Original citation:

Scott, Susan V., Van Reenen, John and Zachariadis, Markos. (2017) The long-term effect of
digital innovation on bank performance : an empirical study of SWIFT adoption in financial
services. Research Policy, 46 (5). pp. 984-1004.

Permanent WRAP URL:

http://wrap.warwick.ac.uk/88927

Copyright and reuse:

The Warwick Research Archive Portal (WRAP) makes this work of researchers of the
University of Warwick available open access under the following conditions.

This article is made available under the Creative Commons Attribution 4.0 International
license (CC BY 4.0) and may be reused according to the conditions of the license. For more
details see: http://creativecommons.org/licenses/by/4.0/

A note on versions:
The version presented in WRAP is the published version, or, version of record, and may be
cited as it appears here.

For more information, please contact the WRAP Team at: [email protected]

warwick.ac.uk/lib-publications
 Research Policy 46 (2017) 984–1004

Contents lists available at ScienceDirect

Research Policy
journal homepage: www.elsevier.com/locate/respol

The long-term effect of digital innovation on bank performance: An

empirical study of SWIFT adoption in financial services
Susan V. Scott a , John Van Reenen b , Markos Zachariadis c,∗
a
Information Systems & Innovation Group, Department of Management, London School of Economics & Political Science, Houghton Street, London, WC2A
2AE, UK
b
MIT Sloan School of Management and Department of Economics, Massachusetts Institute of Technology, CEP, NBER, CEPR, 77 Massachusetts Avenue,
Cambridge, MA 02139, USA
c
Information Systems & Management Group, Warwick Business School, University of Warwick, Scarman Road, Coventry, CV4 7AL, UK

a r t i c l e i n f o a b s t r a c t

Article history: We examine the impact on bank performance of the adoption of SWIFT, a network-based technological
Received 19 May 2015 infrastructure and set of standards for worldwide interbank telecommunication. We construct a new
Received in revised form 7 March 2017 longitudinal dataset of 6848 banks in 29 countries in Europe and the Americas with the full history of
Accepted 20 March 2017
adoption since SWIFT’s initial operations in 1977. Our results suggest that the adoption of SWIFT (i) has
large effects on profitability in the long-term; (ii) these profitability effects are greater for small than for
JEL classification:
large banks; and (iii) exhibits significant network effects on performance. We use an in-depth field study
O33
to better understand the mechanisms underlying the effects on profitability.
N20
© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
Keywords: (http://creativecommons.org/licenses/by/4.0/).
Technology adoption
Bank performance
Financial services
Network innovation
SWIFT

1. Introduction 2007; Kretschmer, 2012). Can strategic investments in certain

information systems provide better explanations than broad analy-
The study of digital innovations and their adoption by organi- ses of a firm’s aggregate IT investment (Aral and Weill, 2007)? The
zations has generated a field of research whose aim is to unravel need for research here is acute because, practitioners and policy
analytical challenges such as the productivity paradox1 and estab- makers have scarce resources with which to base actions, and schol-
lish an empirical knowledge base upon which scholars can build. ars lack the datasets and foundational knowledge claims about
Our research contributes to this on-going effort by focusing on innovation adoption in financial services with which they can draw
the effect that digital network innovation adoption has over time up agendas for future research.
on bank performance. The financial services sector was an early In an effort to address these challenges, we begin by examin-
adopter of key technologies associated with business transfor- ing the approaches used to-date in studies of digital innovation
mation and it is currently one of the most intensive users of adoption in the financial services. We then add to this emerg-
information and communication technologies (ICT). While atten- ing knowledge base by presenting an analysis of adoption data
tion has been drawn to relatively poor gains from ICT investment from SWIFT, the financial digital network innovation and standards
in the financial services sector, findings from relevant research are developed in the 1970s to serve as the infrastructure for worldwide
inconclusive (Roach, 1991; Haynes and Thompson, 2000; Beccalli, interbank payments communication. We construct and analyse a
new dataset comprising SWIFT’s adoption history from 1977 to
2005 matched to bank-level performance data for the US, Canada
and 27 European countries. Our analysis breaks from the majority
∗ Corresponding author.
of past research by utilizing a distinctive longitudinal approach to
E-mail address: [email protected] (M. Zachariadis).
1 investigate claims in the literature. While there is now evidence of
This trend was appropriately characterized by Robert Solow’s famous quote that
‘you can see the computer age everywhere but in the productivity statistics’ (Solow, productivity benefits from ICT adoption, most prior research has
1987), which eventually became known as the “Productivity Paradox” (Roach, 1991; not considered whether these benefits are sustained in the long
Brynjolfsson, 1993).

http://dx.doi.org/10.1016/j.respol.2017.03.010
0048-7333/© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 985

term. Our research focuses on the following questions. Firstly, is Whilst the Schumpeterian approach is useful in describing the
there evidence that ICT adoption generates long-term benefits for link between technological innovation and organizational per-
firms? Secondly, do these benefits accumulate over time? Thirdly, formance, empirical evidence on the magnitude and nature of
do particular kinds of firms benefit more than others in the long the contribution of technology seem to vary considerably across
term? Fourthly, what are the mechanisms underlying these bene- economies, sectors, and firms prompting much discussion about
fits? different measures of economic performance and innovation. For
Our findings determine the timeframe in which benefits from example, at the macro-level, most studies focus on measures of
digital innovation adoption accrue and establish their correspon- economy-wide productivity and labour productivity growth to
dence with network effects. In so doing, we reveal surprising results make claims regarding the aggregate contribution of technology
concerning the performance of small banks relative to large banks. investment (Brynjolfsson and Yang, 1996). A case in point would
Small and medium enterprises (SMEs) are frequently referred to be Gust and Marquez (2004) who analyse data from 13 OECD coun-
as the ‘backbone’ of the economy because they play an impor- tries between 1993 and 2000 and find that ICT expenditure in
tant role in job creation (Brynjolfsson et al., 1994), technology this period is associated with higher productivity growth. Simi-
investment, and GDP growth (Kuan and Chau, 2001) yet research larly, Oliner and Sichel (2000) demonstrate that ICT capital makes
about the effects of ICT adoption on their economic performance is a significant contribution to the output growth rate of the US econ-
sparse. There is a tendency in the adoption literature to treat small omy (between 0.6% and 1.1%) at various intervals during the period
firms as “scaled-down” replicas of larger businesses (Raymond, 1972–1999. Using data from the UK, Oulton (2002) found evidence
1985; Thong et al., 1996) and generalize about them based on of increased ICT contribution to GDP growth (up to 20.7%) for the
large firm only datasets. We find small firms benefit disproportion- years 1979–1998. Gordon (2016) provides a more sceptical per-
ately from SWIFT which is remarkable as this means overcoming spective on the contribution of ICT to US growth, arguing that
scarce resources including relatively limited knowledge of technol- the main effects were all focused in the short window 1996–2004
ogy management (Pfeiffer, 1992; Grandon and Pearson, 2004). period.
Throughout the paper, we complement the quantitative analy- Using economy-wide data is problematic as it is difficult to con-
sis with an in-depth field study to explore the dynamic interplay trol for many other factors. More recent industry-level studies also
between the process of adopting SWIFT and the mechanisms used found notable returns to ICT investments. Based on an analysis of
to realise benefits from that adoption. We argue that this not only 61 industries in the U.S., Stiroh (2002) uncovered evidence sug-
has implications for how firms can leverage ICT-investments but gesting faster productivity growth – both total factor productivity
also suggests insights into adoption strategies for firms navigating (TFP) and average labour productivity (ALP) – in sectors that pro-
the current business landscape in which potentially value-adding duced or used ICT more intensely. While several other studies have
digital infrastructures are an integral part. In the next section, we reported similar conclusions (e.g. Siegel and Griliches, 1992; Berndt
will review the literature upon which we build our study. and Morrison, 1995; etc.), it is apparent that the degree of the effect
varies considerably between countries and industries. Stiroh (2002)
2. ICT adoption and firm performance found the strongest impact in IT-intensive services whereas oth-
ers have found manufacturing to be more important (Baily, 1986;
In the past, ambiguity concerning the economic impact of infor- Roach, 1991). There has also been much recent work at the firm
mation and communication technology adoption or what has been level. Here, most studies reveal a positive and significant correlation
termed the “productivity paradox” was hotly debated. Initial results between the adoption of ICT and business performance. In a series
during the 1980s and 1990s created concerns about whether ICT of analyses using a large sample of company surveys, Brynjolfsson
had any significant effect on economic output, but over the last and Hitt (1993, 1995, 1996), report that ICT capital generates up
couple of decades evidence has mounted confirming that ICT does to 10 times more output than other forms of capital. Although,
yield sizable economic returns at both macro and micro levels other papers have produced similar results that point to a positive
(Brynjolfsson, 1993; Brynjolfsson and Hitt, 1998; Bloom et al., effect from ICT adoption (Jorgenson and Stiroh, 1995; Oliner and
2012). More specifically, a large number of recent studies report Sichel, 1994; etc.), there is less agreement on the magnitude of the
positive results from ICT investments on a range of measures relat- gains. In a meta-analysis of 20 econometric studies, Stiroh (2002),
ing to financial performance (Aral et al., 2006; Bresnahan et al., reports considerable variation with estimates of ICT-elasticity rang-
2002; Brynjolfsson and Hitt, 1996, 2000, 2003; Dewan and Kraemer, ing from −0.06 to 0.24. Even though these are largely attributed
2000).2 to differences in production function specifications, the estima-
These findings are consistent with the Schumpeterian economic tion techniques, and quality of data used, there are other important
theoretical tradition that recognizes the importance of technolog- dimensions such as the timing and span of the sample period, the
ical change and innovation as being the key drivers of economic ICT-measures used (Evangelista, 2000), and the characteristics of
growth and firm performance (see Romer, 1990; David, 1990; the adopters included in the sample. Such variations in findings
Aghion and Howitt, 2007). In this line of work, technological may also be because different types of technologies are lumped
innovation plays a key role in explaining the dynamic properties together as “ICT capital” or “computers.” (Weill, 1992; Barua et al.,
of organizations (Cainelli et al., 2006). According to Schumpeter 1991).
(1943), innovation puts in motion the mechanism of “creative To move forward the research agenda on technology adoption,
destruction” in which technological advances override pre-existing more detailed empirical data is needed, and Anderson et al. (2006)
market conditions. In the process, firms introduce new products, and Jun (2008) emphasise that this need is particularly urgent in
services and organizational processes thus gaining market share at the financial services sector especially with the current wave of
the expense of their non-innovating competitors. Some are then “fintech” innovation where different technologies can have vari-
able to leverage their new competitive position and gradually ous effects on organisations (Evangelista, 2000). To address this, we
accumulate “monopolistic rents”, increasing their profitability still gather detailed firm-level data that incorporates larger samples of
further (Cainelli et al., 2006). companies across longer periods to account for both sample selec-
tion bias and adjustments that take place over time. The availability
of a long observation window offers us the unique opportunity to
2
For a more detailed review of the literature see surveys by Brynjolfsson and Yang
(1996), and Draca et al. (2007).
986 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

gain valuable information regarding the long-term impact of tech- 2.2. Firm size and technology effectiveness
nological innovation on bank performance.
There has been relatively little econometric research on the
effects of ICT on smaller firms. For example, Brynjolfsson and Hitt
(1996) used data from some of the largest US corporations (367
2.1. Long-term outcomes from technological innovation adoption firms generating approximately $1.8 trillion of gross output annu-
ally). Small organizations possess some unique characteristics that
A key debate on the value of ICT has been the effect of technol- matter significantly when new technology is introduced (Raymond,
ogy on long-term profitability and its capacity to create sustainable 1985; Thong et al., 1996; Kuan and Chau, 2001). For example, the
competitive advantage (Clemons and Row, 1991; Clemons, 1986). relative costs and risks from ICT adoption and implementation
A long standing theoretical claim in this literature asserts that can be considerably higher for smaller firms due to their limited
new technology adoption will offer benefits in terms of enhanced resources and lack of knowledge around technology management
cost efficiencies, better product quality, and increased value to (Pfeiffer, 1992; Grandon and Pearson, 2004).
customers but the economic rents and value realised from these In contrast to positive outcomes relating to the introduction and
benefits will not last long due to the high imitability of ICT. Thus, use of ICT in large organizations, research findings regarding the
the ICT applications adopted by firms have the status of “strate- effect of technology in SMEs have been ambiguous at best. Empiri-
gic necessities” and advantages from their early adoption and use cal evidence from the literature on small business ICT suggests that
are lost through imitation and do not lead to profitability increases a number of factors inhibit the uptake of technological innovation
(Clemons and Kimbrough, 1986; Fuentelsaz et al., 2012; Carr, 2003). and impede the benefits of ICT adoption. These include a vital lack
This hypothesis largely relies on the assumption that ICT is highly of financial resources with which to acquire ICT capital, invest in
commoditized and therefore easily replicable at a low cost (Carr, technological skills and achieve systems integration (Pfeiffer, 1992;
2003). In other words, it is expected that technology will be dif- Grandon and Pearson, 2004; Saunders and Clark, 1992). Similar
fused and adopted homogenously – without ‘frictions’ or delays results are also reported by Cragg and King (1993), who identify
across competitor firms, a claim that is disputed in the technological economic costs and shortage of technical knowledge as key barriers
diffusion literature where there are many factors that can prevent to ICT gains in the context of small organizations. Finally, Ballantine
some firms from speedily adopting a technology (Fuentelsaz et al., et al. (1998) identified distinctive features of SMEs such as narrow
2012). The counter claim holds that there are alternative ways with access to capital supplies, absence of business and IT strategy, and
which organizations incorporate ICT into their productive process, greater emphasis on using technology to automate (Zuboff, 1988).
use complementary assets, or reconsider their business strategy in A more optimistic outlook is given by Dos Santos and Peffers (1995)
light of technological change, which can lead to persistent differ- who found inconclusive results regarding firm size and the impact
ences in performance that cannot be accounted for by the strategic of ICT on market share and income gains. Looking at a sample of
necessity hypothesis (Battisti et al., 2009). banks and the benefits from ATM adoption they conclude that there
To address these fundamental arguments around long-term are no economies of scale or scope for this technology that favour
performance and sustainability we need a longitudinal approach larger institutions in particular, however, they did not find any sig-
which enables us to go beyond the short-term effects of technol- nificant results to suggest that such a technology can specifically
ogy adoption to reveal the varying temporal profile and impact of benefit smaller firms either. A similar view is shared by Lacity et al.
innovation. A few papers have attempted to focus on the long-term (2014) who suggest that certain technologies, for example cloud
effects of ICT using micro data. (e.g. Kwon and Stoneman, 1995, for computing can provide equal benefits to both large and smaller
five manufacturing technologies or Haynes and Thompson (2000) firms albeit in different ways.
on Automated Teller Machine (ATM) networks). Of particular interest for our study are hypotheses that con-
We build our study upon two key insights from this prior lit- tradict the generally accepted view and suggest that small
erature both of which centre on the importance of constructing organizations may hold certain advantages over their larger com-
long lags. Firstly, intra-firm diffusion and technological adapta- petitors. For instance, perhaps smaller enterprises can adapt faster
tion often takes time (Tyre and Orlikowski, 1994; Fuentelsaz et al., to internal and external changes in their operating environment,
2009). Second, there is often the need for significant organizational whereas larger organizations may respond slowly to technologi-
changes and learning (Van de Ven, 1986; Brynjolfsson and Yang, cal transformation due to legacy systems that demand substantial
1996; David, 1990). Studies using short lags are unable to capture modifications (Dos Santon and Peffers, 1995). Evidence of this
potential benefits that may accumulate over time from the tech- would be remarkable because it would mean that small busi-
nology investment. As Fuentelsaz et al. (2012) argue, the uneven nesses achieve benefits from ICT adoption that in the long run
patterns of technological diffusion mean that it is possible for bene- outweigh more obvious big firm advantages such as ample financial
fits accrued by adopters (in comparison to non-adopters) to endure resources, ICT expertise, and economies of scale.
for several years depending on the timing of the diffusion process. The literature on the effects of ICT has shown that firms bene-
Thirdly, beyond firm heterogeneity, particular focus must be given fit mostly from cost reductions due to automation and increase of
to the specific characteristics of network innovations and their efficiencies in the production process and less from an increase in
strategic importance for long-term economic performance. revenue streams. Bigger firms are commonly expected to be more
Network externalities can arise when usage benefits increase efficient and thus anticipate better results in this regard (Hall and
with network size (Katz and Shapiro, 1985; Shapiro and Varian, Weiss, 1967), but as we will go on to argue there is also evidence
1999; Farrell and Saloner, 1992). For example, Saloner and that despite their size smaller businesses may also achieve sig-
Shepard’s (1995) show that as more ATMs are installed, the network nificant leverage from ICT adoption. Indeed, some of the existing
size grows, making it hold higher value for cardholders and banks literature points to the realisation of a range of benefits includ-
because the connectivity produced provides more utility. Although ing operational savings, improvements in business processes, the
there are comparable results in other industries (Economides, cultivation of new markets, higher sales turnover and increases in
1996), empirical work on financial services network effects is in profitability (Currie and Parikh, 2006; Kuan and Chau, 2001).
short supply and very focused on ATMs. This is cause for concern In sum, thus far research examining the impact of ICT adoption
in a sector whose history has been defined by network innovations on smaller firms, including mechanisms of value creation and the
and network platforms are undergoing critical development. benefits generated has remained largely inconclusive. This leaves
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 987

considerable scope for the study that we have undertaken here in scarcity of longitudinal studies examining particular ICT innova-
which we ask not only if there is evidence that ICT adoption gen- tions in financial services. In a survey, Frame and White (2004)
erates long-term benefits for firms but whether particular kinds of could only identify eight studies of which six use the same data
firms – small firms – benefit more than others in the long term. on ATM diffusion (Hannan and McDowell, 1984, 1987; Sinha
and Chandrashekaran, 1992; Saloner and Shepherd, 1995; etc.).
2.3. Financial technology and bank performance Although they represent an important body of research, these
studies focus more on the diffusion of specific innovations and
Traditionally, finance has been the highest spender across all less on their impact upon business performance. Thus, Frame and
sectors. For example, data from the U.S. Bureau of Economic Anal- White (2004) conclude that we have a lot of “talk” about financial
ysis measure computer (OCAM) expenditure in financial services innovation but “little action”. In other words, given the size and
between 32.5%-38.7%, from 1979 to 1992 (cf. Brynjolfsson and importance of the financial sector, the number of relevant studies
Yang, 1996; Griliches, 1995). Recent aggregate figures on technol- is surprisingly limited and further scholarly efforts are needed.
ogy investment worldwide place the banking and securities sector In this paper, we combine insights from several different
at the top of ICT spenders’ list with a total expenditure of $486.28 research approaches. As a result, the approach taken here presents
billion– approximately 18% of the total technology investment or some distinct advantages, for example: firstly it proposes the in-
24.8% if we include the insurance industry.3 Consequently, the depth investigation of a particular ICT-related innovation (SWIFT)
implications of ICT adoption and use for the global financial system in the banking sector rather than examining ‘general purpose tech-
have been fundamental. ICT did not only transform transaction pro- nologies’ or ICT broadly, followed by an econometric analysis on
cesses but is also associated with shifting organizational boundaries the impact of SWIFT of bank performance. Secondly, the span of
(Scott and Walsham, 1998), facilitating the creation of new finan- the data allows us to track the effects of SWIFT adoption in a
cial products, changing the nature of work (Barrett and Walsham large sample of banks (6848 in total) across 29 countries. Previ-
1999), globalizing financial markets (Sassen, 2002; Weber, 1994) ous micro-econometric studies have limited themselves to a single
and restructuring the character of financial intermediation (BIS, country4 even though many industries, such as financial services,
2002). are international in scope. Thirdly, based on the whole popula-
Some qualitative case studies have been used to study the tion of SWIFT adopters we are able to track the long-run effects
effects of ICT on financial performance (e.g. Scott and Barrett, 2005; of adoption (up to 30 years) which is important as the impact of
Clemons and Weber, 1990). For example, Autor et al. (2002) exam- innovation is unlikely to be realised in the short run (Geroski et al.,
ine the introduction of automatic image processing on one of the 1993). Finally, in order to explore the impact of SWIFT adoption on
top 20 US banks, arguing that the introduction of complementary bank performance and the value–creating mechanisms that come
organizational changes were crucial in understanding the impact into play once the technology is implemented – for both small
on performance. In Weill and Olson (1989), the authors use six case and large banks − we draw on insights from previous qualitative
studies to investigate the impact that the level of ICT investment research (Scott and Zachariadis, 2012, 2014). The complementary
has on firm performance. Their results, from a series of interviews data was gathered through archival research as well as interviews
with banking professionals, demonstrate the organizational com- with SWIFT employees, bank executives, and domain experts who
plexities involved in defining ICT and difficulties encountered when described the SWIFT implementation process, its cost and potential
searching for an appropriate measure to estimate the impact of benefits for different kinds of financial services organizations.
technology. Such findings are particularly useful in order to under-
stand the richness of processes and technology strategies in specific
3. Overview of SWIFT and research setting
contexts, but are hard to generalise due to their specific nature.
In terms of econometric studies on ICT in financial services,
Our empirical analysis focuses on the adoption of an ‘inter-bank’
Casolaro and Gobbi (2007) estimate profit and cost functions for
financial telecommunication network called SWIFT. Launched in
a panel of 600 Italian banks 1989–2000 and find that ICT capital
1973, SWIFT’s mission was to facilitate correspondence banking
intensive techniques significantly increase total factor productivity
by automating communication between banks through the intro-
(TFP). Jun (2008), examines findings from several studies showing
duction of machine readable encrypted messaging standards, thus,
a positive relationship between ICT and banking performance, and
enabling banks to send funds directly to counterparts at increased
also presents results indicating that ICT investments are associated
speed, in higher volumes, for reduced cost, and with improved secu-
with higher returns on assets in a sample of 22 South Korean secu-
rity (Winder, 1985). In some regards, SWIFT can be compared to
rities firms. Similarly, Anderson et al. (2006) investigate the value
an electronic data interchange (EDI) or co-operative interorganiza-
implications of ICT investments on a panel of 62 Fortune 100 banks
tional system (IOS) allowing trading partners – in this case financial
and find that firm value increased on average with Y2 K spending
institutions – to “exchange structured business information5 elec-
on technology. Also, Parsons et al. (1993) estimate a cost function
tronically” (Iacovou et al., 1995; p.466). SWIFT’s diffusion began
using data from a single large Canadian bank between 1974 and
with European-based banks and gradually moved to countries such
1987 finding a weak but significant correlation between productiv-
as the US and UK (see Fig. 1).6 In this section, we provide a detailed
ity growth and the use of computers. Finally, Alpar and Kim (1990)
explore the impact of ICT on the production of bank services finding
that technology is cost saving, labour saving and capital using.
4
Although these studies are useful in order to understand the An exception to this is Beccalli (2007), who looks at a total of 737 banks in 5
general effect of ICT, treating technology as a single aggregated European countries, however, the panel data used cover only 5 years in total and
fail to identify any long-term effects which may last up to 10 years.
category makes it hard to disentangle which aspects of ICT led to 5
In the case of SWIFT these are financial messages such as instructions for pay-
performance increases and identify the dynamic effects of tech- ments, confirmations, settlement messages, letters of credit, securities transactions,
nology adoption. As a result, many authors have pointed out the and other types of standardized processes.
6
Fig. 1 presents the accumulative diffusion curve of all SWIFT adopters across
eight of the countries in our sample between 1977 and 2006. Even though Germany
led the way until approx. 1985, the US and UK SWIFT population base grew sub-
3
Source: Gardner (September 2015). Manufacturing and natural resources fol- stantially making these two countries the largest SWIFT adopters. As it can be seen
lowed with 476.55 billion USD. These figures are based on real data and partial the diffusion curve does not seem to follow the traditional “S” shape, which could
projections for 2015, however, ICT expenditure in 2014 follows a similar pattern. suggest that the SWIFT diffusion process has not been completed by any means.
988 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Fig. 1. SWIFT diffusion in eight chosen countries (1977–2006).

overview of SWIFT’s proprietary communications platform: its net- aspects. Potential intangible benefits are related to the reduction of
work, the costs and benefits from its adoption, and its mission. We operational risk and fraud (due to the less manual intervention and
conclude by considering the assumptions made by practitioners in more secure transaction environment), enhancement of customer
financial services about the benefits of SWIFT membership. satisfaction, security and resilience, easier regulatory compliance,
SWIFT has established itself as a trusted third party, functioning greater visibility and control (allowing for better cash manage-
as the core gateway especially for large-value payments. Weirdt ment), reliability and timing. These benefits are difficult to measure
et al. (2005) described it “as an obligatory passage point to other and therefore are not directly dealt with in our study, however, we
parts of the transactional infrastructure, which gives it effective would expect that they can contribute towards profitability.
control of the [global] payment system”. Since its founding, there Probably the most obvious tangible benefit, especially in the
has been a working concord among SWIFT members to support its case of larger financial institutions, is the reduction of operat-
operation as a not-for-profit “industry co-operative”, reinvesting ing expenses. While the implementation of SWIFT can be a costly
any surplus in process and product improvement. During its life- investment it is regarded as having a cost-saving effect. At a basic
time, there have been some business and connectivity ‘solutions’ level, SWIFT replaces direct links to corresponded banks with a
in the tech market that engaged in competition, however they only centralised cloud-based solution that allows member to contact
accounted for a small fraction of business, and did not offer a com- anyone on the network (see Fig. 3). However, it also helps to reduce
parable level of service or global coverage, nor they performed its user’s costs by providing automation (through greater standard-
standards and community development roles. ization and interfacing), security, speed, and economies of scale
As the first network innovation of its kind in financial services, thus reducing marginal costs in the long-term via increases in
SWIFT necessitated the development of messaging standards to labour productivity. These benefits extend across numerous busi-
reduce operational complexity and advancements in network secu- ness processes and transactions commonly used in banking such
rity protocols. Today SWIFT operates a highly reliable and secure IP as payments, confirmations, financial reporting, pre-trade, trade,
network (SIPN) that offers a single window access to the financial and post-trade activities. Interviews with financial services pro-
world and allows for interoperability and high end-to-end automa- fessionals enabled us to document this step-by-step progressive
tion (also known as “Straight-Through-Processing”) through a vast roll-out of SWIFT adoption through each business area. They con-
range of standards, technological applications and connectivity firmed that after the initial investment period, long-term operating
solutions. costs decreased as SWIFT became further integrated into their back-
office automated production systems.
The mutuality generated by SWIFT’s industrial cooperative gov-
3.1. Costs and benefits of SWIFT adoption ernance structure ensured a phase of initial reciprocal adoption by
its founding members with subsequent momentum achieved by
The costs and benefits of adoption for different banks can leveraging the counterpart banking relationship. In other words,
vary significantly which means a thorough analysis of deploy- firms compelled trading partners to connect as a condition of busi-
ment (usage needs, interfaces, standards compliance, etc.) must be ness. Large banks were able to assert new terms of business and
undertaken in the pre-implementation phase (see Fig. 2). The main thus realise the economies of scale promised by SWIFT adoption.
fixed costs relate to the original installation which includes all the Other smaller firms were actively recruited as SWIFT executives
items in the implementation phase. Once SWIFT is up and running realized from the outset that network coverage was vital. Indeed,
(post-implementation stage) there are additional costs associated we documented on-going programmes to connect countries and
with maintenance, fees, training, software and hardware upgrades, enroll the widest possible range of financial institutions in our field
and improvement expenses. Going forward there may be further study.
adjustment costs and subsequent software development to inte- SWIFT delivers very high reliability and is now a core part
grate the system with the internal banking processes. of the largely taken-for-granted international financial services
In contrast to the expenses, benefits are typically not realized information systems infrastructure. Indeed, during our field study
until the infrastructure is properly configured and used. The ben- we frequently heard SWIFT referred to as the “plumbing” of
efits that SWIFT can be distinguished into intangible and tangible
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 989

Fig. 2. SWIFT implementation timeline.

Fig. 3. Value added from SWIFT adoption.

Note: Fig. 3 presents a detailed illustration of the mechanism through which SWIFT adds value in comparison to the old system of communication between banks (telex, FAX,
and other proprietary networks).

international financial services. In other words, financial services 4. Data and methods
professionals had come to regard SWIFT as a utility that fulfilled a
basic but low-value facility. Moreover, its continued drive to reward Our main dataset is the entire population of SWIFT adopters
high volumes led one interviewee to claim that SWIFT was built “by worldwide from 1977 to 2006. This consists of the complete list of
big banks, for big banks”. This value neutral “utility” status and the live SWIFT users operating on their “SwiftNet FIN” (or “SNFIN”) net-
claim that its benefits accrue mainly to large financial firms, rep- work – the most popular service and core SWIFT product – across
resents the dominant ‘industry wisdom.’ Neither claim has been 219 countries and territories. Considering the complexity of the
systematically investigated, however. financial systems around the world and the constraints that are
Challenging these active working assumptions lies beyond the placed from national financial regulatory bodies, we also limited
mandate of the actors involved. SWIFT’s attention is on “lean man- our initial analysis to Europe and the Americas. Since 1977, SNFIN
agement” and their main interest is in the analysis of members’ has been adopted by 3380 banks in the 29 countries of our sample.
usage patterns along functional lines (rather than size of firm). By To this panel, we matched information from Bankscope, a global
definition, SWIFT membership relieves firms of the need to anal- database containing information on more than 28,000 public and
yse core transactional network technology so that they can focus private banks (adopters and non-adopters of SWIFT) around the
on developing other kinds of service innovation. Regulators have world. This is compiled by Bureau van Dijk (BVD), a European elec-
oversight but their attention is on audit, systemic risk and com- tronic publisher of business information. The database combines
pliance. The impact of SWIFT adoption on firm performance thus data from seven sources including Fitch Ratings, Capital Intelli-
remains a blind spot both in the financial services industry and the gence, the Economist Intelligence Unit, Moody’s, Standard and
academic literature. Poor’s etc. It includes all the information in the banks’ published
accounts and is reasonably comprehensive in coverage. The prod-
uct of this merge is a unique dataset containing a large sample
990 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table 1
Country Statistics.

(1) (2) (3) (4) (5) (6)

Country name Sample firms Percent Number of SWIFT Percentage of SWIFT Matched adopters in Proportion (%) of
adopters population sample adopters in sample

Austria 230 3.36 100 1.22 69 30.00

Belgium 98 1.43 88 1.08 43 43.39
Canada 83 1.21 62 0.76 31 37.35
Cyprus 29 0.42 38 0.46 15 51.72
Czech Republic 32 0.47 28 0.34 16 50.00
Denmark 121 1.77 59 0.72 41 33.88
Estonia 10 0.15 13 0.16 7 70.00
Finland 19 0.28 22 0.27 7 36.84
France 468 6.83 250 3.06 118 25.21
Germany 1710 24.97 298 3.65 178 10.40
Greece 23 0.34 41 0.50 19 82.60
Hungary 40 0.58 43 0.53 25 62.50
Ireland 70 1.02 81 0.99 37 52.85
Italy 782 11.42 258 3.16 167 21.35
Latvia 23 0.34 27 0.33 23 100.00
Lithuania 10 0.15 12 0.15 10 100.00
Luxembourg 115 1.68 148 1.81 83 72.17
Malta 14 0.20 15 0.18 9 64.28
Netherlands 101 1.47 98 1.20 45 44.55
Norway 88 1.29 34 0.42 17 19.31
Poland 52 0.76 47 0.57 39 75.00
Portugal 47 0.69 45 0.55 31 65.95
Slovakia 21 0.31 20 0.24 14 66.67
Slovenia 20 0.29 23 0.28 17 85.00
Spain 166 2.42 120 1.47 71 42.77
Sweden 139 2.03 35 0.43 14 9.35
Switzerland 539 7.87 270 3.30 162 30.00
United Kingdom 455 6.64 538 6.58 177 38.90
USA 1343 19.61 567 6.94 204 15.19
Total 6848 100.00 3380 41.34 1689 24.66

Notes: Our sample (first column) includes 6848 firms from 29 countries (adopters & non-adopters). Adoption information is from 1977 to 2006. The third column contains
data on the population of SWIFT adopters in the 29 countries of our sample. Column 5 reports the number of adopters by country that were matched in the sample and
column 6 reports what is the proportion (%) of the matched adopters in the whole sample.

of firm-level longitudinal information on ICT adoption and finan- the unconsolidated accounts if we had their consolidated compan-
cial performance. Our financial data run from 1997 (the first year ions and used unconsolidated accounts of a subsidiary when there
that Bankscope was produced) through 2006, but due to the small were no consolidated companions (results were robust when using
number of observations in years 1997 and 2006, we exclude them only consolidated or only unconsolidated accounts).
from our estimations and exploit the years from 1998 to 2005. After Measuring productivity is extremely challenging in the finan-
cleaning7 we are left with an unbalanced panel of 6848 firms and cial sector, mainly due to the difficulties involved in developing an
up to eight years of financial data. adequate price index for value added. In this paper, we focus on
In order for firms not to switch definition (between small and the profit margin defined as gross pre-tax operating profits divided
big) in our sample we first construct an average of the total assets by revenue (“return on sales”) as our key performance measure
for every firm. We then use this median for our split. (we also compare the results to alternative normalizations such as
In Table 1, we present our sample size by country including assets or equity). Accounting profits can diverge from economic
a separate column for SWIFT adopters. While SWIFT is adopted profits, but the two are likely to be correlated at the firm level and
by many organizations including broker/dealers, corporates, cus- there is a tradition in industrial economics which supports using
todians, investment managers, payment and securities market profitability as a key measure of firm performance (Slade, 2004).
infrastructures (i.e. stock exchanges) and other non-financial insti- Table 2 presents descriptive statistics. The median bank in the
tutions, we only keep the matched data from the SWIFT-Bankscope sample is not large: it has 164 employees, sales of $49 million and
merge including the non-adopter firms from Bankscope. Thus, the $5.9 m in profit. The profit margin is 0.13. Note, however, that the
resulting dataset exclusively contains banking institutions of all data is quite skewed as mean sales are $638 m with a standard
kinds existing in Bankscope. Looking down column (1), we see that deviation of $3,702 m. The other parts of the table break down
almost a quarter of firms are in Germany (mostly due to the popu- the descriptive statistics by firm size and country. Profits are the
larity of local savings banks known as Sparkassen) and almost one difference of revenues and costs, so we also present results that dis-
fifth in the US. Other countries which have many banks in the sam- aggregate profitability into the revenue and cost components. We
ple are the UK (6.6%), France (6.8%), Switzerland (7.9%) and Italy also examine the change in employment following SWIFT adoption
(11.4%).8 In order to avoid any duplication in our data we excluded as a further outcome.
Our main indicator of diffusion is simply an adoption dummy
equal to unity in the year that the bank adopts SwiftNet Fin marked
7
We clean our dataset from extreme negative and positive values that appear in by the end of the post-implementation and the start of the “Live”
our factor inputs. We also avoid dropping the data by winsorising our performance phase (recall Fig. 2). It is unclear when the pre-implementation
variables on the top and bottom percentiles. Results are similar if we simply trim
the outliers.
8
The 29 countries and 3380 SWIFT adopters in our database cover 41.34% of
the entire SWIFT population globally which is about 8176 firms in 219 countries the Bankscope database. As mentioned above, SWIFT is adopted by a number of
and territories. From the 3380 SWIFT adopters we managed to match 1689 onto non-banking organizations that are not included in Bankscope (about half).
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 991

Table 2 An econometric problem that arises while trying to estimate the

Descriptive statistics.
effects of technology adoption on firm performance is unobserved
Obs. Median Mean Stdev heterogeneity. This occurs when there are many factors correlated
Variables with firm performance that we do not measure. In our case, this may
Total assets (m$) 29970 729.43 10300 62000 create an upwards bias for the coefficient of SWIFT. We assume that
Total sales (m$) 29970 49.705 637.6793 3701.933 these unobserved factors stay constant over time and we treat them
Pre-tax profits (m$) 29970 5.936 104.894 759.944 as fixed effects. Then we proceed with our estimation by including
Employees 29970 164 1460.54 8479.635
a full set of firm-level dummy variables (the within-groups esti-
Operating expenses (m$) 29901 20.1 259.637 1674.894
mator). A problem with the fixed effects estimator is that it will
Ratios
exacerbate classical measurement error causing the SWIFT coeffi-
Profit margin 29970 0.1384 0.1522 0.1524
Return on assets 29970 0.61 0.7822 0.9919
cient to be attenuated towards zero. But since this is administrative
Return on equity 29946 7.43 8.5566 8.1563 data there is probably little measurement error. A second concern,
Cost to income 29789 67.13 68.337 29.8822 however, is that there may still be unobserved shocks, so that SWIFT
Small Firms adoption is correlated with the error term, εit .In the absence of an
Total assets (m$) 14300 255.992 294.235 211.316 instrumental variable it is difficult to do much about this, but the
Total sales (m$) 14300 17.273 22.053 27.465 fact that the main effects come not from the current variables but
Pre-tax profits (m$) 14300 1.808 3.396 9.0938 the long-lagged variables gives some reassurance that the positive
Profit margin 14300 0.1308 0.1418 0.156
effects we identify are not due to endogeneity bias.
Big Firms Finally, note that all standard errors are clustered by firm to
Total assets (m$) 15670 2429.1 19400 84700
allow for arbitrary patterns of autocorrelation over time (serial cor-
Total sales (m$) 15670 162.521 1199.482 5054.607
Pre-tax profits (m$) 15670 22.23 197.518 1042.361 relation is typical in firm panels even after removing fixed effects).
Profit margin 15670 0.1459 0.1618 0.1483

Notes: Sample includes 6848 firms in 29 countries, from 1998 to 2005; m$ = Millions 6. Analysis and discussion of findings
of US Dollars. Small and Big firms are split according to the overall median of the
Total Assets sample. In order for firms not to switch definition (between small and
big) in our sample we first construct an average of the total assets for every firm. In the next part of the paper we report on our analysis and dis-
We then use this median for our split. cuss the implications of our results for the research questions in
our study. In the first sub-section, we consider whether there is
evidence that ICT adoption generates long-term benefits for banks
phase begins so we are careful to test for the exact timing (see and whether these benefits accumulate over time. The timeline
below). In particular, it is likely that the benefits of SWIFT will not that is produced is important because whereas previous studies
be observed in the first year, but instead there will be a longer- rest on aggregated or cross sectional data, our findings are based
term dynamic at work between the introduction of SWIFT and its upon a longitudinal panel of data for a particular network technol-
eventual effect on the bottom line. The fact that we have the entire ogy. Next, we focus in more detail on whether particular kinds of
history of all adoptions of SWIFT is helpful here because we are firms benefit more than others in the long-term. Finally, we inves-
able to construct long lags back to 1978 for each firm. In other tigate the mechanisms through which SWIFT adoption adds value
words, we are able (in 1998) to include up to a twenty-year dis- to banks in the long-run and identify, among other things, network
tributed lag for SWIFT adoption to examine the dynamic effects on effects to be of significance. For each section we draw on insights
firm performance. from prior in-depth field studies, and discuss how we might inter-
We do not have data on the intensity of usage of SWIFT for the pret the benefits that we have identified and their implications for
whole period − some proxies exist in one or two years but these further research.
are not consistent across countries. Consequently, we focus on the
simple adoption dummy as is standard in the diffusion literature.
6.1. The long-term effects of SWIFT adoption
5. Modelling strategy
Table 3 reports our basic regression results using the specifi-
cation in equation (1). Column (1) simply regresses profitability
The main equation of interest is:
on a nine-year distributed lag of SWIFT adoption (all columns
(II/S)it = Lj=0 ˛j SWIFT i,t−j + ˇ1 Xit + i + Tt + εit (1) include year and firm dummies). SWIFT appears to have a signifi-
cant impact on firm profitability for up to 9 years. Lags at ten years
Where (˘/S)it is the profit margin, the ratio of pre-tax profits to and beyond were insignificant. As shown at the base of the col-
sales of firm i at time t. Xit denotes a vector of control variables umn the sum of the SWIFT coefficients are significantly different
such as the log of total assets to employees as proxy for the fact from zero (p-value = 0.0018) and the coefficients are jointly sig-
that firms of different capital intensity have different profit sales nificant (p-value = 0.0006). The dynamics are interesting: there is
margins (e.g. if there are high fixed costs gross margins will be little effect, even a negative coefficient in some of the early years
higher). We include a full set of firm fixed-effects, i to control of SWIFT on profits. The larger effects do not materialise for several
for permanent unobserved heterogeneity (the country dummies years. We illustrate these dynamic effects in Fig. 4 which presents
are absorbed into this) and time dummies to control for macro- the cumulative effect of SWIFT over time. The figure illustrates that
economic shocks, Tt . Finally, εit is an idiosyncratic error term whose positive returns are not clearly visible until two years after SWIFT
properties we discuss below. SWIFTit is an adoption variable that adoption and only gradually build up the long-run effect of 0.0823,
is a binary dummy variable taking the value of one in the year of which is sizeable.
the “go-live” phase of adoption and all years after (and zero in the Column (2) includes the capital-labor ratio as an additional
years before the go-live year). We allow a distributed lag up to L control, whose coefficient is positive and highly significant. The
on this where empirically we estimate that L is approximately 9, in dynamics are illustrated in Fig. 5 and show an even slower build-
other words it takes about a decade for the full effect of SWIFT to up of profit margins than the previous column – the long-run effect
play out on bank performance. This was also generally confirmed falls to 0.07. Column (3) includes a lead in SWIFT to pick up whether
by the people we interviewed in banks. there were costs in the year prior to the “go-live” year of SWIFT
992 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table 3
SWIFT adoption and firm Performance (Pre-Tax Profits divided by Sales).

(1) (2) (3) (4) (5)

Estimation method OLS OLS OLS OLS OLS
Sample All firms Small firms Large firms
Dependent variable /Sit /Sit /Sit /Sit /Sit
 Assets
 ***
log Employees
– 0.0358 0.0358*** 0.0543*** 0.0268***
it
(0.0076) (0.0076) (0.013) (0.0093)
SWIFTit+1 – – 0.0155 – –
(0.0138)
SWIFTit 0.0011 0.0006 −0.0087 0.0012 0.0023
(0.0177) (0.0171) (0.0162) (0.0273) (0.0216)
SWIFTit-1 −0.0087 −0.0106 −0.0102 −0.0001 −0.0187
(0.0205) (0.0201) (0.0201) (0.0309) (0.0261)
SWIFTit-2 0.0269* 0.0243 0.0246 0.0376 0.0114
(0.016) (0.0159) (0.0158) (0.0288) (0.0155)
SWIFTit-3 −0.0051 −0.0057 −0.0056 0.0059 −0.0162
(0.0133) (0.0132) (0.0132) (0.023) (0.014)
SWIFTit-4 0.0263** 0.026** 0.0262** 0.0411** 0.0105
(0.0113) (0.0112) (0.0112) (0.0173) (0.0141)
SWIFTit-5 0.038 0.002 0.002 −0.0002 0.0028
(0.0108) (0.0108) (0.0108) (0.0161) (0.0145)
SWIFTit-6 0.0057 0.0055 0.0055 0.0027 0.0074
(0.011) (0.011) (0.011) (0.0172) (0.0141)
SWIFTit-7 0.0105 0.0098 0.0098 0.0147 0.0062
(0.0104) (0.0104) (0.0104) (0.0171) (0.0132)
SWIFTit-8 0.0034 0.0026 0.0026 −0.0015 0.0042
(0.0089) (0.0089) (0.0089) (0.016) (0.0105)
SWIFTit-9 0.0184** 0.0154** 0.0155** 0.0167 0.0138
(0.0077) (0.0076) (0.0076) (0.0135) (0.009)
Long-Run impact of SWIFT (Sum of coefficients) 0.0823 0.07 0.0618 0.1181 0.0238
Mean of Dependent variable 0.1522 0.1522 0.1522 0.1382 0.1652
Significance of the sum of SWIFT coef. (Prob > F) 0.0006 0.003 0.01 0.0024 0.4053
Joint significance of SWIFT coef. (Prob > F) 0.0018 0.0093 0.0096 0.0215 0.6004
Firm fixed effects Yes Yes Yes Yes Yes
Number of firms 5615 5615 5615 2832 2783
Number of obs. 29970 29970 29970 14300 15670
R2 0.6694 0.6726 0.6727 0.6622 0.6852

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Standard errors in brackets are clustered by firm. All equations include a full set of country and year dummies.
The dependent variable in all columns (/Sit ) is the Profit Margin denoting Pre-tax Profits over Total Revenues (Sales). In all columns we include a 9-year lag structure to test
the long-term effect of SWIFT on firm performance. In column 3 we have also constructed a lead to investigate the causal direction of SWIFT adoption and firm performance.
In columns 4 & 5, we split our data between “Small” and “Big” firms we use a mean of the Total Assets of each firm as size indicator to make the categorisation. The time
period of our sample is 1998–2005 (eight years).

Fig. 4. Long-term returns from SWIFT. Notes: Fig. 4 is a graphical representation of column (1) in Table 3. It presents the long run effect of SWIFT adoption on Profit Margin
in the whole sample. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Adoption data run from 1977 to 2006 and financial data from 1998 to
2005.
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 993

Fig. 5. Long-term returns from SWIFT (controlling for capital intensity). Notes: Fig. 5 is a graphical representation of column (2) in Table 3. It presents the long run effect of
SWIFT adoption on Profit Margin in the whole sample. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Adoption data run from 1977 to 2006
and financial data from 1998 to 2005.

Fig. 6. Long-term returns from SWIFT (controlling for capital intensity and future adoption). Notes: Fig. 6 is a graphical representation of column (3) in Table 3. It presents
the long run effect of SWIFT adoption on Profit Margin in the whole sample. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Adoption data
run from 1977 to 2006 and financial data from 1998 to 2005.

(Fig. 6). The coefficient is insignificant and actually positive rather The positive and statistically significant effect of SWIFT adoption
than negative. This suggests either that the costs before the go-live on firm performance (measured in terms of profit margin) is impor-
point are insubstantial or that most of these are captured in the tant because it challenges the practitioner notion of “the plumbing”.
year when the go live period occurs (given that the implementation In other words, it overturns the prevalent assumption that adopt-
period may just be months). The first three columns all suggest a ing a network innovation like SWIFT is analogous to connecting
positive long-run impact of SWIFT on performance of between 0.06 a neutral system of ‘pipes’ with little or no effect. Archival mate-
and 0.08. Taking the lower bound, this is still an increase in prof- rial and interviews with practitioners involved in the founding of
itability of around 40% throughout the ten years of SWIFT adoption, SWIFT emphasised the effectiveness of its original mission to “kill
which is large.9 telex”, eliminate telegrams, and stop facsimiles. Having such an
easily communicated objective seems to have played a key role in
adoption providing historical insight for innovation strategists. As
for the focus of our study, rally cries to “kill telex” help us to under-
stand increases in efficiency but are there further insights from our
9
The results are robust even when we put in country dummy and year dummy field study that can speak to more generative mechanisms at work?
interactions (year*country) for all years and countries.
994 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

As a ground-breaking network innovation, many problems were banks”. Interviews with ‘SWIFT pioneers’ and founding members
encountered not least of which was the realisation that multi- indicate that small banks were integral to its formation.11 Accord-
jurisdictional legal agreements would need to be drawn up if the ing to our analysis, if we take a long-term view of SWIFT adoption
status of its international messaging systems was going to be rec- we find consistent and significant evidence that small banks have
ognized. What emerged was a “community of practice” (Wenger, both benefitted from SWIFT membership and been an asset to other
1998) in which expertise of many kinds was pooled and made members in the SWIFT network.
accessible to members. As one interviewee put it: “. . .the unique-
ness of SWIFT was and still is today the absolute impressive
accessibility. . . The great idea behind SWIFT was that this would be 6.3. Other outcomes: sales, expenses and labour
a platform for all financial institutions and that its advantage terms
of cooperation should outweigh the individual benefits”. Banks that Table 4 presents the estimates of three other outcome variables:
were normally competitors sent their staff to inter-organizational ln(Sales), ln(costs) and ln(labor-capital ratio). As in the previous
workshops to collaboratively examine the technical challenges of tables, in all columns, we control for firm fixed effects and we
network innovation ranging from legal/regulatory issues, industry include a full set of country and year dummies. The dynamic
standards, process reengineering, organizational restructuring, and responses are graphically presented in Figs. 9–11.
new approaches to management. A domino effect of process inno- Column (1) of Table 4 presents the sales equation. Sales are
vation followed as SWIFT updates rolled out and legacy systems positively and significantly associated with SWIFT adoption: the
were replaced as part of the adoption process. long-run effect of SWIFT on sales is 41 log points, implying the
firm sales increase by approximately 50% ( = [exp(0.407) − 1] × 100)
6.2. Firm size and financial performance over the decade when SWIFT was adopted. These results are con-
sistent with our prior qualitative findings that SWIFT creates new
In columns (4) and (5) we repeat the analysis by splitting the revenue streams and results into an increase in sales.
sample into larger and smaller firms based on median assets.10 The second column uses costs – operating expenses – as a depen-
The specifications are identical to column (2) and the dynamic dent variable. Controlling for assets, we find that the first two years’
responses are plotted in Figs. 7 and 8. The coefficients are much expenses actually increase and start to decrease only from the
larger for smaller firms than bigger ones: smaller firms have a long- third year after adoption. The long-run effect is negatively corre-
run SWIFT effect of 0.12 whereas this is only 0.02 for larger firms. lated with SWIFT adoption and is statistically significant. The initial
Since the margins are larger for bigger firms the implied propor- increase the long-term decrease of the costs is consistent with our
tionate effect is even greater for the small firms than the large firms. story of how SWIFT affects firm operating expenses. While SWIFT in
A possible explanation for this is that the larger firms have to bear a various cases demands a considerable initial amount of investment
lot more re-organization costs because of their legacy proprietary to implement and use, it substitutes different inputs that account
systems. However, this does not explain the long-term difference for a large piece of the operating costs. From the results, we can pre-
in performance resulting from SWIFT adoption. sume that operating costs fall by approximately 20% in the 10-year
In summary, and taking all columns together, we have two key period following SWIFT adoption. This is smaller than the pro-
results. First, there seems to be a positive and statistically signif- portionate increase in revenues, suggesting that SWIFT increases
icant effect of SWIFT adoption on firm performance (measured in profits both by reducing costs and increasing demand, but the effect
terms of profit margin), and this effect appears to be substantial in is stronger on revenues.
magnitude. This is consistent with other recent findings on ICT and In column (3) we use the ratio of employees over assets as a
firm performance. Secondly this effect is much higher on smaller dependent variable. There appears to be a substantial shakeout of
firms rather than big firms. workers relative to capital following SWIFT adoption, presumably
The result for small firms is important because by mapping this because SWIFT enables reductions in manpower. The results here
original data set onto a timeline for adoption benefits we are not are also statistically significant.
only able to contribute baseline findings to the scholarly knowl- The above findings suggest that the value-added mechanism
edge base but also raise a number of key questions for further from SWIFT adoption does not only concern the reduction of
research. For example, going forward, do we need to differenti- operating expenses (due to efficiencies) as one would expect
ate between network technologies such as EDI and ATM (the focus from the adoption of a production technology (Fuentelsaz et al.,
of prior literature) and network innovations which through their 2012). Through the enhancement of banks’ communication capa-
membership associate small firms with a mode of governance, audit bilities and the development of new standards, SWIFT provided the
and accountability that would otherwise be out of reach? If as a opportunity to increase firms’ transactional capacity, develop new
senior member of the executive team has suggested, SWIFT is “the products and services and create new markets that led to further
first cloud”, can we claim that particular forms of network innova- revenue gains. One of the interviewees at a small bank, which has
tion serve, as Lacity et al. (2014) suggest, as “a great equaliser for been back and forth with the decision to adopt SWIFT solely on the
SMEs [providing] an unprecedented opportunity to access econom- basis of cost reductions, said “. . .then there was a feedback from
ically the same IT infrastructure and software as large-sized firms”? our marketing effort, which said that we would be able to increase
Does SWIFT membership amplify small scale specialist offering our business, increase our volume, if we have SWIFT.” Being a net-
with world-class infrastructure? work innovation, the benefits from SWIFT adoption rely largely on
Our results mean we are also able to better reflect upon prac- the number of network adopters that a bank is able to transact with
titioner claims that SWIFT was developed “by big banks, for big over the SWIFT network. In the next section, we make an attempt
to identify any existing network effects that boost the marginal
benefit of SWIFT adopters as the size of the network grows.
10
Median assets were calculated using all available financial data from 1998 to
2005. The results largely stay the same if we split our sample based on the number
of employees instead of using total assets as a size indicator. In the Appendix A
11
we also include a breakdown of our sample in terciles reporting figures for small, One described SWIFT’s approach to members at the time of its found as follows:
medium and large banks (Table A9). The accumulative benefit decreases as the size “. . .everybody paid the same tariff and they paid the same entry fee, it was really,
of the firms in the sample increase confirming our initial findings that smaller firms you know, everybody’s on the same level, the small [local] bank [. . .] to the big U.S.
benefit more from the adoption of SWIFT in the long-term. or big European bank”.
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 995

Fig. 7. Long-term returns in small firms. Notes: Fig. 7 is a graphical representation of column (4) in Table 3. It presents the long run effect of SWIFT adoption on Profit Margin
in the sub-sample of Small firms. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Adoption data run from 1977 to 2006 and financial data
from 1998 to 2005.

Fig. 8. Long-term returns in big firms. Notes: Fig. 8 is a graphical representation of column (5) in Table 3. It presents the long run effect of SWIFT adoption on Profit Margin
in the sub-sample of Big firms. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Adoption data run from 1977 to 2006 and financial data from
1998 to 2005.

6.4. SWIFT network externalities grows by 10 in a country in a year, firms are going to benefit from
another 0.002 increase on their profit margin (1.3%).12
In Table 5, we augment equation (1) to include a network vari- Despite the challenges, we can say that network effects were sig-
able defined as the cumulated aggregate number of SWIFT adopters nificant to some degree for every firm that adopts SWIFT. In other
in a country in a year from the entire SWIFT population. Columns words, every time a firm adopts SWIFT, other firms increase their
(1) and (2) report the coefficients for network effects and lagged profit margin. Taken with our other findings, it seems clear that
network effects respectively. In both columns we find a positive over time firms experience more than the straight forward replace-
and significant coefficient on the network variable. Even though ment of one technology (telex) with another (SWIFT). ‘Over time’ is
the coefficients seem small, they suggest a considerable profitabil- the key phrase here because we show that network effects are not
ity effect if the number of the adopters increases rapidly every year
in each country. The literal interpretation of the current results is
that, for every additional firm that adopts SWIFT in a specific coun-
12
The full network effects are hard to credibly estimate as many are international
try, other adopters will increase their average profit margin ratio
rather than national. Unfortunately, the aggregate number of adopters is collinear
by approximately 0.0002. If the number of adopters for example with the time dummies so cannot be separately identified.
996 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table 4
SWIFT adoption and sales, expenses and labour capital.

(1) (2) (3)

Estimation method OLS OLS OLS 
Employees
Dependent variable log(Sales)it log(Opex)it log Assets it

log(Assets)it – 0.633*** –
(0.031)
SWIFTit 0.0386 0.0174 −0.0162
(0.0646) (0.0473) (0.0468)
SWIFTit-1 0.1569** 0.0219 −0.0527
(0.0714) (0.0395) (0.0385)
SWIFTit-2 0.0813** −0.0217 −0.0708**
(0.0373) (0.0249) (0.0278)
SWIFTit-3 0.0449 −0.0473 −0.0165
(0.0308) (0.0303) (0.0312)
SWIFTit-4 0.0605** −0.0146 −0.0067
(0.0247) (0.0249) (0.0239)
SWIFTit-5 0.0064 −0.0217 −0.0504*
(0.0247) (0.024) (0.0293)
SWIFTit-6 −0.0078 −0.0131 −0.0043
(0.0271) (0.0258) (0.0298)
SWIFTit-7 −0.0095 −0.0385* −0.0196
(0.0239) (0.0221) (0.0239)
SWIFTit-8 0.003 −0.0102 −0.0232
(0.0252) (0.0202) (0.0236)
SWIFTit-9 0.0327 −0.0662*** −0.0846***
(0.0338) (0.025) (0.0275)
Long-Run impact of SWIFT (Sum of coefficients) 0.4072 −0.1939 −0.345
Mean of Dependent variable (level) 653.6921 254.3271 −8.395
Significance of the sum of SWIFT coef. (Prob > F) 0.0000 0.0023 0.0000
Joint significance of SWIFT coef. (Prob > F) 0.0022 0.0036 0.0007
Firm fixed effects Yes Yes Yes
Number of firms 6727 6720 5620
Number of obs. 39395 39259 30039
R2 0.9722 0.9822 0.9429

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Standard errors in brackets are clustered by firm. All equations include a full set of country and year dummies.
The dependent variable in column 1 is the log of total revenues, in column 2 the log of operating expenses, and in column 3 the log of employees over assets. The time period
is 1998–2005.

Fig. 9. SWIFT effect on total sales. Notes: Fig. 9 is a graphical representation of column (1) in Table 4. It presents the long run effect of SWIFT adoption on the Total Revenues
(Sales) of the firms of the whole sample. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Descriptive Statistics of our variables are reported
in Table 2. Adoption data run from 1977 to 2006 and financial data from 1998 to 2005.

some singular force of nature but rather that when network innova- across small businesses is external pressure from trading partners
tions become industry standards their sustainability depends upon (especially larger institutions) seeking to reduce ongoing operat-
a continued capacity to produce value for their members. ing costs with counterparties (Kuan and Chau, 2001). As one of the
Returning to our findings about small firms above, the qualita- original SWIFT implementation team put it, “If you were not on
tive data that we gathered is in line with evidence from Iacovou SWIFT but, you know, 25 other banks were, you were going to lose
et al. (1995) who found that the primary reason for EDI adoption the business, because the other banks who were ‘on’ are not going
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 997

Fig. 10. SWIFT effect on operating expenses.

Notes: Fig. 10 is a graphical representation of column (2) in Table 4. It presents the long run effect of SWIFT adoption on the Operating Expenses of the firms in the whole
sample. Again we can observe the dynamic effect of SWIFT. According to our analyses on SWIFT, expenses are expected to increase the first two years of the technology
implementation. After that, operating expenses experience a drop since automates a list of processes in the organisations. Our full sample includes 6848 firms in 29 countries
(adopters & non-adopters). Descriptive Statistics of our variables are reported in Table 2. Adoption data run from 1977 to 2006 and financial data from 1998 to 2005.

Fig. 11. SWIFT effect on labour/assets. Notes: Fig. 11 is a graphical representation of column (3) in Table 4. Here we observe a fall in the numbers of employees relatively to
the assets of the firm sample. Our full sample includes 6848 firms in 29 countries (adopters & non-adopters). Description Statistics of our variables are reported in Table 2.
Adoption data run from 1977 to 2006 and financial data from 1998 to 2005.

to go back to sending telexes.” Complying with such pressure may a SWIFT banking community and a non-SWIFT banking commu-
have brought initial difficulties for small banks but over time they nity, the SWIFT banking community would have suffered as much
benefit from access to a wider range of potential clients with whom as the non-SWIFT banking community· · · I mean it’s like making a
they can securely transact. road only for BMW’s, it doesn’t fit, the philosophy is wrong.” Do our
We close our exploration of innovation adoption and network findings about small firms and network effects point to the possi-
effects by posing a question for further research: does our study of bility that relational governance has the potential to influence how
SWIFT adoption suggest (following Bonardi and Durand, 2003) that adoption benefits evolve over time?
network effects can be managed? As one interviewee said when
asked about the active on-boarding of small bank members during
the early years of SWIFT adoption: “In general, during the develop- 7. Conclusion
ment period, the biggest support came from the small-to-medium
size banks who always have been the pro SWIFT–If we had created In this paper, we focus on the impact of ICT adoption on the per-
formance of firms in the financial services sector using an original
998 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table 5 ing firm fixed effects. Our main results show that the returns from
SWIFT network effects.
SWIFT can take up to ten years to be fully realised. As expected for
(1) (2) most technology investments, we observe an extremely weak or
Estimation method OLS OLS negative result within the first few years of the adoption of SWIFT.
Sample All firms
This is consistent with the idea that it takes years to fully implement
Dependent variable /Sit /Sit
technological and organizational changes and even longer for firms
log(Assets)it 0.0362*** 0.0362*** to enjoy the benefits of adopting these innovations. Additionally,
(0.0076) (0.0076)
the profitability effects of SWIFT derive mainly from an increase in
Network Effectjt 0.0001808*** –
(0.0000575) sales, not just a fall in long-term operating expenses (due to fewer
Network Effectjt-1 – 0.0002741*** employees per unit of capital). We believe that the most interesting
(0.0000547) result to emerge from our analysis − beyond the positive impact on
SWIFTit 0.001 0.0013
performance (measured in terms of profit margin) and evidence of
(0.017) (0.017)
SWIFTit-1 −0.0102 −0.0107 network effects − is the realisation that smaller firms benefit from
(0.02) (0.02) relatively higher returns than the larger ones.
SWIFTit-2 0.025 0.025 There are many outstanding issues and areas for further research
(0.0158) (0.0158) here which will intrigue scholars of ICT adoption, particularly those
SWIFTit-3 −0.0053 −0.0054
specialising in financial services. Using additional data from an in-
(0.0132) (0.0132)
SWIFTit-4 0.0264** 0.0264** depth field study, we explore potential generative mechanisms that
(0.0112) (0.0112) may be implicated in the benefits of adoption. We point to the
SWIFTit-5 0.0023 0.0023 ‘domino’ effect as SWIFT network innovations are enfolded into
(0.0108) (0.0109)
“productive process advances” (Fuentelsaz et al., 2009; p.1174),
SWIFTit-6 0.006 0.0061
(0.011) (0.011)
the value that access to a community of practices has for SME’s,
SWIFTit-7 0.0106 0.0108 and SWIFT’s role as a non-state actor working with legislators to
(0.0103) (0.0104) write the legal terms for international business when gaps between
SWIFTit-8 0.0033 0.0034 national jurisdictions were discovered. Aware that scholars are
(0.0089) (0.0089)
interested in the specific characteristics of network innovations
SWIFTit-9 0.0176** 0.0183**
(0.0076) (0.0076) in order to understand their distinctive strategic importance (Katz
Long-Run impact of SWIFT (Sum of coefficients) 0.0765 0.0776 and Shapiro, 1985; Farrell and Saloner, 1985), we note that SWIFT
Mean of Dependent variable 0.1517 0.1517 is an industry cooperative. In other words, SWIFT was funded
Significance of the sum of SWIFT coef. (Prob > F) 0.0011 0.0009
and developed through voluntary action rather than being cre-
Joint significance of SWIFT coef. (Prob > F) 0.0027 0.0020
Firm fixed effects Yes Yes
ated by regulatory instruction and as a mutual it rebates profit
Number of firms 5615 5615 to its members each year. This leads us to ask not only whether
Number of obs. 29970 29970 SWIFT membership provides (especially small firms) access with
R2 0.6730 0.6735 a mode of governance, audit and accountability that would other-
Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Standard errors wise be out of reach but also whether this points to the possibility of
in brackets are clustered by firm. All equations include a full set of country and using relational governance to manage network effects. Finally, we
year dummies. The time period is 1998–2005. The Network Effect variable is the
overturn a number of assumptions about SWIFT in the practitioner
aggregate number of SWIFT adopters by country j and per year t from 1977 to 2006.
The Global stock variable is the accumulated number of SWIFT adopters globally community. We would suggest that it is important for financial
per annum. Global stock * SWIFTit is the interaction between Global stock and the services organizations not only to ‘know your customer’ but also
adoption dummy SWIFTit . to ‘know thyself’. Our findings reinforce the need to explore, sub-
stantiate and then pursue evidence-based strategy when making
dataset from SWIFT, one of the first and probably the most widely ICT investments in the evolving financial services technoscape.
used network technologies in the banking world. SWIFT has been
part of the core financial infrastructure for over 40 years; it pro- Acknowledgements
vides global information processing services, transmits more than
20 million transaction messages a day, and effectively serves as the We would like to thank Peter Ware from The SWIFT Institute
world’s most reliable third party network. Yet for most of its history for supplying the adoption data and for many insights. The views
it has ‘flown under the radar’. Most people will have a SWIFT code here do not in any way reflect the views of SWIFT and no finan-
in the corner of their bank account statement and many will have cial support was received from SWIFT for this study. The Economic
used it when making an international funds transfer but few know and Social Research Council (ESRC) has provided financial support
anything about its founding, development, operations or impact. through the Centre for Economic Performance at the London School
Using an uncommonly rich longitudinal panel of data on 6848 of Economics & Political Science.
banks in 29 countries in Europe and North America we construct
long lags and investigate the dynamics characterizing the effect of Appendix A. Definitions, descriptive statistics, and selected
SWIFT adoption on firm performance. We provide robust evidence results using alternative specifications
that SWIFT adoption has a positive and significant association with
firm performance even after controlling for many factors, includ- See Tables A1–A8 .
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 999

Table A1
SWIFT adoption and firm performance.

(1) (2) (3)

Estimation method OLS OLS OLS
Sample All firms

Dependent variable ROAit ROEit C/Iit

SWIFTit 0.0343 −0.5141 0.7632

(0.0854) (0.742) (3.9223)
SWIFTit-1 −0.1119 −0.8495 −1.3863
(0.0971) (0.7396) (4.7286)
SWIFTit-2 0.0223 0.5512 −5.6947
(0.085) (0.6489) (4.633)
SWIFTit-3 0.023 0.2414 −0.51
(0.0782) (0.6768) (4.0055)
SWIFTit-4 0.139* 0.9697 −3.0426
(0.0737) (0.611) (3.0944)
SWIFTit-5 0.0035 0.1633 2.1921
(0.0706) (0.6274) (2.518)
SWIFTit-6 0.0418 −0.1904 −1.6541
(0.0643) (0.5242) (3.178)
SWIFTit-7 0.0445 0.5961 −0.7196
(0.0618) (0.5321) (1.8542)
SWIFTit-8 0.0055 −0.3878 −2.2
(0.0621) (0.5107) (1.3238)
SWIFTit-9 0.0236 0.5366 −1.6793
(0.0619) (0.5092) (1.4007)
Long-Run impact of SWIFT (Sum of coefficients) 0.2257 1.1165 −13.9313
Mean of Dependent variable 0.7751 8.3774 68.149
Significance of the sum of SWIFT coef. (Prob > F) 0.1238 0.3507 0.0023
Joint significance of SWIFT coef. (Prob > F) 0.4117 0.3680 0.1101
Firm fixed effects Yes Yes Yes
Number of firms 6754 6739 6624
Number of obs. 39661 39602 38785
R2 0.6258 0.6052 0.5830

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. Standard errors in brackets are clustered by firm. All equations include a full set of country and year dummies.
The dependent variable in column 1 is Return on Assets, in column 2 Return on Equity, and in column 3 Cost to Income Ratio. The time period is 1998–2005.

Table A2
Definition of measures.

Measures Ratio Description

Performance
Profit Margin Pretax Income/Sales (PCM) Increase in PCM indicates higher Profits generated by Sales
Return on Assets Pretax Income/Assets (ROA) Increase in ROA indicates higher Profits generated by Assets
Return on Equity Pretax Income/Equity (ROE) Increase in ROE indicates higher Profits generated by Equity

Efficiency
Cost to Income Operating Expenses (OPEX)/Operating Income Efficiency Inverse Ratio: the higher the ratio, the worse the perceived efficiency

Notes: This table includes the definitions of the main ratios that we use in our analysis. We have categorized them into two distinct groups: Performance measures and
Efficiency measures. Performance ratios measure the returns of SWIFT, whereas, efficiency ratios measure the costs behaviour of the banks relative to their Profits and Assets.
Efficiency ratios are both inverse ratios.

Table A3
Descriptive statistics on SWIFT adopters.

Mean Size (Total Assets in millions$) Mean Profit Margin Number of Adopters Number of Firms

SWIFT adopters
early 1977–1999 22200 0.1648 1108
late 2000–2006 9459.479 0.134 263

In our sample
1998–2005 10100 0.1352 329
1998–1999 9519.264 0.1546 98
2000–2005 10400 0.1264 231

Non-adopters
whole sample 5945.563 0.1496 5159

Notes: Sample includes 6848 firms in 29 countries, from 1998 to 2005; millions$ = Millions of US Dollars. Figures reported here are the ones used in the regressions in Table
A4. In total 6530 firms are being utilized by the regressions from which 1371 are SWIFT adopters (1108 + 263), and 5159 are non-adopters. We can observe that early adopters
that adopted SWIFT before 2000 are bigger in size than firms that adopted SWIFT after 2000. The smallest firms in our sample are the ones that haven’t adopted SWIFT yet,
however, these are not necessarily the least profitable in terms of Profit Margins.
1000 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table A4
SWIFT adoption and firm performance − robustness checks.

(1) (2) (3) (4)

Estimation method OLS OLS OLS OLS
Sample All firms SWIFT adopters
Dependent variable /Sit /Sit /Sit /Sit
 Assets
 *** ***
log Employees
0.0404 – 0.0509 –
it
(0.0084) (0.0098)
*
log(Assets)it – 0.0119 – 0.0133
(0.0066) (0.0098)
SWIFTit 0.0014 −0.0023 −0.0012 −0.0052
(0.0168) (0.0153) (0.017) (0.0155)
SWIFTit-1 −0.0015 −0.0054 −0.0106 −0.0052
(0.0191) (0.0178) (0.02) (0.0177)
SWIFTit-2 0.0172 0.0133 0.0227 0.012
(0.0157) (0.0141) (0.0158) (0.0141)
SWIFTit-3 −0.0073 0.0072 −0.0069 0.0055
(0.0125) (0.0121) (0.0133) (0.0122)
SWIFTit-4 0.0206* 0.0184* 0.025** 0.017
(0.0112) (0.0104) (0.0112) (0.0105)
SWIFTit-5 0.0015 0.001 0.0015 0.0009
(0.01) (0.0106) (0.0109) (0.0106)
SWIFTit-6 0.0032 0.0028 0.0056 0.0021
(0.0095) (0.0103) (0.011) (0.0104)
SWIFTit-7 0.0027 0.0096 0.0086 0.0081
(0.0098) (0.0096) (0.0104) (0.0097)
SWIFTit-8 0.0009 0.0098 0.0028 0.0093
(0.009) (0.0093) (0.0091) (0.0093)
SWIFTit-9 0.0167** 0.013 0.0128 0.0107
(0.0077) (0.0083) (0.0079) (0.0086)
Long-Run impact of SWIFT (Sum of coefficients) 0.0555 0.0672 0.0604 0.0554
Mean of Dependent variable 0.1522 0.1517 0.1593 0.1584
Significance of the sum of SWIFT coef. (Prob > F) 0.0155 0.0020 0.0274 0.0296
Joint significance of SWIFT coef. (Prob > F) 0.2009 0.0299 0.1387 0.3271
Firm fixed effects Yes Yes Yes Yes
Number of firms 5615 6727 1371 1642
Number of obs. 29970 39393 7320 9357
R2 0.6919 0.6778 0.6337 0.6271

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. In this table we perform some robustness check on the relationship between SWIFT adoption and firm
performance. Standard errors in brackets are robust to heteroskedacity and autocorrelation of unknown form and are clustered by firm. All equations include a full set of
country and year dummies and in column 1 their interaction. In all columns we include a 9-year lag structure to test the long-run effect of SWIFT on firm performance. We
test our data using the whole sample (col. 1 and 2), and SWIFT adopters sample (col. 3 and 4).

Table A5
SWIFT adoption and firm performance − robustness checks.

(1) (2) (3) (4) (5) (6)

Estimation method OLS OLS OLS OLS OLS OLS
Sample Commercial Banks Non-Commercial Banks Selection of Banks

Dependent variable /Sit /Sit /Sit /Sit /Sit /Sit

 Assets
 *** *** ***
log Employees
0.0395 – 0.0319 – 0.0391 –
it
(0.013) (0.0088) (0.009)
log(Assets)it – 0.0137 – 0.0098 – 0.016**
(0.01) (0.0089) (0.0076)
SWIFTit 0.0005 −0.0076 −0.001 −0.0004 −0.0086 −0.0103
(0.0261) (0.0235) (0.0207) (0.0179) (0.0209) (0.0184)
SWIFTit-1 −0.0052 −0.0041 −0.0194 −0.0085 −0.0205 −0.0166
(0.0255) (0.0234) (0.0332) (0.0277) (0.0241) (0.0209)
SWIFTit-2 0.0209 0.0116 0.0273 0.0108 0.0287 0.0163
(0.0175) (0.0166) (0.0308) (0.0255) (0.0181) (0.016)
SWIFTit-3 −0.0043 0.012 −0.0141 −0.0089 −0.0096 0.0072
(0.0165) (0.0158) (0.0208) (0.0172) (0.0146) (0.0136)
SWIFTit-4 0.0319** 0.0235** 0.0055 0.0005 0.0327*** 0.0241**
(0.0137) (0.0128) (0.0181) (0.018) (0.0117) (0.0112)
SWIFTit-5 0.0000 −0.0018 0.0018 0.0017 −0.0013 −0.0005
(0.0137) (0.0134) (0.0156) (0.0169) (0.0117) (0.0114)
SWIFTit-6 0.0116 0.0112 −0.014 −0.0193 0.0063 0.0057
(0.0127) (0.0131) (0.0211) (0.0164) (0.0111) (0.0108)
SWIFTit-7 0.0191 0.0122 −0.0171 −0.0037 0.0087 0.0053
(0.0117) (0.0113) (0.0213) (0.0181) (0.0105) (0.0099)
SWIFTit-8 −0.0014 0.005 0.0044 0.0125 −0.0004 0.006
(0.0107) (0.0109) (0.0161) (0.0176) (0.0094) (0.0093)
SWIFTit-9 0.0171* 0.0175* 0.0041 −0.0059 0.0153* 0.0195**
(0.0092) (0.0098) (0.0139) (0.0155) (0.0082) (0.0086)
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 1001

Table A5 (Continued)

(1) (2) (3) (4) (5) (6)

Estimation method OLS OLS OLS OLS OLS OLS
Sample Commercial Banks Non-Commercial Banks Selection of Banks

Dependent variable /Sit /Sit /Sit /Sit /Sit /Sit

Long-Run impact of SWIFT (Sum of coefficients) 0.0902 0.0796 −0.0225 −0.0211 0.0513 0.0567
Mean of Dependent variable 0.1717 0.1629 0.1445 0.1474 0.1601 0.1554
Significance of the sum of SWIFT coef. (Prob > F) 0.0045 0.0085 0.5608 0.5222 0.0507 0.0193
Joint significance of SWIFT coef. (Prob > F) 0.0059 0.0460 0.9422 0.9811 0.0110 0.0170
Firm fixed effects Yes Yes Yes Yes Yes Yes
Number of firms 1565 1910 4050 4817 3903 4659
Number of obs. 8516 10840 21454 28553 21272 27516
R2 0.6552 0.6452 0.6829 0.6165 0.6625 0.6575

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. In this table we perform some robustness check on the relationship between SWIFT adoption and firm
performance. Standard errors in brackets are robust to heteroskedacity and autocorrelation of unknown form and are clustered by firm. All equations include a full set of
country and year dummies and in column 1 their interaction. In all columns we include a 9-year lag structure to test the long-run effect of SWIFT on firm performance.
We test our data using Commercial Banks sample (col. 1 and 2), Non-Commercial Banks and other financial institutions sample (col. 3 and 4), and a selection of Financial
Institutions1 (col. 5 and 6).

Table A6
SWIFT adoption and firm performance − robustness checks.

(1) (2) (3) (4)

Estimation method OLS OLS OLS OLS
Sample All firms All firms

Dependent variable /Sit /Sit /Sit /Sit

 Assets 
log Employees 0.023** (0.009) – – –
 Assets it
log Employees
0.0149** – 0.0247*** –
it−1
(0.0073) (0.0076)
log(Assets)it – −0.0008 – –
(0.009)
log(Assets)it-1 – 0.0011 – 0.0007
(0.0073) (0.0073)
SWIFTit 0.0113 0.008 0.0115 0.0079
(0.0214) (0.0171) (0.0214) (0.0171)
SWIFTit-1 −0.025 −0.016 −0.0274 −0.0161
(0.0243) (0.019) (0.024) (0.019)
SWIFTit-2 0.0127 0.009 0.016 0.0091
(0.0199) (0.0164) (0.0191) (0.0164)
SWIFTit-3 0.004 0.0093 0.0028 0.0093
(0.0143) (0.0135) (0.014) (0.0135)
SWIFTit-4 0.0238** 0.0191* 0.0172 0.019*
(0.0107) (0.011) (0.0113) (0.011)
SWIFTit-5 0.0053 0.0061 0.0117 0.0061
(0.0104) (0.0106) (0.0109) (0.0105)
SWIFTit-6 −0.0029 −0.0081 −0.0074 0.0081
(0.0109) (0.0105) (0.0106) (0.0105)
SWIFTit-7 0.0078 0.0086 0.0115 −0.0085
(0.0109) (0.0098) (0.0107) (0.0098)
SWIFTit-8 −0.0043 0.0012 −0.005 0.0012
(0.0092) (0.009) (0.0093) (0.009)
SWIFTit-9 0.0144 0.0087 0.0135 0.0087
(0.0088) (0.0095) (0.0087) (0.095)
Long-Run impact of SWIFT (Sum of coefficients) 0.0471 0.0458 0.0443 0.0457
Mean of Dependent variable 0.1582 0.1545 0.1554 0.1545
Significance of the sum of SWIFT coef. (Prob > F) 0.0859 0.0652 0.1134 0.065
Joint significance of SWIFT coef. (Prob > F) 0.1023 0.3932 0.1338 0.391
Firm fixed effects Yes Yes Yes Yes
Number of firms 5061 6504 5253 6504
Number of obs. 23347 32554 24746 32554
R2 0.7054 0.6938 0.6956 0.6938

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. In this table we perform some robustness check on the relationship between SWIFT adoption and firm
performance. Standard errors in brackets are robust to heteroskedacity and autocorrelation of unknown form and are clustered by firm.
1002 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Table A7
Bank specialisations in the sample.

Bank Specialisations Number of Firms SWIFT Adopters (firms) SWIFT Non-adopters (firms)

Bank Holding & Holding Companiesa 717 38 679

Central Banks 28 27 1
Commercial Banksa 1927 1034 893
Cooperative Banksa 1620 87 1533
Investment Banks/Securities Housesa 498 226 272
Islamic Banks 1 1 0
Medium & Long Term Credit Banks 48 18 30
Multi-lateral Governmental Banks 2 1 1
Non-banking Credit Institutions 463 52 411
Real Estate/Mortgage Banks 189 31 158
Savings Banks 1280 138 1142
Specialised Governmental Credit Institutions 75 36 39
6848 1689 5159

Notes: Sample includes 6848 firms (205,440 observations) in 29 countries, from 1977 to 2006.
a
These banks are included in the sample for columns (5) and (6) in Table A5.

Table A8
SWIFT early and late adopters.

(1) (2) (3) (4) (5)

Estimation method OLS OLS OLS OLS OLS

Sample All firms Years ≤2001 Years ≥2002
Type of SWIFT adopters Early Early Late All All

Dependent 
variable
Assets
/Sit /Sit /Sit /Sit /Sit
log Employees
0.029*** 0.0297*** 0.0351*** 0.0283*** 0.044***
it
(0.0078) (0.0078) (0.01) (0.0098) (0.0169)
SWIFTit −0.0096 −0.0105 −0.0014 −0.0046 0.0148
(0.0245) (0.0245) (0.0194) (0.0258) (0.0351)
SWIFTit-1 −0.0069 −0.0078 −0.0206 −0.0047 −0.0158
(0.0247) (0.0247) (0.0272) (0.0192) (0.0425)
SWIFTit-2 −0.0024 −0.0037 0.0439* 0.0024 0.0592*
(0.0205) (0.0205) (0.023) (0.0202) (0.0345)
SWIFTit-3 −0.0155 −0.017 0.0134 0.0071 0.0075
(0.0185) (0.0185) (0.0155) (0.016) (0.0184)
SWIFTit-4 0.0278* 0.0269* 0.035* 0.0149 0.0692***
(0.0146) (0.0146) (0.0181) (0.0181) (0.0214)
SWIFTit-5 0.0042 0.0172 0.0102 0.0339* −0.0025
(0.0121) (0.0107) (0.0213) (0.0187) (0.0166)
SWIFTit-6 0.0084 0.0143 0.0018
(0.0112) (0.0186) (0.0198)
SWIFTit-7 0.0093 0.022 0.0033
(0.0104) (0.0206) (0.0138)
SWIFTit-8 0.0024 0.0291 −0.0082
(0.0089) (0.0202) (0.0097)
SWIFTit-9 0.0157** 0.0073 0.0072
(0.0076) (0.0155) (0.0131)
Long-Run impact of SWIFT (Sum of coefficients) 0.0335 0.005 0.0806 0.1216 0.1366
Mean of Dependent variable 0.1531 0.1386 0.1491 0.1406 0.1616
Significance of the sum of SWIFT coef. (Prob > F) 0.4066 0.8986 0.0087 0.0122 0.0414
Joint significance of SWIFT coef. (Prob > F) 0.0405 0.0915 0.0647 0.2090 0.1089
Firm fixed effects Yes Yes Yes Yes Yes
Number of firms 5352 5352 4507 4233 5260
Number of obs. 28654 28654 23966 13354 16616
R2 0.6729 0.6725 0.6903 0.7725 0.7775

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. In this table we investigate the SWIFT-effects in two separate sub-samples: early SWIFT adopters (from 1977
to 1999) in columns 1 and 2, and late SWIFT adopters (from 2000 to 2006) in column 3. In columns 4 and 5 we run two additional regressions using a sample of all the
observations prior 2002 and after 2002 respectively. Standard errors in brackets are robust to heteroskedacity and autocorrelation of unknown form and are clustered by
firm. All equations include a full set of country and year dummies.
 S.V. Scott et al. / Research Policy 46 (2017) 984–1004 1003

Table A9
Small and large firms in terciles.

(1) (2) (3)

Estimation method OLS OLS OLS

Dependent variable /Sit
Firm size 
Assets
Small Q1 Medium Q2 Large Q3
log Employees
0.0750*** 0.0285*** 0.0239**
it
(0.0200) (0.0086) (0.0118)
SWIFTit −0.0291 0.0048 0.0204
(0.0348) (0.0312) (0.0228)
SWIFTit-1 −0.0174 0.0284 −0.0305
(0.0403) (0.0417) (0.0258)
SWIFTit-2 0.0605 0.0121 0.0003
(0.0373) (0.0292) (0.0173)
SWIFTit-3 0.0102 −0.0129 −0.0159
(0.0314) (0.0215) (0.0165)
SWIFTit-4 0.0684*** 0.0096 0.0179
(0.0239) (0.0178) (0.0162)
SWIFTit-5 0.0023 0.0147 −0.0123
(0.0244) (0.0170) (0.0158)
SWIFTit-6 −0.0149 0.0065 0.0188
(0.0299) (0.0151) (0.0161)
SWIFTit-7 0.0059 0.0393** −0.0183
(0.0268) (0.0155) (0.0152)
SWIFTit-8 0.0042 −0.0134 0.0165
(0.0257) (0.0138) (0.0103)
SWIFTit-9 0.0259** −0.0018 0.0173
(0.0214) (0.0109) (0.0107)
Long-Run impact of SWIFT (Sum of coefficients) 0.2069 0.0711 0.0142
Mean of Dependent variable 0.1400 0.1475 0.1682
Significance of the sum of SWIFT coef. (Prob > F) 0.0090 0.0442 0.6731
Joint significance of SWIFT coef. (Prob > F) 0.0103 0.0436 0.6071
Firm fixed effects Yes Yes Yes
Number of firms 1827 1982 1807
Number of obs. 8464 11672 9842
R2 0.6473 0.6866 0.6936

Notes: * significant at 10%, ** significant at 5%, *** significant at 1%. In this table we split our data in terciles between “Small”, “Medium” and “Large” firms. We use a mean of
the Total Assets of each firm as size indicator to make the categorisation. The time period of our sample is 1998–2005 (eight years). Standard errors in brackets are robust to
heteroskedacity and autocorrelation of unknown form and are clustered by firm. All equations include a full set of country and year dummies.

1.10 References Bresnahan, Timothy F., Brynjolfsson, Erik, Hitt, Lorin M., 2002. Information
technology, workplace organization and the demand for skilled labor. Q. J.
Aghion, Philippe, Howitt, Peter, 2007. Capital, innovation, and growth accounting. Econ. 117 (1), 339–376.
Oxford Rev. Econ. Policy 23 (1), 79–93. Brynjolfsson, Erik, Hitt, Lorin M., 1993. Is information systems spending
Alpar, Paul, Kim, Moshe, 1990. A microeconomic approach to the measurement of productive? new evidence and new results. In: In The Proceedings of the 14th
information technology value. J. Manag. Inf. Syst. 7 (2), 55–69. International Conference on Information Systems, Orlando, FL.
Anderson, Mark C., Banker, Rajiv D., Ravindran, Sury, 2006. Value implications of Brynjolfsson, Erik, Hitt, Lorin M., 1995. Computers as a factor of production: the
investments in information technology. Manage. Sci. 52 (9), 1359–1376. role of differences among firms. Econ. Innov. New Technol. 3 (3), 183–199.
Aral, Sinan, Weill, Peter, 2007. IT assets, organizational capabilities, and firm Brynjolfsson, Erik, Hitt, Lorin M., 1996. Paradox lost? firm-level evidence on the
performance: how resource allocations and organizational differences explain returns to information systems spending. Manage. Sci. 42 (4), 541–558.
performance variation. Organ. Sci. 18 (5), 763–780. Brynjolfsson, Erik, Hitt, Lorin M., 1998. Beyond the productivity paradox. Commun.
Aral, Sinan, Brynjolfsson, Erik, Wu, D.J., 2006. Which came first, IT of productivity? ACM 41 (8), 49–55.
the virtuous cycle of investment and use in enterprise systems. In: In Brynjolfsson, E., Hitt, L.M., 2000. Beyond computation: information technology,
Proceedings of the 27th International Conference on Information Systems, organizational transformation and business performance. J. Econ. Perspect. 14
Milwaukee, pp. 1–22. (4), 23–48.
Autor, David H., Levy, Frank, Murnane, Richard J., 2002. Upstairs, downstairs: Brynjolfsson, Erik, Hitt, Lorin M., 2003. Computing productivity: firm-level
computers and skills on two floors of a large bank. Ind. Labor Relat. Rev. 55, evidence. Rev. Econ. Stat. 85 (4), 793–808.
432–447. Brynjolfsson, Erik, Yang, Shinkyu, 1996. Information technology and productivity:
BIS, 2002. Statistics on Payment Systems in the Group of Ten Countries. Basel. a review of the literature. Adv. Comput. 43, 179–214.
Baily, Martin Neil, 1986. What has happened to productivity growth? Science 234, Brynjolfsson, Erik, Malone, T.W., Gurvaxani, V., Kambil, A., 1994. Does information
443–451. technology lead to smaller firms? Manage. Sci. 40 (12), 1628–1644.
Ballantine, J., Levy, M., Powel, P., 1998. Evaluating information systems in small and Brynjolfsson, Erik, 1993. The productivity paradox of information technology.
medium-sized enterprises: issues and evidence. Eur. J. Inf. Syst. 7, 241–251. Commun. ACM 36 (12), 66–77.
Barrett, M., Walsham, G., 1999. Electronic trading and work transformation in the Cainelli, G., Evangelista, R., Savona, M., 2006. Innovation and economic
London insurance market. Inf. Syst. Res. 10 (1), 1–22. performance in services: a firm-level analysis. Camb. J. Econ. 30, 435–458.
Barua A., Kriebel C., Mukhopadhyay T., 1991. Information Technology and Business Carr, Nicholas G., 2003. IT doesn’t matter. Harv. Bus. Rev., 41–49.
Value: An Analytic and Empirical Investigation. University of Texas at Austin Casolaro, Luca, Gobbi, Giorgio, 2007. Information technology and productivity
Working Paper May. changes in the banking industry. Econ. Notes 36 (1), 43–76.
Battisti, G., Canepa, A., Stoneman, Paul, 2009. E-business usage across and within Clemons, Eric K., Kimbrough, Steven O., 1986. Information systems,
firms in the UK: profitability, externalities and policy. Res. Policy 38, 133–143. telecommunications, and their effects on industrial organization. In
Beccalli, Elena, 2007. Does IT investment improve bank performance? evidence Proceedings of the Seventh International Conference on Information Systems,
from Europe. J. Bank. Finance 31 (7), 2205–2230. 99–108.
Berndt, Ernst R., Morrison, Catherine J., 1995. High-tech capital formation and Clemons, Eric K., Row, M., 1991. Sustaining IT advantage: the role of structural
economic performance in U.S. manufacturing industries: an exploratory differences. MIS Q. 15 (3), 274–292.
analysis. J. Econ. 65, 9–43. Clemons, Eric K., Weber, Bruce W., 1990. London’s big bang: a case study of
Bloom, Nicholas, Sadun, Raffaella, Van Reenen, John, 2012. Americans do it better: information technology, competitive impact, and organizational change. J.
US multinationals and the productivity miracle. Am. Econ. Rev. 102 (1), Manage. Inf. Syst. 6 (4), 41–60.
167–201. Clemons, Erik K., 1986. Information systems for sustainable competitive
Bonardi, J.P., Durand, R., 2003. Managing network effects in high-tech markets. advantage. Inf. Manage. 11, 131–136.
Acad. Manage. Exec. 17 (4), 40–52.
1004 S.V. Scott et al. / Research Policy 46 (2017) 984–1004

Cragg, P., King, M., 1993. Small-Firm computing: motivators and inhibitors. MIS Q. The resurgence of growth in the late 1990: is information technology the story?,
17 (1), 47–60. 2000. J. Econ. Perspect. 14 (4), 3–22.
Currie, W., Parikh, M., 2006. Value creation in web services: an integrative model. Oulton, N., 2002. ICT and productivity growth in the UK. Oxf. Rev. Econ. Policy 18
The J. Strategic Inf. Syst. 15 (2), 153–174. (3), 363–379.
David, Paul A., 1990. The dynamo and the computer: a historical perspective on the Parsons, Gotlieb, Denny, 1993. Productivity and computers in Canadian banking. J.
modern productivity paradox. Am. Econ. Rev. 80 (2), 35–61. Prod. Anal., 95–114.
Dewan, Sanjeev, Kraemer, Kenneth L., 2000. Information technology and Pfeiffer, H.K.C., 1992. The Diffusion of Electronic Data Interchange. Springer-Verlag,
productivity: evidence from country-level data. Manage. Sci. 46 (4), 548–562. New York, NY.
Dos Santos, B., Peffers, K., 1995. Rewards to investors in innovative information Raymond, Louis., 1985. Organizational characteristics and MIS success in the
technology applications: first movers and early followers in ATMs. Organ. Sci. context of small business. MIS Q. 9 (1), 37–52.
6, 241–259. Roach, Stephen S., 1991. Services under siege: the restructuring imperative. Harv.
Draca, Mirko, Sadun, Raffaella, Van Reenen, John, 2007. Productivity and ICT: a Bus. Rev. 39 (September–October) (2), 82–92.
review of the evidence. In: Mansell, R., Avgerou, C., Quah, D., Silverstone, R. Romer, Paul M., 1990. Endogenous technological change. J. Polit. Econ. 98 (5),
(Eds.), The Oxford Handbook of Information and Communication Technologies. 71–102.
Oxford University Press, Oxford, pp. 100–147. Saloner, Garth, Shepard, Andrea, 1995. Adoption of technologies with network
Economides, Nicholas, 1996. The economics of networks. Int. J. Ind Organ. 14 (6), effects: an empirical examination of the adoption of automated teller
673–699. machines. RAND J. Econ. 26 (3), 479.
Evangelista, R., 2000. Sectoral patterns of technological change in services. Econ. Sassen, S., 2002. Global Networks, Linked Cities. Routledge, London.
Innov. New Technol. 9, 183–221. Saunders, C., Clark, S., 1992. EDI adoption and implementation: a focus on
Farrell, Joseph, Saloner, Garth, 1985. Standardization, compatibility, and interorganizational linkages. Inf. Resour. Manage. J. 5 (1), 9–19.
innovation. RAND J. Econ. 16 (1), 70. Schumpeter, Joseph A., 1943. Capitalism, Socialism and Democracy. Routledge,
Installed base and compatibility: innovation, product preannouncements, and London and New York.
predation., 1992. Am. Econ. Rev. 76, 940–955. Scott, Susan V., Barrett, M., 2005. Strategic risk positioning as sensemaking in
Frame, Scott W., White, Lawrence J., 2004. Empirical studies of financial crisis: the adoption of electronic trading at the London international financial
innovation: lots of talk, little action? J. Econ. Lit. 42 (1), 116–144. futures and options exchange. J. Strategic Inf. Syst. 14 (1), 45–68.
Fuentelsaz, Lucio, Gómez, Jaime, Palomas, Sergio, 2009. The effects of new Scott, Susan V., Walsham, G., 1998. Shifting boundaries and new technologies: a
technologies on productivity: an intrafirm diffusion-based assessment. Res. case study in the UK banking sector. In: In Proceedings of the International
Policy 38 (7), 1172–1180. Conference of Information Systems, Helsinki, Finland, pp. 177–187.
Production technologies and financial performance: the effect of uneven diffusion Scott, Susan V., Zachariadis, Markos, 2012. Origins and development of SWIFT,
among competitors., 2012. Res. Policy 41 (2), 401–413. 1973–2009. Bus. Hist. 54 (3), 462–482.
Geroski, Paul, Machin, Steve, Van Reenen, John, 1993. The profitability of 2014. The Society for Worldwide Interbank Financial Telecommunication (SWIFT):
innovating firms. RAND J. Econ. 24 (2), 198–211. Cooperative Governance for Network Innovation, Standards, and Community.
Gordon, Robert J., 2016. The Rise and Fall of Economic Growth: The U.S. Standard Routledge, Taylor & Francis Group, London; New York, NY.
of Living Since the Civil War. Princeton University Press, NJ. Shapiro, C., Varian, H., 1999. Information Rules: A Strategic Guide to the Network
Grandon, Elizabeth E., Pearson, J. Michael, 2004. Electronic commerce adoption: an Economy. Business School Press, Boston, MA: Harvard.
empirical study of small and medium US businesses. Inf. Manage. 42 (1), Siegel, Donald, Griliches, Zvi, 1992. Purchased services, outsourcing, computers,
197–216. and productivity in manufacturing. In: the Service Sectors (Ed.), In Output
Griliches, Zvi, 1995. Comments on measurement issues in relating IT expenditures Measurement Zvi Griliches. National Bureau of Economic Research, Inc.,
to productivity growth. Econ. Innov. New Technol. 3, 317–321. 429-60.
Gust, Christopher, Marquez, Jaime, 2004. International comparisons of Sinha, Rajiv K., Chandrashekaran, Murali, 1992. A split hazard model for analyzing
productivity growth: the role of information technology and regulatory the diffusion of innovations. J. Mark. Res. 29 (1), 116–127.
practices. Labour Econ. 1 (1), 33–58. Slade, Margaret, 2004. Models of firm-level profitability. Int. J. Ind Organ. 22,
Hall, M., Weiss, L.W., 1967. Firm size and profitability. Rev. Econ. Stat. 49 (8), 289–308.
319–331. Solow, Robert., 1987. We’d Better Watch Out. New York Times.
Hannan, Timothy H., McDowell, John M., 1984. The determinants of technology Stiroh, Kevin, 2002. Information technology and the US productivity revival: what
adoption: the case of the banking firm. RAND J. Econ. 15 (3), 328–335. do the industry data say? Am. Econ. Rev. 92 (5), 1559–1576.
Hannan, Timothy H., McDowell, John M., 1987. Rival precedence and the dynamics Thong, J.Y.L., Yap, C.S., Raman, K.S., 1996. Top management support, external
of technology adoption: an empirical analysis. Economica 54 (214), 155–171. expertise and information systems implementation in small businesses. Inf.
Haynes, Michelle, Thompson, Steve, 2000. The productivity impact of IT Syst. Res. 7 (2), 248–267.
deployment: an empirical evaluation of ATM introduction. Oxf. Bull. Econ. Stat. Tyre, M.J., Orlikowski, W.J., 1994. Windows of opportunity: temporal patterns of
62 (5), 607–619. technological adaptation in organizations. Organ. Sci. 5 (1), 98–118.
Iacovou, Charalambos L., Benbasat, Izak, Dexter, Albert S., 1995. Electronic data Van de Ven, Andrew, 1986. Central problems in the management of innovation.
interchange and small organizations: adoption and impact of technology. MIS Manage. Sci. 32 (5), 590–607.
Q. 19 (4), 465. Weber, Bruce W., 1994. Information technology in the major international
Jorgenson, Dale W., Stiroh, Kevin, 1995. Computers and growth. Econ. Innov. New financial markets. In: Deans, Candace P., Karwan, Kirk R. (Eds.), In Global
Technol. 3, 295–316. Information Systems and Technology: Focus on the Organization and Its
Jun, Sangjoon, 2008. The link between IT investment and securities firms’ returns Functional Areas. IGI Publishing, pp. 132–166.
in korea. J. Econ. Res. 13, 1–43. Weill, Peter, Olson, Margrethe H., 1989. Managing investment in information
Katz, Michael L., Shapiro, Carl, 1985. Network externalities, competition, and technology: mini case examples and implications. MIS Q., 3–18.
compatibility. Am. Econ. Rev. 75 (3), 424–440. Weill, Peter, 1992. The relationship between investment in information
Kretschmer, Tobias, 2012. Information and Communication Technologies and technology and firm performance: a study of the valve manufacturing sector.
Productivity Growth: A Survey of the Literature. OECD Digital Economy Papers Inf. Syst. Res. 3 (4), 307–333.
No. 195: 1–27. Weirdt, D., Hadji-Ashrafi, M.S., Randall, P., Scott, S.V., 2005. Standards: Building a
Kuan, K., Chau, P., 2001. A perception-based model of EDI adoption in small Generic Template. Dialogue, The SWIFT Community.
businesses using technology–organization–environment framework. Inf. Wenger, E., 1998. Communities of Practice: Learning, Meaning, and Identity.
Manage. 38 (8), 507–521. Cambridge University Press, Cambridge, UK.
Kwon, M.J., Stoneman, Paul, 1995. The impact of technology adoption on firm Winder, R., 1985. Too SWIFT for Comfort. Euromoney January: 55–56.
productivity. Econ. Innov. New Technol., 219–233. Zuboff, Shoshanna, 1988. In the Age of the Smart Machine: The Future of Work and
Lacity M.C., Reynolds P., Khan S., Willcocks L., 2014. Cloud Services: The Great Power. Basic Books, New York.
Equalizer for Small and Medium-Sized Enterprises? The Outsourcing Unit
Working Research Paper Series, LSE Paper 14/1.
Oliner, Stephen D., Sichel, Daniel E., 1994. Computers and output growth revisited:
how big is the puzzle? Brook. Pap. Econ. Act. 2, 273–334.