Empirical Implementation of Theoretical Models of Strategic Interaction and Dynamic Behavior

We recover valuations of time using detailed data from a large ride-hail platform, where drivers bid on trips and consumers choose between a set of rides with different prices and waiting times. Leveraging a consumer panel, we estimate demand as a function of both prices and waiting times and use the resulting estimates to recover heterogeneity in the value of time at the individual level. We study the welfare implications of platform pricing policies that take advantage of this heterogeneity. In particular, we compare the consumers’, drivers’, and platform’s welfare under different forms of price discrimination. Taking into account drivers’ optimal reaction to the platform’s pricing policy, total surplus falls by 6% under personalized pricing relative to the current mechanism. However, total surplus grows by 33% compared to the case in which the platform does not incorporate consumer information into its pricing.

We analyze a dynamic, informationally-restricted auction employed by a rental-car company to sell unwanted vehicles. At these auctions, which last two minutes, bids are submitted over the Internet in continuous time under the ascending-price format. Unlike at English or Japanese auctions, here participants are unable to observe either the number of other bidders at the auction or the bids of those opponents: at any time in the auction, each participant only knows the history of his own bids, and whether his current bid is the high one. At an auction’s end, the highest bidder wins the vehicle on sale, and pays his bid. In our data, we note that a substantial portion of bidding occurs early in these sales, even though a game-theoretic analysis suggests that the informational restrictions create strong incentives for bid sniping—that is, waiting to submit a bid in the final seconds of the auction. If all bidders used a non-informative bid-sniping strategy, then the outcome of the auction is equivalent to that of a one-shot, first-price, sealed-bid auction. Participants, however, have an incentive to bid early at these auctions—to learn the current high bid, which a bidder can only discover when he holds that high bid. Yet early bidding can lead to their collective disadvantage—inducing the participants to bid more than they would at a one-shot, first-price, sealed-bid auction. We develop theoretical and empirical models of bidding strategies at these auctions and estimate bidders’ beliefs and distributions of valuations for used cars, and then employ our model and estimated distribution of valuations to compare auction revenues under the dynamic auction with counterfactual revenues from a one-shot, sealed-bid auction with and without a reserve price under the same informational restrictions—namely, bidders do not know the number of opponents they face at an auction.

We examine identification of differentiated products demand when one has “micro data” linking individual consumers’ characteristics and choices. Our model nests standard specifications featuring rich observed and unobserved consumer heterogeneity as well as product/market-level unobservables that introduce the problem of econometric endogeneity. Previous work establishes identification of such models using market-level data and instruments for all prices and quantities. Micro data provides a panel structure that facilitates richer demand specifications and reduces requirements on both the number and types of instrumental variables. We address identification of demand in the standard case in which non-price product characteristics are assumed exogenous, but also cover identification of demand elasticities and other key features when product characteristics are endogenous. We discuss implications of these results for applied work.

Using a unique data set from the sale of train tickets in Sweden, we study the “declining price anomaly” in sequential auctions theoretically and empirically. First, our reduced form analysis suggests that a model with ambiguity averse bidders fits the observed bidder behavior in these auctions better than other existing models. Motivated by this, we study sequential second-price auctions that closely resemble the auction mechanism for train tickets and assume bidders have maxmin expected utilities over multiple priors. In the unique symmetric equilibrium, bidders use their worst-case conditional beliefs to evaluate their payoffs. The equilibrium generates declining prices due to an underestimation of future payoffs that is brought on by ambiguity aversion. We also provide a new revenue ranking between some common multi-unit auction formats and show that ambiguity raises the seller’s revenue despite declining prices.Finally, we non-parametrically estimate both the true distribution of valuations and the worst-case beliefs using a novel identification technique that exploits bidders’ inter-temporal first order conditions. Our estimation uncovers a first-order stochastic dominance relationship between beliefs and the true distribution, which is consistent with ambiguity aversion. Our counterfactuals show that, while ambiguity increases the seller’s revenue by at least 18% compared to the common prior case, switching to sequential first-price auctions would further increase revenue by at least 11%.

This paper explores the extent to which information design can increase the auctioneer’s revenue in the US offshore oil/gas lease auctions. In our setting, firms first submit cash bids in a first-price sealed-bid auction; the winning firm then decides whether to explore a tract, and the government receives a royalty payment on the production value of the tract. We first document that when all the bids are revealed to the winning bidder after the auction, the exploration rate is positively correlated with the losing bids, suggesting that the winning bidder utilizes the rivals’ bids to infer their private information on the tract’s production value. We then characterize the equilibrium bidding strategy in an environment in which the auctioneer can design and commit to how to reveal information on the losing bids to the winning bidder. Our counterfactual exercises show that information design can significantly improve the auctioneer’s revenue. For example, fully withholding information on the losing bids results in an increase in the per-tract profit of between $204K and $901K.

Retired couples dissave housing wealth at a much slower rate than singles, conditional on income. This paper studies mechanisms through which marital transitions affect housing decisions of retirees. We develop and estimate a life-cycle savings model where marital transitions affect long-term care arrangements, bequest motives, and eligibility for means-tested welfare programs. We find that the key driver behind the stark difference in dissaving of housing wealth between retired couples and singles varies substantially by income. For low-income households, how means-tested public insurance treats housing has the most impact on their housing decisions. For middle- and high income households, family caregiving and bequest motives are the dominant driver, respectively. Our counterfactual policy experiments show that the current structure of the Medicaid estate recovery program which exempts housing wealth only for couples is more desirable than alternative rules, such as extending the homestead exemption to singles or providing the exemption to singles only. By inducing lower-income couples to decumulate housing wealth at a slower rate, the current Medicaid program reduces impoverishment risk in retirement.

We develop and estimate a life-cycle consumption savings model in which observed genetic variation is allowed to affect wealth accumulation through several distinct channels. We focus on genetic markers that predict educational attainment, aggregated into a predictive index called a polygenic score. Based on substantial reduced form evidence, we allow variation in these endowments to affect earnings, the disutility of labor, stock market participation costs, and idiosyncratic rates of return on risky investments. The model incorporates endogenous retirement and a realistic social security system. Parameter estimates suggest that, in addition to earnings, genetic differences are significantly associated with both the disutility of labor and risky asset returns. Counterfactual policy exercises indicate that two ways to lower costs of an aging population (extending the age of retirement or cutting social security benefits) have different distributional consequences with respect to genetic heterogeneity. The results highlight how untargeted policies can significantly alter genetic gradients in economic outcomes.

Substantial earnings differences exist across majors with those that pay well also assigning lower grades and higher workloads. We show that the stricter grading policies in STEM and economics courses reduce STEM enrollment, especially for women. To show this, we estimate a model of student demand for courses and optimal effort choices of students given professor grading policies. Professor grading policies are treated as equilibrium objects that in part depend on student demand for courses. Restrictions on grading policies that equalize average grades across classes reduce the STEM gender gap and increase overall enrollment in STEM classes.

Does charter school expansion improve learning outcomes for the average student? Absent plausibly random variation in the presence of charters at the market-level, separate literatures have focused on 1) identifying treatment effects on students who choose charter schools; 2) estimating competitive impacts of charter schools on students who remain in public schools. This paper develops and estimates an empirical framework that puts these two pieces together in order to evaluate the impact of charters on education quality in the aggregate. In the model, students choose schools as a function of distance to their residence, type, and quality; public schools supply quality given demand; and charter schools choose locations in the market. Charter schools are also heterogeneous according to whether they offer an alternative education program, such as project-based learning. To estimate the model parameters, we use rich student-level microdata from North Carolina and target spatial difference-in-differences effects of exposure to charter schools on public schools' value-added. The estimates allow us to quantify the competitive incentives facing public schools and characterize strategic differentiation by charters in terms of elasticities of substitution. We use the estimated model to perform several policy analyses of interest, including one that assesses the aggregate effects of North Carolina's lifting of its statewide charter school cap in 2011. We find modest but meaningful gains in learning experienced by the average student due to charter school expansion. The gains are concentrated among inframarginal students, who benefit from competition, but the test score effects are negative for those students induced into charter schooling by the expansion.

This paper presents an empirical framework to study the assignment of children into foster homes and its implications on placement outcomes. The empirical application uses a novel dataset of confidential foster care records from Los Angeles County, CA. The estimates of the empirical model are used to examine policy interventions aimed at improving placement outcomes. In general, it is observed that market thickness tends to improve expected placement outcomes. If placements were assigned across all the administrative regions of the county, the model predicts that (i) the average number of foster homes children go through before exiting foster care would decrease by 8% and (ii) the distance between foster homes and children’s schools would be reduced by 54%.

Using rich data on hourly physical productivity and five thousand ownership changes from US power plants, we study the effects of mergers and acquisitions on efficiency and provide evidence on the mechanisms. We find that acquired plants experience an average of four percent efficiency increase five to eight months after the acquisition. Three-quarters of this efficiency gain is explained by increased productive efficiency; the rest comes from improved capacity management at the plant level and allocative efficiency at the portfolio level. Our findings suggest that acquisitions reallocate assets to more productive uses: we find that high productive firms buy underperforming assets from low productive firms and make the acquired asset almost as productive as their existing assets after the acquisition. Finally, investigating the mechanism, the evidence suggests that acquired plants achieve higher efficiency through low-cost operational improvements rather than high-cost capital investments. 

How does supporting early clean technologies affect the long-run transition away from dirty technologies? Early policy action generates immediate environmental benefits from increased adoption of available efficient products, but may result in intertemporal substitution away from later products with greater potential for reducing externalities. I examine how standards and subsidies supporting early advancements in lighting efficiency (halogens, CFLs) impacted the adoption of later products with higher efficiency (LEDs). I estimate a model of residential lighting demand, using structural methods adapted from dynamic models to capture how the market size and distribution of consumer heterogeneity depended endogenously on the history of past purchases. Counterfactual simulations suggest that delaying the implementation of standards from 2012 to 2018 would result in 36% greater LED sales over this period, while delaying the phase-out of CFL subsidies from 2012 to 2018 would result in 20% fewer LEDs sold. Across a range of specifications, I find that environmental benefits from some early policy action outweigh the environmental cost of reduced LEDs adoption; the overall environmental externality is minimized when standards are implemented in 2012 and CFL subsidies are phased-out after 2014. Sensitivity analyses around alternative technology lifetimes, externalities, and innovation responses identify conditions under which early policy intervention would be counterproductive.

We investigate the relationship between the interest rates offered to consumers in a deposit banking market and the contact that banks in that market have with each other in other markets. We show, in a simple theoretical model, that such overlapping relationships lead to less competitive behavior by banks. Furthermore, we empirically test this result across U.S. deposit banking markets and find that markets in which banks have many other points of contact with each other act significantly less competitive. Our results are particularly alarming as multimarket contact has increased significantly over the last two decades while the passthrough rate between the Federal Funds rate and deposit banking rates has fallen dramatically.

In this paper, we show that product design shapes search frictions and study how intermediaries may leverage this channel to increase their rents. In the US, the majority of municipal bonds are designed via negotiations between a local government and its underwriter. They are then traded in a decentralized market, where the underwriter often also acts as an intermediary. Exploiting variations in state regulations that limit government officials' conflicts of interest, we provide evidence that bond design from the government's perspective involves a trade-off between flexibility and liquidity, but the underwriter benefits from designing and trading complex bonds. Motivated by these findings, we build and estimate a model of bond origination and trades to quantify market inefficiency driven by underwriters' role in intermediating trades and discuss policy implications.

This paper studies how to design electricity markets to reduce emissions and prevent blackouts. Zero-emission renewable energy sources, such as wind and solar, are intermittent, which can lead to blackouts if the addition of renewables causes more reliable power plants to retire. To quantify the impact of electricity market policies, I build a structural equilibrium model of investment and dis-investment in generators of different energy sources. Oligopolistic firms make dynamic decisions to build or retire generators based on the profits they receive from wholesale electricity markets, which respond to the composition of generators supplying electricity. Using data from the electricity market in Western Australia, I estimate this model and use it to simulate investment and production under counterfactual policies. Carbon taxes reduce emissions but, for certain values, can result in an increase in blackouts by causing retirement of coal and gas plants. Subsidizing capacity prevents this from occurring, but at the expense of higher emissions. Using both policies together, however, keeps reliable, emissions-intensive generators in the market and prevents them from being used unless necessary, substantially lowering emissions while keeping the likelihood of blackouts low. I also explore alternative environmental policies, which are less effective at reducing emissions but have a lower cost to consumers.