROUP
i
Dummy 1 0.58 0.49 0 1
RSUB
it
Dummy 0 0.07 0.25 0 1
Notes: Time and industry dummies omitted.
We use the variables in monetary units measured in thousands of euros in the regression analysis but display
them in millions in this table for better readability.
9
Note that it is not useful to estimate a fixed effects Tobit model, as the maximum likelihood estimator is
not consistent (see, e.g., Cameron and Trivedi, 2005).
534 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
yit 5maxð0;c1xit b1
xij1ai1uit Þ
with
uit jxi;ai,N0;s2
u;
ai,N0;s2
a:
The appropriateness of the Wooldridge model, i.e. that the assumption of uncorrelated-
ness between the firm-specific effect and the regressors is not valid, will be tested by the
joint significance of the x-variables’ ‘within’ means.
We estimate separate models for our three dependent variables of interest. The basic
model for R&D investment is specified as:
10
R&D*
it
Kit
5b01b1
DEBTi;t22
Ki;t21
1b2
WCAPi;t21
Ki;t21
1b3lnKi;t211b4lnKi;t212
1b5lnAGEi;t1b6lnAGE i;t21b7GROUP i1+
23
k58
bkINDik 1+
32
s524
bsts1ci1uit
and
R&Dit
Kit
5(R&D*
it
Kit if xit b1ci1uit >0
0 otherwise
The equations for the two components of internal R&D, i.e. RESEARCH and
DEVELOPMENT are specified analogously.
5. Estimation results
The results of the random effects panel and its modification by Wooldridge are presented in
Table 2. The first column presents the results for total in-house R&D investment, columns
2 and 3 present the results for Research and Development investments separately.
For the total R&D investment, we find the expected results: the working capital variable
is positively significant and debt is negatively significant. Thus, firms are financial
constrained by their internal resources. In addition, with higher values of debt, investment
is reduced even further. We interpret this as indication that the higher the default risk, the
less likely potential investors are to provide further capital for R&D investment.
If we split R and D into its two components, we find interesting differences. Debt is only
negatively significant for R but not for D, and the positive, significant coefficient of
WCAP/K is larger in the R equation than in the D equation. Thus, the results from the
disaggregated models for R and D reveal that these liquidity effects constraining R&D stem
to a large extent from research activities only. In order to evaluate the magnitude of the
impact of WCAP on investment, we calculate marginal effects. We are interested in
studying how E(yjx) changes if xchanges, i.e. we take the censoring into account.
11
We
calculate the marginal effects under the assumption that c
i
50 at the mean of the
covariates, and obtain standard errors using the delta method (cf. Greene, 2000, for
10
Although the model contains a time-invariant firm-specific effect we also include the time-invariant
regressors GROUP and the industry dummies, as the firm-specific effect is treated as a random component in
the estimation. Including time invariant regressors decreaseS the error term variance (see Wooldridge, 2002,
P. 541).
11
Note that the coefficients in a Tobit model indicate @E(y*jx)/@x.
Industrial research versus development investment 535
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Ta b l e 2 . Estimation results from Tobit regressions (3,686 obser vations of 952 firms) on R&D expenditures per unit tangible assets (K)
Variable
Random-effects panel model Wooldridge estimator
R and D Research Development R and D Research Development
DEBT
i,t–2
/K
i,t–1
–0.004** (0.002) –0.003*** (0.001) –0.002 (0.002) –0.005** (0.002) –0.004*** (0.001) –0.002 (0.002)
WCAP
i,t–1
/K
i,t–1
0.127*** (0.027) 0.096*** (0.014) 0.054** (0.025) 0.119*** (0.038) 0.117*** (0.020) 0.034 (0.035)
ln(K
i,t–1
) –4.483*** (0.445) –1.511*** (0.227) –2.628*** (0.419) –3.773*** (0.851) –1.122*** (0.343) –2.388*** (0.619)
[ln(K
i,t–1
)]
2
0.248*** (0.445) 0.093*** (0.015) 0.147*** (0.028) –0.196*** (0.045) 0.061** (0.022) 0.121*** (0.041)
ln(AGE
i,t–1
) 1.001 (1.618) –0.062 (0.824) 0.139 (1.491) 6.663* (3.815) 8.426*** (1.949) –3.135 (3.476)
[ln(AGE
i,t–1
)]
2
–0.219 (0.273) –0.059 (0.140) –0.020 (0.253) –0.883 (0.970) –1.647*** (0.491) 1.195 (0.884)
GROUP
i
2.037*** (0.403) 0.320 (0.206) 1.337*** (0.376) 1.338*** (0.419) 0.263 (0.208) 1.276*** (0.384)
Joint
significance of
time dummies
x
2
(8) 511.11 x
2
(8) 515.44** x
2
(8) 534.63*** x
2
(8) 511.62 x
2
(8) 510.21 x
2
(8) 525.58***
Joint
significance of
industry dummies
x
2
(15) 536.55*** x
2
(15) 516.37 x
2
(15) 539.66*** x
2
(15) 532.12** x
2
(15) 516.36 x
2
(15) 537.39***
Joint significance
of
within means
---x
2
(6) 513.35*** x
2
(6) 553.54*** x
2
(6) 54.84
Log-likelihood –11,186.808 –7,638.383 –9,843.164 –11,061.958 –7,611.771 –9,840.744
r0.319 0.405 0.458 0.445 0.403 0.458
No. of censored
observations
541 1,085 839 541 1,085 839
Notes: Standard errors are in parentheses.
Significance levels: ***1%, **5%, and *10%.
All models include an intercept (not presented).
Coefficients of variables’ within means in Wooldridge model are omitted.
The value of rindicates the share of the total variance which is due to the cross-sectional variation.
536 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
technical details). The marginal effect of WCAP amounts to 0.064 (SE 50.014) in the R
equation and to 0.024 (SE 50.011) in the D equation, which yields a significant difference
between the two effect at the 5% level. Thus, we can conclude that the firms have to rely
even more on internal resources for financing R within their R&D project, or in other
words suffer more from financial constraints for R than for D. This may be due to the fact
that ‘D’ occurs later in the R&D process and is closer to the market, i.e. closer to yielding
returns. Hence, firms may cut ‘R’, before they reduce ‘D’ if they are constrained by the
credit market and have to rely on scarce internal funds.
12
With respect to the control variables we find a non-linear relationship between the
dependent variables and firm size as measured by lnK. The minimum value of this U-
shaped curve is around the 75th percentile of the lnKdistribution for all models. Thus,
smaller firms and larger firms, in terms of lnK, invest relatively more into R&D than
medium-sized firms. As expected, we also find that firms associated with a group invest
more than stand-alone companies. The age variables are insignificant in all models.
When we turn to the Wooldridge models, we find very similar results as in the random
effects models, hence, we discuss these only briefly. First, the tests that xand care
uncorrelated are rejected in the R&D and R equation, but not in the D equation (tests on
joint significance on within means in Table 2). Thus, we would prefer the Wooldridge
generalisation of the random effects model over the ‘traditional’ random effects model for
the R&D and R equations. Although the traditional random effects model is rejected in two
cases, the results on WCAP and DEBT remain basically the same in the Wooldridge
generalisation, except that debt becomes insignificant in the R&D equation. This may be
caused by relatively high correlations between the time varying x
it
and the within-firm
means that are employed to relax the assumption that xand care not correlated.
In Table 3, we present the regression on R and D investment again, but there we have
augmented the regression with the SUBR dummy, i.e. the indicator for the receipt of public
subsidies for ‘basic and strategic’ research. We also use this dummy to estimate to separate
slope parameters for WCAP. On the one hand, we estimate a coefficient for the research-
subsidised firms (WCAP 3SUBR). On the other hand, a separate slope for the firms that
did not receive a research subsidy [WCAP 3(1–SUBR)] is estimated. We expect that the
non-subsidised firms are financially constrained, and thus have a positive and significant
coefficient estimate. The slope for the subsidised firms should be smaller or even
insignificant, that is, firms show no or less sensitivity to their internal funds. If one
compares the results on the main regressors of interest with Table 2, it shows that the
magnitude of the coefficients is quite stable, and thus the earlier results persist. The
coefficient of SUBR is positively significant in the R equation. Thus, recipients of such
subsidies are able to invest significantly more into R due to the subsidy itself.
13
Even more
interesting, however, we find that WCAP 3(1–SUBR) is positively significant as in the
basic model. The slope coefficient for the subsidised firms (WKAP 3SUBR), however, is
not significant, i.e. these firms do not suffer from financial constraints. When testing
whether the coefficients of (WKAP 3SUBR) and WCAP 3(1–SUBR) are significantly
12
As working capital may be industry specific, we test the robustness of our results to possible industry
specificity by including a working capital variable that rescales the firm’s individual working capital ratio by
the average of this ratio in the firm’s industry. This variable was included in the regression in place of the
original working capital to tangible assets ratio. It turned out that all results presented above hold.
13
Note that the coefficient of SUBR is not significant in the Wooldridge extension of the random effects
model. However, this is due to the fact that the subsidy dummy shows little within-time variation. The within
mean of the subsidy dummy (not displayed in the table) is significant at the 1% level with a coefficient of 2.93.
Industrial research versus development investment 537
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Ta b l e 3 . Estimation results from Tobit regressions (3,686 observations of 952 firms) on R&D expenditures per unit tangible assets (K) with SUBR
Variable
Random-effects panel model Wooldridge estimator
Research Development Research Development
DEBT
i,t–2
/K
i,t–1
–0.003*** (0.001) –0.002 (0.002) –0.004*** (0.001) –0.002 (0.002)
SUBR
i,t–1
0.819*** (0.292) 0.120 (0.525) 0.270 (0.312) –0.047 (0.564)
SUBR
i,t–1
3(WCAP
i,t–1
/K
i,t–1
) 0.019 (0.035) 0.058 (0.061) 0.036 (0.038) 0.038 (0.066)
(1–SUBR)
i,t–1
3(WCAP
i,t–1
/K
i,t–1
) 0.103*** (0.014) 0.053 (0.025) 0.122*** (0.021) 0.033 (0.036)
ln(K
i,t–1
) –1.475*** (0.226) –2.625*** (0.419) –1.307*** (0.342) –2.409*** (0.620)
[ln(K
i,t–1
)]
2
0.091*** (0.015) 0.147*** (0.028) 0.067*** (0.022) 0.123*** (0.041)
ln(AGE
i,t–1
) –0.086 (0.819) 0.145 (1.491) 8.543*** (1.944) –3.261 (3.477)
[ln(AGE
i,t–1
)]
2
–0.054 (0.139) –0.021 (0.253) –1.704*** (0.489) 1.181 (0.885)
GROUP
i
0.324 (0.205) 1.341*** (0.376) 0.346* (0.208) 1.298*** (0.386)
Joint significance
of time dummies
x
2
(8) 514.90* x
2
(8) 534.61*** x
2
(8) 510.72 x
2
(8) 525.42***
Joint significance
of industry dummies
x
2
(15) 516.21 x
2
(15) 539.55*** x
2
(15) 516.25 x
2
(15) 537.28***
Joint significance
of within means
–– x
2
(7) 571.59*** x
2
(7) 55.11
Log-likelihood –7,633.656 –9,843.119 –7,598.082 –9,840.563
r0.399 0.458 0.392 0.458
No. of censored
observations
1,085 839 1,085 839
Notes: Standard errors are in parentheses.
Significance levels: ***1%, **5%, and *10%.
All models include an intercept (not presented).
Coefficients of variables’ within means in Wooldridge model are omitted.
The value of rindicates the share of the total variance which is due to the cross-sectional variation.
538 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
different from each other, we reject H
0
(5stating that they are equal) at the 5%
significance level. Thus, in general, firms depend highly on their internal resources, but for
subsidy recipients this effect is offset. On the one hand, this may indicate that the grants are
typically large enough to conduct the desired research project at full scale. On the other
hand, it may also hint at the so-called certification effect of subsidies. Lerner (1999) argued
that US SBIR
14
awardees achieve better access to external capital due to the fact that they
received an SBIR award, as it involves a thorough peer-evaluation procedure of the
submitted research proposal. For instance, venture capitalists and banks may use that as
a positive signal of the quality of the investment project.
It is noteworthy that the findings on the subsidy dummy should be interpreted with some
care. It is known that firms may self-select themselves into subsidy schemes (see, e.g., the
survey by David et al., 2008) and thus we may overestimate the effect of SUBR. Although
our panel regressions control for unobserved heterogeneity among firms and that should
take care of the potential selection bias to a certain extent, correcting for selection properly
would require instrumenting the SUBR variable. We experimented with pooled cross-
sectional regressions, but it turned out that it is important to control for firm-specific
effects in this study (see the high share of total variance explained by mere cross-sectional
variation).
15
Finally, we would like to note that it would be preferableto consider dynamic specifications
of our estimation equation as theoretical investment models are typically based on the inter-
temporal optimisation problem between the size of investment and the level of (R&D)
capital stock. Econometrically this would require dealing with the so-called initial conditions
problem and the inclusion of a lagged dependent variable (see, e.g., Wooldridge, 2005,
or Bond and Van Reenen, 2007, for an overview). The application of such a method
would require four observations per firm for conducting meaningful panel regressions.
Our data, however, is not rich enough. As our panel is not only unbalanced but has also
gaps in the time-series structure (as the firms did not necessarily respond to the surveys
in adjacent years), we found it impossible to estimate a dynamic model. Additional
concerns arise with respect to the validity of the methods above for unbalanced panels in
general.
6. Conclusion
Previous literature provided vast evidence for liquidity constraints of industrial R&D
investments due to information asymmetries. Complexity, specificity and uncertainty of
returns may make it particularly difficult for outsiders to judge the potential value and firms
may be reluctant to reveal details of their projects to potential investors. Furthermore, the
intangibility of the asset that is being created (knowledge) may limit firms’ access to external
funds for R&D. Hence, firms have to rely on internal liquidity to fund their R&D activities.
Although the presence of financing constraints for R&D has been shown in numerous
empirical studies, R&D has usually been viewed as an inseparable process (possibly due to
limited data availability). However, characteristics usually attributed to R&D activities in
general, such as intangibility and uncertainty are very likely to be more applicable to ‘R’
14
The Small Business Innovation and Research program is the largest policy scheme for supporting
commercial research and innovation in US history, and has been in place since 1982.
15
Although it is theoretically possible to estimate a censored panel data model incorporating endogeneity
(see Honore
´and Hu, 2004), we felt that an application of that estimator would be beyond the scope of this
paper.
Industrial research versus development investment 539
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
than to ‘D’. Research projects are usually characterised by being ‘far from the market’ and
may induce higher externalities decreasing the likelihood of profitability. Moreover,
development takes place at a much later stage of the R&D process, building on previously
generated knowledge.
This article aimed to complement previous studies by explicitly taking the heterogeneity
of the two components of R&D into account. By compartmentalising industrial R&D
activity into its components, we argue that financial constraints affect ‘R’ more than ‘D’.
Our empirical study indeed reveals that ‘R’ investment is more sensitive to the firms’
operating liquidity than ‘D’ investments, indicating that firms have to rely even more on
internal funds for financing their research compared to their development activities.
Looking at aggregated R&D expenditures of the firm would not have revealed this effect.
Thus, estimating different investment equations for R&D and ‘R’ and ‘D’ separately
illustrates that financing constraints may be more binding for ‘R’ than for ‘D’.
Governments, as for example in Flanders, are aware of the need for special support of
industrial research, and grant higher shares of total cost in terms of subsidies to ‘basic and
strategic’ research than for experimental development or prototyping. We find that
research subsidy recipients indeed invest more into ‘R’ than other firms, and that their
investment is also less sensitive to internal liquidity.
This has interesting consequences for policy. While policy makers also seek to increase
the outcome of funded projects, it may happen that very challenging research projects are
not awarded a subsidy, as submitted project applications are often evaluated by peer-review
according to criteria such as ‘technological content’ and ‘expected economic value’ or
similar criteria. As highly basic research projects may score low on the latter criterion,
government agents may behave similarly to private lenders when it comes to project
selection. Therefore, even funding for R&D may still inhibit investment into R.
As one rough check, we analysed the submitted grant requests and the grant rate of those
projects submitted under the three different schemes present in Flanders: ‘(strategic) basic
research’ (5‘R’ in our terms), ‘mixed projects’ and ‘experimental development and
prototyping’ (5‘D’ in our terms). As Table 4 shows there is slight evidence that basic
research projects are indeed rejected more frequently. While the total grant rate amounts to
81%, ‘R’ projects are only granted at a rate of 75%, but ‘D’ projects at a rate of 82%.
However, this evidence is not clear cut as ‘mixed projects’ are granted most frequently, at
a rate of 91%.
It would be highly interesting to find comparable figures from other countries to evaluate
national policies across countries to learn more about policy practices, and eventually learn
Table 4. Grant rate of submitted project proposal (by type) in Flanders
Total
(Strategic) basic
research
Mixed
projects
Experimental
development
and prototyping
No. submitted
projects
3,506 1,389 829 1,288
Grant rate 81% 75% 91% 82%
Notes: Data were kindly provided by IWT Flanders (own calculations).
A ‘grant’ may imply a, typically downward, adjustment of accepted total project cost by the agency compared
the originally submitted project budget.
540 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
more about the implications of our findings for levels of ‘R’ and ‘D’ in the European
Union. This could be especially interesting against the backdrop of the EC Lisbon agenda
(European Commission, 2005), which, among other issues, states the intention to raise the
R&D/GDP ratio to 3% in the European Union. If many countries favour projects that are
more of a ‘D’ nature a shift to funding a higher proportion of ‘R’ projects may yield higher
total R&D in Europe as firms are more constrained in ‘R’ according to our econometric
results.
Bibliography
Aghion, P. and Howitt, P. 1992. A model of growth through creative destruction, Econometrica,
vol. 60, 323–51
Anton, J. and Yao, D. 2002. The sale of ideas: strategic disclosure, property rights, and
contracting, Review of Economic Studies, vol. 69, no. 3, 513–31
Arrow, K. 1962. Economic Welfare and the Allocation of Resources for Invention, pp. 609–26, in
Nelson, R. (ed.), The Rate and Direction of Inventive Activity: Economic and Social Factors,
Cambridge, MA, Universities-National Bureau
Bhagat, S. and Welch, I. 1995. Corporate research and development investments: international
comparisons, Journal of Accounting and Economics, vol. 19, 443–70
Bhattacharya, S. and Ritter, J. 1983. Innovation and communication: signaling with partial
disclosure, Review of Economic Studies, vol. 50, no. 2, 331–46
Berger, A. and Udell, G. 1998. The economies of small business finance: the roles of private
equity and debt markets in the financial growth cycle, Jour nal of Banking and Finance, vol. 22,
613–73
Bond, S. and Van Reenen, J. 2007. Microeconometric models of investment and employment,
pp. 4417–98, in Heckman, J. J. and Leamer, E. E. (eds), Handbook of Econometrics, vol. 6,
London, North-Holland
Bougheas, S., Go¨rg, H. and Strobl, E. 2003. Is R & D financially constrained? Theory and
evidence from Irish manufacturing, Review of Industrial Organization, vol. 22, no. 2, 159–74
Cameron, A. C. and Trivedi, P. K. 2005. Microeconometrics: Methods and Applications, New York,
Cambridge University Press
Carpenter, R. and Petersen, B. C. 2002. Capital market imperfections, high-tech investment,
and new equity financing, Economic Jour nal, vol. 112, 54–72
Chiao, C. 2002. Relationship between debt, R&D and physical investment, evidence from U.S.
firm-level data, Applied Financial Economics, vol. 12, 105–21
Coe, D. T. and Helpman, E. 1995. International R&D spillovers, European Economic Review, vol.
39, 859–87
Czarnitzki, D. and Hottenrott, H. 2010. R&D investment and financing constraints of small and
medium-sized firms, Small Business Economics, DOI: 10.1007/s11187-009-9189-3
Czarnitzki, D., Kraft, K. and Thorwarth, S. 2009. The knowledge production of ‘R’ and ‘D’,
Economics Letters, vol. 105, no. 1, 141–3
Deloof, M. 1998. Internal capital markets, bank borrowing, and financing constraints: evidence
from Belgian firms, Journal of Business Finance and Accounting, vol. 25, nos 7 & 8, 945–68
Deloof, M. and Jegers, M. 1999. Trade credit, corporate groups, and the financing of Belgian
firms, Journal of Business Finance and Accounting, vol. 26, nos 7 & 8, 945–66
Dhanani, A. 2005. Corporate dividend policy: the views of British financial managers, Journal of
Business Finance and Accounting, vol. 32, nos 7 & 8, 1625–72
David, P., O’Brien, J. and Yoshikawa, T. 2008. The implications of debt heterogeneity for R&D
investment and firm performance, Academy of Management Journal, vol. 51, 165–81
European Commission 2005. Common Actions for Growth and Employment: The Community Lisbon
Programme, Brussels
Fazzari, S. M. and Petersen, B. C. 1993. Working capital and fixed investments: new evidence on
financing constraints, RAND Journal of Economics, vol. 24, no. 3, 328–42
Fazzari, S., Hubbard, R. and Petersen, B. C. 1988. Financing constraints and corporate
investment, Brookings Papers on Economic Activity, vol. 1, 141–206
Industrial research versus development investment 541
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Greene, W. H. 2000. Econometric Analysis, Upper Saddle River, Prentice-Hall
Griliches, Z. 1980. R&D and the productivity slowdown, American Economic Review, vol. 70,
343–8
Griliches, Z. 1986. Productivity, R and D, and basic research at the firm level in the 1970s,
American Economic Review, vol. 76, no. 1, 141–54
Griliches, Z. 1992. The search for R&D spillovers, Scandinavian Journal of Economics, vol. 94,
Supplement, 29–74
Griliches, Z. and Mairesse, J. 1984. Productivity and R&D at the firm level, in Griliches, Z. (ed.),
R&D, Patents, and Productivity, Chicago, University of Chicago Press
Griliches, Z. and Mairesse, J. 1990. R&D and productivity growth: comparing Japanese and U.S.
manufacturing firms, in Hulten, C. (ed.), Productivity Growth in Japan and the United States,
Chicago, University of Chicago Press
Grossman, G. and Helpman, E. 1991. Innovation and Growth in the World Economy, Ch. 3,
Cambridge, MA, MIT Press
Hall, B. H. 1990. The impact of corporate restructuring on industrial research and development,
Brooking Papers on Economic Activity, vol. 1, 85–136
Hall, B. H. 1992. ‘Investment and Research and Development at the Firm Level: Does the Source of
Financing Matter?’, NBER Working Paper 4096
Hall, B. H. 2008. The financing of innovation, ch. 15, pp. 409–30 in Shane, S. (ed.), Blackwell
Handbook of Technology and Innovation Management, Oxford, Blackwell Publishers.
Hall, B. H. and Kruiniker, H. 1995. ‘The Role of Working Capital in the Investment
Process’, unpublished manuscript, Berkeley. http://www.econ.berkeley.edu/,bhhall/papers/
HallKruiniker94%20workcap.pdf
Hall, B. H. and Mairesse, J. 1995. Exploring the relationship between R&D and productivity in
French manufacturing firms, Journal of Econometrics, vol. 65, 263–93
Hao, K. Y. and Jaffe, A. B. 1993. Effect of liquidity on firms’ R&D spending, Economics of
Innovation and New Technology, vol. 2, no. 4, 275–82
Harhoff, D. 1998. Are there financing constraints for R&D and investment in German
manufacturing firms?, Annales d’ Economie et de Statistique, vol. 49/50, 421–56
Honore
´, B. and Hu, L. 2004. Estimation of cross sectional and panel data censored regression
models with endogeneity, Journal of Econometrics, vol. 122, 293–316
Hubbard, R. 1998. Capital–market imperfections and investment, Journal of Economic Literature,
vol. 36, 193–225
Himmelberg, C. and Peterson, B. 1994. R&D and internal finance: a panel study of small firms
in high-tech industries, Review of Economics and Statistics, vol. 76, 38–51
Jensen, M. 1986. Agency cost of free cash flow, corporate finance, and takeovers, American
Economic Review, vol. 76, 323–39
Jensen, M. and Meckling, W. 1976. The theory of the Firm: managerial behavior, agency cost,
and ownership structure, Journal of Financial Economics, vol. 3, 305–60
Jones, C. I. 1995. R&D-based models of economic growth, Journal of Political Economy, vol. 103,
759–84
Kamien, M. I. and Schwartz, N. L. 1978. Self-financing of an R&D project, American Economic
Review, vol. 68, no. 3, 252–61
Kaplan, S. and Zingales, L. 1997. Do investment-cash flow sensitivities provide useful measures
of financing constraints?, Quarterly Journal of Economics, vol. 112, no. 1, 169–215
Kaplan, S. and Zingales, L. 2000. Investment-cash flow sensitivities are not valid measures of
financing constraints, Quarterly Journal of Economics, vol. 115, no. 2, 707–12
Karlsson, M., Trygg, L. and Elfstro¨m, B.-O. 2004. Measuring R&D productivity:
complementing the picture by focusing on research activities, Technovation, vol. 24, 179–86
Leland, H. and Pyle, D. 1977. Informational asymmetries, financial structure, and financial
intermediation, Journal of Finance, vol. 32, 371–87
Lerner, J. 1999. The government as venture capitalist: the long-run impact of the SBIR program,
Journal of Business, vol. 72, 285–318
Lichtenberg, F. R. 1993. R&D investment and international productivity differences, pp. 47–68
in Siebert, H. (ed.), Economic Growth in the World Economy, Tubingen, Mohr
Lichtenberg, F. and Van Pottelsberghe de la Potterie, B. 1998. International R&D spillovers:
a comment, European Economic Review, vol. 42, 1483–91
542 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Link, A. N. 1982. An analysis of the composition of R&D spending, Southern Economic Journal,
vol. 49, no. 2, 342–9
Long, W. F. and Ravenscraft, D. J. 1993. LBOs, debt and R&D intensity, Strategic Management
Journal, vol. 14, 119–35
Mansfield, E. 1980. Basic research and productivity increase in manufacturing, American
Economic Review, vol. 70, 863–73
Mansfield, E. 1981. Composition of R&D expenditures: relationship to size of firm,
concentration and innovative output, Review of Economics and Statistics, vol. 63, no. 4, 610–14
Mansfield, E., Rapoport, J., Schnee, J., Wagner, S. and Hamburger, M. 1971. Research and
Innovation in the Modern Corporation, New York, Norton
Mansfield, E., Romeo, A., Villani, E., Wagner, S. and Husic, F. 1977. The Production and
Application of New Industrial Technology, New York, Norton
Meyer, J. R. and Kuh, E. 1957. The Investment Decision: An Empirical Study, Harvard University
Press, Cambridge, MA.
Myers, S. and Majluf, N. 1984. Corporate financing and investment decisions when firms have
information that investors do not, Journal of Financial Economics, vol. 13, 187–221
Nadiri, I. 1993, ‘Innovations and Technological Spillovers’, NBER Working Paper No. 4423,
Cambridge, MA
OECD 1993, 2002. Frascati Manual: Proposed Standard Practice for Surveys on Research and
Experimental Development, Luxembourg
Ogawa, K. 2007. Debt, R&D investment and technological progress: a panel study of Japanese
manufacturing firms’ behavior during the 1990s, Journal of the Japanese and International
Economies, vol. 21, no. 4, 403–23
Park, W. 1995. International R&D spillovers and OECD economic growth, Economic Inquir y, vol.
33, 571–91
Romer, P. M. 1990. Endogenous technological change, Journal of Political Economy, vol. 98, no.
2, 71–102
Schankerman, M. 1981. The effects of double-counting and expensing on the measured returns
to R&D, Review of Economics and Statistics, vol. 63, no. 3, 454–8
Seiler, R. 1965. Improving the Effectiveness of Research and Development, New York, McGraw-Hill
Soete, L. and Patel, P. 1985. Recherche-de
´veloppement, importations technologiques et
croissance economique, Revue Economique, vol. 36, 975–1000
Solow, R. M. 1957. Technical change and the aggregate production function, Review of Economics
and Statistics, vol. 39, 312–20
Stiglitz, J. 1969. Theory of innovation – discussion, American Economic Review, vol. 59, 44–6
Stiglitz, J. and Weiss, A. 1981. Credit rationing in markets with imperfect information, American
Economic Review, vol. 71, 393–410, Brussels
Usher, D. 1964. The welfare economics of invention, Econometrica, vol. 31, 279–87
Van Pottelsberghe de la Potterie, B. and Guellec, D. 2004. ‘From R&D to Productivity Growth:
Do the Institutional Settings and the Source of Funds of R&D Matter?’, Brussels, CEB
Working Paper No. 04/010, Brussels.
Wooldridge, J. M. 2002. Econometric Analysis of Cross Section and Panel Data, Cambridge, MA,
MIT Press
Wooldridge, J. M. 2005. Simple solutions to the initial conditions problem for dynamic,
nonlinear panel data models with unobserved heterogeneity, Journal of Applied Econometrics,
vol. 20, 39–54
Industrial research versus development investment 543
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
Appendix
Table A1. Industr y classifications
Industry NACE rev. 2008 Description Frequency %
1 10, 11, 12 Food and Tobacco 273 7.41
2 13, 14, 15 Textiles, Clothing and Leather 224 6.08
3 16, 31 Wood and Furniture 111 3.01
4 17, 18 Paper 99 2.69
5 19, 20 Chemicals 292 7.92
6 21 Pharmaceuticals 76 2.06
7 22 Rubber and Plastic 148 4.02
8 24, 25, 33 Metal 309 8.38
9 27, 28 Machines and Equipment 464 12.59
10 26 ICT 269 7.30
11 29, 30 Transport 127 3.45
12 41 Building and Construction 108 2.93
13 1, 5, 23, 37, 35, 32 Miscellaneous Industries 237 6.43
14 45, 46, 47, 49, 55, 58 Commerce and Transport 257 6.97
15 59, 64, 68, 69, 71–79 Other Services 452 12.26
16 61, 62 Software Development and
Communication
240 6.51
3,686 100.00
544 D. Czarnitzki et al.
at Zentrum fuer Europaeische Wirtschaftsforschung (ZEW) GmbH - Bibliothek on February 23, 2012http://cje.oxfordjournals.org/Downloaded from
