PATENTING, INNOVATIVE TRAINING AND FIRM PERFORMANCE i Maksim Belitski a,b a SPEA, Indiana University, USA b International Business School, Anglia Ruskin University, UK Email: [email protected]; [email protected]Yulia Rodionova, Leicester Business School, De Montfort University, UK Email: [email protected]Abstract This study assesses the returns to patenting and training for a panel of 4049 innovators in the UK during 2002-2009 and quantifies the incentives that patent protection provides for investment in training. When controlling for firm- and industry-specific characteristics, patent and training premiums are positive; however, returns to training vary across firm age and time. Our findings contradict the common-place assumption that there is inducement to knowledge expenditure from patent protection. These results further the understanding of managers and policy-makers on the importance of knowledge expenditure, and demonstrate that the majority of innovations are not protected by patents. JEL classification: L20, L26, O31, O34, O38 Keywords: Innovation, Patenting, Training, Patent propensity, Firm performance
52
Embed
Paper_Patenting, Innovative Training and Firm Performance performance
NES 20th Anniversary Conference, Dec 13-16, 2012 Article "Patenting, Innovative Training and Firm Performance" presented by Maksim Belitski at the NES 20th Anniversary Conference. Authors: Maksim Belitski, SPEA, Indiana University, USA; International Business School, Anglia Ruskin University, UK; Yulia Rodionova, Leicester Business School, De Montfort University, UK
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
PATENTING, INNOVATIVE TRAINING AND FIRM
PERFORMANCEi
Maksim Belitski a,b
a SPEA, Indiana University, USA
b International Business School, Anglia Ruskin University, UK
Training may also affect psychological characteristics of entrepreneurs by providing more
motivation through skill acquisition (Begley and Boyd, 1987). A resource-constrained
manager would be interested in finding out how much extra revenue could be generated from
additional investments in training.
One comprehensive review of training literature during the 80s and 90s is done by Bartel
(2000). More recently, Aguinis and Kraiger (2009) review the training literature focusing on
the benefits of training and development for individuals and teams, organizations, and society
during the 2000s. Authors call for further empirical research regarding organizational-level
benefits of training saying it is “not nearly as abundant as the literature on individual- and
team-level benefits.” They further contend that “not only have there been few empirical
studies showing firm-level impact of training, but those studies use unclear causal link back
to training activities.” Existing empirical studies analyzing the impact of training on firm
performance concentrate on general measures of training, rather than on the expenditure on
training specifically for innovation (Marotta et al., 2007; Acemoglu 1997). A summary of
empirical research on the impact of training (broadly defined) on productivity related to our
study is presented in Table 1 and presents mixed evidence.
Based on these arguments associated with the returns to investment in training we expect
that innovative outcomes are positively affected by increase in the knowledge expenditure,
because of the specific nature of training (Thornhill, 2006; Hansson, 2007). Thus, we posit:
Hypothesis 2: All else equal, investment in innovative training increases innovative
outcomes.
2.3. Patent-training relationship and training determinants
We start the discussion on patent-knowledge investment relationship with a recent work of
Rosenbusch et. al (2011) on venturing approach to innovation. They argue that ‘venturing
approach reflects the widespread assumption that in order to be successful, the entrepreneur
needs to have an innovative edge to compete against bigger, well-established incumbents’
Rosenbusch et. al (2011 p.441). In doing so the entrepreneur will use different forms of legal
and strategic protection of their innovation, looking to increase the investment in knowledge
if intellectual property rights allow for effective protection of innovative outcomes. Yet, there
is no sufficient empirical evidence to support a direct link between protection of innovation
and further investment in knowledge (e.g. training expenditure, R&D, market research).
Recently, using survey data for the U.S. manufacturing sector Arora et al., (2008) analyzed
the effect of patenting on R&D with a model linking a firm's R&D effort with its decision to
patent. Their study recognizes that R&D and patenting affect one another and are both driven
by many of the same factors. ‘Patent protection stimulates R&D across all manufacturing
industries, albeit with the magnitude of that effect varying substantially’ Arora et al., (2008:
p.1153). Almeida and Teixeira (2007) found patents positively impact on knowledge intensity
for the set of less developed countries whereas no statistically significant effect emerges in
the case of ‘higher developed converge clubs’.
No work has been done on investigating patent - innovative training link being a part of
knowledge expenditure and our study aims to bridge this gap. We hypothesize:
Hypothesis 3: All else equal, patent protection has a positive impact on firms’ innovative
training.
Regarding the drivers of training, our paper employs standard controls as found in much of
the literature (e.g., Bishop, 1991, 1997; Galia and Legros, 2004; Baldwin and Johnson, 1995),
subject to their availability in our data, including firm size; global nature of activities; number
of competitors in the industry; cooperation with universities, public and government research
bodies; ownership type; adoption of a patent; and industry dummies (e.g., Parker and
Coleman, 1999; Barrett and O’Connell, 2001).
Table 1. Existing estimates of the impact of training on firm’s performance (sorted by year of study). Study (Year) Dataset Method Performance measure Data type/ Sample size Results
Bassi (1984) Longitudinal Manpower survey
(1975-1978)
Fixed /random
effects Worker earnings
Earnings of white and non-white
males and females
While women are found to benefit significantly from manpower
training programs, no such effect was found for men
Ichniowski et al.
(1987)
Interviews of 45 steel finishing
lines in the US OLS, Fixed effects Productivity
2190 observations from 36 lines
owned by 17 steel companies
Positive effect of high and low incidence of training on
productivity in steel finishing lines
Bishop (1991) Survey by the Nat. Center for
Research in Vocat. Educational
Cross-sectional
OLS and difference Productivity growth 2594 firms
Returns on investment on 100 hours of new hire training ranged
Doubling of worker training reduces scrap rates by 7%; this is
worth $15,000.
Bartel (1994) Columbia HR Survey (1986) OLS, Probit Value added per worker 155 US enterprises in 1986 Firms operating at less than expected labour productivity
implemented training which resulted in 6% higher productivity
Tan and Batra
(1995) World Bank survey
2
OLS; Probit Log of Value added 300-56,000 firms by country
Predicted training has positive effect on value added; effects range
from 2.8% to 71% per year
Huselid (1995) 1992 survey of human resource
practices
Cross-section, as
well as Fixed effects
Tobin’s Q and gross rate
of return on capital 968 firms
High performance practices had significant effect in cross-sections
but disappeared in the fixed effects study
Black and
Lynch (1996)
National Employers Survey
(1994)
Cross-sectional
OLS
Dollar value of sales,
receipts or shipments in
1993
617 firms, matched with the Census
Bureau’s Longitudinal Research
Database for the panel study
Per cent of formal off-the job training in manufacturing, as well as
computer training in non-manufacturing sector is positively related
to productivity in the cross-section.
Black and
Lynch (2001)
EQW National Employers
survey (1987-1993)
Panel, First
differences Productivity Panel data for 1987 to 1993 Number of workers trained in a firm is not linked to productivity.
Barrett and
O’Connell
(2001)
Surveys of enterprises in
Ireland in 1993 and 1996-7
OLS and First
differencing panel Productivity
Surveys of enterprises in Ireland in
1993 and 1996-7
General and all training is positively related to productivity;
specific training has no significant impact.
Guerrero and
Barraud-Didier
(2004)
Guerrero and Barraud-
Didier questionnaire Interview
Performance, employee
productivity
1530 human resource directors
working in large companies in France
4.6% of the variance in financial performance was explained by
training (via social and organizational performance)
Cassidy et al.
(2005)
Total Factor Productivity
Survey (1999 – 2002)
Panel data fixed
effects estimation
Total Factor
Productivity
Foreign-owned and indigenous Irish
manufacturing with > 10 workers
Plants engaged in training have a TFP advantage of 0.3
Per cent, ceteris paribus
Ubeda Garcıa
(2005) Ubeda Garcıa questionnaire Interview
Level of satisfaction;
labor productivity
78 Spanish firms with more than 100
employees.
Training programs oriented toward human capital development are
related to employee, customer, business performance
Thornhill (2006) Survey of Canadian
Manufacturing firms
Weighted Heckman,
Logit, OLS
Innovation; Revenue
growth 845 firms
Training is not statistically significant for either group; Training
positive significant for innovation
Hansson (2007) The Cranet survey OLS, Probit the top 10%; upper
/lower half; profitability.
5,824 private-sector firms in 26
countries
Positive relationship between the number of employees receiving
training and being in top 10% of profitability among other firms.
Source: Bartel (2000), Aguinis and Kraiger (2009) with the authors’ additions and compilation.
2.3. Theoretical Model.
As the starting point of our analysis we modify a theoretical model developed by
Arora et al. (2012) which is used to analyze the private returns to patenting and inducement
for R&D incorporating the trade-offs of holding a patent postulated by Schankermann (1998).
From the CIS we first create a measure of the total revenue from new products (NPR) which
is total revenue (TR) times a share of revenues from new products. We consider as new
products those products that are new to the industry – and not just to the firm.
TR P1Q1 (1.1)
where P1 = average price of products and Q1 = average quantity of products. We
assume that
TR=P1Q1= PQ (1-) + PQ (1.2)
where P is the price of products and Q is the quantity of products sold. This equation
says that the total revenue is a weighted average by of revenue created with and without
patent protection, and that the revenue for items with a patent protection is greater following
Schankerman (1998). is the share of products for which patent protection was sought, i.e.
patent propensity; its estimates are not available at ONS UK and Intellectual Property Office
UK (IPO UK) data, because of no special surveys undertaken; and is the patent premium.
We assume a production function linking the share of new product innovations to
investment in innovative training, N1 = f(T) (Black and Lynch, 1996). Note that T is the
amount of money spent on training for product innovation, not the total training expenditure.
Combining with (1.2) and (1.1), we get (1.3). Taking logs, and transforming the model (1.3)
into econometric form we get (1.4), where lowercases denote natural logs:
NPR = N1 P Q (1 - + ) = f(T) P1Q (1 - + ) (1.3)
npr = p + q + ln(1- + ) + ln(f(T)) + εi (1.4)
where f(T) is thought of as an analogue of total factor productivity in a growth model.
We assume f’(T)>0 which means that NPR is an increasing function of training.
Now we can estimate (1.4) as a non-linear least squares (where is not known and
is a parameter to be estimated). The econometric model of (1.4) becomes (1.5), where A =
p+q + intercept. For simplicity we assume f(T)=T.
npri = A + b1 ln(Ti) + ln(1- i + i) + εi (1.5)
There are two issues. First, (1.5) imposes a specific non-linear specification, albeit
one that naturally follows. Second, T is endogenous. In particular, it will depend upon
unobserved firm specific differences in price and quantity. Put differently, demand shocks
(which affect p and q) will also affect innovative training expenditure. This can easily be seen
by writing p = p+ , where p is the average (across firms) price and is a firm specific
component of price. All else equal, if is high, T will be higher too. The obvious way out is
to find an instrument for T. A natural instrument for (1.5) is any variable that affects cost of
inputs, provided those are independent of demand shocks. We have explored measures from
the CIS, such as the importance of increased capacity for production or service provision to
product (good or service) and/or process innovations introduced scaled (0-3); and the
importance of knowledge factors as constraints to innovation activities or influencing a
decision not to innovate, scaled (0-3). We also attempted to find the Arellano-Bond type
instruments (e.g., Arellano and Bover, 1995) i.e. the first lagged values of innovative training
expenditure; however the sample has considerably decreased increasing the selection bias.
We modify the original model (1.5), given our data constraints and the limited
information available in the following way:
npri = A+ b1ln(Ti) + ln(1- i *(1-))+ εi = A + b1ln(Ti) + i (-1)+ εi (1.6)
where the last equality holds since in the vicinity of x=0, y=ln(1+x) can be
approximated by y=x.
Since patent propensity i is observed (equals 1 for a firm holding a patent and zero
when patent protection is not used) we can quantify the returns to patenting in addition to
establishing a direction of a relationship between patent protection (holding a patent) and the
NPR. We assume innovative firms to be identical and therefore can be interpreted as the
average patent propensity for the entire firm population. Thus, for each firm i to compute
NPRi we can use the average propensity to patent from the population of firms . Now we
can rewrite (1.6) as the reduced form
npri = A + B1ln(Ti) + B2xi + ei (1.7)
Therefore, xi= i and 0<i<1 and B2= (-1) = B2+1 (1.8)
Assuming firms choose their innovative training investments to maximize returns, so that
actual NPR and T are jointly determined by underlying firm and industry characteristics
(denoted by X) thus the estimating equation becomes:
Ti = C1 + Xi i + Bixi+ e2 (1.9)
npri = C2 + Xi i + B1ln(Ti) + B2xi+ e2 (1.10)
where C1 , C2 are vectors of intercept terms in equations (1.9) and (1.10) respectively, i
is a vector of unknown coefficients of the exogenous variables in equation (1.9), i is a vector
of unknown coefficients of the exogenous variables in equation (1.10), Xi is a vector of
exogenous variables (controls) in both equations; npr is new product revenue that serves as
dependent variable in equation (1.10); T is innovative training expenditure is endogenous
variable in equation (1.10) and therefore a dependent variable in the first stage of 2SLS
estimation in equation (1.9). Note that (1.10) is similar to (1.7). However, by estimating (1.9)
and (1.10) together in a cross section, we accomplish two objectives. First, we improve the
efficiency of the estimate, because parameters are estimated together in the two equations.
Second, we are able to estimate the incentives offered for innovative training due to patent
protection and the other factors. The econometric model of equation (1.10) based on the panel
data is as follows:
nprit = C + Xit + B1ln(Tit) + B2xit+ eit (1.11)
eit =vi + uit (1.12)
where i denotes a reporting unit (i=1, …,N) and t - the time period (t=1,..,T); C is a
vector of intercept terms, it is a vector of unknown coefficients of the exogenous variables,
Xit is a vector of exogenous variables (controls); Tit and xit are the variables of interest:
training expenditure and patent protection of a firm i in period t. The error term eit consists of
the unobserved individual-specific effects, vi and the observation-specific errors, uit.
Our study is subject to certain limitations. We do not analyze all different ways that
patenting might affect innovation; however, we do analyze NPR due to the existence of
patent protection and for different enterprise age. Given our main focus is on studying the
private returns to innovative training. Thus, while we control for training spillovers including
patenting, we do not model the impact of training on those spillovers. Nor do we consider the
impact of training on entry and associated innovation.
3. Data and Methodology
3.1. Identification Strategy
In general, many indices could be used to measure innovation (Acs and Audretsch, 1987a,
1987b; Arora et al., 2008). Commonly used indicators of innovation outcome based on the
CIS data include percentage sales of products that are new to the market or to the firm or
significantly improved compared to sales of other products. A review of the advantages and
disadvantages of such indicators and some of the studies that employ them is provided by
Vásquez-Urriago et al. (2011). Their main advantages are that they provide a measure of the
economic success of innovations, are applicable to all sectors, allow types of innovations to
be distinguished, and allow the definition of continuous variables, which contribute to the
development of econometric analyses (Negassi, 2004). Their limitations are that they are
sensitive to product life cycles and markets, which may differ in the context of competing
companies (Kleinknecht et al., 2002; Frenz and Ietto-Gillies, 2009). The number or a share of
products in the market gauged the success of firms in developing and introducing new
products is used as a substitute for a share of new products and therefore, new product
revenue. This measure was among the most widely used indicators of the firm’s innovative
outputs (Deeds and Hill, 1999; Harmon et al., 1997; George et al., 2002). New products were
viewed as the forerunners of a company’s future market offerings, and key stakeholders were
likely to weigh this variable heavily in determining the company’s viability (George et al.,
2002). For the robustness check in this study two indicators are explored: sales of products
that are new to the market per employed (in 000s £) and new product revenue per employee 3.
We define patent premium as the additional revenue from been able to protect its
innovation on the assumption that firms earn more per unit on innovations that are protected
by patents (Arora et al., 2008). Training premium is defined as the additional revenue from
knowledge expenditure in a form of innovative training and education aimed to improve
personnel skills, abilities and productivity of the innovative companies. Innovative training
and training for innovation in our study are used interchangeably.
Regarding the cross-section estimation methodology (equation 1.9 and 1.10) we employ
parametric techniques including Two-stage least squares (2SLS) and Tobit estimation to
evaluate the training premium and returns to patenting. First, 2SLS is used to deal with
potential endogeneity of training expenditure. Second, our dependent variable is double
censored, as firms can have none or all sales from products that are new to the market (new to
the market products per employee). There are several different ways of estimating such a
variable using parametric techniques (e.g., Wooldridge, 2003; p. 565). A double censored IV
Tobit model will account for this fact. This is used in several of the empirical analyses
(Negassi, 2004; Faems et al., 2005; Laursen and Salter, 2006). Tobit approach does not
invalidate 2SLS estimation, however it allows estimating the effect of training expenditure
for those firms whose NPR is strictly greater than zero and in terms of propensity changes
rather than elasticities. In effect, tobit estimation models a dual decision making process: in
our case, firms’ that have NPR equal zero and non-zero; and, if non-zero, how much to sell.
In this way, tobit estimation addresses the potential endogeneity of our independent variables
that would arise if the self-selection of firms into innovative product sales were to be omitted
from the model.
In panel data estimation (equation 1.11) we employ both non-instrumented (Pooled OLS,
Random and Fixed effects, Maximum-likelihood estimation) and instrumented approaches
(IV Random and Fixed effects and Baltagi Random Effects) with training expenditure being
instrumented. We use both instrumented and non-instrumented approaches with various
econometric estimation techniques as a robustness check of our results.
3.2. Data and variable description
The dataset used in this paper is based on two independent, albeit mergeable, datasets,
which is the CIS5 conducted bi-annually and BSD conducted annually by the ONS UK. We
further discuss several particularities of the data. First, since the survey is CIS-based, the
study can be replicated in the other 27 European Union Members, which will enable the
development of stylized facts. Our study could also be useful for North America to
demonstrate the analyses of data available for researchers on innovation and R&D (e.g.
Branzei and Vertinsky, 2006). Second, there is an inconsistency in the survey questions
between CIS4-5 and CIS6 on patent protection. Data on patent protection is available only for
the period of 2002-2006. Third, we use panel data estimation with a split by venture age to
deal with unobserved heterogeneity across the firms of different age and sectors. The
definition of a new venture (firm) varies across studies (Zahra, 1996; Rosenbusch et al.,
2011). Within the scope of this analysis, we use an average age of 10 years as a cut-off point
between young and mature firms. Fourth, the instruments chosen are treated with caution as
the integrated effect can moderate the relationship between training expenditure and firm
innovative outcome (Zhuang et al., 2009)4.
To date there have been four rounds of CIS taken place with the latest in 2009. CIS 4
covers the period 2002-2004 and includes 24.93% matched firms from 16240 firms originally
available from ONS. The CIS 5 and 6 cover 2004-2006 and 2007-2007 periods and result
28% merge from about 14000 originally available on CIS5-6 surveys. Top 5 sectors
presented in CIS4-6 panel data presented in Table 2 and the venture size - in Table 3.