Session 10: Empirical Market Design

Healthcare is often allocated without prices, sacrificing efficiency in the interest of equity. Wait times then typically serve as a substitute rationing mechanism, creating their own distinct efficiency and distributional consequences. I study these issues in the context of the Veterans Health Administration (VA), which provides healthcare that is largely free but congested, and the Choice Act, a large-scale policy intervention that subsidized access to non-VA providers to reduce this congestion. Using variation in Choice Act eligibility in both patient-level and clinic-level difference-in-differences designs, I find that the price reduction for eligible veterans led to substitution away from the VA, an increase in overall healthcare utilization and spending, and reduced wait times at VA clinics in equilibrium. I then use the policy-induced price and wait time variation to estimate the joint distribution of patients’ willingness-to-pay and willingness-to-wait. I find that rationing via wait times redistributes access to healthcare to lower socioeconomic status veterans, but at a large efficiency cost (-23%). By contrast, I find that a coarsely targeted, modest increase in copayments increases consumer surplus by more than the Choice Act, at lower cost to the VA, while disproportionately benefitting low-income veterans.

In the face of limited health care resources, waiting time often serves as a rationing mechanism in health systems around the world. We evaluate the efficiency and equity consequences of rationing via queues in the context of surgical care. Focusing on the U.S. private insurance market, we first develop a new measure of wait time that captures the complete patient trajectory—including visits for primary care, laboratory testing, and medical imaging—along the path to surgery. Exploiting exogenous variation in these waits due to weekly congestion within a patient’s insurance network, we show that patients who wait a month more spend 5.9% more on hospital care, are 3.1% more likely to be readmitted to a hospital, and fill 6.6% more opioid prescriptions in the six months following a surgery. We further demonstrate misallocation of wait times relative to the planner’s ideal: we identify heterogeneous effects of waiting, but those patient groups who suffer the highest costs from delay do not uniformly experience shorter waits. For a set of common surgeries, we illustrate how reassigning patient priorities in the queue—say, by providing physicians information on the treatment effects of waiting by patient type—could substantially reduce payments to hospitals and improve overall patient health outcomes.   

Many centralized assignment systems seek to not only provide good matches for participants’ current needs, but also to accommodate changing preferences and circumstances. We study the problem of designing such a mechanism in the context of Norway’s system for dynamically allocating patients to general practitioners (GPs). We provide direct evidence of misallocation under the current system––patients sitting on waitlists for each others’ GPs, but who cannot trade––and propose an alternative mechanism that adapts the Top-Trading Cycles (TTC) algorithm to a dynamic environment. Because of patients’ dynamic incentives, dynamic TTC raises novel incentive and distributional concerns relative to the static case. We then estimate a structural model of switching behavior and GP choice and empirically evaluate how this mechanism would perform relative to the status quo. While introducing TTC would on average reduce waiting times and increase patient welfare—with especially large benefits for young and female patients—patients endowed with undesirable GPs would be harmed. Adjustments to the priority system can avoid harming this group while preserving most of the gains from TTC.

Carbon offsets could reduce the global costs of carbon abatement but there is little evidence on how much they truly reduce emissions. We study carbon offsets sold by firms under the Clean Development Mechanism (CDM) in China by matching offset projects proposed to the United Nations to panel data on emissions and output for manufacturing firms. We have two main findings. First, the CDM attempts to screen out projects that would be profitable without offset payments by rejecting proposed projects with higher stated returns. Second, offset-selling firms steeply increase emissions after registering an offset project, relative to similar firms that proposed a project but did not follow-through. We explain this increase in emissions by jointly modeling the firm decision to propose an offset project and the Board’s decision of whether to approve. In the model, CDM firms increase emissions due to a combination of the selection of higher-growth firms into abatement project investment and the causal effect of higher productivity, post investment, on firm scale and therefore emissions.

We evaluate the extent to which Google’s dominance in web search is driven by higher quality, default effects, and/or economies of scale in data. We develop a model of consumer choice with quality preferences, inertia, and learning effects, and estimate it using a randomized experiment with desktop internet users. Facilitating active choice between search engines has almost no effect on market shares, but changing the address bar default has persistent market share effects of about 25 percent after four weeks in pilot data. Early results show that remarkably, paying Google users to try Bing for two weeks causes about 20 percent to stay with Bing afterwards, while paying Bing users to try Google causes a similar share to stay with Google. Using internal Microsoft search logs, we estimate that Bing’s click- through rates on less-common search queries improve by 7 percent when the amount of data doubles. In counterfactual simulations, choice screens have very limited effects, switching Chrome defaults to Bing reduces welfare, and sharing Google’s click-and-query data with Microsoft increases Bing’s market share only slightly.

This paper investigates the impact on consumers of the widespread manipulation of reputation systems by sellers on two-sided online platforms. We focus on a relevant empirical setting, the use of fake product reviews on e-commerce platforms, which can affect consumer welfare via two channels. First, rating manipulation deceives consumers directly, causing them to buy lower quality products and pay higher prices for the products with manipulated ratings. Second, the presence of rating manipulation lowers trust in ratings, which may result in worse product matches if consumers place too little weight on quality ratings. This decrease in trust may also increase price competition and benefit consumers by lowering prices on high quality products whose quality is less easily observed. We formally model how consumers form beliefs about quality from product ratings and how these beliefs are affected by the presence of fake reviews. We use incentivized survey experiments to measure beliefs about fake review prevalence. Our model of product quality is incorporated into an empirical model of consumer demand for products and how demand is shifted by ratings, reviews, and prices. The model is estimated using a large and novel dataset of products observed buying fake reviews to manipulate their Amazon ratings. We use counterfactual policy simulations in which fake reviews are removed and consumer beliefs adjust accordingly to explore the effectiveness and welfare and profit implications of different methods of regulating fake reviews.

In many online markets, platforms engage in platform design by choosing product recommendation systems and selectively emphasizing certain product characteristics. I analyze the welfare effects of personalized recommendations in the context of the online market for hotel rooms using clickstream data from Expedia Group. This paper highlights a tradeoff between match quality and price competition. Personalized recommendations can improve consumer welfare through the “long-tail effect,” where consumers find products that better match their tastes. However, sellers, facing demand from better-matched consumers, may be incentivized to increase prices. To understand the welfare effects of personalized recommendations, I develop a structural model of consumer demand, product recommendation systems, and hotel pricing behavior. The structural model accounts for the fact that prices impact demand directly through consumers’ disutility of price and indirectly through positioning by the recommendation system. I find that ignoring seller price adjustments would cause considerable differences in the estimated impact of personalization. Without price adjustments, personalization would increase consumer surplus by 2.3% of total booking revenue (~$0.9 billion). However, once sellers update prices, personalization would lead to a welfare loss, with consumer surplus decreasing by 5% of booking revenue (~$2 billion).

Racial segregation is an enduring feature of U.S. K-12 education. Up to half of it originates within schools due to how classrooms are formed. This paper develops an empirical framework to understand the implications of discretionary classroom formation in competitive education markets. I leverage a school competition reform to document via an event-study that in anticipation of a competitive shock, public schools both raise their academic quality and change students’ assignments to classrooms such that classroom segregation increases. I then estimate an empirical model of school choice and competition to understand whether schools choose their level of classroom segregation so as to differentiate horizontally, thereby relaxing vertical competition on costly academic quality. The model’s novelty is that it embeds classroom segregation both on the demand side, as a dimension that parents have preferences over, and on the supply side, as a margin of differentiation that schools choose directly alongside academic quality. I estimate preferences for classroom segregation so as to rationalize the reduced-form effects of competition identified through the event-study. I use the model to evaluate a policy that requires schools to form racially integrated classrooms, given the composition of the student body at the school. I find that the policy raises aggregate academic quality and the average test score in equilibrium. Magnitude-wise, present value lifetime earnings rise by up to $1,620 per student. Since the schools that increase academic quality the most are located in non-white areas, learning gains accrue mostly to non-white students, decreasing the racial test score gap by 2%.

This paper asks how search costs, limited awareness of schools, misperceptions of schools’ attributes, and inaccurate beliefs over unknown schools affect families’ search and application decisions in Chile’s nation wide school choice process. We combine novel data on search activity with a panel of household surveys, administrative application data, randomized information experiments, and a model of demand and sequential search with subjective beliefs. Descriptively, households hold inaccurate beliefs and misperceptions along multiple dimensions which distort the perceived returns to search. Most importantly, they do not know all schools, and misperceive quality ratings of the schools they know and like. Improving the search technology would raise households’ search effort and welfare. Correcting misperceptions about known schools’ observables would cause students to match to schools with higher quality, equal to what can be achieved under a full-information benchmark. Models with out misperceptions would incorrectly predict quality reductions.

Centralized assignment systems are a popular policy tool to improve fairness and efficiency in allocating students to public college seats. In most implementations, however, private college admissions remain decentralized, which may give high socio-economic status (SES) students a strategic advantage in the centralized public match because high-SES students derive higher value from expensive private alternatives. I empirically study application behavior and the allocation of students in markets where only public college seats are centrally assigned with new data from the college match in Albania. Using a policy change that incorporated all private colleges in the centralized platform, which differentially shifted outside alternatives by SES, I find that when private colleges operate outside the match, high-SES students apply to more selective portfolios and enroll in more selective public pro- grams, but the selectivity gap in applications shrinks after the policy change. I build and estimate a model of applications and matriculation that uses the unique institutional features of the Albanian college admissions to disentangle the effects of heterogeneous beliefs, preferences, and outside options on choice, and evaluate the distributional consequences of counterfactual admissions design. I find that removing outside options reverses the welfare gap in favor of lower-SES students, but at the expense of overall market efficiency. This is driven by the fact that outside options dampen the distortionary effects of list size restrictions and incorrect beliefs on choice.

Although the literature on assignment mechanisms emphasizes the importance of efficiency based on agents’ preferences, policymakers may want to achieve different goals. For instance, school districts may want to affect student learning outcomes but must take teacher welfare into account when assigning teachers to students in classrooms and schools. This paper studies both the potential efficiency and equity test-score gains from within-district reassignment of teachers to classrooms using novel data that allows us to observe decisions of both teachers and principals in the teacher internal-transfer process, and test-scores of students from the ob- served assignments. We jointly model student achievement and teacher and school principal decisions to account for potential selection on test-score gains and to predict teacher effectiveness in unobserved matches. Teachers, but not principals, are averse to assignment based on the teachers’ comparative advantage. Estimates from counterfactual assignments of teachers to classrooms imply that, under a constraint not to reduce any retained teacher’s welfare, average student test scores could rise by 7% of a standard deviation. Although both high and low achievers would experience average gains under this counterfactual, gains would be larger for high-achieving students.

We exploit data on the universe of public-school teachers and students in Per ́u to establish that wage rigidity makes teachers choose schools based on non-pecuniary factors, magnifying the existing urban-rural gap in student achievement. Leveraging a reform in the teacher compensation structure, we provide causal evidence that increasing salaries in less desirable locations is effective at improving student learning by attracting higher-quality teachers. We then build and estimate a model of teacher sorting across schools and student achievement production, whereby teachers are heterogeneous in their preferences over non-wage attributes and their comparative advantages in teaching different student types. Counterfactual compensation policies that leverage information about teachers’ preferences and value-added can result in a substantially more efficient and equitable allocation by inducing teachers to sort based on their comparative advantage.

In many decision-making environments, consumers' or investors' choice sets are designed by intermediaries with potentially misaligned preferences. This paper proposes an empirical framework to recover the preferences of such intermediaries from variation in the observed choice sets. The problem of designing a menu of options is modelled as a tractable multiple-discrete choice problem, which, given intermediaries and product characteristics, predicts the probability that a given product will be included in a menu. Intermediaries' preferences are then identified by matching observed and predicted inclusion probabilities. The second part of the paper examines the US market of employer-sponsored 401(k) retirement plans. It applies the framework to quantify the preference misalignment between employer sponsors, who select a menu of mutual funds to include in their 401(k), and worker investors, who allocate their retirement contributions across the funds available. Model estimates imply that investors' marginal valuations for several fund characteristics, including past performance and fees, are three to four times larger than sponsors' valuations. Aligning sponsors' and investors' preferences reduces plan expenses, reduces the number of funds offered and increases the number of index funds available. Furthermore, because model estimates suggest that a sizable share of investors is inactive, counterfactual policies mandating the inclusion of low-cost default options can generate considerable fee savings for worker investors.

Full automation using Artifical Intelligence (AI) predictions may not be optimal if humans can access contextual information. We study human-AI collaboration using an information experiment with professional radiologists. Results show that providing (i) AI predictions does not always improve performance, whereas (ii) contextual information does. Radiologists do not realize the gains from AI assistance because of errors in belief updating – they underweight AI predictions and treat their own information and AI predictions as statistically independent. Unless these mistakes can be corrected, the optimal human-AI collaboration design delegates cases either to humans or to AI, but rarely to AI assisted humans.