Globalization of Software

Development

Ashish Arora
Carnegie Mellon University

Matej Drev
Carnegie Mellon University

Chris Forman
Carnegie Mellon University

1
Motivation
 Rate of globalization of software services has
been high
 Recent field study work has suggested more
software R&D activities are taking place abroad
 India: Growth of innovative sector of small niche
companies, particularly in embedded systems (e.g.,
Athreye 2005, Arora 2006)
 Ireland, China: Similar trends in increasing inventive
activity (Sands 2005, Tschang and Xue 2005)
 Brazil, Israel: Long have had product-based software
industry (Breznitz 2005, Botelho, Stefanuto, and Veloso
2005 )
 Need systematic measure of inventive activity
2
What we do
 Examine geographic distribution of the site
of software patenting activity worldwide
(using inventor location)
 Examine extent to which patents invented
abroad are assigned to US MNCs
 Decompose software patenting activity by
“industry”
 This is not easy because of way USPTO
classifies patents
 Seek to examine how user innovation shapes
location of inventive activity
3
What we find
 Site of inventive activity in software
continues to be concentrated in US and
predominately assigned to US firms
 Though an increasing share of patents
invented abroad are assigned to US firms, little
evidence of shift to offshore location for
patents assigned to US firms
 Inventive activity in software outside of
the US tends to be more concentrated in
software whose development requires less
interaction with users
4
Use of US patent data to measure location of
inventive activity
 What exactly is a software patent? Since software
is often an input into other inventions, it is found
throughout the USPTO classification system
 Several methods have been proposed to identify
software patents
 Classification-based (Graham and Mowery 2003, 2005;
Hall and MacGarvie 2006)
 Key word based (Bessen and Hunt 2004)
 Intersection of two (Cockburn and MacGarvie 2006; Hall
and MacGarvie 2006)
 We use a combination of methods

5
 Inventive activity in software continues
to be concentrated in the US
5000
Number of Patents

4000

3000

2000

1000

0
1988 1992 1996 2000 2004
Year
Japan Other G-7
All Others US

6
US Software Patents invented in US and Other Countries
Source: USPTO data and author’s calculations
 Few software patents have been granted to
inventors in the software underdogs
Number of Patents

100

80

60

40

20

0
1988 1992 1996 2000 2004
Year
Brazil India
China Ireland
Israel

Number of US Software Patents Invented in Underdog Countries

7

Source: USPTO data and author’s calculations

 More software patents in the Asian Tigers,
but many are electronics-related
Number of Patents

150

100

50

0
1988 1992 1996 2000 2004
Year
Korea Taiwan
Singapore Hong Kong

Number of US Software Patents Invented in East Asian Tigers

Source: USPTO data and author’s calculations

 An increasing share of patents invented in
Percentage of Total Patents
underdogs are assigned to US firms
100
80
60
40
Percentage of Total Patents

20
0
1988 1992 1996 2000 2004
Year

100
80
60
40
20
0
1988 1992 1996 2000 2004
Year
Asian Tigers Other G-7
Underdogs All Others

Percent of Patents Assigned to Firms in Region for US Software Patents

Invented in Software Underdogs (Top Panel US; Bottom Panel All Others)
9

Source: USPTO data and author’s calculations

How does globalization of software
invention vary by industry?
 Some dimensions that may influence
where to locate inventive activity in
software
 Distance to technical frontier: Proximity to US
universities and highly skilled software labor
(e.g., Thursby and Thursby 2006)
 Proximity to lead users: Transition of new
inventions to usable economic products often
is a difficult process, often requires user inputs
(e.g., Rosenberg 1963; Rosenberg 1983)

10
Importance of lead user innovation
varies by software industry
 User co-invention has been found to be
particularly important to enterprise
software that is embedded in business
processes (Bresnahan and Greenstein
1996)
 Likely less important for new software
tools and technical enhancements that are
not connected with business processes

11
Classifying software patents
 Existing classification systems (e.g., USPTO, IPC)
are based upon technological rather than market
differences
 Need to derive independent classification system
based upon software markets
 We use text mining to perform this classification
 Use features (words) of patents to group similar patents
together
 Cockburn and MacGarvie (2006) develop another
classification system based on USPTO classes and
citation patterns
 Our results are robust to use of their classification
system
12
Use of text mining to classify software
patents
 Use industry classification system in Corptech
database of technology companies
 e.g., artificial intelligence, warehousing/distribution,
transportation software, utility software
 Aggregated some classes together which were small and
similar
 Develop training set of patents for which we
know the classification
 Patents in single industry firms
 Feature selection: words in patent abstract and
title (also experimented with description, claims)
 Also used USPTO and IPC classes as features
13
 Percent of all software patents by
category
Perc ent Patents by Sof tw are Category , Worldw ide

60.00% 56.87%

50.00%

40.00%

30.00% 25.67%

20.00% 17.46%

10.00%

0.00%
Bus ines s Sof tw are Other Tec hnic al Sof tw are

Percent of Patents Invented in Each Software Category Worldwide,

1989-2005 14

Source: USPTO data and author’s calculations

 Software Underdogs are less likely to
patent in enterprise software
Relative Contribution to Innovation Input in Software Underdogs vs. U.S.

120.00%
108.93%

100.00% 93.48%
% Sof.Und. / % U.S.

80.48%
80.00%

60.00%

40.00%

20.00%

0.00%
Business Software Other Technical Software

Percent of Patents Invented in Each Software Category in

Underdogs, normalized by comparable US percentage, 1989-200515

Source: USPTO data and author’s calculations

 However, there is variation within the
underdogs: Israel
Relative Contribution to Innovation Input in Israel vs. U.S.

120.00% 113.50%

100.00%
86.55%
% Sof.Und. / % U.S.

80.00% 75.81%

60.00%

40.00%

20.00%

0.00%
Business Software Other Technical Software

Percent of Patents Invented in Each Software Category in Israel,

normalized by comparable US percentage, 1989-2005 16

Source: USPTO data and author’s calculations

 Variation within the underdogs: Ireland
Relative Contribution to Innovation Input in Ireland vs. U.S.

140.00%
120.11%
120.00%

100.00%
% Sof.Und. / % U.S.

81.06%
80.00%
68.33%

60.00%

40.00%

20.00%

0.00%
Business Software Other Technical Software

Percent of Patents Invented in Each Software Category in Ireland,

normalized by comparable US percentage, 1989-2005 17

Source: USPTO data and author’s calculations

 Variation within the underdogs: China
Relative Contribution to Innovation Input in China vs. U.S.

110.00%

104.31%
105.00%
101.53%
% Sof.Und. / % U.S.

100.00%

95.00%

90.00% 88.68%

85.00%

80.00%
Business Softw are Other Technical Softw are

Percent of Patents Invented in Each Software Category in China,

normalized by comparable US percentage, 1989-2005 18

Source: USPTO data and author’s calculations

 However, there is variation within the
underdogs: India
Relativ e Contribution to Innov ation Input in India v s . U.S.

160.00%
140.87%
140.00%
% Sof.Und. / % U.S.

120.00%
105.69%
100.00%
79.80%
80.00%

60.00%

40.00%

20.00%

0.00%
Bus ines s Sof tw are Other Tec hnic al Sof tw are

Percent of Patents Invented in Each Software Category in India,

normalized by comparable US percentage, 1989-2005 19

Source: USPTO data and author’s calculations

 Asian tigers also predominately patent in
technical areas
Relative Contribution to Innovation Input in Asian Tigers vs. U.S.

140.00%
127.51%

120.00%

100.00%
% Sof.Und. / % U.S.

80.00% 75.11%

60.00% 55.98%

40.00%

20.00%

0.00%
Business Software Other Technical Software

Percent of Patents Invented in Each Software Category in Asian 20

Tigers, normalized by comparable US percentage, 1989-2005
Source: USPTO data and author’s calculations
Other industrialized countries are relatively more
likely to patent in enterprise software
Relative Contribution to Innovation Input in G8 (w/o US) vs. U.S.

140.00%

120.00% 115.14%

101.28%
100.00%
% Sof.Und. / % U.S.

80.00%
65.59%
60.00%

40.00%

20.00%

0.00%
Business Software Other Technical Software

21
Percent of Patents Invented in Each Software Category in non-US
G8, normalized by comparable US percentage, 1989-2005
Conclusions
 Considerable evidence that US continues
to lead in inventive activity in software
 US MNCs account for large and increasing
share of patents invented in software
underdogs
 However, relatively little evidence of shift in
inventive activity assigned to US firms abroad
 Demonstrate that lead user innovation
shapes location of inventive activity
 Inventive activity offshore is primarily in
“technical” software industries
22
Next Steps
 Field work to further understand the
mechanisms driving our results
 Additional empirical analysis of the factors
shaping the site of inventive activity in
software

23
Thanks!

24
Backup Slides

25
 Investigation of Potential Home Country
Bias: European Software Data

European Patent Office Software Patent

Grants by Country of the Assignee and Year
of Application 26
Source: Thoma and Torrisi (2006)