Latest Research Publications

This paper studies the unintended effect of English language requirement on educational inequality by investigating how the staggered rollout of English listening tests in China’s high-stakes National College Entrance Exam (NCEE) affected the rural–urban gap in college access. Leveraging administrative data covering the universe of NCEE participants between 1999 and 2003, we find that the introduction of English listening tests significantly lowered rural students’ exam score percentile ranks relative to their urban counterparts, resulting in a 30% increase in the rural–urban gap in college access. Our back-of-the-envelope calculations suggest that, as a result of this policy change, more than 54,000 rural students lost college seats to their urban peers between 1999 and 2003, and another 11,000 rural students who elite colleges could have admitted ended up in non-elite colleges, causing them significant future income losses.

This paper explores the effects of public information such as accounting earnings in a competitive lending setting with risk shifting. Debt financing creates incentives for borrowers to take on excessive risks, in particular in bad states of the world. If a privately informed inside creditor bids against outside creditors to extend a loan, public information levels the playing field, which affects the bidding and risk shifting. Nonetheless, a perfect public signal would yield the least efficient outcome: introducing some measurement noise alleviates risk shifting by subjecting outside creditors to the winner’s curse, allowing borrowers in bad states cheaper access to loans. However, for pessimistic priors about the borrower, greater public signal precision can alleviate risk shifting, at the margin. We discuss implications for financial reporting regulations along the business cycle and for creditor turnover.

In 2013, the U.S. Consumer Financial Protection Bureau released a database of consumer complaints filed against banks under its supervision (“CFPB banks”). We find that after the disclosure, rival banks exhibit a greater increase in mortgage approval rates in markets with more intensive mortgage complaints about CFPB banks. The effect is weaker when rivals have more expertise in the local market, are less concerned about credit risk due to mortgage sales, and locate in areas with more alternative information about the CFPB banks. The effect is concentrated in severe complaints and complaints related to loan underwriting practices. In addition to approving more loans, rivals also open more branches and are more likely to post a job opening in these markets. The findings suggest that these banks learn from the nonfinancial disclosures about operational deficiencies of peers (i.e., CFPB banks) in local markets, which alleviates their adverse selection concern about expanding.

We introduce a novel stakeholder-oriented framework that highlights variance in the application and expression of gender bias in the upper echelons. Directed by their relationship with the firm’s leadership, we theorize that stakeholders’ appraisals of top female leaders map onto a categorical and complex continuum. At the “categorical” end of this continuum, stakeholders neither have access nor are attentive to capability cues from the leader, increasing their reliance on stereotypes and gender biases in their leader evaluations. At the “complex” end of the continuum, stakeholders have access and are attentive to capability cues from the leader, decreasing their reliance on stereotypes and increasing their ability to accurately evaluate the leader. Between these ends, stakeholders evaluate female leaders by applying stereotypes and striving for accuracy to varying degrees. Each region on this continuum is linked to an array of behavioral responses, directed by stakeholders toward a target leader, that differ in valence and intensity. Our framework has significant implications for understanding a variety of social biases beyond gender, and enables the development of tailored strategies that can be used to facilitate accurate leader evaluations by all stakeholders.

Disputes over transactions on two-sided platforms are common and usually arbitrated through platforms’ customer service departments or third-party service providers. This paper studies crowd-judging, a novel crowdsourcing mechanism whereby users (buyers and sellers) volunteer as jurors to decide disputes arising from the platform. Using a rich data set from the dispute resolution center at Taobao, a leading Chinese e-commerce platform, we aim to understand this innovation and propose and analyze potential operational improvements with a focus on in-group bias (buyer jurors favor the buyer, likewise for sellers). Platform users, especially sellers, share the perception that in-group bias is prevalent and systematically sways case outcomes as the majority of users on such platforms are buyers, undermining the legitimacy of crowd-judging. Our empirical findings suggest that such concern is not completely unfounded: on average, a seller juror is approximately 10% likelier (than a buyer juror) to vote for a seller. Such bias is aggravated among cases that are decided by a thin margin and when jurors perceive that their in-group’s interests are threatened. However, the bias diminishes as jurors gain experience: a user’s bias reduces by nearly 95% as experience grows from zero to the sample median level. Incorporating these findings and juror participation dynamics in a simulation study, the paper delivers three managerial insights. First, under the existing voting policy, in-group bias influences the outcomes of no more than 2% of cases. Second, simply increasing crowd size through either a larger case panel or aggressively recruiting new jurors may not be efficient in reducing the adverse effect of in-group bias. Finally, policies that allocate cases dynamically could simultaneously mitigate the impact of in-group bias and nurture a more sustainable juror pool.

Price reductions take either an integrated form (e.g., a discount shown directly on the price tag) or a non-integrated form (e.g., a discount contained in a coupon sent to consumers and thus separate from the price tag). This research examines how non-integrated versus integrated promotions influence choices among vertically differentiated products. Under an integrated promotion (e.g., $10 off) applicable to multiple products (e.g., original list prices: $50 vs. $30), consumers directly compare these products’ post-promotion final prices displayed on their price tags (after a reduction of $10: $40 vs. $20). In contrast, under a non-integrated promotion of the same monetary value, consumers simply compare these products’ original list prices ($50 vs. $30) and neglect their post-promotion final prices, which require calculations. The list prices ($50 vs. $30; relative to the final prices: $40 vs. $20) as a basis for price comparison reduce the perceived price difference between these products. Consequently, a non-integrated promotion (compared to an integrated promotion) increases consumers’ choice of higher-priced products. A series of experiments (N = 5,187) demonstrate this effect and support the final price neglect mechanism. Furthermore, although attenuated, this effect still emerges for price reductions of a smaller magnitude or in a percent-off format.

Compliance-driven investments in technology—or “RegTech”—are growing rapidly. To understand the effects on the financial sector, we study firms’ responses to new internal control requirements. Affected firms make significant investments in ERP and hardware. These expenditures then enable complementary investments that are leveraged for noncompliance purposes, leading to modest savings from avoided customer complaints and misconduct. IT budgets rise and profits fall, especially at small firms, and acquisition activity and market concentration increase. Our results illustrate how regulation can directly and indirectly affect technology adoption, which in turn affects noncompliance functions and market structure.

Stochastic kriging has been widely employed for simulation metamodeling to predict the response surface of complex simulation models. However, its use is limited to cases where the design space is low-dimensional because in general the sample complexity (i.e., the number of design points required for stochastic kriging to produce an accurate prediction) grows exponentially in the dimensionality of the design space. The large sample size results in both a prohibitive sample cost for running the simulation model and a severe computational challenge due to the need to invert large covariance matrices. Based on tensor Markov kernels and sparse grid experimental designs, we develop a novel methodology that dramatically alleviates the curse of dimensionality. We show that the sample complexity of the proposed methodology grows only slightly in the dimensionality, even under model misspecification. We also develop fast algorithms that compute stochastic kriging in its exact form without any approximation schemes. We demonstrate via extensive numerical experiments that our methodology can handle problems with a design space of more than 10,000 dimensions, improving both prediction accuracy and computational efficiency by orders of magnitude relative to typical alternative methods in practice.

This paper considers a moral hazard problem where the agent can choose any output distribution with a support in a given compact set. The agent's effort-cost is smooth and increasing in first-order stochastic dominance. To analyze this model, we develop a generalized notion of the first-order approach applicable to optimization problems over measures. We demonstrate each output distribution can be implemented and identify those contracts that implement that distribution. These contracts are characterized by a simple first-order condition for each output that equates the agent's marginal cost of changing the implemented distribution around that output with its marginal benefit. Furthermore, the agent's wage is shown to be increasing in output. Finally, we consider the problem of a profit-maximizing principal and provide a first-order characterization of principal-optimal distributions.