machine learning

  • 详情 The Two-System View of Cognition and Investor Choice: Evidence from Mutual Fund Launch Livestreams
    This paper examines the interplay between investors’ intuition (System 1) and analytical processing (System 2) in investment decisions. Using livestream promotions for mutual fund offerings in China over 2020–2024, we study a setting without prior performance records or portfolio disclosures. Favorable impressions formed from presenters’ emotional displays—vocal tone, facial expressiveness, and body movement—are positively related to fund subscriptions. This relation is weaker when livestreams convey complex information or feature fund managers, consistent with greater System 2 processing. The attenuation is less pronounced when System 2 faces capacity constraints from concurrent livestreams, extreme market returns, or longer livestreams.
  • 详情 Picking Winners or Building Resilience? The Impact of China's AI Industrial Policy on Firm-Level Supply Chain Resilience
    Using data from Chinese A-share listed companies(2010-2023), this study examines how China's AI Innovation Application Pilot Zone Policy, a place-based industrial initiative that designates AI "winners", affects firm-level supply chain resilience. We employ a difference-indifferences framework combined with double machine learning to mitigate selection bias and high-dimensional confounding. Our findings reveal that the policy significantly bolsters firm-level supply chain resilience, with results robust to parallel trend and placebo tests. The mechanism operates through improved access to financial resources and greater operational efficiency. The effects are amplified among firms with greater digital readiness and managerial expertise in technology. A comprehensive heterogeneity analysis reveals systematically stronger gains across dimensions including non-state-owned, headquarters in first-tier cities, high-tech, growth-stage, higher digital maturity, ESG performance, and located in lower marketization regions. Our study challenges skepticism about industrial policy and provides actionable insights for policymakers and managers.
  • 详情 Linking agricultural land use efficiency and trade policy friendliness to embodied carbon transfers:Evidence from China-ASEAN agricultural trade
    Agricultural land use efficiency (ALUE) shapes global trade-related carbon transfer. However, whether and how efficiency improvements translate into carbon emission reductions remain unclear, particularly under different trade policy friendliness environments. Therefore, based on the theories of Jevons Paradox, New Institutional Economics and Trade Facilitation, this study applies Double Machine Learning with mediation and moderation analyses to explore this ALUE-trade policy friendliness-net embodied carbon transfer (NECT) relationship. Based on data on China-ASEAN agricultural trade from 2002 to 2021, we find: (1) Nearly half of export and import carbon are concentrated in countries with high trade policy friendliness toward China. (2) For ASEAN countries, while improvements in ALUE generally increase NECT to China, this effect is reversed in high-friendliness countries, where efficiency gains reduce net transfer. (3) Improvements in ALUE increase ASEAN countries’ NECT to China via an export-expansion channel, while trade-policy friendliness toward China dampens the latter stage of this mediation.
  • 详情 Regional FinTech Development and Corporate Greenwashing: Authenticating Green Information or Driving Green Transformation?
    Mounting pressures to achieve green transformation are pressuring firms to resort to greenwashing, thereby obscuring their actual environmental footprints. Notably, Financial Technology (FinTech) is transforming green governance in China by enabling multi-stakeholder collaboration among governments, financial institutions, and other entities. However, existing studies predominantly focus on a single type of actor, such as banks, and have yet to systematically investigate the mechanisms through which FinTech curbs corporate greenwashing. This study bridges this gap by proposing a framework of regional FinTech development and employing a double machine learning model for causal inference. It investigates whether the governance mechanism of FinTech operates through authenticating green information at the surface-level or driving green transformation at the root-cause level. The findings demonstrate that regional FinTech development curbs corporate greenwashing not only by authenticating green information through mitigated information asymmetry but also by driving corporate green transformation through bolstered internal controls and green innovation. Further heterogeneity analysis reveals that this effect is more pronounced for non-state-owned enterprises and firms in heavily polluting or regulated industries. Moreover, the effect is amplified in environments with lower levels of market development and intense media scrutiny.
  • 详情 When Markowitz Meets Machine: Optimization of Large Portfolios with High-Dimensional Stock Characteristics
    We propose a novel portfolio optimization framework that transforms the classic Markowitz mean-variance optimization into an unconstrained machine learning task. Our approach, Markowitz Meets Machine (MMM), directly utilizes high-dimensional firm characteristics as inputs to generate portfolio weights, without requiring explicit estimation of expected returns or covariance matrix. The MMM model balances the mean-variance tradeoff but circumvents many implementation challenges of traditional mean-variance optimization (MVO). Using China A-share market data with 267 proprietary signals, we construct long-only optimal portfolios that target outperformance relative to the CSI 300 and CSI 500 benchmarks. The MMM portfolio outperforms the CSI 500 benchmark by approximately 20% annually with tracking errors around 5%, significantly exceeding the performance of traditional multi-step MVO approaches after accounting for transaction costs. Our results demonstrate the importance of integrated modeling of expected returns, risk, and optimization.
  • 详情 Multi‑Source Heterogeneous Data Fusion and Stock Market Predictability: Evidence from Chinese Investor Sentiment and Market Indicators
    Investor sentiment constitutes a pivotal nonrational determinant of asset pricing within capital markets. Nevertheless, achieving high-precision sentiment analysis regarding Chinese financial short texts and effectively integrating these metrics into predictive modeling remains a formidable challenge. Focusing on Vanke A within the Chinese A-share market, this study proposes a quantitative framework for investor sentiment analysis that amalgamates textual commentary from the East Money Guba forum with high-frequency trading data from CSMAR, leveraging a hybrid architecture of BERT deep learning and CatBoost regression. By meticulously constructing a high-fidelity annotated corpus, the BERT model achieves a text sentiment classification accuracy of 94.19%, significantly outperforming conventional machine learning paradigms. Predicated on these classifications, a dynamic sentiment index is constructed, revealing that sentiment indicators exhibit a distinct leading effect on closing indices with a horizon of one to two days. Furthermore, by synthesizing sentiment factors with historical transaction characteristics, a temporally enhanced CatBoost prediction model is established. Following rigorous hyperparameter optimization, the model yields a goodness-of-fit ( ) of 0.953 and a mean squared error (MSE) of 0.0097, demonstrating exceptional predictive precision and robustness. The findings underscore that sentiment factors possess substantial explanatory and predictive power regarding short-term stock market volatility, thereby furnishing robust empirical evidence and methodological scaffolding for the development of quantitative investment strategies and financial risk monitoring systems that incorporate natural language processing technologies.
  • 详情 Is there any sentiment or animals’ spirits in the financial markets’ zoo? A narrative machine-learning analysis of the relationship between news in media and returns on the Chinese commodity futures market
    This study examines the relationship between price discovery and commodity-related media sentiment in the Chinese futures market through a tone-based natural language processing framework. We integrate a customized Chinese BERT model into a BiLSTM network with an Attention mechanism, enhancing semantic representation through contextual embeddings and salient feature emphasis. Analyzing over 850,000 news articles, we find sentiment provides significant incremental predictive power for commodity portfolio returns. We finally design a long-short trading strategy that consistently outperforms several benchmark approaches. The results emphasize that deep semantic sentiment analysis offers substantive predictive value for asset returns.
  • 详情 The Impact of Investor Sentiment on Financing Constraints: Evidence from Generative Artificial Intelligence–Based Measurement
    Against a backdrop of economic transformation and industrial upgrading, financing constraints have become a significant barrier to the growth of microeconomic entities. Investor sentiment, a recognized source of irrationality in capital markets, is understood to influence the financing environment through various channels. Concurrently, advances in generative artificial intelligence (AI) are generating novel methods for the precise measurement of this sentiment. Leveraging data from Chinese A-share listed corporations from 2011 to 2023, this study constructs a double machine learning framework to examine the impact of investor sentiment on financing constraints and its underlying transmission mechanisms. Our findings indicate that elevated investor sentiment significantly reduces these constraints. This effect operates primarily through an enhanced capacity for debt financing, a reduction in market scrutiny, and a stimulation of consumer demand that boosts corporate revenue. Conversely, periods of subdued sentiment tend to trigger risk-averse behavior, intensify liquidity pressures, and further restrict access to capital. A heterogeneity analysis reveals that the effects of sentiment vary across industries, ownership structures, and regional levels of information infrastructure. The impact is particularly pronounced within the real estate sector and in more digitally advanced regions. Interestingly, the effect does not fundamentally differ across ownership types, suggesting a degree of market neutrality. In sum, this study offers fresh empirical evidence on the mechanisms through which sentiment drives financing behavior. It also expands the application of generative AI, demonstrating its value as a tool for financial research.
  • 详情 Nature Transition Risk
    We investigate how investors assess transition risks arising from the discretionary regulatory mandates for nature conservation. We focus on the UN Biodiversity Conference in Kunming, China, a policy shock signaling heightened political commitment to protection. During the conference, firms exhibited negative abnormal stock returns on average. Using a machine learning approach, we identify historical nature-related subsidies to capture firm-level variations in expectations of mandated conservation engagement, beyond measuring corporate environmental (ecological) impacts. This metric negatively correlates with returns during the conference, particularly for firms with high operational volatility, and those in regions with limited fiscal flexibility and stringent conservation pressures. Investor concerns primarily stem from policy uncertainty, without positive earning expectations from conservation. Our findings underscore the role of authoritative discretion in shaping nature transition risks.
  • 详情 Machine-Learning-Augmented Government Bond Portfolio Management: Evidence from China's Treasury Market
    We study machine-learning forecasts and portfolio allocation for Chinese government bonds. Penalised linear regressions and gradient-boosted trees predict next-month total returns for eight maturity buckets. Their forecasts feed mean–variance, Black–Litterman, and risk-budgeting implementations with weight, duration, and trading-cost controls; convexity, tail-risk, and attribution constraints are analysed separately as theoretical extensions. From January 2019 through December 2025, the three optimised cash portfolios earn net excess-return Sharpe ratios of 1.94–2.26 and maximum drawdowns of 0.18–0.25%. Three long-only bond-selection variants attain Sharpe ratios of 1.39–1.70. A futures-hedged arbitrage variant attains 1.08 under an 80% hedge-P&L scaling and basis-cost assumption. A stylised 10-year futures overlay reduces rather than improves cash-portfolio Sharpe ratios.
0.102034s