学术论文

Many emerging economies employ preferential credit policies that target selected sectors. This paper quantifies the implications of such policies for aggregate productivity and welfare. Using Chinese firm-level data from 2009–2020, we first document that sectors with higher markups receive larger credit subsidies and exhibit higher revenue-based productivity. Motivated by these facts, we develop a multi-sector quantitative model with endogenously determined markups and calibrate it to match the distribution of sales both within and across sectors. We find that preferential credit subsidies raise aggregate productivity and welfare by reallocating market shares toward high-markup sectors. These gains persist in an extended framework with endogenous firm entry.

We examine whether and how common ownership affects Environmental, Social, and Governance (ESG) ratings—an important research question given the increasing use of these ratings in investment decisions and corporate evaluations. We find that companies with major shareholders in common with the rating agency (“sister firms”) tend to receive higher ESG ratings. When a company becomes a sister firm through a change in the rating agency's ownership structure, its rating from that agency is subsequently upgraded, whereas its ESG ratings from other agencies remain unchanged. Sister firms exhibit greater rating disagreements across agencies than other firms. The higher ESG ratings for sister firms are partly attributable to the transfer of immaterial positive ESG information through common owners. The common ownership effect is more pronounced when the owner can exert a greater influence on the rating agency. Moreover, sister firms with initially elevated ratings demonstrate poorer future ESG performance. Overall, our findings suggest that owners can affect ESG ratings of their portfolio companies in a way consistent with their influence and interest.

Using data on Internet news reading, we measure fund-level attention to both aggregate and firm-specific news and relate it to fund portfolio allocation decisions. In the time series, we find that funds shift attention toward macroeconomic news during periods of high aggregate volatility. Those funds that exhibit stronger attention-reallocation patterns earn higher future returns. In the cross-section of fund portfolios, fund attention is positively related to stock holdings. Furthermore, fund attention to a stock increases the value-add of that position to the fund's performance. This relationship is stronger using fund attention to more value-relevant news articles.

We investigate investors' voluntary disclosure decisions under uncertainty about their information endowment. In our model, an investor may receive initial evidence about a target firm. Conditional on learning the initial evidence, the investor may receive additional evidence that helps them interpret the initial evidence. The investor takes a position in the firm's stock, then voluntarily discloses some or all of their findings, and finally closes their position after the disclosure. We present two main findings. First, the investor will always disclose the initial evidence, even though the market is uncertain about whether the investor possesses such evidence. Second, the investor's disclosure strategy of the additional evidence increases stock price volatility: they disclose extreme news and withhold moderate news. Due to the withholding of the additional evidence, misleading disclosure arises as an equilibrium outcome, where the investor's report decreases (increases) price despite their news being good (bad). These results remain robust when considering the target firm's endogenous response to the investor's report.

In this paper, we propose a Deep-DiD method that incorporates two deep neural networks in a difference-in-difference (DiD) framework to estimate heterogeneous treatment effects (HTEs). The dual-network architecture contains one neural network modeling HTEs as a nonparametric function of pretreatment features and another neural network capturing individual and time fixed effects. Through a series of simulations, we show that our method can uncover the true HTEs with high accuracy under various settings and demonstrates more robust estimation performance compared with existing methods like linear models and random forests. We apply this method to an empirical setting where a large video-sharing platform introduced a “Creator Signing Program” aimed at signing creators and motivating them to generate more high-quality video content. Leveraging a matched data set of signed and unsigned creators, we employ our Deep-DiD method to estimate the HTEs of the signing program. Our method can help the platform optimize creator selection by identifying creators with the highest-estimated treatment effects. Through out-of-sample tests, we show that creators selected by the Deep-DiD method experience substantially larger actual performance jumps than those selected by the platform. Creator selection based on the Deep-DiD method also consistently outperforms that based on linear models.

Many important platforms, particularly in healthcare, hospitality, and content streaming, depend on a small number of strategic participants on the “seller” side. As a result, individual-level bargaining and participant-specific network effects are central to platform growth and profitability. This paper introduces a novel modeling framework that integrates bargaining outcomes with heterogeneous direct network and cross-network effects to capture platform evolution. We estimate participant-level direct network and cross-network effects based on the attributes of strategic sellers and assess how these shape bargaining outcomes. By modeling both time-varying, participant-specific network effects, which influence market growth, and their impact on bargaining outcomes, which affect profitability, our framework enables platforms to evaluate growth strategies. We apply the model to data from a major Chinese online healthcare platform connecting hospitals and consumers for health checkups. We find substantial heterogeneity in hospitals’ network effects, which drive variations in bargaining outcomes. Hospitals with stronger network effects negotiate lower commission rates, whereas the platform secures higher commission rates in markets where it holds a larger market share. Through policy simulations, we explore strategies including seeding, targeting sequence, and market entry. Our findings highlight key trade-offs between growing market size and maximizing profitability, offering insights for platforms built on negotiated partnerships.

The remarkable success in personalized recommendations on digital platforms has sparked interest in extending this advancement to physical spaces. In response, our study introduces a generalized recommendation problem, named point-of-interest (POI) recommendations in physical spaces with pedestrian movement (P3M). A critical yet under-investigated impediment in addressing P3M is exposure bias: When the exposure likelihood of items to users is unevenly distributed, indiscriminately treating all unobserved user–item interactions as negative feedback introduces bias to the learning of recommender systems. Unlike existing debiasing literature on digital platforms, we focus on the unique source of uneven exposure in physical spaces, arising from the dynamic interaction between pedestrian movement and spatial layout. To address this issue, we propose a novel recommendation method, unbiased movement-aware pairwise ranking (UMPR), which considers dynamic pedestrian movement to achieve unbiased POI recommendations. Specifically, we formulate an unbiased pairwise learning framework, propose a movement-aware recommendation model, and devise an alternating learning algorithm to optimize model parameters. Using real-world mall data, we demonstrate that our method outperforms state-of-the-art benchmarks in delivering store recommendations for pedestrian shoppers. Further investigations confirm that the improved recommendation performance translates into added monetary value while maintaining humanistic fairness across customers and store tenants. Overall, this study underscores the significance of addressing exposure bias through adequate spatial movement modeling, paving the way for effective recommendations in the physical landscape.

This study examines the impact of discrete emotional expression (i.e., expression of anxiety, sadness, anger, disgust, love, joy, surprise, and anticipation) on the differential diffusion of online content in social media networks. We conducted an analysis on a random sample of 387,486 online articles and their corresponding diffusion cascades, involving more than six million unique individuals, on a major online social networking platform. Our investigation focused on the relationships between discrete emotional expression and the diffusion of online articles, specifically the structural properties of diffusion cascades, such as size, depth, maximum breadth, and structural virality. We employed various econometric model specifications, and our results robustly demonstrate that articles expressing higher levels of anxiety, love, and surprise reach a larger number of individuals and diffuse more deeply, broadly, and virally. In contrast, expression of anger, sadness, and joy exhibit the opposite effect. Additionally, we find that articles with different emotional expression tend to spread differently based on individual characteristics and social ties. Our findings offer valuable insights into the diffusion and regulation of online content from the perspectives of emotional expression and social networks.

Latent variable models are popularly used to measure latent embedding factors from large-scale assessment data. Beyond understanding these latent factors, the covariate effect on responses controlling for latent factors is also of great scientific interest and has wide applications, such as evaluating the fairness of educational testing, where the covariate effect reflects whether a test question is biased toward certain individual characteristics (e.g., gender and race), taking into account their latent abilities. However, the large sample sizes and high-dimensional responses pose challenges to developing efficient methods and drawing valid inferences. Moreover, to accommodate the commonly encountered discrete responses, generalized latent factor models are often assumed, adding further complexity. To address these challenges, we consider a covariate-adjusted generalized factor model and develop novel and interpretable conditions to address the identifiability issue. Based on the identifiability conditions, we propose a joint maximum likelihood estimation method and establish estimation consistency and asymptotic normality results for the covariate effects. Furthermore, we derive estimation and inference results for latent factors and the factor loadings. We illustrate the finite sample performance of the proposed method through extensive numerical studies and an educational assessment dataset from the Programme for International Student Assessment (PISA).