Program Report: Industrial Organization, 2012
The NBER's Program on Industrial Organization (IO) begins its third decade with a core of 60 program members, including 15 whose primary affiliations are in another NBER Program. The Program's meetings attract submissions from a large and diverse set of researchers, and are lively sessions with 75 to 90 scholars typically in attendance. The IO Program produces important applied research on a broad range of industries and topics, increasingly at the intersection with such other NBER Programs as Environmental and Energy Economics, Productivity, and Health Care. That commonality is recognized with frequent joint program meetings and contiguous Summer Institute sessions with other NBER groups. In 2012, IO Program members Aviv Nevo and Ariel Pakes delivered the annual Summer Institute Methods Lectures1, focusing on the econometrics of demand estimation and related methodologies.
This report describes work in just three of the Program's areas: modeling consumer choice; the industrial organization of the digital economy; and lessons for designing government auctions. Readers interested in exploring the broader range of NBER work in IO are encouraged to visit https://www.nber.org/papersbyprog/IO.html
Consumer Choice
Empirical economists in the field of IO have devoted substantial attention to modeling the determinants of demand across a variety of settings. For some time, NBER researchers have been active in the design, innovation, and evaluation of methods to estimate demand based on neoclassical theories of consumer utility maximization. Nevo and Pakes discussed this in their 2012 Methods Lectures1 and dozens of NBER Working Papers have been published in this area.2 In recent years, empirical researchers increasingly have turned their attention to analyzing the underpinnings of individual choice, for example characterizing the implications of deviations from standard neoclassical models of optimization behavior and the role of information in markets.
Consumer behavior
The detailed microdata that are the mainstay of much empirical IO research have proved useful for identifying departures from conventional models of consumer utility maximization. A body of work in this area has looked at automobile purchases, one of the most significant consumer purchase decisions for most households. Meghan Busse, Florian Zettlemeyer, and co-author Duncan Simester (w13140) document consumer responses to "price cues" in the context of a Big Three automaker "Employee Discount Pricing" promotion in the summer of 2005. They find that consumers responded to this promotion with unprecedented increases in new car purchases, even though prices during the promotion were not substantially lower than immediately prior to it. Indeed, sales increased even for some models with higher prices during the promotion. While the researchers point out that this behavior can be consistent with rational reliance on (noisy) price signals, their results are cautionary for those who would model consumers as responding primarily to observed prices. In another paper on auto purchases, Nicola Lacetera, Devin Pope, and Justin Sydnor (w17030) look at heuristic information processing in used car purchases. They find that sale prices drop discontinuously at exactly 10,000 mile odometer readings, consistent with customers focusing on the left-most digit of the odometer reading rather than incorporating the full odometer reading into their valuation. They estimate $2.4 billion of mispricing as a result. Busse and Pope and their co-authors (w18212) use a sample of 40 million vehicle purchases and 4 million house purchases to explore the role of projection bias the tendency to over-predict the degree to which one's future tastes will resemble one's current tastes in purchasing behavior. They find that weather at the time of purchase overly influences purchase decisions for these major durables. They meticulously explore alternative explanations for this finding, and their results rule out explanations grounded in neoclassical utility maximization. For example, spring or fall days that are unusually warm and sunny induce additional convertible sales, which are not merely time-shifted. Moreover, the convertibles purchased on such days are more likely to be traded in quickly, consistent with misestimating future tastes.
Justine Hastings and Jesse Shapiro (w18248) analyze "mental accounting" in household purchases of gasoline. Their results consistently reject the null hypothesis that households treat spending on gasoline as fungible with other income. Instead, when gasoline prices rise, consumers disproportionately substitute to (less expensive) lower octane gasoline, far more than the substitution that occurs for similar income effects from non-gasoline price sources; the converse is true when gas prices rise. This complements work that Hastings has done with other collaborators (w13614) on how households adjust grocery purchases when gasoline prices change. Andrei Shleifer and his collaborators (w17947) develop a model of context-dependent consumer choice focused on "salient attributes" that is consistent with this mental accounting behavior, and use their model to study discounts in a variety of settings.
Better understanding of consumer choice is an important input to modeling firm decisions. Julio Rotemberg (w13754 ) models the implications for firms and policymakers of consumers who do not make effective use of price information, and then suffer ex post regret or anger as a consequence. Hastings and Shapiro (w18248) argue that supplier response to the consumer octane adjustment behavior they find in gasoline markets may contribute to an observed inverse relationship between gasoline prices and retailer markups over time.
The role of imperfect information
The rich theoretical literature on markets with imperfect information recently has been married to an increasing body of empirical work exploring how imperfect information affects markets. In one paper, Ginger Jin and David Dranove (w15644) review the theoretical and empirical research on product quality disclosure and certification. Jin and her collaborators (w14252) also explore how information about the properties of a new pharmaceutical is diffused to doctors and patients, and they consider a range of information sources including academic articles, advertising, media reports, FDA updates, and individual patient experiences. How information is presented, in addition to its content, can have substantial impacts on consumer responses: Hastings and various collaborators have shown this in the context of workers' choice among pension investment options in Mexico's privatized social security system (w14538) and in parents' decisions on public school choice programs (w12995). Similarly, Phillip Leslie and Alan Sorensen's work with Bryan Bollinger (w15648 ) on posting calorie counts in restaurant chains demonstrates the importance of how information is presented to consumers.
In many markets, search does not appear to be effective in matching consumers to the lowest-price or highest match quality product. Glenn Ellison's research with Alexander Wolitzky (w15237 ) argues that this may in part reflect actions taken by firms to impede effective search. In their model of "obfuscation," firms selling homogeneous goods find it individually rational to invest in actions that make it more difficult for consumers to learn about their product and full product price, because this reduces competition across firms and sustains higher mark-ups. Bruce Carlin and coauthor Florian Ederer (w17895) model oligopolists' product proliferation responses to the possibility of consumer search fatigue, the notion that search is not only costly but also tiring, potentially leading consumers to break from searching in some periods. Robert Hall and Susan Woodward (w16007) argue that mortgage broker decisions by borrowers suggest substantial deviation from optimal search behavior, and are indicative of buyer confusion, not only on how to assess complex menus of broker charges but even about the potential benefits of search among brokers. They conclude that current disclosure policies have done little to mitigate that confusion.
Industrial Organization of the Digital Economy
The digital economy has exploded in the two decades since the IO Program's January 2001 conference on e-commerce,3 along with economic research on its characteristics and the implications for firm strategies and traditional retail markets. Jonathan Levin (w16852) examines the literature in this area and describes the economic implications of key features of the digital economy: an unusual combination of substantial economies of scale with customer personalization; the ability to collect large volumes of detailed data about customers, their behavior, and preferences; and the rapid pace of innovation facilitated by seller experimentation. He notes the critical role of economic theory in the design and analysis of these markets, and the platform that these markets offer for empirical research on the digital marketplace and as a setting in which to test models of imperfect competition. Below are results from just three strands of IO research that explore these and other themes in online markets: the design of online strategies by firms; mechanisms to address asymmetric information about online seller quality; and the implications of digital distribution for producers and "bricks and mortar" retailers.4
Designing online strategies
As Levin notes, digital marketplaces offer new challenges and new possibilities for firms. David Reiley and his collaborators analyze online auctions in their chapter for the Handbook of Economics and Information Systems (w12785), focusing on the theory, experimental research, and empirical analysis of online retail auctions such as eBay.5 This work describes the responsiveness of bidder strategies to seller strategies, and its implication for optimal design of online auction markets. It also addresses endogenous innovation in markets such as eBay, highlighting the importance of considering dynamic implications of auction design for the viability of platforms.
One of the most active online markets involves "position auctions" which are conducted by search engines such as Google, Bing, or Yahoo to allocate to advertisers the "sponsored link" positions on a search response page. Susan Athey and Ellison (w15253) emphasize the two-sided market aspect of these auctions. Bidders (advertisers) care about how consumers respond to advertising links, and those responses in turn are affected by the mechanism that sellers (search engines) use to allocate advertisers to positions. Enriching the analysis to include consumer search behavior yields a number of insights not present in conventional auction models, such as the benefits of high reserve prices to exclude lower match quality ads and the informational inefficiencies that can be induced by weighting bids by customer click-through rates.
In online markets, experimentation is facilitated and rewarded. Search engine firms rely on substantial experimentation, in addition to economy theory, to enhance profits through the design of their auctions. Liran Einav, Levin, and their collaborators (w17385) document the activity of eBay sellers to improve their strategies through both active and passive experimentation. The ease of experimentation online is a boon to researchers as well. Reiley et al. (w12785) and Levin (w16852) describe a number of academic studies that have taken advantage of online platforms to construct field experiments to investigate consumer behavior, pricing strategies, advertising effectiveness, and the implications of auction design, some of which are described below.
Asymmetric information on the internet
The growth in online markets has elevated interest in the effect of asymmetric information on seller quality, and has provided new tools for its empirical investigation. While online markets may reduce search costs and offer greater apparent pricing transparency, their heightened anonymity of exchange exacerbates the problem of asymmetric information between buyers and sellers, particularly with respect to seller quality or trustworthiness. Seller reputation can mitigate asymmetric information, and often is established online through buyer feedback mechanisms, such as eBay's well-studied feedback system. Third-party certification provides an alternative to feedback or reputation mechanisms. Jin and her co-authors (w17955) study the effectiveness of certification authorities for online pharmacies, used by many consumers to reduce drug acquisition costs. For four of the five popular brand-name drugs they ordered from online pharmacies, drugs labeled as branded were authentic versions for all delivering pharmacies, whether certified or not, but prices at certified U.S. pharmacies were roughly 50 percent higher than were prices at non-U.S. certified pharmacies. This suggests considerable cost to consumers from complying with FDA warnings to avoid all foreign websites, perhaps without concomitant consumer benefit. For the fifth drug -- Viagra -- certified pharmacy prices and quality were roughly identical regardless of country, while uncertified pharmacies offered both a lower price and a lower probability of receiving an authentic branded product. This suggests a potential advantage to buying only from certified websites; the authors' online survey of over 2500 consumers suggests that is what more than 40 percent of consumers who purchase drugs online do. Jin and collaborators study the role of price signals and regulation across international pharmaceutical markets in w16854 and w18073.
Erzo Luttmer, Asim Khwaja, Rajkamal Iyer, and Kelly Shue (w15242), as well as Jin and co-author Seth Freedman (w16855), use data from the online peer-to-peer lending platform Prosper.com to investigate the ability of lenders to screen the creditworthiness of prospective borrowers. Iyer et al. find that lenders respond to coarse information in the Prosper.com profiles in order to infer much of the information that would have been accessible from (unreported) individual-level credit scores. Freedman and Jin report that lenders on Prosper.com generally underestimate the credit risk of borrowers, but learn significantly from their own experiences on the site. Newer cohorts of lenders underestimate less, suggesting some diffusion of learning across cohorts. Of particular interest is the convergence the researchers note between online and more traditional offline sources of credit: as lower quality subprime borrowers have been increasingly excluded from funding on Prosper.com, the site has competed more directly with conventional lending institutions such as banks. Competition between online and offline outlets is also the subject of other work by Freedman and Jin on peer-to-peer lending, and is the topic of a broad research agenda by other NBER researchers, to which I turn next.
Interactions with offline markets
The rapid growth of the internet-based economy over the past 15 years has dramatic implications for both producers and "bricks and mortar" retailers. Early research in this area focused on pricing impacts of online search and ecommerce. More recent research has highlighted the impact of the internet on the allocation of sales across retailers, and the entry/exit decisions of firms, and product choice decisions by producers.
Einav, Levin, and their co-authors (w18018) explore the impact of sales taxes on consumers' choices of online retailers, which is of considerable policy interest. Their analysis of eBay customer responses to sales taxes suggests considerable sensitivity: a single percentage point increase in a customer's home state sales tax implies an increase of nearly 2 percent in online purchasing from other states, and a decline of roughly 3 to 4 percent in online purchasing from home-state sellers. The authors also note increased density of sellers on the low-tax side of state borders.
Ali Hortaçsu and Chad Syverson and their collaborators (w14166) examine impacts on physical retailers for three of the sectors they expect to be most affected by the internet: auto dealers, bookstores, and travel agencies. They find that online shopping has shifted the distribution of revenues among physical bookstore and auto dealers from small retailers to larger retailers, and that smaller retailers disproportionately have exited as the fraction of consumers using online shopping increased. Travel agencies experienced the same reallocation away from small outlets, but for that industry the trend appeared to be national, a function of changes in airlines' distribution systems and not dependent on local consumer online shopping patterns.
Igal Hendel, Nevo, and co-author Francois Ortalo-Magne (w13360) compare the impact on home sellers of using conventional versus online sales outlets in a study of the 2004 housing market in Madison, Wisconsin. They find no sales price difference across houses sold through traditional realtors using the Multiple Listing Service (MLS) and those sold by owners using the online FSBO.com website. However, houses on the MLS are both more likely to sell and are quicker to sell, conditional on a transaction, which is consistent with improved matching on the still-larger MLS network.
The internet facilitates search not only on price but also on provider and product quality. And, online review systems allow consumers to register feedback on physical outlets. Jin and her collaborators (w18567) analyze restaurant ratings on Yelp.com, focusing on the optimal way for a review site such as Yelp to construct aggregate ratings from individual feedback. The usefulness of user feedback depends on its credibility, though, and assessing credibility can be difficult. Judith Chevalier, Dina Mayzlin, and Yaniv Dover (w18340) investigate the incidence of review manipulation by comparing the distribution of hotel reviews on Tripadvisor.com, which allows anyone to post a review, to those on Expedia.com, which restrict reviews to consumers who have made a booking at that hotel through Expedia. They find that on TripAdvisor.com, small independent hotels have more five-star reviews, and their neighboring hotels have more one- and two-star reviews, consistent with their predictions of ex ante incentives for review manipulation. While these results suggest that review manipulation may be economically significant, the authors note that the overall level of manipulation is relatively low, ensuring that the platform still communicates useful content.
The music and video industries have been among those argued to be most affected by the internet, in large part because of the producers' greater difficulty in enforcing intellectual property protection of their content online. In principle, unauthorized distribution of online content may have both demand contraction effects (by substituting for purchased content) and demand expansion effects (by increasing potential consumer awareness of the product, effectively advertising). Joel Waldfogel (w13497) explores these twin effects on television viewing in a study of unauthorized (primarily YouTube) and authorized (primarily network) web distribution of television shows. Using a survey of university students, he finds that internet access induces a modest substitution away from traditional television which is more than offset by a strong demand expansion effect: overall time on network-controlled sites (television and network websites) increases by 1.5 hours/week. Julie Mortimer, Sorensen, and co-author Chris Nosko (w16507) find that musical artists have reacted to the decline in album sales that is associated with unauthorized file-sharing by increasing their live performances. Less well-known or popular musicians among the more than 1800 artists they study experience significant increases in concert revenue in the post-Napster era, in part offsetting the lost album revenues, and perhaps reflecting greater awareness of their music by potential fans. Waldfogel (w16882) assembles a novel dataset to explore whether reduced album revenues have led to reductions in the production of new albums, and he concludes that there is no discernible decline in quantity or quality post-Napster. That is consistent with lower costs of bringing new works to market and growth of independent labels. Finally, Leslie and Sorensen find that the expansion of ticket resale markets for major rock concerts, facilitated by online resale sites, improves the allocation of tickets to high valuation buyers. However, half of the gains are dissipated through higher transactions costs, so resale buyers end up with little of the potential surplus (w15476).
Designing Government Auctions
Governments and quasi-public agencies use auctions in a wide variety of settings, including: competitive procurement; sales or leases of publicly-owned assets, such as mineral and timber rights on public land and spectrum allocation; wholesale electricity purchases and sales; and the allocation of pollution permits under some cap and trade programs. Economic theory has made fundamental contributions to the design of many of these auctions, and empirical research has contributed to evaluation of their operation and guiding improvements in their execution. NBER research has played a role in both fronts, and the NBER Working Group on Market Design, led by Athey and Parag Pathak, focuses on these and related issues.
Since 1994, Federal Communication Commission spectrum auctions have been used to allocate billions of dollars in spectrum rights. Patrick Bajari and his co-author Jungwon Yeo (w14441) describe how FCC auction design has evolved over time to mitigate concerns about tacit collusion by bidders. The researchers examine patterns in the bidding data from four large spectrum auctions and conclude that later auctions do, in fact, exhibit fewer examples of strategies most likely associated with potential collusion. Their analysis gives a flavor of the considerable complexity that is involved in bidding in spectrum auctions. Jeremy Bulow, Levin, and co-author Paul Milgrom (w14765) describe the potential for economic and game-theoretic modeling to help bidders devise successful strategies in the face of that complexity. They illustrate such potential by demonstrating how a new entrant used it in the 2006 90MHz auction, which contributed to the firm's success in purchasing nationwide spectrum coverage at one-third the price paid by incumbents, thus saving more than a billion dollars.
Much of the empirical work in government auctions done by NBER researchers has focused on U.S. Forest Service timber auctions, which can generate more than $1 billion annually. James Roberts and Andrew Sweeting (w17624) focus on when sellers should use auctions, comparing the expected relative performance of a simultaneous bid auction to a setting in which sellers invite buyers to make offers sequentially. Athey and Levin work with various collaborators to analyze the design and operation of timber auctions. In one paper (w14590) they compare performance under two different auction formats: sealed bid auctions, which attract more small bidders, and open outcry auctions. Their calibrated model suggests that sealed bid auctions generate greater expected revenue for the Forest Service, and it focuses attention on bidder competitiveness as a critical choice in auction format. In a more recent paper (w16851) they turn to the set-asides and subsidies that the government frequently uses for preferenced bidders, most often small or minority-owned businesses, in procurement or natural resource auctions. Their analysis shows that restricting entry to small businesses is associated with significant revenue and efficiency costs; replacing the restriction with a bidder subsidy would increase revenue, efficiency, and the profit of small bidders, with minimal impact on large firm profitability.
Highway construction procurement contracts are a significant state-level activity, imposing substantial direct costs to finance road construction and repair, and substantial indirect costs on drivers who are subject to delays and longer commutes while construction projects are underway. Bajari and Greg Lewis have developed a research agenda that investigates how to design procurement contracts to more effectively align the incentives of contractors with those of the highway department and drivers. In one paper (w14855), they evaluate scoring auctions used by the California Department of Transportation to provide explicit time-to-completion incentives in contract awards. They estimate substantial welfare gains from the incentive contracts, although direct outlays by the Department of Transportation also increase through their effect on the winning bid. Their model suggests even larger potential gains from an optimally designed policy. In a more recent paper (w17647), they develop a model of contractor adaptation to productivity shocks, incorporating time incentives in an optimal contract design. They combine this with day-level information on work plans, progress, and delays for Minnesota highway projects to explore empirically the role of adaptation and delay, and illustrate the impact of alternative incentive structures on outcomes.
*The numbers in parentheses throughout this report refer to NBER Working Papers.
1. Video and slides available.
2. See for example, recent contributions by program members Patrick Bajari, Steven Berry, Jean-Pierre Dube, Jeremy Fox, Philip Haile, Christopher Knittel, Julie Mortimer, Aviv Nevo, and Stephen Ryan, among others.
3. Published as a special issue of the Journal of Industrial Economics, Vol. 49, no. 4, December 2001.
4. Many IO program members are active in the Productivity Program's new Economics of Digitization and Copyright Initiative, which brings together researchers from a diverse set of fields to study this sector of the economy. This report focuses on work in this area by members of the IO program; additional working papers can be found on the NBER's website.
5. In Handbook of Economics and Information Systems, Terrence Hendershott, ed, Elsevier Science, 2006.
6. Published as Maris Goldmanis, Ali Hortaçsu, Chad Syverson, and Önsel Emre, "E-Commerce and the Market Structure of Retail Industries," NBER Working Paper No. 14166, July 2008, and Economic Journal, Royal Economic Society, vol. 120(545), (2010), pp. 651-682.