Research Statement – Ali Hortaćsu
Much of my work is focused on empirically assessing the efficiency of markets. In a nutshell, I utilize detailed micro-level data from the markets I study to estimate preference and technology parameters that rationalize individual behavior. I then use the estimated preferences and technological parameters to construct (constrained) “efficient” benchmarks and assess how far observed market outcomes are from efficiency. This comparison also motivates discussions of how market rules can be altered to improve efficiency. I have applied the above framework to many market settings, including financial markets, energy markets, and the Internet, and a variety of market clearing mechanisms, including auctions, matching, and costly search.
Perhaps the clearest sustained strand in my research is my work on markets that clear through multi-unit auctions. Many important markets clear in this fashion: for example, almost all developed country governments auction off their debt using a type of multi-unit auction; most deregulated electricity markets clear using a multi-unit auction; the European Central Bank runs its monetary policy using such an auction. While modeling equilibrium strategic behavior in such markets remains an open challenge, the methods that my co-authors and I have developed demonstrate that doing empirical work achieving the objectives above is feasible.
My first paper on multi-unit auctions is titled “Mechanism Choice and Strategic Bidding in Divisible Good Auctions: An Empirical Analysis of the Turkish Treasury Auction Market,” which was my job market paper. It appeared in the Journal of Political Economy, with very significant contributions by David McAdams to its final version. In this paper, we study Milton Friedman’s question of what type of auction format to use to market government securities. Unfortunately, auction theory does not provide an a priori answer to this question, beyond providing examples demonstrating that the revenue ranking of the various mechanisms is ambiguous. The chief innovation of the paper, therefore, is to utilize individual bidder level data and to recover bidders’ “true” willingness to pay curves that rationalize the observed bids; under the assumption of best-response behavior. To that end, the paper provides econometric methods for constructing estimates of bidders’ preferences. With the true willingness to pay curves in hand, the paper then constructs the efficient “Vickrey” benchmark, and studies how the real-world mechanism departs from this benchmark. The answer, in this particular setting, was “not too much.”
My second paper on this agenda, “Understanding Strategic Models of Bidding in Deregulated Electricity Markets: A Case Study of ERCOT,” studies auctions for real-time electricity procurement in Texas. This paper conducts the “dual” exercise of testing whether best-response behavior is an appropriate assumption to impose on bidders participating in complex multi-unit auction mechanisms. To do this, Steven Puller and I reconstructed the marginal costs of electricity production for each of the bidding generators for every hour. Using these MC estimates, we then constructed “best-response” bids (using info available to bidders at the time of bidding) and compared them to actual bids. What we find is a large amount of variation across bidders in terms of achieving “best-response” behavior. While some bidders were able reap almost all of the achievable best response profits, others were far less successful. While some of this variation was ascribable to learning, most of it appeared to be explainable by scale. Indeed, our field trips to the offices of several of these generation companies revealed that bidders with more money at stake had hired much more sophisticated traders who spoke a very similar language to economists, while the underperformers utilized various heuristics rules-of-thumb (such as arbitrary mark-ups designed to recoup fixed costs) with very little consideration of the residual demand they were facing.
Another contribution of this paper was to show that deviations from “best response” behavior, even though localized to smaller bidders, led to significant inefficiencies in this market when aggregated. Indeed, the production inefficiencies in this market were as much due to the best-response actions of the more “sophisticated” bidders as the non-best response behavior of “unsophisticated” bidders. While learning and some market selection and consolidation led to some improvement, such effects were relatively small.
My next two papers that study multi-unit auctions were co-authored with Jakub Kastl. The first paper, “Valuing Dealers' Informational Advantage: A Study of Canadian Treasury Auctions,” is an in-depth study of front-running in that securities market. It also provides one of the first formal tests of the common versus private values assumptions in a financial market. The paper studies the interaction of primary dealers of Canadian government securities and their customers, who have to route their bids through primary dealers. The dataset we obtained tracked the timing of bid submissions, and importantly, bid modifications by all bidders. Using this information, we were able to test whether primary dealers’ modifications of their bids in response to customer bids were due to learning about the “common” value of the security being auctioned, or whether a purely private values model in which PDs learn about the location and shape of the residual supply curve from customer bids could explain the data. Based on our tests, we were not able to reject the null hypothesis of private values in this market. Given our validation of the private value model, we then embark on the exercise of quantifying the surplus the PD’s derived from observing customers’ “order flow” – which we found to be between 10-30% of dealers’ profits.
My second paper with Jakub Kastl, also co-authored with Nuno Cassola, is titled “The 2007 Subprime Market Crisis Through the Lens of European Central Bank Auctions for Short-Term Funds.” In this project, we studied the European Central Bank’s weekly refinancing auctions of short-term (weekly) repo loans, which was the main conduit of monetary policy in the Eurozone.
What we noticed looking at this data was a sudden and dramatic increase in the bids for ECB loans following August 2007, as reflective of funding shortages in the inter-bank market. However, along with the increase in the level of the bids, the dispersion also increased, suggesting heterogeneity in funding problems across banks. That said, while some of the observed bid increases may reflect a true and sudden shift in the underlying WTP for short-term funding, some banks may have started bidding higher just to remain competitive in the auctions. I.e. the strategic nature of bids may have masked the true heterogeneity of funding troubles across Eurozone banks. Thus, our main methodological contribution in the paper was to dissociate the “strategic” accommodating component of bids to isolate banks’ underlying willingness-to-pay for ECB loans. One interesting result of this exercise was that the estimated WTPs of the bidders were much better predictors of balance sheet troubles at the end of 2007 than the bids were, providing credence to our hypothesis that looking at bids and not at fundamentals masks the true heterogeneity of the crisis. Another practical offshoot of the exercise is that the bidding data – when sufficiently filtered to account for strategic behavior – can provide high-frequency snapshots of the short-term funding rates faced by individual banks in the Eurozone. Since interbank markets are famously opaque, and published rates based on surveys like the LIBOR and EURIBOR are famously corruptible, our bank-level barometer of short term funding rates may be an important temperature gauge for use by Europe’s central bankers.
Along with financial and energy markets, I also have written several papers studying Internet marketplaces. “Winner’s Curse, Reserve Prices, and Endogenous Entry” with Pat Bajari was one of the first papers written on online auctions, and “Matching and Sorting in Online Dating” is one of the first papers written on online dating.
The eBay paper lays down a tractable theoretical framework to study eBay and other – often very similar – online auctions. In particular, it provided an explanation for “sniping” through an interdependent values framework, and also allowed for stochastic entry by participants. The paper then structurally estimates the common value model, and studies the optimality of sellers’ reserve prices (from the profit maximization point of view).
I have written two further papers on eBay: “Dynamics of Seller Reputation: Evidence from eBay” with Luis Cabral focuses on eBay’s reputation system, and provides empirical evidence of “opportunistic” behavior by sellers in response to the incentives provided by the reputation mechanism. The other paper, “The Geography of Trade on eBay and Mercado Libre,” with Asis Martinez-Jerez and Jason Douglas, studies the geographic reach of eBay transactions, under the null hypothesis that the “world is flat,” especially on the Internet. Our analysis reveals that while trade on the Internet is much less distance dependent than offline, a form of “home bias” still persists in that a very significant fraction of transactions lie within the same-city limits. Analyzing the prevalence of this across various product categories suggests that the main culprit for the same-city effect may be the lack of trust. Presumably, eBay buyers prefer to “kick the tires” of their purchases or feel more comfortable when the seller is nearby, should a return be necessary.
My work on online dating, follows the same framework outlined in the beginning of this statement. In “What Makes You Click? Mate Preferences in Online Dating” we utilize detailed data on user actions on a large online dating site to estimate online daters’ preferences over a large set of dating partner preferences, including measures of physical attractiveness. While many of our qualitative findings might not be that surprising, our preference estimates allow us to report the “trade-offs” between different attributes – i.e., how much education/income compensates for physical attractiveness/racial differences.
Our second paper on the topic “Matching and Sorting in Online Dating,” utilized these preference estimates to answer the following questions: (1) how efficient is the dating website compared to matches that would have been proposed by (e.g.) Gale-Shapley if they had been equipped with the preference profiles we estimated for each user, and (2) can the various and commonly documented sorting/homogeny patterns in U.S. marriages be explained by preferences alone – and not by other factors such as search frictions. The answer to (1) is that we achieved allocations that were quite similar to Gale-Shapley stable matches. As for (2), the answer is that preferences alone (scaled up to represent the U.S.) indeed would explain a very large fraction of the sorting patterns we see in the offline world, casting some doubt into the significance of other factors such as search frictions.
Product Markets with Search and Information Frictions
Yet another research agenda I have tried to advance is to empirically study product markets with search and information frictions. My first paper on this was “Product Differentiation, Search Costs and Competition in the Mutual Fund Industry,” with Chad Syverson, where we studied the market for Standard & Poor’s 500 index funds. To our astonishment, we found very large (almost 30-fold) price differentials in this market, where “price” reflects fees paid to money managers. To explain the nature and evolution of price dispersion, we developed a model of utility search by individual investors, which took vertical product differentiation factors into account. We then show that the distribution of search costs rationalizing the observed prices and market shares can be nonparametrically identified from the data, taking into account vertical product differentiation across various funds. Using our estimates of search costs, we found that search costs for investors in the higher quantiles of the search costs distribution increased in the Internet bubble era, which may be rationalized by the fact that a large number of new and inexperienced investors were attracted to invest in the stock market.
In a recent paper titled “Testing Models of Consumer Search Using Data on Web Browsing and Purchasing Behavior,” co-authored with my former Ph.D. student Babur de los Santos and Matthijs Wildenbeest, we use a very detailed web browsing and purchasing data set to study how people shop for books on the Internet. To our knowledge, aside from a number of laboratory experiments, this is one of the first papers that uses data on actual search behavior by consumers to inform a model of consumer search. The paper conducts tests between various search protocols, and proposes and estimates a “hybrid” model of costly search and discrete choice differentiated products to model demand for books.
One of my most recent projects, “Advertising and Competition in Privatized Social Security: The Case of Mexico,” with Chad Syverson and Justine Hastings also focuses on information frictions. Here we study the beginnings of Mexico’s privatized social security system, where the management fees charged by fund managers could be up to 27% of contributions, despite there being 17 competing management companies. Through detailed analysis of administrative data on individuals’ choices, we find that the very higher level of the fees are rationalized by very low elasticity displayed by households, who also display very high sensitivity to advertising and sales efforts by plan providers. Thus, not surprisingly, the market became one where instead of price competition, firms competed through their advertising and sales efforts, shunting millions of people into extremely high cost retirement rates. Using our estimated demand system, we study counterfactual scenarios where advertising is regulated to be equal across firms, and find that this may lead to significant cost savings.
The Organization of Firms
Last, but not least, I have also worked on two projects pertaining to the organizational structure of firms. The first of these papers, “Cementing Relationships: Vertical Integration, Foreclosure, Productivity, and Prices” with Chad Syverson, studies vertical mergers between cement and concrete producers. Such mergers were vehemently opposed in the 60s and 70s, but the “Chicago School” influence on vertical antitrust policy became much more commonplace in the 80s and afterward. Using confidential Census data sets, we find that the vertical mergers in this industry were, on average, efficiency enhancing, leading to lower intermediate and final good prices and larger quantities.
In “Vertical Integration and Input Flows,” we dig deeper into the question of what a “vertical integrated” firm may be, and to study goods shipments across plants belonging to the same firm. We find that, in a large majority of firms that one may label as having “vertically integrated” (which means that the firm owns a plant that produces a good that, according to the 4-digit SIC based Input-Output tables, is used in the production of a good produced in another plant belonging to the firm), there are no goods shipments between the firm’s vertically related plants.
The above result suggests that such “vertical” acquisitions after which there is no explicit goods transfer between plants within the firm may be motivated by the need to facilitate “intangible” input transfers, as suggested by Arrow and Teece. Such “intangibles” may comprise of management and technology know-how, marketing and distribution resources and assets, “branding” capital, etc. The paper concludes with some suggestive evidence in agreement with the intangible inputs theories of Arrow and Teece.