This collection highlights the ongoing research our faculty are actively developing and exploring. We thank our faculty for sharing their work.
Accounting
Does the Director Labor Market Learn from Employees’ Social Media Ratings of the CEO?
Kevin K. Li, Ivy Zhang, and Yong Zhang
Using employee-submitted reviews on Glassdoor.com, we examine whether the director labor market learns from employees’ assessments of CEOs. Employees’ approval ratings of the CEO provide a unique perspective on the CEO as an individual, distinct from evaluations of firm or management team performance. We find that CEOs with higher employee approval ratings subsequently gain more independent directorships, after controlling for firm performance and various firm and CEO characteristics. The effect is not driven by other employee-provided ratings, such as those of business outlook or management team, suggesting that CEO approval ratings likely unveil unique information about CEOs’ personal abilities and skills beyond firm or team performance. Furthermore, the positive relation between CEO approval ratings and directorship gains weakens (strengthens) following quasi-exogenous shocks that decrease the perceived credibility of approval ratings (traditional information sources). This relation also strengthens when outsiders face greater information asymmetry, when firm performance is a noisier signal of CEO quality, or when employee ratings are more informative. Finally, firms appointing CEOs with higher employee ratings as independent directors experience a more positive market reaction to the appointments and a larger improvement in future financial performance, with the effect concentrated in firms where labor is a more critical input. Overall, we provide new evidence on the role of social media postings by rank-and-file employees in informing the external director labor market.
Externalities from Forcing Hospitals to Audit: Evidence from the Single Audit Act
Zhaosong Ruan, Mohan Venkatachalam, Xinyi Xie, and Vincent (Qiru) Zhang
This paper examines the real effects of financial statement audits in non-profit and government hospitals, using changes in audit requirements under the Single Audit Act. Comparing hospitals newly exempted from audits with those consistently audited or never audited, we find that exempted hospitals improved efficiency, measured by a 3.5% reduction in the cost-to-charge ratio. However, this efficiency gain is driven primarily from increased "charges" rather than decreased "costs", suggesting patient overtreatment. An implication of these findings is that forced audits provide a societal benefit by curbing excessive treatment despite reducing reported efficiency. Cross-sectional analyses further reveal that overtreatment post the audit exemption is more pronounced in hospitals with weaker internal controls, greater moral hazard, and a higher tendency toward defensive medicine. Our research highlights the externalities of the Single Audit Act, particularly in the healthcare sector.
Economics
Gaining Control by Losing Control
Jaesoo Kim, Alexander Rodivilov, and Dongsoo Shin
We study an agency model with two sequential tasks: an initial task (e.g., capital-raising), which requires unverifiable effort, and a final task (e.g., implementation), whose environment is privately known only to the agent. The principal can either perform the initial task herself (hands-on management) or delegate it to the agent (hands-off management). We show that when the cost of effort for the initial task is low, hands-on management is optimal, as it mitigates the risk of agent shirking. However, as the cost of effort increases, hands-off management becomes preferable. Under hands-on management, the principal's inability to commit leads her to overinvest in the initial task, which in turn heightens inefficiencies in the final task due to the agent's information rent. In contrast, delegation---despite agency costs---helps better target effort toward favorable project environments and reduces distortion in the output schedule. Our analysis also reveals how effort costs can create systematic biases toward one management style over the other.
The Revenue and Distributional Impacts of Unemployment Insurance Reform: Evidence from California
Mark Duggan, Jonathan Gruber, and Audrey Guo
In the United States, unemployment insurance (UI) is funded through employer-side payroll taxes that are experience-rated based on previous UI claims. States differ significantly with respect to the financing of their programs, and a majority of state programs do not currently meet minimum UI trust fund solvency standards. A common culprit in the least solvent states is a very low tax base of earnings on which UI taxes are levied. We focus on California, the least solvent of the 50 state UI programs, with debt currently to the federal government of $21 billion, and which has the lowest base of taxable earnings at $7000 per year. We use matched employer-employee administrative data to estimate the impact of financing reforms to California’s UI system. We find that raising the taxable earnings base would replenish the state’s UI trust fund and would increase experience rating by reducing the number of systematically subsidized firms. While this improves vertical equity of the UI system, it would also worsen horizontal equity by imposing much larger percentage increases in tax costs on the firms with the fewest layoffs. Alternatively, matching the higher tax base with both higher maximum and lower minimum rates could improve both experience rating and horizontal equity.
The Impact of Preference Programs in Public Procurement: Evidence from Veteran Set-Asides
Audrey Guo and Rodrigo Carril
Veteran-owned businesses are given preferential treatment in the allocation of procurement contracts from the U.S. Department of Veterans Affairs – currently the largest civilian federal agency in terms of procurement spending. We exploit a 2016 Supreme Court ruling that significantly increased the scope of these set-asides, to study the impacts of preference programs on both the targeted businesses and procurement outcomes. The policy change increased the share of contracts awarded to the target population, service-disabled veteran-owned small businesses, and benefited not only preexisting vendors but also new entrants, including those who had previously failed to win contracts. We find no evidence of spillovers to awards by other federal agencies, no decline in competition for awards, and no deterioration of contract execution performance by vendors. These findings suggest that VA set-asides have successfully improved outcomes for the target population without imposing significant costs on the government.
Exchange Rate Narratives
Vito Cormun and Kim Ristolainen
We combine Wall Street Journal news, topic modeling, and generative AI to estimate media narratives associated with the dollar exchange rate. Our findings shed light on the connection between economic fundamentals and the exchange rate, as well as on its absence. We uncover six distinct, largely non-overlapping narratives associated with exchange rate fluctuations since the late 1970s. Narratives centered on U.S. fiscal and monetary policy dominate early in the sample, while narratives associated with geopolitical tensions become more prominent in later years. News on technological change is important for the exchange rate throughout. When used as proxies for investor attention, these narratives significantly improve the explanatory power of fundamentals in exchange rate regressions.
Currency Wars and Trade
Kris James Mitchener and Kirsten Wandschneider
The Great Depression is the canonical case of a widespread currency war, with more than 70 countries devaluing their currencies relative to gold between 1929 and 1936. What were the currency war’s effects on trade flows? We use newly-compiled, high-frequency bilateral trade data and gravity models that account for when and whether trade partners had devalued to identify the effects of the currency war on global trade. Our empirical estimates show that a country’s trade was reduced by more than 21% following devaluation. This negative and statistically significant decline in trade suggests that the currency war destroyed the trade-enhancing benefits of the global monetary standard, ending regime coordination and increasing trade costs.
How do Financial Crises Redistribute Risk?
Kris James Mitchener and Angela Vossmeyer
We examine how financial crises redistribute risk, employing novel empirical methods and micro data from the largest financial crisis of the 20th century – the Great Depression. Using balance-sheet and systemic risk measures at the bank level, we build an econometric model with incidental truncation that jointly considers bank survival, the type of bank closure (consolidations, absorption, and failures), and changes to bank risk. Despite roughly 9,000 bank closures, risk did not leave the financial system; instead, it increased. We show that risk was redistributed to banks that were healthier prior to the financial crisis. A key mechanism driving the redistribution of risk was bank acquisition. Each acquisition increases the balance-sheet and systemic risk of the acquiring bank by 25%. Our findings suggest that financial crises do not quickly purge risk from the system, and that merger policies commonly used to deal with troubled financial institutions during crises have important implications for systemic risk.
Domino Secessions: Evidence from the U.S.
Jean Lacroix, Kris James Mitchener, and Kim Oosterlinck
We analyze how secession movements unfold and the interdependence of regions’ decisions to secede. We first model and then empirically examine how secessions can occur sequentially because the costs of secession decrease with the number of seceders and because regions update their decisions based on whether other regions decide to secede. We verify the existence of these “domino secessions” using the canonical case of the secession of southern U.S. states in the 1860s. We establish that financial markets priced in the costs of secession to geographically- specific assets (state bonds) after Lincoln’s election in the fall of 1860 – long before war broke out. We then show that state bond yields reflect the decreasing costs of secession in two ways. First, as the number of states seceding increased, yields on the bonds of states that had already seceded fell. Second, seceding states with more heterogeneous voters had higher risk premia, reflecting investors beliefs that further sub-secession was more likely in these locations.
Deep Roots: On the Persistence of American Populism
Ze Han, Helen V. Milner, and Kris James Mitchener
Is American populism a persistent political phenomena? Using a new dataset linking county vote shares in the 1890s with recent periods, we show that populist movements in the United States have deep roots. Counties where voters were enthusiastic about populist parties in the late nineteenth century had higher vote shares for Donald Trump in the 2016 and 2020 presidential elections. Exposure to globalization and the intergenerational transmission of political beliefs seem to be mechanisms behind this. Our instrumental variable results imply that globalization fostered populism in the 1890s which in turn laid the ground for populism today. Using individual policy preferences, we show that counties with more individuals holding populist attitudes today are associated with counties voting more populist in the 1890s. Moments of rapid economic change, such as those engendered by globalization, may propel the resurgence of such attitudes, which can then be popularized by charismatic leaders.
Connected Lending of Last Resort
Kris James Mitchener and Eric Monnet
Because of secrecy, little is known about the political economy of central bank lending. Utilizing a novel, hand-collected historical daily dataset on loans to commercial banks, we analyze how personal connections matter for lending of last resort, highlighting the importance of governance for this core function of central banks. We show that, when faced with a banking panic in November 1930, the Banque de France (BdF) lent selectively rather than broadly, providing substantially more liquidity to connected banks – those whose board members were BdF shareholders. The BdF’s selective lending policy failed to internalize a negative externality – that lending would be insufficient to arrest the panic and that distress via contagion would spillover to connected banks. Connected lending of last resort fueled the worst banking crisis in French history, caused an unprecedented government bailout of the central bank, and resulted in loss of shareholder control over the central bank.
Do Pandemics Change Healthcare? Evidence from the Great Influenza
Rui Esteves, Kris James Mitchener, Peter Nencka, and Melissa A. Thomasson
Using newly digitized U.S. city-level data on hospitals, we explore how pandemics alter preferences for healthcare. We find that cities in the top half of the mortality distribution during the Great Influenza of 1918-1919 subsequently increased hospital capacity by 8-10 percent more than cities with lower levels of mortality. This effect, driven by growth in non-governmental hospitals, persisted until 1960. Growth responded most in richer cities, exacerbating inequalities in access to healthcare. Other types of city- level healthcare spending did not respond to pandemic intensity, suggesting that large health shocks may not lead to increased public provision of health services.
A Trilemma for Asset Demand Estimation
William Fuchs, Satoshi Fukuda, and Daniel Neuhann
This paper establishes theoretical limits to identifying asset demand from observational data. We show a trilemma: one cannot simultaneously maintain that (i) prices respect no-arbitrage, (ii) investors care about asset payoffs, and (iii) asset-level demand elasticities can be recovered from supply shocks to individual assets. The trilemma can be resolved only if the econometrician observes at least as many independent quasi-experiments as the dimensionality of the asset span, or else relies on theoretical assumptions that cannot be tested with data. These results provide clear guidance for credible empirical design in asset demand estimation.
Beyond the Fundamentals: How Media-Driven Narratives Influence Cross-Border Capital Flows
Isha Agarwal, Wentong Chen, and Eswar S. Prasad
We provide the first empirical evidence on how media-driven narratives influence cross-border institutional investment flows. Applying natural language processing techniques to one-and-a-half million newspaper articles, we document substantial cross-country variation in sentiment and risk indices constructed from domestic media narratives about China in 15 countries. These narratives significantly affect portfolio flows, even after controlling for macroeconomic and financial fundamentals. This impact is smaller for investors with greater familiarity or private information about China and larger during periods of heightened uncertainty. Political and environmental narratives are as influential as economic narratives. Investors react more sharply to negative narratives than positive ones.
Monetary Policy Transmission in Euroized Countries: Evidence from Emerging Europe
Wentong Chen, Fazurin Jamaludin, Florian Misch, Alex Pienkowski, Mengxue Wang, and Zeju Zhu
This paper studies domestic monetary policy transmission in European countries with a significant share of lending and deposits in foreign currency, referred to as ‘euroized economies’. We find that the impact of domestic monetary policy shocks on both inflation and GDP diminishes with the degree of euroization across countries: the effects are twice as high in non-euroized countries compared to countries in our sample with the highest level of euroization. We further examine the exchange rate, credit and interest rate transmission channels, which are typically less effective in euroized economies. We show that domestic monetary policy has at best limited effects on the exchange rate. In addition, during the post-pandemic monetary tightening episodes, an increase in foreign-currency loans often softened the decline in overall credit growth, and rates of foreign-currency loans have followed the ECB policy rate rather than the domestic ones. By contrast, our analysis suggests that the pass-through to interest rates of domestic currency loans is similar across countries with different levels of euroization.
Foreign Exchange Risk Management Spillovers Across the Production Network
Wentong Chen
This paper uncovers the substantial foreign exchange risks faced by U.S. firms, despite most international trade being invoiced in U.S. dollars. These risks arise due to spillovers through the production network and fluctuations in foreign demand when exchange rates change. Using new hand-collected data from firms’ annual reports, I document that U.S. firms actively hedge foreign exchange risks using financial derivatives. I develop the first model of hedging in a production network and show that hedging by upstream or downstream firms can stabilize a firm’s performance due to shared risk exposures. This positive spillover effect operates through firms’ financial constraints: hedging stabilizes firms’ borrowing costs and the prices they charge connected firms. Exploiting two major USD–Euro exchange rate swings, I find that hedging by connected firms is as effective as a firm’s own hedging in stabilizing performance. Additionally, firms at the extremities of the production and trade network are more likely to hedge. Calibrating the model to U.S. data, I show that these spillover effects boost aggregate output and reduce prices.
Finance
A Meta Reinforcement Learning Approach to Goals-Based Wealth Management
Sanjiv Das, Harshad Khadilkar, Sukrit Mittal, Daniel Ostrov, Deep Srivastav, and Hungjen Wang
Applying concepts related to pre-training foundation models from the realm of generative AI, we develop a meta reinforcement learning approach (denoted MetaRL) that is pre-trained on thousands of goals-based wealth management (GBWM) scenarios. This is analogous to pre-training LLMs on large training datasets. The MetaRL model enables producing near-optimal dynamic decisions for (i) goaltaking and (ii) investment portfolio selection within a few hundredths of a second in inference mode on new investor problems, by not requiring separate training and optimization for each new investor scenario. Using inference with MetaRL delivers expected utilities that are, on average, 97.8% of the optimal expected utilities (determined via dynamic programming). Further, the MetaRL approach can enable solving problems with larger state spaces where dynamic programming becomes computationally infeasible.
Yield Curve Forecasting using Machine Learning and Econometrics: A Comparative Analysis
Sanjiv Das
While machine learning has revolutionized many fields such as natural language processing (NLP) and computer vision, its impact on time-series forecasting is still widely disputed, especially in the finance domain. This paper compares forecasting performance on U.S. Treasury yield curve data across econometrics/time-series analysis, classical machine learning, and deep learning methods, using daily data over 47 years. The Treasury yield curve is important because it is widely used by every participant in the bond markets, which are larger than equity markets. We examine a variety of methods that have not been tested on yield curve forecasting, especially deep learning algorithms. The algorithms include the Autoregressive Integrated Moving Average (ARIMA) model and its extensions, naive benchmarks, ensemble methods, Recurrent Neural Networks (RNNs), and multiple transformers built for forecasting. ARIMA and naive econometric models outperform other models overall, except in one time block. Of the machine learning methods, TimeGPT, LGBM and RNNs perform the best. Furthermore, the paper explores whether stationary or nonstationary data are more appropriate as input to deep learning models.
Augmenting the Funded Ratio: New Metrics for Liability Based Plans
Sanjiv Das, Daniel Ostrov, Deep Srivastav, Anand Radhakrishnan, and Wylie Tollette
The primary metric for the health of a liability based plan (LBP) is the ratio of the LBP's current assets to its present-valued liabilities. This “funded ratio” cannot address some important financial factors, so we suggest three additional metrics of financial health, connected to the probability of fulfilling the plan's liabilities. The first two metrics compare the current assets and projected future contributions to those needed to attain either (1) a specified probability for meeting all the liabilities (SAM, the solvency assets multiple) or (2) specified probabilities for meeting each liability (FAM, the funded assets multiple). The third metric, the risk-free funded ratio (RFFR), uses the STRIPS curve to determine the fraction of the liabilities that can be covered without risk. We implement these metrics, first using Monte Carlo simulation given a fixed investment portfolio strategy, and then using dynamic programming to optimize investment portfolio strategies that maximize SAM and FAM. ARIMA and naive econometric models outperform other models overall, except in one time block. Of the machine learning methods, TimeGPT, LGBM and RNNs perform the best. Furthermore, the paper explores whether stationary or nonstationary data are more appropriate as input to deep learning models.
Does Competitive Pressure Curb Corporate Carbon Emissions?
Ye Cai
We examine whether and how product market competition influences corporate carbon emissions. Using firm-level emissions data from 2002 to 2021 and a forward-looking measure of competition based on product market fluidity, we find that greater competitive pressure is associated with lower Scopes 1, 2, and 3 emissions. The effect is stronger among firms with higher R&D intensity or green patenting activities, and in settings with heightened reputational concerns—such as after the Paris Agreement or in consumer-facing industries. Instrumental variable and difference-in-differences analyses support a causal interpretation. We also find that while higher Scope 1 emissions are generally associated with lower firm value, this negative valuation effect is significantly weaker in more competitive markets. Our findings highlight a previously underexplored role of competition in driving environmental performance and contribute to the literature on climate finance and the real effects of product market rivalry.
Changing Utility Functions and Two-System Economic Models
Hersh Shefrin
This chapter articulates the relative merits of transitioning from a one-system neoclassical based economics framework to a two-system TF&S-framework, where TF&S stands for “thinking, fast and slow.” The chapter focuses on three types of decision problems: (1) intertemporal substitution associated with present bias; (2) the impact of self-control challenges on coincident substitution between different commodities; and (3) changing mental states in respect to substitution between alternative risky prospects. The empirical literature supports making the transition. (1) The neuroeconomics literature is consistent with the two-system planner-doer perspective. (2) The literature on financial literacy suggests that most people are incapable of solving complex dynamic lifetime optimization problems. (3) The literature on behavioral nudging supports the idea that people can be induced to increase savings. (4) The literature on household budgeting supports the notion that people budget by using matrix mental accounting. (5) The literature on windfalls is consistent with two-system theory. (6) The literature on cash dividends suggests that households use mental accounting to separate the value of their stocks into dividend income and capital. (7) The literature on households headed by people in late middle age finds evidence of behavioral differential MPC mental accounting and budgeting-matrix mental accounting. (8) The HANK macroeconomic literature supports the two-system differential MPC hypothesis. (9) The literature on wealth accumulation shows that retired individuals are miserly and typically oversave. (10) The literature on prospect theory provides evidence that investors view risk-seeing in the domain of losses as a self-control problem, and seek commitment devices that will help them with self-control.
Race to Rebalance: Global Supply Chains and Labor Market Regulation
Xiaoxue Hu, Dongxu Li, Rose C Liao, Angie Low, and Carrie Pan
Exploiting 18 million freight shipment records, we show that U.S. firms increase import 6% more from countries that strengthen labor laws through mandatory workforce policy (MWP) regulations, especially from new suppliers. While firms importing from MWP countries experienced increasing operating costs, they also receive favorable employment condition evaluations and have fewer negative social issues reported. Our results are consistent with firms responding to employee preferences: the import increase from MWP countries is more pronounced when majority of firms’ employees are in Democratic states, and in industries with high employee bargaining power. Our results are robust to alternative channels and have important policy implications.
Subscription Lines in Private Equity: General Equilibrium and Pricing
Justin Bocks and Gustavo Schwenkler
We solve the general equilibrium of the market for subscription lines in private equity. If an investment opportunity arrives early, the fund can borrow from a costly subscription line rather than call capital. The fund often does so as capital calls are risky and distort the fund’s performance. In equilibrium, the fund has implicit market power because it can choose not to invest. This yields an inverted relationship between issuer risk and credit spread in the subscription line market. The availability of subscription lines boosts fund and bank returns, reduces funds’ exposure to liquidity risk, and increases their investment probability.
Aggregate Confusion in Crypto Market Data
Gustavo Schwenkler, Aakash Shah, and Darren Yang
The quality of cryptocurrency market data is critical for academic research and financial applications, yet the topic remains understudied. We analyze data from leading vendors and document pervasive mislabeling, measurement errors, and discrepancies in reported market metrics. To address these issues, we propose a novel aggregation methodology that achieves asymptotic accuracy by identifying unreliable data instances. We also introduce a data quality grading system, offering practical guidance for data consumers. Our findings underscore the risks of relying on a single provider. They highlight a possible need for regulation in the market for crypto data.
The Different Networks of Firms Implied By The News
Victor Hilt and Gustavo Schwenkler
The interconnectedness of firms through various networks, such as production, credit, and competition, plays a critical role in determining firm-level and aggregate outcomes. However, data on these connections are often limited. This paper introduces a novel artificial intelligence methodology that extracts explicit firm relationship networks from financial news articles, providing comprehensive and interpretable data across multiple dimensions. Applying this methodology to New York Times articles since 1981, we generate extensive networks that predict key macroeconomic indicators. Our publicly accessible dataset offers valuable insights for future research on firm networks and aggregate fluctuations.
Sharing Economy and Entrepreneurship
Yifei Mao, Xuan Tian, Jiajie Xu, and Kailei Ye
Our paper shows that Airbnb—a sharing economy pioneer—has led to the creation of new, local businesses, which have higher survival rates and better performance both at the creation and in the long-run. Airbnb appears to spur new business creations by increasing passive income for landlords and by boosting tourists spending in the local region, leading to higher local income and the creation of more jobs. Our empirical strategy exploits the staggered arrival of Airbnb across counties, an instrumental variable approach for Airbnb penetration, and a case study using the short-term rental restriction in New York.
Attentive Market Timing
Hong Xiang, Yifei Mao, and Mengdi Zhang
This paper provides evidence that some seasoned equity offerings are motivated by public information. We test this channel in the supply chain setting, where supplier managers are more attentive than outside investors to customer news. We find that supplier firms are more likely to issue seasoned equity when their customer firms have negative earnings surprises. The results are mitigated when there is common scrutiny on the customer-supplier firm pairs by outside investors and analysts. Furthermore, long-run stock market performance appears to be worse for firms that issue seasoned equity following the negative earnings surprise of their customer firms.
Less is more: Institutional investors and Corporate Venture Capital
Dongxu Li, Yifei Mao, Xuan Tian, and Kailei Ye
In this paper, we show that a larger passive institutional ownership helps mitigate managerial entrenchment problems via the lens of corporate venture capital (CVC) investment. Specifically, we find that a plausibly exogenous increase in firms’ passive institutional ownership leads firms to cut their CVC investments in non-core businesses, of higher risk, and low quality. The effect is more pronounced for firms with more severe governance problems. As a result, the cut on CVC investment is associated with higher short-term announcement returns and better long-term operating performance and innovation outcomes. This paper offers novel evidence on a previously under-explored effect of stock market disciplines on firms’ CVC investment.
A Century of Inflation Narratives
Mourad Heddaya, Chenhao Tan, Rob Voigt, Qingcheng Zeng, and Alexander K. Zentefis
We study the evolution of U.S. inflation narratives in American newspapers since 1923. An inflation narrative is an explanation of the causes and/or effects of inflation. Using natural language processing to analyze 4.2 million sentences, we document significant shifts in narrative prevalence across economic eras. We find sharp regional differences as well, with Northern papers emphasizing fiscal causes while Southern papers focusing on supply factors and interest rate effects. Narrative framing also predicts heterogeneity in both short- and long-term consumer inflation expectations across income and education groups, with lower-income households showing greater sensitivity to narratives about the social/political consequences and cost-of-living effects of inflation. These narrative effects in some cases exceed the predictive power of realized inflation itself, suggesting exposure to different media framing may contribute to persistent gaps in inflation expectations across households.
Explaining Models
Kai Hao Yang, Nathan Yoder, and Alexander K. Zentefis
We consider the problem of explaining models to a decision maker (DM) whose payoff depends on a state of the world described by inputs and outputs. A true model specifies the relationship between these inputs and outputs, but is not intelligible to the DM. Instead, the true model must be explained via a simpler model from a finite dimensional set. If the DM maximizes their average payoff, then an explanation using ordinary least squares is as good as understanding the true model itself. However, if the DM maximizes their worst-case payoff, then any explanation is no better than no explanation at all. We discuss how these results apply to policy evaluation and explainable AI.
Information Systems & Analytics
Enhancing Mental Health Service Accessibility on University Campuses: The Impact of Tragic Events
Amber Xiaoyan Liu
The mental health of students has become one of the top concerns in higher education. In contrast to the growing demand for mental health services among students, the prioritization of these services in university budget allocations has lagged significantly behind. Namely, we consider how universities can provide access during tragic events such as student deaths. In such situations, demand for care may significantly increase, and universities need to be able to respond correspondingly to uphold their commitment to students' mental well-being. Leveraging a comprehensive dataset from the Counseling and Psychological Service (CAPS) center of a US university, we address this issue by first rigorously estimating the impact of a tragic event on the demand for mental health services. We find that demand increases significantly, with individuals being 131% more likely to make an appointment on a given day in the immediate days after the tragic event. We further show that among the impacted individuals, those who belong to the proximate cohort of the victims are even more likely to seek mental health services. Second, we propose a multiperiod optimization model that carefully balances the trade-off between addressing immediate demand and anticipating future demand, guiding budget allocation decisions in a rolling-horizon approach over a specified time frame. According to our multiperiod optimization model, properly accounting for this surge in demand can yield substantial gains compared to less sophisticated benchmarks, such as a greedy approach that depletes the remaining budget immediately following a tragic event without accounting for future needs, or a fixed reservation approach that accounts for future needs partially.
How Training Sentiment Shapes GenAI Delegation and Task Performance
Amber Xiaoyan Liu
This study investigates how the sentiment framing of GenAI training materials—positive, neutral, or negative—shapes users’ delegation behaviors and task performance. Drawing on the IS delegation framework, we conceptualize delegation as a process involving appraisal, distribution, and coordination. Through a randomized experiment with 270 university students, we find that positive and neutral training sentiments significantly enhance delegation appraisal, which in turn predicts GenAI adoption and improves task performance. Coordination is modestly influenced by sentiment among adopters, while distribution shows no significant effect. However, neither coordination nor distribution significantly impacts task outcomes. Practically, our findings highlight that sentiment-framed training serves as a meaningful intervention to foster more effective human–AI collaboration.
Voice Your Preferences: The Dual Impact of Personal Voices on Personalized Recommendation
Yaqiong Wang
As recommender systems evolve beyond traditional behavioral data to enhance personalization, understanding the broader impacts of using sensitive personal information becomes critical. This study examines the efficacy and implications of using personal voices in recommender systems. Through two controlled experiments, we first establish the significant correlation between individual vocal features and personal preferences. This correlation is mediated by individual demographic, emotional, and personality traits. These findings lay a theoretical foundation of recommender system design with personal voice beyond traditional behavioral data. Following the discovery, we investigate the impact of personalization strategy disclosure involving different types of personal data. Our results indicate that revealing the use of personal data—particularly voice—significantly reduces users’ perceived recommendation quality and willingness-to-pay. This behavioral backlash highlights a growing tension between personalization and privacy, especially when biometric data is involved. Our findings offer empirical insights for system design and data regulation, emphasizing the significance of transparency strategies in the deployment of recommender systems.
Recommending Composite Items Using Multi-Level Preference Information: A Joint Interaction Modeling Approach
Yaqiong Wang
Recommender systems have been increasingly used across a vast variety of platforms to efficiently and effectively match users with items. As application contexts become more diverse and complex, there is a growing need for more sophisticated recommendation techniques. One example is the composite item (e.g., fashion outfit) recommendation where multiple levels of user preference information might be available and relevant. In this study, we propose JIMA, a joint interaction modeling approach that uses a single model to take advantage of all data from different levels of granularity and incorporate interactions to learn the complex relationships among lower-order (atomic item) and higher-order (composite item) user preferences as well as domain expertise (e.g., on the stylistic fit). Comprehensive experimental results compare the proposed approach with advanced baselines in both offline and online settings to demonstrate its superior performance in composite item recommendation.
The Role of Physical Stores in the Digital Age: Assessing the Product Showcasing Effect and its Mechanisms
Yaqiong Wang
With increasing e-commerce penetration, it is believed that consumers are spending more shopping time online. This raises the question of the relevance of offline stores to retailers in a highly digitized landscape. Furthermore, few studies were able to establish the impact of product showcasing in stores. In light of these challenges, our study assesses whether physical stores remain valuable to retailers and empirically validates the product showcasing effect and its mechanisms. The measurement of the physical stores' economic impacts have been characterized by the difficulty of attributing online sales to product showcasing in stores, as such information is typically not captured. In this study, we address this challenge using a quasi-experiment which takes place through a nationwide retailer expanding its physical presence during the study period. Using product level data under a “triple-differences” framework, we are able to directly capture if a showcased product in the store would increase subsequent online purchase of the same product from customers in the vicinity. We find that online sales for retailers increases as new offline stores open and product showcasing drives the effect. We also find that the product showcasing effect works through cognitive and affective components, which have nuanced impacts on online sales, depending on the product type, location, customer type under consideration. Our research shows that physical stores still play a useful role in the digital age, particularly through its ability to showcase products. Based on the heterogeneous results, retailers should allocate physical store space based on the proportion of inactive to active customers in a sales tract along with the demand forecast of low- and high-involvement products. Retailers could maximize the value of their shelf space by prioritizing physical store space to products with multiple variants to increase the online purchase of a similar product. Finally, retailers should consider their strategic motivation for expanding their physical presence as product showcasing in stores, as that would have implications to where they should launch new stores.
Personal Preference versus Domain Expertise in Composite-Item Recommender Systems
Yaqiong Wang
Recommender systems are integral to online platforms, providing tailored recommendations to users leveraging personal preference, crowd sentiment, or expert opinions. Majority of research and applications in this field has been focusing on recommending single items and the problem of recommending composite-items (i.e., set of items) has been comparatively under explored. In this study, we provide a systematic view of composite-item recommender systems design involving two mechanisms, i.e., personal preference learning algorithms and domain expertise on the compatibility among items in a set. By exploring multiple designs through online experiments, we found the significance of personal preference-driven algorithms and professional knowledge in a complex recommendation context involving composite-items. Experimental results also demonstrated that composite-item recommendation performance can be further improved by integrating the two paradigms. Our findings provide actionable insights on designing composite-item recommender systems.
Do Inspection Delays Lead to Quality Decline? Evidence from U.S. Nursing Homes
Wilson Lin, Lauren Xiaoyuan Lu, and Susan F. Lu
This study examines the impact of inspection delays on nursing home quality using an 8-year panel dataset from 2016 to 2024. Leveraging variation in interinspection times and employing an instrumental variable approach to address endogeneity, we find that longer delays significantly increase health deficiency citations-quality indicators observable only through in-person inspections-while having little to no effect on nurse staffing levels, which are continuously monitored through the payroll-based staffing reporting system. These findings reveal a "teaching-to-the-test" strategy, in which nursing homes prioritize performance on metrics subject to routine monitoring, while neglecting less visible dimensions of care during periods of reduced regulatory oversight. Heterogeneous analysis shows that for-profit facilities are more vulnerable to quality deterioration induced by inspection delays, and that deficiencies are particularly concentrated in domains of resident care and quality of life. The adverse effects are most pronounced when inspection delays exceed 24 months. These insights underscore the importance of timely inspections and point to the need for policies that prioritize oversight of high-risk facilities and essential care areas under resource constraints.
Revisiting the Cause of Algorithm Aversion: Algorithm Feedback Asymmetry in the Field and Lab
Wilson Lin, Song-Hee Kim, and Jordan Tong
Algorithm aversion captures the idea that people may avoid algorithms even when the algorithm generally performs better than their own judgment. Dietvorst et al. (2015) argue and find experimental evidence that a key cause of algorithm aversion is algorithm feedback asymmetry: people lose confidence in algorithms after seeing it err more quickly than they lose confidence in themselves after seeing themselves err. We interpret and test this hypothesis using field data from 221,000 insulin dosing decisions and find the opposite -- patients decrease algorithm use less after an algorithm mistake than they increase it after their own mistake. A controlled laboratory experiment reconciles this contradiction, showing two distinct forms of algorithm feedback asymmetry: one favoring humans in one-shot decisions (as in Dietvorst et al. 2015) and another favoring algorithms in repeated settings (as in our field data). Our findings suggest that performance feedback does not fundamentally drive algorithm aversion and may, in repeated interactions, increase algorithm adoption.
Designing AI-Generated Summaries for Online Video Platforms: Evidence from a Field Experiment
Xiang (Shawn) Wan, Ying Ji, Chaoyue Gao, and Allen Li
As online video content grows exponentially, users need summaries to quickly assess relevance and decide what to watch. This research leverages large language models to automatically generate text summaries for online videos without human input. We evaluate two summarization strategies: information-extractive AI-generated summaries (IAIGS), which present fact-based recaps, and suspense-inducing AI-generated summaries (SAIGS), which withhold key details to spark curiosity. To assess their impact on video consumption, we conducted a field experiment on a video-sharing platform, focusing on two content genres: Science, which primarily addresses users’ instrumental information needs, and Humanities & History, which caters to users’ affective needs. We analyzed engagement metrics for 21,533 videos from 1,545 creators. Our results show that IAIGS reduced video views across genres, driven by information substitution. In contrast, SAIGS had genre-specific effects: they increased engagement with Humanities & History content by sparking curiosity, but decreased engagement with Science content by obstructing information seeking. An online experiment confirmed these patterns and shed light on the underlying mechanisms. Our study highlights the nuanced effects of AI-powered summarization strategies on user engagement across content genres, emphasizing the importance of aligning summary design with content characteristics. The findings provide valuable insights for designing AI-generated summaries to enhance user engagement.
How Do AI-Generated Summaries Affect User-Generated Content? Evidence from a Quasi-Experiment on a Restaurant Review Platform
Xiang (Shawn) Wan, Ying Ji, Chaoyue Gao, and Allen Li
Digital platforms are increasingly deploying AI-generated summaries (AIGS) to manage large volumes of user-generated content. While prior research highlights the benefits of AIGS in enhancing user experience by distilling extensive review content into concise insights, their impact on review contributions to user-generated content platforms remains unclear. On one hand, AIGS may facilitate user contributions by reducing the cognitive effort of writing a new review—the summary can serve as a cognitive scaffold that guides and simplifies the review-writing process. On the other hand, AIGS may discourage contributions by reducing individual recognition due to the prominence of AI-generated summaries. To examine this tension, we leverage a quasi-experiment in which a restaurant review platform introduced AIGS in some cities but not others. We find that the introduction of AIGS leads to a significant decline in review contributions. This effect is primarily driven by reduced perceived social recognition, as contributors feel their individual efforts are less visible and valued after the introduction of AIGS. The negative effect is more pronounced for more popular, longer-tenured restaurants, where new reviews face greater difficulty standing out. Further analysis reveals that the decline is concentrated in positive reviews, whereas neutral and negative reviews remain largely unchanged. Moreover, the decrease in review contributions is largely attributable to a drop in low-quality reviews, which are shorter, contain fewer images, or receive no helpful votes, suggesting that AIGS may help suppress less substantive content. Importantly, we find heterogeneous responses across user status: low-status users reduce their contributions in response to AIGS, whereas high-status users are less affected. These findings highlight the nuanced trade-offs of AI adoption in user-generated content and provide valuable implications for digital platforms to balance algorithmic efficiency with sustained user engagement.
Recommendation Effects in the Presence of Customer Learning and Privacy Concerns
Xiang (Shawn) Wan, Xiaojing Dong, Yuchi Zhang, and Xiaosong Dong
Product recommendation systems are widely used in mobile retailing. While these systems improve through gathering data from user interactions and delivering more precise suggestions, users themselves also learn through repeated app usage and become more self-reliant, potentially reducing their dependence on algorithmic suggestions. This observation raises an important yet underexplored question: how does the effectiveness of recommendations vary with user experience? We address this question using a quasi-experiment conducted on a major mobile shopping platform in Asia. Our findings reveal that the effectiveness of the recommendation engine declines as users gain more experience with the app. To explain this pattern, we propose and validate a dual learning process hypothesis, in which both the user and the platform learn concurrently but at different rates. We show that when user interactions involve product browsing or purchasing, users learn faster than the platform, leading to reduced dependence on recommendations. In contrast, when interactions are limited to session initiation, the platform retains a learning advantage. We also rule out privacy concerns as an alternative explanation. This study underscores the evolving nature of user-platform interactions and offers insights for developing adaptive recommendation strategies that align with users’ experience levels.
Optimal Mechanization Investments in Resource Constrained Farming
Maggie Zhang, and Jayashankar M. Swaminathan
Adopting mechanization in farming is an effective approach to increase agricultural productivity. However, in developing economies, majority of farmers face severe constraints in terms of land size and budget. In this paper, we analyze a farmer's investment decision for a single crop under limited land and budget. We model expected crop production using a Cobb-Douglas function of seed amount and machine capacity and explore the farmer's optimal procurement of these inputs to maximize her profit. For a monopolist farmer, we find that the optimal strategy follows a modified threshold-type policy, which is categorized into three scenarios (1) unconstrained scenario, (2) land-size-constrained scenario and (3) budget-constrained scenario. The relationship between cropped land and mechanization is determined by the policy scenario and budget structure. In the land-size-constrained scenario, land and machine capacity act as complements when the farmer's budget increases with her land size beyond a minimal rate but remains highly limited, while they are substitutes when her budget is moderately limited, independent of her land size or grows minimally with it. In the budget-constrained scenario, however, land and machine capacity serve as complements when the farmer's budget is increasing in her land size. We further explore how variations in seed efficacy and mechanization efficiency affect these decisions and find that, while improved seed efficacy generally incentivizes the planting of more seeds and increased usage of cropping area, and higher mechanization efficiency typically promotes greater mechanization levels, farmers may adopt a counterintuitive strategy when these factors are already high. Extending our analysis to a Cournot competition model, we show that while the threshold policy remains optimal, competition can lead to increased mechanization levels and reduced profits due to overproduction. Additionally, we demonstrate that cooperative formation can enhance profitability for farmers with extreme resource limitations. Our study offers guidance on optimal resource allocation under land and financial constraints as well as the impact of effectiveness of machinery and seeds. Our results offers policymakers actionable insights for designing appropriate subsidies that encourage mechanization under competitive and cooperative settings.
Management & Entrepreneurship
To Thine Own Craft Be True? Engaging with Paradox in Bean-to-bar Chocolate Entrepreneurship
Jo-Ellen Pozner
We investigate how US bean-to-bar craft chocolate entrepreneurs perceive and engage with craft entrepreneurship’s fundamental paradoxes, the conflicting yet connected concerns that stem from tensions between the craft ethos and entrepreneurial imperatives. Through a qualitative analysis of 79 interviews with founders and executives, three salient paradoxes emerge: identity, industrialization, and intention. We find that most entrepreneurs engage in paradoxical thinking, holding craft ideals and entrepreneurial imperatives in tension. Comparing the degree to which our subjects engage in paradoxical thinking across sets of tensions, we derive three paradox-engagement archetypes. Finally, we find that most craft entrepreneurs design their organizations to act as a guardrail, allowing them to engage with both craft and entrepreneurial logics by forestalling opportunities for external stakeholders to exert market-related pressures. Our findings show that entrepreneurs can successfully engage with craft entrepreneurial paradoxes, building sustainable ventures without abandoning their ideals.
Crafting Inclusion: How Innovation Makes Space for New Voices in Craft Entrepreneurship
Jo-Ellen Pozner
In mature consumer industries dominated by homogeneous incumbents, innovation is often driven by craftspeople founding small, specialist firms. While resource partitioning theory explains the emergence of anti-industrial, specialist producers – craft firms – because it operates at the field level, we lack an understanding of the relationship between demographic heterogeneity and specialist foundings. Likewise, research has not yet explored how demographic diversity influences the innovation that drives the founding of craft producers. Using grounded theory methodology to analyze interviews with entrepreneurs in the craft fields of bean-to-bar chocolate and distilled spirits, we explore how women and ethnic minorities navigate entrepreneurship and contribute to innovation, becoming a significant force behind specialist foundings. Our study contributes to resource partitioning theory by examining the human actors behind specialist entrepreneurship and offers insights for understanding diversity's role in driving innovation in mature industries. Moreover, our preliminary findings challenge the assumption that craft fields are inherently masculine and exclusionary, revealing that demographic diversity drives innovation through varied cultural perspectives and experiences.
Exploring the Social Side of Negative Social Evaluation: A Symbolic Interactionist Approach
Jo-Ellen Pozner
In this paper, we use a symbolic interactionist perspective to build a multi-level theory about how negative social evaluations are formed. We begin by describing why there might be inertia in social evaluations of organizations. We then explore how the introduction of negative and inconsonant facts for audiences implicating those organizations are integrated into the public discourse and interpreted by individuals embedded in social contexts. Finally, we describe the process by which collective sensemaking through interaction with other individuals in social groups, even loosely connected ones, leads to downward revisions to social evaluations of those actors.
Social issues in management: A field-level review and consolidation of a heterogenous domain
Andrew McBride
A diverse range of topics, from sustainability to corruption to social injustice, are studied under the banner of social issues in management. Prominent authors and outlets in the field have called for more research attention to these and other organizationally-relevant social issues. Despite the apparent support for studying social issues, the field of management lacks a coherent paradigm for doing so, and there is little indication of whether published research in the field heeds such calls. To provide clarity and direction to social issues research, we conduct a four-stage, field-level review that articulates how and why social issues are studied in management. We first provide a family resemblance conceptualization of social issues by drawing on published research. We then identify the extent to which management research focuses on social issues through a manual analysis of dependent variables (n = 471) and a machine-learning aided analysis of thousands of full-length academic articles (n = 12,573). We then identify the why and how of social issues research by identifying institutional logics that differ in terms of their assumptions, values, and goals: the instrumental, normative, and critical logics. Our review reveals a diverse body of research with no shared definition and a wide range of topics pursued under different institutional logics. We consolidate and organize this broad domain while preserving key differences. Our review highlights opportunities for cross-pollination among the three logics and proposes distinct avenues for future social issues research, depending on the type of social issue and researchers’ preferred logic.
Voluntary Commercial Decarbonization, Self-interest, and Ethics: a Constellation Approach
Ewan Kingston
What should the typical firm do about addressing climate change? Often this question is answered with reference to a North Star: firms should do "their fair share" of the work of decarbonization; or do "everything they can" to address climate change without making a loss. I argue that due to the complexity of the problem, and the adversarial nature of business, positing one moral duty for commercial decarbonization oversimplifies the moral landscape. Instead I propose a "constellation approach" which breaks down the different strategic arguments for taking action on climate and identifies moral elements that supplement or constrain those arguments.
Have You Stopped Pushing Tobacco Yet? A Normative Business Ethics Research Agenda for Responses to Industry-level Controversies
Ewan Kingston
Most discussions in mainstream normative business ethics assume that firms belong to industries that society welcomes or tolerates. However, some industries face industry-level controversies. In these controversies public actors call for an industry to be reduced in size, or even ended completely. The grounds for this can either because the product is allegedly intrinsically immoral, or it allegedly causes or risks harm significant enough to warrant a reduction in the industry’s size. As calls to reduce an industry broaden, the controversy can shift to the proposed regulatory measures. Industry-level controversies generate a number of unique questions about firms’ marketing and collaborative actions. They also make existing questions about firms’ political and research activity more salient. This paper develops the concept of industry-level controversy, and lays out a research agenda for future work in normative business ethics that explores which responses to industry-level controversy made by firms might be justifiable.
Big Buyer Responsibilities for Social Upgrading in Global Supply Chains Under Unbundling and Constrained Influence
Ewan Kingston
An assumption in much philosophical literature on sweatshop ethics is that the individual branded marketers that sell consumer goods either employ sweatshop workers or can strongly influence the conditions under which those workers labor. This over-simplified account of supply chains is a worrying trend in philosophical work that misidentifies the rationale for and details of the responsibilities of big buyers for the labor standards in their supply chains. Throughout this paper, we illustrate how philosophers’ “vertical integration” and “control” assumptions distort our understanding of the internal dynamics within supply chains. In this paper, we assume a more realistic relationship between big buyers, factories, and workers: what we call the constrained influence assumption. Under this assumption big buyers retain the responsibility to work toward social upgrading goals, however fulfilling such a responsibility requires big buyers to collectively cede power to third parties in supply chains in formalized and accountable ways. Recent developments in transnational industrial agreements, such as the International Accord, are promising examples of this accountable ceding of collective power by big buyers.
Channeling Unruly Benefits from Inadequate Regulation: An Alternative to Personal Boycotts in Consumer Ethics
Ewan Kingston
Given labor rights abuses and environmental degradation in the supply chains behind many of the products we are offered, an obvious approach is to conduct (and suggest others conduct) personal boycotts. By “personal boycott” I mean to attempt to select and avoid goods based on how those goods were produced. Since the personal boycott approach must function in marketplaces with myriad offerings, complex supply chains, and major conflicts of interest, it faces steep epistemic, motivational, and political challenges. In response, I propose a largely overlooked alternative to the personal boycott approach which I dub the “unruly benefits approach”. This approach argues that consumers should shift their attention from morally tainted production processes to their own unruly benefit. By “unruly benefit” I mean the extent to which consumers benefit from unreasonably low prices due to poorly regulated global production. The approach suggests consumers’ central pro tanto duty is to channel this “unruly benefit” appropriately. I explain how an unruly benefits approach to consumer ethics might work in practice, justify its component principles and the approach as a whole, and respond to two major objections – that it is unfair to the poor and it condones “buying indulgences” to do wrong.
Marketing
Supervised Mediation Analysis
Judy Zhang
Consumers often engage in multi-step, sequential reasoning when exposed to stimuli. The authors propose a novel approach to understand consumers’ thoughts in a serial mediation framework using both text from open-ended questions and fixed-point ratings data. Unlike parallel mediation, where processes unfold simultaneously, this model assumes treatment variables sequentially affect the topic composition of text responses, which are then related to the rating data and an outcome variable. The proposed model integrates supervision through priors and model structures to guide the identification of mediators and relationships, even in situations where scales are not well developed. Apart from the additional insights revealed from the text and ratings data, the proposed model predictively outperforms alternative models of mediation.
The Impact of Advertising Content on Customer Acquisition and Retention for Subscriptions of Physical Goods: Insights from a Field Experiment
Kirthi Kalyanam, Raphael Thomadsen, and Nan Zhao
We examine the role of advertising content in both the acquisition and the retention of customers in the context of selling subscriptions for physical goods. The dominant strategy currently employed to gain subscription sign-ups is to offer price discounts. The logic of this strategy is that all customers prefer to have lower prices. However, price discounts may not be the most effective motivation in either the short or long run, since the message does not speak to the innate benefits or the underlying motivation for repeat buying. We conduct a four-week email field experiment with a major retailer that sells pet products where we send messages that emphasize different innate benefits of a subscription in addition to price discounts. Causal estimates show that advertising content that emphasizes non-price benefits improves both signs ups and retention rates compared to content that only emphasizes price discounts, especially among customers who have prior experience with this retailer's eCommerce channels. A counterfactual shows that reassigning consumers from the price-only message to a reminder message for one month would add 968 new subscribers in a month and improve subscription revenue by $155k (an increase of 6.1%) over the period of analysis.
Clicks vs. Commitments: The Economic Trade-Offs of Polarizing News Content
Shunyao Yan and Klaus M. Miller
This study empirically investigates how polarizing content impacts user engagement and subscriptions on a major European news website. Using advances in natural language processing, we develop deep learning and large language model-based textual measures of political polarization, tailored to a multi-party political system. By combining the comprehensive supply and demand dataset of this country's leading news site, including user-level clickstream data and subscription records, we examine how consumers engage with polarizing articles. To establish causality, we use an instrumental variable approach. We leverage two theoretically distinct sources of exogenous variation: a Bartik-style instrument that interacts users' stable topic preferences with weekly shifts in content supply and a shock from a federal election for part of the news readers. Our findings reveal a crucial "polarization trap." We find strong causal evidence that an increase in consuming polarizing content boosts user engagement, as measured by time on site, but simultaneously causes a significant decrease in the probability of subscription. Our results suggest that the negative impact of polarization is driven more by its affective dimension than by its ideological dimension, and this effect is most prominent during major political events. Our finding provides a critical insight for publishers: the polarizing content that could boost short-term attention may be harmful to building a loyal, paying subscriber base.