The dissertation consists of three essays on industrial organization. The first two chapters study the shale gas firms’ choices of lease locations in Pennsylvania. The third chapter is an independent study of two-sided matching games, with an emphasis on empirical methods.

The motivation of the first two chapters originates from a policy carried out by the Penn- sylvania government. In November 2010, the Pennsylvania Department of Environmental Protection began posting well-specific production data of the Marcellus Shale. The informa- tion potentially reveals the shale gas reserve distribution. The first two essays seek to study whether the production information disclosure has affected the shale gas operators’ choices of lease locations. In the first chapter, a simple theoretical model is proposed, in which firms sequentially choose between two locations to make their lease investment. The firms con- duct Bayesian updating of the locations’ output and maximize their expected return. The disclosure of the production information is assumed to affect the firms’ prior beliefs, with a higher productivity leading to a greater prior probability that the location has higher poten- tial. Analyses of the two-firm case shows that, when the prior beliefs are corrected adjusted by the production information, the expected number of firms choosing the location with higher output potential goes up. For the multiple-firm case, simulation is run and generates results that are consistent with the two-firm case.

The second chapter examines the model implication using the production and the lease ii

data from the Marcellus Shale in Pennsylvania. A Tobit model is assumed for the monthly incremental acreages of an operator at a municipality. The model is used to study how the acreage is determined by the signals observed by an operator, and how the correlations be- tween the acreage and the signals are affected by the production information. The acreages an operator obtains at the municipality in the past three months is used as a proxy for its pri- vate signal. Estimation shows that without the production information, a one acre increase in the acreages obtained by an operator at a municipality in the past three months leads to a 0.542 acre increase in the latent acreage the operator obtains at the same municipality in the current month. Once the production information is available, for each one percent increase in the municipality’s productivity, that number goes up by 0.013 acres. The estimation also finds that every one percent increase in productivity is associated with a 24.823 acre direct increase in the latent acreages. The findings are consistent with the prediction of the the- oretical mode, which shows that the production information improves the shale gas firms’ resource allocation efficiency.

The third chapter extends the marriage matching model in Choo and Siow (2006) by proposing two empirical models of one-to-one matching games. The first model considers a static one-to-one transferable utility matching game with search frictions, and studies its identification using a nonparametric approach. The second model examines a static one-to- one non-transferable utility matching game. The model is parametrically estimated, allowing heterogeneity by imposing random effects. In particular, a finite mixture model is employed and estimated using the Expectation Maximization algorithm.

I study how information is incorporated into prices. In the first two chapters, I approach this problem by studying the pricing of new products. Retailers face uncertainty about demand for new products, making it challenging to set prices. To learn about demand, retailers rely on information from similar, existing products and demand in similar markets. Thus, retailers have the least knowledge about demand for new products that are not present in any markets and have few substitutes. I show that to address uncertainty retailers launch new products into a subset of markets to learn about demand before introducing a product widely. As retailers gain additional information from similar existing products or prior introductions at other stores, they set prices closer to the optimal level. Additionally, after introducing a new product, retailers continue to learn about demand and refine their pricing strategy. Retailers learn, in part, by sharing information across their network of stores. Information sharing across different geographic regions helps retailers learn about demand if consumers are similar across locations. I find that while information sharing typically leads to prices that are closer to the optimal level, there are cases where information from different geographies has a negative impact on pricing.

In my third chapter, I take these themes in a different direction by studying pricing in financial markets from a theoretical perspective. In modern financial markets, investors seeking to profit from new information compete with high frequency traders (HFTs) who invest in the ability to trade quickly. Using a dynamic model, I find that while HFTs' speed advantage improves market liquidity (as indicated by bid-ask spreads), it also slows down the speed at which new information is incorporated into prices. HFTs decrease the profitability of new information by trading ahead of slower investors, prompting investors to trade less to avoid revealing valuable information. As a result, new information takes longer to be incorporated into prices. However, HFTs also improve market liquidity by ensuring that prices better reflect widely known information.

I study the determinants of the firms’ investment decisions.First, I quantify how building transport infrastructure affects investments. Whereas there is growing evidence that new infrastructure reduces trade costs, we lack a rigorous empirical understanding on how it impacts investments. To make progress, I leverage the U.S. crude oil industry. I specify a structural model of discrete investments centered around the producers’ dynamic trade-off between current revenues and delaying investments to wait for additional transport infrastructure. I bring this model to a comprehensive dataset that allows me to match the drilling activity of oil producers with the construction of pipelines across space and over time. I estimate that pipelines reduce the producers’ transportation costs by 15%, increasing the amount of new oil wells by 28%. In sum, I find that firms substantially increase investments in response to the new transport infrastructure. Then, I study how the firms’ ownership structure affects investments. Vertical integration could increase investments for the integrating firm and reduce investments for its rivals. I measure the investment outcomes of eight vertical mergers in the US motion picture industry using 1997-2019 movie-level and companies’ ownership data. Using a difference-in-difference research design, I estimate that vertical mergers increase investments for the integrating counterpart by 73.5%, while reducing investments for its rivals by 47%. Then I specify a within-firm model of resource allocation in order to separate the role of credit constraints and industry-specific technology from the change in the internalized return from investments. The results support the property rights theory.

This dissertation consists of two essays in the area of Industrial Organization.

In Chapter 1, I investigate the role of network effects in explaining the within-firm rate of technology adoption. I study the conversion of movie distribution and exhibition from 35mm film to digital technology. These industries constitute a hardware-software system with indirect network effects. I specify and estimate a dynamic oligopoly game of digital hardware adoption by movie theaters and digital movies (software) supply by movie distributors. Crucially, theaters' technology-adoption decisions are made at the screen level so diffusion occurs both within and across firms. Counterfactual simulations establish that: (1) at the industry level, diffusion occurs mainly within rather than across firms; (2) differences in technology adoption across firms, which are commonly attributed to scale economies and strategic incentives, are in part due to larger firms' ability to initially adopt the technology at a smaller scale. Therefore, explicitly accounting for intra-firm adoption dynamics is important to better explain aggregate diffusion and firm heterogeneity in technology adoption.

In Chapter 2, I study how non-cartel firms adjust their pricing to the supra-competitive level sustained by a cartel, and in doing so, may harm consumers via so-called "umbrella" damages. Such damages arise, in particular, when contracts are awarded through first-price procurement auctions. This chapter examines the bidding behavior of non-cartel firms bidding against the Texas school milk cartel between 1980 and 1992. Evidence is found that the largest non-cartel firm bid significantly higher when facing the cartel. Structural estimation of damages and inefficiencies due to the cartel agreement reveals that per contract: (1) damages from non-cartel firms overbidding are at least 47% of damages caused by the cartel, (2) when the outcome of the auction is inefficient, damages due to misallocation amount to 64% of cartel damages. Finally, inefficiencies raise the winner's cost by 3.7%. These results shed light on the potential importance of umbrella damages from a civil liability perspective.

In the first two chapters of this dissertation, I study the design of multi-object auctions. Using a large data set from the Brazilian public procurement sector, I show evidence that entry is costly and that the mix of products being auctioned off is a first-order effect to understand firm participation.

In the first chapter, I find evidence that the data is consistent with a theory of selection. The average entrant has a higher product match with the session, and they are closer to auction locations. Distance affects entry decisions negatively: a 1 unit (100km) increase in the distance to the auction location lowers the odds ratio for entry 0.91 times. At the same time, an additional auction in the set of potential auctions of a firm increases the odds for entry 1.62 times.

In terms of variable costs, a 1% increase in the distance to the auction location increases bid by 0.4% to 3.3%. There are also gains of scale in terms of the size of the contracts: a 1% increase in the contract quantity for a given product increases bids by 0.64% to 0.76%. The main force responsible for lowering procurement costs is the presence of additional bidders. I find that an extra bidder can lower costs between 21.2% and 32.4%. These results motivate and feed into the structural model presented in chapter 2.

In the second chapter, I continue to analyze this market with the focus on estimating entry costs and answering policy questions. To do so, I build a novel model of endogenous entry in multi-object auction sessions that allows me to disentangle two forces that affect entry decisions: entry costs, and the menu of items of a given session. The model has two stages. In the first stage, firms decide whether to enter an auction session and pay a fixed cost after observing an imperfect signal of their true cost. In the second stage, both the items for which they can bid and their costs are realized, and the auction takes place. I focus the analysis on type symmetric equilibria, where bidders of the same type follow the same entry strategy. In equilibrium, marginal bidders make zero profits. This condition allows me to link the unobserved entry costs to the observed bid behavior of entrants.

Having derived the equilibrium of the model, I estimate model fundamentals and turn to policy questions. The estimates provide evidence that entry is more attractive to local firms. I find that their cost distribution stochastically dominates the one from non-local firms. Moreover, conditioned on the number of items a firm can participate in, non-local firms face between 3.9% to 6.5% higher entry costs than local firms.

I focus on two counterfactual simulations. In the fully efficient scenario, where firms do not incur any entry costs, I find that procurement costs would be lowered by 22.5% to 40.1%. These are bounds on the maximum cost savings and also quantifies the degree of inefficiency present in this market. The second counterfactual is a partially efficient scenario, where non-local firms face the same entry costs as local firms. This analysis focuses on a selected equilibrium where firms enter the sessions sequentially. Firms are sorted according to a lexicographic order which is determined by the strength of their signal, number of items, and firm type (non-local/local). I find that procurement costs would be lowered by 2.8% to 2.9%. Thus, on this type of equilibrium and by holding on-site auctions, the government indirectly sacrificed some efficiency to the benefit of local firms.

In the third chapter, I study the pricing of platforms that offer consumers the choice between a free package, in which consumers are exposed to advertising, and a premium package, in which they pay to not be exposed to advertisements. I characterize its profit-maximizing and Pigouvian pricing, which allows me to analyze the degree to which the platform incorporates consumers’ distaste for advertising in its pricing scheme, as well as the trade-offs that emerge between the free and paid packages. The results contribute to the discussion of consumers' overexposure to advertising when platforms behave as a social planner and maximize their value.

This dissertation examines questions in industrial organization regarding the evaluation of antitrust and procurement policies. In the first two chapters, I examine whether antitrust agencies should consider the effects of product variety changes due to a merger. In the first chapter I test whether the merger had any effect on product variety directly. I find evidence of product variety changes due to a merger in the context of the MillerCoors merger of 2008. I find that the merged firm decreased the number of brands offered and offset this by increasing product variety in more successful brands. However, under a difference-and-differences framework, I find that product variety declined relative to that of other top competitors.

In the second chapter, I examine the welfare effects of product variety changes and compare them to the welfare effects of prices to see what should be the priority of antitrust agencies. I estimate demand for the MillerCoors merger in the postmerger period, expanding on previous work in the literature. In a set of two counterfactuals, I test the value of new products created after the merger and the value of discontinued products lost after the merger. I find that the merger increased consumer surplus from changes in product variety. Benchmarking this to results in the literature, I find that the effects of product addition and discontinuation are approximately 34% and -4%, respectively, of the consumer welfare effects of the postmerger price changes in the presence of coordinated pricing found in prior work.

In the third chapter, I provide a theoretical framework on how restrictions on firms participating in government procurement auctions affect local government procurement costs, firm participation for procurement decisions and firm investment decisions under the context of the Buy America Policy change in 2015. This policy increased the domestic materials requirement for federal funds to be used for purchases of transit goods, thereby restricting foreign firms from participating in procurement and potentially raising the costs of domestic firms that rely on foreign capital for their products. I find federal transportation funds used for railcars increased while funds used for buses remained flat. I then propose three models to examine the theoretical impact of the policy. The first model estimates a standard first price sealed-bid auction model standard for procurement. The second model separates bidders into two types: one domestic type which draws from a higher mean cost distribution, and one foreign type which draws from a lower mean cost distribution. I find costs and markups increase, while the participation of foreign firms declines.

This dissertation includes three chapters on the health insurance markets established by the Affordable Care Act (ACA), known as exchanges. Chapter 1 estimates the demand for each plan in the California exchange using a discrete choice model. The model incorporates heterogeneity in consumer preferences and in product characteristics, including hospital and primary care physician (PCP) networks. Endogeneity of prices is addressed using networking hospital costs as instruments, and prices for any given plan can vary across consumers within a market. Consumers are highly sensitive to prices, with market shares declining by 3%-5% for just a $1 increase in the premium. Demand also responds to hospital and PCP networks, but to a relatively small degree. Along the take-up margin, a $1 increase in premium subsidy increases take-up by 1.4%.

Chapter 2 uses a structural model of demand and supply to examine how two insurance market regulations--community rating and risk adjustment--affect prices and enrollment in the ACA exchange in California. Without risk adjustment, community rating in the ACA would lead to a significant reduction in enrollment in desirable plans and in take-up overall. Risk adjustment under the ACA roughly restores relative shares across plans to what they would be without community rating; however, the reduction in take-up is not restored. An alternative risk adjustment method can increase enrollment by 3.0% and would have little impact on government spending.

Chapter 3, written jointly with Isaac Menashe and Wesley Yin, examines the impact of information on insurance take-up in the ACA. We exploit experimental variation in the information mailed to 87,000 households in California's exchange to study the role of frictions in insurance take-up. We find that a basic reminder of the enrollment deadline raised enrollment by 1.4 pp (or 16 percent). Compared to the reminder alone, also reporting personalized subsidy benefits increases take-up among low-income individuals, but decreases take-up among higher-income individuals. This is despite reminder-only recipients eventually observing their subsidies before purchase. Finally, the letter interventions induced healthier individuals into the market, lowering aggregate spending risk by 5.9 percent, suggesting these interventions can improve both enrollment and average market risk.

This dissertation analyzes how technological and institutional changes influence economic development. Chapter 1 quantifies the long-term effect of refrigeration on agricultural production. Mechanical refrigeration greatly reduced transportation cost for perishable products but not non-perishable products. I leverage this differential effect to identify the effect of technological change on production with an event-study design. Results show that a one percentile increase in relative suitability of fodder versus wheat production leads to more land area being developed as farmland, higher values of total farm output, as well as higher land values. The effects were driven primarily by the top two quartiles of counties in terms of fodder versus wheat suitability, and most effects persisted until 1960.

Chapter 2 focuses on how the innovation influenced market power in the meatpacking industry. Refrigeration created a highly concentrated meatpacking industry due to the capital-intensive production process. Documents show that the five dominant firms formed a cartel to openly collude and manipulate the market by exploiting the time gap between sellers' shipment decisions and final sales. I leverage an exogenous regulatory change and novel high-frequency data to quantify the welfare effects of the manipulation. Compared to the standard monopsony model, I find that cartel manipulation harmed cattle sellers attracting more cattle to the market and purchases them at a lower average price. The manipulation strategy also harmed downstream consumers by increasing beef prices and thus total household food expenditures.

Chapter 3 uses the historical evolution of fence laws to examine the impact of liability rules on economic development. Fence laws assigned the liability for livestock trespassing to either farmers or ranchers. I use newly compiled data on the evolution of county-level fence laws to analyze the causal relationship between liability rules and land allocation, production decisions, and agricultural productivity. Results show that, by making livestock owners liable for trespassing, the fence-in rule increased farmland development, corn cultivation acreage and yield, and the total value of farm output. I conclude that, with substantial transaction costs, legal institutions that govern liability rules and property rights can have large and persistent effects on economic development.

My dissertation contributes to the structural nonparametric econometrics of auctions and contests with incomplete information. It consists of three chapters.

The first chapter investigates the identification and estimation of an all-pay auction where the object is allocated to the player with the highest bid, and every bidder pays his bid regardless of whether he wins or not. As a baseline model, I consider the setting, where one object is allocated among several risk-neutral participants with independent private values (IPV); however, I also show how the model can be extended to the multiunit case. Moreover, the model is not confined to the IPV paradigm, and I further consider the case where the bidders’ private values are affiliated (APV). In both IPV and APV settings, I prove the identification and derive the consistent estimators of the distribution of the bidders’ valuations using a structural approach similar to that of Guerre et al. (2000). Finally, I consider the model with risk-averse bidders. I prove that in general the model in this set-up is not identified even in the semi-parametric case where the utility function of the bidders is restricted to belong to the class of functions with constant absolute risk aversion (CARA).

The second chapter proves the identification and derives the asymptotically normal estimator of a nonparametric contest of incomplete information with uncertainty. By uncertainty, I mean that the contest success function is not only determined by the bids of the players, but also by the variable, which I call uncertainty, with a nonparametric distribution, unknown to the researcher, but known to the bidders. This work is the first to consider the incomplete information contest with a nonparametric contest success function. The limiting case of the model when there is no uncertainty is an all-pay auction considered in the first chapter. The model with two asymmetric players is examined. First, I recover the distribution of uncertainty using the information on win outcomes and bids. Next, I adopt the structural approach of Guerre et al. (2000) to obtain the distribution of the bidders’ valuations (or types). As an empirical application, I study the U.S. House of Representatives elections. The model provides a method to disentangle two sources of incumbency advantage: a better reputation, and better campaign financing. The former is characterized by the distribution of uncertainty and the latter by the difference in the distributions of candidates’ types. Besides, two counterfactual analyses are performed: I show that the limiting expenditure dominates public campaign financing in terms of lowering total campaign spending as well as the incumbent’s winning probability.

The third chapter is a semiparametric version of the second chapter. In the case when the data is sparse, some restrictions on the nonparametric structure need to be put. In this work, I prove the identification and derive the consistent estimator of a contest of incomplete information, in which an object is allocated according to the serial contest success function. As in previous chapters, I recover the distribution of the bidders’ valuations from the data on observed bids using a structural approach similar to that of Guerre et al. (2000) and He and Huang (2018). As a baseline model, I consider the symmetric contest. Further, the model is extended to account for the bidders’ asymmetry.

In the first chapter of this dissertation (joint with David Atkin, Joaquin Blaum, and Pablo Fajgelbaum), we study the determinants and consequences of discretionary trade policies in Argentina in the 2010s. Many countries utilize discretionary trade policies that target particular firms and products at particular times. Since they are often illegal under WTO rules, the ability to empirically study the drivers and consequences of these policies -and more generally to study hard-to-measure non tariff barriers- has been limited. We observe detailed records from a system of discretionary approvals affecting all imports. The restrictions varied across firms in a manner consistent with the government's rhetoric of encouraging investment and exports. Restrictions also varied across time in an attempt to safeguard the current account, becoming more stringent for initially less-restricted firms when macro conditions worsened. Utilizing these sources of variation to instrument for the quantity restrictions imposed by the policy, we find that firm-level quantitative restrictions increased the prices paid by importers- contrary to the prediction of falling prices generating terms of trade gains from competitive models of trade policy. A trade model with bargaining matched to these responses reveals that high bargaining power of foreign exporters vis-a-vis Argentinian importers shapes the impacts of trade policy.

In the second chapter of this dissertation I study the trends in exposure to natural hazard risk in the United States. In the past two decades, about half of the new homes in the United States were built in environmentally risky areas. Why is new residential development being exposed to such risk? I posit that land-use regulations restricting development in safer areas contribute to this pattern. I show that in most U.S. cities there is a negative correlation between inelastic housing supply and natural hazard risk. Moreover, regressions show that U.S. cities with a less elastic supply of housing in safe areas saw particularly high growth in at-risk areas in past decades: reducing the average housing supply elasticity of safe tracts by one standard deviation is associated with riskier tracts growing 5.7% more. Because the price elasticity of housingsupply is a function of land-use regulations, this pattern suggests that restrictive regulations on housing supply have the potential to drive exposure to a wide range of natural hazard risks all across the United States.

Finally, the third chapter of this dissertation studies the link between urban land-use policies and exposure to wildfire risk in San Diego. In the metropolitan area of San Diego, California, areas unexposed to risk are highly regulated and built out. I estimate a quantitative urban model using detailed spatial data on zoning, density limits, lot size restrictions, wildfire risk, and insurance. In the model, the regulations benefit landowners and reallocate the population to unregulated at-risk areas. These effects depend on estimated disamenities from wildfire risk, insurance access, and the spatial correlations between regulations, wildfire risk, and location amenities. I find that land-use regulations raise city-level rents by an average 28% and explain 7% of the residents living in fire-prone areas. The estimated present-discounted cost of wildfire risk is $14,149 per person, with existing regulations accounting for 10% of that cost. Over the next 40 years, as wildfire risk intensifies, the population grows, and the current land restrictions become more binding, the number of exposed residents will grow by 12%. The results show that institutions that restrain relocating out of harm's way, such as land-use regulations,can limit adaptation to climate change.