Innovation by start-up firms:The influence of the board of directors∗

Christopher F Baum† Hans Lööf‡ Andreas Stephan§ Ingrid Viklund-Ros¶

August 15, 2020

Abstract

The paper examines whether the innovation potential of start-ups is influenced by

existing innovative firms via the board of directors. This channel of spillover is largely

unexplored. Our analysis of over 370,000 firm-year observations on more than 60,000

Swedish start-up firms which entered the market between 2002 and 2013 shows a

positive effect of the boards’ innovation experience on the propensity to apply for

patents, while no impact on trademark registration is found. The results are robust

to controlling for unobserved heterogeneity as well as worker and managerial mobility,

human capital, board diversity, venture capital, patent citations, firm size, region and

industry.

Keywords: start-ups, board of directors, knowledge spillovers, innovation, instrumentalvariable estimation

JEL classification: C36, D24, M13, L21, O33

∗We thank three anonymous reviewers for their constructive suggestions and helpful comments andconference participants at the 8th ZEW/MaCCI Conference on the Economics of Innovation and Patentingin Mannheim May 2019 for their comments on an earlier version of this paper. We have also benefitedfrom comments provided by Lorna Syme, Ali Mohammadi and Christian Thomann. The usual disclaimerapplies.†Boston College, DIW Berlin and CESIS. Corresponding author: Department of Economics, Boston

College, 140 Commonwealth Avenue, Chestnut Hill MA 02467 USA, baum@bc.edu‡Royal Institute of Technology§Jönköping University and DIW Berlin¶Royal Institute of Technology

1 Introduction

Knowledge is a key resource for the competitive advantage of entrepreneurial and innovative

firms (Agarwal, Echambadi, Franco & Sarkar 2004, Kogut & Zander 1992, Audretsch

& Lehmann 2006). In most technology fields, progress draws upon knowledge from a

number of earlier discoveries and experiences (Dosi & Nelson 2010). New entrants lacking

experience can encounter difficulties to overcome this disadvantage without interacting

externally with either organizations or individuals (Dalziel, Gentry & Bowerman 2011,

Jones, Coviello & Tang 2011). However, the capacity of start-ups to access and absorb

external knowledge (Cohen & Levinthal 1990) may be constrained by a limited endowment

of initial knowledge. How can small, young firms solve this dilemma? In this study, we

examine the role of the board of directors for knowledge spillovers to innovative newly

formed firms.

A number of papers (Audretsch & Feldman 1996, Audretsch & Stephan 1996, 1999)

suggest a more comprehensive role of the board of directors to compensate for limited

resources in small businesses. This idea has its roots in the management literature, which

has shown that board members serve as an extension to the management among small and

young firms (Zahra & Filatotchev 2004, Zhang, Baden-Fuller & Pool 2011). In this role,

directors of the board might be able to provide a firm with essential strategic advice and

knowledge derived from their connections to other firms and different networks (Bizjak,

Lemmon & Whitby 2009, Brown 2011, Shropshire 2010). In line with this argument,

Audretsch & Lehmann (2005) find that German initial public offering (IPO) companies’

access to research-intensive universities increases with the human capital of board directors,

proxied by their academic degrees.

Recently, some innovation studies were able to deepen the analysis on board of direc-

tors (BoD) in their role as conduits for spillovers, by using more detailed data, and in

some cases employing improved identification strategies to enable a causal interpretation

of results. Among these papers, Balsmeier, Buchwald & Stiebale (2014) analyze panel

data on the largest public German companies, while Balsmeier, Fleming & Manso (2017)

study major public US-based firms, and Robeson & O’Connor (2013) examine Fortune

1000 firms. Helmers, Patnam & Rau (2017) use data on all publicly listed firms in In-

1

dia, Srinivasan, Wuyts & Mallapragada (2018) apply their analysis on publicly listed U.S.

firms for consumer packaged goods, and Arzubiaga, Kotlar, De Massis, Maseda & Iturralde

(2018) study family businesses. Despite using different measures of innovation, of board

members’ competence, of channels for external networks and of knowledge in the external

networks, and regardless of employing different econometric approaches, all the studies

above come to a similar conclusion: the board of directors can be an important channel for

transferring knowledge to innovative companies. What the existing research to date has

not yet documented is whether this conclusion also holds for start-ups with their specific

characteristics that distinguish them from public and larger firms. Our paper aims to fill

this gap.

The present study uses approaches which are partially found in existing microecono-

metric studies on knowledge spillovers through board members. Our overall analytical ap-

proach is inspired by Audretsch, Lehmann &Warning (2005) and their method of weighting

the board’s absorptive ability using data on the directors’ academic education level. Fol-

lowing Robeson & O’Connor (2013), Hoskisson, Hitt, Johnson & Grossman (2002) and

Arzubiaga, Kotlar, De Massis, Maseda & Iturralde (2018), we share the idea of using the

composition of the background of the board’s directors to study possible influences on

the focal firm’s innovation engagement. Our study utilizes the links of board members

with innovative companies for the weighting calculation. Following Balsmeier, Buchwald

& Stiebale (2014), we employ patent applications, one of the two formal intellectual prop-

erty mechanisms, to identify whether the directors are linked to innovative firms. This

link can be through board membership or employment. Similar to Srinivasan, Wuyts &

Mallapragada (2018) we identify directors serving on one or multiple boards. In line with

Helmers, Patnam & Rau (2017), Balsmeier, Buchwald & Stiebale (2014), Balsmeier, Flem-

ing & Manso (2017) and Srinivasan, Wuyts & Mallapragada (2018), we address potential

endogeneity problems in the empirical analysis.

Our work differs from previous studies in several important aspects which makes this

paper a unique contribution to the understanding of the role of board directors for new, in-

novative entrants. First, we are studying genuine and privately owned start-ups, implying

that there is no internal knowledge transfer between companies within the same ownership

group. We also remove spin-outs from the sample and control for spillovers through mi-

2

gration of management and employees. Second, we are able to track the background of all

employers, all employees and all directors in the new firms and their links to other compa-

nies across the entire economy. Third, we consider outside directors linked to other firms

through employment or membership of the board, as well as inside directors interlocked to

external boards.1 Fourth, we apply both patent applications and trademarks as indicators

of innovation. Our matching rate is above 99% for global patent applications registered

by PATSTAT, and all trademark applications on the European market are identified in

our study. Fifth, we account for board diversity, venture capital and patent citations.

Finally, we are able to conduct a comprehensive sensitivity analysis that tests for different

specification of empirical models and different samples.

Our data consists of 11 annual cohorts of start-ups formed as micro-firms (with a

maximum of 9 employees during the founding year) between 2002 and 2012. In total, the

sample contains 374,198 firm-year observations with detailed information on 61,740 new

entrants to the market. Beyond the availability of comprehensive, high quality data on

firms and their employees and board of directors, the justification for focusing on Sweden

for this study is the country’s position as one of the world’s leading innovative economies.

Global companies like the music-streaming service Spotify, the online-payment firm Klarna

and the gaming company King are all examples of successful Swedish start-ups.2 We are

convinced that the results from Sweden can be generalized to other countries with one-tier

‘Anglo-American’ board systems.

The empirical analysis using an instrumental variable approach shows that start-ups

with directors linked to innovative firms are more likely than other young entrepreneurial

firms to generate new ideas, measured by patent applications. When using trademarks as

an innovation indicator for established and new firms, we find no spillover effect. However,

a causal influence can be established by incumbents with trademarks on the propensity of

start-ups to apply for patents.

The rest of the paper is organized as follows. Section 2 provides the theoretical sup-1Inside directors are employed by the firm, while outside directors are not. Through firm- or industry-

specific experience, outside directors are potentially associated with higher information quality (Rutherford& Buchholtz 2007), valuable strategic advice (Kor & Sundaramurthy 2009), first-hand knowledge, expertiseand scarce information not easily acquired elsewhere (Balsmeier, Buchwald & Stiebale 2014).

2According to the European Innovation Scoreboard for 2019, Sweden held the first place in Europe fol-lowed by Finland, Denmark and the Netherlands, see https://ec.europa.eu/growth/industry/policy/innovation/scoreboards_en retrieved on 30th June 2020.

3

port for the study and formulates hypotheses to be tested. Section 3 presents our data,

variables and descriptive statistics. The identification strategy and empirical approach are

described in Section 4. In Section 5, the empirical models are evaluated followed by a

section that conducts sensitivity tests. The final section summarizes our main results, dis-

cusses implications for management and policymakers, and suggests directions for further

studies.

2 Theory and hypotheses

Economically viable new firms are found to be positively associated with innovation ca-

pability (Lentz & Mortensen 2008, Neumark, Wall & Zhang 2011, Kerr, Akcigit, Bloom,

Acemoglu et al. 2012, Decker, Haltiwanger, Jarmin & Miranda 2014, Akcigit & Kerr 2018,

Bloom, Jones, Van Reenen & Webb 2020). Understanding why some new firms become

innovators while most other start-ups do not has therefore gained considerable attention

from both researchers and policymakers (Baumol 2004, Van Stel, Carree & Thurik 2005,

Grossmann 2009). In this paper, we study the relationship between new and established

firms, and ask whether board members linked to both may be contributing factor.

A theoretical basis for our analysis are the micro-founded neo-Schumpeterian models

developed in recent decades (Kortum & Lerner 2000, Klette & Kortum 2004, Acemoglu

& Cao 2015). Herein, the interplay between R&D-investing incumbent firms and new en-

trepreneurial firms is important not only for the entrants but has a key role for industrial

dynamics and sustained economic growth. In this view, firms are at least partly “standing

upon the shoulders of giants” by drawing upon knowledge from a number of earlier dis-

coveries and experiences in their process of creating new innovations (Caballero & Jaffe

1993). This model has been accompanied by the development of an empirical literature

providing ample evidence on the existence of spillovers, as well as confirming the seminal

work by Cohen & Levinthal (1990) on the importance of a firm’s capacity to recognize,

assimilate and apply knowledge created elsewhere (Chesbrough 2003, Nieto & Quevedo

2005, Kodama 2008, Ferraris, Santoro & Dezi 2017, Oh 2017, Xie, Zou & Qi 2018).

Many previous studies from multiple subdisciplines of economic research, among others

from management (Xie, Zou & Qi 2018), entrepreneurship (Qian & Acs 2013) and the

4

spillover literature (Mancusi 2008, Lim 2009, Fosfuri & Tribó 2008) have confirmed that a

firm’s ability “to recognize the value of new, external information, assimilate it, and apply

it to commercial ends” (Cohen & Levinthal 1990) contributes either directly or indirectly

to a firm’s innovation performance.

While Cohen & Levinthal (1990) found R&D investment to be a primary condition for

a firm’s absorptive capacity of external knowledge, a number of more recent studies have

documented other mechanisms that may facilitate spillovers and reduce frictions due to

factors such as complexity (Spender & Grant 1996), tacitness (Polanyi 1966) and stick-

iness (Von Hippel 1994) of knowledge. They include geographical proximity (Feldman

1994, Breschi & Malerba 2001, Lissoni 2001) and relational networks linking nodes within

innovation systems and value chains (Rodríguez-Pose & Crescenzi 2008, Fritsch & Franke

2004, Tallon 2011, Winkler 2013). Another factor is entrepreneurial mobility, whereby in-

dividuals from successful incumbents move to new ventures (Agarwal, Echambadi, Franco

& Sarkar 2004, Klepper 2007, Howard, Boeker & Andrus 2019).

A common theoretical basis of most of this literature is the notion that knowledge is

embedded in individuals such as workers, engineers, scientists, consultants, and managers.

Personal face-to-face contacts through collaboration, job mobility or geographical proxim-

ity is therefore considered to be crucial for the spillover of complex, tacit, sticky and other

knowledge (Breschi & Lissoni 2001, McCann & Simonen 2005, Henderson 2007, Cantner,

Graf, Herrmann & Kalthaus 2016).

The knowledge spillover theories of entrepreneurship (Audretsch & Keilbach 2005, Qian

& Acs 2013, Ghio, Guerini, Lehmann & Rossi-Lamastra 2015), provide another line of

arguments which can explain firms’ ability to benefit from external knowledge. In this

literature, knowledge capabilities by the board of directors are considered to be a possible

substitute for young companies’ lack of R&D and experiences from yesterday’s failures and

successes transformed into today’s learning routines and stock of knowledge (Audretsch &

Stephan 1996).

Only a relatively few firms in each sector are responsible for the major fraction of re-

search investments, technological development, innovation output and the knowledge stock

in the economy (Cohen & Klepper 1992, Klette & Kortum 2004, Akcigit & Kerr 2018, Coad

2019). Migration and spawning research has examined how these successful innovative in-

5

cumbent firms in a first stage act as training grounds for workers, managers, executives

and founders, subsequently seeding the market with knowledge, skills and experience when

individuals or groups of people start, join or interact with entrepreneurial firms (Klepper

2001, Agarwal, Echambadi, Franco & Sarkar 2004, Campbell, Ganco, Franco & Agarwal

2012). This migration of expertise has been found to influence the quality of knowledge

transfer due to factors such as market overlap, direct engagement and employee hiring

(Howard, Boeker & Andrus 2019). Our lesson from this research is that directors associ-

ated with innovative incumbents may have a larger likelihood of spurring inventiveness of

new firms compared to directors linked to non-innovative companies.

The role of board members for firms’ efficiency and performance has a long tradition

in the corporate governance literature (Jensen & Meckling 1976, Fama & Jensen 1983,

Hermalin & Weisbach 1998). The vast majority of empirical research on the role of boards

has been conducted on large public companies. However, recently there has been an interest

among entrepreneurship researchers to study the board’s role in small businesses. This

research suggests that boards, rather than controlling managers to reduce agency issues,

actually may serve as an extension to the management in small businesses by providing

information, advice and knowledge derived from the directors’ links to other firms and

networks (Zahra & Filatotchev 2004, Bizjak, Lemmon & Whitby 2009, Shropshire 2010,

Brown 2011, Zhang, Baden-Fuller & Pool 2011). So far, few of these studies have been

conducted on representative samples of start-ups and new companies, and virtually none

of these have focused on innovative start-ups. The purpose of this paper is therefore

to contribute to the literature by analyzing directors’ ability to channel knowledge from

innovative incumbents to new innovation-oriented companies.

2.1 Innovation measures

Our study is based on the resource-based view of the firm in regards to ownership or control

of knowledge and intellectual property (IP). It has been suggested that the protection of

knowledge and technology as a competitive advantage is especially important for young

companies. These often lack the control over their ownership and complementary assets

for innovation which, in contrast, established and resourceful companies do have (Teece

1988).

6

A firm’s knowledge or intellectual assets can be protected by patents, trademarks,

copyright, secrecy, complexity or first-mover advantage. Within this set of protection

mechanisms, patents are the most studied mechanism in the literature (for a survey, see

Hall, Helmers, Rogers & Sena 2014). Patents offer a standardized and transparent measure

of inventive activity (Popp 2019). They identify individuals, firms and organizations, and

contain information on the prior knowledge on which the patents are based. Patents provide

a good indicator of R&D (Griliches 1991), and they may also capture non-formal research

investments. Moreover, patents are often a predictor of new product announcements (Artz,

Norman, Hatfield & Cardinal 2010), although with variation across firm sizes (Arundel

2001) and industries (Mansfield 1986). On the other hand, a well-known insight from this

literature is that patents have drawbacks as indicators of innovation and are not always

the most suitable measure of firms’ intellectual assets.

Another instrument for protecting intellectual property, which finds more and more

attention in research, is the registration of trademarks. Patents and trademarks as formal

appropriation mechanisms provide the owner with the exclusive right to use or sell the

invention, and they are found to be both substitute and complementary modes of protec-

tion. (Block, De Vries, Schumann & Sandner 2014, Zhou, Sandner, Martinelli & Block

2016, Veugelers & Schneider 2018). A trademark is a word, symbol, or other expression

used to distinguish a good or service produced by one firm from the goods or services of

other firms (Landes & Posner 1987). Trademarks are increasingly used as an alternative

proxy for innovation, especially in small firms because trademark registration is relatively

inexpensive and straightforward and may therefore suit resource-scarce innovative start-

ups (for a recent survey, see Block, Fisch, Hahn & Sandner 2015). Also, trademarks seem

to have a similar effect as patents on firm value, productivity and survival (Sandner 2009,

Crass 2020).3

Especially for young companies, protection of intellectual property is not solely about

reducing the risk of imitation, infringement and theft of their invention. IP rights may

also have a signalling value to investors and can serve as collateral in financial markets.

Being financially constrained, small firms may lack the resources needed to produce and3For more detailed discussions on the role of intellectual property rights as innovation indicators, see

among others Verhoeven, Bakker & Veugelers (2016), Nagaoka, Motohashi & Goto (2010), Holgersson(2013), Morrar (2014), Gotsch & Hipp (2012) and Mendonça, Pereira & Godinho (2004).

7

commercialize the innovation (Hall & Lerner 2010), and may lack access to financial mar-

kets. Patents and trademarks have also been found to facilitate licensing of the invention,

improving the attraction of brands Veugelers & Schneider (2018) and enhancing reputation

(Audretsch, Bönte & Mahagaonkar 2012, Söderblom, Samuelsson, Wiklund & Sandberg

2015, Colombelli, Grilli, Minola & Mrkajic 2019).

The innovation literature contains a variety of other measures to compare companies’

ability to generate new ideas besides patent applications and trademarks. However, most

of them are not applicable to a study on start-ups such as ours. The firms in our sample

are young and have a maximum of 9 employees in the year of their formation. First, while

R&D is a common measure for studying investment in technological development, it is

less relevant for young and small firms with mainly informal innovation activities. Second,

the European Community Innovation Survey (CIS) has successfully introduced innovation

sales as an innovation measure suitable for both manufacturing and service companies.

However, among innovative start-ups, it is common that the market introduction of new

products or services takes several years, which means they do not have any sales revenue

from innovations. Third, using granted patents instead of patent applications or citation-

weighted patents is not possible due to the long time lag. Fourth, total factor productivity is

not a meaningful measure of innovation and technical change for new and small entrants on

the market. Finally, we are not able to observe intellectual property protection mechanisms

such as secrecy, complexity or first-mover advantage for the start-ups in our sample.

2.2 Hypotheses

While many previous studies on the board of directors have focused on topics such as

the ratio between insiders and outsiders (Hermalin & Weisbach 1998), board diversity

(Schubert & Tavassoli 2020), the moderating role of CEO power (Combs, Ketchen Jr,

Perryman & Donahue 2007) or directors’ experience (Kor & Sundaramurthy 2009), our

paper belongs to the strand of literature that examines the directors’ links to innovative

firms.

We separate the board members into two different categories. The directors in the

first category have no formal connection to external networks consisting of innovative

companies. This division is presented in Table 1.

8

Drawing upon the extensive literature on knowledge spillovers and absorptive capacity,

as well as recent findings from studies on innovation and outside board members (e.g.,

Balsmeier, Buchwald & Stiebale 2014), we formulate the first hypothesis as follows:

Hypothesis 1: Board members not linked to any innovative firm through either employ-

ment or board membership do not affect a start-up firm’s propensity to apply for patents.

The second category contains inside and outside board members and interlocked di-

rectors that are associated to innovative firms through employment, board membership,

or both. We expect that experience from and connections with innovative firms play an

important role in a board member’s ability to recognize, assimilate and absorb knowledge

spillovers from sources in the local, regional or global economy. This leads us to formulate

the second hypothesis:

Hypothesis 2: Board members linked to innovative firms through employment or board

membership positively affect a start-up firm’s propensity to apply for patents.

Our third hypothesis considers the two formal mechanisms for protecting intellectual

assets, patents and trademarks. As briefly discussed above, recent literature suggests

a large degree of similarity between the two protection mechanisms, while trademark is

an attractive choice especially for young firms because of their relatively low costs and

simplicity. We therefore hypothesize that trademarks can be used as a proxy for innovation

similar to patents when examining spillovers through the board of directors in start-up

firms.

Hypothesis 3: There is no significant difference in the importance of patents and trade-

marks as proxies for knowledge transfer via board members from innovative incumbents to

future innovative start-ups.

In the next two sections, we present data, variables and the empirical approach before

testing these hypotheses.

3 Data and variables

The firm-level data used in this study are constructed from several sources. We use the

commercial database Serrano from the company Bisnode to identify 11 cohorts of start-

ups in Sweden formed as micro-firms (9 or fewer employees) over the period 2002–2012.

9

In total, our sample contains 374,198 firm-year observations on 61,740 new entrants. The

sample is restricted to only include limited liability companies as those firms are by law

required to have a board of directors. Our main sample also excludes firms that belong

to business groups and spin-offs4 and new firms with only one employee throughout the

sample period.

Information on boards of directors is retrieved from the Swedish Companies Registra-

tion Office. Using unique identification codes, we merge these two datasets with official

register information provided by Statistics Sweden (SCB) on all firms in Sweden and all in-

dividuals linked to these firms (employer-employee data: EE). The EE data includes exten-

sive statistics on both firms and individuals. We then match patent data from PATSTAT

(EPO) and trademark data from European Union Intellectual Property Office (EUIPO)

with the EE data. We are able to identify more than 99% of the global patent applications

by firms in Sweden, and 100% of their trademarks granted by EUIPO. Finally, we add Ven-

ture Capital data provided by the Swedish House of Finance (SHoF) to the constructed

data set.

We classify incumbent firms as innovative in year t if they applied for a patent in any

of the three previous years. This approach is also used for trademarks as a proxy for

innovation.

According to our definition, a connection between a director on the board of the start-

up firm and an innovative incumbent in a given year exists if the director serves on the

board of the start-up in year t and also is employed in or serves on the board of another

firm defined as innovative in year t.

Table 1 defines the two categories of directors we consider in the analysis. The first

category consists of inside and outside directors without any employment or interlock

connections with innovative firms. In the second category, inside and outside directors have

formal associations with innovative incumbents. Based on this classification, we construct

a dummy variable BCI that indicates whether any of the directors on the start-up’s board

belong to the second category.

Table 2 presents the dependent variables constructed on basis of the classification of4We define start-ups as spin-offs or spin-outs if half or more of the employees were employed in the

same parent firm in the year before firm formation.

10

directors, the instruments and the covariates. The size of each new firm is captured by

the log of total assets. Human capital is represented by the fraction of employees with

three years of university education or more. This corresponds to about a quarter of the

employees in our data. One out of five new companies is located in one of the three Swedish

metropolitan areas: Stockholm, Gothenburg or Malmö.

Table 1: Variable descriptions I: Categories of directors

Category 1:(i)i,t−1 Inside director with no interlocking board(ii)i,t−1 Inside director interlocked with board of one or more non-

innovative firms(iii)i,t−1 Outside director employed in a non-innovative firm and not

interlocked with any board(iv)i,t−1 Outside director employed in a non-innovative firm and in-

terlocked with board of one or more non-innovative firms(v)i,t−1 Outside director not employed in any firm and interlocked

with board of one or more non-innovative firms(vi)i,t−1 Outside director not employed in any firm and not inter-

locked with board of any firm.Category 2:(vii)i,t−1 Inside director interlocked with the board of least one inno-

vative firm(viii)i,t−1 Outside director employed in an innovative firm and with no

interlock(ix)i,t−1 Outside director employed in a non-innovative firm and in-

terlocked with the board of one or more innovative firms(x)i,t−1 Outside director employed in an innovative firm and inter-

locked with the board of one or more non-innovative firms(xi)i,t−1 Outside director employed in an innovative firm and inter-

locked with the board of one or more innovative firms(xii)i,t−1 Outside director not employed in any firm and interlocked

with the board of one or more innovative firms

Table 3 reports the summary statistics for the sample of young firms during the first year

of observation, which is three years after formation. The dependent variable is observed in

year t and the explanatory variables in year t− 1. As could be expected, the proportion of

new firms that apply for patents already during their first years of existence is small. Only

0.2% of the new entrants apply for at least one patent in the third year after the firm was

founded, and 0.3% registered at least one trademark.

In the empirical analysis, we observe companies from three years after foundation up

11

Table 2: Variable descriptions II

Dependent variablesPatenti,t indicator (0/1): firm i applied for one or more patents in

year t.Trademarki,t indicator (0/1): firm i applied for one or more trademarks

in year t.Key determinantBCIi,t−1 equals 1 if any of the directors on the board are connected

to an innovative (patent) firm (Category 2).BCTi,t−1 equals 1 if any of the directors on the board are connected

to an innovative (trademark) firm.InstrumentsNew OBCI1i,t−1 equals 1 if any of the outside directors were newly hired in

a firm with patenting experience in year (t − 1) and wereemployed in a different firm without patenting experience inyear (t− 2), 0 otherwise.

New OBCI2i,t−1 equals 1 if any of the outside directors were newly hired ina firm with patenting experience in year (t − 2) and wereemployed in a different firm without patenting experience inyear (t− 3), 0 otherwise.

New OBCT1i,t−1 equals 1 if any of the outside directors were newly hired ina firm with trademark experience in year (t − 1) and wereemployed in a different firm without trademark experiencein year (t− 2), 0 otherwise.

New OBCT2i,t−1 equals 1 if any of the outside directors were newly hired ina firm with trademark experience in year (t − 2) and wereemployed in a different firm without trademark experiencein year (t− 3), 0 otherwise.

Control variablesBoard sizei,t−1 number of directors on the focal firm’s board.log(Total assets)i,t−1 log of total assets, winsorized.Human capitali,t−1 share of employees with three or more years of university

education.Metroi,t−1 indicator (0/1): focal firm is located in metro area (Stock-

holm, Gothenburg or Malmö).Firm agei,t−1 firm age during the estimation sample, 2–10 years.IE10i,t−1 equals 1 if the share of employees whose last employment

was with an innovative firm > 0.1, 0 otherwise

12

to a maximum of 13 years. Thus, companies created in 2002 may be observed for 13 years

while companies created in 2013 are observed only for one year. The total number of

unique start-ups in our sample seeking patents is just above 150 and the number of unique

companies with trademarks is about 300. In total, we have about 540 and 400 firm–year

observations of patent and trademark applications respectively. Summary statistics for the

full panel of start-ups is provided in Table 4.

4 Empirical models and identification strategy

Our empirical models analyze the influence of directors’ external connections on firms’

propensity to be innovative, using patent applications and granted trademarks as proxies.

The first model estimates the probability that the focal firm applies for a patent in year

t, conditional on the number of board directors of each of the twelve types described in

Table 1. This model is estimated separately for each of those director types to gauge the

importance of their characteristics on the firm’s innovation.

For firm i in industry j and time t, the model is

Pr[Yi,t = 1] = Ψ(αPi,t−1 +X ′i,t−1β + θj + φk + τt + ei,t) (1)

where Yi,t is an indicator of whether the focal firm applied for any patents during year t,

Ψ(·) is the CDF of the normal distribution, and Pi,t−1 is the number of directors on the focal

firm’s board with connections ((i)-(xii)) to innovative firms as specified in Table 1. Xi,t−1

is a vector of firm-specific control variables including firm age, total assets, human capital,

metro location, board size and the share of employees with work experience in innovative

firms, while θj , φk and τt denote industry, cohort and year fixed effects, respectively.

We then extend these descriptive analyses of directors’ linkages to innovative firms

with a second model that accounts for the potential endogeneity of the firm’s choice of

directors. It is likely that the owners of the firm actively seek “high quality” directors to be

appointed to the board by screening characteristics of the potential directors’ employers

and networks. This implies that the presence of directors linked to innovative firms could

be endogenous, as directors with that experience may be more willing to serve on the firm’s

board if it exhibits innovative behavior. Without appropriate instruments for mitigating

13

endogeneity, one cannot establish a causal link between external knowledge via outside

directors and firm performance.

In this second model, we consider a binary indicator signalling the presence of one

or more ‘Category 2’ directors, as defined in Table 1, on the board. To allow for the

potential endogeneity of that measure, we construct two instruments using information on

innovation characteristics of firms other than the focal firm. The first instrument variable

NewOBCI1i,t−1 equals 1 if any of the outside directors who were appointed to the focal

firm’s board in (t − 2) or earlier were newly hired in a firm with patenting experience in

year (t − 1) and were employed in a different firm without patenting experience in year

(t − 2), and equals 0 otherwise. Our second instrument, NewOBCI2i,t−1, equals 1 if any

of the outside directors who were appointed to the focal firm’s board in (t − 3) or earlier

were newly hired in a firm with patenting experience in year (t− 2) and were employed in

a different firm without patenting experience in year (t− 3), and is 0 otherwise.

We assume that the owners of the focal firms cannot foresee that elected directors

will change their place of work in the future, so that the instruments can be considered as

predetermined. Furthermore, we conjecture that there is some stickiness in the composition

of the board, so that current directors are more likely to be candidates and will be reelected

in the following year. Changes in the external directors’ employment can thus be argued

to be exogenous to the focal firm.

For firm i in industry j and time t, the model is

Pr[Yi,t = 1] = Ψ(αBCIi,t−1 +X ′i,t−1β + θj + φk + τt + ei,t) (2)

where Yi,t is an indicator of whether the focal firm applied for any patents during year t, and

BCIi,t−1 indicates whether any of the inside or outside directors on the focal firm’s board

have connections to innovative firms. Xi,t−1 is a vector of firm-specific control variables

including firm age, total assets, human capital, metro location, board size and share of

employees with work experience from innovative firms, θj , +φk and τt denote industry,

cohort and year fixed effects, respectively.

Equation (2) estimates the new firms’ propensity to be innovative, proxied by patent

14

applications. To handle potential endogeneity we specify a second equation:

BCIi,t−1 = X ′i,t−1π0+π1NewOBCI1i,t−1+π2NewOBCI2i,t−1+λj+γk+δt−1+νi,t−1 (3)

Equations (2) and (3) are estimated applying a recursive bivariate probit model. Following

procedures suggested by Wooldridge (2005), Papke & Wooldridge (2008) and Semykina

(2018), we also apply a correlated random effects (CRE) approach by adding firm-specific

time averages of all time-varying covariates to Equations (2) and (3). We also include

firm-specific averages of year dummies to both equations as recommended by Wooldridge

(2019) in the context of unbalanced panels.

5 Results

In this section, we present the results of testing our hypotheses for the influence of the

board of directors on innovation by recently formed firms. The main sample consists of

61,740 unique firms established as limited liability companies between 2002 and 2013 and

their possible links, through employment or board membership, to all other limited liability

companies in the Swedish economy.

Our first set of results is an exploratory analysis, testing the hypothesis that the pres-

ence of board members characterized by the 12 director types influences firms’ innovation,

proxied by patent applications. ‘Category 1’ directors lack links to any innovative firm

through either employment or board membership, while ‘Category 2’ directors exhibit

such links. This facilitates a detailed analysis of the relevance of directors’ characteristics

for firms’ innovation. In this preliminary analysis, we do not account for possible endogene-

ity of directors’ motives for joining the board, so that these results can only be interpreted

as associations. Next, we estimate an instrumental variables Recursive bivariate probit

model for patents and trademarks to analyze the potential causal effects from directors

with links to innovative firms.

15

5.1 Exploratory analysis of the influence of different categories of board

members

Most newly established Swedish firms have no link to external networks through their BoD.

Among start-ups with BoD connections to other firms, the tie is usually to non-innovative

firms. Only a small fraction of start-ups have inside or outside directors associated with

innovative firms via employment or their boards.

In our initial analysis, using a binomial probit model, we examine how the likelihood of

patent applications varies across new firms with the presence of directors of each of the 12

types described in Table 1. The estimates are conditional on firm characteristics, industry

classification, labor mobility from innovative firms, geographical location, and time effects

as specified by Equation (1).

Table 5 presents results for the 12 models corresponding to these director types. The left

panel (Category 1) reports estimates for the presence of directors not linked to innovative

firms. The right panel (Category 2) shows estimates for the directors who are employed in

or are members of the board in innovative firms.

We obtain negative and partially significant estimates for directors belonging to Cat-

egory 1 with exception of sub-group (v): outside directors not employed in any firm and

interlocked with the board of one or more non-innovative firms. Column (v) reveals a

positive and significant association with patent applications. In contrast to the results for

Category 1, the director types in Category 2, capturing links to innovative firms, are all

positive, and four of the six estimates are significantly different from zero.

We obtain positive and highly significant coefficients across all 12 models on board size,

total assets, human capital and employees whose last employment was with an innovative

firm.

Our initial findings based on the exploratory results from these probit models imply

that we can reject Hypothesis 1, which states that board members not linked to any

innovative firm through either employment or board membership do not affect start-up

firms’ propensity to be innovative. These findings suggest that a larger number of non-

innovative directors reduce the probability of patent applications. We conduct a sensitivity

test of this conclusion by including all 12 types of board members in one and the same

16

regression. These results are reported in Table 11 in Appendix 2 and they are in line with

these findings.

Table 6 presents the average marginal effects (AME) for the binomial probit models.

The magnitude of the marginal estimates on the number of directors linked to innovative

firms are within the range of 0.001–0.003. This is equivalent to saying that an additional

director interlocked with or employed by an innovative firm increases the likelihood of a

focal firm to apply for a patent in the next period by 0.1–0.3 percentage points.

The marginal effect on the patent application of recruiting a board member with links

to boards of innovative companies is within the same order of magnitude as the effect

of human capital and employees with experience from previous employment in innovative

companies. However, as the BoD variable is the number of directors while labor mobility

is binary variable and human capital is expressed as a fraction, the coefficients on directors

and employees are not directly comparable.

5.2 Recursive bivariate probit model

The binomial probit estimates presented above indicate the presence of knowledge spillovers

between innovative firms and new firms through board members’ characteristics. The pres-

ence of directors recruited from innovative firms are positively associated with start-up

firms’ likelihood to apply for a patent. However, the results may be caused by reverse

causality: start-ups formed by innovative entrepreneurs are probably more likely to suc-

cessfully recruit directors with links to other innovative companies.

To test our second hypothesis stating that board members linked to innovative firms

through employment or board membership positively affect a start-up firm’s propensity to

be innovative, we need to ensure that the influence of board members is not affected by en-

dogeneity bias. We address this concern by estimating the recursive bivariate probit model

described by equations (2) and (3) in Table 7. The model include external instruments for

the binary variable BCIt−1, which signals whether the focal firm has at least one director

with a connection, through employment or other board appointments, to an innovative

firm. The variable incorporates all six alternative board compositions in Category 2 in

Table 1. The instruments New OBCI1t−1 and New OBCI2t−1, are defined in Table 2.

As the main objective of the paper is to examine the role of BoD for knowledge spillovers,

17

we do not include any instruments for the labor mobility variable which also may capture

diffusion of knowledge.

Table 7 reports estimates of recursive bivariate probit and Correlated Random Effects

(CRE) probit models, including firm-specific time averages of all time-varying variables

to control for firm-specific time invariant effects. As the CRE model also captures unob-

served heterogeneity, conditional on random effects, it may be considered as the preferred

estimator. In both columns, we obtain positive and highly significant coefficients on the

BoD spillover variable (BCI ). The magnitude of the estimates is almost identical in two

columns, and the two instruments (NewOBCI1 and NewOBCI2 ) are highly significant in

both models.

Tests of instrument validity from a linear probability model presented in Table 17

show that we can reject both the null hypothesis of under-identification and of weak in-

struments. Also, we cannot reject the hypothesis testing the overidentifying restrictions.

Taken together the test results suggest that our instruments are valid.

Concerning the controls, there are some differences between the two models. The

variables board size, total assets, human capital and employees recruited from innovative

firms are positive and highly significant in the pooled model, but not significant in the

CRE model. An explanation for the latter is that the CRE model controls for the mean of

all continuous covariates. This implies that the impact on patenting from human capital

and labor migration as well as the size of the board is in accordance with the literature.

Average marginal effects for the recursive bivariate probit models are presented in Table

8. The point estimates suggest that firms with at least one director on the board with a

connection to an innovative firm have a 0.5 percentage point higher probability of applying

for a patent than firms with no connections to innovative firms through their board of

directors.

Results from the instrumental variable model imply that we cannot reject Hypothesis 2:

that board members linked to innovative firms through employment or board membership

positively affect a start-up firm’s propensity to apply for patents.

Our third hypothesis proposes that there is no significant difference in the importance of

patents and trademarks as proxies for knowledge transfer via board members from innova-

tive incumbents to future innovative start-ups. When substituting Trademark applicationt

18

for Patent applicationt (hence BCTt−1 for BCIt−1) we also must construct new instruments,

NewOBCT1t−1 and NewOBCT2t−1, following the same procedure as for NewOBCI1t−1

and New OBCI2t−1, for trademarks rather than patent applications. Here, we find no

support for knowledge spillover from incumbent innovative firms to start-up firms via the

board of directors, as reported in Tables 9 and 10.

The first equation in the recursive model shows that the two external instruments

(NewOBCT1 and NewOBCT2 ) are highly significant in both models. However, the point

estimate for boards linked to firms with trademarks is not significantly different from zero

in either model. We therefore reject Hypothesis 3: that trademarks can be used as a

proxy for firms’ innovation as affected by directors linked to innovative firms. A tentative

explanation for this finding is that patent applications require more technical expertise,

so that knowledge spillovers are more relevant for patent applications than for trademark

applications.

6 Sensitivity analysis

The conclusion from these empirical results is that we find evidence of knowledge trans-

fer via the firm’s directors between established innovative companies and new innovative

companies when patent applications are used as a proxy for firms’ innovation.

In this section, comprehensive sensitivity tests of the patent estimates are performed.

We test for the effect of the board members’ educational background, investigate the

importance of venture capital, explore the impact of patent citations, and evaluate different

restrictions on firm size and definitions of start-up firms. We also study alternative models

and extensions of the estimation sample in Appendix 2, evaluating whether the restrictions

imposed to define that sample are driving the results.

Our first robustness test concerns the results for the two categories of directors re-

ported in Table 5. Estimating separate equations for the different characteristics of board

members, we find that directors not linked to any external innovative firm may negatively

influence the focal firm’s propensity to apply for a patent. This result is confirmed in

Table 11, where all variations of the board members’ characteristics are estimated in one

equation rather than 12 different equations.

19

The second sensitivity test applies instrumental variables techniques to focus on the

results presented in Table 7. A concern with the reported patent estimates is that we retain

the companies that have applied for patents in a given year in our sample. These companies

may be more likely than other firms to apply for patents in subsequent years. As we are

not modeling this potential autocorrelation at the firm level, we evaluate its importance

by restricting the sample to only include firms which have not applied for a patent in prior

years. A firm will be excluded from the sample after its first patent application. Table 12

reports recursive bivariate probit estimates from this reduced sample. The results show

that the magnitude of the BoD indicators’ coefficient estimates are somewhat lower in

both the pooled and CRE model compared to Table 7. However, they are still positive

and highly significant.

Another concern with our main model for patent applications is that we do not control

for the educational background of members of the board.5 As a robustness test, we include

two new variables in Table 13. The first, Board human capital (BHC), measures the fraction

of directors with three or more years of university education. The second is the Blau index

(Blau 1977) capturing the diversity of the Board members Educational Background (BEB).

This measure is defined as:

BEBdiversityi,t−1 = 1 −J∑

j=1

BEB2i,t,j (4)

where BEB is the share of the directors with educational background j of the total number

of directors in firm i in year t. As directors who are also employed in the firm (insiders)

contribute to both variables, we remove the Human Capital variable in this robustness test.

Table 13 shows a positive and significant estimate for board human capital. However, the

inclusion of board characteristics has no impact on the main results.

Extensive research within various strands of management, entrepreneurship and finance

provides evidence on the importance of venture capital (VC) for innovative small businesses.

To investigate whether our results may be driven by access to VC rather than knowledge

spillover from board members, we compare two model specifications in Table 14. The

first column presents results including a binary variable indicating whether the focal firm5We thank an anonymous reviewer for this suggestion.

20

received any VC in year t-1. The second column shows estimates from our main model

excluding those firms that received venture capital in year t-1. Comparing the two columns

reveals that the estimates on the board membership are almost identical. We can therefore

dismiss the concern that omitting VC has biased our results.

Table 15 exploits citation information in the PATSTAT data to address the issue that

innovative start-ups may have gained knowledge through spillovers from companies other

than the companies to which the board members are affiliated. The table estimates our

main model using a modified innovation measure. The dependent variable is set to 1 if the

firm applies for a patent without citing any other patents, and zero otherwise. The results

show that the causal impact from directors on the board connected to innovative firms on

the focal firm’s propensity to apply for patents remains, with the key estimate positive and

highly significant.

While our results presented in Table 5 imply that we cannot reject Hypothesis 2 on

the links between board members and patents, Table 9 does not support Hypothesis 3 on

similarities between patents and trademarks as proxies for knowledge transfer via board

members from innovative incumbents to innovative start-ups.

As an alternative way to test Hypothesis 3, we examine the relationship between incum-

bents with registered trademark protection and start-ups seeking patent protection. This

analysis is reported in Table 16. The table shows that the instruments are positive and

highly significant in the first-stage of the recursive pooled and CRE models. The spillover

measure reported in the second stage is positive and significant at the 1% level in both the

pooled and CRE estimations. This provides evidence that there are knowledge spillovers

not only from established patenting firms to future innovators but also from trademark

companies to start-ups seeking patent protection. We also test the relationship between

established firms with patents and trademark registration by start-ups, and find no sig-

nificant influence. Despite some mixed results from the sensitivity tests on trademarks

reported above, our main conclusion on rejection of Hypothesis 3 remains.

Table 17 presents IV linear probability model estimates to evaluate the validity of the

instruments we have constructed. Two sets of results are reported. The first columns report

first and second stage estimates for the pooled model, while columns 3 and 4 reveals the

corresponding estimates for the CRE approach. The instruments and the board variable

21

are positive and highly significant in both models. Beyond this crucial result, our main

interest is to test validity of the instruments. The Kleibergen–Paap tests of both underi-

dentification and weak instruments6 and the Hansen J test of overidentifying restrictions

provide satisfactory results.

Finally, we test the sensitivity of our results by altering our definition of start-ups,

starting with the entire population of independent firms as sample A1. Relative to our

estimation sample, this represents an increase by almost 60% from about 375,000 firm-year

observations to 590,694 firm-years. In sample A2, we exclude firms that have spun out of

an incumbent firm. This sample is 26% larger than our estimation sample. Sample A3 adds

a restriction on the number of employees during firm formation, excluding all firms with

10 or more employees when formed, and is thereby 18% larger than the estimation sample.

Sample A4 imposes the further restriction of dropping firms with only one employee over

the sample period, and is the sample used in estimation results reported above. The

characteristics of these four samples are described in Appendix 2, Table 19.

In order to assess the impact of these restrictions, we apply the recursive bivariate

probit model on all four samples and compare the results in Appendix 2, Tables 20, 21, 22

and 23. The reported marginal estimates show that the causal impact from board members

affiliated with innovative firms on the likelihood of patent application is positive and highly

significant regardless of sample definition.

.

7 Conclusions

This paper investigates the importance of spillovers from innovative firms to small newly

formed firms in the Swedish economy. The existing spillover research examines factors such

as geographical or industrial clusters, proximity to research milieus, spin-offs from parent

firms, and mobility of labor. We extend the literature by considering the board of directors

as a key linkage between innovative firms and new entrants.

Knowledge spillovers through board members may increase innovative activities of

newly formed firms, but formal tests are hampered by access to data, endogeneity and6See Baum, Schaffer & Stillman (2007) for details of these tests.

22

measurement issues. Our study addresses this problem by exploiting employer-employee

data covering the universe of newly formed Swedish limited enterprises and their employees

as well as all directors of the boards. This allows us to track founders, managers, directors

and employees of different firms over time. It also provides us with a basis for creating

instruments to identify causal relationships in our model.

We find that newly formed firms’ innovation can be enhanced by a network of innovative

firms, and that knowledge spillovers transferred by members of the board of directors affect

new firms’ propensity to be innovative. Selecting directors with experience from patenting

firms increases the probability that the new firm also will apply for patents.

Our paper contributes to the literature on the linkages between the board of directors

and innovation in new firms. For policymakers, the results of the analysis shed light on

the importance of the board of directors in helping new firms to turn their entrepreneurial

orientation into innovation. In most new firms, the board has only a symbolic role to meet

the legal requirements, while others have board members with ties to networks of innova-

tive firms. Although our estimates suggest a modest economic significance of knowledge

transfer via board members, it may be one of several factors that combine to generate new

generations of innovative firms.

Our study shows that trademark registration is the most common method for protect-

ing IP rights among Swedish start-ups. We also find that the probability of obtaining

trademark protection increases with the companies’ human capital intensity, our proxy for

the firm’s R&D intensity, as well as with the proportion of employees recruited from in-

novative companies and localization in metro regions. But the positive effect we find from

board members affiliated with innovative companies, proxied by trademark registrations,

is not statistically significant.

We acknowledge the existence of limitations in our empirical approach that offer oppor-

tunities for future extensions. The results for directors not linked to any innovative firm

are in line with the standard Schumpeterian theory on absorptive capacity, and the find-

ings on directors with innovation experience are in agreement with both entrepreneurial

theory and a small number of empirical studies. However, the difference between patent

applications and trademark applications as indicators of innovation for new firms is in

contrast to expectations derived from the recent literature. This finding deserves further

23

research regarding spillover effects from innovative firms using trademarks as a means of

intellectual property protection.

Another possible direction for future research is to consider additional outcome vari-

ables for the new firms, such as survival, acquisition, productivity and growth. The back-

ground of directors and their prior expertise is another venue for further analysis. For

instance, one could presume that directors have a greater importance on knowledge diffu-

sion between the innovative company and the new company if both firms are in the same

industry and have similar customers.

Acknowledgements

Hans Lööf and Ingrid Viklund Ros thank VINNOVA for financial support.

24

References

Acemoglu, D. & Cao, D. (2015), ‘Innovation by entrants and incumbents’, Journal ofEconomic Theory 157, 255–294.

Agarwal, R., Echambadi, R., Franco, A. M. & Sarkar, M. B. (2004), ‘Knowledge trans-fer through inheritance: Spin-out generation, development, and survival’, Academy ofManagement journal 47(4), 501–522.

Akcigit, U. & Kerr, W. R. (2018), ‘Growth through heterogeneous innovations’, Journalof Political Economy 126(4), 1374–1443.

Artz, K. W., Norman, P. M., Hatfield, D. E. & Cardinal, L. B. (2010), ‘A longitudinal studyof the impact of r&d, patents, and product innovation on firm performance’, Journal ofProduct Innovation Management 27(5), 725–740.

Arundel, A. (2001), ‘The relative effectiveness of patents and secrecy for appropriation’,Research Policy 30(4), 611–624.

Arzubiaga, U., Kotlar, J., De Massis, A., Maseda, A. & Iturralde, T. (2018), ‘En-trepreneurial orientation and innovation in family SMEs: Unveiling the (actual) impactof the Board of Directors’, Journal of Business Venturing 33(4), 455–469.

Audretsch, D. B., Bönte, W. & Mahagaonkar, P. (2012), ‘Financial signaling by innovativenascent ventures: The relevance of patents and prototypes’, Research Policy 41(8), 1407–1421.

Audretsch, D. B. & Feldman, M. P. (1996), ‘R&d spillovers and the geography of innovationand production’, The American Economic Review 86(3), 630–640.

Audretsch, D. B. & Keilbach, M. (2005), ‘The knowledge spillover theory of entrepreneur-ship and economic growth’, Research on Technological Innovation, Management andPolicy 9, 37–54.

Audretsch, D. B. & Lehmann, E. (2005), ‘Does the knowledge spillover theory of en-trepreneurship hold for regions?’, Research Policy 34(8), 1191–1202.

Audretsch, D. B. & Lehmann, E. (2006), ‘Entrepreneurial access and absorption of knowl-edge spillovers: Strategic board and managerial composition for competitive advantage’,Journal of Small Business Management 44(2), 155–166.

Audretsch, D. B., Lehmann, E. E. & Warning, S. (2005), ‘University spillovers and newfirm location’, Research Policy 34(7), 1113–1122.

Audretsch, D. B. & Stephan, P. (1999), ‘How and why does knowledge spill over in biotech-nology’, Innovation, Industry evolution, and Employment pp. 216–229.

Audretsch, D. B. & Stephan, P. E. (1996), ‘Company-scientist locational links: The caseof biotechnology’, The American Economic Review 86(3), 641–652.

Balsmeier, B., Buchwald, A. & Stiebale, J. (2014), ‘Outside directors on the board andinnovative firm performance’, Research Policy 43(10), 1800–1815.

Balsmeier, B., Fleming, L. & Manso, G. (2017), ‘Independent boards and innovation’,Journal of Financial Economics 123(3), 536–557.

25

Baum, C. F., Schaffer, M. E. & Stillman, S. (2007), ‘Enhanced routines for instrumen-tal variables/generalized method of moments estimation and testing’, Stata Journal7(4), 465–506.

Baumol, W. J. (2004), ‘Entrepreneurial enterprises, large established firms and other com-ponents of the free-market growth machine’, Small Business Economics 23(1), 9–21.

Bizjak, J., Lemmon, M. & Whitby, R. (2009), ‘Option backdating and board interlocks’,Review of Financial Studies 22(11), 4821–4847.

Blau, P. M. (1977), Inequality and heterogeneity: A primitive theory of social structure,Vol. 7, Free Press New York.

Block, J. H., De Vries, G., Schumann, J. H. & Sandner, P. (2014), ‘Trademarks and venturecapital valuation’, Journal of Business Venturing 29(4), 525–542.

Block, J. H., Fisch, C. O., Hahn, A. & Sandner, P. G. (2015), ‘Why do SMEs file trade-marks? Insights from firms in innovative industries’, Research Policy 44(10), 1915–1930.

Bloom, N., Jones, C. I., Van Reenen, J. & Webb, M. (2020), ‘Are ideas getting harder tofind?’, American Economic Review 110(4), 1104–44.

Breschi, S. & Lissoni, F. (2001), ‘Knowledge spillovers and local innovation systems: acritical survey’, Industrial and Corporate Change 10(4), 975–1005.

Breschi, S. & Malerba, F. (2001), ‘The geography of innovation and economic clustering:some introductory notes’, Industrial and Corporate Change 10(4), 817–833.

Brown, J. L. (2011), ‘The spread of aggressive corporate tax reporting: A detailed exami-nation of the corporate-owned life insurance shelter’, Accounting Review 86(1), 23–57.

Caballero, R. J. & Jaffe, A. B. (1993), ‘How high are the giants’ shoulders: An empiri-cal assessment of knowledge spillovers and creative destruction in a model of economicgrowth’, NBER macroeconomics annual 8, 15–74.

Campbell, B. A., Ganco, M., Franco, A. M. & Agarwal, R. (2012), ‘Who leaves, whereto, and why worry? employee mobility, entrepreneurship and effects on source firmperformance’, Strategic Management Journal 33(1), 65–87.

Cantner, U., Graf, H., Herrmann, J. & Kalthaus, M. (2016), ‘Inventor networks in renew-able energies: The influence of the policy mix in germany’, Research Policy 45(6), 1165–1184.

Chesbrough, H. W. (2003), Open innovation: The new imperative for creating and profitingfrom technology, Harvard Business Press.

Coad, A. (2019), ‘Persistent heterogeneity of r&d intensities within sectors: Evidence andpolicy implications’, Research Policy 48(1), 37–50.

Cohen, W. M. & Klepper, S. (1992), ‘The anatomy of industry R&D intensity distribu-tions’, American Economic Review pp. 773–799.

Cohen, W. M. & Levinthal, D. A. (1990), ‘Absorptive capacity: A new perspective onlearning and innovation’, Administrative Science Quarterly 35(1), 128–152.

26

Colombelli, A., Grilli, L., Minola, T. & Mrkajic, B. (2019), ‘To what extent do younginnovative companies take advantage of policy support to enact innovation appropriationmechanisms?’, Research Policy p. 103797.

Combs, J. G., Ketchen Jr, D. J., Perryman, A. A. & Donahue, M. S. (2007), ‘The mod-erating effect of ceo power on the board composition–firm performance relationship’,Journal of Management Studies 44(8), 1299–1323.

Crass, D. (2020), ‘Which firms use trademarks? firm-level evidence from germany on therole of distance, product quality and innovation’, Industry and Innovation pp. 1–26.

Dalziel, T., Gentry, R. J. & Bowerman, M. (2011), ‘An integrated agency–resource de-pendence view of the influence of directors’ human and relational capital on firms’ r&dspending’, Journal of Management Studies 48(6), 1217–1242.

Decker, R., Haltiwanger, J., Jarmin, R. & Miranda, J. (2014), ‘The role of entrepreneurshipin us job creation and economic dynamism’, Journal of Economic Perspectives 28(3), 3–24.

Dosi, G. & Nelson, R. R. (2010), Technical change and industrial dynamics as evolutionaryprocesses, in ‘Handbook of the Economics of Innovation’, Vol. 1, Elsevier, pp. 51–127.

Fama, E. F. & Jensen, M. C. (1983), ‘Agency problems and residual claims’, The Journalof Law and Economics 26(2), 327–349.

Feldman, M. P. (1994), The geography of innovation, Vol. 2, Springer Science & BusinessMedia.

Ferraris, A., Santoro, G. & Dezi, L. (2017), ‘How mnc’s subsidiaries may improve theirinnovative performance? the role of external sources and knowledge management capa-bilities’, Journal of Knowledge Management .

Fosfuri, A. & Tribó, J. A. (2008), ‘Exploring the antecedents of potential absorptive ca-pacity and its impact on innovation performance’, Omega 36(2), 173–187.

Fritsch, M. & Franke, G. (2004), ‘Innovation, regional knowledge spillovers and r&d coop-eration’, Research Policy 33(2), 245–255.

Ghio, N., Guerini, M., Lehmann, E. E. & Rossi-Lamastra, C. (2015), ‘The emergence of theknowledge spillover theory of entrepreneurship’, Small Business Economics 44(1), 1–18.

Gotsch, M. & Hipp, C. (2012), ‘Measurement of innovation activities in the knowledge-intensive services industry: a trademark approach’, The Service Industries Journal32(13), 2167–2184.

Griliches, Z. (1991), The search for R&D spillovers, Technical Report 3768, National Bu-reau of Economic Research.

Grossmann, V. (2009), ‘Entrepreneurial innovation and economic growth’, Journal ofMacroeconomics 31(4), 602–613.

Hall, B. H. & Lerner, J. (2010), The financing of r&d and innovation, in ‘Handbook of theEconomics of Innovation’, Vol. 1, Elsevier, pp. 609–639.

27

Hall, B., Helmers, C., Rogers, M. & Sena, V. (2014), ‘The choice between formal andinformal intellectual property: a review’, Journal of Economic Literature 52(2), 375–423.

Helmers, C., Patnam, M. & Rau, P. R. (2017), ‘Do board interlocks increase innovation?evidence from a corporate governance reform in india’, Journal of Banking & Finance80, 51–70.

Henderson, J. V. (2007), ‘Understanding knowledge spillovers’, Regional Science and UrbanEconomics 37(4), 497–508.

Hermalin, B. E. & Weisbach, M. S. (1998), ‘Endogenously chosen boards of directors andtheir monitoring of the ceo’, American Economic Review pp. 96–118.

Holgersson, M. (2013), ‘Patent management in entrepreneurial smes: a literature reviewand an empirical study of innovation appropriation, patent propensity, and motives’,R&D Management 43(1), 21–36.

Hoskisson, R. E., Hitt, M. A., Johnson, R. A. & Grossman, W. (2002), ‘Conflicting voices:The effects of institutional ownership heterogeneity and internal governance on corporateinnovation strategies’, Academy of Management journal 45(4), 697–716.

Howard, M. D., Boeker, W. & Andrus, J. L. (2019), ‘The spawning of ecosystems: Howcohort effects benefit new ventures’, Academy of Management Journal 62(4), 1163–1193.

Jensen, M. & Meckling, W. (1976), ‘Theory of the firm: Management behavior, agencycosts and capital structure’, Journal of financial economics 3(4), 305–60.

Jones, M. V., Coviello, N. & Tang, Y. K. (2011), ‘International entrepreneurship research(1989–2009): a domain ontology and thematic analysis’, Journal of Business Venturing26(6), 632–659.

Kerr, W., Akcigit, U., Bloom, N., Acemoglu, D. et al. (2012), Innovation, reallocation andgrowth, in ‘2012 Meeting Papers’, number 1137, Society for Economic Dynamics.

Klepper, S. (2001), ‘Employee startups in high-tech industries’, Industrial and CorporateChange 10(3), 639–674.

Klepper, S. (2007), ‘Disagreements, spinoffs, and the evolution of detroit as the capital ofthe us automobile industry’, Management Science 53(4), 616–631.

Klette, T. J. & Kortum, S. (2004), ‘Innovating firms and aggregate innovation’, Journal ofPolitical Economy 112(5), 986–1018.

Kodama, T. (2008), ‘The role of intermediation and absorptive capacity in facilitatinguniversity-industry linkages - an empirical study of tama in Japan’, Research Policy37(8), 1224–1240.

Kogut, B. & Zander, U. (1992), ‘Knowledge of the firm, combinative capabilities, and thereplication of technology’, Organization Science 3(3), 383–397.

Kor, Y. Y. & Sundaramurthy, C. (2009), ‘Experience-based human capital and socialcapital of outside directors’, Journal of Management 35(4), 981–1006.

Kortum, S. & Lerner, J. (2000), ‘Assessing the contribution of venture capital to innova-tion’, RAND Journal of Economics pp. 674–692.

28

Landes, W. M. & Posner, R. A. (1987), ‘Trademark law: an economic perspective’, TheJournal of Law and Economics 30(2), 265–309.

Lentz, R. & Mortensen, D. T. (2008), ‘An empirical model of growth through productinnovation’, Econometrica 76(6), 1317–1373.

Lim, K. (2009), ‘The many faces of absorptive capacity: spillovers of copper interconnecttechnology for semiconductor chips’, Industrial and Corporate Change 18(6), 1249–1284.

Lissoni, F. (2001), ‘Knowledge codification and the geography of innovation: the case ofbrescia mechanical cluster’, Research Policy 30(9), 1479–1500.

Mancusi, M. L. (2008), ‘International spillovers and absorptive capacity: A cross-countrycross-sector analysis based on patents and citations’, Journal of International Economics76(2), 155–165.

Mansfield, E. (1986), ‘Patents and innovation: an empirical study’, Management Science32(2), 173–181.

McCann, P. & Simonen, J. (2005), ‘Innovation, knowledge spillovers and local labourmarkets’, Papers in Regional Science 84(3), 465–485.

Mendonça, S., Pereira, T. S. & Godinho, M. M. (2004), ‘Trademarks as an indicator ofinnovation and industrial change’, Research Policy 33(9), 1385–1404.

Morrar, R. (2014), ‘Innovation in services: A literature review’, Technology InnovationManagement Review 4(4).

Nagaoka, S., Motohashi, K. & Goto, A. (2010), Patent statistics as an innovation indicator,in ‘Handbook of the Economics of Innovation’, Vol. 2, Elsevier, pp. 1083–1127.

Neumark, D., Wall, B. & Zhang, J. (2011), ‘Do small businesses create more jobs? newevidence for the united states from the national establishment time series’, The Reviewof Economics and Statistics 93(1), 16–29.

Nieto, M. & Quevedo, P. (2005), ‘Absorptive capacity, technological opportunity, knowl-edge spillovers, and innovative effort’, Technovation 25(10), 1141–1157.

Oh, J.-M. (2017), ‘Absorptive capacity, technology spillovers, and the cross-section of stockreturns’, Journal of Banking & Finance 85, 146–164.

Papke, L. E. & Wooldridge, J. M. (2008), ‘Panel data methods for fractional responsevariables with an application to test pass rates’, Journal of Econometrics 145(1), 121 –133.

Polanyi, M. (1966), ‘The logic of tacit inference’, Philosophy 41(155), 1–18.

Popp, D. (2019), Environmental policy and innovation: a decade of research, TechnicalReport 25631, National Bureau of Economic Research.

Qian, H. & Acs, Z. J. (2013), ‘An absorptive capacity theory of knowledge spillover en-trepreneurship’, Small Business Economics 40(2), 185–197.

Robeson, D. & O’Connor, G. C. (2013), ‘Boards of directors, innovation, and perfor-mance: An exploration at multiple levels’, Journal of Product Innovation Management30(4), 608–625.

29

Rodríguez-Pose, A. & Crescenzi, R. (2008), ‘Research and development, spillovers, innova-tion systems, and the genesis of regional growth in europe’, Regional Studies 42(1), 51–67.

Rutherford, M. A. & Buchholtz, A. K. (2007), ‘Investigating the relationship betweenboard characteristics and board information’, Corporate Governance: An InternationalReview 15(4), 576–584.

Sandner, P. (2009), The market value of r&d, patents, and trademarks, in ‘The Valuationof Intangible Assets’, Springer, pp. 35–72.

Schubert, T. & Tavassoli, S. (2020), ‘Product innovation and educational diversity in topand middle management teams’, Academy of Management Journal 63(1), 272–294.

Semykina, A. (2018), ‘Self-employment among women: Do children matter more than wepreviously thought?’, Journal of Applied Econometrics 33(3), 416–434.

Shropshire, C. (2010), ‘The role of the interlocking director and board receptivity in thediffusion of practices’, Academy of Management Review 35(2), 246–264.

Söderblom, A., Samuelsson, M., Wiklund, J. & Sandberg, R. (2015), ‘Inside the blackbox of outcome additionality: Effects of early-stage government subsidies on resourceaccumulation and new venture performance’, Research Policy 44(8), 1501–1512.

Spender, J.-C. & Grant, R. M. (1996), ‘Knowledge and the firm: Overview’, StrategicManagement Journal 17(S2), 5–9.

Srinivasan, R., Wuyts, S. & Mallapragada, G. (2018), ‘Corporate board interlocks and newproduct introductions’, Journal of Marketing 82(1), 132–148.

Tallon, P. P. (2011), ‘Value chain linkages and the spillover effects of strategic informa-tion technology alignment: A process-level view’, Journal of Management InformationSystems 28(3), 9–44.

Teece, D. J. (1988), ‘Capturing value from technological innovation: Integration, strategicpartnering, and licensing decisions’, Interfaces 18(3), 46–61.

Van Stel, A., Carree, M. & Thurik, R. (2005), ‘The effect of entrepreneurial activity onnational economic growth’, Small Business Economics 24(3), 311–321.

Verhoeven, D., Bakker, J. & Veugelers, R. (2016), ‘Measuring technological novelty withpatent-based indicators’, Research Policy 45(3), 707–723.

Veugelers, R. & Schneider, C. (2018), ‘Which ip strategies do young highly innovative firmschoose?’, Small Business Economics 50(1), 113–129.

Von Hippel, E. (1994), “ ‘sticky information” and the locus of problem solving: implicationsfor innovation’, Management Science 40(4), 429–439.

Winkler, D. (2013), Potential and actual FDI spillovers in global value chains: the roleof foreign investor characteristics, absorptive capacity and transmission channels, TheWorld Bank.

Wooldridge, J. M. (2005), ‘Simple solutions to the initial conditions problem in dynamic,nonlinear panel data models with unobserved heterogeneity’, Journal of Applied Econo-metrics 20(1), 39–54.

30

Wooldridge, J. M. (2019), ‘Correlated random effects models with unbalanced panels’,Journal of Econometrics 211(1), 137–150.

Xie, X., Zou, H. & Qi, G. (2018), ‘Knowledge absorptive capacity and innovation perfor-mance in high-tech companies: A multi-mediating analysis’, Journal of Business Re-search 88, 289–297.

Zahra, S. A. & Filatotchev, I. (2004), ‘Governance of the entrepreneurial threshold firm: aknowledge-based perspective’, Journal of Management Studies 41(5), 885–897.

Zhang, J. J., Baden-Fuller, C. & Pool, J. K. (2011), ‘Resolving the tensions betweenmonitoring, resourcing and strategizing: Structures and processes in high technologyventure boards’, Long Range Planning 44(2), 95–117.

Zhou, H., Sandner, P. G., Martinelli, S. L. & Block, J. H. (2016), ‘Patents, trademarks,and their complementarity in venture capital funding’, Technovation 47, 14–22.

31

Tables

Table 3: Summary statistics. Three years after firm formation.

Variable Mean Std. Dev. Min. Max.

Patent applicationt 0.002 0.049 0 1Patent or trademark applicationt 0.003 0.057 0 1Trademark applicationt 0.001 0.032 0 1BCIt−1 0.022 0.146 0 1BCTt−1 0.025 0.157 0 1NewOBCI1t−1 0.001 0.037 0 1NewOBCI2t−1 0.001 0.031 0 1NewOBCT1t−1 0.002 0.048 0 1NewOBCT2t−1 0.001 0.037 0 1Board sizet−1 1.569 1.052 1 19Log(total assets)t−1 14.143 1.032 6.908 21.371Human Capitalt−1 0.194 0.328 0 1Metro 0.415 0.493 0 1IE10t−1 0.062 0.242 0 1BHCt−1 0.177 0.355 0 1BEB diversityt−1 0.125 0.228 0 1

N 61740

32

Table 4: Summary statistics for full panel

Variable Mean Std. Dev. Min. Max.

Patent applicationt 0.001 0.038 0 1Patent or trademark applicationt 0.002 0.049 0 1Trademark applicationt 0.001 0.031 0 1

BCIt−1 0.016 0.125 0 1BCTt−1 0.021 0.143 0 1NewOBCI1t−1 0.001 0.032 0 1NewOBCI2t−1 0.001 0.028 0 1NewOBCT1t−1 0.002 0.04 0 1NewOBCT2t−1 0.001 0.035 0 1

Board sizet−1 1.523 1.006 1 20Log(total assets)t−1 14.355 1.095 6.908 21.4Human Capitalt−1 0.186 0.319 0 1Metro 0.401 0.49 0 1IE10t−1 0.067 0.25 0 1BHCt−1 0.166 0.348 0 1BEB diversityt−1 0.115 0.221 0 1Year2005 0.093 0.29 0 12006 0.095 0.293 0 12007 0.096 0.295 0 12008 0.097 0.296 0 12009 0.098 0.298 0 12010 0.1 0.301 0 12011 0.097 0.295 0 12012 0.091 0.288 0 12013 0.087 0.281 0 12014 0.083 0.275 0 12015 0.063 0.243 0 1Firm age3 0.165 0.371 0 14 0.146 0.354 0 15 0.13 0.336 0 16 0.114 0.318 0 17 0.101 0.302 0 18 0.087 0.283 0 19 0.073 0.26 0 110 0.061 0.239 0 111 0.051 0.22 0 112 0.042 0.201 0 113 0.029 0.168 0 1

N 374198

33

Table 5: Patent applicationt - Probit estimates

Patent applicationt

Category 1 Category 2(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii)

(i)-(xii) -0.107∗∗∗ -0.058 -0.161∗∗∗ -0.005 0.136∗∗∗ -0.161∗∗∗ 0.308∗∗ 0.014 0.473∗∗∗ 0.147 0.682∗∗∗ 0.632∗∗∗

(0.027) (0.035) (0.035) (0.023) (0.042) (0.055) (0.156) (0.088) (0.043) (0.099) (0.089) (0.106)Board sizet−1 0.354∗∗∗ 0.312∗∗∗ 0.315∗∗∗ 0.306∗∗∗ 0.296∗∗∗ 0.318∗∗∗ 0.309∗∗∗ 0.305∗∗∗ 0.302∗∗∗ 0.307∗∗∗ 0.327∗∗∗ 0.298∗∗∗

(0.034) (0.032) (0.032) (0.032) (0.033) (0.032) (0.033) (0.032) (0.036) (0.032) (0.032) (0.032)Board size 2

t−1 -0.025∗∗∗ -0.021∗∗∗ -0.018∗∗∗ -0.020∗∗∗ -0.020∗∗∗ -0.021∗∗∗ -0.021∗∗∗ -0.020∗∗∗ -0.025∗∗∗ -0.021∗∗∗ -0.025∗∗∗ -0.020∗∗∗

(0.004) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.004) (0.003) (0.004) (0.003)Log(total assets)t−1 0.206∗∗∗ 0.222∗∗∗ 0.209∗∗∗ 0.219∗∗∗ 0.215∗∗∗ 0.217∗∗∗ 0.216∗∗∗ 0.219∗∗∗ 0.199∗∗∗ 0.217∗∗∗ 0.214∗∗∗ 0.216∗∗∗

(0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021)Human Capitalt−1 0.597∗∗∗ 0.629∗∗∗ 0.632∗∗∗ 0.626∗∗∗ 0.629∗∗∗ 0.625∗∗∗ 0.622∗∗∗ 0.625∗∗∗ 0.561∗∗∗ 0.622∗∗∗ 0.591∗∗∗ 0.612∗∗∗

(0.067) (0.070) (0.071) (0.070) (0.070) (0.071) (0.070) (0.070) (0.070) (0.070) (0.070) (0.070)Metro 0.018 0.021 0.008 0.018 0.015 0.019 0.017 0.018 0.010 0.018 0.024 0.015

(0.055) (0.055) (0.056) (0.055) (0.055) (0.055) (0.055) (0.055) (0.056) (0.055) (0.055) (0.055)IE10t−1 0.474∗∗∗ 0.469∗∗∗ 0.462∗∗∗ 0.468∗∗∗ 0.472∗∗∗ 0.465∗∗∗ 0.466∗∗∗ 0.469∗∗∗ 0.440∗∗∗ 0.467∗∗∗ 0.445∗∗∗ 0.464∗∗∗

(0.046) (0.045) (0.046) (0.046) (0.045) (0.045) (0.045) (0.045) (0.047) (0.045) (0.046) (0.046)Constant -6.239∗∗∗ -6.431∗∗∗ -6.104∗∗∗ -6.354∗∗∗ -6.294∗∗∗ -6.297∗∗∗ -6.313∗∗∗ -6.351∗∗∗ -6.006∗∗∗ -6.331∗∗∗ -6.233∗∗∗ -6.242∗∗∗

(0.524) (0.520) (0.517) (0.518) (0.520) (0.516) (0.514) (0.516) (0.516) (0.516) (0.509) (0.514)

Observations 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198Notes: Year, cohort, industry and firm age fixed effects included in all specifications.Clustered standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

34

Table 6: Patent applicationt - Probit AMEa

Patent applicationt

Category 1 Category 2(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii)

(i)-(xii) -0.000∗∗∗ -0.000 -0.001∗∗∗ -0.000 0.001∗∗∗ -0.001∗∗∗ 0.001∗ 0.000 0.002∗∗∗ 0.001 0.003∗∗∗ 0.002∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000)Board sizet−1 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Log(total assets)t−1 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗ 0.001∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Human Capitalt−1 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Metro 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)IE10t−1 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Observations 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198 374198Notes: Year, cohort, industry and firm age fixed effects included in all specifications.Clustered standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.aAverage marginal effects.

35

Table 7: Patent applicationt - Recursive bivariate probit estimates

Pooled CREa

Patent applicationtBCIt−1 1.097∗∗∗ 1.100∗∗∗

(0.113) (0.115)Board sizet−1 0.238∗∗∗ 0.107

(0.037) (0.085)Board size2t−1 -0.022∗∗∗ -0.017∗∗

(0.004) (0.008)Log(total assets)t−1 0.197∗∗∗ 0.026

(0.021) (0.054)Human Capitalt−1 0.494∗∗∗ -0.127

(0.069) (0.156)Metro 0.012 0.008

(0.055) (0.056)IE10t−1 0.411∗∗∗ -0.061

(0.047) (0.065)Constant -5.804∗∗∗ -5.723∗∗∗

(0.505) (0.744)

BCIt−1NewOBCI1t−1 3.831∗∗∗ 3.821∗∗∗

(0.186) (0.185)NewOBCI2t−1 2.362∗∗∗ 2.344∗∗∗

(0.101) (0.101)Board sizet−1 0.494∗∗∗ 0.484∗∗∗

(0.018) (0.029)Board size2t−1 -0.022∗∗∗ -0.025∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.080∗∗∗ 0.017

(0.011) (0.023)Human Capitalt−1 0.503∗∗∗ 0.070

(0.036) (0.069)Metro 0.087∗∗∗ 0.082∗∗∗

(0.025) (0.025)IE10t−1 0.298∗∗∗ -0.227∗∗∗

(0.034) (0.045)Constant -4.293∗∗∗ -4.299∗∗∗

(0.280) (0.356)

ρ -0.215 ∗∗∗ -0.240∗∗∗

(0.055 ) (0.058)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Standard errors inparentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.a

Time averages of all time varying control variablesincluded in both equations.

36

Table 8: Patent applicationt - Recursive bivariate probit AMEa

Pooled CREb

BCIt−1 0.0046∗∗∗ 0.0046∗∗∗

(0.0007) (0.0007)Board sizet−1 0.0004∗∗∗ 0.0000

(0.0001) (0.0002)Log(total assets)t−10.0008∗∗∗ 0.0001

(0.0001) (0.0002)Human Capitalt−1 0.0021∗∗∗ -0.0005

(0.0003) (0.0006)Metro 0.0001 0.0000

(0.0002) (0.0002)IE10t−1 0.0017∗∗∗ -0.0003

(0.0002) (0.0003)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Standard errorsin parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.aAverage marginal effects. b Time averages of all timevarying control variables included in both equations.

37

Table 9: Trademark applicationt - Recursive bivariate probit estimates

Pooled CREa

Trademark applicationtBCTt−1 0.119 0.077

(0.152) (0.149)Board sizet−1 0.189∗∗∗ 0.136∗

(0.031) (0.070)Board size 2

t−1 -0.010∗∗∗ -0.010∗∗

(0.003) (0.005)Log(total assets)t−1 0.252∗∗∗ 0.237∗∗∗

(0.015) (0.071)Human Capitalt−1 0.432∗∗∗ 0.099

(0.067) (0.177)Metro 0.174∗∗∗ 0.168∗∗∗

(0.044) (0.044)IE10t−1 0.219∗∗∗ 0.021

(0.052) (0.114)Constant -7.318∗∗∗ -6.445∗∗∗

(0.457) (0.563)

BCTt−1NewOBCT1t−1 4.114∗∗∗ 4.104∗∗∗

(0.180) (0.180)NewOBCT2t−1 2.310∗∗∗ 2.309∗∗∗

(0.080) (0.080)Board sizet−1 0.521∗∗∗ 0.485∗∗∗

(0.017) (0.027)Board size 2

t−1 -0.022∗∗∗ -0.022∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.056∗∗∗ -0.017

(0.010) (0.022)Human Capitalt−1 0.383∗∗∗ 0.011

(0.031) (0.059)Metro 0.154∗∗∗ 0.149∗∗∗

(0.022) (0.022)IE10t−1 0.137∗∗∗ -0.204∗∗∗

(0.032) (0.042)Constant -3.680∗∗∗ -3.437∗∗∗

(0.222) (0.282)

ρ 0.129 0.014∗

(0.084) (0.083)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Time averages of all time varying con-trol variables included in both equations.

38

Table 10: Trademark applicationt - Recursive bivariate probit AMEa

Pooled CREb

BCTt−1 0.0003 0.0002(0.0004) (0.0004)

Board sizet−1 0.0004∗∗∗ 0.0002∗

(0.0001) (0.0001)Log(total assets)t−10.0007∗∗∗ 0.0007∗∗∗

(0.0001) (0.0002)Human Capitalt−1 0.0012∗∗∗ 0.0003

(0.0002) (0.0005)Metro 0.0005∗∗∗ 0.0005∗∗∗

(0.0001) (0.0001)IE10t−1 0.0006∗∗∗ 0.0001

(0.0001) (0.0003)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Average marginal effects b Time aver-ages of all time varying control variables included inboth equations.

39

Appendix 1

Table 11: Patent applicationt - Probit estimates

(1)Patent applicationt

Category 1:(ii)t−1 -0.015

(0.040)(iii)t−1 -0.061

(0.040)(iv)t−1 0.061∗

(0.033)(v)t−1 -0.100∗

(0.061)(vi)t−1 0.192∗∗∗

(0.048)Category 2:(vii)t−1 0.356∗∗

(0.156)(viii)t−1 0.064

(0.096)(ix)t−1 0.441∗∗∗

(0.050)(x)t−1 0.150

(0.108)(xi)t−1 0.577∗∗∗

(0.093)(xii)t−1 0.582∗∗∗

(0.116)Board sizet−1 0.316∗∗∗

(0.043)Board size 2

t−1 -0.034∗∗∗

(0.005)Log(total assets)t−1 0.182∗∗∗

(0.022)Human Capitalt−1 0.523∗∗∗

(0.070)Metro 0.006

(0.056)IE10t−1 0.416∗∗∗

(0.048)Constant -5.679∗∗∗

(0.525)

Observations 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. a Time averagesof all time varying control variables included in bothequations. Clustered standard errors in parentheses,∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

40

Table 12: Patent applicationt - Recursive bivariate probit estimates - No previous patents

Pooled CREa

Patent applicationtBCIt−1 0.617∗∗∗ 0.721∗∗∗

(0.154) (0.213)Board sizet−1 0.117∗∗∗ 0.243∗

(0.036) (0.130)Board size 2

t−1 -0.010∗∗∗ -0.016(0.003) (0.011)

Log(total assets)t−1 0.091∗∗∗ 0.425∗∗∗

(0.020) (0.091)Human Capitalt−1 0.445∗∗∗ -0.132

(0.073) (0.267)Metro 0.034 0.012

(0.053) (0.059)IE10t−1 0.330∗∗∗ 0.216∗

(0.059) (0.131)Constant -3.637∗∗∗ 3.472∗∗

(0.412) (1.762)

BCIt−1NewOBCI1t−1 3.915∗∗∗ 3.903∗∗∗

(0.193) (0.193)NewOBCI2t−1 2.357∗∗∗ 2.344∗∗∗

(0.102) (0.101)Board sizet−1 0.471∗∗∗ 0.471∗∗∗

(0.019) (0.030)Board size 2

t−1 -0.021∗∗∗ -0.024∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.071∗∗∗ 0.016

(0.012) (0.024)Human Capitalt−1 0.459∗∗∗ 0.047

(0.037) (0.072)Metro 0.084∗∗∗ 0.080∗∗∗

(0.025) (0.026)IE10t−1 0.252∗∗∗ -0.242∗∗∗

(0.036) (0.049)Constant -4.050∗∗∗ -4.063∗∗∗

(0.278) (0.354)

ρ -0.077 -0.129(0.072) (0.103)

Observations 372331 372331Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Time averages of all time varying con-trol variables included in both equations.

41

Table 13: Patent applicationt - Recursive bivariate probit estimates

Pooled CREa

Patent applicationt

BCIt−1 1.104∗∗∗ 1.098∗∗∗

(0.112) (0.113)Board sizet−1 0.251∗∗∗ 0.112

(0.043) (0.097)Board size2t−1 -0.022∗∗∗ -0.017∗∗

(0.004) (0.008)Log(total assets)t−1 0.188∗∗∗ 0.025

(0.021) (0.054)BHCt−1 0.361∗∗∗ 0.218

(0.063) (0.222)BEB diversityt−1 0.005 -0.029

(0.115) (0.235)Metro 0.023 0.026

(0.055) (0.055)IE10t−1 0.419∗∗∗ -0.062

(0.047) (0.064)Constant -5.697∗∗∗ -5.565∗∗∗

(0.508) (0.750)

BCIt−1NewOBCI1t−1 3.762∗∗∗ 3.758∗∗∗

(0.194) (0.194)NewOBCI2t−1 2.271∗∗∗ 2.263∗∗∗

(0.101) (0.101)Board sizet−1 0.399∗∗∗ 0.418∗∗∗

(0.022) (0.035)Board size2t−1 -0.016∗∗∗ -0.021∗∗∗

(0.002) (0.003)Log(total assets)t−1 0.070∗∗∗ 0.015

(0.011) (0.023)BHCt−1 0.753∗∗∗ 0.864∗∗∗

(0.035) (0.123)BEBdiversityt−1 0.526∗∗∗ 0.384∗∗∗

(0.062) (0.113)Metro 0.054∗∗ 0.055∗∗

(0.025) (0.025)IE10t−1 0.297∗∗∗ -0.236∗∗∗

(0.034) (0.045)Constant -4.257∗∗∗ -4.230∗∗∗

(0.287) (0.362)

ρ -0.225∗∗∗ -0.240∗∗∗

(0.055) (0.006)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Time averages of all time varying con-trol variables included in both equations.42

Table 14: Patent applicationt-Recursive bivariate probit estimates

No VC int−1CREa CREa

Patent applicationtBCIt−1 1.046∗∗∗ 1.033∗∗∗

(0.117) (0.119)VCt−1 0.552∗∗∗

(0.155)Board sizet−1 0.115 0.113

(0.086) (0.086)Board size 2

t−1 -0.018∗∗ -0.016∗∗

(0.008) (0.008)Log(total assets)t−1 0.022 0.023

(0.054) (0.055)Human Capitalt−1 -0.131 -0.129

(0.157) (0.157)Metro 0.008 -0.085

(0.056) (0.232)IE10t−1 -0.051 -0.032

(0.063) (0.064)Constant -5.717∗∗∗ -5.701∗∗∗

(0.755) (0.752)

BCIt−1NewOBCI1t−1 3.824∗∗∗ 3.825∗∗∗

(0.185) (0.185)NewOBCI2t−1 2.349∗∗∗ 2.349∗∗∗

(0.100) (0.100)VCt−1 1.427∗∗∗

(0.177)Board sizet−1 0.484∗∗∗ 0.482∗∗∗

(0.029) (0.029)Board size 2

t−1 -0.025∗∗∗ -0.025∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.016 0.015

(0.023) (0.023)Human Capitalt−1 0.069 0.073

(0.070) (0.070)Metro 0.082∗∗∗ -0.043

(0.025) (0.098)IE10t−1 -0.225∗∗∗ -0.231∗∗∗

(0.046) (0.046)Constant -4.284∗∗∗ -4.276∗∗∗

(0.356) (0.355)

ρ -0.220∗∗∗ -0.213∗∗∗

(0.059) (0.059)

Observations 374198 374107Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Time averages of all time varying con-trol variables included in both equations.

43

Table 15: Patent applicationt - Recursive bivariate probit estimates

CREa

Patent applicationt without citation to other patentst

BCIt−1 0.745∗∗∗

(0.153)Board sizet−1 0.097

(0.098)Board size 2

t−1 -0.004(0.007)

Log(total assets)t−1 0.098(0.067)

Human Capitalt−1 -0.231(0.196)

Metro 0.021(0.061)

IE10t−1 -0.176(0.113)

Constant -5.577∗∗∗

(0.763)

BCIt−1NewOBCI1t−1 3.831∗∗∗

(0.185)NewOBCI2t−1 2.349∗∗∗

(0.101)Board sizet−1 0.484∗∗∗

(0.029)Board size 2

t−1 -0.025∗∗∗

(0.002)Log(total assets)t−1 0.017

(0.023)Human Capitalt−1 0.072

(0.069)Metro 0.082∗∗∗

(0.025)IE10t−1 -0.227∗∗∗

(0.045)Constant -4.300∗∗∗

(0.356)

ρ -0.151∗

(0.084)

Observations 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Standard errors inparentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.a

Time averages of all time varying control variablesincluded in both equations.

44

Table 16: Patent applicationt - Recursive bivariate probit estimates

Pooled CREa

Patent applicationtBCTt−1 0.798∗∗∗ 0.789∗∗∗

(0.116) (0.118)Board sizet−1 0.264∗∗∗ 0.154∗

(0.036) (0.084)Board size 2

t−1 -0.022∗∗∗ -0.020∗∗∗

(0.004) (0.008)Log(total assets)t−1 0.204∗∗∗ 0.024

(0.021) (0.054)Human Capitalt−1 0.561∗∗∗ -0.106

(0.070) (0.153)Metro 0.006 -0.000

(0.055) (0.056)IE10 t−1 0.449∗∗∗ -0.075

(0.046) (0.063)Constant -6.047∗∗∗ -5.998∗∗∗

(0.514) (0.742)

BCTt−1NewOBCT1t−1 4.108∗∗∗ 4.097∗∗∗

(0.180) (0.180)NewOBCT2t−1 2.307∗∗∗ 2.305∗∗∗

(0.080) (0.080)Board sizet−1 0.521∗∗∗ 0.486∗∗∗

(0.017) (0.027)Board size 2

t−1 -0.022∗∗∗ -0.022∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.056∗∗∗ -0.018

(0.010) (0.022)Human Capitalt−1 0.384∗∗∗ 0.011

(0.031) (0.059)Metro 0.154∗∗∗ 0.149∗∗∗

(0.022) (0.022)IE1m0 t−1 0.139∗∗∗ -0.206∗∗∗

(0.032) (0.042)Constant -3.677∗∗∗ -3.434∗∗∗

(0.223) (0.282)

ρ -0.178∗∗ -0.197∗∗

(0.057) (0.059)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Standard errors inparentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.a

Time averages of all time varying control variablesincluded in both equations.

45

46

Table 17: Patent applicationt - IV Linear Probability Estimates

Pooled CREa

BCIt−1 Patent applicationt BCIt−1 Patent applicationt

NewOBCI1t−1 0.853∗∗∗ 0.848∗∗∗

(0.012) (0.013)NewOBCI2t−1 0.649∗∗∗ 0.646∗∗∗

(0.025) (0.025)Board sizet−1 0.012∗∗∗ 0.001∗∗∗ 0.017∗∗∗ 0.001

(0.002) (0.000) (0.002) (0.001)Board size2t−1 0.003∗∗∗ -0.000 0.000 -0.000∗∗

(0.000) (0.000) (0.000) (0.000)Log(total assets)t−1 0.003∗∗∗ 0.001∗∗∗ 0.000 -0.000

(0.000) (0.000) (0.001) (0.000)Human Capitalt−1 0.025∗∗∗ 0.003∗∗∗ 0.002 -0.001

(0.002) (0.001) (0.002) (0.001)Metro 0.003∗∗∗ 0.000 0.003∗∗∗ 0.000

(0.001) (0.000) (0.001) (0.000)IE10t−1 0.018∗∗∗ 0.005∗∗∗ -0.010∗∗∗ -0.001

(0.002) (0.001) (0.002) (0.001)BCIt−1 0.039∗∗∗ 0.038∗∗∗

(0.012) (0.011)Constant -0.060∗∗∗ -0.012∗∗∗ -0.061∗∗∗ -0.010∗∗

(0.011) (0.004) (0.014) (0.004)(0.004)

Observations 374198 374198 374198 374198

Kleibergen-Paap rk LM statistic 346.891 346.537χ2 p-value 0.0000 0.0000Kleibergen-Paap rk Wald F statistic 2394.974 2185.480Hansen J statistic 0.343 0.331χ2 p-value 0.558 0.565Notes: Year, cohort, industry and firm age fixed effects included in all specifications.Clustered standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.a

Time averages of all time varying control variables included in both equations.

47

Table 18: Trademark applicationt - Recursive bivariate probit estimates

Pooled CREa

Trademark applicationtBCIt−1 0.149 0.131

(0.193) (0.202)Board sizet−1 0.170∗∗∗ 0.032

(0.029) (0.063)Board size 2

t−1 -0.008∗∗∗ -0.001(0.002) (0.004)

Log(total assets)t−1 0.248∗∗∗ 0.314∗∗∗

(0.015) (0.064)Human Capitalt−1 0.399∗∗∗ -0.064

(0.061) (0.163)Metro 0.205∗∗∗ 0.196∗∗∗

(0.042) (0.042)IE10 t−1 0.225∗∗∗ -0.043

(0.053) (0.114)Constant -7.054∗∗∗ -6.445∗∗∗

(0.437) (0.625)

BCIt−1NewOBCI1t−1 3.841∗∗∗ 3.832∗∗∗

(0.186) (0.185)NewOBCI2t−1 2.366∗∗∗ 2.352∗∗∗

(0.101) (0.100)Board sizet−1 0.493∗∗∗ 0.484∗∗∗

(0.018) (0.029)Board size 2

t−1 -0.022∗∗∗ -0.025∗∗∗

(0.002) (0.002)Log(total assets)t−1 0.080∗∗∗ 0.018

(0.011) (0.023)Human Capitalt−1 0.502∗∗∗ 0.073

(0.036) (0.069)Metro 0.087∗∗∗ 0.082∗∗∗

(0.025) (0.025)IE10 t−1 0.298∗∗∗ -0.225∗∗∗

(0.034) (0.045)Constant -4.295∗∗∗ -4.300∗∗∗

(0.281) (0.356)

ρ 0.149 0.146(0.114) (0.121)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Standard errors inparentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.a

Time averages of all time varying control variablesincluded in both equations.

48

Appendix 2

Table 19: Samples

A: All new firms formed in year t

A1 all independent firmsA2 independent firms, no spin-outsA3 independent firms, no spin-outs, < 10 employees at startA4 independent firms, no spin-outs, < 10 employees at start,

and more than one employee throughout the sample period

Table 20: Sample A1 - Patent applicationt - Recursive bivariate probit AMEa

Pooled CREb

BCIt−1 0.0043∗∗∗ 0.0043∗∗∗

(0.0006) (0.0006)Board sizet−1 0.0004∗∗∗ 0.0000

(0.0000) (0.0001)Log(total assets)t−1 0.0008∗∗∗ 0.0001

(0.0001) (0.0002)Human Capitalt−1 0.0022∗∗∗ -0.0009

(0.0003) (0.0006)Metro -0.0001 -0.0001

(0.0002) (0.0002)IE10t−1 0.0019∗∗∗ -0.0002

(0.0002) (0.0002)

Observations 590694 590694Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Average marginal effects. b Time aver-ages of all time varying control variables included inboth equations.

49

Table 21: Sample A2 - Patent applicationt - Recursive ivariate probit AMEa

Pooled CREb

BCIt−1 0.0045∗∗∗ 0.0045∗∗∗

(0.0007) (0.0007)Board sizet−1 0.0004∗∗∗ -0.0000

(0.0001) (0.0002)Log(total assets)t−1 0.0008∗∗∗ 0.0000

(0.0001) (0.0002)Human Capitalt−1 0.0019∗∗∗ -0.0006

(0.0003) (0.0007)Metro -0.0000 -0.0000

(0.0002) (0.0002)IE10t−1 0.0019∗∗∗ -0.0003

(0.0002) (0.0003)

Observations 484095 484095Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Average marginal effects. b Time aver-ages of all time varying control variables included inboth equations.

Table 22: Sample A3 - Patent applicationt - Recursive ivariate probit AMEa

Pooled CREb

BCIt−1 0.0043∗∗∗ 0.0042∗∗∗

(0.0007) (0.0007)Board sizet−1 0.0004∗∗∗ -0.0000

(0.0001) (0.0002)Log(total assets)t−1 0.0007∗∗∗ 0.0000

(0.0001) (0.0002)Human Capitalt−1 0.0016∗∗∗ -0.0005

(0.0002) (0.0006)Metro -0.0000 -0.0000

(0.0002) (0.0002)IE10t−1 0.0017∗∗∗ -0.0002

(0.0002) (0.0003)

Observations 442162 442162Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Average marginal effects. b Time aver-ages of all time varying control variables included inboth equations.

50

Table 23: Sample A4 - Patent applicationt - Recursive ivariate probit AMEa

Pooled CREb

BCIt−1 0.0046∗∗∗ 0.0046∗∗∗

(0.0007) (0.0007)Board sizet−1 0.0004∗∗∗ 0.0000

(0.0001) (0.0002)Log(total assets)t−1 0.0008∗∗∗ 0.0001

(0.0001) (0.0002)Human Capitalt−1 0.0021∗∗∗ -0.0005

(0.0003) (0.0006)Metro 0.0001 0.0000

(0.0002) (0.0002)IE10t−1 0.0017∗∗∗ -0.0003

(0.0002) (0.0003)

Observations 374198 374198Notes: Year, cohort, industry and firm age fixed ef-fects included in all specifications. Clustered stan-dard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01.a Average marginal effects. b Time aver-ages of all time varying control variables included inboth equations.

51