gains from others’ losses: technology trajectories and the ......economic viability of, and thus...

of 44

Gains from Others’ Losses: Technology Trajectories and the Global Division of Firms

Chia-Hsuan Yang, Rebecca Nugent, Erica R.H. Fuchs

Carnegie Mellon University

This Draft: April 2, 2012

Abstract

After the burst of the telecommunications bubble in March 2000, the majority of U.S.

optoelectronic component firms moved manufacturing offshore. This research explores (1)

whether due to different offshore production economics, firms who move manufacturing

offshore stop or slow U.S.-based R&D activities in the emerging technology necessary to access

larger markets and (2) whether the inventors originally within these offshoring firms, leave, and

continue to innovate in the emerging technology at different institutions. We focus on the 28

leading small- or medium-sized U.S. firms that manufacture optoelectronic components for

telecommunications (18 offshore, 10 not) and the inventors who patent at these firms. We

triangulate hand-classified USPTO patents, firm SEC filings, inventor CVs, and structured

interview data we collect from the firms. Our results show that there is a relationship between

firms’ type and extent of offshoring facilities (fabrication, assembly) and their innovation

directions. In particular, we find that while offshoring is associated with a statistically significant

decrease in innovation in the emerging technology, it can be associated with an increase in all

other types of patenting. While an important minority of inventors of the emerging technology

who worked at offshoring firms depart to a single onshore firm in the same industry (which

subsequently dominates this space), the majority of inventors depart to firms outside our study

scope and stop work in the emerging technology.

****Draft! Please do not cite or circulate without permission of the authors****

of 44

1. Research Background

1.1. Offshoring and Innovation

Offshoring1, the relocation of firms’ business processes to other countries, has become

mainstream in many industries (Bardhan and Kroll 2003; Lewin and Couto 2007). When

offshoring, firms often are pursuing advantages, such as low labor costs (Bardhan and Jaffee

2005), skilled talent access (Baily and Farrell 2004) , unique resources availability (Oviatt and

McDougall 2004), and quick speed to market (Brown and Hagel 2005). Beyond firm advantages

or disadvantages, a growing public debate exists today on the economic impact of offshoring on

the firms’ home nation, especially on innovation and technology change (Pisano and Shih 2009;

Tassey 2010; PCAST 2011; WhiteHouse 2011; Rotman 2012). The renewed focus on the

relationships between offshoring and technology change in the home nation has relevance:

indeed, while the estimate of technology’s contribution has evolved over time, in his nobel-prize

winning article Solow argued that technology change can contribute to over 87% of increases in

productivity and thus economic growth (Solow 1957).

Mainstream economic theory suggests that offshoring, by leveraging the well-known benefits

of comparative advantage and gains from trade, can benefit both firms and their home country.

Related arguments have been made for firms: cost reductions associated with offshoring can

enable firms to earn more and thereby invest more in new technology development within the

domestic economy (Agrawal and Farrell 2003; Baily and Farrell 2004). On the other hand, others

have shown that the separation of processes can be harmful to innovation development and the

national economy. Geographic distance can impede knowledge flows in R&D (Allen 1984) and

reduce opportunities for employees to be physically present, which can in certain contexts be 1 Generally, offshoring is used to refer to the relocation of any productive activities – e.g. services, technological development or production – from a firm’s home country to one or more other nations. For the specific case studied in this paper, we subsequently define offshoring as moving manufacturing processes from the U.S. to low labor cost areas, and in particular developing countries, by firms.

of 44

critical for problem-solving (Tyre and Von Hippel 1997). Other dimensions of distance,

including cultural and administrative distance, can also affect knowledge flows (Ghemawat

2001). Recent studies have also demonstrated that in certain cases when manufacturing goes

offshore, R&D and engineering may also go offshore, following manufacturing, and thus

threaten technological leadership and economic growth in the home country (Fifarek et al. 2008;

Dewey and LeBoeuf 2009). In contrast to these past studies, which have focused on the

challenges of geographic separation and the potential for R&D work to follow, recent work by

Fuchs and others has suggested a direct and very different link between the offshoring of

manufacturing and technology development. In particular, they demonstrate in two industries

that the economic differences between developed and developing locations can reduce the

economic viability of, and thus firms’ incentives to develop, emerging technologies (Fuchs and

Kirchain 2010; Fuchs et al. 2011).

Despite this wealth of past research, few papers to-date have focused on the effect of

offshoring on technology directions – and in particular how that effect may differ at the firm

level versus the industry or national one. Previous research shows that innovation direction can

be influenced by technologists themselves (Dosi, 1982), market demand (Dosi 1982), or process

management, like ISO 9001 (Benner and Tushman 2002). Building on Fuchs and Kirchain

(2010), this paper leverages a case study of the optoelectronics industry to explore the

relationship between offshoring and innovation directions, and in particular the directions of the

offshoring firms versus the directions of individuals that may take that technology into other

non-offshoring firms within or outside the industry.

of 44

1.2. Industry Environment

1.2.1. The U.S., Optoelectronics, and Monolithic Integration

Optoelectronics, a technology combining optical with electronic technologies, is expected

to gradually replace pure electronic technologies in information technology (OIDA 2005;

Schabel 2005). Compared to electronics, optoelectronics enables components to have lower

power consumption, have higher data carrying capacity and have no cross-talk problems by

using photons instead of electronics to transmit information (Holden 2003; OIDA 2005; Shah

2007). As such, optoelectronics technology is become increasingly crucial to meeting market

requirements for higher bandwidth and reliable transmission in a wide range of applications.

In addition to its technical significance, optoelectronics also has the potential to have

significant economic impact. Optoelectronics, as a general purpose technology, has the potential

to be used not only in telecommunications but also widely in biomedical, military, computing

and energy applications (OIDA 2006; Shah 2007). Past research has suggested that this kind of

general purpose technology can drive technical progress and boost economic growth (Bresnahan

and Trajtenberg 1995). Finally, high tech manufacturing industries such as optoelectronics, also

support a disproportionate number of jobs and R&D in the U.S (compared to non-manufacturing

and lower tech industries) (PCAST 2011).

The optoelectronics component manufacturing industry has over the last three decades

faced many challenges. In the 1980s and 1990s, as optoelectronics was revolutionalizing

telecommunications, a firm’s competitiveness was a function of bringing the latest innovation to

market. With the burst of the telecommunications bubble in March 2000, however, firm survival

became a function of costs. With this shift in focus, many optoelectronic component

manufacturers were faced with a dilemma: move manufacturing overseas to developing countries

to reduce costs, or pursue an emerging technology, called monolithic integration, back in the U.S.

of 44

that may enable them to have greater comparative advantage in the longer term in

telecommunications and to move into larger markets beyond telecommunications such as

computing, mobile, and sensing applications. (Fuchs and Kirchain 2010)

Our study focuses on U.S. optoelectronic component manufacturers’ offshoring and

associated innovation in this emerging technology called monolithic integration (discussed in

detail below in section 1.2.2). Monolithic integration is the fabrication of multiple functions on a

single chip using semiconductor processing techniques. While this fabrication of multiple

devices on a single chip was mastered several decades ago in electronic semiconductors for the

case of transistors, it is in its infancy in the photonic semiconducting materials required for

optoelectronic devices. Unlike electronics – where monolithic integration is of devices with

similar material compositions, in optoelectronics, monolithic integration requires the fabrication

of devices with often vastly different material compositions on a single chip. This type of

fabrication poses significant scientific challenges. The potential benefits of monolithic

integration, however, are equally significant. Monolithic integration reduces the size of the

device – thus enabling the many advantages of optoelectroincs to be applicable in new, and much

larger markets beyond telecommunications – including computing, sensing, and mobile

technologies – where device size can be central. In many ways, monolithic integration might be

thought of as a general purpose technology enabler – providing optoelectronics the potential to

expand into this broad array of new markets. These size advantages can also be relevant in

telecommunications, however, do not take precedent when the primary focus is on costs. Finally,

monolithic integration may also have advantages beyond size for reliability, light-weighting, and

in some cases cost (Mickelson et al. 1997; Fuchs and Kirchain 2010).

To help provide broader context for our study, we plotted firm innovation activities in

optoelectronics integration in different nations based on USPTO patents to understand the role of

of 44

U.S. firms in integration internationally. U.S. firms, followed by Japanese firms2, produced

almost half of the patents in integration. (This integration patent count includes both monolithic

and hybrid integration patents – see section 1.2.2 for discussion.) In addition to the U.S. as a

nation being the top annual and cumulative patentor of integrated (monolithic and hybrid)

technologies, U.S. firms are also the vanguards in integrated technology (monolithic and hybrid)

innovation – having started patenting in optoelectroincs integration in 1973, five years prior to

any other nation3. Thus understanding the technology directon of the U.S. firms is also an

important part of understanding trends in optoelectronic innovation internationally.

1.2.2. Integrated vs. Discrete (Non-integrated) Designs

There are three competing designs today in optoelectronic device technology: integrated

designs, which are split into monolithically integrated and hybrid integrated designs, and discrete

designs. Integrated designs combine different components onto one chip; discrete technology

keeps components on separate chips and connects the chips using assembly techniques such as

wirebonding instead of fabricating them on a single integrated circuit. Integrated designs

(monolithic and hybrid) are generally more compact than discrete designs (Mickelson et al.

1997).

To achieve integration, there are two techniques: monolithic and hybrid. Monolithic

techniques fabricate all devices on a single substrate via fabrication processes including

deposition, growth, and etching. In contrast, hybrid techniques, intead of using fabrication, bond

different devices together using techniques such as flip-chip or bump bonding. Of the two, the

monolithic design both requires the most cutting-edge processing and materials technology and

holds the most promise for significant long-term benefit as it significantly reduces component

2 Japanese firms hold just over 30% of the cumulative patents in integrated designs (monolithic and hybrid) for the period of our study. 3 U.S. firms started patenting in integration in 1973, followed by Germany (1978), Italy (1978) and Canada (1979).

of 44

size, thus enabling optoelectronics components be implemented in more market appliations. In

Fuchs 2010, the emerging “integrated” design refers to monolithic integration technology. Fuchs

2010 does not address hybrid technologies. Monolithic technology, which uses fabrication to

integrate devices, is related to frontend (fabrication) manufacturing; hybrid and discrete (non-

integrated) technologies, which use assembly techniques to combine devices, are related to

backend (assembly) manufacturing.

The USPTO patent class for integration (385/14) includes both hybrid and monolithically

integrated patents. In this paper, we hand-classify all integrated patents into monolithic or hybrid

so we can treat them as separate response variables in our regression models and case studies.

2. Research Hypotheses

This research explores how the movement of U.S.-headquartered optoelectronic component

firms’ manufacturing offshore to developing countries is associated with the quantity4, locus5,

and direction6 of innovation within the population of U.S. optoelectronic component

manufacturers. In addition, to unpack the potential relationship between offshoring and activities

beyond the immediate offshoring firms, we also explore the quantity, locus, and direction of

innovation among inventors within the optoelectronic component manufacturers during the same

time period, many of whom were pursuing integrated (monolithic or hybrid) activities.

2.1. Hypotheses – Offshoring and Firms’ Innovation

As discussed earlier, the movement of manufacturing offshore can provide firms with

location-based advantages, such as tax reductions, increased access to and understanding of

4 Here, quantity refers to the total number of patents as well as the rate of patenting. 5 Here, locus refers to an inventors’ location: at an onshore firm in the industry, at an offshore firm in the industry, at a new firm outside the industry, or at an institution other than a firm, such as a university or government lab. 6 Here, direction refers to optoelectronic component innovation in monolithic, hybrid or non‐integrated technologies.

of 44

market preferences (Brown and Hagel 2005), and cheaper access to natural or human resources

(Oviatt and McDougall 2004). These advantages can help firms reduce their manufacturing

costs and free up resources to put into higher value-added activities back in the home nation, like

innovation (Farrell 2003, 2005).

In the case of the optoelectronic industry, manufacturing overseas in developing countries

reduces firms’ overall production costs (see Figure 1). As can also be seen in Figure 1, while the

emerging, monolithically integrated optoelectronic component technology is cheaper to produce

if manufacturing occurs in the U.S.; in contrast, the prevailing, discrete technology is cheaper to

produce if manufacturing is in developing East Asia, due to production differences including

lower labor wage, material and assembly costs overseas. Most importantly, the cost of

producing the discrete optoelectronic technology in Developing East Asia is lower than the cost

of producing the monolithic optoelectronic technology onshore back in the U.S. Thus,

production of the discrete device in developing East Asia is the global optimum from the

perspective of minimizing production costs. These economics alone, however, may not be

enough to change the trajectory of firms. A number of additional factors reduce optoelectronic

firms’ incentives and capability to develop the emerging, monolithically integrated, technology

(Fuchs and Kirchain 2010): First, production of the discrete device in developing East Asia is the

cheapest alternative globally, and firms are likely to focus first on cost-competing in the

telecommunications market with their existing customers rather than pursuing uncertain

opportunities in new markets that may require the size advantages of monolithic integration.

Second, fabrication engineers need to be regularly on the production line during the production

of high-end monolithically integrated optoelectronic devices, and there is currently a lack of

R&D engineers with monolithic integration capabilities in developing East Asia. Third, the

optoelectronic component market is currently too small to support individual firms having

of 44

multiple manufacturing facilities. Given the above context, we hypothesize that after firms go

offshore, they will shift away from the emerging monolithically integrated design but have more

financial and human resources to put towards developing prevailing, assembly based

technologies (hybrid and non-integrated).

Hypothesis 1: Offshoring is associated with a decline in firms’ development of

monolithically integrated designs, but an increase in firms’ development of hybrid integrated

and non-integrated designs.

Figure 1 Offshoring can reduce or remove firms’ economic incentives for developing emerging technology (monolithic integration). (Fuchs and Kirchain 2010)

2.2. Hypotheses – Offshoring and Inventors’ Innovation

Although our patenting data shows firms (compared to, for example, universities or

government laboratories) to be the institutions making the primary contribution to U.S. patenting

activities in optoelectronics, firm incentives alone are not sufficient to explain the quantity and

direction of technology innovation. Mobility of inventors has been shown to be a driver of

regional innovation (Miguélez and Moreno Serrano 2010), and this inter-firm mobility can affect

Prevailing Discrete

Technology U.S.

Emerging Monolithic

Technology U.S.

Prevailing Discrete

Technology D.E.A.

of 44

both the locus and transfer of knowledge (Almeida and Kogut 1999). Therefore, in addition to a

firm-level analysis, we also conduct an analysis of innovation activities at an individual level to

understand whether offshoring by individual firms may be associated with changes in the locus

and direction of technology development in the U.S. more broadly. In particular, individuals may

select different innovation directions than firms by leaving the offshoring firms and working on

the emerging monolithic technologies by themselves or at other firms or institutions. Indeed,

past research has suggested that once an inventor has been in a particular technical area for more

than three years, he or she is likely to continue innovation activities in the same direction despite

institutional changes, or other outside forces (Garud and Rappa 1995; Furman et al. 2010).

Hypothesis 2a: Inventors from offshoring with experience in monolithic integration firms

tend to leave offshoring firms and migrate to non-offshoring firms and non-focus firms.

Hypothesis 2b: These monolithic integration inventors continue innovation activities in

monolithic integration after leaving.

3. Methods

3.1. Data Sources

To explore the hypotheses in the preceding section, we triangulate data from industry

archives, buyers guides, SEC filings, phone-based structured interviews, inventor resumes, and

hand-classified USPTO patent data to quantify firms’ and individuals’ innovation activities,

firms’ extent of offshoring and other control variables at the firm-level and individual-level.

3.1.1. Innovation Activities

To quantify firms’ and individuals’ innovation quantity and direction, we use patent data

from the USPTO database. While not without their limitations, patents have been shown to be a

reasonable indicator of innovation activities (Hall et al. 2001). In past research studies, patents

have been used to evaluate the innovation performance of firms (Fifarek et al. 2008) and

of 44

individuals (Fleming et al. 2007), and to track the mobility of the patent holder (Almeida and

Kogut 1999).

The USPTO patent database has a thorough U.S. patent classification system based on

technical specialties – each patent is classified into one or more technical classes and subclasses

related to its invention (Hall et al. 2001; 2010; USPTO 2010). Based on this classification system

and its detailed definitions for each class as well as the contextual optoelectronics expertise of

one of our authors, we hand-identified patent classes and subclasses within the field of

optoelectronics as well as which of those patents qualify as being in integration (hybrid and

monolithic). We identified in total 11 classes associated with optoelectronics innovation7 and one

subclass 385/14, titled “Integrated Optical Circuit”, that covers integrated optoelectronics

technologies8. According to our classification system, there are 196,439 optoelectronic patents

and 3,326 integrated patents (monolithic and hybrid) in total in the USPTO database up through

December 2010. Our focus firms (those identified as being within our research scope (see 3.1.5))

have 4737 optoelectronic patents and 340 integrated patents. We manually classified the 340

integrated patents into monolithic and hybrid designs leveraging the expertise of one of our

authors in collaboration with a hired research assistant with a bachelors degree in materials

science. Based on this hand-classification, our focus firms have 209 monolithic integrated

patents and 131 hybrid integrated patents.

While USPTO patent data can serve as a proxy for firms’ and individuals’ innovation

activities in certain technology fields, the USPTO database has many challenges, including

ambiguities in the names of assignees and inventors. These ambiguities comes from many

sources, including misspellings, inconsistencies (i.e. middle names, alias, initials, acronym, and 7 The patent classes and subclasses that we identified as optoelectronic innovations are 250/220‐339 and 551, 257/13, 21, 52‐56, 59, 79‐103, 113‐118,184‐189, 225‐234, 257‐258, 290‐294 and 431‐466, 353/, 356/, 359/, 372/, 385/, 398/, 438/24‐25 and 27, 720/ and G9B/7. 8 This subclass, 385/14, includes wide range of integrated designs covering both monolithic and hybrid designs.

of 44

abbreviations) and changes in or different submissions of names (i.e. corp. and LLC.) Many

research groups work on this issue (Hall et al. 2001; Fleming et al. 2007; Bessen 2009). The

Fuchs Lab Optoelectronics Patent Database that we use in this paper builds on the algorithm

used in Fleming (2007) to disambiguate the assignee and inventor names in the optoelectronic

patent database. Building on these initial disambiguation results, we subsequently hand-cleaned

all data associated with our focus firms firms and inventors within these firms.

Finally, it is important to note that there can be a multiple-year gap between the time

patents were filed and the time they were issued. While only 22.24% of patents come out in the

first year, 99.41% of patents have come out after five years. This incompleteness would bias our

observations of firms’ and individuals’ innovation in the final four years of our sample. In an

attempt to address the above issue, we collected information on firms’ patent applications which

are available for the 2000-2011 period. Our goal in collecting this data was to determine

feasibility of an extended time period beyond our current 2006 cut-off (the cut-off being due to

the delay between patents being filed and patents being granted). We found that the current

available application data from the USPTO database is unfortunately not suitable to extend the

length of our observation period. We discuss this data and our analysis thereof in Appendix 1.

Thus for the analyses in this paper, we exclude data post-2006.

3.1.2. Extent of Offshoring

There are several methods to quantify firms’ extent of offshoring9. As used in this paper,

offshoring is the relocation of manufacturing facilities from a home, developed country to a

developing country. We gathered public firms’ offshoring information (i.e. offshoring dates,

applications, and manufacturing sites) from Securities and Exchange Commission (SEC) filings.

9 For example, the most appropriate depiction of a firm’s extent of offshoring might include the initial decision time of moving offshore, the first offshoring date, the length of time offshore, or the percentage of facilities offshore.

of 44

We also conducted structured interviews with private and public firms to collect private firms’

offshoring data and public firms’ offshoring data missing from their SEC filings. To gather this

missing data, we created a structured interview questionnaire that we sent in advance to the firms,

and then completed with ta representative from each company by phone. As a consequence, in

this paper’s analyses, we are able to distinguish between the number of years that only assembly

activities and the number of years that both assembly and fabrication activities were overseas in

a developing country when characterizing firms’ offshoring.

3.1.3. Other Variables

To understand the individual firm environment within which the firms are operating, we

also collect annual firm revenues, profits, R&D expenditures, employee numbers, and merger

and acquisition data from SEC filings and the previously-mentioned structured interviews with

representatives from each firm. In addition, to represent the overall industry environment, we

gained access to industry-wide optoelectronic revenues from market reports and field-wide

patent productivity in optoelectronics, broadly speaking, as well as in optoelectronics integration

from the Fuchs Lab Optoelectronics Database. These data represent control variables that may

also contribute to firms’ innovation. Finally, to better track individuals’ mobility, and associated

quantity, locus and direction of innovation, we collect the professional resumes of our strong

integration inventors10. These resumes, which complement the patent data, allow us to locate

inventors accurately between patents and even if they stop patenting or migrate to other fields.

10 In total we were able to collect 39 of our 106 strong integration inventors’ resumes. We leveraged inventor contact information provided to us by the top three professional societies in optics to contact integration inventors’ from our database. When we were able to reach inventors by phone or email, we would ask them to send us their professional resumes directly. We also were able to reach several inventors on LinkedIn. Again, in reaching out them requested their most current resume. In cases where they didn’t respond we used the data posted on LinkedIn.

of 44

3.1.4. Scope Definition

This study focuses on small and medium sized U.S. firms that manufacture active

optoelectronic components for telecommunication applications and inventors who have patented

within these firms.

3.1.5. FirmScope

To identify all firms whose activities fell within our scope, we triangulated data from

seven sources, ranging from industry market analyses, to buyers guides, to trade association

workshop participation, to industry trade fairs, to the USPTO patent database to identify all firms

manufacturing optoelectronic components for the telecommunications industry11. While these

sources all shed insights into firms in optoelectronics, the raw sources cover different countries,

institutions and market scopes. To deal with these differences, we manually classified firms by

different countries and applications. In addition to the list of firms producing and selling

optoelectronic components for telecommunications markets that we created from these seven

sources, we also wanted to make sure not to overlooks firms that might have innovation activities

for the same industry, but not yet sufficient market presence to show up in the above sources.

After interviews with industry experts we triangulated two sources, OVUM market reports (our

only source that focused exclusively on firms aiming to make optoelectronic components for the

telecommunications industry) and the USPTO database, to identify any remaining optoelectronic

component firms for telecommunications that may not yet have sales but work on integration

(monolithic or hybrid). From this data, we were able to identify 16 firms having at least one

11 Ovum Annual Telecommunications Industry Optoelectronic Component Manufacturers Annual Revenue and Market Share Survey (2005‐2009), LightCounting Optical Transceiver and TOSA/ROSA Vendor List (2010), Reed “Optoelectronics—A Strategic Study of the Worldwide Semiconductor Optoelectronic Component Industry to 2008” (2003‐2005), Strategies Unlimited “Opto 50 ‐ A Review of the World's Leading Manufacturers of Semiconductor‐Based Optoelectronic Components 2000” (1998‐1999), The Optical Society of America (OSA) OFC Buyers’ Guides (2004‐2009), Optoelectronics Industry Development Association (OIDA) Optoelectronic Component Manufacturers for Telecommunications Workshop Attendee List (1998, 2005), United States Patent and Trademark Office (USPTO) (1976‐present)

of 44

integrated patent (monolithic or hybrid) who also reported in the OVUM market reports and

another 12 firms having three and more integrated patents (monolithic or hybrid) that do not

show up in the OVUM market reports but were focused on selling optoelectronic components to

the telecommunications market12. (See Table 1)

Table 1 Research Firm Scope13

Firms Manufacturing Anything Offshore by 2009 Firms Manufacturing Anything Onshore by 2009Name Total Int.

Patent Total Monolithic Patent

2009 Revenue

Name Total Int. Patent

Total Monolithic Patent

2009 Revenue

JDSU† 22 12 $1.29 B Infinera 70 67 $309 M

Finisar† 28 8 $541 M Lightwave * 20 13

Bookham† 25 20 Prima Luci * 6 3

Agilent 22 9 $4.48B SDL 5 5 Avago 16 2 $1.48 B LNL * 4 3

Agere† 14 9 Optronx * 4 0

Avanex† 13 11 Teraconnect *

4 0

Emcore† 11 3 $176 M Xponent * 4 3

NeoPhotonics

*† 10 10 $155 M LSI 3 2 $2.22B

TriQuint† 8 4 $654 M Axsun * 2 0

Kotura *† 7 5 $7 M

OCP† 4 2

Picolight *† 4 4 AFOP 2 0 $30 M

CyOptics *† 2 0 $63 M

Oplink† 2 1 $144 M

Opnext† 2 0 $319 M

New Focus† 1 0 * Private firms

† Offshore post‐1999 Gray: Firms with no monolithically integrated patents ( this use of grey coloring is continued in the text for the remainder of the paper)

12 We used having one integrated patent (monolithic and hybrid) as a threshold for including firms from the OVUM market report to filter out those that never worked on integration even if they have significant market share. We used having three integrated patents (monolithic and hybrid) in the USPTO database as the criteria to select highly innovative firms. We then used a combination of web searches to confirm if these firms are U.S. firms targeting selling optoelectronic components to the telecommunications market. For both the OVUM and the USPTO databases we used publically available resources on the web to confirm which firms had U.S. headquarters, and were thus appropriate to our scope. Finally, we triangulated our results against our five other data sources before finalizing our selection of the 28 firms. 13 Within these 28 focus firms, as of 2011, 12 of the firms no longer exist— four of them exited, two merged with each other, six were acquired, and one was diversified.

of 44

3.1.6. Inventor Scope

Besides firm activities, we also explore the activities of individuals with one or more

optoelectronics patents within our 28 focus firms. To observe inventors’ innovation trajectories

and how these may differ from firms, we track the inventors’ patents and their movement. We

initially divide the inventors who patent at our focus firms into a target group and two control

groups: Our target group is inventors at our focus firms with strong capabilities in integration

(more than two integrated patents), in other words our ‘focus inventors”. In addition we have

two other groups, 1) integration “dabblers”— (inventors at our focus firms who only have one or

two integrated patents, and 2) a control group of non-integrated inventors (inventors at our focus

firms who do not have integrated patents). We chose the cut-off of three integrated patents

(monolithic or hybrid) based on differences we were seeing in inventor activity in the data. We

see this cut-off as representing the difference between inventors who have strong interest or

capabilities in integration innovation (monolithic or hybrid), versus those who may have had

brief involvement in a related project. We have in our target group 106 “focus” inventors (with

more than two patents in integration). Of these focus inventors with strong capabilities in

integration, 99 have one or more monolithic patents, and seven have only hybrid patents.14 In

our two control groups we have 383 integration “dabblers” with two or fewer patents in

integration, and 2907 non-integrated inventors.

Within our integrated inventor group, we divide the integration inventors into four subgroups:

those who have monolithic patenting experience at our focus firms, those who have monolithic

patenting experience prior to being our focus firms but have no monolithic patents at our focus

14 We had already completed the classification of monolithic versus hybrid patent classes for integrated patents within our focus firms. To complete the inventor analysis, we in addition hand‐classified the monolithic versus hybrid patents granted to inventors at firms they worked in before and after working in our focus firms.

of 44

firms, those who have monolithic experience only post being in our focus firms (and no

monolithic patents prior to or at our focus firms), and those who only have hybrid patents at our

focus firms and no prior or post monolithic patenting experience. We further divided each of the

above integration subgroups into two parts: those who ever worked at offshoring and those who

never worked at offshoring firms. In the non-integrated group, we also divided the inventors into

the same two parts: those who ever worked at an offshoring firm and those who never worked at

an offshoring firm.

3.2. Regression Models

Our primary research question focuses on the relationship between patenting and going

offshore. In this section, we present three regression models to address this question. First, we

model the relationship between the firms' number of yearly patents and the extent of their

offshoring using a negative binomial regression (while controlling for firm size, R&D strategy.

the burst of the telecommunications bubble, as well as the current innovation status of the

field). This approach correctly models the distribution of our patent count data and treats the

firms' yearly patents as individual observations. We then turn to a Cox proportional hazards

regression to model the instantaneous rate of patenting over time. This model accounts for a

non-constant patenting rate over the year and allows for firm longitudinal effects over

time. Finally, we similarly use a Cox proportional hazards regression to model the risk of going

offshore each year. This model uses cumulative patent counts to represent innovation capability

and similarly controls for firm size, R&D strategy, the burst of the telecommunications bubble,

and the current innovation status of the field.) The variables and subscripts we use in the models

are described in Table 2 below. In addition to our quantitative analysis, to gain better insights

into each firm’s activities, we leverage the full descriptive data available through the SEC filings

and firm structured interviews to create additional insights into each firm and inform our

of 44

Table 2 Variables and Subscripts

Subscript Description Purpose Role

Y Year

Time

D Day y(d) Year for the day on which a firm filed a patent

A {monolithic, hybrid, non‐integrated} Type of patent (hand classified)

Paten

t

Category

B {integrated, non‐integrated} Field‐wide type of patent

Variable Description Purpose Role

Pa,y Patent count for year (y) of patent type (a) Yearly innovation

Dep

endent

Variable

hpat,a(d) Risk of patenting in day (d) of patent type (a) Innovation risk

hoff(y) Risk of moving any facilities offshore in year (y) Offshoring risk

Ca,y Cumulative patent count in year (y) of patent type (a) Existing innovation capability

Indep

endent

Variable

Ay‐1 Number of full years that firm has only assembly activities offshore (y‐1: one year lag)

Extent of offshoring

By‐1 Number of full years that firm has both assembly and fabrication activities offshore (y‐1: one year lag)

Ay(d)‐1 Number of full years that a firm has only assembly activities offshore (y(d)‐1: one year lag)

By(d)‐1 Number of full years that a firm has both assembly and fabrication activities offshore (y(d)‐1: one year lag)

h0(y) Unspecified baseline rate in Cox offshoring rate models for year (y) Baseline risk

Control

Variable

h0(d) Unspecified baseline rate in Cox patent rate models for day (d)

Qy Revenue for year (y) Firm size lnQy Natural logarithm value of Revenue for year (y)

lnQy(d) Natural logarithm value of Revenue for year (y(d)) Ry‐1 R&D spending for year (y‐1: one year lag)

Firm R&D strategy

lnRy Natural logarithm value of R&D spending for year (y) lnRy(d)‐1 Natural logarithm value of R&D spending for year (y(d))

(y(d) ‐1: one year lag) Fb,y Total patents in the U.S. technical field of optoelectronics

for year (y) of patent type (b) Trends of technical field

Fb,y(d) Total patents in the U.S. technical field of optoelectronics for year (y(d)) of patent type (b)

Trends of technical field

Ty Telecom bubble dummy variable for year (y) Telecom bubble Ty‐1 Telecom bubble dummy variable for year (y)

(y‐1: one year lag) Telecom bubble

Td‐365 Telecom bubble dummy variable for day (d) (d‐365: one year lag)

Telecom bubble

of 44

interpretation of the quantitative results. Finally, we incorporate insights from this descriptive

analysis back into our regression models in terms of fixed effects for firm strategies.

3.2.1. The Relationship between offshoring and innovation

3.2.1.1. Negative Binomial Models

We use negative binomial models15 to gain an insight into the relationship between firms’

innovation activities (monolithic, hybrid and non-integrated patent numbers) and their extent of

offshoring (number of full years that only assembly activities and number of full years that both

assembly and fabrication activities are offshore). In these models, since moving facilities and

products overseas does not happen within a single night or even a single month and often firms

have their old facility still running in the home country for the first few months, we define the

first “full” year of offshoring as the year after offshoring is documented in the SEC filings or

firm structured interviews. In addition, for both our offshoring and our R&D spending variables,

we then use a one-year-lag to account for the delay between when R&D is conducted and

subsequent patenting activities16. As the role a firm’s revenue in a given year may play in its

R&D outcomes that same year versus the next is less clear17, we use an unlagged revenue

variable in our main model, but explore using instead a lagged revenue variable in our robustness

tests. We consider a firm’s yearly revenue as a proxy of firm size 18 and its yearly R&D spending

as a proxy of firms’ R&D strategy. To indicate the overall economic and innovation environment

15 The distribution of patent numbers is right‐skewed count data; therefore we use negative binomial regression model. 16 We learned from our interviews that it takes on average one year after R&D is started for patents to be filed. Thus we would expect the effect of offshoring not to appear immediately but one year later. Similarly, R&D Spending this year will affects firms’ next year patent productivity. 17 Since there is some relationship between R&D spending and revenue, it could be argued to be appropriate to lag the revenue data, as well. That said, revenue in a given year may also affect how many patents a firm can afford to file in that year, and then not lagging the revenue data could be argued to be appropriate. 18 The other possible proxy of firm size is employee number. We have employee numbers from 16 firms and 112 observations for employee data (but revenue data from 19 firms and 158 observations for revenue information). Also, the lack of employee number of certain firms in certain year force us ignore some main integration producers, like Infinera, and therefore bias our regression results a lot.

of 44

in the U.S. optoelectronic industry, we use the total U.S. integrated patent numbers (monolithic

and hybrid) and non-integrated patent numbers as proxies for the overall innovation environment

in the regressions with integrated and non-integrated patents as the dependent variable,

respectively.

Our negative binomial regression models at firm level is as follows:

, ,

In addition to controlling for possible confounding firm effects, we also include firm

fixed-effect variables in our negative binomial regressions to analyze the effect within firms

(Appendix 2).

3.2.1.2. Cox Proportional Hazard Model

In addition to using negative binomial patent count models, we also model the firms’

patenting rates using Cox proportional hazards regression with repeated events clustered by firm

(Sorensen and Stuart 2000; Sosa 2009; Sosa 2011). Unlike the negative binomial model, this

approach does not assume that the patent rate is constant over a year or that a firm’s numbers of

patents each year are independent. Instead we analyze the risk of an event happening (here, a

particular type of patenting19) to a particular firm over time. Like in Sosa 2011, we define the

initial risk of a firm patenting as starting on the day before the firm’s first accepted patent

application file date. The risk of subsequent patenting events starts instantly after the previous

event happens.

We model the risk of patenting for each type of technology (monolithic, hybrid, or non-

integrated) as follows:

, exp ,

19 Our USPTO data sources allow us to access the precise filed date of a patent, and by the filed date we indicated an event happening.

of 44

In this model we use the same theoretical as control variables as the negative binomial

model – revenue, R&D expenditures, total patenting in the field of optoelectronics (integrated

versus non-integrated), and the burst of the telecommunications bubble. We use the natural log

of revenue and R&D spending, highly right-skewed variables, in our patent rate model to

improve model fit. (R&D spending remains lagged by one year.) Note that while the dependent

variable is a daily measure, these controlled variables, except for the telecommunications bubble

burst, are measured yearly.

3.3. Relationship between innovation capability and risk of offshoring

We use Cox proportional hazard regressions to model the risk of a firm going offshore

given the firm’s innovation capability (monolithic, hybrid, and non-integrated cumulative patent

numbers). We set the start of risk for going offshore for each firm as the year when the firm first

provides financial data (with the exception of one firm whose financial data was available for a

full decade prior to the other firms – its start date is set at 1992 for consistency with the general

observation period for the other firms). We observe each firm yearly until either the firm has

exited the market or 2006, the end of our observation period. In this model, going offshore is

defined as moving any type of facility (assembly or fabrication) offshore.

We model the risk of going offshore as follows:

exp , , , ,

,

We continue to use the natural log of revenue and R&D spending to improve model fit. Unlike

the previous negative binomial and patent rate hazard models, we do not lag our R&D spending

as we do not expect the same delayed effect on offshoring that we would expect for a patent

being granted. We include the firm cumulative patent numbers for all three technology types as

of 44

proxies for the firms’ innovation capabilities prior to going offshore. As in the previous models,

we include controls for the burst of the telecommunications bubble and all integrated or non-

integrated patents in each year for the U.S. optoelectronic field.

4. Results

4.1. Firm Regression Results

4.1.1. Relationship between offshoring and innovation

4.1.1.1. Negative Binomial Model

The regression results (Table 3) show that offshoring of just assembly activities and of

both assembly and fabrication activities is associated with a decrease in patenting in monolithic

integration. Controlling for firm revenue, firm R&D spending, the burst of the

telecommunications bubble, and the total integrated optoelectronic patent activity in the U.S.,

each additional year a firm has both assembly and fabrication activities offshore is associated

with a 0.59 decrease in yearly monolithic patents. These results match the economics found in

Fuchs and Kirchain (2010), which show that manufacturing overseas reduces the economic

viability of producing monolithically integrated technologies. Our results then take the Fuchs and

Kirchain (2010) results a step farther, showing that the firms not only lose incentives for

producing the emerging technology – they also reduce their innovation activities back home in

that same technology. It is particularly interesting that the statistically significant relationship

between offshoring and reducing innovation activities in monolithic integration is only found

once the firms move fabrication capabilities offshore – the specific manufacturing capabilities

key to producing the monolithically integrated technology.

The regression results for the relationship between offshoring and innovation in

technologies other than monolithic integration are equally interesting. While the offshoring of

only assembly activities is not associated with a statistically significant change in monolithic

of 44

integration patenting activities, it is associated with an increase in both hybrid integrated and

non-integrated patenting. Controlling again for firm revenue, firm R&D spending, burst of the

telecommunications bubble, and the total non-integrated optoelectronic patent activities in the

U.S., each additional year a firm has only assembly activities offshore is associated with a 0.79

increase in yearly hybrid patents and with a 0.71 increase in yearly non-integrated patents. The

increase in hybrid patenting is not surprising – hybrid integration is in many ways a labor- and

assembly-oriented substitute for monolithic integration that lacks the performance advantages of

monolithic integration could more easily be performed in an overseas facility. The increase in

non-integrated patenting is difficult to interpret without an in-depth understanding of the specific

technologies being patented. While we ran class and subclass analyses on these non-integrated

patents and also read a random sampling from each class, no immediate firm- or industry-level

trends were apparent. If these technologies are advanced technologies in other areas or directions,

this result could be seen as supporting past work by economists that suggests offshoring enables

a firm to save costs and thereby direct more resources toward higher-value-added activities (e.g.

Farell), (even if not to the emerging technology of monolithic integration). That said, with the

exception of one sub-subclass (372/50.1), which seemed to actually belong with monolithic

integration,20 the sample of patents from each major non-integrated class and subclass that we

read seemed oriented largely towards assembly activities.

To begin assessing the robustness of our regression results, we run analyses on two

additional models (Appendix 2: Tables A2.1 and A2.2). First, we explore controlling just for

Infinera. Infinera may be a possible outlier in our firms because it never goes offshore and holds

32% of all of our focus firms’ patents in monolithic integration. These two characteristics of

20 We ran a robustness analysis grouping this sub‐subclass instead with monolithic integration and find the same results reported in both the main model and the subsequent robustness analyses.

of 44

Infinera may overly influence our model. Here, the results of our previous models hold.

Specifically, the effect size of moving both assembly and fabrication offshore on monolithic

integration decreases slightly, we lose marginal significance on moving only assembly offshore,

and the burst of the telecom bubble loses significance. The remaining effects are relatively

similar, and the significance levels do not change. Second, we explore including firm fixed

effects for all firms into our regression to control for individual firm strategies. Here, the results

suggest that the offshoring (whether of just assembly or both assembly and fabrication activities)

is not associated with a statistically significant change in innovation. However, the number of

observations per firm is very small; these results should not be seen as conclusive. As a sanity

check, we also explored the analysis without the telecom bubble burst (which potentially could

exacerbate sample size issues), and find similar results. It is worth noting that in both cases (with

and without the bubble burst) many of the firm fixed effect coefficients are statistically

significant and vary in magnitude and direction. This finding may imply that each firm or distinct

groups of firms may have divergent behavioral trends that are associated with their innovation

trajectories.

Given the above-described results when adding firm fixed effects, we went back to the

data to develop a descriptive understanding of the strategies of individual firms, and whether

these strategies naturally fell into distinct groups. Whereas incompleteness of data in some

control variables limited which observations we were able to include in our regression analyses,

in the descriptive analyses we were not held to the same limitations. Leveraging the more

extensive data available through the SEC filings and our firm structured interviews, we first

created a descriptive write-up of each firm. Using methods common to case study analysis

(Eisenhardt 1989), we then unpacked the trajectories of groupings of firms to help inform our

interpretation of the quantitative results. (See Appendix 3 for our initial ranking of firms by

of 44

different variables, and Appendix 4 for our descriptive analysis.) As described in Appendix 4,

we found that our firms fell into one of five main categories: the “all-onshore strategists,” the

“all-offshore strategists”, the “hedgers” (fabrication onshore, assembly offshore), the “late

offshorers” (don’t go offshore immediately after the burst of the telecommunications bubble in

the first wave of firms, but instead go after 2005), and the “early exiters” (after the burst of the

bubble, either are acquired or go out of business by 2004).

Having completed this descriptive analysis of the firms and their strategies, we then

explored fixed effects associated with these five groupings of firm strategies (Appendix 2: Table

A2.3). Unlike the firm fixed effects, where we suffered from small numbers of observations for

some firms, the robustness analysis leveraging these strategy groupings is more statistically

stable. In this analysis, we are – similar to firm fixed effects – able to adjust for firm strategy,

but now in addition we can see the (adjusted) association between each type of strategy and

patenting. Our results for the relationship between number of years offshore and patenting (both

for offshoring only assembly and for offshoring both assembly and fabrication) are broadly

consistent in magnitude and significance to the primary model shown in Table 3. The inclusion

of firm strategies results in a loss of significance for the telecom bubble burst. While we see no

statistically significant relationship between firm strategy and monolithic patenting, we find, in

keeping with our theory, a negative association between “all-onshore” and “late-offshore”

strategies and non-integrated patenting. Given the low counts in monolithic patenting, we are

not surprised that it would be difficult for these patent counts to see an effect.

As with firm fixed effects, we also explored the analysis without the telecom bubble burst

(which potentially could exacerbate sample size issues). (See Appendix 2 Table A2.4.) With

respect to the relationship between a firm’s strategy and patenting (regardless of number of years

offshore), once we remove the telecommunications bubble, all strategies except keeping all

of 44

manufacturing onshore – specifically, being “all-offshore”, a “hedger”, or an “early exiter” – are

associated with a decrease in monolithic patenting. Again, these results match our theory. As

with the analysis that included the control for the telecommunication bubble, while the “all-

onshore” and “late-offshore” groups do not have a statistically significant association with

monolithic patenting, they have a negative association with non-integrated patenting. These

results suggest that the firms with strategies that keep their fabrication onshore (either everything

onshore, just fabrication, or not going offshore in the first wave immediately after the bubble

burst) may be more likely to have fewer non-integrated patents, compared to firms with more

aggressive offshoring strategies.

In conclusion, the above analyses only present correlations between number of years

offshore and firms’ quantity and direction of innovation. Building on the above strategy fixed

effects, in section 4.1.2 below, we explore what type of firms are more likely to move offshore,

and in particular whether firms with lower capabilities in monolithic integration (or other

common traits like being publicly traded) are more likely to move. This said, it is worth noting

that given the economics shown in Fuchs and Kirchain (2010), even if we were to find that the

weakest firms in innovation (or in monolithic innovation) are those most likely to move overseas,

it may not matter for national innovation outcomes if the firms that stay in the U.S. to pursue the

emerging, monolithically integrated technology cannot survive in the short term. As shown

earlier in Figure 1, firms that move manufacturing overseas are able to produce the old, discrete

technology in Developing East Asia at a significantly lower cost than they can produce either the

emerging, monolithically integrated or the old, discrete technology in the U.S. Thus, firms that

move overseas and produce there the prevailing technology will be able to dramatically out cost-

compete firms that stay in the U.S. and attempt to produce there the emerging, monolithic

of 44

technology until markets (such as the sensor or computing markets) are willing to pay for the

added performance provided by the monolithically integrated technologies.

Table 3 Negative Binomial Regressions on Focus Firms

Monolithic integrated Hybrid integrated Non‐integrated

Assembly Only Offshoring

y‐1

‐0.53 . (0.28)

0.79 ** (0.26)

0.71 *** (0.18)

Both Offshoringy‐1 ‐0.59 ***

(0.15) ‐0.009 (0.09)

‐0.02 (0.06)

Revenuey ‐4.6e‐11

(1.7e‐10) 1.9e‐10 (1.5e‐10)

2.4e‐10 * (1.0e‐10)

R&D Spendingy‐1 1.4e‐09.

(1.1e‐09) ‐3.6e‐10 (1.3e‐09)

3.9e‐10 (7.1e‐10)

U.S. Int. Patenty 0.002

(0.005) 0.03 *** (0.006)

U.S. Non‐int. Patent

y

7.6e‐04 *** (1.2e‐04)

Telecom bubbley‐1 0.92 *

(0.73) ‐0.89 (0.59)

‐0.54 ** (0.21)

Constant ‐1.04 (0.75)

‐4.26 *** (0.69)

‐1.09 (0.61)

N= 152 152 152

AIC 325.21 245.71 1093.8 Robust standard errors are in parentheses.

Signif. codes: ‘***’ p< 0.001; ‘**’ p< 0.01; ‘*’ p< 0.05; ‘.’ p< 0.1

4.1.1.2. Cox Proportional Hazard Model

In contrast to negative binomial models, a Cox proportional hazard model allows us to

incorporate varying patent rates over a year and that a firm’s number of patents across years may

be correlated. The results of the patent hazard models (Table 4) show that offshoring both

assembly and fabrication activities is associated with a decrease in patenting in monolithic

integration. Controlling for offshoring only assembly activities, firm revenue, and firm R&D

spending, each additional year a firm has both assembly and fabrication facilities offshore is

associated with a marginally significant rate decrease of 26% for firms to file a monolithic patent.

These results match the economics found in Fuchs and Kirchain (2010), which show that

manufacturing offshore reduces firms’ incentives for and the economic viability of producing the

of 44

emerging monolithic products. Our results also extend these past findings by suggesting that in

following these economics, firms not only stop producing monolithically integrated products but

also reduce their innovation activities in the emerging technologies. Particularly interesting is

that the statistically significant relationship between offshoring and reducing innovation

activities in monolithic integration is only found once the firms move fabrication capabilities

offshore – the specific manufacturing capabilities key to producing the monolithically integrated

technology.

In addition to the results on the relationship between offshoring and innovation in the

emerging, monolithic technology, we also explore using Cox proportional hazards regression to

model the relationship between offshoring and innovation in the prevailing non-integrated

technologies. Here, interestingly, we find the opposite results from the monolithically integrated

technology. Controlling for offshoring only assembly activities, firm revenue, and firm R&D

spending, each additional year a firm has both assembly and fabrication facilities offshore is

associated with a rate increase of about 27% for firms to file a non-integrated patent. These

results could be seen as supporting past work by economists that suggest offshoring enables a

firm to save costs and thereby direct more resources toward higher-value-added activities (e.g.

Farell), (even if not to the emerging technology of monolithic integration.) It is interesting,

however, that according to this analysis firms must move both assembly and fabrication offshore

to gain to gain the theoretical cost savings that would lead to increased innovations according to

economists. Subsequent work should explore the technical nature of these non-integrated patents

and their disruptiveness compared to the monolithically integrated patents.

Surprisingly, in contrast with the negative binomial model results, here it is moving both

assembly and fabrication offshore (not moving only assembly offshoring) that is associated with

a statistically significant increase in non-integrated patenting. In looking at the descriptive details

of 44

of our data, one firm, Agilent – which moves both assembly and fabrication offshore in 1999,

has a disproportionately large spike in its non-integrated patenting (compared to upward or

downward patenting rates of other firms) at the beginning of its offshoring period. After this

initial boost, however, Agilent’s non-integrated patenting dramatically declines for the duration

of our observation period. This spike in Agilent’s innovation activities combined with Agilent’s

consistently larger numbers of patents may be overly influencing the results, and is worth future

study to understand their strategy21. For example, were they patenting previously unpatented

technologies in preparation for losing employees associated with offshoring?

Table 4 Cox Patent Rate Models on Rate of Patenting in Monolithic and Non‐Integrated Technologies

Monolithic Hybrid Non‐Integrated

Assembly Only Offshoring

y(d)‐1

0.29

(1.04)

1.32

(0.31)

1.57

(0.29)

Both Offshoringy(d)‐1

0.74 .

(0.18)

1.03

(0.07)

1.27 *

(0.11) Revenue

y(d) 0.59 ***

(0.12)

1.09

(0.06)

1.14

(0.15) R&D Spending

y(d)‐1 2.03 ***

(0.18)

1.07

(0.12)

1.26

(0.15) U.S. Int. Patent

y(d) 1.01

(0.009)

1.03 ***

(0.008) U.S. Non‐int. Patent

y(d) 1.001 ***

(1.7e‐05) Telecom bubble

d‐365 0.22 .

(0.84)

0.08 **

(0.93)

0.53 *

(0.27) N= 122 85 2964 Allcoefficientsareinhazardratios.Notethatcoefficientslessthan1areequivalenttonegativeassociations.Allregressionsarerepeatedeventsclusteredbyfirmwithrobuststandarderrors(inparentheses)Signif. codes: ‘***’ p< 0.001; ‘**’ p< 0.01; ‘*’ p< 0.05; ‘.’ p< 0.1

4.1.2. Relationship between innovation capability and risk of offshoring

The Cox proportional hazards regression modeling the risk of going offshore indicates that

cumulative non-integrated patenting (the innovation capability of non-integrated design) is marginally

21 When running the proportional hazards regression without Agilent included, our results become similar to those reported in the negative binomial, where the relationship between moving only assembly offshore and non‐integrated patenting is positive and significant and the relationship between moving both offshore and non‐integrated patenting is non‐significant, but has a negative coefficient.

of 44

associated with an increase in the probability of offshoring any type of facility. Controlling for firms’

cumulative monolithic and hybrid patenting, firm revenue, firm R&D spending, the burst of the

telecommunications bubble, and field-wide trends in integration and non-integration patenting; each

additional cumulative non-integrated patent is associated with a 1% increase in the risk of moving any

facility offshore (marginally significant). Not surprisingly, we also find a positive correlation between the

burst of the telecommunications bubble and offshoring (marginally significant).

These Cox proportional hazard results for probability of offshoring counter the theory that firms

that have low capability in patenting (monolithic, hybrid and non-integrated) would have a higher

probability to go offshore. In contrast, the positive relationship between non-integrated patenting and

going offshore may suggest that firms with a breadth of capabilities perceive that they have greater

competencies to leverage in going offshore. Our results also do not support a theory that firms that were

weak in monolithic integration would be more likely to go offshore. Table 6 below shows correlations

between the three different types of patenting. The results show that a firm’s monolithic, hybrid and non-

integrated cumulative patenting are all positively correlated with each other. These results suggest that a

firm that has a high innovation capability (large cumulative patents) in the non-integrated design is likely

to also have high innovation capability in monolithic and hybrid designs.

To assess the robustness of our results, we also model the rate of a firm moving only assembly

offshore and the risk of moving both assembly and fabrication offshore. In addition, we explore the

relationship between yearly patenting rates (instead of cumulative patent numbers) and the risk of

offshoring for all three definitions of offshoring.22 Similar to the cumulative results, the results of the

offshoring hazard models with yearly patenting rates show that yearly non-integrated patenting is

associated with an increase in the risk of any type of offshoring (significant at the 0.05 level) and an

increase in the risk of moving both assembly and fabrication offshore (marginally significant). None of

the other models found a statistically significant relationship between firm’s patenting and the rate of

22 To recall, our three definitions of offshoring for the offshoring hazard models are any type of facility offshore, offshoring only assembly, or offshoring both assembly and fabrication facilities.

of 44

offshoring. These results suggest that a firm’s innovation capability is either not associated with going

offshore or at best positively associated.

Table 5 Cox proportional hazard regression results on the risk of moving any facility offshore

Rate of offshoring any type of facility

Cumulative monolithic patenty 0.94 (0.05)

Cumulative hybrid patenty 1.22 (0.27)

Cumulative non‐integrated patenty 1.01 . (0.006)

Revenuey 1.39 (0.36)

R&D spendingy 0.64 (0.44)

U.S. Int. Patenty 0.98 (0.03)

U.S. Non‐int. Patenty 1.00 (7e‐4)

Telecom bubbley 22.06 . (1.82)

N 131 Allcoefficientsareinhazardratios.Weincluderobuststandarderrorsinparenthesis.Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Table 6 Correlation between each type of cumulative patents

Cumulative monolithic

Cumulative hybrid Cumulative non‐integrated

Cumulative monolithic 1 0.435 0.398 Cumulative hybrid ‐ 1 0.621 Cumulative non‐integrated

‐ ‐ 1

4.2. Inventor Results

While firms that offshore may reduce their innovation efforts in monolithic integration,

inventors need not take the same trajectory as firms. Table 7 shows the percentage of (or

probability that) people within our target strong integration group versus each of our control

groups (as described earlier under Inventor Scope (section 3.1.6)) take particular actions. The

actions we observe are as follows: (1) eventually leave the focus firm, (2) conditional on leaving

of 44

the focus firm leave to an onshore firm (if at an offshoring firm) versus to an out-of-focus firm,

and (3) conditional on leaving and to what type of firm they go, keep patenting in monolithic

integration.23 We use t-tests to assess whether the observed differences between groups are

statistically significant. (See Table 8 and Appendix 5)

4.2.1. Do the inventors at the offshoring firms with strong capabilities in monolithic integration leave, and keep innovation in the emerging technology going at other institutions?

In analyzing movements of our inventors at the offshoring firms with strong capabilities

in monolithic integration, our results are contrary to our expectations:

On average, 43% of inventors with strong capabilities in monolithic integration (and who

patent at the focus firms in monolithic integration) eventually leave. Contrary to what we might

have expected, there is no statistically significant difference in the leave rates of these inventors

whether they are at an offshoring firm or a non-offshoring firm. In contrast, on average only

17% of inventors with only one or two patents in integration or with no patents in integration

eventually leave their firms (inventors with no patents in integration have a slightly lower

probability to leave if they are at an offshoring firm, but otherwise there are no statistically

significant differences between the various groups without strong patenting capabilities in

monolithic integration).

Of the inventors with strong innovation capabilities in integration who have patented at

23 Note: the total number of “focus” integrated inventors shown in Table 7 is 98 instead of 106 due to us, given limited space, not including in the table two groups of inventors: seven inventors with three or more integrated patents who have monolithic experience only post being in our focus firms (and no monolithic patents prior to or at our focus firms) and one inventor (out of 3388 inventors who ever patent at our focus firms) that didn’t fit into our classification system. The seven inventors (the first group) had only non‐integrated patents at offshoring firms and then go to an onshore firm and patent in monolithic (6 go to Infinera and one goes to Xponent). We discuss the six inventors that go to Infinera in greater detail in section 4.2.2. The one inventor who didn’t fit our classification system currently works at an offshoring firm (Neophotonics) and has no patents at that firm but had

monolithic patents at Lightwave (an on‐shore firm acquired by Neophotonics) and in one out‐of‐focus firm (pre Lightwave) prior to going to Neophotonics. The fact that this inventor goes from an onshore firm to an offshore firm (in this case, due to a acquisition) excludes him from fitting one of the categories in the table.

of 44

the focus firm in monolithic integration and eventually leave, almost all of them (94%) go to out-

of-focus firms (outside the optoelectronic component manufacturing industry). (This statistic is

also similar for our control group—of the inventors at our focus firms with no integration patents

who leave, more than 99% of them go to out-of-focus firms.) And yet, again, contrary to our

expectations, of the inventors with strong monolithic integration capabilities at our offshoring

firms who eventually leave, only 21% of them patent in monolithic integration after leaving.

These results suggest that of the inventors who had been patenting in monolithic integration at

the offshoring firms and leave the industry (to out-of-focus firms) the majority of them end up

not pushing forward efforts in monolithic integration in their new locations.24 Notably, these

results do not indicate that monolithic integration activities are not continued by anyone after

they are abandoned by the offshoring firms, just that the majority of the inventors previously

patenting in monolithic at the offshoring firms are not the ones to do it.

Along these lines, two trends are particularly interesting. First, of the 40% of inventors

with strong capabilities in monolithic integration who eventually leave an offshoring firm, 10%

(2) of these inventors go to an onshore firm, and both of them continue patenting in monolithic

integration. That onshore firm is Infinera. Even more interesting, 70% of inventors with strong

capabilities in monolithic integration, who have prior experience in monolithic integration but

without any patents in monolithic integration at our focus offshoring firms eventually leave.25

Of those who eventually leave the offshoring firms 43% (3) go to an onshore firm, and again all

of them continue patenting in monolithic integration at that onshore firm. Again, this onshore

firm is Infinera. Particularly noteworthy in the above findings is that inventors with prior strong

24 As an aside, a slightly higher percentage of the equivalent inventors from never offshore firms patent in monolithic integration after leaving – 43%. This difference is marginally significant, despite the low sample size. 25 This number compares with, on average 43% of inventors with strong capabilities in monolithic integration but who patent in monolithic integration at the focus firms and on average 17% of inventors with only one or two patents in integration or with no patents in integration eventually leaving their firms.

of 44

Table 7: Comparison between different groups of inventors

Focus Integrated Inventors (>2 Int. Patents)

Remaining Integrated Inventors (1 or 2 Int. Patents)

No Integrated Patents

Patent in Monolithic Patent in Hybrid Patent in Monolithic Patent in Hybrid

@ Focus firms Pre‐focus firm @ Focus firms @ Focus firms Pre‐focus firm @ Focus firms

81 10 7 228 8 147 2907

52 Ever

offshore

29 Never offshore

10 Ever

Offshore

0 Never

Offshore

6 Ever

Offshore

1 Never

Offshore

193 Ever

offshore

35 Never offshore

8 Ever

Offshore

0 Never

Offshore

134

Ever

Offshore

13

Never

Offshore

2685

Ever

offshore

222

Never

offshore

Eventually Leave

40% (21)

48% (14)

70% (7)

0% 33% 0 18% (34)

17% (6)

25% (2)

0 16% 8%

(1)

16% 23%

To On‐shore

10% (2)

NA 43% (3)

NA 50% (1)

NA 0% NA 50% (1)

NA 0% NA 0.23% (1) NA

Patent in Mon. after Leaving

100% NA 100% NA 100% NA 0% NA 100% (1)

NA 0% NA NA NA

To Out‐of‐Focus

90% (19)

100% 57% (4)

0% 50% 0% 100% (34)

100% (6)

50% (1)

0% 100% 100% 99% 100%

Patent in Mon. after Leaving

21% (4)

43% (6)

25% (1)

0% 0% 0% 6% (2)

0% 0% 0% 0% 0% NA NA

Table 8: T‐test bet

Page

tween different grou

e 35 of 44

ups of inventors' ratios of eventually leaaving

of 44

monolithic integration capabilities that were not using those capabilities at the focus firms are

both more likely to leave, and more likely to go to an onshore firm (specifically, Infinera) where

they patent in monolithic integration. (The latter of the two points being marginally significant

on a very small sample size.) Upon initial analysis of the regional locations of these inventors, it

does not look like these inventors would be affected in their states by strongly enforced non-

compete regulations. This finding warrants further exploration as other incentives, such as the

recipient company not wanting to risk patent or knowledge infringement challenges from recent

employers, might be driving these differences.

4.2.2. The Inventors that Go to Infinera

As shown in our firm analysis, Infinera – which holds 21% of integrated patents

(monolithic and hybrid) within our focus firms and 32% of monolithic integration patent within

our focus firms – plays a critical role in monolithic integration innovation. Our findings at the

individual level shows 22% of our 106 focus integration inventors (three or more patents in

integration) and 83% of our inventors with more than 9 integrated patents work at some point at

Infinera. To understand better what led to the creation of such a powerhouse in integration, we

examine the trajectories of inventors who have ever worked at Infinera, and try to unpack what

Infinera’s strategy might have been in hiring these inventors (Table 9).

More than half (52%) of the monolithic integration inventors at Infinera worked

previously at one of our offshoring firms. All of the inventors who worked at our offshoring

firms before going to Infinera left the offshoring firm to go to Infinera one to four years after the

firm moved offshore1. While 42% of these inventors had previously patented in monolithic

integration prior to coming to Infinera (two of these inventors had previously patented in

1 Six of the 16 Infinera inventors come from JDSU, 2 from Avago, 2 from Corvis, 1 from TriQuint, 1 from Agilent, 1 from Intel, 1 from Harmonic Lightwaves, 2 start their patenting careers at Infinera.

of 44

monolithic integration at their offshoring firm, and another three of these inventors had patented

in monolithic integration prior to going to their offshoring firm), a full 58% of the inventors have

Table 9. The Movement of Inventors Who ever worked at Some Point at Infinera

Monolithic Inventors at Infinera

Firm pre‐ Infinera

Headquarters(state)

Leave after firm offshore (years)

Arrival at Infinera,2

Firm post Infinera

Depart Infinera

Drew Perkins (Founder)

(NA) (NA) (NA) 2001 (NA) (NA)

David F. Welch (Founder)

JDSU CA 2 2001 (NA) (NA)

Frank H. Peters Avago CA 2 2001 Tyndall National Institute

2004

Richard P. Schneider Avago CA 2 2001 (NA) (NA)

Stephen Gregory Grubb †

Corvis ‡ MD ‡ (NA) 2001 (NA) (NA)

Matthew L. Mitchell † Corvis ‡ MD ‡ (NA) 2001 (NA) (NA)

Charles H. Hoyner TriQuint OR 1 2002 (NA) (NA)

Radhakrishnan L. Nagarajan

JDSU CA 2 2002 (NA) (NA)

Mark J. Missey JDSU CA 2 2002 (NA) (NA)

Vincent G. Dominc JDSU CA 2 2002 (NA) (NA)

Atual Mathur JDSU CA 2 2002 (NA) (NA)

Andrew Dentai Agere PA 2 2002 (NA) (NA)

Fred Kish Jr. Agilent CA 3 2002 (NA) (NA)

Alan C. Nilsson † Harmonic Lightwave

CA (NA) 2002 (NA) (NA)

Jonas Webjorn † Intel CA (NA) 2002 (NA) (NA)

Robert B. Taylor † (NA) (NA) (NA) 2002 (NA) (NA)

Ting‐Kuang Chiang † (NA) (NA) (NA) 2002 (NA) (NA)

Brent Little MIT MA (NA) 2002 (NA) (NA)

Mehrda Ziari JDSU CA 3 2003 (NA) (NA)

John Hryniewicz Little Optics ‡

MD ‡ (NA) 2005 (NA) (NA)

Wei Chen Avanex CA 4 2007 (NA) (NA)

David Gill Little Optics ‡

MD ‡ (NA) 2007 (NA) (NA)

Wenlu Chen (NA) (NA) (NA) 2007 (NA) (NA)

† Never worked at offshoring firms

‡ In 2006, Infinera took over Corvis’s all‐optical switch product line, including licensing the intellectual property rights to it, and gaining 40 Broadwing employees that had been part of the Corvis business (Corvis changed its name to Broadwing in 2004 after acquiring the Broadwing in 2003, however existing patents kept the Corvis name); Little Optics was acquired by Infinera in 2007 (NA) Not applicable.

2 The time shows when the inventor had his/her first patent at Infinera (based on USPTO database) or when he/she entered Infinera (if his/her resume is available)

of 44

their first patent in monolithic integration only after arriving at Infinera. As can be seen in Table

5, several inventors who worked together in the same firms – in particular JDSU and

Avago/Agilent – move together to Infinera. In the case of JDSU, the inventors follow David

Welch – one of the Infinera’s two founders. David Welch originally worked at SDL, which was

acquired by JDSU in 2001. Three months after the acquisition, David Welch left JDSU and

founded Infinera.

These results are surprising given past literature suggesting that once an inventor has been in

a particular technical area for more than three years, he or she is likely to continue innovation

activities in the same direction despite institutional changes, or other outside forces (Garud and

Rappa 1995; Furman et al. 2010). These results also stand in contrast to the body of literature on

inventors bringing technological expertise from previous firms into new ones (e.g. (Rosenkopf

and Almeida 2003; Song et al. 2003; Klepper and Sleeper 2005; Klepper and Thompson 2010)).

In addition, given the extensive recent literature on the significance of non-compete agreements

(Marx 2009; Marx et al. 2009; Marx et al. 2010; Marx 2011; Marx and Fleming 2011), it is

particularly interesting that Infinera, despite being headquartered in California – a state that does

not enforce non-competes, would not primarily recruit inventors from the offshoring firms with

past experience in monolithic integration. Notably, Infinera does have subsidies in Maryland and

in Pennsylvania, states that may more strictly enforce non-compete agreements. And yet, all of

the inventors Infinera recruits either are part of a legal agreement (one undisclosed agreement

involving the take over of another company’s business line including licensing of patents and

transfer employees, and one formal acquisition) or already come from a company in California.

Leaving aside for a moment Infinera, a partial explanation for why more inventors don’t

continue activities in the emerging technology (monolithic integration) after the burst of the

of 44

bubble may lie in the finding that individual respond to post-employment restraints by taking

“occupational detours” (Marx 2009). Alone, however, this explanation is unsatisfactory, given

that so much of optoelectronic component manufacturing and opportunities in optoelectronics

(both within and outside of telecommunications) lies in California – a state without non-compete

enforcement. Future work should interview inventors with strong capabilities in the emerging

technology that leave the offshoring firm and go to firms outside the telecommunications

industry to understand why more of them don’t continue activities in the emerging technology.

The occupational detours explanation also, alone fails to explain why Infinera wouldn’t focus on

hiring inventors with previous experience in the emerging technology. Future work should also

conduct interviews of Infinera’s founders and inventors to further unpack the story behind why

Infinera largely recruited inventors from their offshoring competitors in optoelectronics

component manufacturing for telecommunications that did not have experience at those firms

(nor in many cases anywhere else) in monolithic integration.

5. Discussion and Conclusion

This study explores the relationship between offshoring and innovation directions. We seek

to understand whether 1) offshoring firms decrease activities in the emerging technologies after

moving offshore, and 2) inventors who had been working on the emerging technologies at the

offshoring firms leave and continue innovation in the emerging technology at other institutions.

We look, in particular, at offshoring in small and medium sized optoelectronic component

manufacturers for the telecommunications industry after the burst of the telecommunications

bubble. We look at three competing technologies: The emerging technology is monolithic

integration. By dramatically reducing form factor size, monolithic integration enables the

benefits of reduced power and increased bandwidth inherent to optoelectronics (in comparison to

of 44

pure electronics) to reach markets outside of telecom including computing, biotechnology,

energy, and military applications. The alternatives are hybrid integration (an intermediate

technology, less advanced than monolithic integration) or discrete technologies (involving

human-based assembly instead of more sophisticed scientific-based combination of functions on

a single semiconductor chip). Both of these latter technologies involve greater labor and

assembly requirements, and lack the critical performance (small form factor) benefits of

monolithic integration for accessing larger markets. Building on Fuchs and Kirchain (2010), we

hypothesize that offshoring will lead to a decrease in innovation in the most advanced

technology, monolithic integration, due to discrete technologies produced in developing East

Asia being cheaper than any other option (including the advanced monolithically integrated

technologies, despite those technologies being the cheapest to produce if only manufacturing in

the U.S.)

Our regressions suggest that moving all manufacturing offshore (both assembly and

fabrication) is associated with a decrease in emerging, monolithic integration innovation

activities. These results match the economics found in Fuchs and Kirchain (2010), which show

that manufacturing overseas reduces the economic viability of producing monolithically

integrated technologies but then take those results a step farther, showing that the firms not only

lose incentives for producing the emerging technology – they also reduce their innovation

activities back home in that same technology. It is particularly interesting that the statistically

significant relationship between offshoring and reducing innovation activities in the emerging

technology is only found once the firms move fabrication capabilities offshore – the specific

manufacturing capabilities key to producing the emerging technology.

of 44

The regression results on the relationship between offshoring and innovation in

technologies other than monolithic integration are equally interesting. While the offshoring of

only assembly activities is not associated with a statistically significant change in innovation in

the emerging technology (monolithic integration), it can be associated with an increase in other

innovation activities (here, hybrid integrated and non-integrated patenting). Additional work is

necessary to understanding the relationship between moving all manufacturing offshore (both

fabrication and assembly) and rates of innovation in areas outside of emerging technologies with

tight links to manufacturing. In the case of the optoelectronics industry, the increase in hybrid

patent counts when only moving assembly facilities offshore is not surprising – hybrid

integration is in many ways a labor- and assembly-oriented substitute for monolithic integration

that while it lacks the performance advantages of monolithic integration, could likely easily be

performed in an overseas facility. The increase in non-integrated patent counts for moving only

assembly offshore (and of non-integrated patenting rates for moving both assembly and

fabrication offshore in the case of Agilent) is difficult to interpret without greater understanding

of the technologies being patented. If these technologies are advanced technologies in other areas

or directions, this result could be seen as supporting past work by economists that suggest

offshoring enables a firm to save costs and thereby direct more resources toward higher-value-

added activities (e.g. Farell), (even if not to the emerging technology of monolithic integration.)

If making this argument, it would be important to understand whether the advanced technologies

did or did not have local manufacturing. On the other hand, if the areas in which firms are

patenting are more incremental activities, such as supporting improvements in assembly, the

most important result may be the decrease in patenting in the emerging technologies.

of 44

At the inventor level, we find that the majority of inventors with strong patenting records

in the emerging technology at the offshoring firms are not the ones who push forward innovation

in the emerging technology after leaving. Only 21% of inventors with strong patenting records

in the emerging technology at the offshoring firms who leave the offshoring firm for a firm

outside the optoelectronic component manufacturing industry continue to patent in the emerging

technology. That said, inventors who go to the onshore firm, Infinera, do push forward patenting

in the emerging technology. More than half of inventors at Infinera were previously at an

offshoring firm, and all of them leave that firm for Infinera one to four years after their previous

firm moves offshore. Interestingly, however, only 42% of these inventors from offshoring firms

have any past experience in the emerging technology (and only 9% have past experience at the

offshoring firm in the emerging technology – the rest have earlier experience.) The rest of the

inventors have their first patent in the emerging technology at Infinera. Subsequent research is

needed to better understand why more inventors with strong capabilities in the emerging

technology do no continue innovation in these areas after leaving the offshoring firm. Likewise,

it will be important to conduct future work to better understand Infinera’s hiring strategy – which

clearly gains from the offshoring firms’ employee losses, although not by focusing on hiring

existing experts in the emerging technology – and how that strategy may have led to Infinera

becoming the telecommunications industry’s leader in the emerging technology that it is today.

References

Agrawal, V. and D. Farrell (2003). "Who wins in offshoring." McKinsey Quarterly: 36‐53. Allen, T. (1984). "Managing the flow of technology: Technology transfer and the dissemination of

technological information within the R&D organization." MIT Press Books 1. Almeida, P. and B. Kogut (1999). "Localization of knowledge and the mobility of engineers in regional

networks." Management Science 45(7): 905‐917. Baily, M. and D. Farrell (2004). "Exploding the myths of offshoring." The McKinsey Quarterly. Bardhan, A. and D. Jaffee (2005). "Innovation, R&D and offshoring." Bardhan, A. and C. Kroll (2003). The new wave of outsourcing. Fisher Center for Real Estate & Urban

Economics Research Report Series No. 1103.

of 44

Benner, M. and M. Tushman (2002). "Process Management and Technological Innovation: A Longitudinal Study of the Photography and Paint Industries." Administrative Science Quarterly 47(4): 676‐709.

Bessen, J. (2009). NBER PDP Project User Documentation: Matching Patent Data to Compustat Firms,

Version: Beta, May. Braga, C., C. Fink and C. Sepulveda (1998). "Intellectual Property Rights and Economic Development."

TechNet. Bresnahan, T. and M. Trajtenberg (1995). "General purpose technologies: "Engines of Growth?"."

Journal of Econometrics 65(1): 83‐108. Brown, J. and J. Hagel (2005). "Innovation blowback: Disruptive management practices from Asia."

McKinsey Quarterly 1: 35‐45. Dewey and LeBoeuf (2009). Maintaining America’s competitive edge: Government policies affecting

semiconductor industry R&D and manufacturing activity, the Semiconductor Industry Association.

Dosi, G. (1982). "Technological paradigms and technological trajectories:: A suggested interpretation of the determinants and directions of technical change." Research policy 11(3): 147‐162.

Eisenhardt, K. (1989). "Building theories from case study research." Academy of management review 14(4): 532‐550.

Fifarek, B., F. Veloso and C. Davidson (2008). "Offshoring technology innovation: A case study of rare‐earth technology." Journal of Operations Management 26(2): 222‐238.

Fleming, L., C. King and A. Juda (2007). "Small worlds and innovation." Organization Science 18(6): 938‐954.

Fuchs, E. and R. Kirchain (2010). Design for location? The impact of manufacturing offshore on technology competitiveness in the optoelectronics industry. Management Science, Working Paper, Carnegie Mellon University, College of Engineering.

Fuchs, E. R. H., F. R. Field, R. Roth, et al. (2011). "Plastic Cars in China? The Significance of Production Location over Markets for Technology Competitiveness." International Journal of Production Economics 132: 79‐92.

Furman, J., F. Murray and S. Stern (2010). Growing Stem Cells: The Impact of U.S. Policy on the Geography and Organization of Scientific Discovery. Opening Up Innovation: Strategy, Organization, and Technology. Imperial College London Business School, Druid Summer Conference. Working Paper Series.

Garud, R. and M. Rappa (1995). "On the persistence of researchers in technological development." Industrial and Corporate Change 4(3): 531.

Ghemawat, P. (2001). "Distance still matters." Harvard Business Review 79(8): 137‐147. Hall, B., A. Jaffe and M. Trajtenberg (2001). The NBER patent citation data file: Lessons, insights and

methodological tools. NBER working paper series. w8498. Holden, H. (2003). "The developing technologies of integrated optical waveguides in printed circuits."

Circuit World 29(4): 42‐50. Klepper, S. and S. Sleeper (2005). "Entry by Spinoffs." Management Science 51(8): 1291‐1306. Klepper, S. and P. Thompson (2010). "Disagreements and intra‐industry spinoffs." International Journal

of Industrial Organization 28(5): 526‐538. Lewin, A. and V. Couto (2007). Next generation offshoring: the globalization of innovation, Duke

University Center for International Business Education and Research. Marx, M. (2009). Good Work if you Can Get It Again: Non‐compete Agreements and Ex‐employees. 2009

Annual Academy of Management Meeting, Chicago, IL, Winner of AOM TIM Division Best Student Paper Award.

of 44

Marx, M. (2011). "The Firms Strike Back: Non‐Compete Agreements and the Mobility of Technical Professionals." American Sociological Review 76(5): 695‐712.

Marx, M. and L. Fleming (2011). Non‐compete Agreements: Barriers to Entry… and Exit? Innovation Policy and the Economy. J. Lerner and S. Stern. Chicago, IL, University of Chicago Press. 12.

Marx, M., J. Singh and L. Fleming (2010). "Regional Disadvantage? Non‐competes and Brain Drain." Review of Economics and Statistics R&R.

Marx, M., D. Strumsky and L. Fleming (2009). "Mobility, Skills, and the Michigan Non‐compete Experiment." Management Science 55(6): 875‐889.

Mickelson, A., N. Basavanhally and Y. Lee (1997). Optoelectronic packaging, Wiley New York. Miguélez, E. and R. Moreno Serrano (2010). Research Networks and Inventors' Mobility as Drivers of

Innovation Evidence from Europe. Documents de Treball (IREA): 1. OIDA (2005). Optical Interconnects: “Thinking Inside of the Box”, Optoelectronic Industry Development

Association. OIDA (2006). Global Optoelectronics Industry Market Report and Forecast. Washington, D.C.,

Optoelectronics Industry Development Association. Oviatt, B. and P. McDougall (2004). "Toward a theory of international new ventures." Journal of

International Business Studies 36(1): 29‐41. PCAST (2011). Report to the President on Ensuring American Leadership in Advanced Manufacturing. Pisano, G. P. and W. C. Shih (2009). "Restoring American Competitiveness." Harvard Business Review

87(7/8): 114‐125. Rosenkopf, L. and P. Almeida (2003). "Overcoming Local Search Through Alliances and Mobility."

Academy of Management Journal 49: 751‐766. Rotman, D. (2012). Can We Build Tomorrow's Breakthroughs? Technology Review, MIT. Schabel, M. (2005). Current State of the Photonics Industry. Microphotonics: Hardware for the

Information Age. L. Kimerling, Cambridge, MA, MIT Microphotonics Center. Shah, J. (2007). "Ultraperformance Nanophotonic Intrachip Communications: UNIC." DARPA/MTO‐

Frontiers of Extreme Computing. Solow, R. (1957). "Technical change and the aggregate production function." The Review of Economics

and Statistics 39(3): 312‐320. Song, J., P. Almeida and G. Wu (2003). "Learning by hiring: When is mobility more likely to fcilitate

interfirm knowledge transfer?" Management Science 49(4): 446‐463. Sorensen, J. B. and T. E. Stuart (2000). "Aging, obsolescence, and organizational innovation."

Administrative Science Quarterly: 81‐112. Sosa, L. (2009). "Application‐Specific R&D Capabilities and the Advantage of Incumbents: Evidence from

the Anticancer Drug Market." Management Science 55(8): 1409‐1422. Sosa, L. (2011). "From Old Competence Destruction to New Competence Access: Evidence from the

Comparison of Two Discontinuities in Anti‐Cancer Drug Discovery." Organization Science Articles in Advance: 1‐17.

Tassey, G. (2010). "Rationales and mechanisms for revitalizing US manufacturing R&D strategies." The Journal of Technology Transfer 35(3): 283‐333.

Tyre, M. and E. Von Hippel (1997). "The situated nature of adaptive learning in organizations." Organization Science 8(1): 71‐83.

USPTO (2010). Overview of the U.S. Patent Classification System (USPC). WhiteHouse. (2011). "Advanced Manufacturing Partnership Press Release " Retrieved January 20, 2012,

from http://www.whitehouse.gov/the‐press‐office/2011/06/24/president‐obama‐launches‐advanced‐manufacturing‐partnership.

Appendix

1. PatentApplicationAnalysis2. RobustAnalysisofRegressionModels3. FirmRankedList4. FirmCaseStudies5. InventorT‐testTables

Appendix1

Appendix 1. Patent Application Analysis

Ouranalysisofthefocusfirms’patentdatashows,onaverage,aonetofouryeargapbetweenpatentapplicationdatesandpatentgrantingdates.Asaconsequence,whilethemajorityofthefirmswentoffshorebetween1997to2007(onein1997,onein1998,onein2000,threein2001,twoin2002,twoin2003,fourin2005,onein2006andtwoin2007),currentlyweonlyareabletoobservethreetonineyearsofpost‐offshoringpatentingbehaviorforseventeenfirmsandhavenoreliablepost‐offshoringpatentingdataforthesevenlatemoverfirms(movedoffshore2005‐2007).Inournegativebinomialpatentcountmodels,theselatemoversaremodeledthenasbeingessentiallyonshoreforourfullobservationperiod(1992to2006).Inanattempttoovercomethislimitationandextendourobservationperiod,weexploredusingpatentapplicationsfromtheUSPTOdatabase,inadditiontograntedpatents,torepresentfirms’innovativeoutcomes.Ourinitialhopewastoincludetheapplicationcountsasapredictorvariable(foryearspriorto2006)orasapredictionforthepatentcountsfrom2007to2010(byusingtheratioofobservedapplicationstograntedpatentspriorto2006).Regretfully,wefoundmanychallengeswhenusingtheUSPTOpublishedapplicationsdata.First,applicationdataisonlyavailableontheonlineUSPTOdatabasefromMarch2001tothepresent.Second,firmshavetherighttoasktheUSPTOtonottopublishtheirapplication.Third,ifafirmdoesnotrequestthattheapplicationnotbepublished,unlesstheyspecificallyrequestthepatentapplicationbepublishedearlier,theapplicationisonlypublishedaftereighteenmonths.Fourth,ifafirm’spatentisgrantedpriortothis18‐monthperiod,evenifthefirmallowedtheapplicationtobepublished,thepatentwillgodirectlytobeingpublishedasagrantedpatentandnotshowupintheapplicationdata.Giventheabove,wewouldexpecttheavailableapplicationdatatoundercountthetruenumberofpatentsgranted(orgoingtobegranted).Indeed,usingtheapplicationdatafrom2001to2006,wefoundthat15ofour28firmshave,inoneormoreyears,aratioofgrantedpatentstopublishedapplicationslargerthanone,suggestingthatmanyoftheirpatentapplicationsforonereasonortheotherwereneverpublished.Wealsofoundthattheseratiosareverydifferentbyfirmandbyyear;thefirm’saveragesix‐yearratiosrangesfrom0to21.33(standarddeviation:0.03to12.07).Tofurtherquantifytheappropriatenessofusingthegrantedpatentstoapplicationsratiotopredictpatentcounts,wecalculatedtheprobabilityofobservingthe2001‐2006patentcountsbyfirmassumingaPoissondistributionwithameanvectorofthefirm’spredictedpatentcounts(usingtheapplicationdata).Weexploredusingratiosbasedonintegratedandnon‐integrated1patent/applicationcountsatthreedifferentlevels(firm,optoelectronicindustryandtheUSPTO).Whenlookingatnon‐integratedpatents,theprobabilityofobservingthe2001‐2006countsgiventhefirm‐levelratiorangesfrom0to0.04(0for21of28firms).Theprobabilityofobservingthe2001‐2006countsgiventheindustry‐levelratiorangesfrom0to2E‐9;usingtheUSPTO‐levelratio,theprobabilitiesrangefrom0to0.02.Again,fortheselatertwolevelsofnon‐integratedpatentratios,wefoundaratioof0for21of28firms.Formonolithicpatents,theprobabilitiesforallthreeratios(firm‐level,industry‐level,andUSPTO‐level)rangefrom0to1(0for19of28firms).Thefirmswithprobabilityoneofobservingtheir2001‐2006countsarethosewhohavenotpatentedinmonolithicintegration.Soalthoughwemaybeabletopredictthepatentcountsofafewspecialcases(i.e.thosewhohaveneverpatentedinanarea),ingeneral,weareunabletosuccessfullypredictthepatentcountsusingavailableapplicationdata.Moreover,theaboveanalysispresumesthatthe2001‐2006ratiosmightbereasonablepredictiontoolsforpatentcountspre‐2006or2007‐2010.

Appendix1

Evenwithin2001‐2006,wesawawidevariationingrantedpatenttoapplicationratios.Thiswidevariationmightbeattributabletodifferentfirmapplicationpublicationstrategiesovertime,particularlyifthefirmshavemovedoffshore.Usingthesameratioofgrantedpatentstoapplicationfrom2001‐2006topredictgrantedpatentseitherpre‐2006or2007‐2010inthepresenceofpublicationstrategychangescouldpotentiallybeverymisleading.Thus,weconcludethecurrentavailableapplicationdatafromtheUSPTOdatabase,unfortunately,isnotsuitabletoextendthelengthofourobservationperiod.

Appendix2

Appendix 2: RobustAnalysisofRegressionModels

TableA2.1NegativeBinomialRegressionswithInfineraFixedEffect

Monolithic Hybrid Non‐integrated

Assembly Only Offshoring y‐1 ‐0.18

(0.23)

0.77 **

(0.26)

0.71 ***

(0.18)

Both Offshoring y‐1 ‐0.41 **

(0.13)

‐0.01

(0.09)

‐0.02

(0.06)

Revenue y ‐6.1e‐11

(1.8e‐10)

1.9e‐10

(1.5e‐10)

2.4e‐10 *

(1.0e‐10)

R&D Spending y‐1 1.5e‐09

(1.2e‐09)

‐4.1e‐11

(1.3e‐09)

4.1e‐10

(7.1e‐10)

U.S. Int. Patent y 0.009

(0.006)

0.03 ***

(0.006)

U.S. Non‐int. Patent y 7.6e‐04 ***

(1.3e‐04)

Telecom bubbley‐1 ‐0.18

(0.45)

‐0.84

(0.60)

‐0.55 *

(0.60)

Infinera 3.08 ***

(0.35)

‐35.39 ***

(1.47)

‐0.12

(0.21)

Constant ‐1.63. (0.83)

‐4.22 ***

(0.69)

‐1.11 .

(0.61)

N= 152 152 152

AIC 313.51 247.03 1095.8

Robust std. errors are in parentheses


Appendix2

TableA2.2NegativeBinomialRegressionswithFirmFixedEffect

Monolithic Hybrid Non‐integrated

Assembly Only

Offshoring y‐1

‐0.37

(0.32)

0.30

(0.24)

0.07

(0.14)

Both Offshoring y‐1 ‐0.03

(0.26)

0.15

(0.21)

‐0.08

(0.08)

Revenue y 2.5e‐10

(2.5e‐10)

2.6e‐10

(2.3e‐10)

1.1e‐10

(1.2e‐10)

R&D Spending y‐1 1.3e‐09

(1.88e‐09)

2.2e‐09

(1.9e‐09)

‐1.9e‐09 *

(9.5e‐10)

U.S. Int. Patent y 0.009 *

(0.006)

0.019 **

(0.007)


(1.1e‐04)

Telecom bubbley‐1 ‐0.01

(0.55)

‐0.53

(0.60)

‐0.14

(0.21)

Agere

N= 8

‐2.78 *

(1.42)

‐5.34 ***

(1.55)

‐0.54

(0.63)

Agilent

N= 10

‐4.55

(2..93)

‐6.78 *

(2.87)

1.21

(1.16)

AFOP

N= 9

‐38.12

(2.2e07) ‐4.60 **

(1.46)

‐2.62 ***

(0.66)

Avanex

N= 8

‐1.06

(0.82)

‐4.04 ***

(1.09)

‐1.59 **

(0.58)

Bookham

N= 11

‐0.43

(0.66)

‐3.34 ***

(0.92)

‐2.16 ***

(0.53)

Cyoptics

N= 7

‐38.40

(2.5e07)

‐3.84 ***

(1.13)

‐5.58 ***

(0.81)

Emcore

N= 11

‐2.52 **

(0.88)

‐2.86 **

(0.87)

‐3.02***

(0.56)

Finisar

N= 10

‐1.39 .

(0.82)

‐2.27 *

(0.87)

‐0.70

(0.55)

Infinera

N= 2

1.20

(0.82)

‐39.18

(4.7e07)

‐2.18 **

(0.72)

JDSU

N= 12

‐1.61 *

(0.72)

‐3.27 ***

(0.86)

0.90 .

(0.49)

LSI

N= 15

‐3.89 ***

(1.03)

‐6.01 ***

(1.34)

‐2.16 ***

(0.45)

NeoPhotonics

N=1

‐38.26

(6.7e07)

‐39.05

(6.7e07)

‐3.78 **

(1.29)

New Focus

N= 5

‐38.06

(3.0e07)

‐4.13 **

(1.35)

‐2.32 ***

(0.64)

Appendix2

Oplink

N= 9

‐3.45 **

(1.25)

‐4.79 ***

(1.32)

‐2.60 ***

(0.58)

Opnext

N=3

‐38.41

(3.9e07)

‐39.34

(3.8e07)

‐4.06 ***

(0.80)

OCP

N=10

‐2.88 **

(1.01)

‐3.92***

(1.07)

‐3.45 ***

(0.59)

SDL

N=7

‐1.12 .

(0.69)

‐38.21

(2.5e07)

‐0.35

(0.45)

TriQuint

N=14

‐2.21 **

(0.83)

‐3.85 ***

(0.99)

‐3.14 ***

(0.56 )

Obs. N= 152 152 152

AIC 297.66 255.49 993.61

Std. errors are in parentheses


Appendix2

TableA2.3NegativeBinomialRegressionsonFocusFirmswithStrategyFixedEffects(withcontrolsforthetelecommunicationsbubble)


Assembly Only

Offshoring y‐1

‐0.41

(0.33)

0.69 *

(0.29)

0.40 .

(0.21) Both Offshoring

y‐1 ‐0.51 **

(0.18)

‐0.09

(0.10)

‐0.12

(0.08) Revenue

y ‐4.2e‐11

(2.1e‐10)

1.4e‐10

(1.5e‐10)

3.1e‐10 **

(1.1e‐10) R&D Spending

y‐1 1.35e‐09

(1.59e‐09)

‐1.5e‐10

(1.2e‐09)

‐2.4e‐10

(7.5e‐10) U.S. Int. Patent

y 0.004 .

(0.007)

0.02 ***

(0.006)


(9.7e‐05) Telecom bubble

y‐1 0.59

(0.68)

‐0.78

(0.54)

‐0.007

(0.24)

Onshore Strategies ‐0.64

(0.91)

‐5.48 ***

(1.33)

‐1.95 ***

(0.47)

All‐offshore Strategies ‐1.31

(0.81)

‐3.55 ***

(0.74)

‐0.95 .

(0.50)

Hedger Strategies ‐1.19

(0.75)

‐4.07 ***

(0.88)

‐0.98 .

(0.52)

Late‐offshore

Strategies

‐1.25

(1.15)

‐3.79 ***

(0.78)

‐2.16 ***

(0.58)

Early Exiters ‐1.38

(0.86)

‐4.89 ***

(1.44)

‐0.15

(0.46)

N= 152 152 152

AIC 331.67 248.13 1059.1

Robuststandarderrorsareinparentheses.Signif.codes:0‘***’0.001‘**’0.01‘*’p<0.05‘.’P<0.1‘’1

Appendix2

TableA2.4NegativeBinomialRegressionsonFocusFirmswithStrategyFixedEffects(withoutcontrolsforthetelecommunicationsbubble)


Assembly Only

Offshoring y‐1

‐0.30

(0.28)

0.49 .

(0.29)

0.40 *

(0.20) Both Offshoring

y‐1 ‐0.45 **

(0.16)

‐0.16 .

(0.10)

‐0.12

(0.07) Revenue

y ‐8.2e‐11

(1.9e‐10)

1.8e‐10

(1.56e‐10)

3.1e‐10 **

(9.9e‐11) R&D Spending

y‐1 1.5e‐09

(1.5e‐09)

‐1.4e‐10

(1.3e‐09)

‐2.4e‐10

(6.7e‐10) U.S. Int. Patent

y 0.008 .

(0.007)

0.02 ***

(0.004)


(1.0e‐04) Onshore Strategies ‐0.74

(0.85)

‐5.17 ***

(1.30)

‐1.94 ***

(0.49)

All‐offshore Strategies ‐1.66 *

(0.82)

‐3.03 ***

(0.65)

‐0.95 .

(0.52)

Hedger Strategies ‐1.50 .

(0.78)

‐3.62 ***

(0.85)

‐0.98 .

(0.55)

Late‐offshore

Strategies

‐1.54

(1.24)

‐3.53 ***

(0.68)

‐2.16 ***

(0.59)

Early Exiters ‐1.68 .

(0.86)

‐4.37 **

(1.21)

‐0.15

(0.48)

N= 152 152 152

AIC 330.4 248.16 1061.5

Robuststandarderrorsareinparentheses.Signif.codes:0‘***’0.001‘**’0.01‘*’p<0.05‘.’P<0.1‘’1

Appendix3

TableA3.1ListofFocusFirmsRankedbyDifferentVariables

Rank Revenue2005 RDSpen2005 Integrated Monolithic Hybrid Non-integrated Integrated/ All patents (%)

Monolithic/ All patents (%)

1 Agilent Agilent Infinera Infinera Finisar Agilent Infinera Infinera 2 LSI Agere JDSU Bookham Avago Finisar NeoPhotonics(2nd) NeoPhotonics 3 Agere LSI Finisar JDSU Agilent JDSU Optronx (2nd) Lightwave 4 JDSU JDSU Bookham Lightwave JDSU Avago Lightwave LNL 5 TriQuint Finisar Agilent Avanex Emcore Agere CyOptics (5th) Kotura 6 Finisar TriQuint Lightwave Agere (6th) Lightwave Avanex LNL (5th) Bookham 7 Bookham Bookham Avago NeoPhotonics(6th) Bookham SDL Kotura Picolight 8 Avanex Opnext Agere Agilent Agere Bookham Prima Luci Prima Luci 9 Opnext Avanex Avanex Finisar Teraconnect Axsun Teraconnect Xponent

10 Emcore Infinera Emcore Kotura Infinera (10th) Oplink Bookham TriQuint 11 OCP Emcore NeoPhotonics Picolight (11th) Optronx (10th) LSI Emcore (11th) Avanex 12 NeoPhotonics OCP TriQuint SDL (11th ) Prima Luci(10th) Emcore Opnext (11th) OCP 13 Oplink NeoPhotonics Kotura TriQuint (11th) TriQuint (10th) AFOP (13th) Picolight (13th) Agere 14 CyOptics CyOptics Prima Luci Emcore (14th) AFOP (14th) TriQuint (13th) TriQuint (13th) Emcore 15 AFOP Oplink SDL LNL (14th) Avanex (14th) New Focus Xponent LSI 16 Infinera AFOP LNL (16th) Prima Luci (14th) Axsun (14th) OCP OCP JDSU 17 OCP (16th) Xponent (14th) CyOptics (14th) Infinera Agere (17th) SDL 18 Optronx (16th) Avago (18th) Kotura (14th) Xponent Avanex (17th) Finisar 19 Picolight(16th) LSI (18th) Opnext (14th) Picolight JDSU Oplink 20 Teraconnect(16th) OCP (18th) OCP (14th) Lightwave Finisar Agilent 21 Xponent (16th) Oplink (21st) LNL (21st) Prima Luci LSI Avago 22 LSI Optronx (22Nd) LSI (21st) Kotura (22nd) Avago Optronx (22Nd) 23 AFOP (23rd) AFOP (22Nd) New Focus(21st) Teraconnect(22nd) AFOP AFOP (22Nd) 24 Axsun (23rd) Axsun (22Nd) Oplink (21st) Opnext SDL Axsun (22Nd) 25 CyOptics (23rd) CyOptics (22Nd) Xponent (21st) NeoPhotonics Agilent CyOptics

(22Nd) 26 Oplink (23rd) New Focus (22Nd) NeoPhotonics(26th) LNL Oplink New Focus

(22Nd) 27 Opnext (23rd) Opnext (22Nd) Picolight(26th) CyOptics (27th) Axsun Opnext (22Nd) 28 New Focus Teraconnect

(22Nd) SDL(26th) Optronx (27th) New Focus Teraconnect

(22Nd)

Appendix4

Appendix 4: Firm Case Study

Thestrongeffectofindividualfirmstrategyonpatentingfoundinthefixedeffectsregressionanalysisarguesforacase‐studyapproachtofurtherunderstandtheindividualstrategiesofeachfirm.Incasestudyresearch,wearenotlimitedtohavingthesamedataforeachfirm,andcanthusleverageabroaderamountandmoreheterogeneoussourcesofdataininformingourconclusions.Inassessingtheindividualfirmstrategieswecategorizetheminto5groups:theall‐onshorestrategists,thealloffshorestrategists,thesplitoffshorestrategists,thelateoffshorers,andtheearlyexiters.

1. The All‐Onshore Strategiests

1.1. The On‐shore Integrator: Infinera

Infineraisthetopintegratedpatentproducer(15%ofallintegratedpatentsinoursample,and23%ofallmonolithicallyintegratedpatentsinoursample)andoneoftheonlytwofirms1forwhomover50%ofitspatentsaremonolithicallyintegrated.Infinerawasfoundedin2000andstayson‐shore.Fromfigure9inAppendix6,wecanseeInfinerastartedtheirinnovationactivitieswithalmostequaleffortsonintegratedandnon‐integrateddesign(5integratedvs.6non‐integrated)in2002.Ofthese5integratedpatents,oneofthemishybriddesignandfourmonolithic.After2003,Infinerashiftsitsfocustomonolithicallyintegrateddesignsandallitsintegratedpatents,whicharetwicethenumberofitsnon‐integratedpatents,aremonolithic.

1.2. The On‐Shore Non‐Integrators: LSI, Axsun, Xponent, Prima Luci

Fourfirms,LSI,Axsun,Xpnent,andPrimaLuci,stayentirelyon‐shore;however,differentfromInfinera,theirpatentingisfocusedonnon‐integratedorhybriddesigns.Noneofthesefirmsareinthetop‐ten,bymagnitude,ofintegrated(monolithicorhybridorboth)patentors.Futureworkshouldmorecarefullyexploreexactlywhatoptoelectroniccomponentsthesefirmsmakeforthetelecommunicationmarket,andhowtheycompete.

2. The All‐Offshore Strategists

OppositetothefirmsthatkeepallproductionintheU.S.,anothersetoffirmsmoveallproductionoverseas.ThesefirmsincludeFinisar,Agilent(Avago),Agere,Oplink,AFOP.Finisarmovedassemblymanufacturingfirstandthenfabricationmanufacturingoffshore.Theotherfirmsmovedassemblyandfabricationoffshoreinthesameyear.

2.1. The Incremental Offshorer: Finisar

LikeInfinera,Finisarisoneofthetopintegratedpatentproducers;however,incontrasttoInfinera,Finisarfocusesitsintegratedinnovationsonhybriddesigns2.Finisarisalsothesecond‐highestnon‐

1AnotherisNeoPhotonics.NeoPhotonicshashalfofitspatentsintegratedandhalfnon‐integrated.Alloftheintegratedpatentsaremonolithic.Itmovedoffshorein2005andseemstoshifttheirfocustonon‐integrateddesign,whichweneedmoredataafter2005toprove.(SeeAppendix6)2Fourpercent(20)ofFinisar’stotalpatents(507)areintegratedand70%ofitsintegratedpatentsarehybrid.

Appendix4

integratedpatentproduceramongourfocusfirms.ThereisaninterestingrelationshipbetweenFinisar’sinnovationactivitiesanditsmanufacturinglocations.Afterthetelecombubble(2000),Finisarstartedtogooffshorein2001.Itfirstmovedonlyitsassemblymanufacturingfollowedbyfabricationmanufacturingin2003.Priortomovingitsfabricationoffshore,Finisarkeptincreasingitsnon‐integratedpatentsandworkingonbothmonolithicandhybriddesigns3.After2003,Finisardecreasesitsnon‐integratedpatentingandreduceitsmonolithicproductivity;duringthesametime,however,itsratioofhybridpatentstototaloptoelectronicpatent(monolithic,hybridandnon‐integrated)increases.4(SeeAppendix6)

2.2. The All‐At‐Once Offshorers: Agilent(Avago), Agere, Oplink, AFOP

Severalfirms(Agilent(Avago)5,Agere,Oplink,AFOP)alsomoveoffshore,but,unlikeFinisar,movetheirfabricationandassemblyoffshoreinthesameyear.Thefirsttwo,AgilentandAgere,havehighrevenues(withinTop3)6andhighnon‐integratedpatentproductivity(withinTop5).Thelasttwo,OplinkandAFOP,havelowerrevenues.Allofthemfocusonnon‐integrateddesignsandiftheyhaveintegrateddesignpatentsthesepatents,exceptforinthecaseofAgere7,focusmoreonhybridtechnologies.InthecaseofAgere,94%ofitspatentsareinnon‐integration;however,withintheintegratedpatentsithas,themajorityismonolithic.AfterAgerewentoffshore,itonlyhadtwointegratedpatents(onehybridonemonolithic)in2003andnoneinthefollowingyears.SimilartoAgere,manyoftheotherfirmsreturntoexclusivelypatentinginnon‐integratedtechnologiesaftertryingbrieflytodointegration(intheircaseslargelyinhybridtechnologies).Overall,theearlierthesefirmsmovedoffshore,thelowerpercentageofmonolithicpatentstheyhad8.

3. Hedging Bets – The Split Offshore Strategists

Inadditiontothegroupswhoareentirelyoffshoreorentirelyon‐shore,thereisanothergroupoffirms–thosethatkeeptheirfabricationon‐shoreandmovetheirassemblyoverseas.ThissetincludesJDSU,TriQuint,AvanexandPicolight.JDSUandTriQuinthavemoreresources(beingthefourthandfifth‐highestrevenueearners).AvanexandPicolightcomparedtoJDSUandTriQuint,

3Before2003,Finisarhavehalfofitsintegratedpatentinhybridtechnologyandhalfinmonolithic.4Finsiar’snon‐integratedpatentsdecreasedfrom127to111to106to39(96%,97%,95%,91%ofitstotaloptoelectronicpatentsrespectively)from2003to2006;itshybridintegratedpatentsincreasedfrom2to3to6to3(40%,75%,100%,75%ofitsintegratepatents;2%,3%,6%7%ofitstotaloptoelectronicpatents)from2003to2006;itsmonolithicintegratedpatentsdecreasedfrom3to1to0to1(60%,25%,0%,25%ofitsintegratepatents;2%,1%,0%,2%ofitstotaloptoelectronicpatents)5WeconsiderAgilentandAvagotogetherbecauseAvagowasdivestedfromAgilentin2005.Originally,AgilentwasdivestedfromHPin1999.6Thefinancialdata,i.e.revenueandR&D,inthefirmcasestudiesareallon2005base.7Here,wedonotconsiderOplink,thoughtheratioofitshybridtomonolithicdesignis1:1,becauseitonlyhasonepatentforeachtypeofintegration.Thenumberofitsintegrateddesignistoofewtoobservethepreferenceforhybridormonolithic.8AFOPmovedoffshorein1997andhad0%ofitspatentsaremonolithic;Agilent(Avago)movedoffshorein1999andthepercentageis0.9%;Oplinkmovedin2001andthepercentageis1.1%;Ageremovedoffshorein2002andthepercentageis4.1%

Appendix4

haverelativelysmallrevenues910.Notably,AvanexmergedwithBookhamin2008;PicolightisaprivatefirmandwasacquiredbyJDSUin2007.

3.1. The Rich Hedgers: JDSU, TriQuint

JDSUandTriQuintbothfocusonnon‐integratedpatentsbutalsoworkonintegrateddesignwithapreferenceformonolithictechnology.JDSUatfirstpatentedonlyinnon‐integrateddesigns.After1994,itstartedalsotopatentinhybriddesigns,andoneyearlateritextendeditsinnovationtomonolithicdesigns.Overall,JDSUfocusesmostonnon‐integratedpatents11butbymagnitudehasthesecondhighestnumberofintegratedpatentsinourfocusgroupandthethirdhighestmonolithic(60%ofJDSU’sintegratedpatentsaremonolithic)12.AfterJDSUmoveassemblyoffshoredin2002,thepercentageofnon‐integratedpatentsdecreasedslightly;thepercentageofhybriddesignincreasedandthatofmonolithicdesignalsoincreased.Thenin2005,JDSU’spatentinginallareasbeginstodrop.TriQuint,whichissmallerthanJDSU13,alsomainlyworkedonnon‐integrateddesigns14.Bypuremagnitude,TriQuintisnotasdominantinintegratedpatentingasJDSU–ithasthe12thhighestnumberofintegratedpatentsand11thhighestnumberofmonolithicpatents;however,likeJDSU,morethanhalf(57%)ofTriQuint’sintegratedpatentsaremonolithic.TriQuintmoveassemblyoffshorein2001.Before2001,TriQuintfocusedexclusivelyonnon‐integratedpatents.Between2001and2003,itstartedtoworkonbothmonolithicandhybriddesigns;thenitstoppedworkinginintegration.InthetwoyearsTriQuinthadmonolithicpatents,itsnon‐integratedpatentnumbersdrop.

3.2. The Poor Hedgers: Avanex and Picolight

AvanexandPicolightalsofocusmainlyonnon‐integrateddesignandwhentheypatentinintegrationtheyhavehighlypreferenceformonolithicdesign15.SimilartothefindingsfromTrQuint,intheearlyyearsasAvanexandPicolightincreasedtheirmonolithicpatentstheirnon‐integratedpatentingdecreased.AfterAvanexwentoffshorein200316,itspatentnumbersdroppedexceptthenumberofhybridpatents.However,Avanex’srevenueroseby4timesin2004andheldthisincreasingtendency.Picolightstartedtopatentinnon‐integratedin1995andinintegrated(onlymonolithic)in1998.Afteritwentoffshorein2003,itspatentproductivityinallareasdecreased.

9TheratioofAvanex’srevenuetoJDSU’srevenueis2%to3%from2000to2003,17%and23%in2004and2005.TheratioofAvanex’srevenuetoTriQuint’srevenueis6%to55%duringthe2000s.Picolightisaprivatefirm.Itsannualrevenuesmightbe$30millionto$50million,whichwerebelow5%ofJDSU’srevenueandaround11%ofTriQuint’s.10Matsumoto,C.,JdsuPicksupPicolightfor$115m,http://www.lightreading.com/document.asp?doc_id=118325,accesedon1195%ofJDSU’Stotalpatentsarenon‐integrated.12JDSUfoundedin1999;however,itcamefromamergeroftwofirms,JDSFitelandUniphase.Thatiswhyithadpatents,revenueandR&Dspendingbeforethedocumentedfoundedtime.13Eventhoughtthetwofirmsarebothtop5inrevenue,JDSUearned1.7to8.5timestherevenueofTriQuinteachyear.1488%ofTriQuint’spatentarenon‐integrated.15Avanexhas85%ofintegratedpatentsasmonolithicandPicolighthas100%ofintegratedpatentsasmonolithic.16Avanexmovedassemblyoffshorein2003butitdidnotmoveactiveopticalcomponentuntil2006.

Appendix4

4. The Late Offshorers: Bookham, Kotura, CyOptics, NeoPhotonics, OCP, Emcore, Opnext

Somefirmswentoffshoreafter2005.Inthesecases,welackcompletepatentdatatoshowthechangeintheirpatentingaftertheiroffshoring.Bookham,Kotura,CyopticsandNeoPhotonicswentoffshorein2005;OCPmovedoffshorein2006;EmcoreandOpnextmovedoffshorein2007.Alloftheselateoffshoringfirmskeeptheirfabricationon‐shore,exceptNeoPhotonics.Allfirmsinthisgrouphavethemajorityoftheirpatentsinnon‐integratedtechnologies,exceptNeoPhotonics.Amongthesefirms,onlyBookhamandNeophotonicshave,bymagnitude,alargenumberofintegratedpatents.Here,Bookham,withthe4thlargestnumberofintegratedpatents,the2ndlargestnumberofmonolithicpatents,andthe7thlargestnumberofhybridpatentsismostinteresting.Initsintegratedpatentinghistory,Bookhamatfirstmainlyfocusesonthemonolithicdesign,butstartingin2000itsmonolithicpatentnumbergraduallydecreaseanditstartstopatentinhybriddesigns.17

5. The Early Exiters

Withinourresearchscope,severalfirmsexitthemarketpriorto2005(butallafterthetelecombubble2000).Someofthemwereacquired(SDLandOptronxbyJDSUin2001and2002;LightwavebyNeoPhotonicsin2003;NewFocusbyBookham),andsomeexitedwithoutbeingacquired(LNLin2004,Teraconnectin2003).Inthisgroup,allofthefirmsareprivate,exceptNewFocus.Amongtheacquiredfirms,SDLstaysonshoreand–whileithasmostlynon‐integrationpatents‐‐itsintegratedpatentsarefocusedonmonolithic.Lightwavealsofocusesonintegration,withmorethanhalfofthoseintegratedpatentsbeingmonolithic.Newfocusisfocusedonnon‐integratedpatentsandmovesoffshorebeforebeingacquired.Futureworkshouldseektobetterunderstandwhatresources(patents,offshoremanufacturingcapabilities,orotherwise)JDSUandBookhamwereseekinginacquiringeachofthesefirms.Bothofthenon‐acquiredexitershadlowernumbersofpatentsthanmostoftheotherfocusfirms.

6. Case Study Summary

Ourin‐depthcasestudiesofthesefirmsdepictsalandscapeoffirmswithvastlydifferentresourcesandstrategies:Infinerastaysonshoreanddominatesinthemostadvancedmonolithicallyintegratedtechnologies,whilethefirmsthatmoveoffshoreeitherfromthestartlackormoveoutofthemostadvancedmonolithicallyintegratedtechnologies.ThesefindingsbroadlysupporttheproductioneconomicspublishedbyFuchsandKirchain2010.Perhapsmostinteresting,however,arethesplitstrategists.AsissuggestedinFuchsandKirchain2010,herethelarge‐resourcefirmsthattakeasplitstrategysurvivewhilethelower‐resourcefirmswithsplitstrategiesareeventuallyacquired.FuchsandKirchain,however,suggestthatthisstrategymaybeuntenableevenforthelarge‐resourcefirms,duetothechallengesofmaintainingastrategywheresuccessinintegrationR&DintheU.S.meansthedeclineofassemblyneedsindevelopingEastAsiaandviceversa.Inourdatato‐date,JDSUappearstosustainbothactivities,althoughitsmonolithicintegrationactivitiesdeclineinourfinalyearofgooddata.InthecaseofTriQuint,whichstartedwithlesscomparative17Kotura,Cyoptics,NeoPhotonics,andOpnextareprivateandtheyhaveasmallertotalnumberofpatentsthanmostofotherfirmsinourfocusgroup.OCPonlyhad3integratedpatents;thelowestacceptabletomakeitintoourfocusfirmscope.

Appendix4

advantageinmonolithicintegrationamongourfocusfirms,iteventuallygetsoutofintegrationpatenting(monolithicandhybrid).ItremainstobeseenhowJDSU’monolithicintegrationactivitieswillfareinthelongterm.

Appendix5

Appendi

Tabl

Table

5

ix5T‐testbe

leA5.1T‐testb

A5.2T‐testbe

etweengrou

betweendiffer

etweendifferen

upsofIndiv

rentgroupsofi

ntgroupsofin

vidualInven

inventors’rati

nventors'ratio

tors

iosofeventual

oofeventually

llyleavingfor

leavingforou

onshorefirms

ut‐of‐focusfirm

s

ms

Appendix5

TableA5.

5

3T‐testbetweeendifferentgroupsofinvenfntors'ratioofpfocusfirms

patentinginmmonolithicafterrleavingforouut‐of‐