Varian (2010) introduces the idea of computer mediated transactions (CMT). What is a CMT and are the four types he talks about? What is surveillan

Computer Mediated Transactions

Author(s): Hal R. Varian

Source: The American Economic Review , May 2010, Vol. 100, No. 2, PAPERS AND PROCEEDINGS OF THE One Hundred Twenty Second Annual Meeting OF THE AMERICAN ECONOMIC ASSOCIATION (May 2010), pp. 1-10

Published by: American Economic Association

Stable URL: https://www.jstor.org/stable/27804953

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected] Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to The American Economic Review

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

American Economie Review: Papers & Proceedings 100 (May 2010): 1-10 http://www.aeaweb.org/articles.php?doi=10.1257/aer. 100.2.1

RICHARD T. ELY LECTURE

Computer Mediated Transactions

By Hal R. Varian*

Every now and then a set of technologies becomes available that sets off a period of "combinatorial innovation." Think of standard ized mechanical parts in the 1800s, the gasoline engine in the early 1900s, electronics in the 1920s, integrated circuits in the 1970s, and the Internet in the last decade or so.

The component parts of these technologies can be combined and recombined by innovators to create new devices and applications. Since these innovators are working in parallel with similar components, it is common to see simul taneous invention. There are many well-known examples, such as the electric light, the airplane, the automobile, and the telephone. Many schol ars have described such periods of innovation, using terms such as "recombinant growth," "general purpose technologies," "cumulative synthesis," and "clusters of innovation."1

The Internet and the Web are wonderful examples of combinatorial innovation. In the last 15 years we have seen a huge proliferation of Web applications, all built from a basic set of component technologies.

The Internet itself was a rather unlikely inno vation; I like to describe it as a "lab experiment that got loose." Since the Internet arose from the research community rather than from the pri vate sector, it had no obvious business model. Other public computer networks, such as AOL, CompuServe, and Minitel, generally used sub scription models but were centrally controlled and offered little scope for innovation at the user level. The Internet won out over these alterna

tives, precisely because it offered a flexible set

*UC Berkeley and Google, 1600 Amphitheatre Parkway, Mountain View, CA 94043 (e-mail: [email protected]

1 See, for example, Weitzman (1998), Bresnahan and Trajtenberg (1995), Bresnahan (forthcoming), Rosenberg (1976), Usher (1998), and Schumpeter (2000).

of component technologies which encouraged combinatorial innovation.

The earlier waves of combinatorial innovation

required decades or more to play out. For exam ple, Hounshell (1984) argues that the Utopian vision of interchangeable parts took well over a century to be realized. The Web was invented in the early 1990s but didn't really become widely used till the mid-1990s. Since then we have seen

a huge number of novel applications, from Web browsers, to search engines, to social networks, to mention just a few examples. As with the Internet, the Web initially had no real business

model but offered a fertile ground for combina torial innovation.

Why was innovation so rapid on the Internet? The reason is that the component parts were all bits. They were programming languages, pro tocols, standards, software libraries, produc tivity tools and the like. There was no time to

manufacture, no inventory management, and no shipping delay. You never run out of HTML, just like you never run out of e-mail. New tools could be sent around the world in seconds, and innovators could combine and recombine these

bits to create new Web applications. This parallel invention has led to a burst of

global innovation in Web applications. It is true that the Internet was an American innovation, but the Web was invented by an Englishman liv ing in Switzerland. Linux, the most used oper ating system on the Web, came from Finland, as did MySQL, a widely used database for Web applications. Skype, which uses the Internet for voice communication, came from Estonia.

Of course there were many other technologies with worldwide innovation such as automobiles, airplanes, photography, and incandescent light ing to name just a few. However, applications for the Internet, which is inherently a commu nications technology, could be developed every where in the world in parallel, leading to the very rapid innovation we have observed.

1

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

2 AEA PAPERS AND PROCEEDINGS MA Y 2010

My interest in this lecture is in the economic aspects of these technological developments. I start with a point so mundane and obvious, it barely seems worth mentioning.

I. Computer Mediated Transactions

Nowadays, most economic transactions involve a computer. Sometimes the computer takes the form of a smart cash register, some times it is part of a sophisticated point of sale system, and sometimes it is a Web site. In each of these cases, the computer creates a record of the transaction.

The record-keeping role was the original motivation for adding the computer to the trans action. Creating a record of transactions is the first step in building an accounting system, thereby enabling a firm to understand how its business is doing.

But now that these computers are in place, they can be used for many other purposes. In this lecture I would like to explore some of the ways that computer mediation can affect eco nomic transactions. I argue that these computer mediated transactions have enabled significant improvements in the way transactions are car ried out and will continue to impact the econ omy for the foreseeable future.

I classify impact of computer mediated trans actions into four main categories.

Facilitate new forms of contract; facilitate data extraction and analysis; facilitate controlled experimentation; facilitate personalization and customiza tion.

II. Enable New Forms of Contract

Contracts are fundamental to commerce. The

simplest commercial contract says "I will do X if you do Y," as in "I will give you $1 if you give

me a cup of coffee." Of course, this requires that the actions be verifiable. Just because I ask

for coffee doesn't mean that I will get some. As Abraham Lincoln supposedly remarked, "If this is coffee, please bring me some tea; but if this is tea, please bring me some coffee."

A computer in the middle of a transaction can observe and verify many aspects of a transaction. The record produced by the computer can allow the contracting parties to condition the contract

on terms that were previously unobservable, thereby allowing for more efficient transactions.

I am not claiming that increased observability will necessarily lead to more efficient contracts. There are counterexamples to the claim that "more information is better" such as the famous

Hirshleifer (1971) example. I am only claiming that additional information allows for more effi cient contracts.

Of course, the study of contracts is a highly developed field in economics, and it is hardly novel to suggest that contractual form depends on what is observable. What is interesting, I think, is the way that progress in information technology enables new contractual forms.

Consider, for example, a rental-car agency that buys insurance based on accident rates, and that accident rates in turn depend on the speed at which a vehicle is operated. All rent ers would prefer to drive within the speed limit if they were compensated with a lower rental fee. However, if there is no way to monitor the speed of the rental car, such a contractual provi sion is unenforceable. Putting a computer trans

mitter in the trunk of the car that records the

vehicle's speed makes the contract enforceable and offers a Pareto improvement on the original arrangements.2

The transportation sector has capitalized on the availability of computerized transmitters to create more efficient contracts in a number of areas.

Car dealers are selling cars with "starter interrupt" devices that inhibit operations if car payments are missed (Press 2006).3 Similar interrupt devices attached to breath analyzers are mandated for drunk driving offenders in many states. Parents can buy a device known as "MyKey" which allows them to limit auto speed, cap the volume on the radio, require seat belt use and encourage other safe-driv ing habits for teenage drivers (Bunkley and Vlasic 2008).4

2 This is a particularly simple case. If drivers have het erogeneous preferences, those who prefer to speed may be made worse off by the availability of such a device.

3 Los Angeles Times. 2006. "For Some High-Risk Auto Buyers, Repo Man Is a High-Tech Gadget." November 18.

4Bunkley, Nick and Bill Vlasic. 2008. "Ensuring Junior Goes for a Mild Ride." New York Times, October 7.

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

VOL. 100 NO. 2 RICHARD T. ELY LECTURE 3

In the economics literature, Hubbard (2000) and Baker and Hubbard (2004) examine a variety of ways that vehicular monitoring systems have impacted the trucking industry.

There are many other examples of computer mediated contracts. One nice example is the work of Dana and Spier (2001) and Mortimer (2008) that describes the efficiency gains result ing from revenue sharing in the videotape rental industry.

Videotapes were originally purchased by retail stores from distributors for about $65 per tape. Since the videos were so expensive, stores bought only a few. As a result, the popular vid eos quickly disappeared from the shelves, mak ing everyone unhappy.

In 1998, retailers and distributors adopted a new business model, a revenue sharing arrange ment in which the stores paid a small upfront fee of $3 to $8 but split the revenue when the video was rented, with 40 percent to 60 percent going to the retailer. Stores no longer had an incentive to economize on purchase, and all parties to the transaction?retailers, distributors, and custom ers?were made better off.

Sharing revenue at point of sale requires that both parties be able to monitor the trans action. The technological innovations of bar code scanning, the computerized cash regis ter, and dial-up computer networks were the technologies that enabled revenue sharing arrangements.

Of course, when a transaction takes place online, revenue sharing is much easier. Online advertising is a case in point where revenue from an advertiser for an ad impression or click may be split among publishers, ad exchanges, ad networks, affiliates and other parties based on contractual formulas.

I have emphasized the benefits from com puters offering more information to contract ing parties, but there are also cases where computers can be used to improve contractual performance by hiding information, by using cryptographic methods. A picturesque exam ple is the "cocaine auction protocol" which describes an auction mechanism designed to hide as much information as possible (Stajano and Anderson 1999).

Finally, I should mention "algorithmic game theory," an exciting hybrid of computer science

and economic theory. This subject brings com putational considerations to game theory (how a particular solution can be computed) and strategic considerations to algorithm design (is a particular algorithm actually incentive com patible?). See Nisan, Roughgarden, Tardos, Vazirani, eds (2007) for a comprehensive col lection of articles and Varian (1995) for an early contribution.5

A. Some History of Monitoring Technologies

Though I have emphasized computer medi ated transactions, the definition of computer can be considered to be quite broad. The earliest example I have been able to find for an account ing technology that enabled new forms of con tract involves Mediterranean shipping circa 3300 BC.

The challenge was how to write a "bill of lad ing" for long distance trade in societies that were pre-literate and pre-numerate. The brilliant solu tion was to introduce small clay tokens, known as "bullae," which were small representations of the material being transported. As each barrel of olive oil was loaded onto a ship, a barrel-shaped token was placed in clay envelope. After the load ing was completed, the envelope was baked in a kiln and given to the ship's captain. At the other end of the voyage, the envelope was broken open and the tokens were compared to the barrels of oil on the ship as they were unloaded. If the numbers matched, the contract was verified. Later marks were scratched on the outside of the bullae that indicated the number of tokens inside, and some authors believe that this innovation led to the

invention of writing between 3400 and 3300 BC. (Glassner, Bahrani and de Miero 2005). A somewhat more recent example is the

invention of the cash register in 1883 by James Ritty. Ritty was a saloon owner who discov ered that his employees were stealing money. He hit upon the idea of building a device which would record each transaction on a paper tape, an invention that he patented under the name of "the incorruptible cashier" (Patent 271,368). Ritty's machine formed the basis for the National Cash Register company, founded in 1884. The NCR device added a cash drawer

5 Varian, Hal R. 2005. "Technology Levels the Business Playing Field," New York Times, August 25.

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

4 AEA PAPERS AND PROCEEDINGS MA Y 2010

and a bell that sounded "ka-ching" whenever the drawer was opened, which alerted the owner to the transaction, thereby discouraging pilfering. It is thought that this improved monitoring tech nology made retailers willing to hire employees outside the immediate family, leading to larger and more efficient establishments. See Yates

(2000) for a more detailed account of the role of office machinery in the development of com mercial enterprises.

B. Online Adverlising

Online advertising serves as a poster child for algorithmic mechanism design. A Pasadena company called GoTo came up with the idea of ranking search results using an auction. Users did not find this particular form of search attractive, so GoTo switched to using their auc tion to rank advertisements. In the original auction, ads were ranked by "bid per click" and advertisers paid the amount they bid. After consultation with auction theorists, GoTo moved to second-price auction: an advertiser paid a price per click determined by the bid of the advertiser in the next lower position. See Battelle (2005) and Levy (2009) for accounts of the development of these auctions.

There is a fundamental divergence of incen tives in advertising. The publisher (the content provider) has space on its the Web page for an ad and it wants to sell these ad impressions to the highest bidders. The advertiser doesn't care directly about ad impressions but does care about visitors to its Web site, and ultimately about sales of its products. So the publisher wants to sell impressions, but the advertiser wants to buy clicks.

This is like an international trade transaction

where the buyer wants to pay in euros but the seller wants to receive dollars. The solution in

both cases is the same: an exchange rate. In the context of online advertising the exchange rate is the predicted clickthrough rate, an estimate of how many clicks a particular ad impression will receive. This allows one to convert the advertis

er's offered bid per click to an equivalent bid per impression, allowing the publisher to sell each impression to the highest bidder.

This mechanism aligns the interests of the buyer and seller but creates other problems. If the advertiser pays only for clicks, then it has no direct incentive to economize on impressions.

But excessive impressions impose an attention cost on users, so some further attention to ad quality is important in order to ensure that ad impressions remain relevant to users. Nowadays, the major providers of search

engine advertising all estimate clickthrough rates along with other measures of ad quality and use auctions to sell ads. Economists have applied the tools of game theory and mecha nism design to analyze the properties of these auctions. See, for example, Athey and Ellison (2007), Edelman, Ostrovsky and Schwartz (2007), Varian (2007), and Varian (2009).

III. Enable Data Extraction and Analysis

The data from computer mediated transac tions can be analyzed and used to improve the performance of future transactions.

An illustrative example is Sabre air passen ger reservation system offered by American Airlines. The original conception in 1953 was simply to automate the creation of an airline res ervation. However, by the time the system was released in 1960, it had become apparent that such a system could also be used to study pat terns in the airline reservation process: the acro nym Sabre stands for Semi-Automatic Business Research Environment (Sabre 2009).

The existence of airline reservation systems enabled sophisticated differential pricing (also known as "yield management") in the trans portation business. See Smith, Leimkuhler and Darrow (1992) for the history of yield manage ment in the airline industry and Talluri and van Ryzin (2004) for a textbook treatment.

Many firms have built data warehouses based on transaction level data which can then be used as input to analytic models for customer behavior. A prominent example is supermarket scanner data, which has been widely used in economic analysis (see Nevo and Wolfram 2002, Hendel and Nevo 2006 for just two examples). Scanner data has also been useful in constructing price indexes (Feenstra, Shapiro, eds. 2003; Farm Foundation 2003), since it allows for much more direct and timely access to prices. The fact that the data is timely is worth emphasizing, since it allows for real time analysis and intervention at both the busi ness and policy level.

Choi and Varian (2009b) and Choi and Varian (2009a) use real-time publicly available

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

VOL. 100 NO. 2 RICHARD T. ELY LECTURE 5

search engine data to predict the current level of economic activity for automobile, real estate, retail trade, travel, and unemployment indica tors. There are many other sources of real-time data, such as credit card data, package delivery data, and financial data. Clements and Hendry (2003) and Castle and Hendry (2009) have coined the term "nowcasting" to describe the use of real-time data to estimate the current state of

the economy. They use a variety of econometric techniques to deal with the problems of variable selection, gaps, lags, structural changes, and so on. One promising development is that much of the real-time data is also available at state and

city levels, allowing for regional macroeco nomic analysis.

In the last 20 years or so, the field of machine learning has made tremendous strides in "data mining." This term was once pejorative (at least among econometricians) but now enjoys a somewhat better reputation due to the excit ing applications developed by computer scien tists and statisticians; see Hastie, Friedman and Tibshirani (2009) for a technical overview. One of the big problems with data mining is over fitting, but various sorts of cross-validation techniques have been developed that mitigate this problem. Econometricians have only begun to utilize these techniques; the previously men tioned work by Castle and Hendry (2009) is noteworthy in this respect.

IV. Enable Experimentation

As econometricians have observed, "if you torture the data long enough it will confess to anything." It is difficult to establish causality from retrospective data analysis, so it is note worthy that computer mediation allows one not only to measure economic activity but also to conduct controlled experiments.

In particular, it is relatively easy to imple ment experiments on Web-based systems. Such experiments can be conducted at the query level, the user level, or the geographic level. In 2008, Google ran 6,000 experiments

involving Web search which resulted in 450-500 changes in the system. Some of these were experiments with the user interface, some were basic changes to the algorithm (Hoff 2009). The ads team at Google ran a similar number of experiments, tweaking everything from the background color of the ads, to the spacing

between the ads and search results, to the under lying ranking algorithm. In the 1980s, Japanese manufacturers touted

their "kaizen" system that allowed for "continu ous improvement" of the production process. In a well-designed Web-based business, one can have continuous improvement of the product itself?the Web site.

Google and other search engines also offer various experimental platforms to advertisers and publishers, such as "ad rotation," which rotates ad cr?atives among various alternatives to choose the one that performs best, and "Web site optimizer," a system that allows Web sites to try out different designs or layouts and deter mine which performs best.

Building a system that allows for experimen tation is critical for future improvement, but it is all too often left out of initial implementation. This is a shame, since it is the early versions of a system that are often in need of the most improvement. Cloud computing, which I will discuss later

in the lecture, offers a model for "software as service," which typically means software which is hosted in a remote data center and accessed via a Web interface. There are numerous advan

tages to this architecture, but one that is not sufficiently appreciated is the fact that it allows for controlled experiments which can in turn lead to continuous improvement of the system. Alternatives, such as packaged software, make experimentation much more difficult.

Ideally, experiments lead to understanding of causal relations that can then be modeled. In

the case of Web applications there are typically two "economic agents": the users and the appli cations. The applications are already modeled via the source code that is used to implement them, so all that is necessary is to model the user behavior. The resulting model will often take the form of a computer simulation which can be used to understand how the system works.

A nice example of this are the Bid Simulator and Bid Forecasting tools offered by Google and Yahoo. These tools give an estimate of the cost and clicks associated with possible bids. The cost per click is determined by the rules of the auction and can be calculated directly; the clicks are part of user behavior and must be estimated statistically. Putting them together gives a model of the auction outcomes.

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

6 AEA PAPERS AND PROCEEDINGS MA Y 2010

A. How Experiments Change Business

The fact that computer mediation drastically reduces the cost of experimentation changes the role of management. As Kohavi, Longbotham, Sommerfield, and Henne (2008) have empha sized, decisions should be based on carefully controlled experiments rather than "the Highest Paid Person's Opinion (HiPPO)."

If experiments are costly, expert opinion by management is a plausible way to make decisions. But when experiments are cheap, they are likely to provide more reliable answers than opinions?even opinions from highly paid people. Furthermore, even when experienced managers have better-than-average opinions, it is likely that there are more useful things for them to do than sit around a table debating about which background colors will appeal to

Web users. The right response from managers to such questions should be "run an experiment."

Businesses have always engaged in experi mentation in one form or another. But the availability of computer mediated transactions makes these experiments much more inexpen sive and flexible than they have been in the past.

V. Customization and Personalization

Finally, computer mediated transactions allow for customization and personalization of the interactions by basing current transactions on earlier transactions or other relevant information.

Instead of a "one size fits all" model, the Web offers a "market of one." Amazon, for example, makes suggestions of things to buy based on your previous purchases, or on purchases of consumers like you. These suggestions can be based on "recommender systems" of various sorts (Resnick and Varian 1997).

Not only content, but prices may also be per sonalized, leading to various forms of differ ential pricing. What are the welfare effects of such personalized pricing? Acquisiti and Varian (2005) examine a model in which firms can con dition prices on past history. They find that the ability of firms to extract surplus is quite limited when consumers are sophisticated. In fact, firms have to offer some sort of "enhanced services"

in order to justify higher prices. Varian (2005b) suggests that there is a "third

welfare theorem" that applies to (admittedly extreme) cases with perfect price discrimination

and free entry: perfect price discrimination results in the optimal amount of output being sold while free entry pushes profits to zero, con ferring all benefits to the consumers. See Ulph and Vulkan (2007) for a theoretical analysis of first-degree price discrimination.

The same sort of personalization can occur in advertising. Search engine advertising is inher ently customized since ads are shown based on the user's query. Google and Yahoo offer ser vices that allow users to specify their areas of interest and then see ads related to those inter

ests. It is also relatively common for advertisers to use various forms of "re-targeting" that allow them to show ads based on previous responses of users to related ads.

VI. Transactions among Workers

My emphasis so far has been on transactions among buyers, sellers and advertisers. But com puters can also mediate transactions among workers. The resulting improvements in com munication and coordination can lead to produc tivity gains, as documented in the large literature on the impact of computers on productivity.

In a series of works, Paul David (1990, 1991a, 1991b) has drawn an extended analogy between the productivity impact of electricity at the end of the nineteenth century and the productivity impact of computing at the end of the twenti eth century.6 Originally factories were powered by waterwheels which drove a shaft, and all the machines in the factory had to connect to this central shaft. The manufacturing process involved moving the piece being assembled from station to station during assembly.

The power source evolved from waterwheels to steam engines to electric motors. Eventually elec tric motors were attached to each machine, which allowed more flexibility in how the machines were arranged within the factory. However, fac tories still stuck to the time-honored arrange

ments, grouping the same sort of machines in the same location?all the lathes in one place, saws in another, and drills in yet another.

It wasn't until Henry Ford invented the assembly line in the first decade of the twenti eth century that the flexibility offered by electric

6 See David (1990), David (1991b), and David (1991a).

This content downloaded from �������������69.166.59.244 on Wed, 03 Nov 2021 19:33:42 UTC�������������

All use subject to https://about.jstor.org/terms

VOL. 100 NO. 2 RICHARD T. ELY LECTURE 7

motors was really appreciated.7 As David (1990) shows, the productivity impact of the assembly line was huge, and over the last century manu facturing has become far more efficient.8

I want to extend David (1990)'s assembly line analogy to examine "knowledge worker

Collepals.com Plagiarism Free Papers

Are you looking for custom essay writing service or even dissertation writing services? Just request for our write my paper service, and we'll match you with the best essay writer in your subject! With an exceptional team of professional academic experts in a wide range of subjects, we can guarantee you an unrivaled quality of custom-written papers.

Get ZERO PLAGIARISM, HUMAN WRITTEN ESSAYS

Why Hire Collepals.com writers to do your paper?

Quality- We are experienced and have access to ample research materials.

We write plagiarism Free Content

Confidential- We never share or sell your personal information to third parties.

Support-Chat with us today! We are always waiting to answer all your questions.