Publications

Collective evaluation processes, which offer individuals an opportunity to assess quality, have transcended mainstream sectors (e.g., books, restaurants) to permeate professional contexts from within and across organizations to the gig economy. This paper introduces a theoretical framework to understand how evaluators’ visibility into prior evaluations influences the subsequent evaluation process: the likelihood of evaluating at all and the value of the evaluations that end up being submitted. Central to this discussion are the conditions under which evaluations converge—are more similar to prior evaluations—or diverge—are less similar—as well as the mechanisms driving observed outcomes. Using a quasinatural experiment on a platform where investment professionals submit and evaluate investment recommendations, I compare evaluations that are made with and without the possibility of prior ratings influencing the subsequent evaluation process. I find that when prior ratings are visible, convergence occurs. The visibility of prior evaluations decreases the likelihood that a subsequent evaluation occurs by about 50%, and subsequent evaluations become 54%–63% closer to the visible rating. Further analysis suggests that peer deference is a dominant mechanism driving convergence, and only professionals with specialized expertise resist peer deference. Notably, there is no evidence that initial ratings are related to long-term performance. Thus, in this context, convergence distorts the available quality signal for a recommendation. These findings underscore how the structure of evaluation processes can perpetuate initial stratification, even among professionals with baseline levels of expertise.

I study whether firms that receive targeted U.S. state-level subsidies are more likely to subsequently engage in corporate misconduct. I find that firms are more likely to engage in misconduct in subsidizing states, but not in other states that they operate in, after receiving state subsidies. Using data on both federal and state enforcement actions, and exploiting the legal principle of dual sovereignty for identification, I show that this finding reflects an increase in the underlying rate of misconduct and that this increase is attributable to lenient state-level misconduct enforcement. Collectively, my findings present evidence of an important consequence of targeted firm-specific subsidies: non-financial misconduct that potentially could impact the very stakeholders subsidies are ostensibly intended to benefit.

The real estate market is highly intermediated, with 90 percent of buyers and sellers hiring an agent to help them transact a house. However, low barriers to entry and fixed commission rates result in a market where inexperienced intermediaries have a large market share, especially following house price booms. Using rich micro-level data on 8.5 million listings and a novel instrumental variables research design, we first show that houses listed for sale by inexperienced real estate agents have a lower probability of selling, and this effect is strongest during the housing bust. We then study the aggregate implications of the distribution of agents’ experience on housing market liquidity by building a dynamic entry and exit model of real estate agents with aggregate shocks. We find that 3.7% more listings would have been sold in a flexible commission equilibrium. Eighty percent of this improvement comes from competition driving down overall seller commissions, while the remaining share can be attributed to commission variation across experience levels.

Housing market conditions are often used as leading indicators of real business cycles. Does the housing market also lead the financial cycle? We address this question by applying deep learning OCR techniques to create a new hand-collected database spanning a century of monthly building permit quantities and valuations for all U.S. states and the 60 largest MSAs. We show that the option to build embedded in permits renders volatility in residential building permit growth (BPG) a strong predictor of aggregate and cross-sectional stock and corporate bond return volatility. This predictability remains even after conditioning on a battery of factors, including corporate and household leverage and firms' exposure through their network of plants to other localized physical risks like natural disasters. Cities with more elastic housing supply consistently predict stock market downturns at 12-month horizons, resulting in new trading strategies to hedge against overbuilding risk.

Sparsity is a central aspect of interpretability in machine learning. Typically, sparsity is measured in terms of the size of a model globally, such as the number of variables it uses. However, this notion of sparsity is not particularly relevant for decision making; someone subjected to a decision does not care about variables that do not contribute to the decision. In this work, we dramatically expand a notion of decision sparsity called the Sparse Explanation Value (SEV) so that its explanations are more meaningful. SEV considers movement along a hypercube towards a reference point. By allowing flexibility in that reference and by considering how distances along the hypercube translate to distances in feature space, we can derive sparser and more meaningful explanations for various types of function classes. We present cluster-based SEV and its variant tree-based SEV, introduce a method that improves credibility of explanations, and propose algorithms that optimize decision sparsity in machine learning models

Limited capacity in the emergency department (ED) secondary to boarding and crowding has resulted in patients receiving care in hallways to provide access to timely evaluation and treatment. However, there are concerns raised by physicians and patients regarding a decrease in patient centered care and quality resulting from hallway care. We sought to explore social risk factors associated with hallway placement and operational outcomes.

We study the cross-sector allocation of R&D resources in a multisector growth model with an innovation network, where one sector's past innovations may benefit other sectors' future innovations. Theoretically, we solve for the optimal allocation of R&D resources. We show a planner valuing long-term growth should allocate more R&D toward central sectors in the innovation network, but the incentive is muted in open economies that benefit more from foreign knowledge spillovers. We derive sufficient statistics for evaluating the welfare gains from improving R&D allocation. Empirically, we build the global innovation network based on patent citations and establish its empirical importance for knowledge spillovers. We evaluate R&D allocative efficiency across countries using model-based sufficient statistics. Japan has the highest allocative efficiency among the advanced economies. For the U.S., improving R&D allocative efficiency to Japan's level could generate more than 19.6% welfare gains.

We study how countries compete to become an international safe asset provider. Gov- ernments in our model issue debt to a common set of investors, resulting in competition as issuance by one country raises required yields for all countries. Governments are tempted ex post to engage in expropriation or capital controls, and can build reputa- tion as a safe asset provider by resisting temptation to do so. We show how increased competition deters countries from building reputation, leaving more countries stuck at low reputation levels and unable to supply safe assets. We derive a model-implied measure of country reputation. We estimate this reputation measure using micro-data on investor portfolio holdings, and use it to track the evolution of countries’ reputa- tion over time. We study how an incumbent safe asset provider, like the U.S., uses its issuance strategy to deter the emergence of competitors.

Science, technology, engineering, and mathematics (STEM) education can be stressful, but uncertainty exists about (a) whether stressful academic settings elevate cortisol, particularly among students from underrepresented racial/ethnic groups, and (b) whether cortisol responses are associated with academic performance. In four classes around the first exam in a gateway college STEM course, we investigated participants' (N = 271) cortisol levels as a function of race/ethnicity and tested whether cortisol responses predicted students' performance. Regardless of race/ethnicity, students' cortisol, on average, declined from the beginning to the end of each class and across the four classes. Among underrepresented minority (URM) students, higher cortisol responses predicted better performance and a lower likelihood of dropping the course. Among non-URM students, there were no such associations. For URM students, lower cortisol responses may have indicated disengagement, whereas higher cortisol responses may have indicated striving. The implication of cortisol responses can depend on how members of a group experience an environment.

We propose that investment strategies should be evaluated based on their net-oftrading- cost return for each level of risk, which we term the “implementable efficient frontier.” While numerous studies use machine learning return forecasts to generate portfolios, their agnosticism toward trading costs leads to excessive reliance on fleeting small-scale characteristics, resulting in poor net returns. We develop a framework that produces a superior frontier by integrating trading-cost-aware portfolio optimization with machine learning. The superior net-of-cost performance is achieved by learning directly about portfolio weights using an economic objective. Further, our model gives rise to a new measure of “economic feature importance.”

Medical crowdfunding is a popular channel for people in need of financial help with paying medical bills to collect donations from large numbers of people. However, large heterogeneity exists in the amount of donations each case receives and such uncertainty in fundraising outcomes hinders making timely treatment plans for patients. It is important to provide early and accurate predictions for medical crowdfunding performance, and help fundraisers engage in timely interventions. In this study, we propose a new approach that effectively combines time-varying features and time-invariant features in a deep learning model. This model provides dynamic predictions of fundraising outcomes, based on fixed case-level attributes and daily updated measures of social media activities. Compared with a rich set of baseline models, our model consistently demonstrates higher predictive accuracy while requiring a shorter observation window of data, thus achieving both accurate and early prediction objectives. We conduct a temporal clustering analysis to analyze the heterogeneous patterns in how the time-varying features relate to fundraising outcomes. In addition, we conduct a simulation analysis to demonstrate that interventions from fundraisers can significantly improve the fundraising performance of disadvantaged cases that are predicted to receive the lowest donation amounts, particularly when the interventions are taken early. These findings show that our deep learning prediction model and the actionable insights can provide timely feedback to fundraisers and promote equal access to resources for all. Our proposed approach is generalizable to different contexts, enabling effective processing of diverse sources of data and informing timely interventions early on.

We analyze optimal monetary policy and its implications for asset prices, when aggregate demand has inertia and responds to asset prices with a lag. If there is a negative output gap, the central bank optimally overshoots aggregate asset prices (asset prices are initially pushed above their steady-state levels consistent with current potential output). Overshooting leads to a temporary disconnect between the performance of financial markets and the real economy, but accelerates the recovery. When there is a lower-bound constraint on the discount rate, overshooting becomes a concave and non-monotonic function of the output gap: the asset price boost is low for a deeply negative initial output gap, grows as the output gap improves over a range, and shrinks toward zero as the output gap improves further. This pattern also implies that good macroeconomic news is better news for asset prices when the output gap is more negative. Finally, we document that during the Covid-19 recovery, the policy-induced overshooting was large—sufficient to explain the high levels of stock and house prices in 2021.

Demand rebounded more rapidly than expected after the COVID pandemic. This interacted with a series of unprecedented supply-side shocks around the shutdown and reopening of economies, along with broad-based increases in commodity prices following the invasion of Ukraine. Inflation spiked to the highest level in decades. In response, central banks tightened monetary policy sharply. This book explores the commonalities and differences in countries’ strategies, as well as lessons for the next inflationary episode.

Demand rebounded more rapidly than expected after the COVID pandemic. This interacted with a series of unprecedented supply-side shocks around the shutdown and reopening of economies, along with broad-based increases in commodity prices following the invasion of Ukraine. Inflation spiked to the highest level in decades. In response, central banks tightened monetary policy sharply. This book explores the commonalities and differences in countries’ strategies, as well as lessons for the next inflationary episode.

A monotone function interval is the set of monotone functions that lie pointwise between two fixed monotone functions. We characterize the set of extreme points of monotone function intervals and apply this to a number of economic settings. First, we leverage the main result to characterize the set of distributions of posterior quantiles that can be induced by a signal, with applications to political economy, Bayesian persuasion, and the psychology of judgment. Second, we combine our characterization with properties of convex optimization problems to unify and generalize seminal results in the literature on security design under adverse selection and moral hazard.