On June 29, 2026, CEIBS hosted the 9th CEIBS Accounting and Finance Symposium at its Shanghai campus, bringing together more than 80 scholars from leading universities across the United States, Europe, and Asia. The one-day event provided a vibrant platform for interdisciplinary exchange on the evolving landscape of finance, accounting, economics, and technology.

Organised by the CEIBS Department of Finance and Accounting, the Symposium has grown into an important forum for scholars to share cutting-edge research and exchange ideas with implications for both academic theory and business practice. The 2026 edition featured work by both established scholars and emerging voices in the field. From more than 130 submissions, seven papers were selected for presentation. Professor Qiang Cheng of Singapore Management University delivered the keynote address, presenting his paper, “The Value of Man in an AI-Assisted Decision Model: Field Evidence from Small Business Lending.”

In his opening remarks, CEIBS Vice President and Co-Dean Professor Zhu Tian underscored the strong alignment between the Symposium and CEIBS’s guiding philosophy of “China Depth, Global Breadth.” He noted that convening prominent international scholars to examine global issues of direct relevance to China reflects CEIBS’s strong commitment to research excellence.

Keynote Address: The Value of Man in an AI-Assisted Decision Model: Field Evidence from Small Business Lending

Presenter: Qiang Cheng, Singapore Management University

As more companies adopt AI-assisted decision models, the question of how humans contribute to such models has become increasingly important for both organisations and employees. Professor Qiang Cheng and co-authors examine this issue by investigating the value of human loan officers in an AI-assisted lending model, in which loan officers make loan approval decisions based on recommendations from an AI model. Using proprietary loan-level data from a lending company that provides loans to small businesses, they find that the incremental contribution of loan officers to loan profits, beyond AI recommendations, increases with the soft information collected and used by loan officers and decreases with agency issues, algorithm aversion, and mental fatigue.

Further analyses show that evaluating loan officers based on both the quantity and quality of approved loans enhances the value of soft information and mitigates the adverse impact of agency issues, but exacerbates the negative effects of algorithm aversion and mental fatigue. The paper sheds light on the conditions under which humans can create or destroy value in AI-assisted decision models, offering timely insights for organisations navigating the integration of AI into decision-making processes.

How Does Big Data Affect Accounting Information Acquisition?

Presenter: Kuo Zhang, Shanghai Jiao Tong University

Discussant: Xuefeng Jiang, Michigan State University

The rise of big data has raised fundamental questions about the continued relevance of traditional accounting information. Contrary to the view that big data displaces traditional reporting, Professor Kuo Zhang and co-authors find that the availability of big data is positively associated with the use of traditional accounting reports, as measured by EDGAR searches by institutional investors. Using the staggered introduction of satellite imagery in a difference-in-differences setting, they show that institutional EDGAR searches increase significantly for covered retailers relative to uncovered firms.

The increase in searches is stronger when satellite data contain more meaningful information. It primarily targets recent filings and filings related to financial results and insider trading, and occurs both well before and immediately after earnings announcements. Consistent with a complementarity view, accounting information and satellite signals better predict future firm performance when used jointly, and institutional trades become more informed when accompanied by greater institutional EDGAR searches following satellite coverage. The findings suggest that investors use traditional accounting information to contextualise and validate insights from big data, highlighting its continued relevance in the big data era.

When LLMs Go Abroad: Foreign Bias in AI Financial Predictions

Presenter: Xiang Yi, The Hong Kong Polytechnic University

Discussant: Travis Chow, University of Hong Kong

Professor Xiang Yi and co-authors document “foreign bias” in AI-generated financial analysis, reversing the classic home bias observed among human investors. They find that ChatGPT, developed by U.S.-based OpenAI, is systematically more optimistic than China-developed DeepSeek about Chinese firms in price predictions, earnings forecasts, and qualitative business descriptions, while its quantitative forecasts are significantly less accurate.

The evidence supports an information-availability mechanism: the bias tracks cross-border news coverage gaps, attenuates for cross-listed firms, disappears when Chinese news is injected at inference, and is absent when both models analyse U.S. firms. Placebo and robustness tests rule out alternative explanations related to model alignment and prompt elicitation. The bias is not inherently optimistic; rather, its direction depends on whether U.S. sources underreport negative or positive news and therefore cannot be predicted in advance. The findings indicate that LLMs trained in different information environments can generate divergent and hard-to-predict signals, with important implications for investors and policymakers as AI increasingly intermediates global markets.

The Information Content of Loan Contract Disclosures

Presenter: Paul Demeré, Bocconi University

Discussant: Weili Ge, University of Washington

Professor Paul Demeré and co-authors examine whether publicly disclosed loan contracts contain information that corporate stakeholders can use to predict borrower outcomes. Loan contracts are the result of negotiations between informed parties following extensive due diligence, but strategic disclosure incentives and reliance on boilerplate language may limit their usefulness to stakeholders.

Using natural language processing techniques applied to a large sample of SEC-filed loan contracts, the authors find that contract terms and language, particularly discussion topics within the contracts, contain incremental predictive information about future borrower performance, investment, bankruptcy risk, and stock return volatility. They also find evidence that loan contract topics are associated with future stock returns and analyst forecast errors, suggesting that many corporate stakeholders struggle to process and use this information in a timely manner. The results indicate that loan contracts contain substantial information that is valuable to corporate stakeholders and highlight the richness of loan contract text.

Beyond Disclosure: Signaling Corporate Ties on Social Media

Presenter: Enshuai Yu, National University of Singapore

Discussant: Cyrus Aghamolla, Rice University

Professor Enshuai Yu and co-authors provide systematic evidence on the relational dimension of social media. The study shows that firms use peer tweets to signal economic ties with other firms. By analysing firms that actively follow and share the developments of their peers on social media, the authors shed light on the characteristics of firms engaging in corporate social signalling, whether such signals predict future strengthened economic ties, and how investors perceive such tweets.

They find that peer tweeting is associated with economic relatedness, demand for legitimacy, and litigation risk. These tweets predict future interfirm collaborations and convey information not readily available through traditional disclosure channels. Peer tweets also generate significantly positive market reactions for both the tweeting and tweeted firms. However, corporate social signalling also has a downside: tweeting firms experience negative market reactions when firms they previously tweeted about subsequently release negative news. These findings highlight the relational dimension of social media beyond its well-studied disclosure and dissemination roles.

Transparency as a Double-Edged Sword: Evidence from Social Score Disclosure in the Agency MBS Market

Presenter: Helen Zhang, University of Minnesota

Discussant: Sterling Huang, NYU Shanghai

Classic disclosure theories predict that transparency improves market liquidity by reducing information asymmetry. Yet for assets whose value depends on fungibility, greater information may instead increase adverse selection and impair market liquidity. Professor Helen Zhang and co-authors examine this dual role of transparency through the lens of social score disclosure in the agency mortgage-backed securities market.

This market provides a unique setting because the same mortgage-backed securities can be traded as individual bonds in the specified pool market or pooled with similar securities and traded through standardised contracts in the to-be-announced market. The authors first show that social scores are informative about prepayment risk. They then contrast the impact of social score disclosure across the specified pool and to-be-announced markets. In the specified pool market, mortgage-backed securities with higher scores, indicating lower prepayment risk, enjoy a price premium and improved market liquidity after disclosure. However, because social score disclosure allows sellers to better identify the riskiest eligible securities to deliver for a given price, the to-be-announced market experiences lower prices and liquidity. The findings provide novel evidence that transparency improves price discovery in the specified pool market but exacerbates adverse selection in the to-be-announced market, highlighting how the effects of transparency depend on market fungibility and depth.

Strategic Forecasts under Ambiguity

Presenter: Wei Shao, CEIBS

Discussant: Huai Zhang, Nanyang Technological University

CEIBS Assistant Professor of Finance Wei Shao and co-authors study how ambiguity shapes analysts’ forecasts by creating an additional layer of information asymmetry between analysts and investors. Because ambiguity-averse investors underreact to good news and overreact to bad news, the model predicts that analysts strategically inflate upward revisions and soften downward revisions in equilibrium. This implies that forecast errors are negatively correlated with positive revisions and positively correlated with negative revisions.

Using analyst forecasts from U.S. and international markets, the authors find support for these predictions at both the individual and consensus levels and across short- and long-term horizons. The findings point to the importance of client-facing incentives in financial intermediation and provide a novel theoretical and empirical framework for understanding analysts’ asymmetric forecast behaviour under ambiguity.

Reporting Data as an Asset

Presenter: Xinyi Zhang, Sun Yat-sen University

Discussant: Cheng Zeng, The Hong Kong Polytechnic University

In 2024, China adopted accounting standards allowing firms to recognise qualified data resources as assets on the balance sheet. Professor Xinyi Zhang and co-authors study the determinants and consequences of data asset recognition and report three main findings. First, firms with greater data intensity, tighter financing constraints, undervaluation concerns, and state ownership are more likely to recognise data assets, indicating both informational and strategic motives. Second, recognition increases the value relevance of net income and book value and strengthens earnings response coefficients. Third, using a proxy derived from large language models, the authors show that recognition raises data-related investment intentions without crowding out other investments.

Overall, the evidence suggests that discretionary recognition with institutional oversight can improve the usefulness of financial statements and spur investment in data assets, informing debates on accounting for intangibles in the digital economy and offering guidance for global standard-setters.