Testing Models of Consumer Search Using Data on Web Browsing and Purchasing Behavior

(with Ali Hortaçsu and Matthijs R. Wildenbeest, American Economic Review, 102(6): 2955–80.)

Using a large dataset on web browsing and purchasing behavior we test to what extent consumers are searching in accordance to various search models. We find that the benchmark model of sequential search with a known price distribution can be rejected based on recall patterns found in the data as well as the absence of dependence of search decisions on prices. Our findings suggest fixed sample size search provides a more accurate description of search behavior. We then utilize the fixed sample size search model to estimate demand elasticities of online bookstores in an environment where store preferences are heterogeneous.

[Published version] [Working paper version]

Optimizing Click-through in Online Rankings for Partially Anonymous Consumers

(with Sergei Koulayev)

While considering differentiated products for purchasing decisions, it is costly for consumers to obtain the necessary information to weigh the various alternatives. The vast amount of information available online has revolutionized the way firms present consumers with product options. Presenting the best alternatives reduces search costs associated with a consumer finding the right product. Heterogeneity in consumers' preference for products with multiple attributes makes it challenging to present a relevant ranking, especially when important characteristics, such as price, differ between the formation of the underlying ranking and the consumer's search process. We use novel data on consumer click-stream behavior from a major web-based hotel comparison platform to estimate a random coefficient discrete choice model. We are then able to infer consumer preferences regarding a set of product attributes and propose an optimal ranking tailored to anonymous consumers with different price sensitivity. We are able to customize rankings by relating price sensitivity to their request parameters, such as the length of stay, number of guests, and day of the week of the stay. In contrast to a myopic popularity-based ranking, our model accounts for the rapidly changing prices that characterize the hotel industry, consumers' expected search strategies including result sorting and filtering, and consumer heterogeneity. The platform must determine which hotel ordering maximizes consumers' click-through rates (CTR) based on the information available to the platform at that time, its assessment of consumers' preferences, and the expected consumer type based on request parameters. We find that consumers' CTRs more than double when consumers are provided consumers with customized rankings that reflect the price/quality trade-off inferred from the consumer's request parameters. We show that the optimal ranking results in consumers' welfare 173 percent greater on average than in the original ranking.

Consumer Search on the Internet

This paper analyzes search frictions in online markets using novel data on the web browsing and purchasing behavior of a large panel of consumers. This dataset is unique in that consumer search behavior prior to a transaction is observed. Although recent models have shown that large price dispersion persists in various markets, and this dispersion is directly related to the magnitude of consumer search costs, little attention has been given to quantifying these costs. I use data on consumers shopping for books online to link prices and consumer search patterns at different bookstores to estimate consumer search costs in the context of an equilibrium search model. The search patterns indicate that consumers visit relatively few firms and exhibit a strong search preference for prominent retailers. I control for search intensities at different retailers during consumers' search process and find that search cost estimates are lower than when assuming consumers sample equally among alternatives. Accounting for unequal consumer search reduces search cost estimates in half from $1.8 to $0.9 per search. I examined the search cost heterogeneity by using a rich set of consumer characteristics and relating them to search patterns and search costs estimates. I use a flexible random effects model in which the number and order of firms visited by the consumer is her optimal ordered choices, allowing search cost cutoffs to depend on regressors. The estimates indicate that consumer search costs are related to their observable characteristics, such as income, where individuals with income greater than $100,000 incur relatively higher search costs.

Search with Learning

(with Ali Hortaçsu and Matthijs R. Wildenbeest, revise and resubmit Journal of Business & Economic Statistics)

This paper provides a method to estimate search costs in an environment in which consumers are uncertain about the price distribution. Consumers learn about the price distribution by Bayesian updating their priors belief. The model provides bounds on the search costs that can rationalize observed search and purchasing behavior. Using individual--specific data on web browsing and purchasing behavior for electronics sold online we show how to use these bounds to estimate search costs. Estimated search costs are sizable and are found to relate to consumer characteristics in intuitive ways. The model outperforms a standard sequential search model in which the price distribution is known to consumers.

What's in a Name? Measuring Prominence, and Its Impact on Organic Traffic from Search Engines

(with Michael R. Baye and and Matthijs R. Wildenbeest)

Organic product search results on Google and Bing do not systematically include information about seller characteristics (e.g., feedback ratings and prices). Consequently, it is often assumed that a retailer's organic traffic is driven by the prominence of its position in the list of search results. We propose a novel measure of the prominence of a retailer's name, and show that it is also an important predictor of the organic traffic retailers enjoy from product searches through Google and Bing. We also show that failure to account for the prominence of retailers' names---as well as the endogeneity of retailers' positions in the list of search results---significantly inflates the estimated impact of screen position on organic clicks.

Estimating Price Discrimination in the Online Distribution Channel

(with Sergei Koulayev and Renáta Kosová)

This article explores price discrimination strategies in the online sales channel. We combine a rich set of consumer search data for hotel rooms at a major online travel consolidator with data on hotels' daily revenue, capacity, and hotel census information that provide a nearly complete picture of the competitive environment for 13 major U.S. hotel markets. Contrary to theoretical predictions on the unambiguous profitability from price discrimination, we find that across different quality segments a significant proportion of hotels forgo the option of price discrimination by not participating in the online distribution channel. We find that hotels with an online distribution channel differ in important ways, and online prices are consistently higher across quality segments and other hotel attributes. The different pricing strategies between online and offline channels across establishments of different quality suggest that the degree of price discrimination depends on the degree of product differentiation. The results suggest behavior considered by Corts (1998), namely, that hotels have asymmetric assessments of consumers' demand elasticities across distribution channels, and they adjust their pricing strategies accordingly. We correct for selection into the online channel and find that it noticeably changes the direction of the effect of online pricing across quality segments. These results suggest that online pricing strategies are opposite high to low quality segments are compared.

Estimating the Effect of the Manufacturer's Suggested Retail Price

(with In Kyung Kim and Dmitry Lubensky)

Although the manufacturer's suggested retail price (MSRP) is widely used for various consumer goods, current research does not conclusively indicate how the price recommendation affects competition of the market and equilibrium prices. We exploit a ban of recommended retail prices on some categories of grocery products in South Korea to explore the effect of MSRP advertisement on retail price. The remarkable aspect of this ban is that it was reversed after a year of being in place, allowing us to observe the effect of allowing the MSRP on prices of previously banned products. Our preliminary results show that the MSRP results in a higher level in distribution channels characterized with a large number of atomized firms.

The Evolution of Product Search

(with Michael R. Baye and and Matthijs R. Wildenbeest, Journal of Law, Economics & Policy, 9(2): 201-221.)

This paper examines the evolution of product search. We provide an overview of product search in the pre-internet era, and discuss how online search evolved from directory based search in the early 1990s to "vertical" search engines by the late 1990s. We also document the prominence of price comparison sites in the mid-2000s, and the challenges these platforms faced through 2010. We then use comScore qSearch data to closely examine trends in product search between 2010 and 2012. We find that, today, the vast majority of shoppers conduct product searches at retailer sites and other marketplaces, whereas traditional price comparison sites have become less important.