The emergence of a series of new economic forms and phenomena in the era of the digital economy poses unprecedented challenges to traditional economic theories, creating an urgent need for theoretical innovation and development. Focusing on the core issues of the digital economy, this paper systematically examines theoretical advances in three foundational domains: the theory of data as a factor of production and value creation, theories of digital economic growth, and the political economy of the digital economy. First, it analyzes how the distinctive attributes of data as a novel production factor and its empowering characteristics drive profound transformations in value theory. Second, it explores the critical contribution of data factors to digital economic growth and the consequent reconstruction of growth theory. Finally, it investigates a range of structural impacts, including the labor-substitution effects of artificial intelligence, asymmetric relationships arising from the platform economy, the widening of social and income inequalities caused by the digital divide, as well as the implications of cross-border data flows and increasingly intense international competition in digital technologies for the global economic structure. This paper aims to offer new perspectives and analytical frameworks for economic theory innovation in the digital economy era, thereby fostering a closer alignment between economic theory and the evolving realities of the digital age.
Network externality is a typical feature of the digital economy. Over recent year, the network analysis literature has employed the framework of network games to move beyond the classical notion that externalities depend solely on aggregate or average social usage, and to characterize the heterogeneity of network externalities along the dimensions of scope, intensity, and directionality. This paper surveys the literature on optimal pricing of goods with network externalities, tracing the evolution and frontier developments under both complete and incomplete information about the externality network. We further discuss the empirical evidence on network externalities, the identification challenges involved, and the new opportunities arising in the digital economy. Finally, we offer reflections and an outlook on the emerging intersection between information economics and social network analysis.
The rapid development of the digital economy has reshaped investors’ decision environments and behavioral patterns by changing how information is generated, presented, and disseminated, thereby affecting asset prices. This paper reviews the literature on investor behavior and asset pricing in the digital economy from three analytical dimensions—information content, information presentation, and information dissemination structure—while also considering the evolution of decision-making agents from individual investors to human groups and algorithmic agents. Existing studies show that social media sentiment, mobile devices, and platform interfaces, by reshaping attention allocation and belief updating, render classic behavioral biases such as the disposition effect and overconfidence highly context-dependent. At the collective level, social networks and recommendation algorithms reinforce homogeneous information diffusion, manifesting as market-wide price co-movements. Meanwhile, with the widespread participation of algorithmic trading and AI agents in information processing and trade execution, market participants have evolved into a multi-agent structure in which humans and intelligent agents coexist. From an asset pricing perspective, these behavioral dynamics cause investor characteristics, collective interactions, and agent behaviors to crystallize into priced risk exposures, making investor factors and agent-behavior factors valuable complements to traditional characteristic-based factors. The digital economy thus affects asset prices not merely by improving informational efficiency, but by reconfiguring how information acts upon human investors and algorithmic agents, thereby altering trading behavior. Future research should establish a unified framework that integrates the multi-dimensional information environment (content, presentation, and dissemination) with the interactive dynamics among multi-agent behavioral subjects (individual humans, groups, and intelligent agents). Such a framework would help systematically elucidate the mechanisms through which investor behavior is translated into asset prices, thereby enhancing our understanding of asset pricing in the digital era.
With the rapid advancement of large language models (LLMs), the large-scale application of qualitative interviews in economics has become feasible. This paper introduces a multi-agent LLM–based approach to semi-structured qualitative interviewing and develops a complete workflow built upon it. In this framework, AI agents serve as interviewers to conduct standardized yet probing semi-structured interviews. Subsequently, LLMs, assisted by human reviewers, extract thematic insights and perform automated coding from the interview transcripts, thus achieving an integrated process from interviewing to coding. We implemented online interviews on three policy-relevant topics—fertility intentions, personal pension contributions, and stock market participation–and compared the performance of text-based and voice-based input modes. A total of 525 interviews were collected, with an average duration of 23 minutes each. The results indicate that, compared with one-off open-ended questioning, AI-led interviews can systematically elicit and track the considerations (themes) activated during multi-turn conversations-including their frequency, turn-level emergence, and co-occurrence patterns-reveal group heterogeneity, and produce results highly correlated with conventional closed-form survey measures. Subjectively, over95% of participants reported positive overall evaluations of the AI interviews, and the majority of participants expressed a preference for AI-conducted interviews. The entire interview and coding pipeline is replicable, auditable, and highly scalable. Finally, we discuss the practical challenges of applying this method in real-world settings and potential strategies to address them.
High-quality data are fundamental to ensuring the reliability and validity of statistical inference. Existing research on interval-valued data has primarily focused on modeling and estimation, while largely overlooking the problem of outlier detection. In this paper, we develop a robust outlier detection framework for high-dimensional interval-valued data based on a modified Mahalanobis distance. Specifically, we replace the covariance matrix in the Mahalanobis distance with a diagonal matrix composed of its diagonal entries, thereby circumventing the difficulties associated with estimating the full covariance matrix for high-dimensional interval data. We further extend the minimum diagonal product (MDP) estimator to the interval-valued setting to obtain a robust diagonal estimate. At the detection stage, the modified Mahalanobis distances are treated as test statistics in a multiple-testing framework, and the outlier detection threshold is adaptively determined using a Benjamini–Hochberg false discovery rate (FDR) control procedure. This approach enables the joint identification of potential outliers at a prespecified FDR level. On the optimization side, we propose a fast iterative algorithm for computing the interval MDP estimator and establish its convergence. Theoretically, we show that the extended estimator preserves the high breakdown-point robustness property of the original MDP estimator. Simulation studies and empirical analyses further demonstrate the effectiveness of the proposed outlier detection method.
In the era of the digital economy, platforms have experienced rapid growth, with both the types and scale of connected agents expanding continuously. Existing theories largely emphasize that an increase in the number of agents can generate social welfare through positive network externalities. However, a potential factor also deserves attention: as the diversity of agent types grows, the heterogeneity among different types may introduce significant negative network externalities, such as reduced matching efficiency, thereby lowering the utility of various agents on the platform. This paper focuses on the negative cross-group network externalities among different categories of agents and develops a pricing model for a monopolistic digital platform serving multiple agent types. The model investigates how the increase in agent types affects both agent utility and platform profit. The analysis shows that, regardless of the number of agent types, in equilibrium, there are agents of every type participating on the platform. Comparative statics reveal that as the number of agent types increases, the number of agents of each type joining the platform and their individual utilities decrease. The effects of the number of agent types on the total number of agents and the platform’s profit depend on the magnitude of network externalities among agents. When the intra-group negative externalities are small, both the total number of agents and the platform’s profit increase with the number of agent types. Further analysis indicates that when agents differ in their intrinsic utilities or in the externalities they impose on other types, some agent types may no longer obtain positive utility from joining the platform and therefore choose not to participate. This study offers a new theoretical perspective for understanding the change of agent utility and platform profit during platform expansion and provides insights into how platforms can better coordinate interactions among heterogeneous agent types.
The rapid development of artificial intelligence (AI), particularly large language models (LLMs), is demonstrating immense potential in the field of advertising content generation. However, its application faces two major challenges: First, the subjectivity and opacity of advertising quality assessment standards; second, the lack of interpretability in the AI-driven evaluation process. This paper proposes an innovative context-aware evaluation system (CAES), driven by a multi-agent framework. Built upon a multi-agent collaboration mechanism, the framework establishes dynamic interactions among an evaluation agent, a gradient engine agent, and an evaluation optimization agent. By learning from real advertising data and user behavior feedback (such as click-through rate, CTR), the system continuously optimizes the advertising content generation process. Through a series of experimental analyses, this study validates the effectiveness of the proposed framework in both advertising effectiveness assessment and content optimization. The CAES achieves an interpretable quantitative assessment of advertising content, thereby overcoming the subjectivity inherent in traditional evaluation systems. Furthermore, it can backpropagate quantified user feedback signals into the semantic space of the generative model, realizing the automatic and continuous optimization of the assessment framework, significantly contributing to the enhancement of advertising evaluation efficiency and interpretability.
Stablecoins serve as a key bridge between the crypto-financial system and the real-world financial system. Their value anchoring mechanisms and potential risks have attracted increasing attention from both academia and policymakers. Based on existing literature, this paper systematically analyzes the value anchoring mechanisms of three mainstream types of stablecoins — fiat-backed, over-collateralized, and algorithmic stablecoins —from a mathematical modeling perspective. Using the 2022 UST de-pegging event as an empirical case, we construct a price dynamics equation and a redemption pressure model based on on-chain data to characterize the risk transmission mechanism. Furthermore, network representation and systemic risk measures are introduced to examine the structural position of stablecoins in the crypto-financial ecosystem and their role in liquidity allocation, price discovery, and risk spillovers. The results suggest that stablecoins represent a form of nested financial innovation that simultaneously relies on external credit support and market expectations. Finally, the paper evaluates current regulatory practices and discusses future institutional evolution.
Promoting the market-oriented pricing of municipal bonds is a critical step in preventing and mitigating local government debt risks. Against the backdrop of persistent pressure on the financial fundamentals of financing platforms, the pricing mechanism of these bonds has revealed a growing problem of credit deviation and risk mismatch, which has become a major concern for financial stability. However, regarding the intrinsic mechanisms that lead to distortions in risk pricing, existing studies mostly focus on macro policies and traditional credit risk frameworks, but relatively few examine how the information environment and institutional expectations shape pricing dynamics through big data and artificial intelligence methods. Drawing on text mining and deep learning techniques, this study investigates the influence of a media sentiment index and implicit guarantee expectations on the pricing of municipal bonds, as well as their interaction effects. The analysis shows that media sentiment significantly compresses issuance spreads, while both the capacity and willingness associated with implicit guarantees amplify pricing deviations and materially weaken the marginal effect of media sentiment in reducing issuance spreads. These findings uncover the interaction mechanism between media information and implicit guarantee expectations in the digital economy era, offering important empirical evidence and policy insights for improving the market-oriented pricing of municipal bonds and regulating implicit guarantee practices.
Social capital refers to networks of reciprocity and mutual assistance formed on the basis of kinship, geographic proximity, and similar ties, together with the embedded resources of trust (credit) and risk sharing they entail. Household gift expenditure plays an important role in sustaining social capital by functioning as quasi-credit and informal finance, yet it imposes substantial economic burdens on many families. In the context of the rapid development of digital inclusive finance, an important question arises: Can innovative formal financial services help relieve households’ financial pressure associated with maintaining social capital? Using data from the China family panel studies (CFPS) matched with the Peking University digital inclusive finance index, this study examines whether digital inclusive finance reduces households’ gift expenditure related to social capital maintenance. The results show that digital inclusive finance significantly lowers household gift spending. Mechanism analyses indicate that digital inclusive finance mitigates households’ credit constraints and enhances the convenience of providing direct economic assistance, thereby reducing their reliance on borrowing within social networks and subsequently decreasing gift expenditure. Further analyses reveal that the reduction of gift spending increases household consumption and weakens social trust within close-knit circles, while leaving intra-household trust and broader societal trust unaffected. This study enriches the literature on the social effects of digital inclusive finance and highlights the need to consider both economic and social outcomes when evaluating financial innovation.
The digital economy, with new-generation information technology as its driver and data as its core factor of production, is profoundly reshaping the global industrial landscape. Against this backdrop, promoting the intelligent transformation of supply chains and enhancing their visibility has become critical for maintaining industrial stability and strengthening corporate core competitiveness. This paper focuses on the frontier issue of supply chain risk management, exploring the use of graph learning technology to provide a new analytical framework for intelligent supply chain management. To this end, we propose a hierarchy- and path-aware network prediction model. The model takes a multi-source, heterogeneous graph that integrates supply chain, industry, and product information as input. Through an innovative composite loss function, it injects domain knowledge, such as hierarchical industry structure and semantic meta-paths, into the training process of a graph neural network to predict potential supply relationships between enterprises. Its performance is comprehensively compared against several mainstream baseline models. Results show that: 1) Compared to mainstream graph embedding and graph neural network methods, our proposed model that integrates domain knowledge achieves significant performance gains of 22.1% on Recall@20 and 17.4% on NDCG@20, exhibiting state-of-the-art performance; 2) The step-wise introduction of hierarchy and meta-path information not only enhances the model’s interpretability but also significantly improves its predictive capability for complex business relationships. This research provides a novel tool for enterprises to conduct data-driven supply chain risk management and strategic partner identification.
The rise of large language model has driven a new wave of transformation in the field of artificial intelligence, significantly enhancing the ability of agents to handle open-ended, dynamic, and complex tasks. Such progress has enabled agents to transition from “rule-based execution” to “autonomous intelligent decision-making”. This study constructs a large language model agent to tackle the exploratory task of optimal risk factor mining, which requires large-scale, multi-stage processing. The agent is designed to overcome the limitations of traditional manual methods in terms of efficiency, objectivity, and timeliness, systematically mining optimal risk factors that are constructed based on reliable methods, and are characterized by clear economic significance and strong predictive power. The effectiveness of the large language model agent is validated in the context of mining endogenous risk factors for small and micro enterprises. The results show that the large language model agent can quickly and cost-effectively extract information from nearly 800 literatures, perform nearly 300,000 calculations, and conduct 252regression tests, ultimately identifying optimal risk factors with both theoretical and statistical significance. This study not only provides an automated and reproducible new paradigm for risk factor mining but also expands the application boundaries of large language model agent in complex professional tasks.
In the era of the digital economy, how to fully leverage data factors to expand corporate innovation boundaries and facilitate high-quality corporate innovation has become a critical research topic. Based on the policy shock of the construction of government open public data platforms, this paper employs data of listed manufacturing firms from 2012 to 2022 and adopts the multi-period DID (difference-in-differences) model to examine the impact of government open public data on the expansion of corporate innovation boundaries and its underlying mechanism. The results indicate that government open public data significantly promotes corporate innovation in emerging technological fields, with the mechanism lying in the reduction of enterprises’ institutional transaction costs and the enhancement of their technological capability accumulation. Further analysis reveals that government open public data exerts a notably positive effect on expanding innovation boundaries in both the digital technology sector and the strategic emerging industries. In addition, the innovation boundary expansion effect of government open public data platforms exhibits significant heterogeneity. From the novel perspective of open public data, this study identifies a crucial factor driving the expansion of corporate innovation boundaries, thereby providing important policy implications for the Chinese government to further improve the open public data system, advance high-quality corporate innovation, and foster new-quality productive forces.
