Digital Economics Avi Goldfarb and Catherine Tucker* - Conferences
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Journal of Economic Literature 2019, 57(1), 3–43
https://doi.org/10.1257/jel.20171452
Digital Economics†
Avi Goldfarb and Catherine Tucker*
Digital technology is the representation of information in bits. This technology has
reduced the cost of storage, computation, and transmission of data. Research on digital
economics examines whether and how digital technology changes economic activity.
In this review, we emphasize the reduction in five distinct economic costs associated
with digital economic activity: search costs, replication costs, transportation costs,
tracking costs, and verification costs. (JEL D24, D83, L86, O33, R41)
1. What Is Digital Economics? economic models change as certain costs
fall substantially and perhaps approach zero.
D igital technology is the representation
of information in bits. This reduces the
cost of storage, computation, and transmis-
We emphasize how this shift in costs can be
divided into five types:
sion of data. Research on digital economics (i) Lower search costs
examines whether and how digital technol-
ogy changes economic activity. (ii) Lower replication costs
Understanding the effects of digital tech-
nology does not require fundamentally new (iii) Lower transportation costs
economic theory. However, it requires a dif-
ferent emphasis. Studying digital economics (iv) Lower tracking costs
starts with the question of “what is differ-
ent?” What is easier to do when information (v) Lower verification costs
is represented by bits rather than atoms?
Digital technology often means that costs Search costs are lower in digital environ-
may constrain economic actions. Therefore, ments, enlarging the potential scope and
digital economics explores how standard quality of search. Digital goods can be rep-
licated at zero cost, meaning they are often
non-rival. The role of geographic distance
* Goldfarb: University of Toronto and NBER. Tucker: changes as the cost of transportation for dig-
Massachusetts Institute of Technology Sloan School of ital goods and information is approximately
Management and NBER. We thank Andrey Fradkin and
Kristina McElheran for helpful comments. We are grate- zero. Digital technologies make it easy to
ful to the Sloan Foundation for its support of the NBER track any one individual’s behavior. Last, dig-
Digitization Initiative, which built the research community ital verification can make it easier to certify
around which this review is based.
†
Go to https://doi.org/10.1257/jel.20171452 to visit the the reputation and trustworthiness of any
article page and view author disclosure statement(s). one individual, firm, or organization in the
34 Journal of Economic Literature, Vol. LVII (March 2019) digital economy. Each of these cost changes affect prices and price dispersion. They affect draws on a different set of w ell-established product variety and media availability. They economic models, primarily search, non-rival change matches in a variety of settings, from goods, transportation cost, price discrimina- labor markets to dating. They have led to an tion, and reputation models. increase in the prevalence of platform-based Early research tested straightforward mod- businesses, and affected the organization of els of lower costs. For example, the search some firms. literature of the late 1990s and early 2000s We next turn to zero replication costs, which built directly on earlier models by Diamond also affect pricing decisions including the (1971) and Varian (1980). As we detail below, decision to provide a good for free. This has empirical work emerged that found some enabled an increase in the provision of public inconsistencies with the simple models, and goods such as Wikipedia, raising a number of so richer models and empirical analysis of the new questions about the motivations for pro- cost reductions developed to take account of viding such goods. Zero replication costs cre- the subtleties of the digital context. ate challenges with respect to excludability. Other authors have also emphasized the Copyright can enforce excludability by using role of lower costs for digital economics the law to overcome the non-rival nature of (e.g., Shapiro and Varian 1998; Borenstein the technology. Consequently, copyright has and Saloner 2001; and Smith, Bailey, and become increasingly important to a variety of Brynjolfsson 2001). Ellison and Ellison businesses and a core policy challenge related (2005) discuss the implications of these lower to digitization. search and transportation costs for indus- Because the cost of transporting infor- trial organization with respect to increasing mation stored in bits is near zero, this has returns, distance, and two-sided markets. changed the role of p lace-based constraints Since their article, the digital economics liter- on economic activity, whether due to costs of ature has grown to contribute to the econom- physical transportation or policy. Digitization ics of crime, the economics of public goods, changes the ways governments can control organizational economics, finance, urban the flow of information, from advertising economics, labor economics, development restrictions to media blackouts. economics, health economics, political econ- We then turn to examine a more recent omy, media economics, public finance, and literature that has identified two other cost international economics. In this sense, we changes: Tracking and verification costs. view digital economics as a way of thinking Tracking costs are the costs associated with that touches many fields of economics. connecting an individual person or firm with In addition to applying across many fields, information about them. Low tracking costs these shifts in costs have transformed many enable novel forms of price discrimination as aspects of the economy. After providing a brief well as new ways to targeting advertising and history of digital technology and the Internet, other information. At the same time, better we discuss each of the cost changes associated tracking has made privacy a key issue, gen- with digitization. In each section, we empha- erating a great deal of research and policy size the key research questions that have discussion. driven the area and how they have evolved, We conclude the discussion of cost and relate them to policy where applicable. changes by detailing changes in verification We begin with a discussion of the effect costs. The rise of online reputation sys- of lower search costs, defined as the costs of tems has facilitated trust and created new looking for information. Lower search costs markets. At the same time, such systems are
Goldfarb and Tucker: Digital Economics 5
imperfect, and can serve as platforms for researchers also developed the particu-
fraud or discrimination. lar packet-switching standards that define
We finish by discussing the consequences Internet communication: the Transmission
of digitization for countries, regions, firms, Control Protocol/Internet Protocol (TCP/
and individuals. Digitization has affected IP). The National Science Foundation (NSF)
productivity, trade, the economic role of cit- began managing a network using that proto-
ies, domestic and international outsourcing, col in the 1980s, building a reliable infra-
consumer surplus, and how people spend structure that was relatively easy to adopt but
their leisure time. also restricted to researchers.
Privatization occurred between 1990 and
1995, leading to the modern commercial
2. Digital Technology: A Brief History
Internet. The commercial Internet diffused
The history of modern computing begins quickly, with universities playing a key role
not with the Internet, but in 1945 with the in the diffusion process (Goldfarb 2006).
commercialization of technologies developed There was near-universal availability and
during World War II (Ceruzzi 2003). These widespread adoption in the United States
first machines focused on rapid calculation by 2000 (Greenstein 2000).1 Over time, new
with little capacity for storing and retrieving technologies have been layered on top of
information. By the early 1950s, magnetic the basic TCP/IP-based Internet, including
core memories enabled efficient digital infor- browsers, search engines, online shopping,
mation storage and perhaps the first real social networks, mobile communications
non-arithmetical benefit of representing
protocols, security standards, customer rela-
information in bits emerged: the lower mar- tionship management systems, and many
ginal cost of reproducing information. Over others. These technologies and others have
time, storage technology, software, and hard- enabled increased collection and use of data.
ware improved so that information processing During this process, there has been an
and reproduction became widespread. The open question of who should control vari-
software and hardware industries grew rap- ous aspects of commercial Internet activity,
idly (Ceruzzi 2003 and Campbell-Kelly 2004). given this historical context of decentral-
Limited communication between ization. Standards are often agreed upon
computers limited their effect on the through committees with representatives
economy. It was with the rise of the from industry and academia. Such standards
Internet—and with it, low-cost, commer- have an influence on which technologies
cial, c omputer-to-computer communica- are widely adopted (Rysman and Simcoe
tion—that the representation of information 2008). Therefore, standards setting creates
in bits began to have a measurable effect winners and losers. Simcoe (2012) examines
on multiple markets. This rise was built on the incentives in standards development for
key inventions developed through US mili- one such standard-setting organization, the
tary funding in the 1960s and 1970s (Hafner Internet Engineering Task Force, demon-
and Lyon 1996 and Greenstein 2015). For strating that the commercialization of the
example, the Defense Advanced Research Internet slowed standards development due
Projects Agency (DARPA) funded the inven-
tion of packet switching, which breaks down
1 This rapid speed of diffusion proved useful for identi-
a long message into shorter messages that
fication in the empirical papers examining the effect of the
can be sent through the network and then Internet on regions, firms, and individuals that we discuss
reassembled upon receipt. DARPA-funded in the penultimate section.6 Journal of Economic Literature, Vol. LVII (March 2019)
to competing commercial interests. Given Therefore, a key theme in the history of
their importance, control of hardware and digital technology is a tension between open-
software standards has been controversial. ness and control. As we discuss below, this
Echoing this question of control, the tension is at the center of much of the digital
earlier literature on the economics of the policy literature with respect to copyright,
Internet focused on pricing the sending of privacy, and discrimination.
information and how it varies with intercon-
nection, competition, and the nature of the
3. Reduction in Search Costs
content (MacKie-Mason and Varian 1994).
In other words, there is a question about Search costs are the costs of looking for
the role of Internet service providers in con- information. Every information-gathering
trolling access. Laffont et al. (2003) empha- activity therefore involves search costs. The
sized how the need for interconnection can basic idea with respect to digital economic
affect prices and welfare. This literature activity is that it is easier to find and compare
emphasized network effects and the chal- information about potential economic trans-
lenges of interconnection (Cremer, Rey, and actions online than offline.
Tirole 2000; Besen et al. 2001; Laffont et al. At the beginning of the commercial
2001; and Caillaud and Jullien 2003). Internet, there was much discussion among
As data transmission became a key aspect economics researchers around how a dra-
of digital technology, the question of net matic reduction in search costs might trans-
neutrality has become a central research form the economy by reducing prices, price
and policy focus. Net neutrality means that dispersion, unemployment, vacancies, and
an Internet service provider should treat all inventories. Alan Greenspan argued that the
data in the same way; regardless of the con- information and communications technology
tent provider or content, companies cannot (ICT) revolution would reduce the severity
pay an Internet service provider to have of business cycles.2 The consequences of low
faster speeds. The net neutrality debate asks search costs were discussed in financial mar-
whether Internet service providers should kets (Barber and Odean 2001), labor markets
exercise control over content. Put differently, (Autor 2001), and retail markets (Borenstein
net neutrality is the norm that Netflix pays the and Saloner 2001 and Bakos 2001). The ideas
same to send a gigabyte of data to one of their in these papers have their roots in the early
customers as a small startup would pay to search literature, which modeled search
send data to the same customer. Internet ser- costs as the costs of gathering information
vices have had a historic norm of net neutral- (Stigler 1961, Diamond 1971, and Varian
ity, though this has been challenged in recent 1980). Reflecting this early focus and solid
years by Internet service providers and policy base of economic understanding, the liter-
makers in the United States and globally. The ature on the effects of lower digital search
net neutrality literature therefore empha- costs is more established than the other parts
sizes the role of the connection intermediary of the digital economics literature.
(Economides and Hermalin 2012; Bourreau,
Kourandi, and Valletti 2015; Choi, Jeon, and
Kim 2015; and Goetz 2017). As shown by Lee
and Wu (2009) and Greenstein, Peitz, and 2 “Information technology has doubtless enhanced
Valletti (2016), the particulars of the model the stability of business operations,” Federal Reserve
Chairman Alan Greenspan, February 26, 1997, testimony
matter, and the costs and benefits of net neu- before Congress. https://www.federalreserve.gov/board-
trality depend on the specific setting. docs/hh/1997/february/testimony.htm.Goldfarb and Tucker: Digital Economics 7
3.1 Are Prices and Price Dispersion Lower and shipping policies. Firms with higher qual-
Online? ity may develop stronger brands and there-
Low search costs make it easier for con- fore command higher prices (Waldfogel and
sumers to compare prices, putting downward Chen 2006).
pressure on prices for similar products. This Firms selling products can also shape the
should reduce both prices and price disper- search process. When consumers search, they
sion. Brynjolfsson and Smith (2000) compare assess multiple dimensions of information:
prices of books and CDs at four Internet-only price, quality, reputation, shipping fees, time
retailers, four offline retailers, and four to delivery, color, etc. Lynch and Ariely (2000)
“hybrid” retailers who had both online and demonstrate this for online wine purchasing
offline stores. They identified twenty books in a laboratory. If price was available on the
and twenty CDs, half of which were best- first page, consumers focused on price. If con-
sellers and half of which were randomly sumers needed to click further to learn the
selected among titles popular enough to be price, other attributes became more import-
sold in most offline stores. They showed that ant for purchase decisions. Fradkin (2017)
online prices for these items were substan- shows that the details of the search process
tially lower than offline prices. Relatively low matter in the context of short-term accommo-
online prices have been shown in a variety of dation platform Airbnb. Structural estimates
other settings, including insurance (Brown of the cost of an extra click in the consumer
and Goolsbee 2002), automotive products search process suggest they are larger than
(Scott Morton, Zettelmeyer, and S ilva-Risso might be supposed (Honka 2014 and De Los
2001), and airlines (Orlov 2011). Santos, Hortacsu, and Wildenbeest 2012).
However, though prices may be lower, This means that consumers stop searching
substantial price dispersion remains. sooner than predicted by models that assume
Brynjolfsson and Smith (2000) show this search costs close to zero.
in their online–offline retail study. Baye, In the presence of search costs, and mul-
Morgan, and Scholten (2004) use evidence tiple dimensions of information, firms can
from thousands of products and prices partly choose which information has the low-
to document large and persistent online est search costs. Ellison and Ellison (2009a)
price dispersion. Orlov (2011) finds that demonstrate that computer memory chip
the Internet increases the intrafirm disper- retailers attract customers with low prices
sion of airline prices, but had no effect on at an online price comparison website, and
interfirm price dispersion. By contrast, the then show customers other (typically higher
development economics literature measur- quality and higher margin) products once
ing the effect of mobile phones on commod- they arrive. Using data from eBay, Dinerstein
ity prices suggests that lower search costs et al. (2018) emphasizes how the design of
reduced price dispersion (Jensen 2007; Aker the search algorithm on eBay affects mark-
2010; and Parker, Ramdas, and Savva 2016). ups charged by eBay sellers. More directly,
Given evidence of the persistence of price Hossain and Morgan (2006) show that online
dispersion online, research turned to explore sellers often hide shipping fees until the final
why price dispersion does not disappear. Of purchase page. Blake et al. (2018) shows a
course comparison of online products does similar phenomenon in the information
not always compare apples to apples. In com- revealed in ticket prices at an online ticket
paring book prices, the book may be the same, platform.
but the retailer is different. Different retailers Therefore, while prices have fallen,
offer different quality, shopping experiences, price dispersion has persisted. The initial8 Journal of Economic Literature, Vol. LVII (March 2019)
redictions of low price dispersion missed
p sold to everyone, while niche products are
the point that search costs are endogenous, sold through long-tail retailers. The increase
and so firms can manipulate the search pro- in tails at the right and left of the distribution
cess in order to sustain higher margins and comes at the cost of products in the middle.
prices. The degree to which search costs generate
more or less variety depends on the search
3.2 How Do Low Search Costs Affect
process endogenously chosen by the firm.
Variety?
Recommendation engines are a key aspect
Low search costs may mean that it is easier of the online search process. Fleder and
to find rare and niche products (Yang 2013). Hosanagar (2009) demonstrate this, show-
In this case, digital search might lead to an ing that algorithms that emphasize “people
increase in the proportion of sales going to who bought this also bought” move the sales
products that are relatively rarely purchased, distribution toward superstars. If many peo-
a phenomenon dubbed “the long tail” by ple buy Harry Potter, this recommendation
Anderson (2006). Using data from a retailer engine will recommend Harry Potter to
with both online and offline channels, everyone else. In contrast, if the algorithm
Brynjolfsson, Hu, and Simester (2011) doc- emphasizes “people who bought this dis-
ument that the variety of products available, proportionately bought,” relatively unusual
and purchased, online is higher than offline. items that demonstrate niche tastes will be
Low search costs may facilitate discovery of sold. Empirically, Tucker and Zhang (2011)
relatively unknown products (Zhang 2018).3 document that popularity information
Low search costs could also generate has asymmetrically large effects for niche
superstar effects (Rosen 1981). If there are products.
vertically differentiated products and the Popularity information affects sales in
marginal cost of production is zero, then general. Many online platforms sort items
homogeneous consumers will all agree by popularity and feature popularity prom-
which product is best and buy it. Consistent inently, reducing search costs for this type
with this, Goldmanis et al. (2010) show that of information. Showing such popularity
the Internet initially led to a relative increase information affects purchase behavior not
in the number of large offline bookstores and only in retail, but also online lending (Zhang
travel agencies. and Liu 2012) and online investing (Agrawal,
Bar-Isaac, Caruana, and Cunat (2012)
Catalini, and Goldfarb 2015).
explain how superstar and long-tail effects The effect on welfare of this change in
may result from a reduction in search costs. variety is not obvious, and so it has been
If products are both vertically and horizon- the subject of a rich discussion in the liter-
tally differentiated, a reduction in search ature. Lower search costs that lead people
costs may lead to an equilibrium where the to buy the products that more closely match
most popular and highest-quality products their preferences should increase welfare.
are produced in high enough quantity to be Consistent with this, Brynjolfsson, Hu, and
Smith (2003) show that increased variety
3 In addition to search costs, variety may increase
increases consumer surplus.
because digital technologies can make inventory systems At the same time, improvements in wel-
more efficient, meaning firms can hold millions of prod- fare may be small. The increase in matching
ucts, especially for digital goods that have no physical pres- of products to preferences is, by definition,
ence. People may also be less inhibited from purchasing
nonstandard items when purchasing on a screen, rather marginal. The new products offered are the
than from a human (Goldfarb et al. 2015). products on the margin of being produced.Goldfarb and Tucker: Digital Economics 9
The superstar effects may be marginal rela- polarization of digital content because the
tive to the consumers who bought products increase in polarization is largest for demo-
in the middle because they were unwilling to graphic groups with the least Internet usage.
pay search costs. For example, Ershov (2017) Polarized media may be less concentrated,
shows that a reduction in search costs in the generating incentives for niche sources to
mobile app market reduced average product intentionally mislead. Allcott and Gentzkow
quality. On balance, however, it also shows (2017) show that false news stories about the
that the increase in variety led to a substan- 2016 presidential election were shared tens
tial increase in overall welfare despite the of millions of times, though they demon-
incremental nature of the new products. strate the fake news was unlikely to have
Aguiar and Waldfogel (2016) suggest that changed the election outcome. Long before
this marginal argument misses the substan- the attention to fake news in the 2016 elec-
tial uncertainty about product quality for tion, Antweiler and Frank (2004) examined
many information goods. In the context how anonymous, and potentially misleading,
of music, they show that several songs and online investing advice affects stock prices.
musicians that seem marginal ex ante ended Low search costs—in the absence of a reli-
up having substantial sales. Therefore, by able quality filter—meant that this informa-
enabling such music to get produced, digital tion could be more easily found and shared.
markets led to a large change in the relative Low online search costs have also trans-
sales of products. Uncertainty in the process formed the way academic research is con-
meant better and more music was created. sumed. McCabe and Snyder (2015) show
A great deal of attention has focused on that JSTOR led to an increase in citations
the increase in variety in consumption of of included articles at the expense of others.
media in particular. The Internet might also Search costs fell, but because they fell more
enable people to only read information that for some articles than others, it changed the
reflects their narrow viewpoint; despite the nature of attention to specific articles and
variety, there is no need to search widely. ideas. More starkly, Ellison (2011) argues
The latter idea has been emphasized by Cass that peer review may be in decline because
Sunstein as an “echo chamber” (Sunstein of low online search costs. In particular, he
2001). Consistent with the idea of wide vari- shows that high-profile researchers do not
ety available but consumption in echo cham- need to rely on academic journals to dissem-
bers, Greenstein and Zhu (2012) examine inate their ideas. They can post online and
the bias of Wikipedia and show that, while, people will find their work. In other words,
on aggregate Wikipedia has become less similar to the superstar effect in products,
politically biased (toward Democrats) over low search costs combined with thousands
time, the bias of articles has not changed of research articles benefit the superstar
much. Instead, the political bias has mainly researchers.
dropped because of the arrival of new, rela-
3.3 How Do Low Search Costs Affect
tively right-wing articles.
Matching?
By contrast, Gentzkow and Shapiro (2011)
show that Internet media consumption is Reduced search costs facilitate exchange
more varied than offline media consumption. more generally, often enabled by large dig-
Therefore, in this context, low search costs ital platforms. Dana and Orlov (2014) show
lead to increased variety. Boxell, Gentzkow, that airlines are better able to fill to c apacity.
and Shapiro (2017) argues that the Internet Ellison et al. (2014) show that online buy-
is unlikely to be responsible for increased ers are better able to find the specific books10 Journal of Economic Literature, Vol. LVII (March 2019)
they want. Kroft and Pope (2014) find online much of the research takes a market design
search through Craigslist decreased rental perspective. For example, Cullen and
apartment and home vacancies (though Farronato (2016) examine an online mar-
they measure no effect on unemploy- ketplace that matches buyers and sellers of
ment). Anenberg and Kung (2015) show domestic tasks, such as cleaning, moving,
that online search enabled the rise of a and simple home repair. They emphasize the
market for truck-based mobile restaurants challenges in growing both the demand and
(“food trucks”). To the extent that the liter- supply sides with respect to variation in the
ature emphasizing matching is distinct from quantity of buyers and sellers over time, econ-
search, the matching literature emphasizes omies of scale in matching, and geographic
that both sides of the market engage in the density. A key result is that demand fluctua-
search process. tions in this two-sided market lead to changes
Related to the above ideas, low search in quantity supplied rather than changes in
costs are likely to increase the quality of prices. Similarly, Hall, Kendrick, and Nosko
matches between buyers and sellers, firms (2016), Farronato and Fradkin (2018), and
and workers, etc. The labor economics lit- Zervas, Proserpio, and Byers (2017) also show
erature has emphasized that the Internet that the responsiveness in quantity supplied
should reduce unemployment and vacancies. to changes in demand conditions is a key
Kuhn and Skuterud (2004) find no effect of aspect of peer-to-peer platforms (specifically,
Internet job search on employment. Kuhn Uber and Airbnb). Low search costs provide
and Mansour (2014) revisit the analysis sev- market demand information that enables sup-
eral years later with updated data and find ply to enter the market when needed.
that individuals that used the Internet in job
3.4 Why Are Digital Platform-based
search were indeed more likely to match to
Businesses So Prevalent?
an employer.
The reduced costs of search have led to the Platforms are intermediaries that
development of online “peer-to-peer” plat- enable exchange between other players.
forms dedicated to facilitating matching. The Digitization has led to an increase in the
variety of such online matching markets is prevalence of platform businesses, even
extraordinary: workers and firms, buyers and beyond the p eer-to-peer platforms discussed
sellers, investors and entrepreneurs, vacant above. Most of the major technology firms
rooms and travelers, charities and donors, can be seen as platform-based businesses.
dog walkers and dog owners, etc. Several of For example, Apple provides hardware and
these markets have been dubbed the “shar- software platforms for others to build appli-
ing economy” because people are able to use cations around. Google provides platforms
unused objects or skills better. Most “sharing for bringing together advertisers and poten-
economy” platforms are not sharing in the tial buyers.
sense learned by kindergarteners: custom- As highlighted in Jullien (2012), there are
ers typically pay for the “shared” services. two main reasons digital markets give rise to
Horton and Zeckhauser (2016) emphasize platforms. First, platforms facilitate match-
that many of these markets are driven by ing. In particular, as in the sharing economy
an unused capacity for durable goods. Low platforms, they provide a structure that can
search costs enable such unused capacity to take advantage of low search costs to create
be filled more efficiently. efficient matches. Often platforms serve as
In a review of the peer-to-peer mar- intermediaries between buyers and sellers, as
kets literature, Einav et al. (2018) note that highlighted in Nocke, Peitz, and Stahl (2007)Goldfarb and Tucker: Digital Economics 11
and Jullien (2012). In the context of a central istance. On the other hand, Garicano (2000)
d
role of matching, a rich theory literature has also shows that low-cost communication
arisen that examines competition and pricing could decrease centralization by enabling
strategy in such platform businesses, with an front-line employees to access information
emphasis on the importance of indirect net- previously only available to senior employ-
work effects (for example Baye and Morgan ees at headquarters. A variety of papers have
2001, Caillaud and Jullien 2003, Weyl 2010, explored nuances in this trade-off within
Hagiu and Jullien 2011, and de Corniere organizations, emphasizing the importance
2016). of the particular technology studied.
Second, platforms increase the efficiency Bloom et al. (2014) test this theory directly,
of trade. They do this through lower search using data on European and American
costs as well as other aspects of digitization manufacturing firms to show that informa-
that we discuss below: low reproduction and tion technology is a centralizing force and
verification costs. Hagiu (2012) emphasizes communication technology is a decentral-
how software platforms enable application izing force. Acemoglu et al. (2007) also dis-
providers to serve a large number of cus- cuss the decentralizing role of information
tomers quickly, with the only requirement technology. For example, Forman and van
that the application serve some particular Zeebroeck (2012) shows that digital commu-
customer need, reproduce at zero cost, and nication increases in research collaboration
rely on the platform and the other applica- across establishments within an organization.
tions to serve other needs. Interoperability Baker and Hubbard (2003) examines the
is therefore a key aspect of platforms. There effect of on-board computers on asset own-
is a large literature on the topic, as reviewed ership in the trucking industry. They empha-
in Farrell and Simcoe (2012). A key contri- size tracking costs more than search costs
bution of this literature is the emphasis on and find that aspects of o n-board comput-
the strategic nature of decisions on interop- ing that improve monitoring pushed truck-
erability and standards (Rysman and Simcoe ing firms to more ownership of trucks while
2008 and Simcoe 2012). A related set of aspects of on-board computing that improve
questions examines whether market partic- real-time location information pushed
ipants will “multi-home” and use multiple trucking firms to less ownership of trucks.
platforms (Rochet and Tirole 2003, Rysman Therefore, while adoption of digital tech-
2007, Halaburda and Yehezkel 2013). nology led to improved efficiency, the effect
on organization of the firm in equilibrium
3.5 How Do Low Search Costs Affect the
depends on the nature of the technology
Organization of the Firm?
and how its specific features affect trade-offs
Lucking-Reiley and Spulber (2001) discuss between competing tensions at the bound-
several hypotheses with respect to the effect ary of the firm. McElheran (2014) examines
of the Internet on firm structure in terms of the decision to centralize or delegate IT
the role of online intermediaries and vertical adoption decisions within firms. Firms with
integration. This literature emphasizes infor- a greater need for integrated processes (dig-
mation flow generally, in which search is one ital or otherwise) delegate less. Forman and
key type of information flow. Garicano (2000) McElheran (2013) show that this tendency is
shows that low-cost digital information flow mitigated by the ease with which IT enables
could increase centralization by enabling coordination across firms, so that disintegra-
headquarters, and organizational leaders, tion of the firm boundary can be seen as an
to understand better what is happening at a extreme form of delegation.12 Journal of Economic Literature, Vol. LVII (March 2019)
In addition to the effect on the domes- be shared without diminishing the original
tic boundaries of the firm, the reduction in information.
search costs (combined with the reduction In the absence of deliberate legal or tech-
in verification costs discussed below) has nological effort to exclude, bits can be repro-
also led to an increase in international hiring duced by anyone—not just the producing
and outsourcing. While international out- firm—at near zero cost without degrading
sourcing is not a new phenomenon (Leamer the quality of the initial good. As Shapiro and
2007), the recent rise of digital international Varian (1998, p. 83) put it, the Internet can
labor market platforms suggests a different be seen as a “giant, out of control copying
avenue for international hiring. Agrawal, machine.”
Lacetera, and Lyons (2016) show that online Nevertheless, the economics of zero mar-
platforms with standardized information dis- ginal cost, n
on-rival goods can shift things in
proportionately benefit workers from devel- favor of producers, consumers, or both. In a
oping countries. The objective information static model, as marginal costs fall the poten-
available online, combined with the ability to tial surplus rises, and so the welfare effect
send the output of the work (typically infor- depends on the final price and associated
mation such as data or software code) for free deadweight loss. The final price and dead-
over long distance helps workers who are far weight loss depend on legal and technological
from the buyer. Such online labor markets tools for exclusion (Cornes and Sandler 1986),
have several important challenges. Using which relate to the ability to track behavior—
data from online labor markets, Lyons (2017) the subject of a later section. In this section,
shows that c ross-cultural international teams we emphasize that the underlying technology
can be less productive because of commu- enables firms and governments to make a
nication challenges. Relatedly, Ghani, Kerr, choice not to exclude. This can allow individ-
and Stanton (2014) show that employers in uals to enjoy the full benefits of the non-rival
the Indian diaspora are more likely to hire nature of information-based goods.
Indians online.
4.1 How Can N on-Rival Digital Goods Be
Priced Profitably?
4. The Replication Cost of Digital Goods
The n on-rival nature of digital goods has
Is Zero
led to questions of how to structure pric-
The key shift in the production function ing of a large variety of n on-rival zero-cost
is not that digital goods have a marginal cost goods, should a producer choose to charge.
of zero. Simple microeconomic models with Bundling occurs when two or more products
zero marginal cost are not so different from are sold together at a single price (Shapiro
models with positive marginal cost. The and Varian 1998, Choi 2012). Bundling
demand curve slopes downward and firms models have a long history in economics.
price where marginal revenue equals zero. Stigler (1964) and Adams and Yellen (1976)
Instead, a key distinction between goods note that a sufficient condition for price dis-
made of atoms and goods made of bits is that crimination benefit of bundling arises when
bits are n
on-rival, meaning that they can be consumers have negatively correlated pref-
consumed by one person without reducing erences. Some people may value an action
the amount or quality available to others. A movie at $10 and a romance at $2. Others
common analogy for n on-rival goods is that may value the romance at $10 and the action
just as one person can start a fire without movie at $2. Selling the bundle at $12 yields
diminishing another’s fire, information can higher profits than selling the action andGoldfarb and Tucker: Digital Economics 13
romance movies separately. The challenge of non-rival public digital goods are open-
for firms is to identify such negative correla- source software and Wikipedia. Both cases
tions in preferences to identify when bun- involve a deliberate decision not to exclude,
dling will increase profits. and applying established models is somewhat
Bakos and Brynjolfsson (1999, 2000) rec- less straightforward than the bundling models
ognize that, under certain assumptions, with highlighted in the preceding subsection.
enough goods and independent preferences, Lerner and Tirole (2002) ask why software
this challenge is overcome. Furthermore, developers would freely share their code with
the non-rival nature of information goods no direct payment. They emphasize two core
means that large numbers of information benefits from open source that do not appear
goods can be bundled without substantially in standard models of public goods. For
increasing costs. Therefore, a simple and individual developers, providing high-qual-
useful insight on the economics of n on-rival ity open-source code is a way to signal their
information goods is that it will sometimes skills to potential employers. For companies,
be optimal to bundle thousands of digital improving the quality of open source software
products together. may allow them to sell other services that
Chu, Leslie, and Sorensen (2011) use an are complementary to open-source software
empirical example to show that the intuition (such as hardware or consulting services)
of Bakos (1999) applies to relatively small at a premium. Underlying these core bene-
numbers of goods in the bundle. There are fits is the non-rival nature of the code: digi-
also strategic reasons to bundle because it can tal distribution through the Internet means
reduce competition (Carbajo, de Meza, and that (high-quality) open-source contributions
Seidmann 1990). When bundling has zero can be widely adopted. The literature on the
marginal cost, such strategic considerations economics of open source that followed has
can become particularly relevant (Carlton, largely supported their hypotheses of career
Gans, and Waldman 2010; Choi 2012). concerns and complementarity (Johnson
Despite the extensive theory work, it is only 2002; Bitzer and Schroder 2005; Mustonen
recently that empirical examples of such mas- 2005; Lerner, Pathak, and Tirole 2006; Henkel
sive bundles appeared in the literature, in the 2009; Xu, Nian, and Cabral 2016).
form of subscription services such as Netflix Wikipedia represents a different import-
for video and Spotify and Apple Music for ant context for the puzzle of why people
music. Aguiar and Waldfogel (2018) show that contribute to digital public goods. Zhang and
Spotify displaces sales but it also displaces Zhu (2011) emphasize social benefits related
“piracy,” or the downloading of music without to breadth of readership. In the context of
permission from the copyright holder. They Chinese-language Wikipedia, they show that
estimate that the reduction in sales and the users care about audience size, and decrease
increase in legal music consumption balance contributions when part of the audience is
each other so that Spotify appears to be reve- blocked due to Chinese government policy.
nue neutral in the 2013–15 time period. Consistent with this idea of a social benefit,
Aaltonen and Seiler (2016) and Kummer,
4.2 What Are the Motivations for Providing
Slivko, and Zhang (2015) together provide
Digital Public Goods?
evidence for a virtuous circle in which more
Information providers can deliberately editing leads to more views and more views
decide not to exclude. It is somewhat of a lead to more editing. Contributions are likely
puzzle why private actors would choose to related to the interests of the contributors:
create public goods. Two prominent examples Wikipedia leaned sharply Democratic early14 Journal of Economic Literature, Vol. LVII (March 2019)
on and has gradually become more neutral (Gordon and Loeb 2002 and G al-Or and
(Greenstein and Zhu 2012). Ghose 2005), especially if costly investments
Nagaraj (2016) suggests the potential for in data security also are a public good.
government sponsorship of digital public While digital technology creates public
goods. He finds that open mapping infor- goods, zero marginal cost of production can
mation led to a substantial increase in min- also create public bads, such as spam (Rao
ing activity, particularly for smaller firms and Reiley 2012) and online crime (Moore,
with fewer resources. Therefore, open data Clayton, and Anderson 2009). These have
enabled a wider set of participants to succeed. led to policy responses such as the US
More generally, the non-rivalrous nature Controlling the Assault of Non-Solicited
of digital technology could enable consum- Pornography And Marketing Act of 2003
ers and workers in developing countries to (CAN-SPAM). Another example of digital
access the same information as people in spam is junk telephone calls, the automation
developed countries, conditional on having of which has been enabled by digital technol-
access to the Internet. In the context of edu- ogies. Petty (2000) and Varian, Wallenberg,
cation, Kremer, Brannen, and Glennerster and Woroch (2005) evaluate the role of the
(2013) argue that information technology can federally sponsored “Do Not Call” list in pre-
improve pedagogy in the developing world. venting potentially intrusive direct sales calls
Underlying their argument is an emphasis and find positive effects.
on non-rival, non-excludable digital informa- That said, the economics of such bads
tion, and the public Internet-based posting are relatively straightforward. In contrast,
of educational materials. Correspondingly, the more challenging policy question for
Acemoglu, Laibson, and List (2014) empha- non-rival digital goods is whether the gov-
sizes that digital education will lead to a more ernment should intervene through copyright
equal distribution of educational resources. policy to enforce excludability despite the
There are, however, situations in which non-rival nature of the goods.
welfare may decrease because of a decision
4.3 How Do Digital Markets Affect
not to exclude digital goods from wide-
Copyright Policy?
spread copying. The decision not to exclude
non-rival goods can reduce the incentives to As the Internet first diffused in the late
produce information goods, a subject we dis- 1990s, copyrights of music (and text) were
cuss below in the context of copyright pol- often ignored as people freely posted copy-
icy. It can also create negative externalities. righted goods online. Because of the non-rival
For example, Acquisti and Tucker (2014) nature of digital information, one posted
show that policies that mandate “open data” copyrighted item could be useful to millions
by government may lead to data leakages of people, potentially replacing sales. At the
(or privacy breaches) that affect individ- same time, music industry revenue began to
uals’ welfare offline. Openness, almost by fall (Waldfogel 2012a) and this was widely
definition, implies a reduction in privacy. blamed on changes brought by the Internet.
Relatedly, Acquisti and Gross (2009) show Optimal enforcement of copyright has
that using public data online makes it pos- therefore been a key focus of the digital eco-
sible to predict an individual’s social security nomics literature. The early work focused
number. This feeds back, in general, to the on the revenue consequences of free online
idea that while n on-excludability may be copying. This was referred to as “file sharing”
attractive in principle, it can lead to ques- to those who believe it should be allowed,
tions of appropriate data security practices and as “piracy” by those who didn’t. TheGoldfarb and Tucker: Digital Economics 15 direct effect of free online copying of media so quality rose. Results are similar in movies is that revenues from the sale of copies of (Waldfogel 2016) and books (Waldfogel and that media fall. At the same time, revenues Reimers 2015). This contrasts with the eco- could rise if the free copies are merely sam- nomic history literature, which suggested that pled and consumers buy what they like (Peitz copyright alone could increase the quality of and Waelbroeck 2006). Revenues could also creative output (Giorcelli and Moser 2016). rise for complementary goods like live per- In addition to affecting incentives to inno- formances (Mortimer, Nosko, and Sorensen vate, digital challenges to copyright protec- 2012). Finally, revenues could rise if the free tion may affect incentives to build on prior copies are limited to developing markets for work. Williams (2013) demonstrates this products with network effects (Takayama point in a different intellectual property 1994). Empirically, though a small num- context and shows that intellectual property ber of studies have found positive effects protections limit follow-on innovation in ( Oberholzer-Gee and Strumpf 2007), most gene sequencing. Heald (2009) shows that studies have found that free online copy- copyrighted music is less used in the movies ing reduces revenues in music (Rob and than non-copyrighted music. Nagaraj (2018) Waldfogel 2006, Zentner 2006, Liebowitz shows that copyright protection of old sports 2008, and Waldfogel 2010), in video (Rob and magazines reduces the quality of Wikipedia Waldfogel 2007; Liebowitz and Zentner 2012; pages decades later. This phenomenon is Danaher, Smith, and Telang 2014b; Danaher not unique to the digital context. Biasi and and Smith 2014; and Peukert, Claussen, and Moser (2018) show that eliminating copy- Kretschmer 2017), and in books (Reimers rights of German books during World War II 2016). This echoes a non-digital historical led to a substantial increase in US scientific literature (Li, MacGarvie, and Moser 2015; output, measured by PhDs in mathematics and MacGarvie and Moser 2015) suggesting and patents that cited the German books. a continuity between policy governing digital Another challenge for copyright policy technologies and earlier policies. driven by the shift in costs of replication is that How does copyright affect the creation of it has made it easier for other firms to repli- new works? This is a more difficult research cate digital content and attempt to aggregate question, as it requires some attempt to it. This practice has been particularly preva- measure counterfactual quality and quan- lent in the news media, where policy makers tity of goods had copyright law not existed have been encouraged to take action to pro- (Varian 2005; Waldfogel 2012b; and Danaher, tect the interests of the newspapers that actu- Smith, and Telang 2014b). Waldfogel (2012a) ally originated this news content. However, addresses this challenge using two measures in general the work in economics that has of music quality: historical “best albums” lists evaluated the effect of these aggregators has and usage information over time. In both been to emphasize that such aggregation pro- cases, he shows that the quality of music motes more exploration, rather than neces- began to decline in the early 1990s and sarily cannibalizing content (Calzada and Gil stopped declining after the arrival of free forthcoming; Chiou and Tucker 2017; Athey, online copying in 1999. Why did quality rise Mobius, and Pal 2017). despite declining revenue? He argues that Overall, copyright law is more import- simultaneously with the decline in revenue ant in digital markets because goods can be came a decline in the cost of producing and copied at zero cost. Stricter enforcement of distributing music. Digitization affected the copyright appears to increase revenue to the supply side as well as the demand side, and copyright holder, increase some incentives
16 Journal of Economic Literature, Vol. LVII (March 2019)
by potential copyright holders to innovate, 5.1 Does Distance Still Matter If
but reduce incentives by others to build Transportation Costs Are Near Zero?
on copyrighted work. Nevertheless, the
literature also shows that, despite ease of Low transportation costs for information
copying, digitization has not killed creative mean that the cost of distribution for digital
industries because production and distribu- goods approaches zero and the difference in
tion costs have fallen and because the tech- cost of nearby and distant communication
nology has caught up to facilitate copyright approaches zero.
enforcement. The potential implications of low transpor-
tation costs have been explored in the popular
press. Cairncross (1997) suggests that this fall
5. Lower Transportation Costs
in the costs of transporting information would
Related to replication being costless, the lead to a “death of distance.” Isolated indi-
cost of transporting information stored in viduals and companies would be able to plug
bits over the Internet is near zero.4 Put dif- into the global economy. Rural consumers
ferently, the cost of distribution for digital would benefit by having access to the same
goods approaches zero and the difference in set of digital products and services as every-
the cost of nearby and distant communica- one else. There would be a global diffusion of
tion approaches zero. knowledge. Friedman (2005) identifies sev-
In addition, digital purchasing technol- eral of the same themes in predicting a “flat
ogies have reduced transportation costs. world,” in which businesses anywhere could
Consumers buy physical goods online, partic- plug into the global supply chain and produce.
ularly when offline purchasing is costly or dif- Being in the United States would not confer
ficult (Goolsbee 2000; Forman, Ghose, and a meaningful advantage relative to India.
Goldfarb 2009; and Brynjolfsson, Hu, and Both Cairncross and Friedman suggested the
Rahman 2009). Furthermore, Pozzi (2013) potential arrival of a global culture, in which
shows that consumers also use online shop- everyone everywhere would consume the
ping to overcome the transportation costs of same information, an idea with its roots in
carrying things from the store. In this way, the McLuhan (1964). This idea is implicit in the
Internet facilitates stockpiling, allowing peo- trade model of Krugman (1979): countries
ple to buy in bulk when a discount appears consume the same goods as transport costs
because delivery means there is no need to approach zero. Rosenblat and Mobius (2004)
carry the large quantity of items purchased. formalize some of these ideas in a different
Therefore, for information, digital goods, context, using a network model of collabora-
and physical goods, transportation costs are tion in which long distance collaboration rises
lower online. but coauthor similarity in other dimensions
(such as field of research) also rises.
4 While transportation costs could be positive and even
A less extreme question than “Is distance
high due to network congestion, in practice this has not dead?” is “Does distance matter more or less
been an issue. Early on, such network congestion was a than it used to?” The most definitive answer
key focus of the literature. For example, one of the first to that question comes from Lendle et al.
volumes on Internet economics, Mcknight and Bailey
(1998), has several articles on congestion pricing. This (2016). They compare cross-border sales
early literature on backbone competition and congestion on eBay with international trade data. They
ended up influencing our understanding of the economics demonstrate that, while distance predicts
of net neutrality discussed above (Cremer, Rey, and Tirole
2000; Laffont et al. 2001; Besen et al. 2001; and Laffont both online and offline trade flows, distance
et al. 2003). matters substantially less on eBay.Goldfarb and Tucker: Digital Economics 17 The digital economic literature has empha- retailers went online first. In their review sized what factors influence the extent to of the literature on online–offline compe- which distance still matters. tition, Lieber and Syverson (2012) provide As Lemley (2003) notes, “No one is ‘in’ some additional evidence that offline options cyberspace” (p. 523). Therefore, offline affect online purchasing. Similarly, in the options matter. Balasubramanian (1998) digital media context, evidence suggests that examines the importance of offline options online media consumption substitutes for, using a circular city/Salop (1979) model with and is replacing, offline media consumption the cost of using the direct retailer as con- (Wallsten 2013 and Gentzkow 2007). stant for all locations, but the cost of using the In addition to the offline option, the fact stores located around the circle dependent on that tastes are spatially correlated also mat- transportation costs. The model shows that ters for the persistent role of distance. Blum the benefit of a direct (online) retailer will be and Goldfarb (2006) examine the interna- largest for those who live far from an offline tional Internet surfing behavior of about retailer. Forman, Ghose, and Wiesenfeld 2,600 American Internet users, and demon- (2008) provide evidence to support this strate that Internet surfing behavior is con- model, demonstrating that when a Walmart or sistent with the w ell-established empirical Barnes & Noble opens offline, people substi- finding in the trade literature that bilateral tute away from purchasing books on Amazon. trade decreases with distance (Overman, A number of other studies also demonstrate Redding, and Venables 2003; Anderson and how offline retail affects online purchasing. van Wincoop 2004; and Disdier and Head Related models include Loginova (2009) and 2008). In other words, even for a product Dinlersoz and Pereira (2007), which examine with zero shipping costs (visiting websites), the role of loyalty to the offline store in driv- people are more likely to visit websites from ing the more price sensitive customers online. nearby countries than from faraway coun- Empirically, Brynjolfsson, Hu, and Rahman tries. This relationship between distance (2009) show that online sales at a women’s and website visits is much higher in taste-de- clothing retailer are lower from places with pendent categories (and loses statistical many offline women’s clothing stores. This significance in the non-taste-dependent cat- effect is driven by the more popular prod- egories). Distance matters because it prox- ucts that are likely to be available in a typical ies for taste similarity. Alaveras and Martens offline store. Choi and Bell (2011) shows that (2015) replicates this core result using much online sales of niche diaper brands are higher richer data on website visits by users in a large in places where they are unlikely to be avail- number of countries. Sinai and Waldfogel able offline. Goolsbee (2001), Prince (2007), (2004) also shows that highly populated areas and Duch-Brown et al. (2017) all show sub- produce more content, and that because stitution between online and offline sales tastes are spatially correlated in the sense of personal computers. Gentzkow (2007) that people are more likely to consume local demonstrates substitution between the media than distant media, people in highly online and offline news in Washington DC. populated areas are particularly likely to go Seamans and Zhu (2014) and Goldfarb and online. This geographically specific nature Tucker (2011a, 2011d) demonstrate substitu- of tastes is also reflected in the consumption tion between online and offline advertising. of digital goods such as music (Ferreira and Gertner and Stillman (2001) show how chan- Waldfogel 2013) and content (Gandal 2006). nel conflict interacts with vertical integration Quan and Williams (2018) demonstrate that and show that vertically integrated apparel accounting for spatial correlation in tastes
You can also read
