This project considers the estimation of nonparametric panel regression models with a latent group structure. It will propose a novel approach which combines a feed-forward neural network machine learning method with the hierarchical agglomerative clustering to estimate both the group structure and the group-specific regression functions. Asymptotic properties of the developed estimators will be established to provide theoretical justification of their soundness. An extensive simulation study will be carried out to investigate the finite-sample performance of the proposed estimation method. An empirical application to stock returns data will be conducted to demonstrate the usefulness of the proposed model and method.

Integrating data from multiple sources is increasingly vital in statistical and business realms. Challenges emerge when these sources possess distinct covariate sets despite sharing latent structures. Three key issues drive our investigation: First, integrative analysis encounters missing covariates, where sources only partially observe variables. Second, partial transfer learning involves sources with heterogeneous, partially transferable coefficients despite measuring the same variables. Third, generalized homogeneity pursuit tackles source-specific covariates while sharing coefficient structures. To surmount these obstacles, we propose a unified penalized likelihood framework leveraging common homogeneity in regression coefficients across diverse sources. Our approach includes an efficient coordinate descent algorithm with established oracle properties and asymptotic normality under standard conditions. Through extensive Monte Carlo simulations and real-business data examples, we aim to showcase the superior estimation and prediction performance of our methods over conventional single-source techniques and naive imputation strategies.

The proposed project studies on high-dimensional dynamic modeling of binary variables. Binary variables play an important role in describing social and economic phenomena. For example, they are used to describe pairwise relationships in social networks and international trade networks. In Finance, binary variables capture events such as stock jumps and co-jumps. These binary outcomes are often high-dimensional and dependent over time. A natural question is how can we model their dynamics and leverage such models for practical applications, such as prediction? This project will explore conditional autoregression type modeling for high-dimensional binary variables. We will develop statistical methodologies for efficient estimation and establish inference framework for the proposed approach. Empirical applications will explore the modeling and prediction based on international trades networks and global GDP data.

The association between log-price increments of exchange-traded equities, as measured by their spot correlation estimated from high-frequency data, exhibits a pronounced upward-sloping and almost piecewise linear relationship at the intraday horizon. There is notably lower—on average less positive—correlation in the morning than in the afternoon. We will develop a nonparametric testing procedure to detect such variation in a correlation process. In a preliminary application in the high frequency data, it suggests that diurnal variation in the correlation process is a nontrivial effect in practice. We will also consider how the conditioning information on macroeconomic news and corporate earnings announcements affect the intraday correlation curve.

The generalized method of moments (GMM) is a key framework for estimating structural economic models. This project develops GMM inference robust to misspecified moment conditions. We first establish an identification condition that does not require prior knowledge of moment validity. This allows consistent parameter estimation even with potentially invalid moments. Next, we employ a penalty-based approach—such as LASSO, ridge, or SCAD—to select valid moments from a candidate set. After consistent selection of moment conditions, we derive valid post-selection inference, ensuring reliable confidence intervals and hypothesis tests. Finally, we extend the method to high-dimensional settings where the number of moments may be large, making it applicable to modern data-rich research environments. Overall, this work provides a theoretically grounded and practical toolkit for robust GMM estimation under realistic empirical conditions.

This research will further examine how livestream developing as regulation mature with video of livestreams. In this research, we will apply advanced technology like video processing and NLP technique in studying the content of livestream to investigate how mutual funds livestream. There are two specific questions we would like to answer. One is whether the content of livestream differs as regulation completes. The other is whether the physical appearance difference of host between poster and actual livestream can influence the investors’ confidence on fund family.

A long-standing debate in finance is whether dividend policies influence stock valuations and returns, or whether dividends merely reflect firm characteristics already priced by the market. While early studies produced mixed evidence, recent research has shifted this debate by documenting predictable price pressures around dividend events, particularly in the U.S., attributing these to temporary spikes in investor demand for dividends (Hartzmark and Solomon, 2019, and 2025). What remains unclear, however, is whether dividends exert persistent effects on relative stock returns beyond temporary price pressure, and whether the timing of dividend events create coordinated, market-wide, or even cross-market movement in returns. If dividends are a salient style characteristic affecting investors’ portfolio choice rather than just firm policy, then we should expect not only a consistent dividend premium across global markets but also synchronized returns tied to clustering of predictable dividend events. We focus on addressing these questions.

We aim to provide new evidence on why the dividend premium exists in international sample by analyzing a previously overlooked mechanism: a dynamic, coordinated, and market-wide demand for dividend-paying stocks anchored around the predictable clustering of ex-dividend events. In addition, we study dividend payment-induced price pressure, shedding fresh insights on the demand-based asset pricing.

EDGAR (Electronic Data Gathering, Analysis, and Retrieval) was introduced by the SEC as an early regulatory technology (“RegTech”) reform designed to enhance capital-market transparency by lowering investors’ costs of accessing mandatory corporate disclosures. By improving the timeliness, accessibility, and verifiability of firm-level information, EDGAR reduced informational frictions and reshaped firms’ external information environments. Prior research documents that EDGAR’s staggered rollout lowered firms’ costs of equity and improved market quality through liquidity, risk-taking, and governance channels Lai, Lin, and Ma (2024), constrained strategic analyst forecasting behavior by making biased forecasts easier to detect ex post Chang, Ljungqvist, and Tseng (2023), and reduced investor disagreement around information events Chang, Ljungqvist, and Tseng (2022). Whether these informational benefits translate into realized primary-market financing outcomes—where firms raise new capital—remains less well understood.

This study examines seasoned equity offerings (SEOs), economically consequential financing events through which firms raise external equity capital. SEOs provide a particularly informative setting because issuance costs are directly observable in offering-related expenses, realized at the time of issuance, and borne by the issuing firm—thereby affecting capital structure choices, financing capacity, and access to external capital. Leveraging the SEC’s staggered EDGAR phase-in as a quasi-natural experiment, we test whether improved disclosure accessibility is associated with changes in firms’ realized SEO issuance costs. By linking a disclosure-access reform to tangible issuance frictions, the study clarifies how RegTech-driven transparency affects firms’ ability to raise external equity capital.

This project aims to develop and validate a real‑time dispersion measure—the inter‑decile return spread (IDR)—derived from cross‑sectional stock returns to detect retail‑driven speculative episodes. We will test the behavioral mechanism whereby elevated dispersion coincides with intensified retail trading shaped by salience, diagnostic expectations, and extrapolative beliefs.

Empirically, we will show that high‑dispersion states are characterized by momentum collapses and dominant short‑term reversals, and that conditioning momentum exposure on dispersion restores profitability in retail‑intensive markets (e.g., China). Building on these insights, we will design and backtest dynamic rotation rules between momentum and short‑term reversal portfolios guided by dispersion states, targeting materially higher Sharpe ratios than static allocations.

To establish external validity, we will extend the analysis internationally, using Google search trends as proxies for retail attention, and test whether dispersion robustly predicts momentum and reversal returns across developed and emerging markets. The project will deliver implementable state‑timing guidelines for portfolio managers and risk controllers, clarifying the behavioral channel through which retail‑driven speculation conditions momentum dynamics.

We investigate a manipulation pattern in concentrated liquidity market makers on BSC, focusing on pools tied to Binance Alpha token incentives. An attacker concentrates liquidity in a single tick and forms a one tick liquidity wall. With a small price move, the attacker keeps the price inside that tick range. The pool can then appear to have low slippage and impermanent loss that seems limited. Users swap within these pools to earn Alpha points and this increases volume. The stable price and high volume attract outside liquidity providers. After enough external liquidity enters, the attacker withdraws liquidity and sells a low value token. This drains the valuable token from the pool and shifts losses to other liquidity providers. We will build a scalable pipeline to detect and measure these attacks. We will design on chain features that separate attacks from normal market making. We will cluster related attacker addresses and quantify prevalence, impact and risk factors. We will release reproducible datasets and tools.

This paper investigates how electric vehicle (EV) charging demand responds to price adjustments and whether these responses vary with climate risk exposure. Utilizing comprehensive 15-minute EV charging data from public parking facilities in Macao SAR, our sample encompasses a citywide network of EV users, including electric bikes, and captures charging volumes for all types of chargers—slow, medium, and fast—thereby offering a holistic understanding of EV charging behavior. To ensure clear identification, we leverage a citywide charging price adjustment policy as a quasi-natural experiment. The empirical analysis takes advantage of the heterogeneity in pricing strategies across different charger types and time-of-day schedules to identify the causal effects of price changes on EV charging demand. Additionally, we examine whether price sensitivity differs between flood-prone (low-lying) and non-flood-prone areas, providing revealed-preference evidence of climate risk awareness in daily energy consumption behavior. Our findings provide new insights into the heterogeneous price elasticity of EV usage and emphasize the importance of considering spatial climate vulnerability when designing urban EV charging pricing policies.

The project examines the balance between tourism development, environmental sustainability, and resident well-being in the Guangdong–Hong Kong–Macau Greater Bay Area (GBA). Nocturnal lighting is central to tourism development, particularly in urban tourism hubs such as Macau, Hong Kong, and Guangzhou. However, it is also attributed to light pollution and rising carbon emissions, leading to rising concerns about environmental costs and the quality of life of residents. To address this tension, the project applies a Spatial Durbin Mediation Model (SDM) to assess how tourism development affects residents’ well-being, with light intensity serving as a key mediating pathway. The analytical framework first supports city-level diagnostics, allowing for structured comparisons across different urban settings and tourism development pathways within the GBA. It then extends beyond single-city analysis by incorporating spatial spillover effects, examining how environmental pressures originating in one city shape resident well-being and health outcomes in neighboring areas. The dual perspective supports both regional coordination and context-sensitive policy learning. The project aims to generate relevant policy insights into light usage efficiency and night-time tourism governance, informing strategies that balance economic vitality with environmental sustainability and resident well-being, with tightened coordination between Macau and the other GBA cities in the environmental aspects.

This project studies the design of transparent and verifiable allocation mechanisms for public housing, with a primary application to Macau’s social housing system. The allocation of public housing is a canonical matching problem between households and indivisible housing units under eligibility, priority, and capacity constraints, in an environment without prices and with strong normative requirements regarding fairness and equal treatment of equals. In practice, such systems rely on randomized priority rules, but the use of administrative or software-based randomization raises concerns about transparency and public trust. Building on market design theory and on historical and contemporary uses of publicly observable drawing of lots, this project models housing allocation as a constrained bipartite matching problem and designs lottery-based mechanisms whose outcomes are transparent, verifiable, and institutionally feasible. The project analyzes the efficiency and fairness properties of sequential and partitioned lots-drawing procedures using graph-theoretic tools, identifies when transparent random mechanisms achieve maximal allocations under housing-specific constraints, and characterizes the limits of such procedures when multiple constraints interact. The theoretical analysis is complemented by simulations calibrated to Macau’s housing context, with the goal of producing concrete, implementable recommendations for the design of public housing allocation, relocation, and exchange mechanisms in Macau.

The rapid adoption of AI-driven wealth management in Macau’s Modern Finance sector—a pillar of the SAR’s ‘1+4’ diversification strategy—creates an urgent need to understand the role of AI bias in manipulating investor behaviour. This proposal investigates how biased Large Language Models (LLMs) facilitate strategic manipulation in financial decision-making. Utilizing a controlled experimental setting, we test whether the directional bias of an LLM and the interactive tone it employs—ranging from analytical to authoritative or emotional—can be used to strategically manipulate investment outcomes, particularly in environments where decision-makers face cognitive constraints. Drawing on the framework of bounded rationality, we hypothesize that as investment tasks become more cognitively demanding, individuals engage in “attribute substitution”, relying on the AI as a decision proxy rather than a supplementary tool. Consequently, we expect the “bite” of the AI’s bias to intensify under high complexity, particularly when delivered through authoritative or emotional tones that bypass analytical scrutiny.

The requested funds from this application will contribute to a research project on imperfect human memory and belief formation.

Cross-border surrogacy tourism has emerged as a technologically mediated transnational reproductive market shaped by regulatory asymmetries, economic inequalities, and divergent moral norms. Advances in assisted reproductive technologies (ARTs), coupled with digital platforms and online brokerage services, have significantly lowered information and transaction costs, enabling individuals and couples to travel across borders to access surrogacy services that are prohibited, restricted, or prohibitively expensive in their home jurisdictions.

This project adopts an interdisciplinary approach to examine the legal governance and ethical implications of cross-border surrogacy tourism, with a particular focus on Asia-Pacific jurisdictions and their interaction with global, technology-enabled reproductive markets. It analyses conflicts of laws concerning parental status, nationality, and liability, alongside ethical issues relating to commodification, exploitation, data transparency, and the welfare of surrogate mothers and children. By integrating comparative legal analysis with social welfare, political economy, and technology-aware perspectives, the project illuminates how digital intermediation and reproductive technologies reshape risk allocation, regulatory capacity, and accountability.

As a seed-funded study, the project will generate pilot findings and conceptual frameworks that support future large-scale grant applications and contribute to evidence-based discussions on governing cross-border reproductive services in a technologically advanced yet ethically constrained global environment.

As smart tourism evolves into an integrated system for digital governance and service delivery, it increasingly facilitates high-value but also high-risk tourism practices. A critical and sensitive application within this domain is regenerative medical tourism (RMT), which combines advanced medical therapies with international travel. Driven by global advancements in regenerative medicine and a regulatory trend toward approval, the international demand for RMT is growing. This study conceptualizes smart RMT as an advanced form of smart tourism mediated primarily by intelligent digital platforms. It examines how these smart tourism systems structure the provision of RMT services, and the subsequent impacts on patient data protection, privacy, health rights, medical efficacy, and overall well-being. By integrating an analysis of RMT digital platforms with relevant legal and ethical frameworks, this study assesses the health, industry, and social impacts of smart RMT to provide recommendations for its regulated and sustainable development.

The rapid evolution of smart tourism demands a shift from static destination management to dynamic, sustainable traveler engagement. This research proposes a novel framework utilizing neuro-symbolic large language models (LLMs) to simulate high-fidelity human mobility for Personal Travel Planning (PTP). Traditional mobility models face significant limitations: rule-based agents lack cognitive depth, while deep learning acts as an uninterpretable “black box.” Conversely, standard LLMs offer exceptional semantic reasoning but often suffer from hallucinations, rendering generated itineraries spatially unfeasible.

To bridge these gaps, our approach integrates the intuitive perception of neural architectures (System 1) with the rigorous, constraint-based reasoning of symbolic systems (System 2). We employ a hierarchical “Narrative-to-Action” architecture, where agents generate motivation-driven narratives that are parsed into formal constraints for numerical solvers. This ensures simulated trajectories are not only personalized but also mathematically valid and physically executable. Furthermore, by incorporating Retrieval-Augmented Generation (RAG) and multi-horizon memory, the framework grounds simulations in real-time data. Ultimately, this provides destination managers with a scalable digital laboratory to optimize visitor flows, enhance satisfaction, and protect fragile cultural heritage sites from the pressures of overtourism.

This project investigates how technology-enabled living environments in smart cities shape tourists’ psychological perceptions and tourism-related decision-making. As high-rise living, smart buildings, and digitally mediated residential environments become increasingly prevalent, these contexts may alter individuals’ embodied experiences by weakening their sense of groundness while simultaneously heightening perceptions of power and control.

Drawing on embodied cognition and environmental psychology, this research conceptualizes daily living environments as a pre-travel psychological input within digital tourism ecosystems. Using surveys, controlled experiments, and virtual environment simulations, the project examines how high-rise versus low-rise and smart versus non-smart living environments influence preferences for destinations, accommodations, activities, and technology-enabled tourism services.

By integrating psychological mechanisms with smart tourism infrastructure, this research advances understanding of how digital transformation reshapes tourism demand formation. The findings will support the development of intelligent destination marketing systems, personalized tourism recommendations, and technology-enhanced hospitality design, contributing directly to smart tourism planning and digital innovation in tourism management.

The advent of generative AI technologies has profoundly impacted consumers in multifaceted ways. Traditionally, consumers have relied on search engines (e.g., Google) and word-of-mouth from both online and offline platforms to gather information for their decision-making processes (Lynch & Zauberman, 2007). The emergence of AI technologies offers a novel decision-making aid (Duan, Edwards, & Dwivedi, 2019). GenAI is capable of providing personalized itineraries and precise recommendations through deep analysis of vast travel data. For instance, travel companies like Expedia and Kayak are integrating their services into generative AI platforms, enabling consumers to solicit travel advice, plan, and book trips directly through these platforms. This research project aims to investigate the factors influencing consumer adoption of and trust in generative AI technologies, and how these technologies impact decision quality and satisfaction.

Guided by the consumer decision-making process frameworks of Engel, Blackwell, and Miniard (1986), and modified by Darley, Blankson, and Luethge (2010), consumer decision-making involves five stages: need recognition, information search, evaluation of alternatives, purchase, and post-purchase evaluation. Generative AI technologies are expected to predominantly influence the information search and alternatives evaluation stages by streamlining these processes through comprehensive and pertinent information provision. This, in turn, may impact decision quality through consumers’ perceived value, satisfaction, and regret after using AI technologies as decision-making aids, compared to traditional aids like online reviews and personal consultations.

This research hypothesizes that assistance from generative AI technologies will lead to better-informed decisions, higher satisfaction, and reduced regret due to its efficient and targeted information provision. However, it may also reduce users’ sense of agency and autonomy, impacting their overall happiness. This research project will provide a nuanced understanding of generative AI technologies’ role in enhancing consumer decision-making and overall well-being and provide actionable suggestions to practitioners in Smart Tourism industry.

This study aims to investigate the evolving use of Chinese vocabulary in Macao’s hotel and food and beverage industries, particularly in the context of smart tourism and multilingual communication. In recent years, Macao’s tourism industry saw a significant recovery, welcoming receiving 28.23 million visitors in 2023, which amounted to approximately 70% of pre-pandemic levels. Notably, tourists from mainland China accounted for 67.5%, making up the majority of visitors. As Macao’s economy heavily relies on tourism, its GDP increased by 80.5% in 2023, largely due to the surge in tourism revenue. In terms of language, Macao implements a “trilingual written and quadrilingual spoken languages” policy, and with the increasing influx of tourists, the use of Chinese, English, and Portuguese in the tourism industry has gradually increased. This trend is especially prominent with the advancement of smart tourism, where digital tools and big data technologies are widely applied. This multilingual environment provides rich linguistic data for studying Chinese vocabulary in Macao’s tourism, while also presenting new challenges and opportunities for Chinese vocabulary research. This study, based on the Generative Lexicon Theory, analyzes the qualia structure of key Mandarin in Macao’s Hotel Industry and Food and Beverage Industry, including the constitutive role, formal role, telic role, agentive role, and evaluative role. Additionally, the project will present vocabulary information in the form of digital labels and publish it on popular new media platforms such as Xiaohongshu, Youku, Bilibili, and YouTube. This project helps to reveal the usage contextual use of these vocabulary.

This project investigates privacy concerns arising from smart tourism, focusing on smart devices in hotel rooms and the development of a privacy risk assessment tool, with a special emphasis on Macao’s unique context as a UNESCO World Heritage site and Asia’s premier integrated resort destination. From check-in to check-out, smart services offer convenience but pose significant risks related to data collection and user privacy. These risks pose critical challenges to Macao’s vision of building a world-class smart tourism ecosystem, potentially deterring tourist engagement and threatening industry growth. In the context of Macao’s tourism industry accounting for 80% of its GDP, any security vulnerabilities in the large amounts of personal information collected by hotel smart home devices will directly impact the international reputation of the ‘World Heritage City.’ As Macao’s ‘Smart City Development Plan’ accelerates, with smart tourism as one of its core scenarios, ensuring privacy protection has become a key issue for sustainable development. The goal of this research is to assess privacy risks, identify vulnerabilities, and propose actionable guidelines. Through a combination of field studies, user surveys, and technical analyses, this project will contribute to the design of a robust privacy risk assessment tool specifically designed for Macao’s smart tourism scenarios. The tool will evaluate risks, identify vulnerabilities, and provide actionable guidelines. The research results will serve as a reference for hotel management, policymakers, and technology developers, ensuring a balance between convenience and privacy protection. This will consolidate Macao’s exemplary position in the smart tourism sector of the Guangdong-Hong Kong-Macao Greater Bay Area and promote high-quality development of the tourism economy.