Essays in Industrial Organization

Abstract

This thesis comprises three chapters on the US trucking industry. The first and second chapters study long-term relationships between shippers and carriers. The third studies how artificial intelligence changes human decision-making in the context of the maintenance of heavy-duty trucks.

The first chapter (joint with Thi Mai Anh Nguyen) provides evidence on the scope and incentive mechanisms of long-term relationships in the US truckload freight industry. In this setting, shippers and carriers engage in repeated interactions under fixed-rate contracts that leave scope for inefficient opportunism. We show that shippers use the threat of relationship termination to deter carriers from short-term opportunism. Carriers respond to the resultant dynamic incentives, behaving more cooperatively when their potential future rents are higher. While shippers and carriers often interact on multiple lanes, we find evidence that shippers' incentive schemes do not take advantage of this multi-lane scope for certain classes of carriers.

The second chapter (also joint with Thi Mai Anh Nguyen) builds on the first, exploring a market-level tradeoff that informal long-term relationships present. On the one hand, relationships capitalize on match-specific efficiency gains and mitigating incentive problems. On the other hand, the prevalence of long-term relationships can also lead to thinner, less efficient spot markets. We develop an empirical framework to quantify the market-level tradeoff between long-term relationships and the spot market. We apply this framework to an economically important setting—the US truckload freight industry—exploiting detailed transaction-level data for estimation. At the relationship level, we find that long-term relationships have large intrinsic benefits over spot transactions. At the market level, we find a strong link between the thickness and the efficiency of the spot market. Overall, the current institution performs fairly well against our first-best benchmarks, achieving 44% of the relationship-level first-best surplus and even more of the market-level first-best surplus. The findings motivate two counterfactuals: (i) a centralized spot market for optimal spot market efficiency and (ii) index pricing for optimal gains from individual long-term relationships. The former results in substantial welfare loss, and the latter leads to welfare gains during periods of high demand.

The third chapter (joint with Maggie Yellen) uses observational data to study how a predictive algorithm changes human decision-making. Using a rich decision-level data set from the maintenance of heavy-duty trucks, we document how skilled technicians' decision-making is changed by the introduction of an algorithm designed to predict the risk of truck breakdowns. We develop and estimate a model of technician decision-making that accounts for variation in monetary and non-monetary costs. Using an embedded neural network, we flexibly estimate technicians' beliefs about the probability of truck breakdowns both before and after the introduction of the algorithm. Comparing these estimated beliefs with an objective breakdown probability, we find that the algorithm significantly improves technicians' ability to predict breakdowns: the algorithm narrows the gap between actual and optimal costs by 79%. All of this gain comes from decreased repair costs, suggesting that the algorithm primarily helps technicians avoid low value repairs.